os_linux.cpp revision 7877:cc8363b030d5
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/vmError.hpp" 72 73// put OS-includes here 74# include <sys/types.h> 75# include <sys/mman.h> 76# include <sys/stat.h> 77# include <sys/select.h> 78# include <pthread.h> 79# include <signal.h> 80# include <errno.h> 81# include <dlfcn.h> 82# include <stdio.h> 83# include <unistd.h> 84# include <sys/resource.h> 85# include <pthread.h> 86# include <sys/stat.h> 87# include <sys/time.h> 88# include <sys/times.h> 89# include <sys/utsname.h> 90# include <sys/socket.h> 91# include <sys/wait.h> 92# include <pwd.h> 93# include <poll.h> 94# include <semaphore.h> 95# include <fcntl.h> 96# include <string.h> 97# include <syscall.h> 98# include <sys/sysinfo.h> 99# include <gnu/libc-version.h> 100# include <sys/ipc.h> 101# include <sys/shm.h> 102# include <link.h> 103# include <stdint.h> 104# include <inttypes.h> 105# include <sys/ioctl.h> 106 107PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC 108 109// if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 110// getrusage() is prepared to handle the associated failure. 111#ifndef RUSAGE_THREAD 112 #define RUSAGE_THREAD (1) /* only the calling thread */ 113#endif 114 115#define MAX_PATH (2 * K) 116 117#define MAX_SECS 100000000 118 119// for timer info max values which include all bits 120#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 121 122#define LARGEPAGES_BIT (1 << 6) 123//////////////////////////////////////////////////////////////////////////////// 124// global variables 125julong os::Linux::_physical_memory = 0; 126 127address os::Linux::_initial_thread_stack_bottom = NULL; 128uintptr_t os::Linux::_initial_thread_stack_size = 0; 129 130int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 131int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 132int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL; 133Mutex* os::Linux::_createThread_lock = NULL; 134pthread_t os::Linux::_main_thread; 135int os::Linux::_page_size = -1; 136const int os::Linux::_vm_default_page_size = (8 * K); 137bool os::Linux::_is_floating_stack = false; 138bool os::Linux::_is_NPTL = false; 139bool os::Linux::_supports_fast_thread_cpu_time = false; 140const char * os::Linux::_glibc_version = NULL; 141const char * os::Linux::_libpthread_version = NULL; 142pthread_condattr_t os::Linux::_condattr[1]; 143 144static jlong initial_time_count=0; 145 146static int clock_tics_per_sec = 100; 147 148// For diagnostics to print a message once. see run_periodic_checks 149static sigset_t check_signal_done; 150static bool check_signals = true; 151 152static pid_t _initial_pid = 0; 153 154// Signal number used to suspend/resume a thread 155 156// do not use any signal number less than SIGSEGV, see 4355769 157static int SR_signum = SIGUSR2; 158sigset_t SR_sigset; 159 160// Used to protect dlsym() calls 161static pthread_mutex_t dl_mutex; 162 163// Declarations 164static void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 165 166// utility functions 167 168static int SR_initialize(); 169 170julong os::available_memory() { 171 return Linux::available_memory(); 172} 173 174julong os::Linux::available_memory() { 175 // values in struct sysinfo are "unsigned long" 176 struct sysinfo si; 177 sysinfo(&si); 178 179 return (julong)si.freeram * si.mem_unit; 180} 181 182julong os::physical_memory() { 183 return Linux::physical_memory(); 184} 185 186//////////////////////////////////////////////////////////////////////////////// 187// environment support 188 189bool os::getenv(const char* name, char* buf, int len) { 190 const char* val = ::getenv(name); 191 if (val != NULL && strlen(val) < (size_t)len) { 192 strcpy(buf, val); 193 return true; 194 } 195 if (len > 0) buf[0] = 0; // return a null string 196 return false; 197} 198 199 200// Return true if user is running as root. 201 202bool os::have_special_privileges() { 203 static bool init = false; 204 static bool privileges = false; 205 if (!init) { 206 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 207 init = true; 208 } 209 return privileges; 210} 211 212 213#ifndef SYS_gettid 214// i386: 224, ia64: 1105, amd64: 186, sparc 143 215 #ifdef __ia64__ 216 #define SYS_gettid 1105 217 #elif __i386__ 218 #define SYS_gettid 224 219 #elif __amd64__ 220 #define SYS_gettid 186 221 #elif __sparc__ 222 #define SYS_gettid 143 223 #else 224 #error define gettid for the arch 225 #endif 226#endif 227 228// Cpu architecture string 229#if defined(ZERO) 230static char cpu_arch[] = ZERO_LIBARCH; 231#elif defined(IA64) 232static char cpu_arch[] = "ia64"; 233#elif defined(IA32) 234static char cpu_arch[] = "i386"; 235#elif defined(AMD64) 236static char cpu_arch[] = "amd64"; 237#elif defined(ARM) 238static char cpu_arch[] = "arm"; 239#elif defined(PPC32) 240static char cpu_arch[] = "ppc"; 241#elif defined(PPC64) 242static char cpu_arch[] = "ppc64"; 243#elif defined(SPARC) 244 #ifdef _LP64 245static char cpu_arch[] = "sparcv9"; 246 #else 247static char cpu_arch[] = "sparc"; 248 #endif 249#elif defined(AARCH64) 250static char cpu_arch[] = "aarch64"; 251#else 252 #error Add appropriate cpu_arch setting 253#endif 254 255 256// pid_t gettid() 257// 258// Returns the kernel thread id of the currently running thread. Kernel 259// thread id is used to access /proc. 260// 261// (Note that getpid() on LinuxThreads returns kernel thread id too; but 262// on NPTL, it returns the same pid for all threads, as required by POSIX.) 263// 264pid_t os::Linux::gettid() { 265 int rslt = syscall(SYS_gettid); 266 if (rslt == -1) { 267 // old kernel, no NPTL support 268 return getpid(); 269 } else { 270 return (pid_t)rslt; 271 } 272} 273 274// Most versions of linux have a bug where the number of processors are 275// determined by looking at the /proc file system. In a chroot environment, 276// the system call returns 1. This causes the VM to act as if it is 277// a single processor and elide locking (see is_MP() call). 278static bool unsafe_chroot_detected = false; 279static const char *unstable_chroot_error = "/proc file system not found.\n" 280 "Java may be unstable running multithreaded in a chroot " 281 "environment on Linux when /proc filesystem is not mounted."; 282 283void os::Linux::initialize_system_info() { 284 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 285 if (processor_count() == 1) { 286 pid_t pid = os::Linux::gettid(); 287 char fname[32]; 288 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 289 FILE *fp = fopen(fname, "r"); 290 if (fp == NULL) { 291 unsafe_chroot_detected = true; 292 } else { 293 fclose(fp); 294 } 295 } 296 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 297 assert(processor_count() > 0, "linux error"); 298} 299 300void os::init_system_properties_values() { 301 // The next steps are taken in the product version: 302 // 303 // Obtain the JAVA_HOME value from the location of libjvm.so. 304 // This library should be located at: 305 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 306 // 307 // If "/jre/lib/" appears at the right place in the path, then we 308 // assume libjvm.so is installed in a JDK and we use this path. 309 // 310 // Otherwise exit with message: "Could not create the Java virtual machine." 311 // 312 // The following extra steps are taken in the debugging version: 313 // 314 // If "/jre/lib/" does NOT appear at the right place in the path 315 // instead of exit check for $JAVA_HOME environment variable. 316 // 317 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 318 // then we append a fake suffix "hotspot/libjvm.so" to this path so 319 // it looks like libjvm.so is installed there 320 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 321 // 322 // Otherwise exit. 323 // 324 // Important note: if the location of libjvm.so changes this 325 // code needs to be changed accordingly. 326 327 // See ld(1): 328 // The linker uses the following search paths to locate required 329 // shared libraries: 330 // 1: ... 331 // ... 332 // 7: The default directories, normally /lib and /usr/lib. 333#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 334 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 335#else 336 #define DEFAULT_LIBPATH "/lib:/usr/lib" 337#endif 338 339// Base path of extensions installed on the system. 340#define SYS_EXT_DIR "/usr/java/packages" 341#define EXTENSIONS_DIR "/lib/ext" 342#define ENDORSED_DIR "/lib/endorsed" 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 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends. 349 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir 350 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir 351 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 352 353 // sysclasspath, java_home, dll_dir 354 { 355 char *pslash; 356 os::jvm_path(buf, bufsize); 357 358 // Found the full path to libjvm.so. 359 // Now cut the path to <java_home>/jre if we can. 360 pslash = strrchr(buf, '/'); 361 if (pslash != NULL) { 362 *pslash = '\0'; // Get rid of /libjvm.so. 363 } 364 pslash = strrchr(buf, '/'); 365 if (pslash != NULL) { 366 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 367 } 368 Arguments::set_dll_dir(buf); 369 370 if (pslash != NULL) { 371 pslash = strrchr(buf, '/'); 372 if (pslash != NULL) { 373 *pslash = '\0'; // Get rid of /<arch>. 374 pslash = strrchr(buf, '/'); 375 if (pslash != NULL) { 376 *pslash = '\0'; // Get rid of /lib. 377 } 378 } 379 } 380 Arguments::set_java_home(buf); 381 set_boot_path('/', ':'); 382 } 383 384 // Where to look for native libraries. 385 // 386 // Note: Due to a legacy implementation, most of the library path 387 // is set in the launcher. This was to accomodate linking restrictions 388 // on legacy Linux implementations (which are no longer supported). 389 // Eventually, all the library path setting will be done here. 390 // 391 // However, to prevent the proliferation of improperly built native 392 // libraries, the new path component /usr/java/packages is added here. 393 // Eventually, all the library path setting will be done here. 394 { 395 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 396 // should always exist (until the legacy problem cited above is 397 // addressed). 398 const char *v = ::getenv("LD_LIBRARY_PATH"); 399 const char *v_colon = ":"; 400 if (v == NULL) { v = ""; v_colon = ""; } 401 // That's +1 for the colon and +1 for the trailing '\0'. 402 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 403 strlen(v) + 1 + 404 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1, 405 mtInternal); 406 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch); 407 Arguments::set_library_path(ld_library_path); 408 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal); 409 } 410 411 // Extensions directories. 412 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 413 Arguments::set_ext_dirs(buf); 414 415 // Endorsed standards default directory. 416 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 417 Arguments::set_endorsed_dirs(buf); 418 419 FREE_C_HEAP_ARRAY(char, buf, mtInternal); 420 421#undef DEFAULT_LIBPATH 422#undef SYS_EXT_DIR 423#undef EXTENSIONS_DIR 424#undef ENDORSED_DIR 425} 426 427//////////////////////////////////////////////////////////////////////////////// 428// breakpoint support 429 430void os::breakpoint() { 431 BREAKPOINT; 432} 433 434extern "C" void breakpoint() { 435 // use debugger to set breakpoint here 436} 437 438//////////////////////////////////////////////////////////////////////////////// 439// signal support 440 441debug_only(static bool signal_sets_initialized = false); 442static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 443 444bool os::Linux::is_sig_ignored(int sig) { 445 struct sigaction oact; 446 sigaction(sig, (struct sigaction*)NULL, &oact); 447 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 448 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 449 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) { 450 return true; 451 } else { 452 return false; 453 } 454} 455 456void os::Linux::signal_sets_init() { 457 // Should also have an assertion stating we are still single-threaded. 458 assert(!signal_sets_initialized, "Already initialized"); 459 // Fill in signals that are necessarily unblocked for all threads in 460 // the VM. Currently, we unblock the following signals: 461 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 462 // by -Xrs (=ReduceSignalUsage)); 463 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 464 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 465 // the dispositions or masks wrt these signals. 466 // Programs embedding the VM that want to use the above signals for their 467 // own purposes must, at this time, use the "-Xrs" option to prevent 468 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 469 // (See bug 4345157, and other related bugs). 470 // In reality, though, unblocking these signals is really a nop, since 471 // these signals are not blocked by default. 472 sigemptyset(&unblocked_sigs); 473 sigemptyset(&allowdebug_blocked_sigs); 474 sigaddset(&unblocked_sigs, SIGILL); 475 sigaddset(&unblocked_sigs, SIGSEGV); 476 sigaddset(&unblocked_sigs, SIGBUS); 477 sigaddset(&unblocked_sigs, SIGFPE); 478#if defined(PPC64) 479 sigaddset(&unblocked_sigs, SIGTRAP); 480#endif 481 sigaddset(&unblocked_sigs, SR_signum); 482 483 if (!ReduceSignalUsage) { 484 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 485 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 486 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 487 } 488 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 489 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 490 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 491 } 492 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 493 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 494 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 495 } 496 } 497 // Fill in signals that are blocked by all but the VM thread. 498 sigemptyset(&vm_sigs); 499 if (!ReduceSignalUsage) { 500 sigaddset(&vm_sigs, BREAK_SIGNAL); 501 } 502 debug_only(signal_sets_initialized = true); 503 504} 505 506// These are signals that are unblocked while a thread is running Java. 507// (For some reason, they get blocked by default.) 508sigset_t* os::Linux::unblocked_signals() { 509 assert(signal_sets_initialized, "Not initialized"); 510 return &unblocked_sigs; 511} 512 513// These are the signals that are blocked while a (non-VM) thread is 514// running Java. Only the VM thread handles these signals. 515sigset_t* os::Linux::vm_signals() { 516 assert(signal_sets_initialized, "Not initialized"); 517 return &vm_sigs; 518} 519 520// These are signals that are blocked during cond_wait to allow debugger in 521sigset_t* os::Linux::allowdebug_blocked_signals() { 522 assert(signal_sets_initialized, "Not initialized"); 523 return &allowdebug_blocked_sigs; 524} 525 526void os::Linux::hotspot_sigmask(Thread* thread) { 527 528 //Save caller's signal mask before setting VM signal mask 529 sigset_t caller_sigmask; 530 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 531 532 OSThread* osthread = thread->osthread(); 533 osthread->set_caller_sigmask(caller_sigmask); 534 535 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 536 537 if (!ReduceSignalUsage) { 538 if (thread->is_VM_thread()) { 539 // Only the VM thread handles BREAK_SIGNAL ... 540 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 541 } else { 542 // ... all other threads block BREAK_SIGNAL 543 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 544 } 545 } 546} 547 548////////////////////////////////////////////////////////////////////////////// 549// detecting pthread library 550 551void os::Linux::libpthread_init() { 552 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 553 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 554 // generic name for earlier versions. 555 // Define macros here so we can build HotSpot on old systems. 556#ifndef _CS_GNU_LIBC_VERSION 557 #define _CS_GNU_LIBC_VERSION 2 558#endif 559#ifndef _CS_GNU_LIBPTHREAD_VERSION 560 #define _CS_GNU_LIBPTHREAD_VERSION 3 561#endif 562 563 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 564 if (n > 0) { 565 char *str = (char *)malloc(n, mtInternal); 566 confstr(_CS_GNU_LIBC_VERSION, str, n); 567 os::Linux::set_glibc_version(str); 568 } else { 569 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 570 static char _gnu_libc_version[32]; 571 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 572 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 573 os::Linux::set_glibc_version(_gnu_libc_version); 574 } 575 576 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 577 if (n > 0) { 578 char *str = (char *)malloc(n, mtInternal); 579 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 580 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 581 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 582 // is the case. LinuxThreads has a hard limit on max number of threads. 583 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 584 // On the other hand, NPTL does not have such a limit, sysconf() 585 // will return -1 and errno is not changed. Check if it is really NPTL. 586 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 587 strstr(str, "NPTL") && 588 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 589 free(str); 590 os::Linux::set_libpthread_version("linuxthreads"); 591 } else { 592 os::Linux::set_libpthread_version(str); 593 } 594 } else { 595 // glibc before 2.3.2 only has LinuxThreads. 596 os::Linux::set_libpthread_version("linuxthreads"); 597 } 598 599 if (strstr(libpthread_version(), "NPTL")) { 600 os::Linux::set_is_NPTL(); 601 } else { 602 os::Linux::set_is_LinuxThreads(); 603 } 604 605 // LinuxThreads have two flavors: floating-stack mode, which allows variable 606 // stack size; and fixed-stack mode. NPTL is always floating-stack. 607 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 608 os::Linux::set_is_floating_stack(); 609 } 610} 611 612///////////////////////////////////////////////////////////////////////////// 613// thread stack 614 615// Force Linux kernel to expand current thread stack. If "bottom" is close 616// to the stack guard, caller should block all signals. 617// 618// MAP_GROWSDOWN: 619// A special mmap() flag that is used to implement thread stacks. It tells 620// kernel that the memory region should extend downwards when needed. This 621// allows early versions of LinuxThreads to only mmap the first few pages 622// when creating a new thread. Linux kernel will automatically expand thread 623// stack as needed (on page faults). 624// 625// However, because the memory region of a MAP_GROWSDOWN stack can grow on 626// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 627// region, it's hard to tell if the fault is due to a legitimate stack 628// access or because of reading/writing non-exist memory (e.g. buffer 629// overrun). As a rule, if the fault happens below current stack pointer, 630// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 631// application (see Linux kernel fault.c). 632// 633// This Linux feature can cause SIGSEGV when VM bangs thread stack for 634// stack overflow detection. 635// 636// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 637// not use this flag. However, the stack of initial thread is not created 638// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 639// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 640// and then attach the thread to JVM. 641// 642// To get around the problem and allow stack banging on Linux, we need to 643// manually expand thread stack after receiving the SIGSEGV. 644// 645// There are two ways to expand thread stack to address "bottom", we used 646// both of them in JVM before 1.5: 647// 1. adjust stack pointer first so that it is below "bottom", and then 648// touch "bottom" 649// 2. mmap() the page in question 650// 651// Now alternate signal stack is gone, it's harder to use 2. For instance, 652// if current sp is already near the lower end of page 101, and we need to 653// call mmap() to map page 100, it is possible that part of the mmap() frame 654// will be placed in page 100. When page 100 is mapped, it is zero-filled. 655// That will destroy the mmap() frame and cause VM to crash. 656// 657// The following code works by adjusting sp first, then accessing the "bottom" 658// page to force a page fault. Linux kernel will then automatically expand the 659// stack mapping. 660// 661// _expand_stack_to() assumes its frame size is less than page size, which 662// should always be true if the function is not inlined. 663 664#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 665 #define NOINLINE 666#else 667 #define NOINLINE __attribute__ ((noinline)) 668#endif 669 670static void _expand_stack_to(address bottom) NOINLINE; 671 672static void _expand_stack_to(address bottom) { 673 address sp; 674 size_t size; 675 volatile char *p; 676 677 // Adjust bottom to point to the largest address within the same page, it 678 // gives us a one-page buffer if alloca() allocates slightly more memory. 