os_linux.cpp revision 7882:694f5e5bb982
1/* 2 * Copyright (c) 1999, 2014, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25// no precompiled headers 26#include "classfile/classLoader.hpp" 27#include "classfile/systemDictionary.hpp" 28#include "classfile/vmSymbols.hpp" 29#include "code/icBuffer.hpp" 30#include "code/vtableStubs.hpp" 31#include "compiler/compileBroker.hpp" 32#include "compiler/disassembler.hpp" 33#include "interpreter/interpreter.hpp" 34#include "jvm_linux.h" 35#include "memory/allocation.inline.hpp" 36#include "memory/filemap.hpp" 37#include "mutex_linux.inline.hpp" 38#include "oops/oop.inline.hpp" 39#include "os_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// Used to protect dlsym() calls 162static pthread_mutex_t dl_mutex; 163 164// Declarations 165static void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 166 167// utility functions 168 169static int SR_initialize(); 170 171julong os::available_memory() { 172 return Linux::available_memory(); 173} 174 175julong os::Linux::available_memory() { 176 // values in struct sysinfo are "unsigned long" 177 struct sysinfo si; 178 sysinfo(&si); 179 180 return (julong)si.freeram * si.mem_unit; 181} 182 183julong os::physical_memory() { 184 return Linux::physical_memory(); 185} 186 187//////////////////////////////////////////////////////////////////////////////// 188// environment support 189 190bool os::getenv(const char* name, char* buf, int len) { 191 const char* val = ::getenv(name); 192 if (val != NULL && strlen(val) < (size_t)len) { 193 strcpy(buf, val); 194 return true; 195 } 196 if (len > 0) buf[0] = 0; // return a null string 197 return false; 198} 199 200 201// Return true if user is running as root. 202 203bool os::have_special_privileges() { 204 static bool init = false; 205 static bool privileges = false; 206 if (!init) { 207 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 208 init = true; 209 } 210 return privileges; 211} 212 213 214#ifndef SYS_gettid 215// i386: 224, ia64: 1105, amd64: 186, sparc 143 216 #ifdef __ia64__ 217 #define SYS_gettid 1105 218 #elif __i386__ 219 #define SYS_gettid 224 220 #elif __amd64__ 221 #define SYS_gettid 186 222 #elif __sparc__ 223 #define SYS_gettid 143 224 #else 225 #error define gettid for the arch 226 #endif 227#endif 228 229// Cpu architecture string 230#if defined(ZERO) 231static char cpu_arch[] = ZERO_LIBARCH; 232#elif defined(IA64) 233static char cpu_arch[] = "ia64"; 234#elif defined(IA32) 235static char cpu_arch[] = "i386"; 236#elif defined(AMD64) 237static char cpu_arch[] = "amd64"; 238#elif defined(ARM) 239static char cpu_arch[] = "arm"; 240#elif defined(PPC32) 241static char cpu_arch[] = "ppc"; 242#elif defined(PPC64) 243static char cpu_arch[] = "ppc64"; 244#elif defined(SPARC) 245 #ifdef _LP64 246static char cpu_arch[] = "sparcv9"; 247 #else 248static char cpu_arch[] = "sparc"; 249 #endif 250#elif defined(AARCH64) 251static char cpu_arch[] = "aarch64"; 252#else 253 #error Add appropriate cpu_arch setting 254#endif 255 256 257// pid_t gettid() 258// 259// Returns the kernel thread id of the currently running thread. Kernel 260// thread id is used to access /proc. 261// 262// (Note that getpid() on LinuxThreads returns kernel thread id too; but 263// on NPTL, it returns the same pid for all threads, as required by POSIX.) 264// 265pid_t os::Linux::gettid() { 266 int rslt = syscall(SYS_gettid); 267 if (rslt == -1) { 268 // old kernel, no NPTL support 269 return getpid(); 270 } else { 271 return (pid_t)rslt; 272 } 273} 274 275// Most versions of linux have a bug where the number of processors are 276// determined by looking at the /proc file system. In a chroot environment, 277// the system call returns 1. This causes the VM to act as if it is 278// a single processor and elide locking (see is_MP() call). 279static bool unsafe_chroot_detected = false; 280static const char *unstable_chroot_error = "/proc file system not found.\n" 281 "Java may be unstable running multithreaded in a chroot " 282 "environment on Linux when /proc filesystem is not mounted."; 283 284void os::Linux::initialize_system_info() { 285 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 286 if (processor_count() == 1) { 287 pid_t pid = os::Linux::gettid(); 288 char fname[32]; 289 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 290 FILE *fp = fopen(fname, "r"); 291 if (fp == NULL) { 292 unsafe_chroot_detected = true; 293 } else { 294 fclose(fp); 295 } 296 } 297 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 298 assert(processor_count() > 0, "linux error"); 299} 300 301void os::init_system_properties_values() { 302 // The next steps are taken in the product version: 303 // 304 // Obtain the JAVA_HOME value from the location of libjvm.so. 305 // This library should be located at: 306 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 307 // 308 // If "/jre/lib/" appears at the right place in the path, then we 309 // assume libjvm.so is installed in a JDK and we use this path. 310 // 311 // Otherwise exit with message: "Could not create the Java virtual machine." 312 // 313 // The following extra steps are taken in the debugging version: 314 // 315 // If "/jre/lib/" does NOT appear at the right place in the path 316 // instead of exit check for $JAVA_HOME environment variable. 317 // 318 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 319 // then we append a fake suffix "hotspot/libjvm.so" to this path so 320 // it looks like libjvm.so is installed there 321 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 322 // 323 // Otherwise exit. 324 // 325 // Important note: if the location of libjvm.so changes this 326 // code needs to be changed accordingly. 327 328 // See ld(1): 329 // The linker uses the following search paths to locate required 330 // shared libraries: 331 // 1: ... 332 // ... 333 // 7: The default directories, normally /lib and /usr/lib. 334#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 335 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 336#else 337 #define DEFAULT_LIBPATH "/lib:/usr/lib" 338#endif 339 340// Base path of extensions installed on the system. 341#define SYS_EXT_DIR "/usr/java/packages" 342#define EXTENSIONS_DIR "/lib/ext" 343 344 // Buffer that fits several sprintfs. 345 // Note that the space for the colon and the trailing null are provided 346 // by the nulls included by the sizeof operator. 347 const size_t bufsize = 348 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 349 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 350 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 351 352 // sysclasspath, java_home, dll_dir 353 { 354 char *pslash; 355 os::jvm_path(buf, bufsize); 356 357 // Found the full path to libjvm.so. 358 // Now cut the path to <java_home>/jre if we can. 359 pslash = strrchr(buf, '/'); 360 if (pslash != NULL) { 361 *pslash = '\0'; // Get rid of /libjvm.so. 362 } 363 pslash = strrchr(buf, '/'); 364 if (pslash != NULL) { 365 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 366 } 367 Arguments::set_dll_dir(buf); 368 369 if (pslash != NULL) { 370 pslash = strrchr(buf, '/'); 371 if (pslash != NULL) { 372 *pslash = '\0'; // Get rid of /<arch>. 373 pslash = strrchr(buf, '/'); 374 if (pslash != NULL) { 375 *pslash = '\0'; // Get rid of /lib. 376 } 377 } 378 } 379 Arguments::set_java_home(buf); 380 set_boot_path('/', ':'); 381 } 382 383 // Where to look for native libraries. 384 // 385 // Note: Due to a legacy implementation, most of the library path 386 // is set in the launcher. This was to accomodate linking restrictions 387 // on legacy Linux implementations (which are no longer supported). 388 // Eventually, all the library path setting will be done here. 389 // 390 // However, to prevent the proliferation of improperly built native 391 // libraries, the new path component /usr/java/packages is added here. 392 // Eventually, all the library path setting will be done here. 393 { 394 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 395 // should always exist (until the legacy problem cited above is 396 // addressed). 397 const char *v = ::getenv("LD_LIBRARY_PATH"); 398 const char *v_colon = ":"; 399 if (v == NULL) { v = ""; v_colon = ""; } 400 // That's +1 for the colon and +1 for the trailing '\0'. 401 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 402 strlen(v) + 1 + 403 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1, 404 mtInternal); 405 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch); 406 Arguments::set_library_path(ld_library_path); 407 FREE_C_HEAP_ARRAY(char, ld_library_path); 408 } 409 410 // Extensions directories. 411 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 412 Arguments::set_ext_dirs(buf); 413 414 FREE_C_HEAP_ARRAY(char, buf); 415 416#undef DEFAULT_LIBPATH 417#undef SYS_EXT_DIR 418#undef EXTENSIONS_DIR 419} 420 421//////////////////////////////////////////////////////////////////////////////// 422// breakpoint support 423 424void os::breakpoint() { 425 BREAKPOINT; 426} 427 428extern "C" void breakpoint() { 429 // use debugger to set breakpoint here 430} 431 432//////////////////////////////////////////////////////////////////////////////// 433// signal support 434 435debug_only(static bool signal_sets_initialized = false); 436static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 437 438bool os::Linux::is_sig_ignored(int sig) { 439 struct sigaction oact; 440 sigaction(sig, (struct sigaction*)NULL, &oact); 441 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 442 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 443 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) { 444 return true; 445 } else { 446 return false; 447 } 448} 449 450void os::Linux::signal_sets_init() { 451 // Should also have an assertion stating we are still single-threaded. 452 assert(!signal_sets_initialized, "Already initialized"); 453 // Fill in signals that are necessarily unblocked for all threads in 454 // the VM. Currently, we unblock the following signals: 455 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 456 // by -Xrs (=ReduceSignalUsage)); 457 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 458 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 459 // the dispositions or masks wrt these signals. 460 // Programs embedding the VM that want to use the above signals for their 461 // own purposes must, at this time, use the "-Xrs" option to prevent 462 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 463 // (See bug 4345157, and other related bugs). 464 // In reality, though, unblocking these signals is really a nop, since 465 // these signals are not blocked by default. 466 sigemptyset(&unblocked_sigs); 467 sigemptyset(&allowdebug_blocked_sigs); 468 sigaddset(&unblocked_sigs, SIGILL); 469 sigaddset(&unblocked_sigs, SIGSEGV); 470 sigaddset(&unblocked_sigs, SIGBUS); 471 sigaddset(&unblocked_sigs, SIGFPE); 472#if defined(PPC64) 473 sigaddset(&unblocked_sigs, SIGTRAP); 474#endif 475 sigaddset(&unblocked_sigs, SR_signum); 476 477 if (!ReduceSignalUsage) { 478 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 479 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 480 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 481 } 482 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 483 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 484 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 485 } 486 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 487 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 488 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 489 } 490 } 491 // Fill in signals that are blocked by all but the VM thread. 492 sigemptyset(&vm_sigs); 493 if (!ReduceSignalUsage) { 494 sigaddset(&vm_sigs, BREAK_SIGNAL); 495 } 496 debug_only(signal_sets_initialized = true); 497 498} 499 500// These are signals that are unblocked while a thread is running Java. 501// (For some reason, they get blocked by default.) 502sigset_t* os::Linux::unblocked_signals() { 503 assert(signal_sets_initialized, "Not initialized"); 504 return &unblocked_sigs; 505} 506 507// These are the signals that are blocked while a (non-VM) thread is 508// running Java. Only the VM thread handles these signals. 509sigset_t* os::Linux::vm_signals() { 510 assert(signal_sets_initialized, "Not initialized"); 511 return &vm_sigs; 512} 513 514// These are signals that are blocked during cond_wait to allow debugger in 515sigset_t* os::Linux::allowdebug_blocked_signals() { 516 assert(signal_sets_initialized, "Not initialized"); 517 return &allowdebug_blocked_sigs; 518} 519 520void os::Linux::hotspot_sigmask(Thread* thread) { 521 522 //Save caller's signal mask before setting VM signal mask 523 sigset_t caller_sigmask; 524 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 525 526 OSThread* osthread = thread->osthread(); 527 osthread->set_caller_sigmask(caller_sigmask); 528 529 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 530 531 if (!ReduceSignalUsage) { 532 if (thread->is_VM_thread()) { 533 // Only the VM thread handles BREAK_SIGNAL ... 534 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 535 } else { 536 // ... all other threads block BREAK_SIGNAL 537 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 538 } 539 } 540} 541 542////////////////////////////////////////////////////////////////////////////// 543// detecting pthread library 544 545void os::Linux::libpthread_init() { 546 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 547 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 548 // generic name for earlier versions. 549 // Define macros here so we can build HotSpot on old systems. 550#ifndef _CS_GNU_LIBC_VERSION 551 #define _CS_GNU_LIBC_VERSION 2 552#endif 553#ifndef _CS_GNU_LIBPTHREAD_VERSION 554 #define _CS_GNU_LIBPTHREAD_VERSION 3 555#endif 556 557 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 558 if (n > 0) { 559 char *str = (char *)malloc(n, mtInternal); 560 confstr(_CS_GNU_LIBC_VERSION, str, n); 561 os::Linux::set_glibc_version(str); 562 } else { 563 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 564 static char _gnu_libc_version[32]; 565 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 566 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 567 os::Linux::set_glibc_version(_gnu_libc_version); 568 } 569 570 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 571 if (n > 0) { 572 char *str = (char *)malloc(n, mtInternal); 573 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 574 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 575 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 576 // is the case. LinuxThreads has a hard limit on max number of threads. 577 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 578 // On the other hand, NPTL does not have such a limit, sysconf() 579 // will return -1 and errno is not changed. Check if it is really NPTL. 580 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 581 strstr(str, "NPTL") && 582 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 583 free(str); 584 os::Linux::set_libpthread_version("linuxthreads"); 585 } else { 586 os::Linux::set_libpthread_version(str); 587 } 588 } else { 589 // glibc before 2.3.2 only has LinuxThreads. 590 os::Linux::set_libpthread_version("linuxthreads"); 591 } 592 593 if (strstr(libpthread_version(), "NPTL")) { 594 os::Linux::set_is_NPTL(); 595 } else { 596 os::Linux::set_is_LinuxThreads(); 597 } 598 599 // LinuxThreads have two flavors: floating-stack mode, which allows variable 600 // stack size; and fixed-stack mode. NPTL is always floating-stack. 601 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 602 os::Linux::set_is_floating_stack(); 603 } 604} 605 606///////////////////////////////////////////////////////////////////////////// 607// thread stack 608 609// Force Linux kernel to expand current thread stack. If "bottom" is close 610// to the stack guard, caller should block all signals. 611// 612// MAP_GROWSDOWN: 613// A special mmap() flag that is used to implement thread stacks. It tells 614// kernel that the memory region should extend downwards when needed. This 615// allows early versions of LinuxThreads to only mmap the first few pages 616// when creating a new thread. Linux kernel will automatically expand thread 617// stack as needed (on page faults). 618// 619// However, because the memory region of a MAP_GROWSDOWN stack can grow on 620// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 621// region, it's hard to tell if the fault is due to a legitimate stack 622// access or because of reading/writing non-exist memory (e.g. buffer 623// overrun). As a rule, if the fault happens below current stack pointer, 624// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 625// application (see Linux kernel fault.c). 626// 627// This Linux feature can cause SIGSEGV when VM bangs thread stack for 628// stack overflow detection. 629// 630// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 631// not use this flag. However, the stack of initial thread is not created 632// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 633// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 634// and then attach the thread to JVM. 635// 636// To get around the problem and allow stack banging on Linux, we need to 637// manually expand thread stack after receiving the SIGSEGV. 638// 639// There are two ways to expand thread stack to address "bottom", we used 640// both of them in JVM before 1.5: 641// 1. adjust stack pointer first so that it is below "bottom", and then 642// touch "bottom" 643// 2. mmap() the page in question 644// 645// Now alternate signal stack is gone, it's harder to use 2. For instance, 646// if current sp is already near the lower end of page 101, and we need to 647// call mmap() to map page 100, it is possible that part of the mmap() frame 648// will be placed in page 100. When page 100 is mapped, it is zero-filled. 649// That will destroy the mmap() frame and cause VM to crash. 650// 651// The following code works by adjusting sp first, then accessing the "bottom" 652// page to force a page fault. Linux kernel will then automatically expand the 653// stack mapping. 654// 655// _expand_stack_to() assumes its frame size is less than page size, which 656// should always be true if the function is not inlined. 657 658#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 659 #define NOINLINE 660#else 661 #define NOINLINE __attribute__ ((noinline)) 662#endif 663 664static void _expand_stack_to(address bottom) NOINLINE; 665 666static void _expand_stack_to(address bottom) { 667 address sp; 668 size_t size; 669 volatile char *p; 670 671 // Adjust bottom to point to the largest address within the same page, it 672 // gives us a one-page buffer if alloca() allocates slightly more memory. 673 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 674 bottom += os::Linux::page_size() - 1; 675 676 // sp might be slightly above current stack pointer; if that's the case, we 677 // will alloca() a little more space than necessary, which is OK. Don't use 678 // os::current_stack_pointer(), as its result can be slightly below current 679 // stack pointer, causing us to not alloca enough to reach "bottom". 680 sp = (address)&sp; 681 682 if (sp > bottom) { 683 size = sp - bottom; 684 p = (volatile char *)alloca(size); 685 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 686 p[0] = '\0'; 687 } 688} 689 690bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 691 assert(t!=NULL, "just checking"); 692 assert(t->osthread()->expanding_stack(), "expand should be set"); 693 assert(t->stack_base() != NULL, "stack_base was not initialized"); 694 695 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 696 sigset_t mask_all, old_sigset; 697 sigfillset(&mask_all); 698 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 699 _expand_stack_to(addr); 700 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 701 return true; 702 } 703 return false; 704} 705 706////////////////////////////////////////////////////////////////////////////// 707// create new thread 708 709static address highest_vm_reserved_address(); 710 711// check if it's safe to start a new thread 712static bool _thread_safety_check(Thread* thread) { 713 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 714 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 715 // Heap is mmap'ed at lower end of memory space. Thread stacks are 716 // allocated (MAP_FIXED) from high address space. Every thread stack 717 // occupies a fixed size slot (usually 2Mbytes, but user can change 718 // it to other values if they rebuild LinuxThreads). 719 // 720 // Problem with MAP_FIXED is that mmap() can still succeed even part of 721 // the memory region has already been mmap'ed. That means if we have too 722 // many threads and/or very large heap, eventually thread stack will 723 // collide with heap. 724 // 725 // Here we try to prevent heap/stack collision by comparing current 726 // stack bottom with the highest address that has been mmap'ed by JVM 727 // plus a safety margin for memory maps created by native code. 728 // 729 // This feature can be disabled by setting ThreadSafetyMargin to 0 730 // 731 if (ThreadSafetyMargin > 0) { 732 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 733 734 // not safe if our stack extends below the safety margin 735 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 736 } else { 737 return true; 738 } 739 } else { 740 // Floating stack LinuxThreads or NPTL: 741 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 742 // there's not enough space left, pthread_create() will fail. If we come 743 // here, that means enough space has been reserved for stack. 744 return true; 745 } 746} 747 748// Thread start routine for all newly created threads 749static void *java_start(Thread *thread) { 750 // Try to randomize the cache line index of hot stack frames. 751 // This helps when threads of the same stack traces evict each other's 752 // cache lines. The threads can be either from the same JVM instance, or 753 // from different JVM instances. The benefit is especially true for 754 // processors with hyperthreading technology. 755 static int counter = 0; 756 int pid = os::current_process_id(); 757 alloca(((pid ^ counter++) & 7) * 128); 758 759 ThreadLocalStorage::set_thread(thread); 760 761 OSThread* osthread = thread->osthread(); 762 Monitor* sync = osthread->startThread_lock(); 763 764 // non floating stack LinuxThreads needs extra check, see above 765 if (!_thread_safety_check(thread)) { 766 // notify parent thread 767 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 768 osthread->set_state(ZOMBIE); 769 sync->notify_all(); 770 return NULL; 771 } 772 773 // thread_id is kernel thread id (similar to Solaris LWP id) 774 osthread->set_thread_id(os::Linux::gettid()); 775 776 if (UseNUMA) { 777 int lgrp_id = os::numa_get_group_id(); 778 if (lgrp_id != -1) { 779 thread->set_lgrp_id(lgrp_id); 780 } 781 } 782 // initialize signal mask for this thread 783 os::Linux::hotspot_sigmask(thread); 784 785 // initialize floating point control register 786 os::Linux::init_thread_fpu_state(); 787 788 // handshaking with parent thread 789 { 790 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 791 792 // notify parent thread 793 osthread->set_state(INITIALIZED); 794 sync->notify_all(); 795 796 // wait until os::start_thread() 797 while (osthread->get_state() == INITIALIZED) { 798 sync->wait(Mutex::_no_safepoint_check_flag); 799 } 800 } 801 802 // call one more level start routine 803 thread->run(); 804 805 return 0; 806} 807 808bool os::create_thread(Thread* thread, ThreadType thr_type, 809 size_t stack_size) { 810 assert(thread->osthread() == NULL, "caller responsible"); 811 812 // Allocate the OSThread object 813 OSThread* osthread = new OSThread(NULL, NULL); 814 if (osthread == NULL) { 815 return false; 816 } 817 818 // set the correct thread state 819 osthread->set_thread_type(thr_type); 820 821 // Initial state is ALLOCATED but not INITIALIZED 822 osthread->set_state(ALLOCATED); 823 824 thread->set_osthread(osthread); 825 826 // init thread attributes 827 pthread_attr_t attr; 828 pthread_attr_init(&attr); 829 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 830 831 // stack size 832 if (os::Linux::supports_variable_stack_size()) { 833 // calculate stack size if it's not specified by caller 834 if (stack_size == 0) { 835 stack_size = os::Linux::default_stack_size(thr_type); 836 837 switch (thr_type) { 838 case os::java_thread: 839 // Java threads use ThreadStackSize which default value can be 840 // changed with the flag -Xss 841 assert(JavaThread::stack_size_at_create() > 0, "this should be set"); 842 stack_size = JavaThread::stack_size_at_create(); 843 break; 844 case os::compiler_thread: 845 if (CompilerThreadStackSize > 0) { 846 stack_size = (size_t)(CompilerThreadStackSize * K); 847 break; 848 } // else fall through: 849 // use VMThreadStackSize if CompilerThreadStackSize is not defined 850 case os::vm_thread: 851 case os::pgc_thread: 852 case os::cgc_thread: 853 case os::watcher_thread: 854 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 855 break; 856 } 857 } 858 859 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 860 pthread_attr_setstacksize(&attr, stack_size); 861 } else { 862 // let pthread_create() pick the default value. 