679 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 680 bottom += os::Linux::page_size() - 1; 681 682 // sp might be slightly above current stack pointer; if that's the case, we 683 // will alloca() a little more space than necessary, which is OK. Don't use 684 // os::current_stack_pointer(), as its result can be slightly below current 685 // stack pointer, causing us to not alloca enough to reach "bottom". 686 sp = (address)&sp; 687 688 if (sp > bottom) { 689 size = sp - bottom; 690 p = (volatile char *)alloca(size); 691 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 692 p[0] = '\0'; 693 } 694} 695 696bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 697 assert(t!=NULL, "just checking"); 698 assert(t->osthread()->expanding_stack(), "expand should be set"); 699 assert(t->stack_base() != NULL, "stack_base was not initialized"); 700 701 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 702 sigset_t mask_all, old_sigset; 703 sigfillset(&mask_all); 704 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 705 _expand_stack_to(addr); 706 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 707 return true; 708 } 709 return false; 710} 711 712////////////////////////////////////////////////////////////////////////////// 713// create new thread 714 715static address highest_vm_reserved_address(); 716 717// check if it's safe to start a new thread 718static bool _thread_safety_check(Thread* thread) { 719 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 720 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 721 // Heap is mmap'ed at lower end of memory space. Thread stacks are 722 // allocated (MAP_FIXED) from high address space. Every thread stack 723 // occupies a fixed size slot (usually 2Mbytes, but user can change 724 // it to other values if they rebuild LinuxThreads). 725 // 726 // Problem with MAP_FIXED is that mmap() can still succeed even part of 727 // the memory region has already been mmap'ed. That means if we have too 728 // many threads and/or very large heap, eventually thread stack will 729 // collide with heap. 730 // 731 // Here we try to prevent heap/stack collision by comparing current 732 // stack bottom with the highest address that has been mmap'ed by JVM 733 // plus a safety margin for memory maps created by native code. 734 // 735 // This feature can be disabled by setting ThreadSafetyMargin to 0 736 // 737 if (ThreadSafetyMargin > 0) { 738 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 739 740 // not safe if our stack extends below the safety margin 741 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 742 } else { 743 return true; 744 } 745 } else { 746 // Floating stack LinuxThreads or NPTL: 747 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 748 // there's not enough space left, pthread_create() will fail. If we come 749 // here, that means enough space has been reserved for stack. 750 return true; 751 } 752} 753 754// Thread start routine for all newly created threads 755static void *java_start(Thread *thread) { 756 // Try to randomize the cache line index of hot stack frames. 757 // This helps when threads of the same stack traces evict each other's 758 // cache lines. The threads can be either from the same JVM instance, or 759 // from different JVM instances. The benefit is especially true for 760 // processors with hyperthreading technology. 761 static int counter = 0; 762 int pid = os::current_process_id(); 763 alloca(((pid ^ counter++) & 7) * 128); 764 765 ThreadLocalStorage::set_thread(thread); 766 767 OSThread* osthread = thread->osthread(); 768 Monitor* sync = osthread->startThread_lock(); 769 770 // non floating stack LinuxThreads needs extra check, see above 771 if (!_thread_safety_check(thread)) { 772 // notify parent thread 773 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 774 osthread->set_state(ZOMBIE); 775 sync->notify_all(); 776 return NULL; 777 } 778 779 // thread_id is kernel thread id (similar to Solaris LWP id) 780 osthread->set_thread_id(os::Linux::gettid()); 781 782 if (UseNUMA) { 783 int lgrp_id = os::numa_get_group_id(); 784 if (lgrp_id != -1) { 785 thread->set_lgrp_id(lgrp_id); 786 } 787 } 788 // initialize signal mask for this thread 789 os::Linux::hotspot_sigmask(thread); 790 791 // initialize floating point control register 792 os::Linux::init_thread_fpu_state(); 793 794 // handshaking with parent thread 795 { 796 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 797 798 // notify parent thread 799 osthread->set_state(INITIALIZED); 800 sync->notify_all(); 801 802 // wait until os::start_thread() 803 while (osthread->get_state() == INITIALIZED) { 804 sync->wait(Mutex::_no_safepoint_check_flag); 805 } 806 } 807 808 // call one more level start routine 809 thread->run(); 810 811 return 0; 812} 813 814bool os::create_thread(Thread* thread, ThreadType thr_type, 815 size_t stack_size) { 816 assert(thread->osthread() == NULL, "caller responsible"); 817 818 // Allocate the OSThread object 819 OSThread* osthread = new OSThread(NULL, NULL); 820 if (osthread == NULL) { 821 return false; 822 } 823 824 // set the correct thread state 825 osthread->set_thread_type(thr_type); 826 827 // Initial state is ALLOCATED but not INITIALIZED 828 osthread->set_state(ALLOCATED); 829 830 thread->set_osthread(osthread); 831 832 // init thread attributes 833 pthread_attr_t attr; 834 pthread_attr_init(&attr); 835 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 836 837 // stack size 838 if (os::Linux::supports_variable_stack_size()) { 839 // calculate stack size if it's not specified by caller 840 if (stack_size == 0) { 841 stack_size = os::Linux::default_stack_size(thr_type); 842 843 switch (thr_type) { 844 case os::java_thread: 845 // Java threads use ThreadStackSize which default value can be 846 // changed with the flag -Xss 847 assert(JavaThread::stack_size_at_create() > 0, "this should be set"); 848 stack_size = JavaThread::stack_size_at_create(); 849 break; 850 case os::compiler_thread: 851 if (CompilerThreadStackSize > 0) { 852 stack_size = (size_t)(CompilerThreadStackSize * K); 853 break; 854 } // else fall through: 855 // use VMThreadStackSize if CompilerThreadStackSize is not defined 856 case os::vm_thread: 857 case os::pgc_thread: 858 case os::cgc_thread: 859 case os::watcher_thread: 860 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 861 break; 862 } 863 } 864 865 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 866 pthread_attr_setstacksize(&attr, stack_size); 867 } else { 868 // let pthread_create() pick the default value. 869 } 870 871 // glibc guard page 872 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 873 874 ThreadState state; 875 876 { 877 // Serialize thread creation if we are running with fixed stack LinuxThreads 878 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 879 if (lock) { 880 os::Linux::createThread_lock()->lock_without_safepoint_check(); 881 } 882 883 pthread_t tid; 884 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 885 886 pthread_attr_destroy(&attr); 887 888 if (ret != 0) { 889 if (PrintMiscellaneous && (Verbose || WizardMode)) { 890 perror("pthread_create()"); 891 } 892 // Need to clean up stuff we've allocated so far 893 thread->set_osthread(NULL); 894 delete osthread; 895 if (lock) os::Linux::createThread_lock()->unlock(); 896 return false; 897 } 898 899 // Store pthread info into the OSThread 900 osthread->set_pthread_id(tid); 901 902 // Wait until child thread is either initialized or aborted 903 { 904 Monitor* sync_with_child = osthread->startThread_lock(); 905 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 906 while ((state = osthread->get_state()) == ALLOCATED) { 907 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 908 } 909 } 910 911 if (lock) { 912 os::Linux::createThread_lock()->unlock(); 913 } 914 } 915 916 // Aborted due to thread limit being reached 917 if (state == ZOMBIE) { 918 thread->set_osthread(NULL); 919 delete osthread; 920 return false; 921 } 922 923 // The thread is returned suspended (in state INITIALIZED), 924 // and is started higher up in the call chain 925 assert(state == INITIALIZED, "race condition"); 926 return true; 927} 928 929///////////////////////////////////////////////////////////////////////////// 930// attach existing thread 931 932// bootstrap the main thread 933bool os::create_main_thread(JavaThread* thread) { 934 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 935 return create_attached_thread(thread); 936} 937 938bool os::create_attached_thread(JavaThread* thread) { 939#ifdef ASSERT 940 thread->verify_not_published(); 941#endif 942 943 // Allocate the OSThread object 944 OSThread* osthread = new OSThread(NULL, NULL); 945 946 if (osthread == NULL) { 947 return false; 948 } 949 950 // Store pthread info into the OSThread 951 osthread->set_thread_id(os::Linux::gettid()); 952 osthread->set_pthread_id(::pthread_self()); 953 954 // initialize floating point control register 955 os::Linux::init_thread_fpu_state(); 956 957 // Initial thread state is RUNNABLE 958 osthread->set_state(RUNNABLE); 959 960 thread->set_osthread(osthread); 961 962 if (UseNUMA) { 963 int lgrp_id = os::numa_get_group_id(); 964 if (lgrp_id != -1) { 965 thread->set_lgrp_id(lgrp_id); 966 } 967 } 968 969 if (os::Linux::is_initial_thread()) { 970 // If current thread is initial thread, its stack is mapped on demand, 971 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 972 // the entire stack region to avoid SEGV in stack banging. 973 // It is also useful to get around the heap-stack-gap problem on SuSE 974 // kernel (see 4821821 for details). We first expand stack to the top 975 // of yellow zone, then enable stack yellow zone (order is significant, 976 // enabling yellow zone first will crash JVM on SuSE Linux), so there 977 // is no gap between the last two virtual memory regions. 978 979 JavaThread *jt = (JavaThread *)thread; 980 address addr = jt->stack_yellow_zone_base(); 981 assert(addr != NULL, "initialization problem?"); 982 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 983 984 osthread->set_expanding_stack(); 985 os::Linux::manually_expand_stack(jt, addr); 986 osthread->clear_expanding_stack(); 987 } 988 989 // initialize signal mask for this thread 990 // and save the caller's signal mask 991 os::Linux::hotspot_sigmask(thread); 992 993 return true; 994} 995 996void os::pd_start_thread(Thread* thread) { 997 OSThread * osthread = thread->osthread(); 998 assert(osthread->get_state() != INITIALIZED, "just checking"); 999 Monitor* sync_with_child = osthread->startThread_lock(); 1000 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1001 sync_with_child->notify(); 1002} 1003 1004// Free Linux resources related to the OSThread 1005void os::free_thread(OSThread* osthread) { 1006 assert(osthread != NULL, "osthread not set"); 1007 1008 if (Thread::current()->osthread() == osthread) { 1009 // Restore caller's signal mask 1010 sigset_t sigmask = osthread->caller_sigmask(); 1011 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1012 } 1013 1014 delete osthread; 1015} 1016 1017////////////////////////////////////////////////////////////////////////////// 1018// thread local storage 1019 1020// Restore the thread pointer if the destructor is called. This is in case 1021// someone from JNI code sets up a destructor with pthread_key_create to run 1022// detachCurrentThread on thread death. Unless we restore the thread pointer we 1023// will hang or crash. When detachCurrentThread is called the key will be set 1024// to null and we will not be called again. If detachCurrentThread is never 1025// called we could loop forever depending on the pthread implementation. 1026static void restore_thread_pointer(void* p) { 1027 Thread* thread = (Thread*) p; 1028 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread); 1029} 1030 1031int os::allocate_thread_local_storage() { 1032 pthread_key_t key; 1033 int rslt = pthread_key_create(&key, restore_thread_pointer); 1034 assert(rslt == 0, "cannot allocate thread local storage"); 1035 return (int)key; 1036} 1037 1038// Note: This is currently not used by VM, as we don't destroy TLS key 1039// on VM exit. 1040void os::free_thread_local_storage(int index) { 1041 int rslt = pthread_key_delete((pthread_key_t)index); 1042 assert(rslt == 0, "invalid index"); 1043} 1044 1045void os::thread_local_storage_at_put(int index, void* value) { 1046 int rslt = pthread_setspecific((pthread_key_t)index, value); 1047 assert(rslt == 0, "pthread_setspecific failed"); 1048} 1049 1050extern "C" Thread* get_thread() { 1051 return ThreadLocalStorage::thread(); 1052} 1053 1054////////////////////////////////////////////////////////////////////////////// 1055// initial thread 1056 1057// Check if current thread is the initial thread, similar to Solaris thr_main. 1058bool os::Linux::is_initial_thread(void) { 1059 char dummy; 1060 // If called before init complete, thread stack bottom will be null. 1061 // Can be called if fatal error occurs before initialization. 1062 if (initial_thread_stack_bottom() == NULL) return false; 1063 assert(initial_thread_stack_bottom() != NULL && 1064 initial_thread_stack_size() != 0, 1065 "os::init did not locate initial thread's stack region"); 1066 if ((address)&dummy >= initial_thread_stack_bottom() && 1067 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) { 1068 return true; 1069 } else { 1070 return false; 1071 } 1072} 1073 1074// Find the virtual memory area that contains addr 1075static bool find_vma(address addr, address* vma_low, address* vma_high) { 1076 FILE *fp = fopen("/proc/self/maps", "r"); 1077 if (fp) { 1078 address low, high; 1079 while (!feof(fp)) { 1080 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1081 if (low <= addr && addr < high) { 1082 if (vma_low) *vma_low = low; 1083 if (vma_high) *vma_high = high; 1084 fclose(fp); 1085 return true; 1086 } 1087 } 1088 for (;;) { 1089 int ch = fgetc(fp); 1090 if (ch == EOF || ch == (int)'\n') break; 1091 } 1092 } 1093 fclose(fp); 1094 } 1095 return false; 1096} 1097 1098// Locate initial thread stack. This special handling of initial thread stack 1099// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1100// bogus value for initial thread. 1101void os::Linux::capture_initial_stack(size_t max_size) { 1102 // stack size is the easy part, get it from RLIMIT_STACK 1103 size_t stack_size; 1104 struct rlimit rlim; 1105 getrlimit(RLIMIT_STACK, &rlim); 1106 stack_size = rlim.rlim_cur; 1107 1108 // 6308388: a bug in ld.so will relocate its own .data section to the 1109 // lower end of primordial stack; reduce ulimit -s value a little bit 1110 // so we won't install guard page on ld.so's data section. 1111 stack_size -= 2 * page_size(); 1112 1113 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1114 // 7.1, in both cases we will get 2G in return value. 1115 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1116 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1117 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1118 // in case other parts in glibc still assumes 2M max stack size. 1119 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1120 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1121 if (stack_size > 2 * K * K IA64_ONLY(*2)) { 1122 stack_size = 2 * K * K IA64_ONLY(*2); 1123 } 1124 // Try to figure out where the stack base (top) is. This is harder. 1125 // 1126 // When an application is started, glibc saves the initial stack pointer in 1127 // a global variable "__libc_stack_end", which is then used by system 1128 // libraries. __libc_stack_end should be pretty close to stack top. The 1129 // variable is available since the very early days. However, because it is 1130 // a private interface, it could disappear in the future. 1131 // 1132 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1133 // to __libc_stack_end, it is very close to stack top, but isn't the real 1134 // stack top. Note that /proc may not exist if VM is running as a chroot 1135 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1136 // /proc/<pid>/stat could change in the future (though unlikely). 1137 // 1138 // We try __libc_stack_end first. If that doesn't work, look for 1139 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1140 // as a hint, which should work well in most cases. 1141 1142 uintptr_t stack_start; 1143 1144 // try __libc_stack_end first 1145 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1146 if (p && *p) { 1147 stack_start = *p; 1148 } else { 1149 // see if we can get the start_stack field from /proc/self/stat 1150 FILE *fp; 1151 int pid; 1152 char state; 1153 int ppid; 1154 int pgrp; 1155 int session; 1156 int nr; 1157 int tpgrp; 1158 unsigned long flags; 1159 unsigned long minflt; 1160 unsigned long cminflt; 1161 unsigned long majflt; 1162 unsigned long cmajflt; 1163 unsigned long utime; 1164 unsigned long stime; 1165 long cutime; 1166 long cstime; 1167 long prio; 1168 long nice; 1169 long junk; 1170 long it_real; 1171 uintptr_t start; 1172 uintptr_t vsize; 1173 intptr_t rss; 1174 uintptr_t rsslim; 1175 uintptr_t scodes; 1176 uintptr_t ecode; 1177 int i; 1178 1179 // Figure what the primordial thread stack base is. Code is inspired 1180 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1181 // followed by command name surrounded by parentheses, state, etc. 1182 char stat[2048]; 1183 int statlen; 1184 1185 fp = fopen("/proc/self/stat", "r"); 1186 if (fp) { 1187 statlen = fread(stat, 1, 2047, fp); 1188 stat[statlen] = '\0'; 1189 fclose(fp); 1190 1191 // Skip pid and the command string. Note that we could be dealing with 1192 // weird command names, e.g. user could decide to rename java launcher 1193 // to "java 1.4.2 :)", then the stat file would look like 1194 // 1234 (java 1.4.2 :)) R ... ... 1195 // We don't really need to know the command string, just find the last 1196 // occurrence of ")" and then start parsing from there. See bug 4726580. 1197 char * s = strrchr(stat, ')'); 1198 1199 i = 0; 1200 if (s) { 1201 // Skip blank chars 1202 do { s++; } while (s && isspace(*s)); 1203 1204#define _UFM UINTX_FORMAT 1205#define _DFM INTX_FORMAT 1206 1207 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1208 // 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 1209 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, 1210 &state, // 3 %c 1211 &ppid, // 4 %d 1212 &pgrp, // 5 %d 1213 &session, // 6 %d 1214 &nr, // 7 %d 1215 &tpgrp, // 8 %d 1216 &flags, // 9 %lu 1217 &minflt, // 10 %lu 1218 &cminflt, // 11 %lu 1219 &majflt, // 12 %lu 1220 &cmajflt, // 13 %lu 1221 &utime, // 14 %lu 1222 &stime, // 15 %lu 1223 &cutime, // 16 %ld 1224 &cstime, // 17 %ld 1225 &prio, // 18 %ld 1226 &nice, // 19 %ld 1227 &junk, // 20 %ld 1228 &it_real, // 21 %ld 1229 &start, // 22 UINTX_FORMAT 1230 &vsize, // 23 UINTX_FORMAT 1231 &rss, // 24 INTX_FORMAT 1232 &rsslim, // 25 UINTX_FORMAT 1233 &scodes, // 26 UINTX_FORMAT 1234 &ecode, // 27 UINTX_FORMAT 1235 &stack_start); // 28 UINTX_FORMAT 1236 } 1237 1238#undef _UFM 1239#undef _DFM 1240 1241 if (i != 28 - 2) { 1242 assert(false, "Bad conversion from /proc/self/stat"); 1243 // product mode - assume we are the initial thread, good luck in the 1244 // embedded case. 1245 warning("Can't detect initial thread stack location - bad conversion"); 1246 stack_start = (uintptr_t) &rlim; 1247 } 1248 } else { 1249 // For some reason we can't open /proc/self/stat (for example, running on 1250 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1251 // most cases, so don't abort: 1252 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1253 stack_start = (uintptr_t) &rlim; 1254 } 1255 } 1256 1257 // Now we have a pointer (stack_start) very close to the stack top, the 1258 // next thing to do is to figure out the exact location of stack top. We 1259 // can find out the virtual memory area that contains stack_start by 1260 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1261 // and its upper limit is the real stack top. (again, this would fail if 1262 // running inside chroot, because /proc may not exist.) 1263 1264 uintptr_t stack_top; 1265 address low, high; 1266 if (find_vma((address)stack_start, &low, &high)) { 1267 // success, "high" is the true stack top. (ignore "low", because initial 1268 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1269 stack_top = (uintptr_t)high; 1270 } else { 1271 // failed, likely because /proc/self/maps does not exist 1272 warning("Can't detect initial thread stack location - find_vma failed"); 1273 // best effort: stack_start is normally within a few pages below the real 1274 // stack top, use it as stack top, and reduce stack size so we won't put 1275 // guard page outside stack. 