863 } 864 865 // glibc guard page 866 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 867 868 ThreadState state; 869 870 { 871 // Serialize thread creation if we are running with fixed stack LinuxThreads 872 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 873 if (lock) { 874 os::Linux::createThread_lock()->lock_without_safepoint_check(); 875 } 876 877 pthread_t tid; 878 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 879 880 pthread_attr_destroy(&attr); 881 882 if (ret != 0) { 883 if (PrintMiscellaneous && (Verbose || WizardMode)) { 884 perror("pthread_create()"); 885 } 886 // Need to clean up stuff we've allocated so far 887 thread->set_osthread(NULL); 888 delete osthread; 889 if (lock) os::Linux::createThread_lock()->unlock(); 890 return false; 891 } 892 893 // Store pthread info into the OSThread 894 osthread->set_pthread_id(tid); 895 896 // Wait until child thread is either initialized or aborted 897 { 898 Monitor* sync_with_child = osthread->startThread_lock(); 899 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 900 while ((state = osthread->get_state()) == ALLOCATED) { 901 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 902 } 903 } 904 905 if (lock) { 906 os::Linux::createThread_lock()->unlock(); 907 } 908 } 909 910 // Aborted due to thread limit being reached 911 if (state == ZOMBIE) { 912 thread->set_osthread(NULL); 913 delete osthread; 914 return false; 915 } 916 917 // The thread is returned suspended (in state INITIALIZED), 918 // and is started higher up in the call chain 919 assert(state == INITIALIZED, "race condition"); 920 return true; 921} 922 923///////////////////////////////////////////////////////////////////////////// 924// attach existing thread 925 926// bootstrap the main thread 927bool os::create_main_thread(JavaThread* thread) { 928 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 929 return create_attached_thread(thread); 930} 931 932bool os::create_attached_thread(JavaThread* thread) { 933#ifdef ASSERT 934 thread->verify_not_published(); 935#endif 936 937 // Allocate the OSThread object 938 OSThread* osthread = new OSThread(NULL, NULL); 939 940 if (osthread == NULL) { 941 return false; 942 } 943 944 // Store pthread info into the OSThread 945 osthread->set_thread_id(os::Linux::gettid()); 946 osthread->set_pthread_id(::pthread_self()); 947 948 // initialize floating point control register 949 os::Linux::init_thread_fpu_state(); 950 951 // Initial thread state is RUNNABLE 952 osthread->set_state(RUNNABLE); 953 954 thread->set_osthread(osthread); 955 956 if (UseNUMA) { 957 int lgrp_id = os::numa_get_group_id(); 958 if (lgrp_id != -1) { 959 thread->set_lgrp_id(lgrp_id); 960 } 961 } 962 963 if (os::Linux::is_initial_thread()) { 964 // If current thread is initial thread, its stack is mapped on demand, 965 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 966 // the entire stack region to avoid SEGV in stack banging. 967 // It is also useful to get around the heap-stack-gap problem on SuSE 968 // kernel (see 4821821 for details). We first expand stack to the top 969 // of yellow zone, then enable stack yellow zone (order is significant, 970 // enabling yellow zone first will crash JVM on SuSE Linux), so there 971 // is no gap between the last two virtual memory regions. 972 973 JavaThread *jt = (JavaThread *)thread; 974 address addr = jt->stack_yellow_zone_base(); 975 assert(addr != NULL, "initialization problem?"); 976 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 977 978 osthread->set_expanding_stack(); 979 os::Linux::manually_expand_stack(jt, addr); 980 osthread->clear_expanding_stack(); 981 } 982 983 // initialize signal mask for this thread 984 // and save the caller's signal mask 985 os::Linux::hotspot_sigmask(thread); 986 987 return true; 988} 989 990void os::pd_start_thread(Thread* thread) { 991 OSThread * osthread = thread->osthread(); 992 assert(osthread->get_state() != INITIALIZED, "just checking"); 993 Monitor* sync_with_child = osthread->startThread_lock(); 994 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 995 sync_with_child->notify(); 996} 997 998// Free Linux resources related to the OSThread 999void os::free_thread(OSThread* osthread) { 1000 assert(osthread != NULL, "osthread not set"); 1001 1002 if (Thread::current()->osthread() == osthread) { 1003 // Restore caller's signal mask 1004 sigset_t sigmask = osthread->caller_sigmask(); 1005 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1006 } 1007 1008 delete osthread; 1009} 1010 1011////////////////////////////////////////////////////////////////////////////// 1012// thread local storage 1013 1014// Restore the thread pointer if the destructor is called. This is in case 1015// someone from JNI code sets up a destructor with pthread_key_create to run 1016// detachCurrentThread on thread death. Unless we restore the thread pointer we 1017// will hang or crash. When detachCurrentThread is called the key will be set 1018// to null and we will not be called again. If detachCurrentThread is never 1019// called we could loop forever depending on the pthread implementation. 1020static void restore_thread_pointer(void* p) { 1021 Thread* thread = (Thread*) p; 1022 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread); 1023} 1024 1025int os::allocate_thread_local_storage() { 1026 pthread_key_t key; 1027 int rslt = pthread_key_create(&key, restore_thread_pointer); 1028 assert(rslt == 0, "cannot allocate thread local storage"); 1029 return (int)key; 1030} 1031 1032// Note: This is currently not used by VM, as we don't destroy TLS key 1033// on VM exit. 1034void os::free_thread_local_storage(int index) { 1035 int rslt = pthread_key_delete((pthread_key_t)index); 1036 assert(rslt == 0, "invalid index"); 1037} 1038 1039void os::thread_local_storage_at_put(int index, void* value) { 1040 int rslt = pthread_setspecific((pthread_key_t)index, value); 1041 assert(rslt == 0, "pthread_setspecific failed"); 1042} 1043 1044extern "C" Thread* get_thread() { 1045 return ThreadLocalStorage::thread(); 1046} 1047 1048////////////////////////////////////////////////////////////////////////////// 1049// initial thread 1050 1051// Check if current thread is the initial thread, similar to Solaris thr_main. 1052bool os::Linux::is_initial_thread(void) { 1053 char dummy; 1054 // If called before init complete, thread stack bottom will be null. 1055 // Can be called if fatal error occurs before initialization. 1056 if (initial_thread_stack_bottom() == NULL) return false; 1057 assert(initial_thread_stack_bottom() != NULL && 1058 initial_thread_stack_size() != 0, 1059 "os::init did not locate initial thread's stack region"); 1060 if ((address)&dummy >= initial_thread_stack_bottom() && 1061 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) { 1062 return true; 1063 } else { 1064 return false; 1065 } 1066} 1067 1068// Find the virtual memory area that contains addr 1069static bool find_vma(address addr, address* vma_low, address* vma_high) { 1070 FILE *fp = fopen("/proc/self/maps", "r"); 1071 if (fp) { 1072 address low, high; 1073 while (!feof(fp)) { 1074 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1075 if (low <= addr && addr < high) { 1076 if (vma_low) *vma_low = low; 1077 if (vma_high) *vma_high = high; 1078 fclose(fp); 1079 return true; 1080 } 1081 } 1082 for (;;) { 1083 int ch = fgetc(fp); 1084 if (ch == EOF || ch == (int)'\n') break; 1085 } 1086 } 1087 fclose(fp); 1088 } 1089 return false; 1090} 1091 1092// Locate initial thread stack. This special handling of initial thread stack 1093// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1094// bogus value for initial thread. 1095void os::Linux::capture_initial_stack(size_t max_size) { 1096 // stack size is the easy part, get it from RLIMIT_STACK 1097 size_t stack_size; 1098 struct rlimit rlim; 1099 getrlimit(RLIMIT_STACK, &rlim); 1100 stack_size = rlim.rlim_cur; 1101 1102 // 6308388: a bug in ld.so will relocate its own .data section to the 1103 // lower end of primordial stack; reduce ulimit -s value a little bit 1104 // so we won't install guard page on ld.so's data section. 1105 stack_size -= 2 * page_size(); 1106 1107 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1108 // 7.1, in both cases we will get 2G in return value. 1109 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1110 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1111 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1112 // in case other parts in glibc still assumes 2M max stack size. 1113 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1114 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1115 if (stack_size > 2 * K * K IA64_ONLY(*2)) { 1116 stack_size = 2 * K * K IA64_ONLY(*2); 1117 } 1118 // Try to figure out where the stack base (top) is. This is harder. 1119 // 1120 // When an application is started, glibc saves the initial stack pointer in 1121 // a global variable "__libc_stack_end", which is then used by system 1122 // libraries. __libc_stack_end should be pretty close to stack top. The 1123 // variable is available since the very early days. However, because it is 1124 // a private interface, it could disappear in the future. 1125 // 1126 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1127 // to __libc_stack_end, it is very close to stack top, but isn't the real 1128 // stack top. Note that /proc may not exist if VM is running as a chroot 1129 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1130 // /proc/<pid>/stat could change in the future (though unlikely). 1131 // 1132 // We try __libc_stack_end first. If that doesn't work, look for 1133 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1134 // as a hint, which should work well in most cases. 1135 1136 uintptr_t stack_start; 1137 1138 // try __libc_stack_end first 1139 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1140 if (p && *p) { 1141 stack_start = *p; 1142 } else { 1143 // see if we can get the start_stack field from /proc/self/stat 1144 FILE *fp; 1145 int pid; 1146 char state; 1147 int ppid; 1148 int pgrp; 1149 int session; 1150 int nr; 1151 int tpgrp; 1152 unsigned long flags; 1153 unsigned long minflt; 1154 unsigned long cminflt; 1155 unsigned long majflt; 1156 unsigned long cmajflt; 1157 unsigned long utime; 1158 unsigned long stime; 1159 long cutime; 1160 long cstime; 1161 long prio; 1162 long nice; 1163 long junk; 1164 long it_real; 1165 uintptr_t start; 1166 uintptr_t vsize; 1167 intptr_t rss; 1168 uintptr_t rsslim; 1169 uintptr_t scodes; 1170 uintptr_t ecode; 1171 int i; 1172 1173 // Figure what the primordial thread stack base is. Code is inspired 1174 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1175 // followed by command name surrounded by parentheses, state, etc. 1176 char stat[2048]; 1177 int statlen; 1178 1179 fp = fopen("/proc/self/stat", "r"); 1180 if (fp) { 1181 statlen = fread(stat, 1, 2047, fp); 1182 stat[statlen] = '\0'; 1183 fclose(fp); 1184 1185 // Skip pid and the command string. Note that we could be dealing with 1186 // weird command names, e.g. user could decide to rename java launcher 1187 // to "java 1.4.2 :)", then the stat file would look like 1188 // 1234 (java 1.4.2 :)) R ... ... 1189 // We don't really need to know the command string, just find the last 1190 // occurrence of ")" and then start parsing from there. See bug 4726580. 1191 char * s = strrchr(stat, ')'); 1192 1193 i = 0; 1194 if (s) { 1195 // Skip blank chars 1196 do { s++; } while (s && isspace(*s)); 1197 1198#define _UFM UINTX_FORMAT 1199#define _DFM INTX_FORMAT 1200 1201 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1202 // 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 1203 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, 1204 &state, // 3 %c 1205 &ppid, // 4 %d 1206 &pgrp, // 5 %d 1207 &session, // 6 %d 1208 &nr, // 7 %d 1209 &tpgrp, // 8 %d 1210 &flags, // 9 %lu 1211 &minflt, // 10 %lu 1212 &cminflt, // 11 %lu 1213 &majflt, // 12 %lu 1214 &cmajflt, // 13 %lu 1215 &utime, // 14 %lu 1216 &stime, // 15 %lu 1217 &cutime, // 16 %ld 1218 &cstime, // 17 %ld 1219 &prio, // 18 %ld 1220 &nice, // 19 %ld 1221 &junk, // 20 %ld 1222 &it_real, // 21 %ld 1223 &start, // 22 UINTX_FORMAT 1224 &vsize, // 23 UINTX_FORMAT 1225 &rss, // 24 INTX_FORMAT 1226 &rsslim, // 25 UINTX_FORMAT 1227 &scodes, // 26 UINTX_FORMAT 1228 &ecode, // 27 UINTX_FORMAT 1229 &stack_start); // 28 UINTX_FORMAT 1230 } 1231 1232#undef _UFM 1233#undef _DFM 1234 1235 if (i != 28 - 2) { 1236 assert(false, "Bad conversion from /proc/self/stat"); 1237 // product mode - assume we are the initial thread, good luck in the 1238 // embedded case. 1239 warning("Can't detect initial thread stack location - bad conversion"); 1240 stack_start = (uintptr_t) &rlim; 1241 } 1242 } else { 1243 // For some reason we can't open /proc/self/stat (for example, running on 1244 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1245 // most cases, so don't abort: 1246 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1247 stack_start = (uintptr_t) &rlim; 1248 } 1249 } 1250 1251 // Now we have a pointer (stack_start) very close to the stack top, the 1252 // next thing to do is to figure out the exact location of stack top. We 1253 // can find out the virtual memory area that contains stack_start by 1254 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1255 // and its upper limit is the real stack top. (again, this would fail if 1256 // running inside chroot, because /proc may not exist.) 1257 1258 uintptr_t stack_top; 1259 address low, high; 1260 if (find_vma((address)stack_start, &low, &high)) { 1261 // success, "high" is the true stack top. (ignore "low", because initial 1262 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1263 stack_top = (uintptr_t)high; 1264 } else { 1265 // failed, likely because /proc/self/maps does not exist 1266 warning("Can't detect initial thread stack location - find_vma failed"); 1267 // best effort: stack_start is normally within a few pages below the real 1268 // stack top, use it as stack top, and reduce stack size so we won't put 1269 // guard page outside stack. 1270 stack_top = stack_start; 1271 stack_size -= 16 * page_size(); 1272 } 1273 1274 // stack_top could be partially down the page so align it 1275 stack_top = align_size_up(stack_top, page_size()); 1276 1277 if (max_size && stack_size > max_size) { 1278 _initial_thread_stack_size = max_size; 1279 } else { 1280 _initial_thread_stack_size = stack_size; 1281 } 1282 1283 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1284 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1285} 1286 1287//////////////////////////////////////////////////////////////////////////////// 1288// time support 1289 1290// Time since start-up in seconds to a fine granularity. 1291// Used by VMSelfDestructTimer and the MemProfiler. 1292double os::elapsedTime() { 1293 1294 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1295} 1296 1297jlong os::elapsed_counter() { 1298 return javaTimeNanos() - initial_time_count; 1299} 1300 1301jlong os::elapsed_frequency() { 1302 return NANOSECS_PER_SEC; // nanosecond resolution 1303} 1304 1305bool os::supports_vtime() { return true; } 1306bool os::enable_vtime() { return false; } 1307bool os::vtime_enabled() { return false; } 1308 1309double os::elapsedVTime() { 1310 struct rusage usage; 1311 int retval = getrusage(RUSAGE_THREAD, &usage); 1312 if (retval == 0) { 1313 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); 1314 } else { 1315 // better than nothing, but not much 1316 return elapsedTime(); 1317 } 1318} 1319 1320jlong os::javaTimeMillis() { 1321 timeval time; 1322 int status = gettimeofday(&time, NULL); 1323 assert(status != -1, "linux error"); 1324 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1325} 1326 1327#ifndef CLOCK_MONOTONIC 1328 #define CLOCK_MONOTONIC (1) 1329#endif 1330 1331void os::Linux::clock_init() { 1332 // we do dlopen's in this particular order due to bug in linux 1333 // dynamical loader (see 6348968) leading to crash on exit 1334 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1335 if (handle == NULL) { 1336 handle = dlopen("librt.so", RTLD_LAZY); 1337 } 1338 1339 if (handle) { 1340 int (*clock_getres_func)(clockid_t, struct timespec*) = 1341 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1342 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1343 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1344 if (clock_getres_func && clock_gettime_func) { 1345 // See if monotonic clock is supported by the kernel. Note that some 1346 // early implementations simply return kernel jiffies (updated every 1347 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1348 // for nano time (though the monotonic property is still nice to have). 1349 // It's fixed in newer kernels, however clock_getres() still returns 1350 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1351 // resolution for now. Hopefully as people move to new kernels, this 1352 // won't be a problem. 1353 struct timespec res; 1354 struct timespec tp; 1355 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1356 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1357 // yes, monotonic clock is supported 1358 _clock_gettime = clock_gettime_func; 1359 return; 1360 } else { 1361 // close librt if there is no monotonic clock 1362 dlclose(handle); 1363 } 1364 } 1365 } 1366 warning("No monotonic clock was available - timed services may " \ 1367 "be adversely affected if the time-of-day clock changes"); 1368} 1369 1370#ifndef SYS_clock_getres 1371 #if defined(IA32) || defined(AMD64) 1372 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1373 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1374 #else 1375 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1376 #define sys_clock_getres(x,y) -1 1377 #endif 1378#else 1379 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1380#endif 1381 1382void os::Linux::fast_thread_clock_init() { 1383 if (!UseLinuxPosixThreadCPUClocks) { 1384 return; 1385 } 1386 clockid_t clockid; 1387 struct timespec tp; 1388 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1389 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1390 1391 // Switch to using fast clocks for thread cpu time if 1392 // the sys_clock_getres() returns 0 error code. 1393 // Note, that some kernels may support the current thread 1394 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1395 // returned by the pthread_getcpuclockid(). 1396 // If the fast Posix clocks are supported then the sys_clock_getres() 1397 // must return at least tp.tv_sec == 0 which means a resolution 1398 // better than 1 sec. This is extra check for reliability. 1399 1400 if (pthread_getcpuclockid_func && 1401 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1402 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1403 _supports_fast_thread_cpu_time = true; 1404 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1405 } 1406} 1407 1408jlong os::javaTimeNanos() { 1409 if (os::supports_monotonic_clock()) { 1410 struct timespec tp; 1411 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1412 assert(status == 0, "gettime error"); 1413 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1414 return result; 1415 } else { 1416 timeval time; 1417 int status = gettimeofday(&time, NULL); 1418 assert(status != -1, "linux error"); 1419 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1420 return 1000 * usecs; 1421 } 1422} 1423 1424void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1425 if (os::supports_monotonic_clock()) { 1426 info_ptr->max_value = ALL_64_BITS; 1427 1428 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1429 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1430 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1431 } else { 1432 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1433 info_ptr->max_value = ALL_64_BITS; 1434 1435 // gettimeofday is a real time clock so it skips 1436 info_ptr->may_skip_backward = true; 1437 info_ptr->may_skip_forward = true; 1438 } 1439 1440 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1441} 1442 1443// Return the real, user, and system times in seconds from an 1444// arbitrary fixed point in the past. 1445bool os::getTimesSecs(double* process_real_time, 1446 double* process_user_time, 1447 double* process_system_time) { 1448 struct tms ticks; 1449 clock_t real_ticks = times(&ticks); 1450 1451 if (real_ticks == (clock_t) (-1)) { 1452 return false; 1453 } else { 1454 double ticks_per_second = (double) clock_tics_per_sec; 1455 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1456 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1457 *process_real_time = ((double) real_ticks) / ticks_per_second; 1458 1459 return true; 1460 } 1461} 1462 1463 1464char * os::local_time_string(char *buf, size_t buflen) { 1465 struct tm t; 1466 time_t long_time; 1467 time(&long_time); 1468 localtime_r(&long_time, &t); 1469 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1470 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1471 t.tm_hour, t.tm_min, t.tm_sec); 1472 return buf; 1473} 1474 1475struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1476 return localtime_r(clock, res); 1477} 1478 1479//////////////////////////////////////////////////////////////////////////////// 1480// runtime exit support 1481 1482// Note: os::shutdown() might be called very early during initialization, or 1483// called from signal handler. Before adding something to os::shutdown(), make 1484// sure it is async-safe and can handle partially initialized VM. 1485void os::shutdown() { 1486 1487 // allow PerfMemory to attempt cleanup of any persistent resources 1488 perfMemory_exit(); 1489 1490 // needs to remove object in file system 1491 AttachListener::abort(); 1492 1493 // flush buffered output, finish log files 1494 ostream_abort(); 1495 1496 // Check for abort hook 1497 abort_hook_t abort_hook = Arguments::abort_hook(); 1498 if (abort_hook != NULL) { 1499 abort_hook(); 1500 } 1501 1502} 1503 1504// Note: os::abort() might be called very early during initialization, or 1505// called from signal handler. Before adding something to os::abort(), make 1506// sure it is async-safe and can handle partially initialized VM. 1507void os::abort(bool dump_core) { 1508 os::shutdown(); 1509 if (dump_core) { 1510#ifndef PRODUCT 1511 fdStream out(defaultStream::output_fd()); 1512 out.print_raw("Current thread is "); 1513 char buf[16]; 1514 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1515 out.print_raw_cr(buf); 1516 out.print_raw_cr("Dumping core ..."); 1517#endif 1518 ::abort(); // dump core 1519 } 1520 1521 ::exit(1); 1522} 1523 1524// Die immediately, no exit hook, no abort hook, no cleanup. 1525void os::die() { 1526 // _exit() on LinuxThreads only kills current thread 1527 ::abort(); 1528} 1529 1530 1531// This method is a copy of JDK's sysGetLastErrorString 1532// from src/solaris/hpi/src/system_md.c 1533 1534size_t os::lasterror(char *buf, size_t len) { 1535 if (errno == 0) return 0; 1536 1537 const char *s = ::strerror(errno); 1538 size_t n = ::strlen(s); 1539 if (n >= len) { 1540 n = len - 1; 1541 } 1542 ::strncpy(buf, s, n); 1543 buf[n] = '\0'; 1544 return n; 1545} 1546 1547intx os::current_thread_id() { return (intx)pthread_self(); } 1548int os::current_process_id() { 1549 1550 // Under the old linux thread library, linux gives each thread 1551 // its own process id. Because of this each thread will return 1552 // a different pid if this method were to return the result 1553 // of getpid(2). Linux provides no api that returns the pid 1554 // of the launcher thread for the vm. This implementation 1555 // returns a unique pid, the pid of the launcher thread 1556 // that starts the vm 'process'. 1557 1558 // Under the NPTL, getpid() returns the same pid as the 1559 // launcher thread rather than a unique pid per thread. 1560 // Use gettid() if you want the old pre NPTL behaviour. 1561 1562 // if you are looking for the result of a call to getpid() that 1563 // returns a unique pid for the calling thread, then look at the 1564 // OSThread::thread_id() method in osThread_linux.hpp file 1565 1566 return (int)(_initial_pid ? _initial_pid : getpid()); 1567} 1568 1569// DLL functions 1570 1571const char* os::dll_file_extension() { return ".so"; } 1572 1573// This must be hard coded because it's the system's temporary 1574// directory not the java application's temp directory, ala java.io.tmpdir. 1575const char* os::get_temp_directory() { return "/tmp"; } 1576 1577static bool file_exists(const char* filename) { 1578 struct stat statbuf; 1579 if (filename == NULL || strlen(filename) == 0) { 1580 return false; 1581 } 1582 return os::stat(filename, &statbuf) == 0; 1583} 1584 1585bool os::dll_build_name(char* buffer, size_t buflen, 1586 const char* pname, const char* fname) { 1587 bool retval = false; 1588 // Copied from libhpi 1589 const size_t pnamelen = pname ? strlen(pname) : 0; 1590 1591 // Return error on buffer overflow. 1592 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1593 return retval; 1594 } 1595 1596 if (pnamelen == 0) { 1597 snprintf(buffer, buflen, "lib%s.so", fname); 1598 retval = true; 1599 } else if (strchr(pname, *os::path_separator()) != NULL) { 1600 int n; 1601 char** pelements = split_path(pname, &n); 1602 if (pelements == NULL) { 1603 return false; 1604 } 1605 for (int i = 0; i < n; i++) { 1606 // Really shouldn't be NULL, but check can't hurt 1607 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1608 continue; // skip the empty path values 1609 } 1610 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1611 if (file_exists(buffer)) { 1612 retval = true; 1613 break; 1614 } 1615 } 1616 // release the storage 1617 for (int i = 0; i < n; i++) { 1618 if (pelements[i] != NULL) { 1619 FREE_C_HEAP_ARRAY(char, pelements[i]); 1620 } 1621 } 1622 if (pelements != NULL) { 1623 FREE_C_HEAP_ARRAY(char*, pelements); 1624 } 1625 } else { 1626 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1627 retval = true; 1628 } 1629 return retval; 1630} 1631 1632// check if addr is inside libjvm.so 1633bool os::address_is_in_vm(address addr) { 1634 static address libjvm_base_addr; 1635 Dl_info dlinfo; 1636 1637 if (libjvm_base_addr == NULL) { 1638 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1639 libjvm_base_addr = (address)dlinfo.dli_fbase; 1640 } 1641 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1642 } 1643 1644 if (dladdr((void *)addr, &dlinfo) != 0) { 1645 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1646 } 1647 1648 return false; 1649} 1650 1651bool os::dll_address_to_function_name(address addr, char *buf, 1652 int buflen, int *offset) { 1653 // buf is not optional, but offset is optional 1654 assert(buf != NULL, "sanity check"); 1655 1656 Dl_info dlinfo; 1657 1658 if (dladdr((void*)addr, &dlinfo) != 0) { 1659 // see if we have a matching symbol 1660 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1661 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1662 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1663 } 1664 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1665 return true; 1666 } 1667 // no matching symbol so try for just file info 1668 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1669 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1670 buf, buflen, offset, dlinfo.dli_fname)) { 1671 return true; 1672 } 1673 } 1674 } 1675 1676 buf[0] = '\0'; 1677 if (offset != NULL) *offset = -1; 1678 return false; 1679} 1680 1681struct _address_to_library_name { 1682 address addr; // input : memory address 1683 size_t buflen; // size of fname 1684 char* fname; // output: library name 1685 address base; // library base addr 1686}; 1687 1688static int address_to_library_name_callback(struct dl_phdr_info *info, 1689 size_t size, void *data) { 1690 int i; 1691 bool found = false; 1692 address libbase = NULL; 1693 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1694 1695 // iterate through all loadable segments 1696 for (i = 0; i < info->dlpi_phnum; i++) { 1697 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1698 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1699 // base address of a library is the lowest address of its loaded 1700 // segments. 1701 if (libbase == NULL || libbase > segbase) { 1702 libbase = segbase; 1703 } 1704 // see if 'addr' is within current segment 1705 if (segbase <= d->addr && 1706 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1707 found = true; 1708 } 1709 } 1710 } 1711 1712 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1713 // so dll_address_to_library_name() can fall through to use dladdr() which 1714 // can figure out executable name from argv[0]. 1715 if (found && info->dlpi_name && info->dlpi_name[0]) { 1716 d->base = libbase; 1717 if (d->fname) { 1718 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1719 } 1720 return 1; 1721 } 1722 return 0; 1723} 1724 1725bool os::dll_address_to_library_name(address addr, char* buf, 1726 int buflen, int* offset) { 1727 // buf is not optional, but offset is optional 1728 assert(buf != NULL, "sanity check"); 1729 1730 Dl_info dlinfo; 1731 struct _address_to_library_name data; 1732 1733 // There is a bug in old glibc dladdr() implementation that it could resolve 1734 // to wrong library name if the .so file has a base address != NULL. Here 1735 // we iterate through the program headers of all loaded libraries to find 1736 // out which library 'addr' really belongs to. This workaround can be 1737 // removed once the minimum requirement for glibc is moved to 2.3.x. 1738 data.addr = addr; 1739 data.fname = buf; 1740 data.buflen = buflen; 1741 data.base = NULL; 1742 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1743 1744 if (rslt) { 1745 // buf already contains library name 1746 if (offset) *offset = addr - data.base; 1747 return true; 1748 } 1749 if (dladdr((void*)addr, &dlinfo) != 0) { 1750 if (dlinfo.dli_fname != NULL) { 1751 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1752 } 1753 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1754 *offset = addr - (address)dlinfo.dli_fbase; 1755 } 1756 return true; 1757 } 1758 1759 buf[0] = '\0'; 1760 if (offset) *offset = -1; 1761 return false; 1762} 1763 1764// Loads .dll/.so and 1765// in case of error it checks if .dll/.so was built for the 1766// same architecture as Hotspot is running on 1767 1768 1769// Remember the stack's state. The Linux dynamic linker will change 1770// the stack to 'executable' at most once, so we must safepoint only once. 1771bool os::Linux::_stack_is_executable = false; 1772 1773// VM operation that loads a library. This is necessary if stack protection 1774// of the Java stacks can be lost during loading the library. If we 1775// do not stop the Java threads, they can stack overflow before the stacks 1776// are protected again. 1777class VM_LinuxDllLoad: public VM_Operation { 1778 private: 1779 const char *_filename; 1780 char *_ebuf; 1781 int _ebuflen; 1782 void *_lib; 1783 public: 1784 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1785 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1786 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1787 void doit() { 1788 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1789 os::Linux::_stack_is_executable = true; 1790 } 1791 void* loaded_library() { return _lib; } 1792}; 1793 1794void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1795 void * result = NULL; 1796 bool load_attempted = false; 1797 1798 // Check whether the library to load might change execution rights 1799 // of the stack. If they are changed, the protection of the stack 1800 // guard pages will be lost. We need a safepoint to fix this. 1801 // 1802 // See Linux man page execstack(8) for more info. 1803 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1804 ElfFile ef(filename); 1805 if (!ef.specifies_noexecstack()) { 1806 if (!is_init_completed()) { 1807 os::Linux::_stack_is_executable = true; 1808 // This is OK - No Java threads have been created yet, and hence no 1809 // stack guard pages to fix. 1810 // 1811 // This should happen only when you are building JDK7 using a very 1812 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1813 // 1814 // Dynamic loader will make all stacks executable after 1815 // this function returns, and will not do that again. 1816 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1817 } else { 1818 warning("You have loaded library %s which might have disabled stack guard. " 1819 "The VM will try to fix the stack guard now.\n" 1820 "It's highly recommended that you fix the library with " 1821 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1822 filename); 1823 1824 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1825 JavaThread *jt = JavaThread::current(); 1826 if (jt->thread_state() != _thread_in_native) { 1827 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1828 // that requires ExecStack. Cannot enter safe point. Let's give up. 1829 warning("Unable to fix stack guard. Giving up."); 1830 } else { 1831 if (!LoadExecStackDllInVMThread) { 1832 // This is for the case where the DLL has an static 1833 // constructor function that executes JNI code. We cannot 1834 // load such DLLs in the VMThread. 1835 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1836 } 1837 1838 ThreadInVMfromNative tiv(jt); 1839 debug_only(VMNativeEntryWrapper vew;) 1840 1841 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1842 VMThread::execute(&op); 1843 if (LoadExecStackDllInVMThread) { 1844 result = op.loaded_library(); 1845 } 1846 load_attempted = true; 1847 } 1848 } 1849 } 1850 } 1851 1852 if (!load_attempted) { 1853 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1854 } 1855 1856 if (result != NULL) { 1857 // Successful loading 1858 return result; 1859 } 1860 1861 Elf32_Ehdr elf_head; 1862 int diag_msg_max_length=ebuflen-strlen(ebuf); 1863 char* diag_msg_buf=ebuf+strlen(ebuf); 1864 1865 if (diag_msg_max_length==0) { 1866 // No more space in ebuf for additional diagnostics message 1867 return NULL; 1868 } 1869 1870 1871 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1872 1873 if (file_descriptor < 0) { 1874 // Can't open library, report dlerror() message 1875 return NULL; 1876 } 1877 1878 bool failed_to_read_elf_head= 1879 (sizeof(elf_head)!= 1880 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1881 1882 ::close(file_descriptor); 1883 if (failed_to_read_elf_head) { 1884 // file i/o error - report dlerror() msg 1885 return NULL; 1886 } 1887 1888 typedef struct { 1889 Elf32_Half code; // Actual value as defined in elf.h 1890 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1891 char elf_class; // 32 or 64 bit 1892 char endianess; // MSB or LSB 1893 char* name; // String representation 1894 } arch_t; 1895 1896#ifndef EM_486 1897 #define EM_486 6 /* Intel 80486 */ 1898#endif 1899#ifndef EM_AARCH64 1900 #define EM_AARCH64 183 /* ARM AARCH64 */ 1901#endif 1902 1903 static const arch_t arch_array[]={ 1904 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1905 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1906 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1907 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1908 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1909 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1910 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1911 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1912#if defined(VM_LITTLE_ENDIAN) 1913 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"}, 1914#else 1915 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1916#endif 1917 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1918 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1919 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1920 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1921 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1922 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1923 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1924 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1925 }; 1926 1927#if (defined IA32) 1928 static Elf32_Half running_arch_code=EM_386; 1929#elif (defined AMD64) 1930 static Elf32_Half running_arch_code=EM_X86_64; 1931#elif (defined IA64) 1932 static Elf32_Half running_arch_code=EM_IA_64; 1933#elif (defined __sparc) && (defined _LP64) 1934 static Elf32_Half running_arch_code=EM_SPARCV9; 1935#elif (defined __sparc) && (!defined _LP64) 1936 static Elf32_Half running_arch_code=EM_SPARC; 1937#elif (defined __powerpc64__) 1938 static Elf32_Half running_arch_code=EM_PPC64; 1939#elif (defined __powerpc__) 1940 static Elf32_Half running_arch_code=EM_PPC; 1941#elif (defined ARM) 1942 static Elf32_Half running_arch_code=EM_ARM; 1943#elif (defined S390) 1944 static Elf32_Half running_arch_code=EM_S390; 1945#elif (defined ALPHA) 1946 static Elf32_Half running_arch_code=EM_ALPHA; 1947#elif (defined MIPSEL) 1948 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1949#elif (defined PARISC) 1950 static Elf32_Half running_arch_code=EM_PARISC; 1951#elif (defined MIPS) 1952 static Elf32_Half running_arch_code=EM_MIPS; 1953#elif (defined M68K) 1954 static Elf32_Half running_arch_code=EM_68K; 1955#elif (defined AARCH64) 1956 static Elf32_Half running_arch_code=EM_AARCH64; 1957#else 1958 #error Method os::dll_load requires that one of following is defined:\ 1959 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64 1960#endif 1961 1962 // Identify compatability class for VM's architecture and library's architecture 1963 // Obtain string descriptions for architectures 1964 1965 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1966 int running_arch_index=-1; 1967 1968 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1969 if (running_arch_code == arch_array[i].code) { 1970 running_arch_index = i; 1971 } 1972 if (lib_arch.code == arch_array[i].code) { 1973 lib_arch.compat_class = arch_array[i].compat_class; 1974 lib_arch.name = arch_array[i].name; 1975 } 1976 } 1977 1978 assert(running_arch_index != -1, 1979 "Didn't find running architecture code (running_arch_code) in arch_array"); 1980 if (running_arch_index == -1) { 1981 // Even though running architecture detection failed 1982 // we may still continue with reporting dlerror() message 1983 return NULL; 1984 } 1985 1986 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1987 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1988 return NULL; 1989 } 1990 1991#ifndef S390 1992 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1993 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1994 return NULL; 1995 } 1996#endif // !S390 1997 1998 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1999 if (lib_arch.name!=NULL) { 2000 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2001 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2002 lib_arch.name, arch_array[running_arch_index].name); 2003 } else { 2004 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2005 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2006 lib_arch.code, 2007 arch_array[running_arch_index].name); 2008 } 2009 } 2010 2011 return NULL; 2012} 2013 2014void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 2015 int ebuflen) { 2016 void * result = ::dlopen(filename, RTLD_LAZY); 2017 if (result == NULL) { 2018 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2019 ebuf[ebuflen-1] = '\0'; 2020 } 2021 return result; 2022} 2023 2024void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 2025 int ebuflen) { 2026 void * result = NULL; 2027 if (LoadExecStackDllInVMThread) { 2028 result = dlopen_helper(filename, ebuf, ebuflen); 2029 } 2030 2031 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2032 // library that requires an executable stack, or which does not have this 2033 // stack attribute set, dlopen changes the stack attribute to executable. The 2034 // read protection of the guard pages gets lost. 2035 // 2036 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2037 // may have been queued at the same time. 2038 2039 if (!_stack_is_executable) { 2040 JavaThread *jt = Threads::first(); 2041 2042 while (jt) { 2043 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2044 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2045 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2046 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2047 warning("Attempt to reguard stack yellow zone failed."); 2048 } 2049 } 2050 jt = jt->next(); 2051 } 2052 } 2053 2054 return result; 2055} 2056 2057// glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2058// chances are you might want to run the generated bits against glibc-2.0 2059// libdl.so, so always use locking for any version of glibc. 2060// 2061void* os::dll_lookup(void* handle, const char* name) { 2062 pthread_mutex_lock(&dl_mutex); 2063 void* res = dlsym(handle, name); 2064 pthread_mutex_unlock(&dl_mutex); 2065 return res; 2066} 2067 2068void* os::get_default_process_handle() { 2069 return (void*)::dlopen(NULL, RTLD_LAZY); 2070} 2071 2072static bool _print_ascii_file(const char* filename, outputStream* st) { 2073 int fd = ::open(filename, O_RDONLY); 2074 if (fd == -1) { 2075 return false; 2076 } 2077 2078 char buf[32]; 2079 int bytes; 2080 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2081 st->print_raw(buf, bytes); 2082 } 2083 2084 ::close(fd); 2085 2086 return true; 2087} 2088 2089void os::print_dll_info(outputStream *st) { 2090 st->print_cr("Dynamic libraries:"); 2091 2092 char fname[32]; 2093 pid_t pid = os::Linux::gettid(); 2094 2095 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2096 2097 if (!_print_ascii_file(fname, st)) { 2098 st->print("Can not get library information for pid = %d\n", pid); 2099 } 2100} 2101 2102int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 2103 FILE *procmapsFile = NULL; 2104 2105 // Open the procfs maps file for the current process 2106 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 2107 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 2108 char line[PATH_MAX + 100]; 2109 2110 // Read line by line from 'file' 2111 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 2112 u8 base, top, offset, inode; 2113 char permissions[5]; 2114 char device[6]; 2115 char name[PATH_MAX + 1]; 2116 2117 // Parse fields from line 2118 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name); 2119 2120 // Filter by device id '00:00' so that we only get file system mapped files. 2121 if (strcmp(device, "00:00") != 0) { 2122 2123 // Call callback with the fields of interest 2124 if(callback(name, (address)base, (address)top, param)) { 2125 // Oops abort, callback aborted 2126 fclose(procmapsFile); 2127 return 1; 2128 } 2129 } 2130 } 2131 fclose(procmapsFile); 2132 } 2133 return 0; 2134} 2135 2136void os::print_os_info_brief(outputStream* st) { 2137 os::Linux::print_distro_info(st); 2138 2139 os::Posix::print_uname_info(st); 2140 2141 os::Linux::print_libversion_info(st); 2142 2143} 2144 2145void os::print_os_info(outputStream* st) { 2146 st->print("OS:"); 2147 2148 os::Linux::print_distro_info(st); 2149 2150 os::Posix::print_uname_info(st); 2151 2152 // Print warning if unsafe chroot environment detected 2153 if (unsafe_chroot_detected) { 2154 st->print("WARNING!! "); 2155 st->print_cr("%s", unstable_chroot_error); 2156 } 2157 2158 os::Linux::print_libversion_info(st); 2159 2160 os::Posix::print_rlimit_info(st); 2161 2162 os::Posix::print_load_average(st); 2163 2164 os::Linux::print_full_memory_info(st); 2165} 2166 2167// Try to identify popular distros. 2168// Most Linux distributions have a /etc/XXX-release file, which contains 2169// the OS version string. Newer Linux distributions have a /etc/lsb-release 2170// file that also contains the OS version string. Some have more than one 2171// /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2172// /etc/redhat-release.), so the order is important. 2173// Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2174// their own specific XXX-release file as well as a redhat-release file. 2175// Because of this the XXX-release file needs to be searched for before the 2176// redhat-release file. 2177// Since Red Hat has a lsb-release file that is not very descriptive the 2178// search for redhat-release needs to be before lsb-release. 2179// Since the lsb-release file is the new standard it needs to be searched 2180// before the older style release files. 2181// Searching system-release (Red Hat) and os-release (other Linuxes) are a 2182// next to last resort. The os-release file is a new standard that contains 2183// distribution information and the system-release file seems to be an old 2184// standard that has been replaced by the lsb-release and os-release files. 2185// Searching for the debian_version file is the last resort. It contains 2186// an informative string like "6.0.6" or "wheezy/sid". Because of this 2187// "Debian " is printed before the contents of the debian_version file. 2188void os::Linux::print_distro_info(outputStream* st) { 2189 if (!_print_ascii_file("/etc/oracle-release", st) && 2190 !_print_ascii_file("/etc/mandriva-release", st) && 2191 !_print_ascii_file("/etc/mandrake-release", st) && 2192 !_print_ascii_file("/etc/sun-release", st) && 2193 !_print_ascii_file("/etc/redhat-release", st) && 2194 !_print_ascii_file("/etc/lsb-release", st) && 2195 !_print_ascii_file("/etc/SuSE-release", st) && 2196 !_print_ascii_file("/etc/turbolinux-release", st) && 2197 !_print_ascii_file("/etc/gentoo-release", st) && 2198 !_print_ascii_file("/etc/ltib-release", st) && 2199 !_print_ascii_file("/etc/angstrom-version", st) && 2200 !_print_ascii_file("/etc/system-release", st) && 2201 !_print_ascii_file("/etc/os-release", st)) { 2202 2203 if (file_exists("/etc/debian_version")) { 2204 st->print("Debian "); 2205 _print_ascii_file("/etc/debian_version", st); 2206 } else { 2207 st->print("Linux"); 2208 } 2209 } 2210 st->cr(); 2211} 2212 2213void os::Linux::print_libversion_info(outputStream* st) { 2214 // libc, pthread 2215 st->print("libc:"); 2216 st->print("%s ", os::Linux::glibc_version()); 2217 st->print("%s ", os::Linux::libpthread_version()); 2218 if (os::Linux::is_LinuxThreads()) { 2219 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2220 } 2221 st->cr(); 2222} 2223 2224void os::Linux::print_full_memory_info(outputStream* st) { 2225 st->print("\n/proc/meminfo:\n"); 2226 _print_ascii_file("/proc/meminfo", st); 2227 st->cr(); 2228} 2229 2230void os::print_memory_info(outputStream* st) { 2231 2232 st->print("Memory:"); 2233 st->print(" %dk page", os::vm_page_size()>>10); 2234 2235 // values in struct sysinfo are "unsigned long" 2236 struct sysinfo si; 2237 sysinfo(&si); 2238 2239 st->print(", physical " UINT64_FORMAT "k", 2240 os::physical_memory() >> 10); 2241 st->print("(" UINT64_FORMAT "k free)", 2242 os::available_memory() >> 10); 2243 st->print(", swap " UINT64_FORMAT "k", 2244 ((jlong)si.totalswap * si.mem_unit) >> 10); 2245 st->print("(" UINT64_FORMAT "k free)", 2246 ((jlong)si.freeswap * si.mem_unit) >> 10); 2247 st->cr(); 2248} 2249 2250void os::pd_print_cpu_info(outputStream* st) { 2251 st->print("\n/proc/cpuinfo:\n"); 2252 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2253 st->print(" <Not Available>"); 2254 } 2255 st->cr(); 2256} 2257 2258void os::print_siginfo(outputStream* st, void* siginfo) { 2259 const siginfo_t* si = (const siginfo_t*)siginfo; 2260 2261 os::Posix::print_siginfo_brief(st, si); 2262#if INCLUDE_CDS 2263 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2264 UseSharedSpaces) { 2265 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2266 if (mapinfo->is_in_shared_space(si->si_addr)) { 2267 st->print("\n\nError accessing class data sharing archive." \ 2268 " Mapped file inaccessible during execution, " \ 2269 " possible disk/network problem."); 2270 } 2271 } 2272#endif 2273 st->cr(); 2274} 2275 2276 2277static void print_signal_handler(outputStream* st, int sig, 2278 char* buf, size_t buflen); 2279 2280void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2281 st->print_cr("Signal Handlers:"); 2282 print_signal_handler(st, SIGSEGV, buf, buflen); 2283 print_signal_handler(st, SIGBUS , buf, buflen); 2284 print_signal_handler(st, SIGFPE , buf, buflen); 2285 print_signal_handler(st, SIGPIPE, buf, buflen); 2286 print_signal_handler(st, SIGXFSZ, buf, buflen); 2287 print_signal_handler(st, SIGILL , buf, buflen); 2288 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2289 print_signal_handler(st, SR_signum, buf, buflen); 2290 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2291 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2292 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2293 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2294#if defined(PPC64) 2295 print_signal_handler(st, SIGTRAP, buf, buflen); 2296#endif 2297} 2298 2299static char saved_jvm_path[MAXPATHLEN] = {0}; 2300 2301// Find the full path to the current module, libjvm.so 2302void os::jvm_path(char *buf, jint buflen) { 2303 // Error checking. 2304 if (buflen < MAXPATHLEN) { 2305 assert(false, "must use a large-enough buffer"); 2306 buf[0] = '\0'; 2307 return; 2308 } 2309 // Lazy resolve the path to current module. 2310 if (saved_jvm_path[0] != 0) { 2311 strcpy(buf, saved_jvm_path); 2312 return; 2313 } 2314 2315 char dli_fname[MAXPATHLEN]; 2316 bool ret = dll_address_to_library_name( 2317 CAST_FROM_FN_PTR(address, os::jvm_path), 2318 dli_fname, sizeof(dli_fname), NULL); 2319 assert(ret, "cannot locate libjvm"); 2320 char *rp = NULL; 2321 if (ret && dli_fname[0] != '\0') { 2322 rp = realpath(dli_fname, buf); 2323 } 2324 if (rp == NULL) { 2325 return; 2326 } 2327 2328 if (Arguments::sun_java_launcher_is_altjvm()) { 2329 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2330 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2331 // If "/jre/lib/" appears at the right place in the string, then 2332 // assume we are installed in a JDK and we're done. Otherwise, check 2333 // for a JAVA_HOME environment variable and fix up the path so it 2334 // looks like libjvm.so is installed there (append a fake suffix 2335 // hotspot/libjvm.so). 2336 const char *p = buf + strlen(buf) - 1; 2337 for (int count = 0; p > buf && count < 5; ++count) { 2338 for (--p; p > buf && *p != '/'; --p) 2339 /* empty */ ; 2340 } 2341 2342 if (strncmp(p, "/jre/lib/", 9) != 0) { 2343 // Look for JAVA_HOME in the environment. 2344 char* java_home_var = ::getenv("JAVA_HOME"); 2345 if (java_home_var != NULL && java_home_var[0] != 0) { 2346 char* jrelib_p; 2347 int len; 2348 2349 // Check the current module name "libjvm.so". 