1276 stack_top = stack_start; 1277 stack_size -= 16 * page_size(); 1278 } 1279 1280 // stack_top could be partially down the page so align it 1281 stack_top = align_size_up(stack_top, page_size()); 1282 1283 if (max_size && stack_size > max_size) { 1284 _initial_thread_stack_size = max_size; 1285 } else { 1286 _initial_thread_stack_size = stack_size; 1287 } 1288 1289 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1290 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1291} 1292 1293//////////////////////////////////////////////////////////////////////////////// 1294// time support 1295 1296// Time since start-up in seconds to a fine granularity. 1297// Used by VMSelfDestructTimer and the MemProfiler. 1298double os::elapsedTime() { 1299 1300 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1301} 1302 1303jlong os::elapsed_counter() { 1304 return javaTimeNanos() - initial_time_count; 1305} 1306 1307jlong os::elapsed_frequency() { 1308 return NANOSECS_PER_SEC; // nanosecond resolution 1309} 1310 1311bool os::supports_vtime() { return true; } 1312bool os::enable_vtime() { return false; } 1313bool os::vtime_enabled() { return false; } 1314 1315double os::elapsedVTime() { 1316 struct rusage usage; 1317 int retval = getrusage(RUSAGE_THREAD, &usage); 1318 if (retval == 0) { 1319 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); 1320 } else { 1321 // better than nothing, but not much 1322 return elapsedTime(); 1323 } 1324} 1325 1326jlong os::javaTimeMillis() { 1327 timeval time; 1328 int status = gettimeofday(&time, NULL); 1329 assert(status != -1, "linux error"); 1330 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1331} 1332 1333#ifndef CLOCK_MONOTONIC 1334 #define CLOCK_MONOTONIC (1) 1335#endif 1336 1337void os::Linux::clock_init() { 1338 // we do dlopen's in this particular order due to bug in linux 1339 // dynamical loader (see 6348968) leading to crash on exit 1340 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1341 if (handle == NULL) { 1342 handle = dlopen("librt.so", RTLD_LAZY); 1343 } 1344 1345 if (handle) { 1346 int (*clock_getres_func)(clockid_t, struct timespec*) = 1347 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1348 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1349 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1350 if (clock_getres_func && clock_gettime_func) { 1351 // See if monotonic clock is supported by the kernel. Note that some 1352 // early implementations simply return kernel jiffies (updated every 1353 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1354 // for nano time (though the monotonic property is still nice to have). 1355 // It's fixed in newer kernels, however clock_getres() still returns 1356 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1357 // resolution for now. Hopefully as people move to new kernels, this 1358 // won't be a problem. 1359 struct timespec res; 1360 struct timespec tp; 1361 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1362 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1363 // yes, monotonic clock is supported 1364 _clock_gettime = clock_gettime_func; 1365 return; 1366 } else { 1367 // close librt if there is no monotonic clock 1368 dlclose(handle); 1369 } 1370 } 1371 } 1372 warning("No monotonic clock was available - timed services may " \ 1373 "be adversely affected if the time-of-day clock changes"); 1374} 1375 1376#ifndef SYS_clock_getres 1377 #if defined(IA32) || defined(AMD64) 1378 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1379 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1380 #else 1381 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1382 #define sys_clock_getres(x,y) -1 1383 #endif 1384#else 1385 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1386#endif 1387 1388void os::Linux::fast_thread_clock_init() { 1389 if (!UseLinuxPosixThreadCPUClocks) { 1390 return; 1391 } 1392 clockid_t clockid; 1393 struct timespec tp; 1394 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1395 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1396 1397 // Switch to using fast clocks for thread cpu time if 1398 // the sys_clock_getres() returns 0 error code. 1399 // Note, that some kernels may support the current thread 1400 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1401 // returned by the pthread_getcpuclockid(). 1402 // If the fast Posix clocks are supported then the sys_clock_getres() 1403 // must return at least tp.tv_sec == 0 which means a resolution 1404 // better than 1 sec. This is extra check for reliability. 1405 1406 if (pthread_getcpuclockid_func && 1407 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1408 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1409 _supports_fast_thread_cpu_time = true; 1410 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1411 } 1412} 1413 1414jlong os::javaTimeNanos() { 1415 if (os::supports_monotonic_clock()) { 1416 struct timespec tp; 1417 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1418 assert(status == 0, "gettime error"); 1419 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1420 return result; 1421 } else { 1422 timeval time; 1423 int status = gettimeofday(&time, NULL); 1424 assert(status != -1, "linux error"); 1425 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1426 return 1000 * usecs; 1427 } 1428} 1429 1430void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1431 if (os::supports_monotonic_clock()) { 1432 info_ptr->max_value = ALL_64_BITS; 1433 1434 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1435 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1436 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1437 } else { 1438 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1439 info_ptr->max_value = ALL_64_BITS; 1440 1441 // gettimeofday is a real time clock so it skips 1442 info_ptr->may_skip_backward = true; 1443 info_ptr->may_skip_forward = true; 1444 } 1445 1446 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1447} 1448 1449// Return the real, user, and system times in seconds from an 1450// arbitrary fixed point in the past. 1451bool os::getTimesSecs(double* process_real_time, 1452 double* process_user_time, 1453 double* process_system_time) { 1454 struct tms ticks; 1455 clock_t real_ticks = times(&ticks); 1456 1457 if (real_ticks == (clock_t) (-1)) { 1458 return false; 1459 } else { 1460 double ticks_per_second = (double) clock_tics_per_sec; 1461 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1462 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1463 *process_real_time = ((double) real_ticks) / ticks_per_second; 1464 1465 return true; 1466 } 1467} 1468 1469 1470char * os::local_time_string(char *buf, size_t buflen) { 1471 struct tm t; 1472 time_t long_time; 1473 time(&long_time); 1474 localtime_r(&long_time, &t); 1475 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1476 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1477 t.tm_hour, t.tm_min, t.tm_sec); 1478 return buf; 1479} 1480 1481struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1482 return localtime_r(clock, res); 1483} 1484 1485//////////////////////////////////////////////////////////////////////////////// 1486// runtime exit support 1487 1488// Note: os::shutdown() might be called very early during initialization, or 1489// called from signal handler. Before adding something to os::shutdown(), make 1490// sure it is async-safe and can handle partially initialized VM. 1491void os::shutdown() { 1492 1493 // allow PerfMemory to attempt cleanup of any persistent resources 1494 perfMemory_exit(); 1495 1496 // needs to remove object in file system 1497 AttachListener::abort(); 1498 1499 // flush buffered output, finish log files 1500 ostream_abort(); 1501 1502 // Check for abort hook 1503 abort_hook_t abort_hook = Arguments::abort_hook(); 1504 if (abort_hook != NULL) { 1505 abort_hook(); 1506 } 1507 1508} 1509 1510// Note: os::abort() might be called very early during initialization, or 1511// called from signal handler. Before adding something to os::abort(), make 1512// sure it is async-safe and can handle partially initialized VM. 1513void os::abort(bool dump_core) { 1514 os::shutdown(); 1515 if (dump_core) { 1516#ifndef PRODUCT 1517 fdStream out(defaultStream::output_fd()); 1518 out.print_raw("Current thread is "); 1519 char buf[16]; 1520 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1521 out.print_raw_cr(buf); 1522 out.print_raw_cr("Dumping core ..."); 1523#endif 1524 ::abort(); // dump core 1525 } 1526 1527 ::exit(1); 1528} 1529 1530// Die immediately, no exit hook, no abort hook, no cleanup. 1531void os::die() { 1532 // _exit() on LinuxThreads only kills current thread 1533 ::abort(); 1534} 1535 1536 1537// This method is a copy of JDK's sysGetLastErrorString 1538// from src/solaris/hpi/src/system_md.c 1539 1540size_t os::lasterror(char *buf, size_t len) { 1541 if (errno == 0) return 0; 1542 1543 const char *s = ::strerror(errno); 1544 size_t n = ::strlen(s); 1545 if (n >= len) { 1546 n = len - 1; 1547 } 1548 ::strncpy(buf, s, n); 1549 buf[n] = '\0'; 1550 return n; 1551} 1552 1553intx os::current_thread_id() { return (intx)pthread_self(); } 1554int os::current_process_id() { 1555 1556 // Under the old linux thread library, linux gives each thread 1557 // its own process id. Because of this each thread will return 1558 // a different pid if this method were to return the result 1559 // of getpid(2). Linux provides no api that returns the pid 1560 // of the launcher thread for the vm. This implementation 1561 // returns a unique pid, the pid of the launcher thread 1562 // that starts the vm 'process'. 1563 1564 // Under the NPTL, getpid() returns the same pid as the 1565 // launcher thread rather than a unique pid per thread. 1566 // Use gettid() if you want the old pre NPTL behaviour. 1567 1568 // if you are looking for the result of a call to getpid() that 1569 // returns a unique pid for the calling thread, then look at the 1570 // OSThread::thread_id() method in osThread_linux.hpp file 1571 1572 return (int)(_initial_pid ? _initial_pid : getpid()); 1573} 1574 1575// DLL functions 1576 1577const char* os::dll_file_extension() { return ".so"; } 1578 1579// This must be hard coded because it's the system's temporary 1580// directory not the java application's temp directory, ala java.io.tmpdir. 1581const char* os::get_temp_directory() { return "/tmp"; } 1582 1583static bool file_exists(const char* filename) { 1584 struct stat statbuf; 1585 if (filename == NULL || strlen(filename) == 0) { 1586 return false; 1587 } 1588 return os::stat(filename, &statbuf) == 0; 1589} 1590 1591bool os::dll_build_name(char* buffer, size_t buflen, 1592 const char* pname, const char* fname) { 1593 bool retval = false; 1594 // Copied from libhpi 1595 const size_t pnamelen = pname ? strlen(pname) : 0; 1596 1597 // Return error on buffer overflow. 1598 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1599 return retval; 1600 } 1601 1602 if (pnamelen == 0) { 1603 snprintf(buffer, buflen, "lib%s.so", fname); 1604 retval = true; 1605 } else if (strchr(pname, *os::path_separator()) != NULL) { 1606 int n; 1607 char** pelements = split_path(pname, &n); 1608 if (pelements == NULL) { 1609 return false; 1610 } 1611 for (int i = 0; i < n; i++) { 1612 // Really shouldn't be NULL, but check can't hurt 1613 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1614 continue; // skip the empty path values 1615 } 1616 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1617 if (file_exists(buffer)) { 1618 retval = true; 1619 break; 1620 } 1621 } 1622 // release the storage 1623 for (int i = 0; i < n; i++) { 1624 if (pelements[i] != NULL) { 1625 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1626 } 1627 } 1628 if (pelements != NULL) { 1629 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1630 } 1631 } else { 1632 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1633 retval = true; 1634 } 1635 return retval; 1636} 1637 1638// check if addr is inside libjvm.so 1639bool os::address_is_in_vm(address addr) { 1640 static address libjvm_base_addr; 1641 Dl_info dlinfo; 1642 1643 if (libjvm_base_addr == NULL) { 1644 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1645 libjvm_base_addr = (address)dlinfo.dli_fbase; 1646 } 1647 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1648 } 1649 1650 if (dladdr((void *)addr, &dlinfo) != 0) { 1651 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1652 } 1653 1654 return false; 1655} 1656 1657bool os::dll_address_to_function_name(address addr, char *buf, 1658 int buflen, int *offset) { 1659 // buf is not optional, but offset is optional 1660 assert(buf != NULL, "sanity check"); 1661 1662 Dl_info dlinfo; 1663 1664 if (dladdr((void*)addr, &dlinfo) != 0) { 1665 // see if we have a matching symbol 1666 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1667 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1668 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1669 } 1670 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1671 return true; 1672 } 1673 // no matching symbol so try for just file info 1674 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1675 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1676 buf, buflen, offset, dlinfo.dli_fname)) { 1677 return true; 1678 } 1679 } 1680 } 1681 1682 buf[0] = '\0'; 1683 if (offset != NULL) *offset = -1; 1684 return false; 1685} 1686 1687struct _address_to_library_name { 1688 address addr; // input : memory address 1689 size_t buflen; // size of fname 1690 char* fname; // output: library name 1691 address base; // library base addr 1692}; 1693 1694static int address_to_library_name_callback(struct dl_phdr_info *info, 1695 size_t size, void *data) { 1696 int i; 1697 bool found = false; 1698 address libbase = NULL; 1699 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1700 1701 // iterate through all loadable segments 1702 for (i = 0; i < info->dlpi_phnum; i++) { 1703 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1704 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1705 // base address of a library is the lowest address of its loaded 1706 // segments. 1707 if (libbase == NULL || libbase > segbase) { 1708 libbase = segbase; 1709 } 1710 // see if 'addr' is within current segment 1711 if (segbase <= d->addr && 1712 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1713 found = true; 1714 } 1715 } 1716 } 1717 1718 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1719 // so dll_address_to_library_name() can fall through to use dladdr() which 1720 // can figure out executable name from argv[0]. 1721 if (found && info->dlpi_name && info->dlpi_name[0]) { 1722 d->base = libbase; 1723 if (d->fname) { 1724 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1725 } 1726 return 1; 1727 } 1728 return 0; 1729} 1730 1731bool os::dll_address_to_library_name(address addr, char* buf, 1732 int buflen, int* offset) { 1733 // buf is not optional, but offset is optional 1734 assert(buf != NULL, "sanity check"); 1735 1736 Dl_info dlinfo; 1737 struct _address_to_library_name data; 1738 1739 // There is a bug in old glibc dladdr() implementation that it could resolve 1740 // to wrong library name if the .so file has a base address != NULL. Here 1741 // we iterate through the program headers of all loaded libraries to find 1742 // out which library 'addr' really belongs to. This workaround can be 1743 // removed once the minimum requirement for glibc is moved to 2.3.x. 1744 data.addr = addr; 1745 data.fname = buf; 1746 data.buflen = buflen; 1747 data.base = NULL; 1748 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1749 1750 if (rslt) { 1751 // buf already contains library name 1752 if (offset) *offset = addr - data.base; 1753 return true; 1754 } 1755 if (dladdr((void*)addr, &dlinfo) != 0) { 1756 if (dlinfo.dli_fname != NULL) { 1757 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1758 } 1759 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1760 *offset = addr - (address)dlinfo.dli_fbase; 1761 } 1762 return true; 1763 } 1764 1765 buf[0] = '\0'; 1766 if (offset) *offset = -1; 1767 return false; 1768} 1769 1770// Loads .dll/.so and 1771// in case of error it checks if .dll/.so was built for the 1772// same architecture as Hotspot is running on 1773 1774 1775// Remember the stack's state. The Linux dynamic linker will change 1776// the stack to 'executable' at most once, so we must safepoint only once. 1777bool os::Linux::_stack_is_executable = false; 1778 1779// VM operation that loads a library. This is necessary if stack protection 1780// of the Java stacks can be lost during loading the library. If we 1781// do not stop the Java threads, they can stack overflow before the stacks 1782// are protected again. 1783class VM_LinuxDllLoad: public VM_Operation { 1784 private: 1785 const char *_filename; 1786 char *_ebuf; 1787 int _ebuflen; 1788 void *_lib; 1789 public: 1790 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1791 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1792 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1793 void doit() { 1794 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1795 os::Linux::_stack_is_executable = true; 1796 } 1797 void* loaded_library() { return _lib; } 1798}; 1799 1800void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1801 void * result = NULL; 1802 bool load_attempted = false; 1803 1804 // Check whether the library to load might change execution rights 1805 // of the stack. If they are changed, the protection of the stack 1806 // guard pages will be lost. We need a safepoint to fix this. 1807 // 1808 // See Linux man page execstack(8) for more info. 1809 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1810 ElfFile ef(filename); 1811 if (!ef.specifies_noexecstack()) { 1812 if (!is_init_completed()) { 1813 os::Linux::_stack_is_executable = true; 1814 // This is OK - No Java threads have been created yet, and hence no 1815 // stack guard pages to fix. 1816 // 1817 // This should happen only when you are building JDK7 using a very 1818 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1819 // 1820 // Dynamic loader will make all stacks executable after 1821 // this function returns, and will not do that again. 1822 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1823 } else { 1824 warning("You have loaded library %s which might have disabled stack guard. " 1825 "The VM will try to fix the stack guard now.\n" 1826 "It's highly recommended that you fix the library with " 1827 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1828 filename); 1829 1830 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1831 JavaThread *jt = JavaThread::current(); 1832 if (jt->thread_state() != _thread_in_native) { 1833 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1834 // that requires ExecStack. Cannot enter safe point. Let's give up. 1835 warning("Unable to fix stack guard. Giving up."); 1836 } else { 1837 if (!LoadExecStackDllInVMThread) { 1838 // This is for the case where the DLL has an static 1839 // constructor function that executes JNI code. We cannot 1840 // load such DLLs in the VMThread. 1841 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1842 } 1843 1844 ThreadInVMfromNative tiv(jt); 1845 debug_only(VMNativeEntryWrapper vew;) 1846 1847 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1848 VMThread::execute(&op); 1849 if (LoadExecStackDllInVMThread) { 1850 result = op.loaded_library(); 1851 } 1852 load_attempted = true; 1853 } 1854 } 1855 } 1856 } 1857 1858 if (!load_attempted) { 1859 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1860 } 1861 1862 if (result != NULL) { 1863 // Successful loading 1864 return result; 1865 } 1866 1867 Elf32_Ehdr elf_head; 1868 int diag_msg_max_length=ebuflen-strlen(ebuf); 1869 char* diag_msg_buf=ebuf+strlen(ebuf); 1870 1871 if (diag_msg_max_length==0) { 1872 // No more space in ebuf for additional diagnostics message 1873 return NULL; 1874 } 1875 1876 1877 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1878 1879 if (file_descriptor < 0) { 1880 // Can't open library, report dlerror() message 1881 return NULL; 1882 } 1883 1884 bool failed_to_read_elf_head= 1885 (sizeof(elf_head)!