2350 p = strrchr(buf, '/'); 2351 if (p == NULL) { 2352 return; 2353 } 2354 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2355 2356 rp = realpath(java_home_var, buf); 2357 if (rp == NULL) { 2358 return; 2359 } 2360 2361 // determine if this is a legacy image or modules image 2362 // modules image doesn't have "jre" subdirectory 2363 len = strlen(buf); 2364 assert(len < buflen, "Ran out of buffer room"); 2365 jrelib_p = buf + len; 2366 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2367 if (0 != access(buf, F_OK)) { 2368 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2369 } 2370 2371 if (0 == access(buf, F_OK)) { 2372 // Use current module name "libjvm.so" 2373 len = strlen(buf); 2374 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2375 } else { 2376 // Go back to path of .so 2377 rp = realpath(dli_fname, buf); 2378 if (rp == NULL) { 2379 return; 2380 } 2381 } 2382 } 2383 } 2384 } 2385 2386 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2387 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2388} 2389 2390void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2391 // no prefix required, not even "_" 2392} 2393 2394void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2395 // no suffix required 2396} 2397 2398//////////////////////////////////////////////////////////////////////////////// 2399// sun.misc.Signal support 2400 2401static volatile jint sigint_count = 0; 2402 2403static void UserHandler(int sig, void *siginfo, void *context) { 2404 // 4511530 - sem_post is serialized and handled by the manager thread. When 2405 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2406 // don't want to flood the manager thread with sem_post requests. 2407 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2408 return; 2409 } 2410 2411 // Ctrl-C is pressed during error reporting, likely because the error 2412 // handler fails to abort. Let VM die immediately. 2413 if (sig == SIGINT && is_error_reported()) { 2414 os::die(); 2415 } 2416 2417 os::signal_notify(sig); 2418} 2419 2420void* os::user_handler() { 2421 return CAST_FROM_FN_PTR(void*, UserHandler); 2422} 2423 2424class Semaphore : public StackObj { 2425 public: 2426 Semaphore(); 2427 ~Semaphore(); 2428 void signal(); 2429 void wait(); 2430 bool trywait(); 2431 bool timedwait(unsigned int sec, int nsec); 2432 private: 2433 sem_t _semaphore; 2434}; 2435 2436Semaphore::Semaphore() { 2437 sem_init(&_semaphore, 0, 0); 2438} 2439 2440Semaphore::~Semaphore() { 2441 sem_destroy(&_semaphore); 2442} 2443 2444void Semaphore::signal() { 2445 sem_post(&_semaphore); 2446} 2447 2448void Semaphore::wait() { 2449 sem_wait(&_semaphore); 2450} 2451 2452bool Semaphore::trywait() { 2453 return sem_trywait(&_semaphore) == 0; 2454} 2455 2456bool Semaphore::timedwait(unsigned int sec, int nsec) { 2457 2458 struct timespec ts; 2459 // Semaphore's are always associated with CLOCK_REALTIME 2460 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2461 // see unpackTime for discussion on overflow checking 2462 if (sec >= MAX_SECS) { 2463 ts.tv_sec += MAX_SECS; 2464 ts.tv_nsec = 0; 2465 } else { 2466 ts.tv_sec += sec; 2467 ts.tv_nsec += nsec; 2468 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2469 ts.tv_nsec -= NANOSECS_PER_SEC; 2470 ++ts.tv_sec; // note: this must be <= max_secs 2471 } 2472 } 2473 2474 while (1) { 2475 int result = sem_timedwait(&_semaphore, &ts); 2476 if (result == 0) { 2477 return true; 2478 } else if (errno == EINTR) { 2479 continue; 2480 } else if (errno == ETIMEDOUT) { 2481 return false; 2482 } else { 2483 return false; 2484 } 2485 } 2486} 2487 2488extern "C" { 2489 typedef void (*sa_handler_t)(int); 2490 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2491} 2492 2493void* os::signal(int signal_number, void* handler) { 2494 struct sigaction sigAct, oldSigAct; 2495 2496 sigfillset(&(sigAct.sa_mask)); 2497 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2498 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2499 2500 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2501 // -1 means registration failed 2502 return (void *)-1; 2503 } 2504 2505 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2506} 2507 2508void os::signal_raise(int signal_number) { 2509 ::raise(signal_number); 2510} 2511 2512// The following code is moved from os.cpp for making this 2513// code platform specific, which it is by its very nature. 2514 2515// Will be modified when max signal is changed to be dynamic 2516int os::sigexitnum_pd() { 2517 return NSIG; 2518} 2519 2520// a counter for each possible signal value 2521static volatile jint pending_signals[NSIG+1] = { 0 }; 2522 2523// Linux(POSIX) specific hand shaking semaphore. 2524static sem_t sig_sem; 2525static Semaphore sr_semaphore; 2526 2527void os::signal_init_pd() { 2528 // Initialize signal structures 2529 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2530 2531 // Initialize signal semaphore 2532 ::sem_init(&sig_sem, 0, 0); 2533} 2534 2535void os::signal_notify(int sig) { 2536 Atomic::inc(&pending_signals[sig]); 2537 ::sem_post(&sig_sem); 2538} 2539 2540static int check_pending_signals(bool wait) { 2541 Atomic::store(0, &sigint_count); 2542 for (;;) { 2543 for (int i = 0; i < NSIG + 1; i++) { 2544 jint n = pending_signals[i]; 2545 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2546 return i; 2547 } 2548 } 2549 if (!wait) { 2550 return -1; 2551 } 2552 JavaThread *thread = JavaThread::current(); 2553 ThreadBlockInVM tbivm(thread); 2554 2555 bool threadIsSuspended; 2556 do { 2557 thread->set_suspend_equivalent(); 2558 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2559 ::sem_wait(&sig_sem); 2560 2561 // were we externally suspended while we were waiting? 2562 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2563 if (threadIsSuspended) { 2564 // The semaphore has been incremented, but while we were waiting 2565 // another thread suspended us. We don't want to continue running 2566 // while suspended because that would surprise the thread that 2567 // suspended us. 2568 ::sem_post(&sig_sem); 2569 2570 thread->java_suspend_self(); 2571 } 2572 } while (threadIsSuspended); 2573 } 2574} 2575 2576int os::signal_lookup() { 2577 return check_pending_signals(false); 2578} 2579 2580int os::signal_wait() { 2581 return check_pending_signals(true); 2582} 2583 2584//////////////////////////////////////////////////////////////////////////////// 2585// Virtual Memory 2586 2587int os::vm_page_size() { 2588 // Seems redundant as all get out 2589 assert(os::Linux::page_size() != -1, "must call os::init"); 2590 return os::Linux::page_size(); 2591} 2592 2593// Solaris allocates memory by pages. 2594int os::vm_allocation_granularity() { 2595 assert(os::Linux::page_size() != -1, "must call os::init"); 2596 return os::Linux::page_size(); 2597} 2598 2599// Rationale behind this function: 2600// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2601// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2602// samples for JITted code. Here we create private executable mapping over the code cache 2603// and then we can use standard (well, almost, as mapping can change) way to provide 2604// info for the reporting script by storing timestamp and location of symbol 2605void linux_wrap_code(char* base, size_t size) { 2606 static volatile jint cnt = 0; 2607 2608 if (!UseOprofile) { 2609 return; 2610 } 2611 2612 char buf[PATH_MAX+1]; 2613 int num = Atomic::add(1, &cnt); 2614 2615 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2616 os::get_temp_directory(), os::current_process_id(), num); 2617 unlink(buf); 2618 2619 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2620 2621 if (fd != -1) { 2622 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2623 if (rv != (off_t)-1) { 2624 if (::write(fd, "", 1) == 1) { 2625 mmap(base, size, 2626 PROT_READ|PROT_WRITE|PROT_EXEC, 2627 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2628 } 2629 } 2630 ::close(fd); 2631 unlink(buf); 2632 } 2633} 2634 2635static bool recoverable_mmap_error(int err) { 2636 // See if the error is one we can let the caller handle. This 2637 // list of errno values comes from JBS-6843484. I can't find a 2638 // Linux man page that documents this specific set of errno 2639 // values so while this list currently matches Solaris, it may 2640 // change as we gain experience with this failure mode. 2641 switch (err) { 2642 case EBADF: 2643 case EINVAL: 2644 case ENOTSUP: 2645 // let the caller deal with these errors 2646 return true; 2647 2648 default: 2649 // Any remaining errors on this OS can cause our reserved mapping 2650 // to be lost. That can cause confusion where different data 2651 // structures think they have the same memory mapped. The worst 2652 // scenario is if both the VM and a library think they have the 2653 // same memory mapped. 2654 return false; 2655 } 2656} 2657 2658static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2659 int err) { 2660 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2661 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2662 strerror(err), err); 2663} 2664 2665static void warn_fail_commit_memory(char* addr, size_t size, 2666 size_t alignment_hint, bool exec, 2667 int err) { 2668 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2669 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2670 alignment_hint, exec, strerror(err), err); 2671} 2672 2673// NOTE: Linux kernel does not really reserve the pages for us. 2674// All it does is to check if there are enough free pages 2675// left at the time of mmap(). This could be a potential 2676// problem. 2677int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2678 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2679 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2680 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2681 if (res != (uintptr_t) MAP_FAILED) { 2682 if (UseNUMAInterleaving) { 2683 numa_make_global(addr, size); 2684 } 2685 return 0; 2686 } 2687 2688 int err = errno; // save errno from mmap() call above 2689 2690 if (!recoverable_mmap_error(err)) { 2691 warn_fail_commit_memory(addr, size, exec, err); 2692 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2693 } 2694 2695 return err; 2696} 2697 2698bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2699 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2700} 2701 2702void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2703 const char* mesg) { 2704 assert(mesg != NULL, "mesg must be specified"); 2705 int err = os::Linux::commit_memory_impl(addr, size, exec); 2706 if (err != 0) { 2707 // the caller wants all commit errors to exit with the specified mesg: 2708 warn_fail_commit_memory(addr, size, exec, err); 2709 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2710 } 2711} 2712 2713// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2714#ifndef MAP_HUGETLB 2715 #define MAP_HUGETLB 0x40000 2716#endif 2717 2718// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2719#ifndef MADV_HUGEPAGE 2720 #define MADV_HUGEPAGE 14 2721#endif 2722 2723int os::Linux::commit_memory_impl(char* addr, size_t size, 2724 size_t alignment_hint, bool exec) { 2725 int err = os::Linux::commit_memory_impl(addr, size, exec); 2726 if (err == 0) { 2727 realign_memory(addr, size, alignment_hint); 2728 } 2729 return err; 2730} 2731 2732bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2733 bool exec) { 2734 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2735} 2736 2737void os::pd_commit_memory_or_exit(char* addr, size_t size, 2738 size_t alignment_hint, bool exec, 2739 const char* mesg) { 2740 assert(mesg != NULL, "mesg must be specified"); 2741 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2742 if (err != 0) { 2743 // the caller wants all commit errors to exit with the specified mesg: 2744 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2745 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2746 } 2747} 2748 2749void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2750 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2751 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2752 // be supported or the memory may already be backed by huge pages. 2753 ::madvise(addr, bytes, MADV_HUGEPAGE); 2754 } 2755} 2756 2757void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2758 // This method works by doing an mmap over an existing mmaping and effectively discarding 2759 // the existing pages. However it won't work for SHM-based large pages that cannot be 2760 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2761 // small pages on top of the SHM segment. This method always works for small pages, so we 2762 // allow that in any case. 2763 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2764 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2765 } 2766} 2767 2768void os::numa_make_global(char *addr, size_t bytes) { 2769 Linux::numa_interleave_memory(addr, bytes); 2770} 2771 2772// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2773// bind policy to MPOL_PREFERRED for the current thread. 2774#define USE_MPOL_PREFERRED 0 2775 2776void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2777 // To make NUMA and large pages more robust when both enabled, we need to ease 2778 // the requirements on where the memory should be allocated. MPOL_BIND is the 2779 // default policy and it will force memory to be allocated on the specified 2780 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2781 // the specified node, but will not force it. Using this policy will prevent 2782 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2783 // free large pages. 2784 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2785 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2786} 2787 2788bool os::numa_topology_changed() { return false; } 2789 2790size_t os::numa_get_groups_num() { 2791 int max_node = Linux::numa_max_node(); 2792 return max_node > 0 ? max_node + 1 : 1; 2793} 2794 2795int os::numa_get_group_id() { 2796 int cpu_id = Linux::sched_getcpu(); 2797 if (cpu_id != -1) { 2798 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2799 if (lgrp_id != -1) { 2800 return lgrp_id; 2801 } 2802 } 2803 return 0; 2804} 2805 2806size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2807 for (size_t i = 0; i < size; i++) { 2808 ids[i] = i; 2809 } 2810 return size; 2811} 2812 2813bool os::get_page_info(char *start, page_info* info) { 2814 return false; 2815} 2816 2817char *os::scan_pages(char *start, char* end, page_info* page_expected, 2818 page_info* page_found) { 2819 return end; 2820} 2821 2822 2823int os::Linux::sched_getcpu_syscall(void) { 2824 unsigned int cpu; 2825 int retval = -1; 2826 2827#if defined(IA32) 2828 #ifndef SYS_getcpu 2829 #define SYS_getcpu 318 2830 #endif 2831 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2832#elif defined(AMD64) 2833// Unfortunately we have to bring all these macros here from vsyscall.h 2834// to be able to compile on old linuxes. 2835 #define __NR_vgetcpu 2 2836 #define VSYSCALL_START (-10UL << 20) 2837 #define VSYSCALL_SIZE 1024 2838 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2839 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2840 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2841 retval = vgetcpu(&cpu, NULL, NULL); 2842#endif 2843 2844 return (retval == -1) ? retval : cpu; 2845} 2846 2847// Something to do with the numa-aware allocator needs these symbols 2848extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2849extern "C" JNIEXPORT void numa_error(char *where) { } 2850extern "C" JNIEXPORT int fork1() { return fork(); } 2851 2852 2853// If we are running with libnuma version > 2, then we should 2854// be trying to use symbols with versions 1.1 2855// If we are running with earlier version, which did not have symbol versions, 2856// we should use the base version. 2857void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2858 void *f = dlvsym(handle, name, "libnuma_1.1"); 2859 if (f == NULL) { 2860 f = dlsym(handle, name); 2861 } 2862 return f; 2863} 2864 2865bool os::Linux::libnuma_init() { 2866 // sched_getcpu() should be in libc. 2867 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2868 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2869 2870 // If it's not, try a direct syscall. 2871 if (sched_getcpu() == -1) { 2872 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2873 (void*)&sched_getcpu_syscall)); 2874 } 2875 2876 if (sched_getcpu() != -1) { // Does it work? 2877 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2878 if (handle != NULL) { 2879 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2880 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2881 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2882 libnuma_dlsym(handle, "numa_max_node"))); 2883 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2884 libnuma_dlsym(handle, "numa_available"))); 2885 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2886 libnuma_dlsym(handle, "numa_tonode_memory"))); 2887 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2888 libnuma_dlsym(handle, "numa_interleave_memory"))); 2889 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2890 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2891 2892 2893 if (numa_available() != -1) { 2894 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2895 // Create a cpu -> node mapping 2896 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2897 rebuild_cpu_to_node_map(); 2898 return true; 2899 } 2900 } 2901 } 2902 return false; 2903} 2904 2905// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2906// The table is later used in get_node_by_cpu(). 2907void os::Linux::rebuild_cpu_to_node_map() { 2908 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2909 // in libnuma (possible values are starting from 16, 2910 // and continuing up with every other power of 2, but less 2911 // than the maximum number of CPUs supported by kernel), and 2912 // is a subject to change (in libnuma version 2 the requirements 2913 // are more reasonable) we'll just hardcode the number they use 2914 // in the library. 2915 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2916 2917 size_t cpu_num = os::active_processor_count(); 2918 size_t cpu_map_size = NCPUS / BitsPerCLong; 2919 size_t cpu_map_valid_size = 2920 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2921 2922 cpu_to_node()->clear(); 2923 cpu_to_node()->at_grow(cpu_num - 1); 2924 size_t node_num = numa_get_groups_num(); 2925 2926 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2927 for (size_t i = 0; i < node_num; i++) { 2928 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2929 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2930 if (cpu_map[j] != 0) { 2931 for (size_t k = 0; k < BitsPerCLong; k++) { 2932 if (cpu_map[j] & (1UL << k)) { 2933 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2934 } 2935 } 2936 } 2937 } 2938 } 2939 } 2940 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2941} 2942 2943int os::Linux::get_node_by_cpu(int cpu_id) { 2944 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2945 return cpu_to_node()->at(cpu_id); 2946 } 2947 return -1; 2948} 2949 2950GrowableArray<int>* os::Linux::_cpu_to_node; 2951os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2952os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2953os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2954os::Linux::numa_available_func_t os::Linux::_numa_available; 2955os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2956os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2957os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2958unsigned long* os::Linux::_numa_all_nodes; 2959 2960bool os::pd_uncommit_memory(char* addr, size_t size) { 2961 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2962 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2963 return res != (uintptr_t) MAP_FAILED; 2964} 2965 2966static address get_stack_commited_bottom(address bottom, size_t size) { 2967 address nbot = bottom; 2968 address ntop = bottom + size; 2969 2970 size_t page_sz = os::vm_page_size(); 2971 unsigned pages = size / page_sz; 2972 2973 unsigned char vec[1]; 2974 unsigned imin = 1, imax = pages + 1, imid; 2975 int mincore_return_value = 0; 2976 2977 assert(imin <= imax, "Unexpected page size"); 2978 2979 while (imin < imax) { 2980 imid = (imax + imin) / 2; 2981 nbot = ntop - (imid * page_sz); 2982 2983 // Use a trick with mincore to check whether the page is mapped or not. 2984 // mincore sets vec to 1 if page resides in memory and to 0 if page 2985 // is swapped output but if page we are asking for is unmapped 2986 // it returns -1,ENOMEM 2987 mincore_return_value = mincore(nbot, page_sz, vec); 2988 2989 if (mincore_return_value == -1) { 2990 // Page is not mapped go up 2991 // to find first mapped page 2992 if (errno != EAGAIN) { 2993 assert(errno == ENOMEM, "Unexpected mincore errno"); 2994 imax = imid; 2995 } 2996 } else { 2997 // Page is mapped go down 2998 // to find first not mapped page 2999 imin = imid + 1; 3000 } 3001 } 3002 3003 nbot = nbot + page_sz; 3004 3005 // Adjust stack bottom one page up if last checked page is not mapped 3006 if (mincore_return_value == -1) { 3007 nbot = nbot + page_sz; 3008 } 3009 3010 return nbot; 3011} 3012 3013 3014// Linux uses a growable mapping for the stack, and if the mapping for 3015// the stack guard pages is not removed when we detach a thread the 3016// stack cannot grow beyond the pages where the stack guard was 3017// mapped. If at some point later in the process the stack expands to 3018// that point, the Linux kernel cannot expand the stack any further 3019// because the guard pages are in the way, and a segfault occurs. 3020// 3021// However, it's essential not to split the stack region by unmapping 3022// a region (leaving a hole) that's already part of the stack mapping, 3023// so if the stack mapping has already grown beyond the guard pages at 3024// the time we create them, we have to truncate the stack mapping. 3025// So, we need to know the extent of the stack mapping when 3026// create_stack_guard_pages() is called. 3027 3028// We only need this for stacks that are growable: at the time of 3029// writing thread stacks don't use growable mappings (i.e. those 3030// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3031// only applies to the main thread. 3032 3033// If the (growable) stack mapping already extends beyond the point 3034// where we're going to put our guard pages, truncate the mapping at 3035// that point by munmap()ping it. This ensures that when we later 3036// munmap() the guard pages we don't leave a hole in the stack 3037// mapping. This only affects the main/initial thread 3038 3039bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3040 if (os::Linux::is_initial_thread()) { 3041 // As we manually grow stack up to bottom inside create_attached_thread(), 3042 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3043 // we don't need to do anything special. 3044 // Check it first, before calling heavy function. 3045 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3046 unsigned char vec[1]; 3047 3048 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3049 // Fallback to slow path on all errors, including EAGAIN 3050 stack_extent = (uintptr_t) get_stack_commited_bottom( 3051 os::Linux::initial_thread_stack_bottom(), 3052 (size_t)addr - stack_extent); 3053 } 3054 3055 if (stack_extent < (uintptr_t)addr) { 3056 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3057 } 3058 } 3059 3060 return os::commit_memory(addr, size, !ExecMem); 3061} 3062 3063// If this is a growable mapping, remove the guard pages entirely by 3064// munmap()ping them. If not, just call uncommit_memory(). This only 3065// affects the main/initial thread, but guard against future OS changes 3066// It's safe to always unmap guard pages for initial thread because we 3067// always place it right after end of the mapped region 3068 3069bool os::remove_stack_guard_pages(char* addr, size_t size) { 3070 uintptr_t stack_extent, stack_base; 3071 3072 if (os::Linux::is_initial_thread()) { 3073 return ::munmap(addr, size) == 0; 3074 } 3075 3076 return os::uncommit_memory(addr, size); 3077} 3078 3079static address _highest_vm_reserved_address = NULL; 3080 3081// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3082// at 'requested_addr'. If there are existing memory mappings at the same 3083// location, however, they will be overwritten. If 'fixed' is false, 3084// 'requested_addr' is only treated as a hint, the return value may or 3085// may not start from the requested address. Unlike Linux mmap(), this 3086// function returns NULL to indicate failure. 3087static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3088 char * addr; 3089 int flags; 3090 3091 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3092 if (fixed) { 3093 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3094 flags |= MAP_FIXED; 3095 } 3096 3097 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3098 // touch an uncommitted page. Otherwise, the read/write might 3099 // succeed if we have enough swap space to back the physical page. 3100 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3101 flags, -1, 0); 3102 3103 if (addr != MAP_FAILED) { 3104 // anon_mmap() should only get called during VM initialization, 3105 // don't need lock (actually we can skip locking even it can be called 3106 // from multiple threads, because _highest_vm_reserved_address is just a 3107 // hint about the upper limit of non-stack memory regions.) 3108 if ((address)addr + bytes > _highest_vm_reserved_address) { 3109 _highest_vm_reserved_address = (address)addr + bytes; 3110 } 3111 } 3112 3113 return addr == MAP_FAILED ? NULL : addr; 3114} 3115 3116// Don't update _highest_vm_reserved_address, because there might be memory 3117// regions above addr + size. If so, releasing a memory region only creates 3118// a hole in the address space, it doesn't help prevent heap-stack collision. 3119// 3120static int anon_munmap(char * addr, size_t size) { 3121 return ::munmap(addr, size) == 0; 3122} 3123 3124char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3125 size_t alignment_hint) { 3126 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3127} 3128 3129bool os::pd_release_memory(char* addr, size_t size) { 3130 return anon_munmap(addr, size); 3131} 3132 3133static address highest_vm_reserved_address() { 3134 return _highest_vm_reserved_address; 3135} 3136 3137static bool linux_mprotect(char* addr, size_t size, int prot) { 3138 // Linux wants the mprotect address argument to be page aligned. 