= 1886 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1887 1888 ::close(file_descriptor); 1889 if (failed_to_read_elf_head) { 1890 // file i/o error - report dlerror() msg 1891 return NULL; 1892 } 1893 1894 typedef struct { 1895 Elf32_Half code; // Actual value as defined in elf.h 1896 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1897 char elf_class; // 32 or 64 bit 1898 char endianess; // MSB or LSB 1899 char* name; // String representation 1900 } arch_t; 1901 1902#ifndef EM_486 1903 #define EM_486 6 /* Intel 80486 */ 1904#endif 1905#ifndef EM_AARCH64 1906 #define EM_AARCH64 183 /* ARM AARCH64 */ 1907#endif 1908 1909 static const arch_t arch_array[]={ 1910 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1911 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1912 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1913 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1914 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1915 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1916 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1917 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1918#if defined(VM_LITTLE_ENDIAN) 1919 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"}, 1920#else 1921 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1922#endif 1923 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1924 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1925 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1926 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1927 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1928 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1929 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1930 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1931 }; 1932 1933#if (defined IA32) 1934 static Elf32_Half running_arch_code=EM_386; 1935#elif (defined AMD64) 1936 static Elf32_Half running_arch_code=EM_X86_64; 1937#elif (defined IA64) 1938 static Elf32_Half running_arch_code=EM_IA_64; 1939#elif (defined __sparc) && (defined _LP64) 1940 static Elf32_Half running_arch_code=EM_SPARCV9; 1941#elif (defined __sparc) && (!defined _LP64) 1942 static Elf32_Half running_arch_code=EM_SPARC; 1943#elif (defined __powerpc64__) 1944 static Elf32_Half running_arch_code=EM_PPC64; 1945#elif (defined __powerpc__) 1946 static Elf32_Half running_arch_code=EM_PPC; 1947#elif (defined ARM) 1948 static Elf32_Half running_arch_code=EM_ARM; 1949#elif (defined S390) 1950 static Elf32_Half running_arch_code=EM_S390; 1951#elif (defined ALPHA) 1952 static Elf32_Half running_arch_code=EM_ALPHA; 1953#elif (defined MIPSEL) 1954 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1955#elif (defined PARISC) 1956 static Elf32_Half running_arch_code=EM_PARISC; 1957#elif (defined MIPS) 1958 static Elf32_Half running_arch_code=EM_MIPS; 1959#elif (defined M68K) 1960 static Elf32_Half running_arch_code=EM_68K; 1961#elif (defined AARCH64) 1962 static Elf32_Half running_arch_code=EM_AARCH64; 1963#else 1964 #error Method os::dll_load requires that one of following is defined:\ 1965 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64 1966#endif 1967 1968 // Identify compatability class for VM's architecture and library's architecture 1969 // Obtain string descriptions for architectures 1970 1971 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1972 int running_arch_index=-1; 1973 1974 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1975 if (running_arch_code == arch_array[i].code) { 1976 running_arch_index = i; 1977 } 1978 if (lib_arch.code == arch_array[i].code) { 1979 lib_arch.compat_class = arch_array[i].compat_class; 1980 lib_arch.name = arch_array[i].name; 1981 } 1982 } 1983 1984 assert(running_arch_index != -1, 1985 "Didn't find running architecture code (running_arch_code) in arch_array"); 1986 if (running_arch_index == -1) { 1987 // Even though running architecture detection failed 1988 // we may still continue with reporting dlerror() message 1989 return NULL; 1990 } 1991 1992 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1993 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1994 return NULL; 1995 } 1996 1997#ifndef S390 1998 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1999 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 2000 return NULL; 2001 } 2002#endif // !S390 2003 2004 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2005 if (lib_arch.name!=NULL) { 2006 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2007 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2008 lib_arch.name, arch_array[running_arch_index].name); 2009 } else { 2010 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2011 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2012 lib_arch.code, 2013 arch_array[running_arch_index].name); 2014 } 2015 } 2016 2017 return NULL; 2018} 2019 2020void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 2021 int ebuflen) { 2022 void * result = ::dlopen(filename, RTLD_LAZY); 2023 if (result == NULL) { 2024 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2025 ebuf[ebuflen-1] = '\0'; 2026 } 2027 return result; 2028} 2029 2030void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 2031 int ebuflen) { 2032 void * result = NULL; 2033 if (LoadExecStackDllInVMThread) { 2034 result = dlopen_helper(filename, ebuf, ebuflen); 2035 } 2036 2037 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2038 // library that requires an executable stack, or which does not have this 2039 // stack attribute set, dlopen changes the stack attribute to executable. The 2040 // read protection of the guard pages gets lost. 2041 // 2042 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2043 // may have been queued at the same time. 2044 2045 if (!_stack_is_executable) { 2046 JavaThread *jt = Threads::first(); 2047 2048 while (jt) { 2049 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2050 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2051 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2052 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2053 warning("Attempt to reguard stack yellow zone failed."); 2054 } 2055 } 2056 jt = jt->next(); 2057 } 2058 } 2059 2060 return result; 2061} 2062 2063// glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2064// chances are you might want to run the generated bits against glibc-2.0 2065// libdl.so, so always use locking for any version of glibc. 2066// 2067void* os::dll_lookup(void* handle, const char* name) { 2068 pthread_mutex_lock(&dl_mutex); 2069 void* res = dlsym(handle, name); 2070 pthread_mutex_unlock(&dl_mutex); 2071 return res; 2072} 2073 2074void* os::get_default_process_handle() { 2075 return (void*)::dlopen(NULL, RTLD_LAZY); 2076} 2077 2078static bool _print_ascii_file(const char* filename, outputStream* st) { 2079 int fd = ::open(filename, O_RDONLY); 2080 if (fd == -1) { 2081 return false; 2082 } 2083 2084 char buf[32]; 2085 int bytes; 2086 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2087 st->print_raw(buf, bytes); 2088 } 2089 2090 ::close(fd); 2091 2092 return true; 2093} 2094 2095void os::print_dll_info(outputStream *st) { 2096 st->print_cr("Dynamic libraries:"); 2097 2098 char fname[32]; 2099 pid_t pid = os::Linux::gettid(); 2100 2101 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2102 2103 if (!_print_ascii_file(fname, st)) { 2104 st->print("Can not get library information for pid = %d\n", pid); 2105 } 2106} 2107 2108int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 2109 FILE *procmapsFile = NULL; 2110 2111 // Open the procfs maps file for the current process 2112 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 2113 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 2114 char line[PATH_MAX + 100]; 2115 2116 // Read line by line from 'file' 2117 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 2118 u8 base, top, offset, inode; 2119 char permissions[5]; 2120 char device[6]; 2121 char name[PATH_MAX + 1]; 2122 2123 // Parse fields from line 2124 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name); 2125 2126 // Filter by device id '00:00' so that we only get file system mapped files. 2127 if (strcmp(device, "00:00") != 0) { 2128 2129 // Call callback with the fields of interest 2130 if(callback(name, (address)base, (address)top, param)) { 2131 // Oops abort, callback aborted 2132 fclose(procmapsFile); 2133 return 1; 2134 } 2135 } 2136 } 2137 fclose(procmapsFile); 2138 } 2139 return 0; 2140} 2141 2142void os::print_os_info_brief(outputStream* st) { 2143 os::Linux::print_distro_info(st); 2144 2145 os::Posix::print_uname_info(st); 2146 2147 os::Linux::print_libversion_info(st); 2148 2149} 2150 2151void os::print_os_info(outputStream* st) { 2152 st->print("OS:"); 2153 2154 os::Linux::print_distro_info(st); 2155 2156 os::Posix::print_uname_info(st); 2157 2158 // Print warning if unsafe chroot environment detected 2159 if (unsafe_chroot_detected) { 2160 st->print("WARNING!! "); 2161 st->print_cr("%s", unstable_chroot_error); 2162 } 2163 2164 os::Linux::print_libversion_info(st); 2165 2166 os::Posix::print_rlimit_info(st); 2167 2168 os::Posix::print_load_average(st); 2169 2170 os::Linux::print_full_memory_info(st); 2171} 2172 2173// Try to identify popular distros. 2174// Most Linux distributions have a /etc/XXX-release file, which contains 2175// the OS version string. Newer Linux distributions have a /etc/lsb-release 2176// file that also contains the OS version string. Some have more than one 2177// /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2178// /etc/redhat-release.), so the order is important. 2179// Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2180// their own specific XXX-release file as well as a redhat-release file. 2181// Because of this the XXX-release file needs to be searched for before the 2182// redhat-release file. 2183// Since Red Hat has a lsb-release file that is not very descriptive the 2184// search for redhat-release needs to be before lsb-release. 2185// Since the lsb-release file is the new standard it needs to be searched 2186// before the older style release files. 2187// Searching system-release (Red Hat) and os-release (other Linuxes) are a 2188// next to last resort. The os-release file is a new standard that contains 2189// distribution information and the system-release file seems to be an old 2190// standard that has been replaced by the lsb-release and os-release files. 2191// Searching for the debian_version file is the last resort. It contains 2192// an informative string like "6.0.6" or "wheezy/sid". Because of this 2193// "Debian " is printed before the contents of the debian_version file. 2194void os::Linux::print_distro_info(outputStream* st) { 2195 if (!_print_ascii_file("/etc/oracle-release", st) && 2196 !_print_ascii_file("/etc/mandriva-release", st) && 2197 !_print_ascii_file("/etc/mandrake-release", st) && 2198 !_print_ascii_file("/etc/sun-release", st) && 2199 !_print_ascii_file("/etc/redhat-release", st) && 2200 !_print_ascii_file("/etc/lsb-release", st) && 2201 !_print_ascii_file("/etc/SuSE-release", st) && 2202 !_print_ascii_file("/etc/turbolinux-release", st) && 2203 !_print_ascii_file("/etc/gentoo-release", st) && 2204 !_print_ascii_file("/etc/ltib-release", st) && 2205 !_print_ascii_file("/etc/angstrom-version", st) && 2206 !_print_ascii_file("/etc/system-release", st) && 2207 !_print_ascii_file("/etc/os-release", st)) { 2208 2209 if (file_exists("/etc/debian_version")) { 2210 st->print("Debian "); 2211 _print_ascii_file("/etc/debian_version", st); 2212 } else { 2213 st->print("Linux"); 2214 } 2215 } 2216 st->cr(); 2217} 2218 2219void os::Linux::print_libversion_info(outputStream* st) { 2220 // libc, pthread 2221 st->print("libc:"); 2222 st->print("%s ", os::Linux::glibc_version()); 2223 st->print("%s ", os::Linux::libpthread_version()); 2224 if (os::Linux::is_LinuxThreads()) { 2225 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2226 } 2227 st->cr(); 2228} 2229 2230void os::Linux::print_full_memory_info(outputStream* st) { 2231 st->print("\n/proc/meminfo:\n"); 2232 _print_ascii_file("/proc/meminfo", st); 2233 st->cr(); 2234} 2235 2236void os::print_memory_info(outputStream* st) { 2237 2238 st->print("Memory:"); 2239 st->print(" %dk page", os::vm_page_size()>>10); 2240 2241 // values in struct sysinfo are "unsigned long" 2242 struct sysinfo si; 2243 sysinfo(&si); 2244 2245 st->print(", physical " UINT64_FORMAT "k", 2246 os::physical_memory() >> 10); 2247 st->print("(" UINT64_FORMAT "k free)", 2248 os::available_memory() >> 10); 2249 st->print(", swap " UINT64_FORMAT "k", 2250 ((jlong)si.totalswap * si.mem_unit) >> 10); 2251 st->print("(" UINT64_FORMAT "k free)", 2252 ((jlong)si.freeswap * si.mem_unit) >> 10); 2253 st->cr(); 2254} 2255 2256void os::pd_print_cpu_info(outputStream* st) { 2257 st->print("\n/proc/cpuinfo:\n"); 2258 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2259 st->print(" <Not Available>"); 2260 } 2261 st->cr(); 2262} 2263 2264void os::print_siginfo(outputStream* st, void* siginfo) { 2265 const siginfo_t* si = (const siginfo_t*)siginfo; 2266 2267 os::Posix::print_siginfo_brief(st, si); 2268#if INCLUDE_CDS 2269 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2270 UseSharedSpaces) { 2271 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2272 if (mapinfo->is_in_shared_space(si->si_addr)) { 2273 st->print("\n\nError accessing class data sharing archive." \ 2274 " Mapped file inaccessible during execution, " \ 2275 " possible disk/network problem."); 2276 } 2277 } 2278#endif 2279 st->cr(); 2280} 2281 2282 2283static void print_signal_handler(outputStream* st, int sig, 2284 char* buf, size_t buflen); 2285 2286void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2287 st->print_cr("Signal Handlers:"); 2288 print_signal_handler(st, SIGSEGV, buf, buflen); 2289 print_signal_handler(st, SIGBUS , buf, buflen); 2290 print_signal_handler(st, SIGFPE , buf, buflen); 2291 print_signal_handler(st, SIGPIPE, buf, buflen); 2292 print_signal_handler(st, SIGXFSZ, buf, buflen); 2293 print_signal_handler(st, SIGILL , buf, buflen); 2294 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2295 print_signal_handler(st, SR_signum, buf, buflen); 2296 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2297 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2298 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2299 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2300#if defined(PPC64) 2301 print_signal_handler(st, SIGTRAP, buf, buflen); 2302#endif 2303} 2304 2305static char saved_jvm_path[MAXPATHLEN] = {0}; 2306 2307// Find the full path to the current module, libjvm.so 2308void os::jvm_path(char *buf, jint buflen) { 2309 // Error checking. 2310 if (buflen < MAXPATHLEN) { 2311 assert(false, "must use a large-enough buffer"); 2312 buf[0] = '\0'; 2313 return; 2314 } 2315 // Lazy resolve the path to current module. 2316 if (saved_jvm_path[0] != 0) { 2317 strcpy(buf, saved_jvm_path); 2318 return; 2319 } 2320 2321 char dli_fname[MAXPATHLEN]; 2322 bool ret = dll_address_to_library_name( 2323 CAST_FROM_FN_PTR(address, os::jvm_path), 2324 dli_fname, sizeof(dli_fname), NULL); 2325 assert(ret, "cannot locate libjvm"); 2326 char *rp = NULL; 2327 if (ret && dli_fname[0] != '\0') { 2328 rp = realpath(dli_fname, buf); 2329 } 2330 if (rp == NULL) { 2331 return; 2332 } 2333 2334 if (Arguments::sun_java_launcher_is_altjvm()) { 2335 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2336 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2337 // If "/jre/lib/" appears at the right place in the string, then 2338 // assume we are installed in a JDK and we're done. Otherwise, check 2339 // for a JAVA_HOME environment variable and fix up the path so it 2340 // looks like libjvm.so is installed there (append a fake suffix 2341 // hotspot/libjvm.so). 2342 const char *p = buf + strlen(buf) - 1; 2343 for (int count = 0; p > buf && count < 5; ++count) { 2344 for (--p; p > buf && *p != '/'; --p) 2345 /* empty */ ; 2346 } 2347 2348 if (strncmp(p, "/jre/lib/", 9) != 0) { 2349 // Look for JAVA_HOME in the environment. 2350 char* java_home_var = ::getenv("JAVA_HOME"); 2351 if (java_home_var != NULL && java_home_var[0] != 0) { 2352 char* jrelib_p; 2353 int len; 2354 2355 // Check the current module name "libjvm.so". 2356 p = strrchr(buf, '/'); 2357 if (p == NULL) { 2358 return; 2359 } 2360 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2361 2362 rp = realpath(java_home_var, buf); 2363 if (rp == NULL) { 2364 return; 2365 } 2366 2367 // determine if this is a legacy image or modules image 2368 // modules image doesn't have "jre" subdirectory 2369 len = strlen(buf); 2370 assert(len < buflen, "Ran out of buffer room"); 2371 jrelib_p = buf + len; 2372 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2373 if (0 != access(buf, F_OK)) { 2374 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2375 } 2376 2377 if (0 == access(buf, F_OK)) { 2378 // Use current module name "libjvm.so" 2379 len = strlen(buf); 2380 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2381 } else { 2382 // Go back to path of .so 2383 rp = realpath(dli_fname, buf); 2384 if (rp == NULL) { 2385 return; 2386 } 2387 } 2388 } 2389 } 2390 } 2391 2392 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2393 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2394} 2395 2396void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2397 // no prefix required, not even "_" 2398} 2399 2400void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2401 // no suffix required 2402} 2403 2404//////////////////////////////////////////////////////////////////////////////// 2405// sun.misc.Signal support 2406 2407static volatile jint sigint_count = 0; 2408 2409static void UserHandler(int sig, void *siginfo, void *context) { 2410 // 4511530 - sem_post is serialized and handled by the manager thread. When 2411 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2412 // don't want to flood the manager thread with sem_post requests. 2413 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2414 return; 2415 } 2416 2417 // Ctrl-C is pressed during error reporting, likely because the error 2418 // handler fails to abort. Let VM die immediately. 2419 if (sig == SIGINT && is_error_reported()) { 2420 os::die(); 2421 } 2422 2423 os::signal_notify(sig); 2424} 2425 2426void* os::user_handler() { 2427 return CAST_FROM_FN_PTR(void*, UserHandler); 2428} 2429 2430class Semaphore : public StackObj { 2431 public: 2432 Semaphore(); 2433 ~Semaphore(); 2434 void signal(); 2435 void wait(); 2436 bool trywait(); 2437 bool timedwait(unsigned int sec, int nsec); 2438 private: 2439 sem_t _semaphore; 2440}; 2441 2442Semaphore::Semaphore() { 2443 sem_init(&_semaphore, 0, 0); 2444} 2445 2446Semaphore::~Semaphore() { 2447 sem_destroy(&_semaphore); 2448} 2449 2450void Semaphore::signal() { 2451 sem_post(&_semaphore); 2452} 2453 2454void Semaphore::wait() { 2455 sem_wait(&_semaphore); 2456} 2457 2458bool Semaphore::trywait() { 2459 return sem_trywait(&_semaphore) == 0; 2460} 2461 2462bool Semaphore::timedwait(unsigned int sec, int nsec) { 2463 2464 struct timespec ts; 2465 // Semaphore's are always associated with CLOCK_REALTIME 2466 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2467 // see unpackTime for discussion on overflow checking 2468 if (sec >= MAX_SECS) { 2469 ts.tv_sec += MAX_SECS; 2470 ts.tv_nsec = 0; 2471 } else { 2472 ts.tv_sec += sec; 2473 ts.tv_nsec += nsec; 2474 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2475 ts.tv_nsec -= NANOSECS_PER_SEC; 2476 ++ts.tv_sec; // note: this must be <= max_secs 2477 } 2478 } 2479 2480 while (1) { 2481 int result = sem_timedwait(&_semaphore, &ts); 2482 if (result == 0) { 2483 return true; 2484 } else if (errno == EINTR) { 2485 continue; 2486 } else if (errno == ETIMEDOUT) { 2487 return false; 2488 } else { 2489 return false; 2490 } 2491 } 2492} 2493 2494extern "C" { 2495 typedef void (*sa_handler_t)(int); 2496 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2497} 2498 2499void* os::signal(int signal_number, void* handler) { 2500 struct sigaction sigAct, oldSigAct; 2501 2502 sigfillset(&(sigAct.sa_mask)); 2503 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2504 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2505 2506 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2507 // -1 means registration failed 2508 return (void *)-1; 2509 } 2510 2511 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2512} 2513 2514void os::signal_raise(int signal_number) { 2515 ::raise(signal_number); 2516} 2517 2518// The following code is moved from os.cpp for making this 2519// code platform specific, which it is by its very nature. 2520 2521// Will be modified when max signal is changed to be dynamic 2522int os::sigexitnum_pd() { 2523 return NSIG; 2524} 2525 2526// a counter for each possible signal value 2527static volatile jint pending_signals[NSIG+1] = { 0 }; 2528 2529// Linux(POSIX) specific hand shaking semaphore. 2530static sem_t sig_sem; 2531static Semaphore sr_semaphore; 2532 2533void os::signal_init_pd() { 2534 // Initialize signal structures 2535 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2536 2537 // Initialize signal semaphore 2538 ::sem_init(&sig_sem, 0, 0); 2539} 2540 2541void os::signal_notify(int sig) { 2542 Atomic::inc(&pending_signals[sig]); 2543 ::sem_post(&sig_sem); 2544} 2545 2546static int check_pending_signals(bool wait) { 2547 Atomic::store(0, &sigint_count); 2548 for (;;) { 2549 for (int i = 0; i < NSIG + 1; i++) { 2550 jint n = pending_signals[i]; 2551 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2552 return i; 2553 } 2554 } 2555 if (!wait) { 2556 return -1; 2557 } 2558 JavaThread *thread = JavaThread::current(); 2559 ThreadBlockInVM tbivm(thread); 2560 2561 bool threadIsSuspended; 2562 do { 2563 thread->set_suspend_equivalent(); 2564 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2565 ::sem_wait(&sig_sem); 2566 2567 // were we externally suspended while we were waiting? 