3139 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3140 3141 // According to SUSv3, mprotect() should only be used with mappings 3142 // established by mmap(), and mmap() always maps whole pages. Unaligned 3143 // 'addr' likely indicates problem in the VM (e.g. trying to change 3144 // protection of malloc'ed or statically allocated memory). Check the 3145 // caller if you hit this assert. 3146 assert(addr == bottom, "sanity check"); 3147 3148 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3149 return ::mprotect(bottom, size, prot) == 0; 3150} 3151 3152// Set protections specified 3153bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3154 bool is_committed) { 3155 unsigned int p = 0; 3156 switch (prot) { 3157 case MEM_PROT_NONE: p = PROT_NONE; break; 3158 case MEM_PROT_READ: p = PROT_READ; break; 3159 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3160 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3161 default: 3162 ShouldNotReachHere(); 3163 } 3164 // is_committed is unused. 3165 return linux_mprotect(addr, bytes, p); 3166} 3167 3168bool os::guard_memory(char* addr, size_t size) { 3169 return linux_mprotect(addr, size, PROT_NONE); 3170} 3171 3172bool os::unguard_memory(char* addr, size_t size) { 3173 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3174} 3175 3176bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3177 size_t page_size) { 3178 bool result = false; 3179 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3180 MAP_ANONYMOUS|MAP_PRIVATE, 3181 -1, 0); 3182 if (p != MAP_FAILED) { 3183 void *aligned_p = align_ptr_up(p, page_size); 3184 3185 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3186 3187 munmap(p, page_size * 2); 3188 } 3189 3190 if (warn && !result) { 3191 warning("TransparentHugePages is not supported by the operating system."); 3192 } 3193 3194 return result; 3195} 3196 3197bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3198 bool result = false; 3199 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3200 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3201 -1, 0); 3202 3203 if (p != MAP_FAILED) { 3204 // We don't know if this really is a huge page or not. 3205 FILE *fp = fopen("/proc/self/maps", "r"); 3206 if (fp) { 3207 while (!feof(fp)) { 3208 char chars[257]; 3209 long x = 0; 3210 if (fgets(chars, sizeof(chars), fp)) { 3211 if (sscanf(chars, "%lx-%*x", &x) == 1 3212 && x == (long)p) { 3213 if (strstr (chars, "hugepage")) { 3214 result = true; 3215 break; 3216 } 3217 } 3218 } 3219 } 3220 fclose(fp); 3221 } 3222 munmap(p, page_size); 3223 } 3224 3225 if (warn && !result) { 3226 warning("HugeTLBFS is not supported by the operating system."); 3227 } 3228 3229 return result; 3230} 3231 3232// Set the coredump_filter bits to include largepages in core dump (bit 6) 3233// 3234// From the coredump_filter documentation: 3235// 3236// - (bit 0) anonymous private memory 3237// - (bit 1) anonymous shared memory 3238// - (bit 2) file-backed private memory 3239// - (bit 3) file-backed shared memory 3240// - (bit 4) ELF header pages in file-backed private memory areas (it is 3241// effective only if the bit 2 is cleared) 3242// - (bit 5) hugetlb private memory 3243// - (bit 6) hugetlb shared memory 3244// 3245static void set_coredump_filter(void) { 3246 FILE *f; 3247 long cdm; 3248 3249 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3250 return; 3251 } 3252 3253 if (fscanf(f, "%lx", &cdm) != 1) { 3254 fclose(f); 3255 return; 3256 } 3257 3258 rewind(f); 3259 3260 if ((cdm & LARGEPAGES_BIT) == 0) { 3261 cdm |= LARGEPAGES_BIT; 3262 fprintf(f, "%#lx", cdm); 3263 } 3264 3265 fclose(f); 3266} 3267 3268// Large page support 3269 3270static size_t _large_page_size = 0; 3271 3272size_t os::Linux::find_large_page_size() { 3273 size_t large_page_size = 0; 3274 3275 // large_page_size on Linux is used to round up heap size. x86 uses either 3276 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3277 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3278 // page as large as 256M. 3279 // 3280 // Here we try to figure out page size by parsing /proc/meminfo and looking 3281 // for a line with the following format: 3282 // Hugepagesize: 2048 kB 3283 // 3284 // If we can't determine the value (e.g. /proc is not mounted, or the text 3285 // format has been changed), we'll use the largest page size supported by 3286 // the processor. 3287 3288#ifndef ZERO 3289 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3290 ARM_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M); 3291#endif // ZERO 3292 3293 FILE *fp = fopen("/proc/meminfo", "r"); 3294 if (fp) { 3295 while (!feof(fp)) { 3296 int x = 0; 3297 char buf[16]; 3298 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3299 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3300 large_page_size = x * K; 3301 break; 3302 } 3303 } else { 3304 // skip to next line 3305 for (;;) { 3306 int ch = fgetc(fp); 3307 if (ch == EOF || ch == (int)'\n') break; 3308 } 3309 } 3310 } 3311 fclose(fp); 3312 } 3313 3314 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3315 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3316 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3317 proper_unit_for_byte_size(large_page_size)); 3318 } 3319 3320 return large_page_size; 3321} 3322 3323size_t os::Linux::setup_large_page_size() { 3324 _large_page_size = Linux::find_large_page_size(); 3325 const size_t default_page_size = (size_t)Linux::page_size(); 3326 if (_large_page_size > default_page_size) { 3327 _page_sizes[0] = _large_page_size; 3328 _page_sizes[1] = default_page_size; 3329 _page_sizes[2] = 0; 3330 } 3331 3332 return _large_page_size; 3333} 3334 3335bool os::Linux::setup_large_page_type(size_t page_size) { 3336 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3337 FLAG_IS_DEFAULT(UseSHM) && 3338 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3339 3340 // The type of large pages has not been specified by the user. 3341 3342 // Try UseHugeTLBFS and then UseSHM. 3343 UseHugeTLBFS = UseSHM = true; 3344 3345 // Don't try UseTransparentHugePages since there are known 3346 // performance issues with it turned on. This might change in the future. 3347 UseTransparentHugePages = false; 3348 } 3349 3350 if (UseTransparentHugePages) { 3351 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3352 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3353 UseHugeTLBFS = false; 3354 UseSHM = false; 3355 return true; 3356 } 3357 UseTransparentHugePages = false; 3358 } 3359 3360 if (UseHugeTLBFS) { 3361 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3362 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3363 UseSHM = false; 3364 return true; 3365 } 3366 UseHugeTLBFS = false; 3367 } 3368 3369 return UseSHM; 3370} 3371 3372void os::large_page_init() { 3373 if (!UseLargePages && 3374 !UseTransparentHugePages && 3375 !UseHugeTLBFS && 3376 !UseSHM) { 3377 // Not using large pages. 3378 return; 3379 } 3380 3381 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3382 // The user explicitly turned off large pages. 3383 // Ignore the rest of the large pages flags. 3384 UseTransparentHugePages = false; 3385 UseHugeTLBFS = false; 3386 UseSHM = false; 3387 return; 3388 } 3389 3390 size_t large_page_size = Linux::setup_large_page_size(); 3391 UseLargePages = Linux::setup_large_page_type(large_page_size); 3392 3393 set_coredump_filter(); 3394} 3395 3396#ifndef SHM_HUGETLB 3397 #define SHM_HUGETLB 04000 3398#endif 3399 3400char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3401 char* req_addr, bool exec) { 3402 // "exec" is passed in but not used. Creating the shared image for 3403 // the code cache doesn't have an SHM_X executable permission to check. 3404 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3405 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3406 3407 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3408 return NULL; // Fallback to small pages. 3409 } 3410 3411 key_t key = IPC_PRIVATE; 3412 char *addr; 3413 3414 bool warn_on_failure = UseLargePages && 3415 (!FLAG_IS_DEFAULT(UseLargePages) || 3416 !FLAG_IS_DEFAULT(UseSHM) || 3417 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3418 char msg[128]; 3419 3420 // Create a large shared memory region to attach to based on size. 3421 // Currently, size is the total size of the heap 3422 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3423 if (shmid == -1) { 3424 // Possible reasons for shmget failure: 3425 // 1. shmmax is too small for Java heap. 3426 // > check shmmax value: cat /proc/sys/kernel/shmmax 3427 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3428 // 2. not enough large page memory. 3429 // > check available large pages: cat /proc/meminfo 3430 // > increase amount of large pages: 3431 // echo new_value > /proc/sys/vm/nr_hugepages 3432 // Note 1: different Linux may use different name for this property, 3433 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3434 // Note 2: it's possible there's enough physical memory available but 3435 // they are so fragmented after a long run that they can't 3436 // coalesce into large pages. Try to reserve large pages when 3437 // the system is still "fresh". 3438 if (warn_on_failure) { 3439 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3440 warning("%s", msg); 3441 } 3442 return NULL; 3443 } 3444 3445 // attach to the region 3446 addr = (char*)shmat(shmid, req_addr, 0); 3447 int err = errno; 3448 3449 // Remove shmid. If shmat() is successful, the actual shared memory segment 3450 // will be deleted when it's detached by shmdt() or when the process 3451 // terminates. If shmat() is not successful this will remove the shared 3452 // segment immediately. 3453 shmctl(shmid, IPC_RMID, NULL); 3454 3455 if ((intptr_t)addr == -1) { 3456 if (warn_on_failure) { 3457 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3458 warning("%s", msg); 3459 } 3460 return NULL; 3461 } 3462 3463 return addr; 3464} 3465 3466static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3467 int error) { 3468 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3469 3470 bool warn_on_failure = UseLargePages && 3471 (!FLAG_IS_DEFAULT(UseLargePages) || 3472 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3473 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3474 3475 if (warn_on_failure) { 3476 char msg[128]; 3477 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3478 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3479 warning("%s", msg); 3480 } 3481} 3482 3483char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3484 char* req_addr, 3485 bool exec) { 3486 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3487 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3488 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3489 3490 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3491 char* addr = (char*)::mmap(req_addr, bytes, prot, 3492 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3493 -1, 0); 3494 3495 if (addr == MAP_FAILED) { 3496 warn_on_large_pages_failure(req_addr, bytes, errno); 3497 return NULL; 3498 } 3499 3500 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3501 3502 return addr; 3503} 3504 3505char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3506 size_t alignment, 3507 char* req_addr, 3508 bool exec) { 3509 size_t large_page_size = os::large_page_size(); 3510 3511 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3512 3513 // Allocate small pages. 3514 3515 char* start; 3516 if (req_addr != NULL) { 3517 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3518 assert(is_size_aligned(bytes, alignment), "Must be"); 3519 start = os::reserve_memory(bytes, req_addr); 3520 assert(start == NULL || start == req_addr, "Must be"); 3521 } else { 3522 start = os::reserve_memory_aligned(bytes, alignment); 3523 } 3524 3525 if (start == NULL) { 3526 return NULL; 3527 } 3528 3529 assert(is_ptr_aligned(start, alignment), "Must be"); 3530 3531 if (MemTracker::tracking_level() > NMT_minimal) { 3532 // os::reserve_memory_special will record this memory area. 3533 // Need to release it here to prevent overlapping reservations. 3534 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3535 tkr.record((address)start, bytes); 3536 } 3537 3538 char* end = start + bytes; 3539 3540 // Find the regions of the allocated chunk that can be promoted to large pages. 3541 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3542 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3543 3544 size_t lp_bytes = lp_end - lp_start; 3545 3546 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3547 3548 if (lp_bytes == 0) { 3549 // The mapped region doesn't even span the start and the end of a large page. 3550 // Fall back to allocate a non-special area. 3551 ::munmap(start, end - start); 3552 return NULL; 3553 } 3554 3555 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3556 3557 3558 void* result; 3559 3560 if (start != lp_start) { 3561 result = ::mmap(start, lp_start - start, prot, 3562 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3563 -1, 0); 3564 if (result == MAP_FAILED) { 3565 ::munmap(lp_start, end - lp_start); 3566 return NULL; 3567 } 3568 } 3569 3570 result = ::mmap(lp_start, lp_bytes, prot, 3571 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3572 -1, 0); 3573 if (result == MAP_FAILED) { 3574 warn_on_large_pages_failure(req_addr, bytes, errno); 3575 // If the mmap above fails, the large pages region will be unmapped and we 3576 // have regions before and after with small pages. Release these regions. 3577 // 3578 // | mapped | unmapped | mapped | 3579 // ^ ^ ^ ^ 3580 // start lp_start lp_end end 3581 // 3582 ::munmap(start, lp_start - start); 3583 ::munmap(lp_end, end - lp_end); 3584 return NULL; 3585 } 3586 3587 if (lp_end != end) { 3588 result = ::mmap(lp_end, end - lp_end, prot, 3589 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3590 -1, 0); 3591 if (result == MAP_FAILED) { 3592 ::munmap(start, lp_end - start); 3593 return NULL; 3594 } 3595 } 3596 3597 return start; 3598} 3599 3600char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3601 size_t alignment, 3602 char* req_addr, 3603 bool exec) { 3604 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3605 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3606 assert(is_power_of_2(alignment), "Must be"); 3607 assert(is_power_of_2(os::large_page_size()), "Must be"); 3608 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3609 3610 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3611 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3612 } else { 3613 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3614 } 3615} 3616 3617char* os::reserve_memory_special(size_t bytes, size_t alignment, 3618 char* req_addr, bool exec) { 3619 assert(UseLargePages, "only for large pages"); 3620 3621 char* addr; 3622 if (UseSHM) { 3623 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3624 } else { 3625 assert(UseHugeTLBFS, "must be"); 3626 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3627 } 3628 3629 if (addr != NULL) { 3630 if (UseNUMAInterleaving) { 3631 numa_make_global(addr, bytes); 3632 } 3633 3634 // The memory is committed 3635 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3636 } 3637 3638 return addr; 3639} 3640 3641bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3642 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3643 return shmdt(base) == 0; 3644} 3645 3646bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3647 return pd_release_memory(base, bytes); 3648} 3649 3650bool os::release_memory_special(char* base, size_t bytes) { 3651 bool res; 3652 if (MemTracker::tracking_level() > NMT_minimal) { 3653 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3654 res = os::Linux::release_memory_special_impl(base, bytes); 3655 if (res) { 3656 tkr.record((address)base, bytes); 3657 } 3658 3659 } else { 3660 res = os::Linux::release_memory_special_impl(base, bytes); 3661 } 3662 return res; 3663} 3664 3665bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3666 assert(UseLargePages, "only for large pages"); 3667 bool res; 3668 3669 if (UseSHM) { 3670 res = os::Linux::release_memory_special_shm(base, bytes); 3671 } else { 3672 assert(UseHugeTLBFS, "must be"); 3673 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3674 } 3675 return res; 3676} 3677 3678size_t os::large_page_size() { 3679 return _large_page_size; 3680} 3681 3682// With SysV SHM the entire memory region must be allocated as shared 3683// memory. 3684// HugeTLBFS allows application to commit large page memory on demand. 3685// However, when committing memory with HugeTLBFS fails, the region 3686// that was supposed to be committed will lose the old reservation 3687// and allow other threads to steal that memory region. Because of this 3688// behavior we can't commit HugeTLBFS memory. 3689bool os::can_commit_large_page_memory() { 3690 return UseTransparentHugePages; 3691} 3692 3693bool os::can_execute_large_page_memory() { 3694 return UseTransparentHugePages || UseHugeTLBFS; 3695} 3696 3697// Reserve memory at an arbitrary address, only if that area is 3698// available (and not reserved for something else). 3699 3700char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3701 const int max_tries = 10; 3702 char* base[max_tries]; 3703 size_t size[max_tries]; 3704 const size_t gap = 0x000000; 3705 3706 // Assert only that the size is a multiple of the page size, since 3707 // that's all that mmap requires, and since that's all we really know 3708 // about at this low abstraction level. If we need higher alignment, 3709 // we can either pass an alignment to this method or verify alignment 3710 // in one of the methods further up the call chain. See bug 5044738. 3711 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3712 3713 // Repeatedly allocate blocks until the block is allocated at the 3714 // right spot. Give up after max_tries. Note that reserve_memory() will 3715 // automatically update _highest_vm_reserved_address if the call is 3716 // successful. The variable tracks the highest memory address every reserved 3717 // by JVM. It is used to detect heap-stack collision if running with 3718 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3719 // space than needed, it could confuse the collision detecting code. To 3720 // solve the problem, save current _highest_vm_reserved_address and 3721 // calculate the correct value before return. 3722 address old_highest = _highest_vm_reserved_address; 3723 3724 // Linux mmap allows caller to pass an address as hint; give it a try first, 3725 // if kernel honors the hint then we can return immediately. 3726 char * addr = anon_mmap(requested_addr, bytes, false); 3727 if (addr == requested_addr) { 3728 return requested_addr; 3729 } 3730 3731 if (addr != NULL) { 3732 // mmap() is successful but it fails to reserve at the requested address 3733 anon_munmap(addr, bytes); 3734 } 3735 3736 int i; 3737 for (i = 0; i < max_tries; ++i) { 3738 base[i] = reserve_memory(bytes); 3739 3740 if (base[i] != NULL) { 3741 // Is this the block we wanted? 3742 if (base[i] == requested_addr) { 3743 size[i] = bytes; 3744 break; 3745 } 3746 3747 // Does this overlap the block we wanted? Give back the overlapped 3748 // parts and try again. 3749 3750 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3751 if (top_overlap >= 0 && top_overlap < bytes) { 3752 unmap_memory(base[i], top_overlap); 3753 base[i] += top_overlap; 3754 size[i] = bytes - top_overlap; 3755 } else { 3756 size_t bottom_overlap = base[i] + bytes - requested_addr; 3757 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3758 unmap_memory(requested_addr, bottom_overlap); 3759 size[i] = bytes - bottom_overlap; 3760 } else { 3761 size[i] = bytes; 3762 } 3763 } 3764 } 3765 } 3766 3767 // Give back the unused reserved pieces. 3768 3769 for (int j = 0; j < i; ++j) { 3770 if (base[j] != NULL) { 3771 unmap_memory(base[j], size[j]); 3772 } 3773 } 3774 3775 if (i < max_tries) { 3776 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3777 return requested_addr; 3778 } else { 3779 _highest_vm_reserved_address = old_highest; 3780 return NULL; 3781 } 3782} 3783 3784size_t os::read(int fd, void *buf, unsigned int nBytes) { 3785 return ::read(fd, buf, nBytes); 3786} 3787 3788size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) { 3789 return ::pread(fd, buf, nBytes, offset); 3790} 3791 3792// Short sleep, direct OS call. 3793// 3794// Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3795// sched_yield(2) will actually give up the CPU: 3796// 3797// * Alone on this pariticular CPU, keeps running. 3798// * Before the introduction of "skip_buddy" with "compat_yield" disabled 3799// (pre 2.6.39). 3800// 3801// So calling this with 0 is an alternative. 3802// 3803void os::naked_short_sleep(jlong ms) { 3804 struct timespec req; 3805 3806 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3807 req.tv_sec = 0; 3808 if (ms > 0) { 3809 req.tv_nsec = (ms % 1000) * 1000000; 3810 } else { 3811 req.tv_nsec = 1; 3812 } 3813 3814 nanosleep(&req, NULL); 3815 3816 return; 3817} 3818 3819// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3820void os::infinite_sleep() { 3821 while (true) { // sleep forever ... 3822 ::sleep(100); // ... 100 seconds at a time 3823 } 3824} 3825 3826// Used to convert frequent JVM_Yield() to nops 3827bool os::dont_yield() { 3828 return DontYieldALot; 3829} 3830 3831void os::naked_yield() { 3832 sched_yield(); 3833} 3834 3835//////////////////////////////////////////////////////////////////////////////// 3836// thread priority support 3837 3838// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3839// only supports dynamic priority, static priority must be zero. For real-time 3840// applications, Linux supports SCHED_RR which allows static priority (1-99). 3841// However, for large multi-threaded applications, SCHED_RR is not only slower 3842// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3843// of 5 runs - Sep 2005). 3844// 3845// The following code actually changes the niceness of kernel-thread/LWP. It 3846// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3847// not the entire user process, and user level threads are 1:1 mapped to kernel 3848// threads. It has always been the case, but could change in the future. For 3849// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3850// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3851 3852int os::java_to_os_priority[CriticalPriority + 1] = { 3853 19, // 0 Entry should never be used 3854 3855 4, // 1 MinPriority 3856 3, // 2 3857 2, // 3 3858 3859 1, // 4 3860 0, // 5 NormPriority 3861 -1, // 6 3862 3863 -2, // 7 3864 -3, // 8 3865 -4, // 9 NearMaxPriority 3866 3867 -5, // 10 MaxPriority 3868 3869 -5 // 11 CriticalPriority 3870}; 3871 3872static int prio_init() { 3873 if (ThreadPriorityPolicy == 1) { 3874 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3875 // if effective uid is not root. Perhaps, a more elegant way of doing 3876 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3877 if (geteuid() != 0) { 3878 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3879 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3880 } 3881 ThreadPriorityPolicy = 0; 3882 } 3883 } 3884 if (UseCriticalJavaThreadPriority) { 3885 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3886 } 3887 return 0; 3888} 3889 3890OSReturn os::set_native_priority(Thread* thread, int newpri) { 3891 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 3892 3893 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3894 return (ret == 0) ? OS_OK : OS_ERR; 3895} 3896 3897OSReturn os::get_native_priority(const Thread* const thread, 3898 int *priority_ptr) { 3899 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 3900 *priority_ptr = java_to_os_priority[NormPriority]; 3901 return OS_OK; 3902 } 3903 3904 errno = 0; 3905 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3906 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3907} 3908 3909// Hint to the underlying OS that a task switch would not be good. 3910// Void return because it's a hint and can fail. 3911void os::hint_no_preempt() {} 3912 3913//////////////////////////////////////////////////////////////////////////////// 3914// suspend/resume support 3915 3916// the low-level signal-based suspend/resume support is a remnant from the 3917// old VM-suspension that used to be for java-suspension, safepoints etc, 3918// within hotspot. Now there is a single use-case for this: 3919// - calling get_thread_pc() on the VMThread by the flat-profiler task 3920// that runs in the watcher thread. 3921// The remaining code is greatly simplified from the more general suspension 3922// code that used to be used. 3923// 3924// The protocol is quite simple: 3925// - suspend: 3926// - sends a signal to the target thread 3927// - polls the suspend state of the osthread using a yield loop 3928// - target thread signal handler (SR_handler) sets suspend state 3929// and blocks in sigsuspend until continued 3930// - resume: 3931// - sets target osthread state to continue 3932// - sends signal to end the sigsuspend loop in the SR_handler 3933// 3934// Note that the SR_lock plays no role in this suspend/resume protocol. 3935 3936static void resume_clear_context(OSThread *osthread) { 3937 osthread->set_ucontext(NULL); 3938 osthread->set_siginfo(NULL); 3939} 3940 3941static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 3942 ucontext_t* context) { 3943 osthread->set_ucontext(context); 3944 osthread->set_siginfo(siginfo); 3945} 3946 3947// Handler function invoked when a thread's execution is suspended or 3948// resumed. We have to be careful that only async-safe functions are 3949// called here (Note: most pthread functions are not async safe and 3950// should be avoided.) 3951// 3952// Note: sigwait() is a more natural fit than sigsuspend() from an 3953// interface point of view, but sigwait() prevents the signal hander 3954// from being run. libpthread would get very confused by not having 3955// its signal handlers run and prevents sigwait()'s use with the 3956// mutex granting granting signal. 3957// 3958// Currently only ever called on the VMThread and JavaThreads (PC sampling) 3959// 3960static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3961 // Save and restore errno to avoid confusing native code with EINTR 3962 // after sigsuspend. 3963 int old_errno = errno; 3964 3965 Thread* thread = Thread::current(); 3966 OSThread* osthread = thread->osthread(); 3967 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3968 3969 os::SuspendResume::State current = osthread->sr.state(); 3970 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3971 suspend_save_context(osthread, siginfo, context); 3972 3973 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 3974 os::SuspendResume::State state = osthread->sr.suspended(); 3975 if (state == os::SuspendResume::SR_SUSPENDED) { 3976 sigset_t suspend_set; // signals for sigsuspend() 3977 3978 // get current set of blocked signals and unblock resume signal 3979 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3980 sigdelset(&suspend_set, SR_signum); 3981 3982 sr_semaphore.signal(); 3983 // wait here until we are resumed 3984 while (1) { 3985 sigsuspend(&suspend_set); 3986 3987 os::SuspendResume::State result = osthread->sr.running(); 3988 if (result == os::SuspendResume::SR_RUNNING) { 3989 sr_semaphore.signal(); 3990 break; 3991 } 3992 } 3993 3994 } else if (state == os::SuspendResume::SR_RUNNING) { 3995 // request was cancelled, continue 3996 } else { 3997 ShouldNotReachHere(); 3998 } 3999 4000 resume_clear_context(osthread); 4001 } else if (current == os::SuspendResume::SR_RUNNING) { 4002 // request was cancelled, continue 4003 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4004 // ignore 4005 } else { 4006 // ignore 4007 } 4008 4009 errno = old_errno; 4010} 4011 4012 4013static int SR_initialize() { 4014 struct sigaction act; 4015 char *s; 4016 // Get signal number to use for suspend/resume 4017 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4018 int sig = ::strtol(s, 0, 10); 4019 if (sig > 0 || sig < _NSIG) { 4020 SR_signum = sig; 4021 } 4022 } 4023 4024 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4025 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4026 4027 sigemptyset(&SR_sigset); 4028 sigaddset(&SR_sigset, SR_signum); 4029 4030 // Set up signal handler for suspend/resume 4031 act.sa_flags = SA_RESTART|SA_SIGINFO; 4032 act.sa_handler = (void (*)(int)) SR_handler; 4033 4034 // SR_signum is blocked by default. 4035 // 4528190 - We also need to block pthread restart signal (32 on all 4036 // supported Linux platforms). Note that LinuxThreads need to block 4037 // this signal for all threads to work properly. So we don't have 4038 // to use hard-coded signal number when setting up the mask. 4039 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4040 4041 if (sigaction(SR_signum, &act, 0) == -1) { 4042 return -1; 4043 } 4044 4045 // Save signal flag 4046 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4047 return 0; 4048} 4049 4050static int sr_notify(OSThread* osthread) { 4051 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4052 assert_status(status == 0, status, "pthread_kill"); 4053 return status; 4054} 4055 4056// "Randomly" selected value for how long we want to spin 4057// before bailing out on suspending a thread, also how often 4058// we send a signal to a thread we want to resume 4059static const int RANDOMLY_LARGE_INTEGER = 1000000; 4060static const int RANDOMLY_LARGE_INTEGER2 = 100; 4061 4062// returns true on success and false on error - really an error is fatal 4063// but this seems the normal response to library errors 4064static bool do_suspend(OSThread* osthread) { 4065 assert(osthread->sr.is_running(), "thread should be running"); 4066 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4067 4068 // mark as suspended and send signal 4069 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4070 // failed to switch, state wasn't running? 4071 ShouldNotReachHere(); 4072 return false; 4073 } 4074 4075 if (sr_notify(osthread) != 0) { 4076 ShouldNotReachHere(); 4077 } 4078 4079 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4080 while (true) { 4081 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4082 break; 4083 } else { 4084 // timeout 4085 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4086 if (cancelled == os::SuspendResume::SR_RUNNING) { 4087 return false; 4088 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4089 // make sure that we consume the signal on the semaphore as well 4090 sr_semaphore.wait(); 4091 break; 4092 } else { 4093 ShouldNotReachHere(); 4094 return false; 4095 } 4096 } 4097 } 4098 4099 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4100 return true; 4101} 4102 4103static void do_resume(OSThread* osthread) { 4104 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4105 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4106 4107 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4108 // failed to switch to WAKEUP_REQUEST 4109 ShouldNotReachHere(); 4110 return; 4111 } 4112 4113 while (true) { 4114 if (sr_notify(osthread) == 0) { 4115 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4116 if (osthread->sr.is_running()) { 4117 return; 4118 } 4119 } 4120 } else { 4121 ShouldNotReachHere(); 4122 } 4123 } 4124 4125 guarantee(osthread->sr.is_running(), "Must be running!"); 4126} 4127 4128/////////////////////////////////////////////////////////////////////////////////// 4129// signal handling (except suspend/resume) 4130 4131// This routine may be used by user applications as a "hook" to catch signals. 4132// The user-defined signal handler must pass unrecognized signals to this 4133// routine, and if it returns true (non-zero), then the signal handler must 4134// return immediately. If the flag "abort_if_unrecognized" is true, then this 4135// routine will never retun false (zero), but instead will execute a VM panic 4136// routine kill the process. 4137// 4138// If this routine returns false, it is OK to call it again. This allows 4139// the user-defined signal handler to perform checks either before or after 4140// the VM performs its own checks. Naturally, the user code would be making 4141// a serious error if it tried to handle an exception (such as a null check 4142// or breakpoint) that the VM was generating for its own correct operation. 4143// 4144// This routine may recognize any of the following kinds of signals: 4145// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4146// It should be consulted by handlers for any of those signals. 4147// 4148// The caller of this routine must pass in the three arguments supplied 4149// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4150// field of the structure passed to sigaction(). This routine assumes that 4151// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4152// 4153// Note that the VM will print warnings if it detects conflicting signal 4154// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4155// 4156extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4157 siginfo_t* siginfo, 4158 void* ucontext, 4159 int abort_if_unrecognized); 4160 4161void signalHandler(int sig, siginfo_t* info, void* uc) { 4162 assert(info != NULL && uc != NULL, "it must be old kernel"); 4163 int orig_errno = errno; // Preserve errno value over signal handler. 4164 JVM_handle_linux_signal(sig, info, uc, true); 4165 errno = orig_errno; 4166} 4167 4168 4169// This boolean allows users to forward their own non-matching signals 4170// to JVM_handle_linux_signal, harmlessly. 4171bool os::Linux::signal_handlers_are_installed = false; 4172 4173// For signal-chaining 4174struct sigaction os::Linux::sigact[MAXSIGNUM]; 4175unsigned int os::Linux::sigs = 0; 4176bool os::Linux::libjsig_is_loaded = false; 4177typedef struct sigaction *(*get_signal_t)(int); 4178get_signal_t os::Linux::get_signal_action = NULL; 4179 4180struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4181 struct sigaction *actp = NULL; 4182 4183 if (libjsig_is_loaded) { 4184 // Retrieve the old signal handler from libjsig 4185 actp = (*get_signal_action)(sig); 4186 } 4187 if (actp == NULL) { 4188 // Retrieve the preinstalled signal handler from jvm 4189 actp = get_preinstalled_handler(sig); 4190 } 4191 4192 return actp; 4193} 4194 4195static bool call_chained_handler(struct sigaction *actp, int sig, 4196 siginfo_t *siginfo, void *context) { 4197 // Call the old signal handler 4198 if (actp->sa_handler == SIG_DFL) { 4199 // It's more reasonable to let jvm treat it as an unexpected exception 4200 // instead of taking the default action. 4201 return false; 4202 } else if (actp->sa_handler != SIG_IGN) { 4203 if ((actp->sa_flags & SA_NODEFER) == 0) { 4204 // automaticlly block the signal 4205 sigaddset(&(actp->sa_mask), sig); 4206 } 4207 4208 sa_handler_t hand; 4209 sa_sigaction_t sa; 4210 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4211 // retrieve the chained handler 4212 if (siginfo_flag_set) { 4213 sa = actp->sa_sigaction; 4214 } else { 4215 hand = actp->sa_handler; 4216 } 4217 4218 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4219 actp->sa_handler = SIG_DFL; 4220 } 4221 4222 // try to honor the signal mask 4223 sigset_t oset; 4224 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4225 4226 // call into the chained handler 4227 if (siginfo_flag_set) { 4228 (*sa)(sig, siginfo, context); 4229 } else { 4230 (*hand)(sig); 4231 } 4232 4233 // restore the signal mask 4234 pthread_sigmask(SIG_SETMASK, &oset, 0); 4235 } 4236 // Tell jvm's signal handler the signal is taken care of. 4237 return true; 4238} 4239 4240bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4241 bool chained = false; 4242 // signal-chaining 4243 if (UseSignalChaining) { 4244 struct sigaction *actp = get_chained_signal_action(sig); 4245 if (actp != NULL) { 4246 chained = call_chained_handler(actp, sig, siginfo, context); 4247 } 4248 } 4249 return chained; 4250} 4251 4252struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4253 if ((((unsigned int)1 << sig) & sigs) != 0) { 4254 return &sigact[sig]; 4255 } 4256 return NULL; 4257} 4258 4259void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4260 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4261 sigact[sig] = oldAct; 4262 sigs |= (unsigned int)1 << sig; 4263} 4264 4265// for diagnostic 4266int os::Linux::sigflags[MAXSIGNUM]; 4267 4268int os::Linux::get_our_sigflags(int sig) { 4269 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4270 return sigflags[sig]; 4271} 4272 4273void os::Linux::set_our_sigflags(int sig, int flags) { 4274 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4275 sigflags[sig] = flags; 4276} 4277 4278void os::Linux::set_signal_handler(int sig, bool set_installed) { 4279 // Check for overwrite. 4280 struct sigaction oldAct; 4281 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4282 4283 void* oldhand = oldAct.sa_sigaction 4284 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4285 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4286 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4287 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4288 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4289 if (AllowUserSignalHandlers || !set_installed) { 4290 // Do not overwrite; user takes responsibility to forward to us. 4291 return; 4292 } else if (UseSignalChaining) { 4293 // save the old handler in jvm 4294 save_preinstalled_handler(sig, oldAct); 4295 // libjsig also interposes the sigaction() call below and saves the 4296 // old sigaction on it own. 4297 } else { 4298 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4299 "%#lx for signal %d.", (long)oldhand, sig)); 4300 } 4301 } 4302 4303 struct sigaction sigAct; 4304 sigfillset(&(sigAct.sa_mask)); 4305 sigAct.sa_handler = SIG_DFL; 4306 if (!set_installed) { 4307 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4308 } else { 4309 sigAct.sa_sigaction = signalHandler; 4310 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4311 } 4312 // Save flags, which are set by ours 4313 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4314 sigflags[sig] = sigAct.sa_flags; 4315 4316 int ret = sigaction(sig, &sigAct, &oldAct); 4317 assert(ret == 0, "check"); 4318 4319 void* oldhand2 = oldAct.sa_sigaction 4320 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4321 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4322 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4323} 4324 4325// install signal handlers for signals that HotSpot needs to 4326// handle in order to support Java-level exception handling. 4327 4328void os::Linux::install_signal_handlers() { 4329 if (!signal_handlers_are_installed) { 4330 signal_handlers_are_installed = true; 4331 4332 // signal-chaining 4333 typedef void (*signal_setting_t)(); 4334 signal_setting_t begin_signal_setting = NULL; 4335 signal_setting_t end_signal_setting = NULL; 4336 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4337 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4338 if (begin_signal_setting != NULL) { 4339 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4340 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4341 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4342 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4343 libjsig_is_loaded = true; 4344 assert(UseSignalChaining, "should enable signal-chaining"); 4345 } 4346 if (libjsig_is_loaded) { 4347 // Tell libjsig jvm is setting signal handlers 4348 (*begin_signal_setting)(); 4349 } 4350 4351 set_signal_handler(SIGSEGV, true); 4352 set_signal_handler(SIGPIPE, true); 4353 set_signal_handler(SIGBUS, true); 4354 set_signal_handler(SIGILL, true); 4355 set_signal_handler(SIGFPE, true); 4356#if defined(PPC64) 4357 set_signal_handler(SIGTRAP, true); 4358#endif 4359 set_signal_handler(SIGXFSZ, true); 4360 4361 if (libjsig_is_loaded) { 4362 // Tell libjsig jvm finishes setting signal handlers 4363 (*end_signal_setting)(); 4364 } 4365 4366 // We don't activate signal checker if libjsig is in place, we trust ourselves 4367 // and if UserSignalHandler is installed all bets are off. 4368 // Log that signal checking is off only if -verbose:jni is specified. 4369 if (CheckJNICalls) { 4370 if (libjsig_is_loaded) { 4371 if (PrintJNIResolving) { 4372 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4373 } 4374 check_signals = false; 4375 } 4376 if (AllowUserSignalHandlers) { 4377 if (PrintJNIResolving) { 4378 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4379 } 4380 check_signals = false; 4381 } 4382 } 4383 } 4384} 4385 4386// This is the fastest way to get thread cpu time on Linux. 4387// Returns cpu time (user+sys) for any thread, not only for current. 4388// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4389// It might work on 2.6.10+ with a special kernel/glibc patch. 4390// For reference, please, see IEEE Std 1003.1-2004: 4391// http://www.unix.org/single_unix_specification 4392 4393jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4394 struct timespec tp; 4395 int rc = os::Linux::clock_gettime(clockid, &tp); 4396 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4397 4398 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4399} 4400 4401///// 4402// glibc on Linux platform uses non-documented flag 4403// to indicate, that some special sort of signal 4404// trampoline is used. 4405// We will never set this flag, and we should 4406// ignore this flag in our diagnostic 4407#ifdef SIGNIFICANT_SIGNAL_MASK 4408 #undef SIGNIFICANT_SIGNAL_MASK 4409#endif 4410#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4411 4412static const char* get_signal_handler_name(address handler, 4413 char* buf, int buflen) { 4414 int offset; 4415 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4416 if (found) { 4417 // skip directory names 4418 const char *p1, *p2; 4419 p1 = buf; 4420 size_t len = strlen(os::file_separator()); 4421 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4422 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4423 } else { 4424 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4425 } 4426 return buf; 4427} 4428 4429static void print_signal_handler(outputStream* st, int sig, 4430 char* buf, size_t buflen) { 4431 struct sigaction sa; 4432 4433 sigaction(sig, NULL, &sa); 4434 4435 // See comment for SIGNIFICANT_SIGNAL_MASK define 4436 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4437 4438 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4439 4440 address handler = (sa.sa_flags & SA_SIGINFO) 4441 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4442 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4443 4444 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4445 st->print("SIG_DFL"); 4446 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4447 st->print("SIG_IGN"); 4448 } else { 4449 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4450 } 4451 4452 st->print(", sa_mask[0]="); 4453 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4454 4455 address rh = VMError::get_resetted_sighandler(sig); 4456 // May be, handler was resetted by VMError? 4457 if (rh != NULL) { 4458 handler = rh; 4459 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4460 } 4461 4462 st->print(", sa_flags="); 4463 os::Posix::print_sa_flags(st, sa.sa_flags); 4464 4465 // Check: is it our handler? 4466 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4467 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4468 // It is our signal handler 4469 // check for flags, reset system-used one! 4470 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4471 st->print( 4472 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4473 os::Linux::get_our_sigflags(sig)); 4474 } 4475 } 4476 st->cr(); 4477} 4478 4479 4480#define DO_SIGNAL_CHECK(sig) \ 4481 do { \ 4482 if (!sigismember(&check_signal_done, sig)) { \ 4483 os::Linux::check_signal_handler(sig); \ 4484 } \ 4485 } while (0) 4486 4487// This method is a periodic task to check for misbehaving JNI applications 4488// under CheckJNI, we can add any periodic checks here 4489 4490void os::run_periodic_checks() { 4491 if (check_signals == false) return; 4492 4493 // SEGV and BUS if overridden could potentially prevent 4494 // generation of hs*.log in the event of a crash, debugging 4495 // such a case can be very challenging, so we absolutely 4496 // check the following for a good measure: 4497 DO_SIGNAL_CHECK(SIGSEGV); 4498 DO_SIGNAL_CHECK(SIGILL); 4499 DO_SIGNAL_CHECK(SIGFPE); 4500 DO_SIGNAL_CHECK(SIGBUS); 4501 DO_SIGNAL_CHECK(SIGPIPE); 4502 DO_SIGNAL_CHECK(SIGXFSZ); 4503#if defined(PPC64) 4504 DO_SIGNAL_CHECK(SIGTRAP); 4505#endif 4506 4507 // ReduceSignalUsage allows the user to override these handlers 4508 // see comments at the very top and jvm_solaris.h 4509 if (!ReduceSignalUsage) { 4510 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4511 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4512 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4513 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4514 } 4515 4516 DO_SIGNAL_CHECK(SR_signum); 4517 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4518} 4519 4520typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4521 4522static os_sigaction_t os_sigaction = NULL; 4523 4524void os::Linux::check_signal_handler(int sig) { 4525 char buf[O_BUFLEN]; 4526 address jvmHandler = NULL; 4527 4528 4529 struct sigaction act; 4530 if (os_sigaction == NULL) { 4531 // only trust the default sigaction, in case it has been interposed 4532 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4533 if (os_sigaction == NULL) return; 4534 } 4535 4536 os_sigaction(sig, (struct sigaction*)NULL, &act); 4537 4538 4539 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4540 4541 address thisHandler = (act.sa_flags & SA_SIGINFO) 4542 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4543 : CAST_FROM_FN_PTR(address, act.sa_handler); 4544 4545 4546 switch (sig) { 4547 case SIGSEGV: 4548 case SIGBUS: 4549 case SIGFPE: 4550 case SIGPIPE: 4551 case SIGILL: 4552 case SIGXFSZ: 4553 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4554 break; 4555 4556 case SHUTDOWN1_SIGNAL: 4557 case SHUTDOWN2_SIGNAL: 4558 case SHUTDOWN3_SIGNAL: 4559 case BREAK_SIGNAL: 4560 jvmHandler = (address)user_handler(); 4561 break; 4562 4563 case INTERRUPT_SIGNAL: 4564 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4565 break; 4566 4567 default: 4568 if (sig == SR_signum) { 4569 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4570 } else { 4571 return; 4572 } 4573 break; 4574 } 4575 4576 if (thisHandler != jvmHandler) { 4577 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4578 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4579 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4580 // No need to check this sig any longer 4581 sigaddset(&check_signal_done, sig); 4582 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4583 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4584 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4585 exception_name(sig, buf, O_BUFLEN)); 4586 } 4587 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4588 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4589 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4590 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4591 // No need to check this sig any longer 4592 sigaddset(&check_signal_done, sig); 4593 } 4594 4595 // Dump all the signal 4596 if (sigismember(&check_signal_done, sig)) { 4597 print_signal_handlers(tty, buf, O_BUFLEN); 4598 } 4599} 4600 4601extern void report_error(char* file_name, int line_no, char* title, 4602 char* format, ...); 4603 4604extern bool signal_name(int signo, char* buf, size_t len); 4605 4606const char* os::exception_name(int exception_code, char* buf, size_t size) { 4607 if (0 < exception_code && exception_code <= SIGRTMAX) { 4608 // signal 4609 if (!signal_name(exception_code, buf, size)) { 4610 jio_snprintf(buf, size, "SIG%d", exception_code); 4611 } 4612 return buf; 4613 } else { 4614 return NULL; 4615 } 4616} 4617 4618// this is called _before_ the most of global arguments have been parsed 4619void os::init(void) { 4620 char dummy; // used to get a guess on initial stack address 4621// first_hrtime = gethrtime(); 4622 4623 // With LinuxThreads the JavaMain thread pid (primordial thread) 4624 // is different than the pid of the java launcher thread. 4625 // So, on Linux, the launcher thread pid is passed to the VM 4626 // via the sun.java.launcher.pid property. 4627 // Use this property instead of getpid() if it was correctly passed. 4628 // See bug 6351349. 4629 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4630 4631 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4632 4633 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4634 4635 init_random(1234567); 4636 4637 ThreadCritical::initialize(); 4638 4639 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4640 if (Linux::page_size() == -1) { 4641 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4642 strerror(errno))); 4643 } 4644 init_page_sizes((size_t) Linux::page_size()); 4645 4646 Linux::initialize_system_info(); 4647 4648 // main_thread points to the aboriginal thread 4649 Linux::_main_thread = pthread_self(); 4650 4651 Linux::clock_init(); 4652 initial_time_count = javaTimeNanos(); 4653 4654 // pthread_condattr initialization for monotonic clock 4655 int status; 4656 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4657 if ((status = pthread_condattr_init(_condattr)) != 0) { 4658 fatal(err_msg("pthread_condattr_init: %s", strerror(status))); 4659 } 4660 // Only set the clock if CLOCK_MONOTONIC is available 4661 if (os::supports_monotonic_clock()) { 4662 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4663 if (status == EINVAL) { 4664 warning("Unable to use monotonic clock with relative timed-waits" \ 4665 " - changes to the time-of-day clock may have adverse affects"); 4666 } else { 4667 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status))); 4668 } 4669 } 4670 } 4671 // else it defaults to CLOCK_REALTIME 4672 4673 pthread_mutex_init(&dl_mutex, NULL); 4674 4675 // If the pagesize of the VM is greater than 8K determine the appropriate 4676 // number of initial guard pages. The user can change this with the 4677 // command line arguments, if needed. 