2568 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2569 if (threadIsSuspended) { 2570 // The semaphore has been incremented, but while we were waiting 2571 // another thread suspended us. We don't want to continue running 2572 // while suspended because that would surprise the thread that 2573 // suspended us. 2574 ::sem_post(&sig_sem); 2575 2576 thread->java_suspend_self(); 2577 } 2578 } while (threadIsSuspended); 2579 } 2580} 2581 2582int os::signal_lookup() { 2583 return check_pending_signals(false); 2584} 2585 2586int os::signal_wait() { 2587 return check_pending_signals(true); 2588} 2589 2590//////////////////////////////////////////////////////////////////////////////// 2591// Virtual Memory 2592 2593int os::vm_page_size() { 2594 // Seems redundant as all get out 2595 assert(os::Linux::page_size() != -1, "must call os::init"); 2596 return os::Linux::page_size(); 2597} 2598 2599// Solaris allocates memory by pages. 2600int os::vm_allocation_granularity() { 2601 assert(os::Linux::page_size() != -1, "must call os::init"); 2602 return os::Linux::page_size(); 2603} 2604 2605// Rationale behind this function: 2606// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2607// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2608// samples for JITted code. Here we create private executable mapping over the code cache 2609// and then we can use standard (well, almost, as mapping can change) way to provide 2610// info for the reporting script by storing timestamp and location of symbol 2611void linux_wrap_code(char* base, size_t size) { 2612 static volatile jint cnt = 0; 2613 2614 if (!UseOprofile) { 2615 return; 2616 } 2617 2618 char buf[PATH_MAX+1]; 2619 int num = Atomic::add(1, &cnt); 2620 2621 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2622 os::get_temp_directory(), os::current_process_id(), num); 2623 unlink(buf); 2624 2625 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2626 2627 if (fd != -1) { 2628 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2629 if (rv != (off_t)-1) { 2630 if (::write(fd, "", 1) == 1) { 2631 mmap(base, size, 2632 PROT_READ|PROT_WRITE|PROT_EXEC, 2633 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2634 } 2635 } 2636 ::close(fd); 2637 unlink(buf); 2638 } 2639} 2640 2641static bool recoverable_mmap_error(int err) { 2642 // See if the error is one we can let the caller handle. This 2643 // list of errno values comes from JBS-6843484. I can't find a 2644 // Linux man page that documents this specific set of errno 2645 // values so while this list currently matches Solaris, it may 2646 // change as we gain experience with this failure mode. 2647 switch (err) { 2648 case EBADF: 2649 case EINVAL: 2650 case ENOTSUP: 2651 // let the caller deal with these errors 2652 return true; 2653 2654 default: 2655 // Any remaining errors on this OS can cause our reserved mapping 2656 // to be lost. That can cause confusion where different data 2657 // structures think they have the same memory mapped. The worst 2658 // scenario is if both the VM and a library think they have the 2659 // same memory mapped. 2660 return false; 2661 } 2662} 2663 2664static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2665 int err) { 2666 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2667 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2668 strerror(err), err); 2669} 2670 2671static void warn_fail_commit_memory(char* addr, size_t size, 2672 size_t alignment_hint, bool exec, 2673 int err) { 2674 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2675 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2676 alignment_hint, exec, strerror(err), err); 2677} 2678 2679// NOTE: Linux kernel does not really reserve the pages for us. 2680// All it does is to check if there are enough free pages 2681// left at the time of mmap(). This could be a potential 2682// problem. 2683int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2684 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2685 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2686 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2687 if (res != (uintptr_t) MAP_FAILED) { 2688 if (UseNUMAInterleaving) { 2689 numa_make_global(addr, size); 2690 } 2691 return 0; 2692 } 2693 2694 int err = errno; // save errno from mmap() call above 2695 2696 if (!recoverable_mmap_error(err)) { 2697 warn_fail_commit_memory(addr, size, exec, err); 2698 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2699 } 2700 2701 return err; 2702} 2703 2704bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2705 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2706} 2707 2708void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2709 const char* mesg) { 2710 assert(mesg != NULL, "mesg must be specified"); 2711 int err = os::Linux::commit_memory_impl(addr, size, exec); 2712 if (err != 0) { 2713 // the caller wants all commit errors to exit with the specified mesg: 2714 warn_fail_commit_memory(addr, size, exec, err); 2715 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2716 } 2717} 2718 2719// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2720#ifndef MAP_HUGETLB 2721 #define MAP_HUGETLB 0x40000 2722#endif 2723 2724// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2725#ifndef MADV_HUGEPAGE 2726 #define MADV_HUGEPAGE 14 2727#endif 2728 2729int os::Linux::commit_memory_impl(char* addr, size_t size, 2730 size_t alignment_hint, bool exec) { 2731 int err = os::Linux::commit_memory_impl(addr, size, exec); 2732 if (err == 0) { 2733 realign_memory(addr, size, alignment_hint); 2734 } 2735 return err; 2736} 2737 2738bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2739 bool exec) { 2740 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2741} 2742 2743void os::pd_commit_memory_or_exit(char* addr, size_t size, 2744 size_t alignment_hint, bool exec, 2745 const char* mesg) { 2746 assert(mesg != NULL, "mesg must be specified"); 2747 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2748 if (err != 0) { 2749 // the caller wants all commit errors to exit with the specified mesg: 2750 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2751 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2752 } 2753} 2754 2755void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2756 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2757 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2758 // be supported or the memory may already be backed by huge pages. 2759 ::madvise(addr, bytes, MADV_HUGEPAGE); 2760 } 2761} 2762 2763void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2764 // This method works by doing an mmap over an existing mmaping and effectively discarding 2765 // the existing pages. However it won't work for SHM-based large pages that cannot be 2766 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2767 // small pages on top of the SHM segment. This method always works for small pages, so we 2768 // allow that in any case. 2769 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2770 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2771 } 2772} 2773 2774void os::numa_make_global(char *addr, size_t bytes) { 2775 Linux::numa_interleave_memory(addr, bytes); 2776} 2777 2778// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2779// bind policy to MPOL_PREFERRED for the current thread. 2780#define USE_MPOL_PREFERRED 0 2781 2782void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2783 // To make NUMA and large pages more robust when both enabled, we need to ease 2784 // the requirements on where the memory should be allocated. MPOL_BIND is the 2785 // default policy and it will force memory to be allocated on the specified 2786 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2787 // the specified node, but will not force it. Using this policy will prevent 2788 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2789 // free large pages. 2790 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2791 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2792} 2793 2794bool os::numa_topology_changed() { return false; } 2795 2796size_t os::numa_get_groups_num() { 2797 int max_node = Linux::numa_max_node(); 2798 return max_node > 0 ? max_node + 1 : 1; 2799} 2800 2801int os::numa_get_group_id() { 2802 int cpu_id = Linux::sched_getcpu(); 2803 if (cpu_id != -1) { 2804 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2805 if (lgrp_id != -1) { 2806 return lgrp_id; 2807 } 2808 } 2809 return 0; 2810} 2811 2812size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2813 for (size_t i = 0; i < size; i++) { 2814 ids[i] = i; 2815 } 2816 return size; 2817} 2818 2819bool os::get_page_info(char *start, page_info* info) { 2820 return false; 2821} 2822 2823char *os::scan_pages(char *start, char* end, page_info* page_expected, 2824 page_info* page_found) { 2825 return end; 2826} 2827 2828 2829int os::Linux::sched_getcpu_syscall(void) { 2830 unsigned int cpu; 2831 int retval = -1; 2832 2833#if defined(IA32) 2834 #ifndef SYS_getcpu 2835 #define SYS_getcpu 318 2836 #endif 2837 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2838#elif defined(AMD64) 2839// Unfortunately we have to bring all these macros here from vsyscall.h 2840// to be able to compile on old linuxes. 2841 #define __NR_vgetcpu 2 2842 #define VSYSCALL_START (-10UL << 20) 2843 #define VSYSCALL_SIZE 1024 2844 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2845 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2846 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2847 retval = vgetcpu(&cpu, NULL, NULL); 2848#endif 2849 2850 return (retval == -1) ? retval : cpu; 2851} 2852 2853// Something to do with the numa-aware allocator needs these symbols 2854extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2855extern "C" JNIEXPORT void numa_error(char *where) { } 2856extern "C" JNIEXPORT int fork1() { return fork(); } 2857 2858 2859// If we are running with libnuma version > 2, then we should 2860// be trying to use symbols with versions 1.1 2861// If we are running with earlier version, which did not have symbol versions, 2862// we should use the base version. 2863void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2864 void *f = dlvsym(handle, name, "libnuma_1.1"); 2865 if (f == NULL) { 2866 f = dlsym(handle, name); 2867 } 2868 return f; 2869} 2870 2871bool os::Linux::libnuma_init() { 2872 // sched_getcpu() should be in libc. 2873 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2874 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2875 2876 // If it's not, try a direct syscall. 2877 if (sched_getcpu() == -1) { 2878 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2879 (void*)&sched_getcpu_syscall)); 2880 } 2881 2882 if (sched_getcpu() != -1) { // Does it work? 2883 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2884 if (handle != NULL) { 2885 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2886 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2887 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2888 libnuma_dlsym(handle, "numa_max_node"))); 2889 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2890 libnuma_dlsym(handle, "numa_available"))); 2891 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2892 libnuma_dlsym(handle, "numa_tonode_memory"))); 2893 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2894 libnuma_dlsym(handle, "numa_interleave_memory"))); 2895 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2896 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2897 2898 2899 if (numa_available() != -1) { 2900 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2901 // Create a cpu -> node mapping 2902 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2903 rebuild_cpu_to_node_map(); 2904 return true; 2905 } 2906 } 2907 } 2908 return false; 2909} 2910 2911// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2912// The table is later used in get_node_by_cpu(). 2913void os::Linux::rebuild_cpu_to_node_map() { 2914 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2915 // in libnuma (possible values are starting from 16, 2916 // and continuing up with every other power of 2, but less 2917 // than the maximum number of CPUs supported by kernel), and 2918 // is a subject to change (in libnuma version 2 the requirements 2919 // are more reasonable) we'll just hardcode the number they use 2920 // in the library. 2921 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2922 2923 size_t cpu_num = os::active_processor_count(); 2924 size_t cpu_map_size = NCPUS / BitsPerCLong; 2925 size_t cpu_map_valid_size = 2926 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2927 2928 cpu_to_node()->clear(); 2929 cpu_to_node()->at_grow(cpu_num - 1); 2930 size_t node_num = numa_get_groups_num(); 2931 2932 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2933 for (size_t i = 0; i < node_num; i++) { 2934 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2935 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2936 if (cpu_map[j] != 0) { 2937 for (size_t k = 0; k < BitsPerCLong; k++) { 2938 if (cpu_map[j] & (1UL << k)) { 2939 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2940 } 2941 } 2942 } 2943 } 2944 } 2945 } 2946 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2947} 2948 2949int os::Linux::get_node_by_cpu(int cpu_id) { 2950 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2951 return cpu_to_node()->at(cpu_id); 2952 } 2953 return -1; 2954} 2955 2956GrowableArray<int>* os::Linux::_cpu_to_node; 2957os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2958os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2959os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2960os::Linux::numa_available_func_t os::Linux::_numa_available; 2961os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2962os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2963os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2964unsigned long* os::Linux::_numa_all_nodes; 2965 2966bool os::pd_uncommit_memory(char* addr, size_t size) { 2967 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2968 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2969 return res != (uintptr_t) MAP_FAILED; 2970} 2971 2972static address get_stack_commited_bottom(address bottom, size_t size) { 2973 address nbot = bottom; 2974 address ntop = bottom + size; 2975 2976 size_t page_sz = os::vm_page_size(); 2977 unsigned pages = size / page_sz; 2978 2979 unsigned char vec[1]; 2980 unsigned imin = 1, imax = pages + 1, imid; 2981 int mincore_return_value = 0; 2982 2983 assert(imin <= imax, "Unexpected page size"); 2984 2985 while (imin < imax) { 2986 imid = (imax + imin) / 2; 2987 nbot = ntop - (imid * page_sz); 2988 2989 // Use a trick with mincore to check whether the page is mapped or not. 2990 // mincore sets vec to 1 if page resides in memory and to 0 if page 2991 // is swapped output but if page we are asking for is unmapped 2992 // it returns -1,ENOMEM 2993 mincore_return_value = mincore(nbot, page_sz, vec); 2994 2995 if (mincore_return_value == -1) { 2996 // Page is not mapped go up 2997 // to find first mapped page 2998 if (errno != EAGAIN) { 2999 assert(errno == ENOMEM, "Unexpected mincore errno"); 3000 imax = imid; 3001 } 3002 } else { 3003 // Page is mapped go down 3004 // to find first not mapped page 3005 imin = imid + 1; 3006 } 3007 } 3008 3009 nbot = nbot + page_sz; 3010 3011 // Adjust stack bottom one page up if last checked page is not mapped 3012 if (mincore_return_value == -1) { 3013 nbot = nbot + page_sz; 3014 } 3015 3016 return nbot; 3017} 3018 3019 3020// Linux uses a growable mapping for the stack, and if the mapping for 3021// the stack guard pages is not removed when we detach a thread the 3022// stack cannot grow beyond the pages where the stack guard was 3023// mapped. If at some point later in the process the stack expands to 3024// that point, the Linux kernel cannot expand the stack any further 3025// because the guard pages are in the way, and a segfault occurs. 3026// 3027// However, it's essential not to split the stack region by unmapping 3028// a region (leaving a hole) that's already part of the stack mapping, 3029// so if the stack mapping has already grown beyond the guard pages at 3030// the time we create them, we have to truncate the stack mapping. 3031// So, we need to know the extent of the stack mapping when 3032// create_stack_guard_pages() is called. 3033 3034// We only need this for stacks that are growable: at the time of 3035// writing thread stacks don't use growable mappings (i.e. those 3036// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3037// only applies to the main thread. 3038 3039// If the (growable) stack mapping already extends beyond the point 3040// where we're going to put our guard pages, truncate the mapping at 3041// that point by munmap()ping it. This ensures that when we later 3042// munmap() the guard pages we don't leave a hole in the stack 3043// mapping. This only affects the main/initial thread 3044 3045bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3046 if (os::Linux::is_initial_thread()) { 3047 // As we manually grow stack up to bottom inside create_attached_thread(), 3048 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3049 // we don't need to do anything special. 3050 // Check it first, before calling heavy function. 3051 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3052 unsigned char vec[1]; 3053 3054 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3055 // Fallback to slow path on all errors, including EAGAIN 3056 stack_extent = (uintptr_t) get_stack_commited_bottom( 3057 os::Linux::initial_thread_stack_bottom(), 3058 (size_t)addr - stack_extent); 3059 } 3060 3061 if (stack_extent < (uintptr_t)addr) { 3062 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3063 } 3064 } 3065 3066 return os::commit_memory(addr, size, !ExecMem); 3067} 3068 3069// If this is a growable mapping, remove the guard pages entirely by 3070// munmap()ping them. If not, just call uncommit_memory(). This only 3071// affects the main/initial thread, but guard against future OS changes 3072// It's safe to always unmap guard pages for initial thread because we 3073// always place it right after end of the mapped region 3074 3075bool os::remove_stack_guard_pages(char* addr, size_t size) { 3076 uintptr_t stack_extent, stack_base; 3077 3078 if (os::Linux::is_initial_thread()) { 3079 return ::munmap(addr, size) == 0; 3080 } 3081 3082 return os::uncommit_memory(addr, size); 3083} 3084 3085static address _highest_vm_reserved_address = NULL; 3086 3087// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3088// at 'requested_addr'. If there are existing memory mappings at the same 3089// location, however, they will be overwritten. If 'fixed' is false, 3090// 'requested_addr' is only treated as a hint, the return value may or 3091// may not start from the requested address. Unlike Linux mmap(), this 3092// function returns NULL to indicate failure. 3093static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3094 char * addr; 3095 int flags; 3096 3097 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3098 if (fixed) { 3099 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3100 flags |= MAP_FIXED; 3101 } 3102 3103 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3104 // touch an uncommitted page. Otherwise, the read/write might 3105 // succeed if we have enough swap space to back the physical page. 3106 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3107 flags, -1, 0); 3108 3109 if (addr != MAP_FAILED) { 3110 // anon_mmap() should only get called during VM initialization, 3111 // don't need lock (actually we can skip locking even it can be called 3112 // from multiple threads, because _highest_vm_reserved_address is just a 3113 // hint about the upper limit of non-stack memory regions.) 3114 if ((address)addr + bytes > _highest_vm_reserved_address) { 3115 _highest_vm_reserved_address = (address)addr + bytes; 3116 } 3117 } 3118 3119 return addr == MAP_FAILED ? NULL : addr; 3120} 3121 3122// Don't update _highest_vm_reserved_address, because there might be memory 3123// regions above addr + size. If so, releasing a memory region only creates 3124// a hole in the address space, it doesn't help prevent heap-stack collision. 3125// 3126static int anon_munmap(char * addr, size_t size) { 3127 return ::munmap(addr, size) == 0; 3128} 3129 3130char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3131 size_t alignment_hint) { 3132 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3133} 3134 3135bool os::pd_release_memory(char* addr, size_t size) { 3136 return anon_munmap(addr, size); 3137} 3138 3139static address highest_vm_reserved_address() { 3140 return _highest_vm_reserved_address; 3141} 3142 3143static bool linux_mprotect(char* addr, size_t size, int prot) { 3144 // Linux wants the mprotect address argument to be page aligned. 3145 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3146 3147 // According to SUSv3, mprotect() should only be used with mappings 3148 // established by mmap(), and mmap() always maps whole pages. Unaligned 3149 // 'addr' likely indicates problem in the VM (e.g. trying to change 3150 // protection of malloc'ed or statically allocated memory). Check the 3151 // caller if you hit this assert. 3152 assert(addr == bottom, "sanity check"); 3153 3154 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3155 return ::mprotect(bottom, size, prot) == 0; 3156} 3157 3158// Set protections specified 3159bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3160 bool is_committed) { 3161 unsigned int p = 0; 3162 switch (prot) { 3163 case MEM_PROT_NONE: p = PROT_NONE; break; 3164 case MEM_PROT_READ: p = PROT_READ; break; 3165 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3166 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3167 default: 3168 ShouldNotReachHere(); 3169 } 3170 // is_committed is unused. 3171 return linux_mprotect(addr, bytes, p); 3172} 3173 3174bool os::guard_memory(char* addr, size_t size) { 3175 return linux_mprotect(addr, size, PROT_NONE); 3176} 3177 3178bool os::unguard_memory(char* addr, size_t size) { 3179 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3180} 3181 3182bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3183 size_t page_size) { 3184 bool result = false; 3185 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3186 MAP_ANONYMOUS|MAP_PRIVATE, 3187 -1, 0); 3188 if (p != MAP_FAILED) { 3189 void *aligned_p = align_ptr_up(p, page_size); 3190 3191 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3192 3193 munmap(p, page_size * 2); 3194 } 3195 3196 if (warn && !result) { 3197 warning("TransparentHugePages is not supported by the operating system."); 3198 } 3199 3200 return result; 3201} 3202 3203bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3204 bool result = false; 3205 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3206 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3207 -1, 0); 3208 3209 if (p != MAP_FAILED) { 3210 // We don't know if this really is a huge page or not. 3211 FILE *fp = fopen("/proc/self/maps", "r"); 3212 if (fp) { 3213 while (!feof(fp)) { 3214 char chars[257]; 3215 long x = 0; 3216 if (fgets(chars, sizeof(chars), fp)) { 3217 if (sscanf(chars, "%lx-%*x", &x) == 1 3218 && x == (long)p) { 3219 if (strstr (chars, "hugepage")) { 3220 result = true; 3221 break; 3222 } 3223 } 3224 } 3225 } 3226 fclose(fp); 3227 } 3228 munmap(p, page_size); 3229 } 3230 3231 if (warn && !result) { 3232 warning("HugeTLBFS is not supported by the operating system."); 3233 } 3234 3235 return result; 3236} 3237 3238// Set the coredump_filter bits to include largepages in core dump (bit 6) 3239// 3240// From the coredump_filter documentation: 3241// 3242// - (bit 0) anonymous private memory 3243// - (bit 1) anonymous shared memory 3244// - (bit 2) file-backed private memory 3245// - (bit 3) file-backed shared memory 3246// - (bit 4) ELF header pages in file-backed private memory areas (it is 3247// effective only if the bit 2 is cleared) 3248// - (bit 5) hugetlb private memory 3249// - (bit 6) hugetlb shared memory 3250// 3251static void set_coredump_filter(void) { 3252 FILE *f; 3253 long cdm; 3254 3255 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3256 return; 3257 } 3258 3259 if (fscanf(f, "%lx", &cdm) != 1) { 3260 fclose(f); 3261 return; 3262 } 3263 3264 rewind(f); 3265 3266 if ((cdm & LARGEPAGES_BIT) == 0) { 3267 cdm |= LARGEPAGES_BIT; 3268 fprintf(f, "%#lx", cdm); 3269 } 3270 3271 fclose(f); 3272} 3273 3274// Large page support 3275 3276static size_t _large_page_size = 0; 3277 3278size_t os::Linux::find_large_page_size() { 3279 size_t large_page_size = 0; 3280 3281 // large_page_size on Linux is used to round up heap size. x86 uses either 3282 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3283 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3284 // page as large as 256M. 3285 // 3286 // Here we try to figure out page size by parsing /proc/meminfo and looking 3287 // for a line with the following format: 3288 // Hugepagesize: 2048 kB 3289 // 3290 // If we can't determine the value (e.g. /proc is not mounted, or the text 3291 // format has been changed), we'll use the largest page size supported by 3292 // the processor. 3293 3294#ifndef ZERO 3295 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3296 ARM_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M); 3297#endif // ZERO 3298 3299 FILE *fp = fopen("/proc/meminfo", "r"); 3300 if (fp) { 3301 while (!feof(fp)) { 3302 int x = 0; 3303 char buf[16]; 3304 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3305 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3306 large_page_size = x * K; 3307 break; 3308 } 3309 } else { 3310 // skip to next line 3311 for (;;) { 3312 int ch = fgetc(fp); 3313 if (ch == EOF || ch == (int)'\n') break; 3314 } 3315 } 3316 } 3317 fclose(fp); 3318 } 3319 3320 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3321 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3322 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3323 proper_unit_for_byte_size(large_page_size)); 3324 } 3325 3326 return large_page_size; 3327} 3328 3329size_t os::Linux::setup_large_page_size() { 3330 _large_page_size = Linux::find_large_page_size(); 3331 const size_t default_page_size = (size_t)Linux::page_size(); 3332 if (_large_page_size > default_page_size) { 3333 _page_sizes[0] = _large_page_size; 3334 _page_sizes[1] = default_page_size; 3335 _page_sizes[2] = 0; 3336 } 3337 3338 return _large_page_size; 3339} 3340 3341bool os::Linux::setup_large_page_type(size_t page_size) { 3342 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3343 FLAG_IS_DEFAULT(UseSHM) && 3344 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3345 3346 // The type of large pages has not been specified by the user. 3347 3348 // Try UseHugeTLBFS and then UseSHM. 3349 UseHugeTLBFS = UseSHM = true; 3350 3351 // Don't try UseTransparentHugePages since there are known 3352 // performance issues with it turned on. This might change in the future. 3353 UseTransparentHugePages = false; 3354 } 3355 3356 if (UseTransparentHugePages) { 3357 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3358 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3359 UseHugeTLBFS = false; 3360 UseSHM = false; 3361 return true; 3362 } 3363 UseTransparentHugePages = false; 3364 } 3365 3366 if (UseHugeTLBFS) { 3367 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3368 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3369 UseSHM = false; 3370 return true; 3371 } 3372 UseHugeTLBFS = false; 3373 } 3374 3375 return UseSHM; 3376} 3377 3378void os::large_page_init() { 3379 if (!UseLargePages && 3380 !UseTransparentHugePages && 3381 !UseHugeTLBFS && 3382 !UseSHM) { 3383 // Not using large pages. 3384 return; 3385 } 3386 3387 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3388 // The user explicitly turned off large pages. 3389 // Ignore the rest of the large pages flags. 3390 UseTransparentHugePages = false; 3391 UseHugeTLBFS = false; 3392 UseSHM = false; 3393 return; 3394 } 3395 3396 size_t large_page_size = Linux::setup_large_page_size(); 3397 UseLargePages = Linux::setup_large_page_type(large_page_size); 3398 3399 set_coredump_filter(); 3400} 3401 3402#ifndef SHM_HUGETLB 3403 #define SHM_HUGETLB 04000 3404#endif 3405 3406char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3407 char* req_addr, bool exec) { 3408 // "exec" is passed in but not used. Creating the shared image for 3409 // the code cache doesn't have an SHM_X executable permission to check. 3410 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3411 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3412 3413 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3414 return NULL; // Fallback to small pages. 3415 } 3416 3417 key_t key = IPC_PRIVATE; 3418 char *addr; 3419 3420 bool warn_on_failure = UseLargePages && 3421 (!FLAG_IS_DEFAULT(UseLargePages) || 3422 !FLAG_IS_DEFAULT(UseSHM) || 3423 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3424 char msg[128]; 3425 3426 // Create a large shared memory region to attach to based on size. 3427 // Currently, size is the total size of the heap 3428 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3429 if (shmid == -1) { 3430 // Possible reasons for shmget failure: 3431 // 1. shmmax is too small for Java heap. 3432 // > check shmmax value: cat /proc/sys/kernel/shmmax 3433 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3434 // 2. not enough large page memory. 3435 // > check available large pages: cat /proc/meminfo 3436 // > increase amount of large pages: 3437 // echo new_value > /proc/sys/vm/nr_hugepages 3438 // Note 1: different Linux may use different name for this property, 3439 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3440 // Note 2: it's possible there's enough physical memory available but 3441 // they are so fragmented after a long run that they can't 3442 // coalesce into large pages. Try to reserve large pages when 3443 // the system is still "fresh". 3444 if (warn_on_failure) { 3445 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3446 warning("%s", msg); 3447 } 3448 return NULL; 3449 } 3450 3451 // attach to the region 3452 addr = (char*)shmat(shmid, req_addr, 0); 3453 int err = errno; 3454 3455 // Remove shmid. If shmat() is successful, the actual shared memory segment 3456 // will be deleted when it's detached by shmdt() or when the process 3457 // terminates. If shmat() is not successful this will remove the shared 3458 // segment immediately. 3459 shmctl(shmid, IPC_RMID, NULL); 3460 3461 if ((intptr_t)addr == -1) { 3462 if (warn_on_failure) { 3463 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3464 warning("%s", msg); 3465 } 3466 return NULL; 3467 } 3468 3469 return addr; 3470} 3471 3472static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3473 int error) { 3474 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3475 3476 bool warn_on_failure = UseLargePages && 3477 (!FLAG_IS_DEFAULT(UseLargePages) || 3478 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3479 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3480 3481 if (warn_on_failure) { 3482 char msg[128]; 3483 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3484 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3485 warning("%s", msg); 3486 } 3487} 3488 3489char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3490 char* req_addr, 3491 bool exec) { 3492 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3493 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3494 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3495 3496 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3497 char* addr = (char*)::mmap(req_addr, bytes, prot, 3498 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3499 -1, 0); 3500 3501 if (addr == MAP_FAILED) { 3502 warn_on_large_pages_failure(req_addr, bytes, errno); 3503 return NULL; 3504 } 3505 3506 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3507 3508 return addr; 3509} 3510 3511char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3512 size_t alignment, 3513 char* req_addr, 3514 bool exec) { 3515 size_t large_page_size = os::large_page_size(); 3516 3517 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3518 3519 // Allocate small pages. 3520 3521 char* start; 3522 if (req_addr != NULL) { 3523 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3524 assert(is_size_aligned(bytes, alignment), "Must be"); 3525 start = os::reserve_memory(bytes, req_addr); 3526 assert(start == NULL || start == req_addr, "Must be"); 3527 } else { 3528 start = os::reserve_memory_aligned(bytes, alignment); 3529 } 3530 3531 if (start == NULL) { 3532 return NULL; 3533 } 3534 3535 assert(is_ptr_aligned(start, alignment), "Must be"); 3536 3537 if (MemTracker::tracking_level() > NMT_minimal) { 3538 // os::reserve_memory_special will record this memory area. 3539 // Need to release it here to prevent overlapping reservations. 3540 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3541 tkr.record((address)start, bytes); 3542 } 3543 3544 char* end = start + bytes; 3545 3546 // Find the regions of the allocated chunk that can be promoted to large pages. 3547 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3548 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3549 3550 size_t lp_bytes = lp_end - lp_start; 3551 3552 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3553 3554 if (lp_bytes == 0) { 3555 // The mapped region doesn't even span the start and the end of a large page. 3556 // Fall back to allocate a non-special area. 3557 ::munmap(start, end - start); 3558 return NULL; 3559 } 3560 3561 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3562 3563 3564 void* result; 3565 3566 if (start != lp_start) { 3567 result = ::mmap(start, lp_start - start, prot, 3568 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3569 -1, 0); 3570 if (result == MAP_FAILED) { 3571 ::munmap(lp_start, end - lp_start); 3572 return NULL; 3573 } 3574 } 3575 3576 result = ::mmap(lp_start, lp_bytes, prot, 3577 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3578 -1, 0); 3579 if (result == MAP_FAILED) { 3580 warn_on_large_pages_failure(req_addr, bytes, errno); 3581 // If the mmap above fails, the large pages region will be unmapped and we 3582 // have regions before and after with small pages. Release these regions. 3583 // 3584 // | mapped | unmapped | mapped | 3585 // ^ ^ ^ ^ 3586 // start lp_start lp_end end 3587 // 3588 ::munmap(start, lp_start - start); 3589 ::munmap(lp_end, end - lp_end); 3590 return NULL; 3591 } 3592 3593 if (lp_end != end) { 3594 result = ::mmap(lp_end, end - lp_end, prot, 3595 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3596 -1, 0); 3597 if (result == MAP_FAILED) { 3598 ::munmap(start, lp_end - start); 3599 return NULL; 3600 } 3601 } 3602 3603 return start; 3604} 3605 3606char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3607 size_t alignment, 3608 char* req_addr, 3609 bool exec) { 3610 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3611 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3612 assert(is_power_of_2(alignment), "Must be"); 3613 assert(is_power_of_2(os::large_page_size()), "Must be"); 3614 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3615 3616 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3617 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3618 } else { 3619 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3620 } 3621} 3622 3623char* os::reserve_memory_special(size_t bytes, size_t alignment, 3624 char* req_addr, bool exec) { 3625 assert(UseLargePages, "only for large pages"); 3626 3627 char* addr; 3628 if (UseSHM) { 3629 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3630 } else { 3631 assert(UseHugeTLBFS, "must be"); 3632 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3633 } 3634 3635 if (addr != NULL) { 3636 if (UseNUMAInterleaving) { 3637 numa_make_global(addr, bytes); 3638 } 3639 3640 // The memory is committed 3641 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3642 } 3643 3644 return addr; 3645} 3646 3647bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3648 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3649 return shmdt(base) == 0; 3650} 3651 3652bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3653 return pd_release_memory(base, bytes); 3654} 3655 3656bool os::release_memory_special(char* base, size_t bytes) { 3657 bool res; 3658 if (MemTracker::tracking_level() > NMT_minimal) { 3659 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3660 res = os::Linux::release_memory_special_impl(base, bytes); 3661 if (res) { 3662 tkr.record((address)base, bytes); 3663 } 3664 3665 } else { 3666 res = os::Linux::release_memory_special_impl(base, bytes); 3667 } 3668 return res; 3669} 3670 3671bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3672 assert(UseLargePages, "only for large pages"); 3673 bool res; 3674 3675 if (UseSHM) { 3676 res = os::Linux::release_memory_special_shm(base, bytes); 3677 } else { 3678 assert(UseHugeTLBFS, "must be"); 3679 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3680 } 3681 return res; 3682} 3683 3684size_t os::large_page_size() { 3685 return _large_page_size; 3686} 3687 3688// With SysV SHM the entire memory region must be allocated as shared 3689// memory. 3690// HugeTLBFS allows application to commit large page memory on demand. 3691// However, when committing memory with HugeTLBFS fails, the region 3692// that was supposed to be committed will lose the old reservation 3693// and allow other threads to steal that memory region. Because of this 3694// behavior we can't commit HugeTLBFS memory. 3695bool os::can_commit_large_page_memory() { 3696 return UseTransparentHugePages; 3697} 3698 3699bool os::can_execute_large_page_memory() { 3700 return UseTransparentHugePages || UseHugeTLBFS; 3701} 3702 3703// Reserve memory at an arbitrary address, only if that area is 3704// available (and not reserved for something else). 3705 3706char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3707 const int max_tries = 10; 3708 char* base[max_tries]; 3709 size_t size[max_tries]; 3710 const size_t gap = 0x000000; 3711 3712 // Assert only that the size is a multiple of the page size, since 3713 // that's all that mmap requires, and since that's all we really know 3714 // about at this low abstraction level. If we need higher alignment, 3715 // we can either pass an alignment to this method or verify alignment 3716 // in one of the methods further up the call chain. See bug 5044738. 3717 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3718 3719 // Repeatedly allocate blocks until the block is allocated at the 3720 // right spot. Give up after max_tries. Note that reserve_memory() will 3721 // automatically update _highest_vm_reserved_address if the call is 3722 // successful. The variable tracks the highest memory address every reserved 3723 // by JVM. It is used to detect heap-stack collision if running with 3724 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3725 // space than needed, it could confuse the collision detecting code. To 3726 // solve the problem, save current _highest_vm_reserved_address and 3727 // calculate the correct value before return. 3728 address old_highest = _highest_vm_reserved_address; 3729 3730 // Linux mmap allows caller to pass an address as hint; give it a try first, 3731 // if kernel honors the hint then we can return immediately. 3732 char * addr = anon_mmap(requested_addr, bytes, false); 3733 if (addr == requested_addr) { 3734 return requested_addr; 3735 } 3736 3737 if (addr != NULL) { 3738 // mmap() is successful but it fails to reserve at the requested address 3739 anon_munmap(addr, bytes); 3740 } 3741 3742 int i; 3743 for (i = 0; i < max_tries; ++i) { 3744 base[i] = reserve_memory(bytes); 3745 3746 if (base[i] != NULL) { 3747 // Is this the block we wanted? 3748 if (base[i] == requested_addr) { 3749 size[i] = bytes; 3750 break; 3751 } 3752 3753 // Does this overlap the block we wanted? Give back the overlapped 3754 // parts and try again. 3755 3756 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3757 if (top_overlap >= 0 && top_overlap < bytes) { 3758 unmap_memory(base[i], top_overlap); 3759 base[i] += top_overlap; 3760 size[i] = bytes - top_overlap; 3761 } else { 3762 size_t bottom_overlap = base[i] + bytes - requested_addr; 3763 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3764 unmap_memory(requested_addr, bottom_overlap); 3765 size[i] = bytes - bottom_overlap; 3766 } else { 3767 size[i] = bytes; 3768 } 3769 } 3770 } 3771 } 3772 3773 // Give back the unused reserved pieces. 3774 3775 for (int j = 0; j < i; ++j) { 3776 if (base[j] != NULL) { 3777 unmap_memory(base[j], size[j]); 3778 } 3779 } 3780 3781 if (i < max_tries) { 3782 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3783 return requested_addr; 3784 } else { 3785 _highest_vm_reserved_address = old_highest; 3786 return NULL; 3787 } 3788} 3789 3790size_t os::read(int fd, void *buf, unsigned int nBytes) { 3791 return ::read(fd, buf, nBytes); 3792} 3793 3794// Short sleep, direct OS call. 3795// 3796// Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3797// sched_yield(2) will actually give up the CPU: 3798// 3799// * Alone on this pariticular CPU, keeps running. 3800// * Before the introduction of "skip_buddy" with "compat_yield" disabled 3801// (pre 2.6.39). 3802// 3803// So calling this with 0 is an alternative. 