4678 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4679 StackYellowPages = 1; 4680 StackRedPages = 1; 4681 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4682 } 4683 4684 // retrieve entry point for pthread_setname_np 4685 Linux::_pthread_setname_np = 4686 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4687 4688} 4689 4690// To install functions for atexit system call 4691extern "C" { 4692 static void perfMemory_exit_helper() { 4693 perfMemory_exit(); 4694 } 4695} 4696 4697// this is called _after_ the global arguments have been parsed 4698jint os::init_2(void) { 4699 Linux::fast_thread_clock_init(); 4700 4701 // Allocate a single page and mark it as readable for safepoint polling 4702 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4703 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page"); 4704 4705 os::set_polling_page(polling_page); 4706 4707#ifndef PRODUCT 4708 if (Verbose && PrintMiscellaneous) { 4709 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", 4710 (intptr_t)polling_page); 4711 } 4712#endif 4713 4714 if (!UseMembar) { 4715 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4716 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4717 os::set_memory_serialize_page(mem_serialize_page); 4718 4719#ifndef PRODUCT 4720 if (Verbose && PrintMiscellaneous) { 4721 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", 4722 (intptr_t)mem_serialize_page); 4723 } 4724#endif 4725 } 4726 4727 // initialize suspend/resume support - must do this before signal_sets_init() 4728 if (SR_initialize() != 0) { 4729 perror("SR_initialize failed"); 4730 return JNI_ERR; 4731 } 4732 4733 Linux::signal_sets_init(); 4734 Linux::install_signal_handlers(); 4735 4736 // Check minimum allowable stack size for thread creation and to initialize 4737 // the java system classes, including StackOverflowError - depends on page 4738 // size. Add a page for compiler2 recursion in main thread. 4739 // Add in 2*BytesPerWord times page size to account for VM stack during 4740 // class initialization depending on 32 or 64 bit VM. 4741 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4742 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4743 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4744 4745 size_t threadStackSizeInBytes = ThreadStackSize * K; 4746 if (threadStackSizeInBytes != 0 && 4747 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4748 tty->print_cr("\nThe stack size specified is too small, " 4749 "Specify at least %dk", 4750 os::Linux::min_stack_allowed/ K); 4751 return JNI_ERR; 4752 } 4753 4754 // Make the stack size a multiple of the page size so that 4755 // the yellow/red zones can be guarded. 4756 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4757 vm_page_size())); 4758 4759 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4760 4761#if defined(IA32) 4762 workaround_expand_exec_shield_cs_limit(); 4763#endif 4764 4765 Linux::libpthread_init(); 4766 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4767 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4768 Linux::glibc_version(), Linux::libpthread_version(), 4769 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4770 } 4771 4772 if (UseNUMA) { 4773 if (!Linux::libnuma_init()) { 4774 UseNUMA = false; 4775 } else { 4776 if ((Linux::numa_max_node() < 1)) { 4777 // There's only one node(they start from 0), disable NUMA. 4778 UseNUMA = false; 4779 } 4780 } 4781 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4782 // we can make the adaptive lgrp chunk resizing work. If the user specified 4783 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4784 // disable adaptive resizing. 4785 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4786 if (FLAG_IS_DEFAULT(UseNUMA)) { 4787 UseNUMA = false; 4788 } else { 4789 if (FLAG_IS_DEFAULT(UseLargePages) && 4790 FLAG_IS_DEFAULT(UseSHM) && 4791 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4792 UseLargePages = false; 4793 } else { 4794 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing"); 4795 UseAdaptiveSizePolicy = false; 4796 UseAdaptiveNUMAChunkSizing = false; 4797 } 4798 } 4799 } 4800 if (!UseNUMA && ForceNUMA) { 4801 UseNUMA = true; 4802 } 4803 } 4804 4805 if (MaxFDLimit) { 4806 // set the number of file descriptors to max. print out error 4807 // if getrlimit/setrlimit fails but continue regardless. 4808 struct rlimit nbr_files; 4809 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4810 if (status != 0) { 4811 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4812 perror("os::init_2 getrlimit failed"); 4813 } 4814 } else { 4815 nbr_files.rlim_cur = nbr_files.rlim_max; 4816 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4817 if (status != 0) { 4818 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4819 perror("os::init_2 setrlimit failed"); 4820 } 4821 } 4822 } 4823 } 4824 4825 // Initialize lock used to serialize thread creation (see os::create_thread) 4826 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4827 4828 // at-exit methods are called in the reverse order of their registration. 4829 // atexit functions are called on return from main or as a result of a 4830 // call to exit(3C). There can be only 32 of these functions registered 4831 // and atexit() does not set errno. 4832 4833 if (PerfAllowAtExitRegistration) { 4834 // only register atexit functions if PerfAllowAtExitRegistration is set. 4835 // atexit functions can be delayed until process exit time, which 4836 // can be problematic for embedded VM situations. Embedded VMs should 4837 // call DestroyJavaVM() to assure that VM resources are released. 4838 4839 // note: perfMemory_exit_helper atexit function may be removed in 4840 // the future if the appropriate cleanup code can be added to the 4841 // VM_Exit VMOperation's doit method. 4842 if (atexit(perfMemory_exit_helper) != 0) { 4843 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4844 } 4845 } 4846 4847 // initialize thread priority policy 4848 prio_init(); 4849 4850 return JNI_OK; 4851} 4852 4853// Mark the polling page as unreadable 4854void os::make_polling_page_unreadable(void) { 4855 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 4856 fatal("Could not disable polling page"); 4857 } 4858} 4859 4860// Mark the polling page as readable 4861void os::make_polling_page_readable(void) { 4862 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4863 fatal("Could not enable polling page"); 4864 } 4865} 4866 4867int os::active_processor_count() { 4868 // Linux doesn't yet have a (official) notion of processor sets, 4869 // so just return the number of online processors. 4870 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4871 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4872 return online_cpus; 4873} 4874 4875void os::set_native_thread_name(const char *name) { 4876 if (Linux::_pthread_setname_np) { 4877 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 4878 snprintf(buf, sizeof(buf), "%s", name); 4879 buf[sizeof(buf) - 1] = '\0'; 4880 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 4881 // ERANGE should not happen; all other errors should just be ignored. 4882 assert(rc != ERANGE, "pthread_setname_np failed"); 4883 } 4884} 4885 4886bool os::distribute_processes(uint length, uint* distribution) { 4887 // Not yet implemented. 4888 return false; 4889} 4890 4891bool os::bind_to_processor(uint processor_id) { 4892 // Not yet implemented. 4893 return false; 4894} 4895 4896/// 4897 4898void os::SuspendedThreadTask::internal_do_task() { 4899 if (do_suspend(_thread->osthread())) { 4900 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4901 do_task(context); 4902 do_resume(_thread->osthread()); 4903 } 4904} 4905 4906class PcFetcher : public os::SuspendedThreadTask { 4907 public: 4908 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4909 ExtendedPC result(); 4910 protected: 4911 void do_task(const os::SuspendedThreadTaskContext& context); 4912 private: 4913 ExtendedPC _epc; 4914}; 4915 4916ExtendedPC PcFetcher::result() { 4917 guarantee(is_done(), "task is not done yet."); 4918 return _epc; 4919} 4920 4921void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4922 Thread* thread = context.thread(); 4923 OSThread* osthread = thread->osthread(); 4924 if (osthread->ucontext() != NULL) { 4925 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4926 } else { 4927 // NULL context is unexpected, double-check this is the VMThread 4928 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4929 } 4930} 4931 4932// Suspends the target using the signal mechanism and then grabs the PC before 4933// resuming the target. Used by the flat-profiler only 4934ExtendedPC os::get_thread_pc(Thread* thread) { 4935 // Make sure that it is called by the watcher for the VMThread 4936 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4937 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4938 4939 PcFetcher fetcher(thread); 4940 fetcher.run(); 4941 return fetcher.result(); 4942} 4943 4944int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, 4945 pthread_mutex_t *_mutex, 4946 const struct timespec *_abstime) { 4947 if (is_NPTL()) { 4948 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4949 } else { 4950 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4951 // word back to default 64bit precision if condvar is signaled. Java 4952 // wants 53bit precision. Save and restore current value. 4953 int fpu = get_fpu_control_word(); 4954 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4955 set_fpu_control_word(fpu); 4956 return status; 4957 } 4958} 4959 4960//////////////////////////////////////////////////////////////////////////////// 4961// debug support 4962 4963bool os::find(address addr, outputStream* st) { 4964 Dl_info dlinfo; 4965 memset(&dlinfo, 0, sizeof(dlinfo)); 4966 if (dladdr(addr, &dlinfo) != 0) { 4967 st->print(PTR_FORMAT ": ", addr); 4968 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 4969 st->print("%s+%#x", dlinfo.dli_sname, 4970 addr - (intptr_t)dlinfo.dli_saddr); 4971 } else if (dlinfo.dli_fbase != NULL) { 4972 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4973 } else { 4974 st->print("<absolute address>"); 4975 } 4976 if (dlinfo.dli_fname != NULL) { 4977 st->print(" in %s", dlinfo.dli_fname); 4978 } 4979 if (dlinfo.dli_fbase != NULL) { 4980 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4981 } 4982 st->cr(); 4983 4984 if (Verbose) { 4985 // decode some bytes around the PC 4986 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 4987 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 4988 address lowest = (address) dlinfo.dli_sname; 4989 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4990 if (begin < lowest) begin = lowest; 4991 Dl_info dlinfo2; 4992 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 4993 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 4994 end = (address) dlinfo2.dli_saddr; 4995 } 4996 Disassembler::decode(begin, end, st); 4997 } 4998 return true; 4999 } 5000 return false; 5001} 5002 5003//////////////////////////////////////////////////////////////////////////////// 5004// misc 5005 5006// This does not do anything on Linux. This is basically a hook for being 5007// able to use structured exception handling (thread-local exception filters) 5008// on, e.g., Win32. 5009void 5010os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 5011 JavaCallArguments* args, Thread* thread) { 5012 f(value, method, args, thread); 5013} 5014 5015void os::print_statistics() { 5016} 5017 5018int os::message_box(const char* title, const char* message) { 5019 int i; 5020 fdStream err(defaultStream::error_fd()); 5021 for (i = 0; i < 78; i++) err.print_raw("="); 5022 err.cr(); 5023 err.print_raw_cr(title); 5024 for (i = 0; i < 78; i++) err.print_raw("-"); 5025 err.cr(); 5026 err.print_raw_cr(message); 5027 for (i = 0; i < 78; i++) err.print_raw("="); 5028 err.cr(); 5029 5030 char buf[16]; 5031 // Prevent process from exiting upon "read error" without consuming all CPU 5032 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5033 5034 return buf[0] == 'y' || buf[0] == 'Y'; 5035} 5036 5037int os::stat(const char *path, struct stat *sbuf) { 5038 char pathbuf[MAX_PATH]; 5039 if (strlen(path) > MAX_PATH - 1) { 5040 errno = ENAMETOOLONG; 5041 return -1; 5042 } 5043 os::native_path(strcpy(pathbuf, path)); 5044 return ::stat(pathbuf, sbuf); 5045} 5046 5047bool os::check_heap(bool force) { 5048 return true; 5049} 5050 5051// Is a (classpath) directory empty? 5052bool os::dir_is_empty(const char* path) { 5053 DIR *dir = NULL; 5054 struct dirent *ptr; 5055 5056 dir = opendir(path); 5057 if (dir == NULL) return true; 5058 5059 // Scan the directory 5060 bool result = true; 5061 char buf[sizeof(struct dirent) + MAX_PATH]; 5062 while (result && (ptr = ::readdir(dir)) != NULL) { 5063 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5064 result = false; 5065 } 5066 } 5067 closedir(dir); 5068 return result; 5069} 5070 5071// This code originates from JDK's sysOpen and open64_w 5072// from src/solaris/hpi/src/system_md.c 5073 5074int os::open(const char *path, int oflag, int mode) { 5075 if (strlen(path) > MAX_PATH - 1) { 5076 errno = ENAMETOOLONG; 5077 return -1; 5078 } 5079 5080 // All file descriptors that are opened in the Java process and not 5081 // specifically destined for a subprocess should have the close-on-exec 5082 // flag set. If we don't set it, then careless 3rd party native code 5083 // might fork and exec without closing all appropriate file descriptors 5084 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5085 // turn might: 5086 // 5087 // - cause end-of-file to fail to be detected on some file 5088 // descriptors, resulting in mysterious hangs, or 5089 // 5090 // - might cause an fopen in the subprocess to fail on a system 5091 // suffering from bug 1085341. 5092 // 5093 // (Yes, the default setting of the close-on-exec flag is a Unix 5094 // design flaw) 5095 // 5096 // See: 5097 // 1085341: 32-bit stdio routines should support file descriptors >255 5098 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5099 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5100 // 5101 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5102 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5103 // because it saves a system call and removes a small window where the flag 5104 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5105 // and we fall back to using FD_CLOEXEC (see below). 5106#ifdef O_CLOEXEC 5107 oflag |= O_CLOEXEC; 5108#endif 5109 5110 int fd = ::open64(path, oflag, mode); 5111 if (fd == -1) return -1; 5112 5113 //If the open succeeded, the file might still be a directory 5114 { 5115 struct stat64 buf64; 5116 int ret = ::fstat64(fd, &buf64); 5117 int st_mode = buf64.st_mode; 5118 5119 if (ret != -1) { 5120 if ((st_mode & S_IFMT) == S_IFDIR) { 5121 errno = EISDIR; 5122 ::close(fd); 5123 return -1; 5124 } 5125 } else { 5126 ::close(fd); 5127 return -1; 5128 } 5129 } 5130 5131#ifdef FD_CLOEXEC 5132 // Validate that the use of the O_CLOEXEC flag on open above worked. 5133 // With recent kernels, we will perform this check exactly once. 5134 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5135 if (!O_CLOEXEC_is_known_to_work) { 5136 int flags = ::fcntl(fd, F_GETFD); 5137 if (flags != -1) { 5138 if ((flags & FD_CLOEXEC) != 0) 5139 O_CLOEXEC_is_known_to_work = 1; 5140 else 5141 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5142 } 5143 } 5144#endif 5145 5146 return fd; 5147} 5148 5149 5150// create binary file, rewriting existing file if required 5151int os::create_binary_file(const char* path, bool rewrite_existing) { 5152 int oflags = O_WRONLY | O_CREAT; 5153 if (!rewrite_existing) { 5154 oflags |= O_EXCL; 5155 } 5156 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5157} 5158 5159// return current position of file pointer 5160jlong os::current_file_offset(int fd) { 5161 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5162} 5163 5164// move file pointer to the specified offset 5165jlong os::seek_to_file_offset(int fd, jlong offset) { 5166 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5167} 5168 5169// This code originates from JDK's sysAvailable 5170// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5171 5172int os::available(int fd, jlong *bytes) { 5173 jlong cur, end; 5174 int mode; 5175 struct stat64 buf64; 5176 5177 if (::fstat64(fd, &buf64) >= 0) { 5178 mode = buf64.st_mode; 5179 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5180 // XXX: is the following call interruptible? If so, this might 5181 // need to go through the INTERRUPT_IO() wrapper as for other 5182 // blocking, interruptible calls in this file. 5183 int n; 5184 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5185 *bytes = n; 5186 return 1; 5187 } 5188 } 5189 } 5190 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5191 return 0; 5192 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5193 return 0; 5194 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5195 return 0; 5196 } 5197 *bytes = end - cur; 5198 return 1; 5199} 5200 5201// Map a block of memory. 5202char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5203 char *addr, size_t bytes, bool read_only, 5204 bool allow_exec) { 5205 int prot; 5206 int flags = MAP_PRIVATE; 5207 5208 if (read_only) { 5209 prot = PROT_READ; 5210 } else { 5211 prot = PROT_READ | PROT_WRITE; 5212 } 5213 5214 if (allow_exec) { 5215 prot |= PROT_EXEC; 5216 } 5217 5218 if (addr != NULL) { 5219 flags |= MAP_FIXED; 5220 } 5221 5222 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5223 fd, file_offset); 5224 if (mapped_address == MAP_FAILED) { 5225 return NULL; 5226 } 5227 return mapped_address; 5228} 5229 5230 5231// Remap a block of memory. 5232char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5233 char *addr, size_t bytes, bool read_only, 5234 bool allow_exec) { 5235 // same as map_memory() on this OS 5236 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5237 allow_exec); 5238} 5239 5240 5241// Unmap a block of memory. 5242bool os::pd_unmap_memory(char* addr, size_t bytes) { 5243 return munmap(addr, bytes) == 0; 5244} 5245 5246static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5247 5248static clockid_t thread_cpu_clockid(Thread* thread) { 5249 pthread_t tid = thread->osthread()->pthread_id(); 5250 clockid_t clockid; 5251 5252 // Get thread clockid 5253 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5254 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5255 return clockid; 5256} 5257 5258// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5259// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5260// of a thread. 5261// 5262// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5263// the fast estimate available on the platform. 5264 5265jlong os::current_thread_cpu_time() { 5266 if (os::Linux::supports_fast_thread_cpu_time()) { 5267 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5268 } else { 5269 // return user + sys since the cost is the same 5270 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5271 } 5272} 5273 5274jlong os::thread_cpu_time(Thread* thread) { 5275 // consistent with what current_thread_cpu_time() returns 5276 if (os::Linux::supports_fast_thread_cpu_time()) { 5277 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5278 } else { 5279 return slow_thread_cpu_time(thread, true /* user + sys */); 5280 } 5281} 5282 5283jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5284 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5285 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5286 } else { 5287 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5288 } 5289} 5290 5291jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5292 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5293 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5294 } else { 5295 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5296 } 5297} 5298 5299// -1 on error. 5300static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5301 pid_t tid = thread->osthread()->thread_id(); 5302 char *s; 5303 char stat[2048]; 5304 int statlen; 5305 char proc_name[64]; 5306 int count; 5307 long sys_time, user_time; 5308 char cdummy; 5309 int idummy; 5310 long ldummy; 5311 FILE *fp; 5312 5313 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5314 fp = fopen(proc_name, "r"); 5315 if (fp == NULL) return -1; 5316 statlen = fread(stat, 1, 2047, fp); 5317 stat[statlen] = '\0'; 5318 fclose(fp); 5319 5320 // Skip pid and the command string. Note that we could be dealing with 5321 // weird command names, e.g. user could decide to rename java launcher 5322 // to "java 1.4.2 :)", then the stat file would look like 5323 // 1234 (java 1.4.2 :)) R ... ... 5324 // We don't really need to know the command string, just find the last 5325 // occurrence of ")" and then start parsing from there. See bug 4726580. 5326 s = strrchr(stat, ')'); 5327 if (s == NULL) return -1; 5328 5329 // Skip blank chars 5330 do { s++; } while (s && isspace(*s)); 5331 5332 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5333 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5334 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5335 &user_time, &sys_time); 5336 if (count != 13) return -1; 5337 if (user_sys_cpu_time) { 5338 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5339 } else { 5340 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5341 } 5342} 5343 5344void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5345 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5346 info_ptr->may_skip_backward = false; // elapsed time not wall time 5347 info_ptr->may_skip_forward = false; // elapsed time not wall time 5348 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5349} 5350 5351void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5352 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5353 info_ptr->may_skip_backward = false; // elapsed time not wall time 5354 info_ptr->may_skip_forward = false; // elapsed time not wall time 5355 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5356} 5357 5358bool os::is_thread_cpu_time_supported() { 5359 return true; 5360} 5361 5362// System loadavg support. Returns -1 if load average cannot be obtained. 5363// Linux doesn't yet have a (official) notion of processor sets, 5364// so just return the system wide load average. 5365int os::loadavg(double loadavg[], int nelem) { 5366 return ::getloadavg(loadavg, nelem); 5367} 5368 5369void os::pause() { 5370 char filename[MAX_PATH]; 5371 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5372 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5373 } else { 5374 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5375 } 5376 5377 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5378 if (fd != -1) { 5379 struct stat buf; 5380 ::close(fd); 5381 while (::stat(filename, &buf) == 0) { 5382 (void)::poll(NULL, 0, 100); 5383 } 5384 } else { 5385 jio_fprintf(stderr, 5386 "Could not open pause file '%s', continuing immediately.\n", filename); 5387 } 5388} 5389 5390 5391// Refer to the comments in os_solaris.cpp park-unpark. The next two 5392// comment paragraphs are worth repeating here: 5393// 5394// Assumption: 5395// Only one parker can exist on an event, which is why we allocate 5396// them per-thread. Multiple unparkers can coexist. 5397// 5398// _Event serves as a restricted-range semaphore. 5399// -1 : thread is blocked, i.e. there is a waiter 5400// 0 : neutral: thread is running or ready, 5401// could have been signaled after a wait started 5402// 1 : signaled - thread is running or ready 5403// 5404// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5405// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5406// For specifics regarding the bug see GLIBC BUGID 261237 : 5407// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5408// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5409// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5410// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5411// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5412// and monitorenter when we're using 1-0 locking. All those operations may result in 5413// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5414// of libpthread avoids the problem, but isn't practical. 5415// 5416// Possible remedies: 5417// 5418// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5419// This is palliative and probabilistic, however. If the thread is preempted 5420// between the call to compute_abstime() and pthread_cond_timedwait(), more 5421// than the minimum period may have passed, and the abstime may be stale (in the 5422// past) resultin in a hang. Using this technique reduces the odds of a hang 5423// but the JVM is still vulnerable, particularly on heavily loaded systems. 5424// 5425// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5426// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5427// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5428// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5429// thread. 5430// 5431// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5432// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5433// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5434// This also works well. In fact it avoids kernel-level scalability impediments 5435// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5436// timers in a graceful fashion. 5437// 5438// 4. When the abstime value is in the past it appears that control returns 5439// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5440// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5441// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5442// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5443// It may be possible to avoid reinitialization by checking the return 5444// value from pthread_cond_timedwait(). In addition to reinitializing the 5445// condvar we must establish the invariant that cond_signal() is only called 5446// within critical sections protected by the adjunct mutex. This prevents 5447// cond_signal() from "seeing" a condvar that's in the midst of being 5448// reinitialized or that is corrupt. Sadly, this invariant obviates the 5449// desirable signal-after-unlock optimization that avoids futile context switching. 