3804// 3805void os::naked_short_sleep(jlong ms) { 3806 struct timespec req; 3807 3808 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3809 req.tv_sec = 0; 3810 if (ms > 0) { 3811 req.tv_nsec = (ms % 1000) * 1000000; 3812 } else { 3813 req.tv_nsec = 1; 3814 } 3815 3816 nanosleep(&req, NULL); 3817 3818 return; 3819} 3820 3821// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3822void os::infinite_sleep() { 3823 while (true) { // sleep forever ... 3824 ::sleep(100); // ... 100 seconds at a time 3825 } 3826} 3827 3828// Used to convert frequent JVM_Yield() to nops 3829bool os::dont_yield() { 3830 return DontYieldALot; 3831} 3832 3833void os::naked_yield() { 3834 sched_yield(); 3835} 3836 3837//////////////////////////////////////////////////////////////////////////////// 3838// thread priority support 3839 3840// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3841// only supports dynamic priority, static priority must be zero. For real-time 3842// applications, Linux supports SCHED_RR which allows static priority (1-99). 3843// However, for large multi-threaded applications, SCHED_RR is not only slower 3844// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3845// of 5 runs - Sep 2005). 3846// 3847// The following code actually changes the niceness of kernel-thread/LWP. It 3848// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3849// not the entire user process, and user level threads are 1:1 mapped to kernel 3850// threads. It has always been the case, but could change in the future. For 3851// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3852// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3853 3854int os::java_to_os_priority[CriticalPriority + 1] = { 3855 19, // 0 Entry should never be used 3856 3857 4, // 1 MinPriority 3858 3, // 2 3859 2, // 3 3860 3861 1, // 4 3862 0, // 5 NormPriority 3863 -1, // 6 3864 3865 -2, // 7 3866 -3, // 8 3867 -4, // 9 NearMaxPriority 3868 3869 -5, // 10 MaxPriority 3870 3871 -5 // 11 CriticalPriority 3872}; 3873 3874static int prio_init() { 3875 if (ThreadPriorityPolicy == 1) { 3876 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3877 // if effective uid is not root. Perhaps, a more elegant way of doing 3878 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3879 if (geteuid() != 0) { 3880 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3881 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3882 } 3883 ThreadPriorityPolicy = 0; 3884 } 3885 } 3886 if (UseCriticalJavaThreadPriority) { 3887 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3888 } 3889 return 0; 3890} 3891 3892OSReturn os::set_native_priority(Thread* thread, int newpri) { 3893 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 3894 3895 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3896 return (ret == 0) ? OS_OK : OS_ERR; 3897} 3898 3899OSReturn os::get_native_priority(const Thread* const thread, 3900 int *priority_ptr) { 3901 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 3902 *priority_ptr = java_to_os_priority[NormPriority]; 3903 return OS_OK; 3904 } 3905 3906 errno = 0; 3907 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3908 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3909} 3910 3911// Hint to the underlying OS that a task switch would not be good. 3912// Void return because it's a hint and can fail. 3913void os::hint_no_preempt() {} 3914 3915//////////////////////////////////////////////////////////////////////////////// 3916// suspend/resume support 3917 3918// the low-level signal-based suspend/resume support is a remnant from the 3919// old VM-suspension that used to be for java-suspension, safepoints etc, 3920// within hotspot. Now there is a single use-case for this: 3921// - calling get_thread_pc() on the VMThread by the flat-profiler task 3922// that runs in the watcher thread. 3923// The remaining code is greatly simplified from the more general suspension 3924// code that used to be used. 3925// 3926// The protocol is quite simple: 3927// - suspend: 3928// - sends a signal to the target thread 3929// - polls the suspend state of the osthread using a yield loop 3930// - target thread signal handler (SR_handler) sets suspend state 3931// and blocks in sigsuspend until continued 3932// - resume: 3933// - sets target osthread state to continue 3934// - sends signal to end the sigsuspend loop in the SR_handler 3935// 3936// Note that the SR_lock plays no role in this suspend/resume protocol. 3937 3938static void resume_clear_context(OSThread *osthread) { 3939 osthread->set_ucontext(NULL); 3940 osthread->set_siginfo(NULL); 3941} 3942 3943static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 3944 ucontext_t* context) { 3945 osthread->set_ucontext(context); 3946 osthread->set_siginfo(siginfo); 3947} 3948 3949// Handler function invoked when a thread's execution is suspended or 3950// resumed. We have to be careful that only async-safe functions are 3951// called here (Note: most pthread functions are not async safe and 3952// should be avoided.) 3953// 3954// Note: sigwait() is a more natural fit than sigsuspend() from an 3955// interface point of view, but sigwait() prevents the signal hander 3956// from being run. libpthread would get very confused by not having 3957// its signal handlers run and prevents sigwait()'s use with the 3958// mutex granting granting signal. 3959// 3960// Currently only ever called on the VMThread and JavaThreads (PC sampling) 3961// 3962static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3963 // Save and restore errno to avoid confusing native code with EINTR 3964 // after sigsuspend. 3965 int old_errno = errno; 3966 3967 Thread* thread = Thread::current(); 3968 OSThread* osthread = thread->osthread(); 3969 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3970 3971 os::SuspendResume::State current = osthread->sr.state(); 3972 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3973 suspend_save_context(osthread, siginfo, context); 3974 3975 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 3976 os::SuspendResume::State state = osthread->sr.suspended(); 3977 if (state == os::SuspendResume::SR_SUSPENDED) { 3978 sigset_t suspend_set; // signals for sigsuspend() 3979 3980 // get current set of blocked signals and unblock resume signal 3981 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3982 sigdelset(&suspend_set, SR_signum); 3983 3984 sr_semaphore.signal(); 3985 // wait here until we are resumed 3986 while (1) { 3987 sigsuspend(&suspend_set); 3988 3989 os::SuspendResume::State result = osthread->sr.running(); 3990 if (result == os::SuspendResume::SR_RUNNING) { 3991 sr_semaphore.signal(); 3992 break; 3993 } 3994 } 3995 3996 } else if (state == os::SuspendResume::SR_RUNNING) { 3997 // request was cancelled, continue 3998 } else { 3999 ShouldNotReachHere(); 4000 } 4001 4002 resume_clear_context(osthread); 4003 } else if (current == os::SuspendResume::SR_RUNNING) { 4004 // request was cancelled, continue 4005 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4006 // ignore 4007 } else { 4008 // ignore 4009 } 4010 4011 errno = old_errno; 4012} 4013 4014 4015static int SR_initialize() { 4016 struct sigaction act; 4017 char *s; 4018 // Get signal number to use for suspend/resume 4019 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4020 int sig = ::strtol(s, 0, 10); 4021 if (sig > 0 || sig < _NSIG) { 4022 SR_signum = sig; 4023 } 4024 } 4025 4026 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4027 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4028 4029 sigemptyset(&SR_sigset); 4030 sigaddset(&SR_sigset, SR_signum); 4031 4032 // Set up signal handler for suspend/resume 4033 act.sa_flags = SA_RESTART|SA_SIGINFO; 4034 act.sa_handler = (void (*)(int)) SR_handler; 4035 4036 // SR_signum is blocked by default. 4037 // 4528190 - We also need to block pthread restart signal (32 on all 4038 // supported Linux platforms). Note that LinuxThreads need to block 4039 // this signal for all threads to work properly. So we don't have 4040 // to use hard-coded signal number when setting up the mask. 4041 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4042 4043 if (sigaction(SR_signum, &act, 0) == -1) { 4044 return -1; 4045 } 4046 4047 // Save signal flag 4048 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4049 return 0; 4050} 4051 4052static int sr_notify(OSThread* osthread) { 4053 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4054 assert_status(status == 0, status, "pthread_kill"); 4055 return status; 4056} 4057 4058// "Randomly" selected value for how long we want to spin 4059// before bailing out on suspending a thread, also how often 4060// we send a signal to a thread we want to resume 4061static const int RANDOMLY_LARGE_INTEGER = 1000000; 4062static const int RANDOMLY_LARGE_INTEGER2 = 100; 4063 4064// returns true on success and false on error - really an error is fatal 4065// but this seems the normal response to library errors 4066static bool do_suspend(OSThread* osthread) { 4067 assert(osthread->sr.is_running(), "thread should be running"); 4068 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4069 4070 // mark as suspended and send signal 4071 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4072 // failed to switch, state wasn't running? 4073 ShouldNotReachHere(); 4074 return false; 4075 } 4076 4077 if (sr_notify(osthread) != 0) { 4078 ShouldNotReachHere(); 4079 } 4080 4081 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4082 while (true) { 4083 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4084 break; 4085 } else { 4086 // timeout 4087 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4088 if (cancelled == os::SuspendResume::SR_RUNNING) { 4089 return false; 4090 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4091 // make sure that we consume the signal on the semaphore as well 4092 sr_semaphore.wait(); 4093 break; 4094 } else { 4095 ShouldNotReachHere(); 4096 return false; 4097 } 4098 } 4099 } 4100 4101 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4102 return true; 4103} 4104 4105static void do_resume(OSThread* osthread) { 4106 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4107 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4108 4109 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4110 // failed to switch to WAKEUP_REQUEST 4111 ShouldNotReachHere(); 4112 return; 4113 } 4114 4115 while (true) { 4116 if (sr_notify(osthread) == 0) { 4117 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4118 if (osthread->sr.is_running()) { 4119 return; 4120 } 4121 } 4122 } else { 4123 ShouldNotReachHere(); 4124 } 4125 } 4126 4127 guarantee(osthread->sr.is_running(), "Must be running!"); 4128} 4129 4130/////////////////////////////////////////////////////////////////////////////////// 4131// signal handling (except suspend/resume) 4132 4133// This routine may be used by user applications as a "hook" to catch signals. 4134// The user-defined signal handler must pass unrecognized signals to this 4135// routine, and if it returns true (non-zero), then the signal handler must 4136// return immediately. If the flag "abort_if_unrecognized" is true, then this 4137// routine will never retun false (zero), but instead will execute a VM panic 4138// routine kill the process. 4139// 4140// If this routine returns false, it is OK to call it again. This allows 4141// the user-defined signal handler to perform checks either before or after 4142// the VM performs its own checks. Naturally, the user code would be making 4143// a serious error if it tried to handle an exception (such as a null check 4144// or breakpoint) that the VM was generating for its own correct operation. 4145// 4146// This routine may recognize any of the following kinds of signals: 4147// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4148// It should be consulted by handlers for any of those signals. 4149// 4150// The caller of this routine must pass in the three arguments supplied 4151// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4152// field of the structure passed to sigaction(). This routine assumes that 4153// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4154// 4155// Note that the VM will print warnings if it detects conflicting signal 4156// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4157// 4158extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4159 siginfo_t* siginfo, 4160 void* ucontext, 4161 int abort_if_unrecognized); 4162 4163void signalHandler(int sig, siginfo_t* info, void* uc) { 4164 assert(info != NULL && uc != NULL, "it must be old kernel"); 4165 int orig_errno = errno; // Preserve errno value over signal handler. 4166 JVM_handle_linux_signal(sig, info, uc, true); 4167 errno = orig_errno; 4168} 4169 4170 4171// This boolean allows users to forward their own non-matching signals 4172// to JVM_handle_linux_signal, harmlessly. 4173bool os::Linux::signal_handlers_are_installed = false; 4174 4175// For signal-chaining 4176struct sigaction os::Linux::sigact[MAXSIGNUM]; 4177unsigned int os::Linux::sigs = 0; 4178bool os::Linux::libjsig_is_loaded = false; 4179typedef struct sigaction *(*get_signal_t)(int); 4180get_signal_t os::Linux::get_signal_action = NULL; 4181 4182struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4183 struct sigaction *actp = NULL; 4184 4185 if (libjsig_is_loaded) { 4186 // Retrieve the old signal handler from libjsig 4187 actp = (*get_signal_action)(sig); 4188 } 4189 if (actp == NULL) { 4190 // Retrieve the preinstalled signal handler from jvm 4191 actp = get_preinstalled_handler(sig); 4192 } 4193 4194 return actp; 4195} 4196 4197static bool call_chained_handler(struct sigaction *actp, int sig, 4198 siginfo_t *siginfo, void *context) { 4199 // Call the old signal handler 4200 if (actp->sa_handler == SIG_DFL) { 4201 // It's more reasonable to let jvm treat it as an unexpected exception 4202 // instead of taking the default action. 4203 return false; 4204 } else if (actp->sa_handler != SIG_IGN) { 4205 if ((actp->sa_flags & SA_NODEFER) == 0) { 4206 // automaticlly block the signal 4207 sigaddset(&(actp->sa_mask), sig); 4208 } 4209 4210 sa_handler_t hand; 4211 sa_sigaction_t sa; 4212 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4213 // retrieve the chained handler 4214 if (siginfo_flag_set) { 4215 sa = actp->sa_sigaction; 4216 } else { 4217 hand = actp->sa_handler; 4218 } 4219 4220 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4221 actp->sa_handler = SIG_DFL; 4222 } 4223 4224 // try to honor the signal mask 4225 sigset_t oset; 4226 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4227 4228 // call into the chained handler 4229 if (siginfo_flag_set) { 4230 (*sa)(sig, siginfo, context); 4231 } else { 4232 (*hand)(sig); 4233 } 4234 4235 // restore the signal mask 4236 pthread_sigmask(SIG_SETMASK, &oset, 0); 4237 } 4238 // Tell jvm's signal handler the signal is taken care of. 4239 return true; 4240} 4241 4242bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4243 bool chained = false; 4244 // signal-chaining 4245 if (UseSignalChaining) { 4246 struct sigaction *actp = get_chained_signal_action(sig); 4247 if (actp != NULL) { 4248 chained = call_chained_handler(actp, sig, siginfo, context); 4249 } 4250 } 4251 return chained; 4252} 4253 4254struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4255 if ((((unsigned int)1 << sig) & sigs) != 0) { 4256 return &sigact[sig]; 4257 } 4258 return NULL; 4259} 4260 4261void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4262 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4263 sigact[sig] = oldAct; 4264 sigs |= (unsigned int)1 << sig; 4265} 4266 4267// for diagnostic 4268int os::Linux::sigflags[MAXSIGNUM]; 4269 4270int os::Linux::get_our_sigflags(int sig) { 4271 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4272 return sigflags[sig]; 4273} 4274 4275void os::Linux::set_our_sigflags(int sig, int flags) { 4276 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4277 sigflags[sig] = flags; 4278} 4279 4280void os::Linux::set_signal_handler(int sig, bool set_installed) { 4281 // Check for overwrite. 4282 struct sigaction oldAct; 4283 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4284 4285 void* oldhand = oldAct.sa_sigaction 4286 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4287 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4288 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4289 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4290 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4291 if (AllowUserSignalHandlers || !set_installed) { 4292 // Do not overwrite; user takes responsibility to forward to us. 4293 return; 4294 } else if (UseSignalChaining) { 4295 // save the old handler in jvm 4296 save_preinstalled_handler(sig, oldAct); 4297 // libjsig also interposes the sigaction() call below and saves the 4298 // old sigaction on it own. 4299 } else { 4300 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4301 "%#lx for signal %d.", (long)oldhand, sig)); 4302 } 4303 } 4304 4305 struct sigaction sigAct; 4306 sigfillset(&(sigAct.sa_mask)); 4307 sigAct.sa_handler = SIG_DFL; 4308 if (!set_installed) { 4309 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4310 } else { 4311 sigAct.sa_sigaction = signalHandler; 4312 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4313 } 4314 // Save flags, which are set by ours 4315 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4316 sigflags[sig] = sigAct.sa_flags; 4317 4318 int ret = sigaction(sig, &sigAct, &oldAct); 4319 assert(ret == 0, "check"); 4320 4321 void* oldhand2 = oldAct.sa_sigaction 4322 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4323 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4324 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4325} 4326 4327// install signal handlers for signals that HotSpot needs to 4328// handle in order to support Java-level exception handling. 4329 4330void os::Linux::install_signal_handlers() { 4331 if (!signal_handlers_are_installed) { 4332 signal_handlers_are_installed = true; 4333 4334 // signal-chaining 4335 typedef void (*signal_setting_t)(); 4336 signal_setting_t begin_signal_setting = NULL; 4337 signal_setting_t end_signal_setting = NULL; 4338 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4339 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4340 if (begin_signal_setting != NULL) { 4341 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4342 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4343 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4344 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4345 libjsig_is_loaded = true; 4346 assert(UseSignalChaining, "should enable signal-chaining"); 4347 } 4348 if (libjsig_is_loaded) { 4349 // Tell libjsig jvm is setting signal handlers 4350 (*begin_signal_setting)(); 4351 } 4352 4353 set_signal_handler(SIGSEGV, true); 4354 set_signal_handler(SIGPIPE, true); 4355 set_signal_handler(SIGBUS, true); 4356 set_signal_handler(SIGILL, true); 4357 set_signal_handler(SIGFPE, true); 4358#if defined(PPC64) 4359 set_signal_handler(SIGTRAP, true); 4360#endif 4361 set_signal_handler(SIGXFSZ, true); 4362 4363 if (libjsig_is_loaded) { 4364 // Tell libjsig jvm finishes setting signal handlers 4365 (*end_signal_setting)(); 4366 } 4367 4368 // We don't activate signal checker if libjsig is in place, we trust ourselves 4369 // and if UserSignalHandler is installed all bets are off. 4370 // Log that signal checking is off only if -verbose:jni is specified. 4371 if (CheckJNICalls) { 4372 if (libjsig_is_loaded) { 4373 if (PrintJNIResolving) { 4374 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4375 } 4376 check_signals = false; 4377 } 4378 if (AllowUserSignalHandlers) { 4379 if (PrintJNIResolving) { 4380 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4381 } 4382 check_signals = false; 4383 } 4384 } 4385 } 4386} 4387 4388// This is the fastest way to get thread cpu time on Linux. 4389// Returns cpu time (user+sys) for any thread, not only for current. 4390// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4391// It might work on 2.6.10+ with a special kernel/glibc patch. 