5450// 5451// I'm also concerned that some versions of NTPL might allocate an auxilliary 5452// structure when a condvar is used or initialized. cond_destroy() would 5453// release the helper structure. Our reinitialize-after-timedwait fix 5454// put excessive stress on malloc/free and locks protecting the c-heap. 5455// 5456// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5457// It may be possible to refine (4) by checking the kernel and NTPL verisons 5458// and only enabling the work-around for vulnerable environments. 5459 5460// utility to compute the abstime argument to timedwait: 5461// millis is the relative timeout time 5462// abstime will be the absolute timeout time 5463// TODO: replace compute_abstime() with unpackTime() 5464 5465static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5466 if (millis < 0) millis = 0; 5467 5468 jlong seconds = millis / 1000; 5469 millis %= 1000; 5470 if (seconds > 50000000) { // see man cond_timedwait(3T) 5471 seconds = 50000000; 5472 } 5473 5474 if (os::supports_monotonic_clock()) { 5475 struct timespec now; 5476 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5477 assert_status(status == 0, status, "clock_gettime"); 5478 abstime->tv_sec = now.tv_sec + seconds; 5479 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5480 if (nanos >= NANOSECS_PER_SEC) { 5481 abstime->tv_sec += 1; 5482 nanos -= NANOSECS_PER_SEC; 5483 } 5484 abstime->tv_nsec = nanos; 5485 } else { 5486 struct timeval now; 5487 int status = gettimeofday(&now, NULL); 5488 assert(status == 0, "gettimeofday"); 5489 abstime->tv_sec = now.tv_sec + seconds; 5490 long usec = now.tv_usec + millis * 1000; 5491 if (usec >= 1000000) { 5492 abstime->tv_sec += 1; 5493 usec -= 1000000; 5494 } 5495 abstime->tv_nsec = usec * 1000; 5496 } 5497 return abstime; 5498} 5499 5500void os::PlatformEvent::park() { // AKA "down()" 5501 // Transitions for _Event: 5502 // -1 => -1 : illegal 5503 // 1 => 0 : pass - return immediately 5504 // 0 => -1 : block; then set _Event to 0 before returning 5505 5506 // Invariant: Only the thread associated with the Event/PlatformEvent 5507 // may call park(). 5508 // TODO: assert that _Assoc != NULL or _Assoc == Self 5509 assert(_nParked == 0, "invariant"); 5510 5511 int v; 5512 for (;;) { 5513 v = _Event; 5514 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5515 } 5516 guarantee(v >= 0, "invariant"); 5517 if (v == 0) { 5518 // Do this the hard way by blocking ... 5519 int status = pthread_mutex_lock(_mutex); 5520 assert_status(status == 0, status, "mutex_lock"); 5521 guarantee(_nParked == 0, "invariant"); 5522 ++_nParked; 5523 while (_Event < 0) { 5524 status = pthread_cond_wait(_cond, _mutex); 5525 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5526 // Treat this the same as if the wait was interrupted 5527 if (status == ETIME) { status = EINTR; } 5528 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5529 } 5530 --_nParked; 5531 5532 _Event = 0; 5533 status = pthread_mutex_unlock(_mutex); 5534 assert_status(status == 0, status, "mutex_unlock"); 5535 // Paranoia to ensure our locked and lock-free paths interact 5536 // correctly with each other. 5537 OrderAccess::fence(); 5538 } 5539 guarantee(_Event >= 0, "invariant"); 5540} 5541 5542int os::PlatformEvent::park(jlong millis) { 5543 // Transitions for _Event: 5544 // -1 => -1 : illegal 5545 // 1 => 0 : pass - return immediately 5546 // 0 => -1 : block; then set _Event to 0 before returning 5547 5548 guarantee(_nParked == 0, "invariant"); 5549 5550 int v; 5551 for (;;) { 5552 v = _Event; 5553 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5554 } 5555 guarantee(v >= 0, "invariant"); 5556 if (v != 0) return OS_OK; 5557 5558 // We do this the hard way, by blocking the thread. 5559 // Consider enforcing a minimum timeout value. 5560 struct timespec abst; 5561 compute_abstime(&abst, millis); 5562 5563 int ret = OS_TIMEOUT; 5564 int status = pthread_mutex_lock(_mutex); 5565 assert_status(status == 0, status, "mutex_lock"); 5566 guarantee(_nParked == 0, "invariant"); 5567 ++_nParked; 5568 5569 // Object.wait(timo) will return because of 5570 // (a) notification 5571 // (b) timeout 5572 // (c) thread.interrupt 5573 // 5574 // Thread.interrupt and object.notify{All} both call Event::set. 5575 // That is, we treat thread.interrupt as a special case of notification. 5576 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false. 5577 // We assume all ETIME returns are valid. 5578 // 5579 // TODO: properly differentiate simultaneous notify+interrupt. 5580 // In that case, we should propagate the notify to another waiter. 5581 5582 while (_Event < 0) { 5583 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5584 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5585 pthread_cond_destroy(_cond); 5586 pthread_cond_init(_cond, os::Linux::condAttr()); 5587 } 5588 assert_status(status == 0 || status == EINTR || 5589 status == ETIME || status == ETIMEDOUT, 5590 status, "cond_timedwait"); 5591 if (!FilterSpuriousWakeups) break; // previous semantics 5592 if (status == ETIME || status == ETIMEDOUT) break; 5593 // We consume and ignore EINTR and spurious wakeups. 5594 } 5595 --_nParked; 5596 if (_Event >= 0) { 5597 ret = OS_OK; 5598 } 5599 _Event = 0; 5600 status = pthread_mutex_unlock(_mutex); 5601 assert_status(status == 0, status, "mutex_unlock"); 5602 assert(_nParked == 0, "invariant"); 5603 // Paranoia to ensure our locked and lock-free paths interact 5604 // correctly with each other. 5605 OrderAccess::fence(); 5606 return ret; 5607} 5608 5609void os::PlatformEvent::unpark() { 5610 // Transitions for _Event: 5611 // 0 => 1 : just return 5612 // 1 => 1 : just return 5613 // -1 => either 0 or 1; must signal target thread 5614 // That is, we can safely transition _Event from -1 to either 5615 // 0 or 1. 5616 // See also: "Semaphores in Plan 9" by Mullender & Cox 5617 // 5618 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5619 // that it will take two back-to-back park() calls for the owning 5620 // thread to block. This has the benefit of forcing a spurious return 5621 // from the first park() call after an unpark() call which will help 5622 // shake out uses of park() and unpark() without condition variables. 5623 5624 if (Atomic::xchg(1, &_Event) >= 0) return; 5625 5626 // Wait for the thread associated with the event to vacate 5627 int status = pthread_mutex_lock(_mutex); 5628 assert_status(status == 0, status, "mutex_lock"); 5629 int AnyWaiters = _nParked; 5630 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5631 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5632 AnyWaiters = 0; 5633 pthread_cond_signal(_cond); 5634 } 5635 status = pthread_mutex_unlock(_mutex); 5636 assert_status(status == 0, status, "mutex_unlock"); 5637 if (AnyWaiters != 0) { 5638 // Note that we signal() *after* dropping the lock for "immortal" Events. 5639 // This is safe and avoids a common class of futile wakeups. In rare 5640 // circumstances this can cause a thread to return prematurely from 5641 // cond_{timed}wait() but the spurious wakeup is benign and the victim 5642 // will simply re-test the condition and re-park itself. 5643 // This provides particular benefit if the underlying platform does not 5644 // provide wait morphing. 5645 status = pthread_cond_signal(_cond); 5646 assert_status(status == 0, status, "cond_signal"); 5647 } 5648} 5649 5650 5651// JSR166 5652// ------------------------------------------------------- 5653 5654// The solaris and linux implementations of park/unpark are fairly 5655// conservative for now, but can be improved. They currently use a 5656// mutex/condvar pair, plus a a count. 5657// Park decrements count if > 0, else does a condvar wait. Unpark 5658// sets count to 1 and signals condvar. Only one thread ever waits 5659// on the condvar. Contention seen when trying to park implies that someone 5660// is unparking you, so don't wait. And spurious returns are fine, so there 5661// is no need to track notifications. 5662 5663// This code is common to linux and solaris and will be moved to a 5664// common place in dolphin. 5665// 5666// The passed in time value is either a relative time in nanoseconds 5667// or an absolute time in milliseconds. Either way it has to be unpacked 5668// into suitable seconds and nanoseconds components and stored in the 5669// given timespec structure. 5670// Given time is a 64-bit value and the time_t used in the timespec is only 5671// a signed-32-bit value (except on 64-bit Linux) we have to watch for 5672// overflow if times way in the future are given. Further on Solaris versions 5673// prior to 10 there is a restriction (see cond_timedwait) that the specified 5674// number of seconds, in abstime, is less than current_time + 100,000,000. 5675// As it will be 28 years before "now + 100000000" will overflow we can 5676// ignore overflow and just impose a hard-limit on seconds using the value 5677// of "now + 100,000,000". This places a limit on the timeout of about 3.17 5678// years from "now". 5679 5680static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5681 assert(time > 0, "convertTime"); 5682 time_t max_secs = 0; 5683 5684 if (!os::supports_monotonic_clock() || isAbsolute) { 5685 struct timeval now; 5686 int status = gettimeofday(&now, NULL); 5687 assert(status == 0, "gettimeofday"); 5688 5689 max_secs = now.tv_sec + MAX_SECS; 5690 5691 if (isAbsolute) { 5692 jlong secs = time / 1000; 5693 if (secs > max_secs) { 5694 absTime->tv_sec = max_secs; 5695 } else { 5696 absTime->tv_sec = secs; 5697 } 5698 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5699 } else { 5700 jlong secs = time / NANOSECS_PER_SEC; 5701 if (secs >= MAX_SECS) { 5702 absTime->tv_sec = max_secs; 5703 absTime->tv_nsec = 0; 5704 } else { 5705 absTime->tv_sec = now.tv_sec + secs; 5706 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5707 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5708 absTime->tv_nsec -= NANOSECS_PER_SEC; 5709 ++absTime->tv_sec; // note: this must be <= max_secs 5710 } 5711 } 5712 } 5713 } else { 5714 // must be relative using monotonic clock 5715 struct timespec now; 5716 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5717 assert_status(status == 0, status, "clock_gettime"); 5718 max_secs = now.tv_sec + MAX_SECS; 5719 jlong secs = time / NANOSECS_PER_SEC; 5720 if (secs >= MAX_SECS) { 5721 absTime->tv_sec = max_secs; 5722 absTime->tv_nsec = 0; 5723 } else { 5724 absTime->tv_sec = now.tv_sec + secs; 5725 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5726 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5727 absTime->tv_nsec -= NANOSECS_PER_SEC; 5728 ++absTime->tv_sec; // note: this must be <= max_secs 5729 } 5730 } 5731 } 5732 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5733 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5734 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5735 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5736} 5737 5738void Parker::park(bool isAbsolute, jlong time) { 5739 // Ideally we'd do something useful while spinning, such 5740 // as calling unpackTime(). 5741 5742 // Optional fast-path check: 5743 // Return immediately if a permit is available. 5744 // We depend on Atomic::xchg() having full barrier semantics 5745 // since we are doing a lock-free update to _counter. 5746 if (Atomic::xchg(0, &_counter) > 0) return; 5747 5748 Thread* thread = Thread::current(); 5749 assert(thread->is_Java_thread(), "Must be JavaThread"); 5750 JavaThread *jt = (JavaThread *)thread; 5751 5752 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5753 // Check interrupt before trying to wait 5754 if (Thread::is_interrupted(thread, false)) { 5755 return; 5756 } 5757 5758 // Next, demultiplex/decode time arguments 5759 timespec absTime; 5760 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all 5761 return; 5762 } 5763 if (time > 0) { 5764 unpackTime(&absTime, isAbsolute, time); 5765 } 5766 5767 5768 // Enter safepoint region 5769 // Beware of deadlocks such as 6317397. 5770 // The per-thread Parker:: mutex is a classic leaf-lock. 5771 // In particular a thread must never block on the Threads_lock while 5772 // holding the Parker:: mutex. If safepoints are pending both the 5773 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5774 ThreadBlockInVM tbivm(jt); 5775 5776 // Don't wait if cannot get lock since interference arises from 5777 // unblocking. Also. check interrupt before trying wait 5778 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5779 return; 5780 } 5781 5782 int status; 5783 if (_counter > 0) { // no wait needed 5784 _counter = 0; 5785 status = pthread_mutex_unlock(_mutex); 5786 assert(status == 0, "invariant"); 5787 // Paranoia to ensure our locked and lock-free paths interact 5788 // correctly with each other and Java-level accesses. 5789 OrderAccess::fence(); 5790 return; 5791 } 5792 5793#ifdef ASSERT 5794 // Don't catch signals while blocked; let the running threads have the signals. 5795 // (This allows a debugger to break into the running thread.) 5796 sigset_t oldsigs; 5797 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5798 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5799#endif 5800 5801 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5802 jt->set_suspend_equivalent(); 5803 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5804 5805 assert(_cur_index == -1, "invariant"); 5806 if (time == 0) { 5807 _cur_index = REL_INDEX; // arbitrary choice when not timed 5808 status = pthread_cond_wait(&_cond[_cur_index], _mutex); 5809 } else { 5810 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5811 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime); 5812 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5813 pthread_cond_destroy(&_cond[_cur_index]); 5814 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr()); 5815 } 5816 } 5817 _cur_index = -1; 5818 assert_status(status == 0 || status == EINTR || 5819 status == ETIME || status == ETIMEDOUT, 5820 status, "cond_timedwait"); 5821 5822#ifdef ASSERT 5823 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5824#endif 5825 5826 _counter = 0; 5827 status = pthread_mutex_unlock(_mutex); 5828 assert_status(status == 0, status, "invariant"); 5829 // Paranoia to ensure our locked and lock-free paths interact 5830 // correctly with each other and Java-level accesses. 5831 OrderAccess::fence(); 5832 5833 // If externally suspended while waiting, re-suspend 5834 if (jt->handle_special_suspend_equivalent_condition()) { 5835 jt->java_suspend_self(); 5836 } 5837} 5838 5839void Parker::unpark() { 5840 int status = pthread_mutex_lock(_mutex); 5841 assert(status == 0, "invariant"); 5842 const int s = _counter; 5843 _counter = 1; 5844 if (s < 1) { 5845 // thread might be parked 5846 if (_cur_index != -1) { 5847 // thread is definitely parked 5848 if (WorkAroundNPTLTimedWaitHang) { 5849 status = pthread_cond_signal(&_cond[_cur_index]); 5850 assert(status == 0, "invariant"); 5851 status = pthread_mutex_unlock(_mutex); 5852 assert(status == 0, "invariant"); 5853 } else { 5854 status = pthread_mutex_unlock(_mutex); 5855 assert(status == 0, "invariant"); 5856 status = pthread_cond_signal(&_cond[_cur_index]); 5857 assert(status == 0, "invariant"); 5858 } 5859 } else { 5860 pthread_mutex_unlock(_mutex); 5861 assert(status == 0, "invariant"); 5862 } 5863 } else { 5864 pthread_mutex_unlock(_mutex); 5865 assert(status == 0, "invariant"); 5866 } 5867} 5868 5869 5870extern char** environ; 5871 5872#ifndef __NR_fork 5873 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) AARCH64_ONLY(1079) 5874#endif 5875 5876#ifndef __NR_execve 5877 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) AARCH64_ONLY(221) 5878#endif 5879 5880// Run the specified command in a separate process. Return its exit value, 5881// or -1 on failure (e.g. can't fork a new process). 5882// Unlike system(), this function can be called from signal handler. It 5883// doesn't block SIGINT et al. 5884int os::fork_and_exec(char* cmd) { 5885 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5886 5887 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5888 // pthread_atfork handlers and reset pthread library. All we need is a 5889 // separate process to execve. Make a direct syscall to fork process. 5890 // On IA64 there's no fork syscall, we have to use fork() and hope for 5891 // the best... 5892 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5893 IA64_ONLY(fork();) 5894 5895 if (pid < 0) { 5896 // fork failed 5897 return -1; 5898 5899 } else if (pid == 0) { 5900 // child process 5901 5902 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5903 // first to kill every thread on the thread list. Because this list is 5904 // not reset by fork() (see notes above), execve() will instead kill 5905 // every thread in the parent process. We know this is the only thread 5906 // in the new process, so make a system call directly. 5907 // IA64 should use normal execve() from glibc to match the glibc fork() 5908 // above. 5909 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5910 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5911 5912 // execve failed 5913 _exit(-1); 5914 5915 } else { 5916 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5917 // care about the actual exit code, for now. 5918 5919 int status; 5920 5921 // Wait for the child process to exit. This returns immediately if 5922 // the child has already exited. */ 5923 while (waitpid(pid, &status, 0) < 0) { 5924 switch (errno) { 5925 case ECHILD: return 0; 5926 case EINTR: break; 5927 default: return -1; 5928 } 5929 } 5930 5931 if (WIFEXITED(status)) { 5932 // The child exited normally; get its exit code. 5933 return WEXITSTATUS(status); 5934 } else if (WIFSIGNALED(status)) { 5935 // The child exited because of a signal 5936 // The best value to return is 0x80 + signal number, 5937 // because that is what all Unix shells do, and because 5938 // it allows callers to distinguish between process exit and 5939 // process death by signal. 5940 return 0x80 + WTERMSIG(status); 5941 } else { 5942 // Unknown exit code; pass it through 5943 return status; 5944 } 5945 } 5946} 5947 5948// is_headless_jre() 5949// 5950// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5951// in order to report if we are running in a headless jre 5952// 5953// Since JDK8 xawt/libmawt.so was moved into the same directory 5954// as libawt.so, and renamed libawt_xawt.so 5955// 5956bool os::is_headless_jre() { 5957 struct stat statbuf; 5958 char buf[MAXPATHLEN]; 5959 char libmawtpath[MAXPATHLEN]; 5960 const char *xawtstr = "/xawt/libmawt.so"; 5961 const char *new_xawtstr = "/libawt_xawt.so"; 5962 char *p; 5963 5964 // Get path to libjvm.so 5965 os::jvm_path(buf, sizeof(buf)); 5966 5967 // Get rid of libjvm.so 5968 p = strrchr(buf, '/'); 5969 if (p == NULL) { 5970 return false; 5971 } else { 5972 *p = '\0'; 5973 } 5974 5975 // Get rid of client or server 5976 p = strrchr(buf, '/'); 5977 if (p == NULL) { 5978 return false; 5979 } else { 5980 *p = '\0'; 5981 } 5982 5983 // check xawt/libmawt.so 5984 strcpy(libmawtpath, buf); 5985 strcat(libmawtpath, xawtstr); 5986 if (::stat(libmawtpath, &statbuf) == 0) return false; 5987 5988 // check libawt_xawt.so 5989 strcpy(libmawtpath, buf); 5990 strcat(libmawtpath, new_xawtstr); 5991 if (::stat(libmawtpath, &statbuf) == 0) return false; 5992 5993 return true; 5994} 5995 5996// Get the default path to the core file 5997// Returns the length of the string 5998int os::get_core_path(char* buffer, size_t bufferSize) { 5999 /* 6000 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 6001 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 6002 */ 6003 const int core_pattern_len = 129; 6004 char core_pattern[core_pattern_len] = {0}; 6005 6006 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 6007 if (core_pattern_file != -1) { 6008 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 6009 ::close(core_pattern_file); 6010 6011 if (ret > 0) { 6012 char *last_char = core_pattern + strlen(core_pattern) - 1; 6013 6014 if (*last_char == '\n') { 6015 *last_char = '\0'; 6016 } 6017 } 6018 } 6019 6020 if (strlen(core_pattern) == 0) { 6021 return 0; 6022 } 6023 6024 char *pid_pos = strstr(core_pattern, "%p"); 6025 size_t written; 6026 6027 if (core_pattern[0] == '/') { 6028 written = jio_snprintf(buffer, bufferSize, core_pattern); 6029 } else { 6030 char cwd[PATH_MAX]; 6031 6032 const char* p = get_current_directory(cwd, PATH_MAX); 6033 if (p == NULL) { 6034 assert(p != NULL, "failed to get current directory"); 6035 return 0; 6036 } 6037 6038 if (core_pattern[0] == '|') { 6039 written = jio_snprintf(buffer, bufferSize, 6040 "\"%s\" (or dumping to %s/core.%d)", 6041 &core_pattern[1], p, current_process_id()); 6042 } else { 6043 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 6044 } 6045 } 6046 6047 if ((written >= 0) && (written < bufferSize) 6048 && (pid_pos == NULL) && (core_pattern[0] != '|')) { 6049 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 6050 6051 if (core_uses_pid_file != -1) { 6052 char core_uses_pid = 0; 6053 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 6054 ::close(core_uses_pid_file); 6055 6056 if (core_uses_pid == '1'){ 6057 jio_snprintf(buffer + written, bufferSize - written, 6058 ".%d", current_process_id()); 6059 } 6060 } 6061 } 6062 6063 return strlen(buffer); 6064} 6065 6066/////////////// Unit tests /////////////// 6067 6068#ifndef PRODUCT 6069 6070#define test_log(...) \ 6071 do { \ 6072 if (VerboseInternalVMTests) { \ 6073 tty->print_cr(__VA_ARGS__); \ 6074 tty->flush(); \ 6075 } \ 6076 } while (false) 6077 6078class TestReserveMemorySpecial : AllStatic { 6079 public: 6080 static void small_page_write(void* addr, size_t size) { 6081 size_t page_size = os::vm_page_size(); 6082 6083 char* end = (char*)addr + size; 6084 for (char* p = (char*)addr; p < end; p += page_size) { 6085 *p = 1; 6086 } 6087 } 6088 6089 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6090 if (!UseHugeTLBFS) { 6091 return; 6092 } 6093 6094 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6095 6096 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6097 6098 if (addr != NULL) { 6099 small_page_write(addr, size); 6100 6101 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6102 } 6103 } 6104 6105 static void test_reserve_memory_special_huge_tlbfs_only() { 6106 if (!UseHugeTLBFS) { 6107 return; 6108 } 6109 6110 size_t lp = os::large_page_size(); 6111 6112 for (size_t size = lp; size <= lp * 10; size += lp) { 6113 test_reserve_memory_special_huge_tlbfs_only(size); 6114 } 6115 } 6116 6117 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) { 6118 if (!UseHugeTLBFS) { 6119 return; 6120 } 6121 6122 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")", 6123 size, alignment); 6124 6125 assert(size >= os::large_page_size(), "Incorrect input to test"); 6126 6127 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6128 6129 if (addr != NULL) { 6130 small_page_write(addr, size); 6131 6132 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6133 } 6134 } 6135 6136 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) { 6137 size_t lp = os::large_page_size(); 6138 size_t ag = os::vm_allocation_granularity(); 6139 6140 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6141 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment); 6142 } 6143 } 6144 6145 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6146 size_t lp = os::large_page_size(); 6147 size_t ag = os::vm_allocation_granularity(); 6148 6149 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp); 6150 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag); 6151 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2); 6152 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2); 6153 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag); 6154 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag); 6155 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2); 6156 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10); 6157 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2); 6158 } 6159 6160 static void test_reserve_memory_special_huge_tlbfs() { 6161 if (!UseHugeTLBFS) { 6162 return; 6163 } 6164 6165 test_reserve_memory_special_huge_tlbfs_only(); 6166 test_reserve_memory_special_huge_tlbfs_mixed(); 6167 } 6168 6169 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6170 if (!UseSHM) { 6171 return; 6172 } 6173 6174 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6175 6176 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6177 6178 if (addr != NULL) { 6179 assert(is_ptr_aligned(addr, alignment), "Check"); 6180 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6181 6182 small_page_write(addr, size); 6183 6184 os::Linux::release_memory_special_shm(addr, size); 6185 } 6186 } 6187 6188 static void test_reserve_memory_special_shm() { 6189 size_t lp = os::large_page_size(); 6190 size_t ag = os::vm_allocation_granularity(); 6191 6192 for (size_t size = ag; size < lp * 3; size += ag) { 6193 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6194 test_reserve_memory_special_shm(size, alignment); 6195 } 6196 } 6197 } 6198 6199 static void test() { 6200 test_reserve_memory_special_huge_tlbfs(); 6201 test_reserve_memory_special_shm(); 6202 } 6203}; 6204 6205void TestReserveMemorySpecial_test() { 6206 TestReserveMemorySpecial::test(); 6207} 6208 6209#endif 6210