4392// For reference, please, see IEEE Std 1003.1-2004: 4393// http://www.unix.org/single_unix_specification 4394 4395jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4396 struct timespec tp; 4397 int rc = os::Linux::clock_gettime(clockid, &tp); 4398 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4399 4400 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4401} 4402 4403///// 4404// glibc on Linux platform uses non-documented flag 4405// to indicate, that some special sort of signal 4406// trampoline is used. 4407// We will never set this flag, and we should 4408// ignore this flag in our diagnostic 4409#ifdef SIGNIFICANT_SIGNAL_MASK 4410 #undef SIGNIFICANT_SIGNAL_MASK 4411#endif 4412#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4413 4414static const char* get_signal_handler_name(address handler, 4415 char* buf, int buflen) { 4416 int offset; 4417 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4418 if (found) { 4419 // skip directory names 4420 const char *p1, *p2; 4421 p1 = buf; 4422 size_t len = strlen(os::file_separator()); 4423 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4424 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4425 } else { 4426 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4427 } 4428 return buf; 4429} 4430 4431static void print_signal_handler(outputStream* st, int sig, 4432 char* buf, size_t buflen) { 4433 struct sigaction sa; 4434 4435 sigaction(sig, NULL, &sa); 4436 4437 // See comment for SIGNIFICANT_SIGNAL_MASK define 4438 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4439 4440 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4441 4442 address handler = (sa.sa_flags & SA_SIGINFO) 4443 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4444 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4445 4446 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4447 st->print("SIG_DFL"); 4448 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4449 st->print("SIG_IGN"); 4450 } else { 4451 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4452 } 4453 4454 st->print(", sa_mask[0]="); 4455 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4456 4457 address rh = VMError::get_resetted_sighandler(sig); 4458 // May be, handler was resetted by VMError? 4459 if (rh != NULL) { 4460 handler = rh; 4461 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4462 } 4463 4464 st->print(", sa_flags="); 4465 os::Posix::print_sa_flags(st, sa.sa_flags); 4466 4467 // Check: is it our handler? 4468 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4469 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4470 // It is our signal handler 4471 // check for flags, reset system-used one! 4472 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4473 st->print( 4474 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4475 os::Linux::get_our_sigflags(sig)); 4476 } 4477 } 4478 st->cr(); 4479} 4480 4481 4482#define DO_SIGNAL_CHECK(sig) \ 4483 do { \ 4484 if (!sigismember(&check_signal_done, sig)) { \ 4485 os::Linux::check_signal_handler(sig); \ 4486 } \ 4487 } while (0) 4488 4489// This method is a periodic task to check for misbehaving JNI applications 4490// under CheckJNI, we can add any periodic checks here 4491 4492void os::run_periodic_checks() { 4493 if (check_signals == false) return; 4494 4495 // SEGV and BUS if overridden could potentially prevent 4496 // generation of hs*.log in the event of a crash, debugging 4497 // such a case can be very challenging, so we absolutely 4498 // check the following for a good measure: 4499 DO_SIGNAL_CHECK(SIGSEGV); 4500 DO_SIGNAL_CHECK(SIGILL); 4501 DO_SIGNAL_CHECK(SIGFPE); 4502 DO_SIGNAL_CHECK(SIGBUS); 4503 DO_SIGNAL_CHECK(SIGPIPE); 4504 DO_SIGNAL_CHECK(SIGXFSZ); 4505#if defined(PPC64) 4506 DO_SIGNAL_CHECK(SIGTRAP); 4507#endif 4508 4509 // ReduceSignalUsage allows the user to override these handlers 4510 // see comments at the very top and jvm_solaris.h 4511 if (!ReduceSignalUsage) { 4512 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4513 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4514 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4515 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4516 } 4517 4518 DO_SIGNAL_CHECK(SR_signum); 4519 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4520} 4521 4522typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4523 4524static os_sigaction_t os_sigaction = NULL; 4525 4526void os::Linux::check_signal_handler(int sig) { 4527 char buf[O_BUFLEN]; 4528 address jvmHandler = NULL; 4529 4530 4531 struct sigaction act; 4532 if (os_sigaction == NULL) { 4533 // only trust the default sigaction, in case it has been interposed 4534 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4535 if (os_sigaction == NULL) return; 4536 } 4537 4538 os_sigaction(sig, (struct sigaction*)NULL, &act); 4539 4540 4541 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4542 4543 address thisHandler = (act.sa_flags & SA_SIGINFO) 4544 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4545 : CAST_FROM_FN_PTR(address, act.sa_handler); 4546 4547 4548 switch (sig) { 4549 case SIGSEGV: 4550 case SIGBUS: 4551 case SIGFPE: 4552 case SIGPIPE: 4553 case SIGILL: 4554 case SIGXFSZ: 4555 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4556 break; 4557 4558 case SHUTDOWN1_SIGNAL: 4559 case SHUTDOWN2_SIGNAL: 4560 case SHUTDOWN3_SIGNAL: 4561 case BREAK_SIGNAL: 4562 jvmHandler = (address)user_handler(); 4563 break; 4564 4565 case INTERRUPT_SIGNAL: 4566 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4567 break; 4568 4569 default: 4570 if (sig == SR_signum) { 4571 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4572 } else { 4573 return; 4574 } 4575 break; 4576 } 4577 4578 if (thisHandler != jvmHandler) { 4579 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4580 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4581 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4582 // No need to check this sig any longer 4583 sigaddset(&check_signal_done, sig); 4584 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4585 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4586 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4587 exception_name(sig, buf, O_BUFLEN)); 4588 } 4589 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4590 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4591 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4592 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4593 // No need to check this sig any longer 4594 sigaddset(&check_signal_done, sig); 4595 } 4596 4597 // Dump all the signal 4598 if (sigismember(&check_signal_done, sig)) { 4599 print_signal_handlers(tty, buf, O_BUFLEN); 4600 } 4601} 4602 4603extern void report_error(char* file_name, int line_no, char* title, 4604 char* format, ...); 4605 4606extern bool signal_name(int signo, char* buf, size_t len); 4607 4608const char* os::exception_name(int exception_code, char* buf, size_t size) { 4609 if (0 < exception_code && exception_code <= SIGRTMAX) { 4610 // signal 4611 if (!signal_name(exception_code, buf, size)) { 4612 jio_snprintf(buf, size, "SIG%d", exception_code); 4613 } 4614 return buf; 4615 } else { 4616 return NULL; 4617 } 4618} 4619 4620// this is called _before_ the most of global arguments have been parsed 4621void os::init(void) { 4622 char dummy; // used to get a guess on initial stack address 4623// first_hrtime = gethrtime(); 4624 4625 // With LinuxThreads the JavaMain thread pid (primordial thread) 4626 // is different than the pid of the java launcher thread. 4627 // So, on Linux, the launcher thread pid is passed to the VM 4628 // via the sun.java.launcher.pid property. 4629 // Use this property instead of getpid() if it was correctly passed. 4630 // See bug 6351349. 4631 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4632 4633 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4634 4635 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4636 4637 init_random(1234567); 4638 4639 ThreadCritical::initialize(); 4640 4641 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4642 if (Linux::page_size() == -1) { 4643 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4644 strerror(errno))); 4645 } 4646 init_page_sizes((size_t) Linux::page_size()); 4647 4648 Linux::initialize_system_info(); 4649 4650 // main_thread points to the aboriginal thread 4651 Linux::_main_thread = pthread_self(); 4652 4653 Linux::clock_init(); 4654 initial_time_count = javaTimeNanos(); 4655 4656 // pthread_condattr initialization for monotonic clock 4657 int status; 4658 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4659 if ((status = pthread_condattr_init(_condattr)) != 0) { 4660 fatal(err_msg("pthread_condattr_init: %s", strerror(status))); 4661 } 4662 // Only set the clock if CLOCK_MONOTONIC is available 4663 if (os::supports_monotonic_clock()) { 4664 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4665 if (status == EINVAL) { 4666 warning("Unable to use monotonic clock with relative timed-waits" \ 4667 " - changes to the time-of-day clock may have adverse affects"); 4668 } else { 4669 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status))); 4670 } 4671 } 4672 } 4673 // else it defaults to CLOCK_REALTIME 4674 4675 pthread_mutex_init(&dl_mutex, NULL); 4676 4677 // If the pagesize of the VM is greater than 8K determine the appropriate 4678 // number of initial guard pages. The user can change this with the 4679 // command line arguments, if needed. 4680 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4681 StackYellowPages = 1; 4682 StackRedPages = 1; 4683 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4684 } 4685 4686 // retrieve entry point for pthread_setname_np 4687 Linux::_pthread_setname_np = 4688 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4689 4690} 4691 4692// To install functions for atexit system call 4693extern "C" { 4694 static void perfMemory_exit_helper() { 4695 perfMemory_exit(); 4696 } 4697} 4698 4699// this is called _after_ the global arguments have been parsed 4700jint os::init_2(void) { 4701 Linux::fast_thread_clock_init(); 4702 4703 // Allocate a single page and mark it as readable for safepoint polling 4704 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4705 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page"); 4706 4707 os::set_polling_page(polling_page); 4708 4709#ifndef PRODUCT 4710 if (Verbose && PrintMiscellaneous) { 4711 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", 4712 (intptr_t)polling_page); 4713 } 4714#endif 4715 4716 if (!UseMembar) { 4717 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4718 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4719 os::set_memory_serialize_page(mem_serialize_page); 4720 4721#ifndef PRODUCT 4722 if (Verbose && PrintMiscellaneous) { 4723 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", 4724 (intptr_t)mem_serialize_page); 4725 } 4726#endif 4727 } 4728 4729 // initialize suspend/resume support - must do this before signal_sets_init() 4730 if (SR_initialize() != 0) { 4731 perror("SR_initialize failed"); 4732 return JNI_ERR; 4733 } 4734 4735 Linux::signal_sets_init(); 4736 Linux::install_signal_handlers(); 4737 4738 // Check minimum allowable stack size for thread creation and to initialize 4739 // the java system classes, including StackOverflowError - depends on page 4740 // size. Add a page for compiler2 recursion in main thread. 4741 // Add in 2*BytesPerWord times page size to account for VM stack during 4742 // class initialization depending on 32 or 64 bit VM. 4743 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4744 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4745 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4746 4747 size_t threadStackSizeInBytes = ThreadStackSize * K; 4748 if (threadStackSizeInBytes != 0 && 4749 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4750 tty->print_cr("\nThe stack size specified is too small, " 4751 "Specify at least %dk", 4752 os::Linux::min_stack_allowed/ K); 4753 return JNI_ERR; 4754 } 4755 4756 // Make the stack size a multiple of the page size so that 4757 // the yellow/red zones can be guarded. 4758 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4759 vm_page_size())); 4760 4761 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4762 4763#if defined(IA32) 4764 workaround_expand_exec_shield_cs_limit(); 4765#endif 4766 4767 Linux::libpthread_init(); 4768 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4769 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4770 Linux::glibc_version(), Linux::libpthread_version(), 4771 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4772 } 4773 4774 if (UseNUMA) { 4775 if (!Linux::libnuma_init()) { 4776 UseNUMA = false; 4777 } else { 4778 if ((Linux::numa_max_node() < 1)) { 4779 // There's only one node(they start from 0), disable NUMA. 4780 UseNUMA = false; 4781 } 4782 } 4783 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4784 // we can make the adaptive lgrp chunk resizing work. If the user specified 4785 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4786 // disable adaptive resizing. 4787 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4788 if (FLAG_IS_DEFAULT(UseNUMA)) { 4789 UseNUMA = false; 4790 } else { 4791 if (FLAG_IS_DEFAULT(UseLargePages) && 4792 FLAG_IS_DEFAULT(UseSHM) && 4793 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4794 UseLargePages = false; 4795 } else { 4796 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing"); 4797 UseAdaptiveSizePolicy = false; 4798 UseAdaptiveNUMAChunkSizing = false; 4799 } 4800 } 4801 } 4802 if (!UseNUMA && ForceNUMA) { 4803 UseNUMA = true; 4804 } 4805 } 4806 4807 if (MaxFDLimit) { 4808 // set the number of file descriptors to max. print out error 4809 // if getrlimit/setrlimit fails but continue regardless. 4810 struct rlimit nbr_files; 4811 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4812 if (status != 0) { 4813 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4814 perror("os::init_2 getrlimit failed"); 4815 } 4816 } else { 4817 nbr_files.rlim_cur = nbr_files.rlim_max; 4818 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4819 if (status != 0) { 4820 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4821 perror("os::init_2 setrlimit failed"); 4822 } 4823 } 4824 } 4825 } 4826 4827 // Initialize lock used to serialize thread creation (see os::create_thread) 4828 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4829 4830 // at-exit methods are called in the reverse order of their registration. 4831 // atexit functions are called on return from main or as a result of a 4832 // call to exit(3C). There can be only 32 of these functions registered 4833 // and atexit() does not set errno. 4834 4835 if (PerfAllowAtExitRegistration) { 4836 // only register atexit functions if PerfAllowAtExitRegistration is set. 4837 // atexit functions can be delayed until process exit time, which 4838 // can be problematic for embedded VM situations. Embedded VMs should 4839 // call DestroyJavaVM() to assure that VM resources are released. 4840 4841 // note: perfMemory_exit_helper atexit function may be removed in 4842 // the future if the appropriate cleanup code can be added to the 4843 // VM_Exit VMOperation's doit method. 4844 if (atexit(perfMemory_exit_helper) != 0) { 4845 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4846 } 4847 } 4848 4849 // initialize thread priority policy 4850 prio_init(); 4851 4852 return JNI_OK; 4853} 4854 4855// Mark the polling page as unreadable 4856void os::make_polling_page_unreadable(void) { 4857 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 4858 fatal("Could not disable polling page"); 4859 } 4860} 4861 4862// Mark the polling page as readable 4863void os::make_polling_page_readable(void) { 4864 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4865 fatal("Could not enable polling page"); 4866 } 4867} 4868 4869int os::active_processor_count() { 4870 // Linux doesn't yet have a (official) notion of processor sets, 4871 // so just return the number of online processors. 4872 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4873 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4874 return online_cpus; 4875} 4876 4877void os::set_native_thread_name(const char *name) { 4878 if (Linux::_pthread_setname_np) { 4879 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 4880 snprintf(buf, sizeof(buf), "%s", name); 4881 buf[sizeof(buf) - 1] = '\0'; 4882 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 4883 // ERANGE should not happen; all other errors should just be ignored. 4884 assert(rc != ERANGE, "pthread_setname_np failed"); 4885 } 4886} 4887 4888bool os::distribute_processes(uint length, uint* distribution) { 4889 // Not yet implemented. 4890 return false; 4891} 4892 4893bool os::bind_to_processor(uint processor_id) { 4894 // Not yet implemented. 4895 return false; 4896} 4897 4898/// 4899 4900void os::SuspendedThreadTask::internal_do_task() { 4901 if (do_suspend(_thread->osthread())) { 4902 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4903 do_task(context); 4904 do_resume(_thread->osthread()); 4905 } 4906} 4907 4908class PcFetcher : public os::SuspendedThreadTask { 4909 public: 4910 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4911 ExtendedPC result(); 4912 protected: 4913 void do_task(const os::SuspendedThreadTaskContext& context); 4914 private: 4915 ExtendedPC _epc; 4916}; 4917 4918ExtendedPC PcFetcher::result() { 4919 guarantee(is_done(), "task is not done yet."); 4920 return _epc; 4921} 4922 4923void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4924 Thread* thread = context.thread(); 4925 OSThread* osthread = thread->osthread(); 4926 if (osthread->ucontext() != NULL) { 4927 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4928 } else { 4929 // NULL context is unexpected, double-check this is the VMThread 4930 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4931 } 4932} 4933 4934// Suspends the target using the signal mechanism and then grabs the PC before 4935// resuming the target. Used by the flat-profiler only 4936ExtendedPC os::get_thread_pc(Thread* thread) { 4937 // Make sure that it is called by the watcher for the VMThread 4938 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4939 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4940 4941 PcFetcher fetcher(thread); 4942 fetcher.run(); 4943 return fetcher.result(); 4944} 4945 4946int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, 4947 pthread_mutex_t *_mutex, 4948 const struct timespec *_abstime) { 4949 if (is_NPTL()) { 4950 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4951 } else { 4952 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4953 // word back to default 64bit precision if condvar is signaled. Java 4954 // wants 53bit precision. Save and restore current value. 4955 int fpu = get_fpu_control_word(); 4956 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4957 set_fpu_control_word(fpu); 4958 return status; 4959 } 4960} 4961 4962//////////////////////////////////////////////////////////////////////////////// 4963// debug support 4964 4965bool os::find(address addr, outputStream* st) { 4966 Dl_info dlinfo; 4967 memset(&dlinfo, 0, sizeof(dlinfo)); 4968 if (dladdr(addr, &dlinfo) != 0) { 4969 st->print(PTR_FORMAT ": ", addr); 4970 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 4971 st->print("%s+%#x", dlinfo.dli_sname, 4972 addr - (intptr_t)dlinfo.dli_saddr); 4973 } else if (dlinfo.dli_fbase != NULL) { 4974 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4975 } else { 4976 st->print("<absolute address>"); 4977 } 4978 if (dlinfo.dli_fname != NULL) { 4979 st->print(" in %s", dlinfo.dli_fname); 4980 } 4981 if (dlinfo.dli_fbase != NULL) { 4982 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4983 } 4984 st->cr(); 4985 4986 if (Verbose) { 4987 // decode some bytes around the PC 4988 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 4989 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 4990 address lowest = (address) dlinfo.dli_sname; 4991 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4992 if (begin < lowest) begin = lowest; 4993 Dl_info dlinfo2; 4994 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 4995 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 4996 end = (address) dlinfo2.dli_saddr; 4997 } 4998 Disassembler::decode(begin, end, st); 4999 } 5000 return true; 5001 } 5002 return false; 5003} 5004 5005//////////////////////////////////////////////////////////////////////////////// 5006// misc 5007 5008// This does not do anything on Linux. This is basically a hook for being 5009// able to use structured exception handling (thread-local exception filters) 5010// on, e.g., Win32. 5011void 5012os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 5013 JavaCallArguments* args, Thread* thread) { 5014 f(value, method, args, thread); 5015} 5016 5017void os::print_statistics() { 5018} 5019 5020int os::message_box(const char* title, const char* message) { 5021 int i; 5022 fdStream err(defaultStream::error_fd()); 5023 for (i = 0; i < 78; i++) err.print_raw("="); 5024 err.cr(); 5025 err.print_raw_cr(title); 5026 for (i = 0; i < 78; i++) err.print_raw("-"); 5027 err.cr(); 5028 err.print_raw_cr(message); 5029 for (i = 0; i < 78; i++) err.print_raw("="); 5030 err.cr(); 5031 5032 char buf[16]; 5033 // Prevent process from exiting upon "read error" without consuming all CPU 5034 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5035 5036 return buf[0] == 'y' || buf[0] == 'Y'; 5037} 5038 5039int os::stat(const char *path, struct stat *sbuf) { 5040 char pathbuf[MAX_PATH]; 5041 if (strlen(path) > MAX_PATH - 1) { 5042 errno = ENAMETOOLONG; 5043 return -1; 5044 } 5045 os::native_path(strcpy(pathbuf, path)); 5046 return ::stat(pathbuf, sbuf); 5047} 5048 5049bool os::check_heap(bool force) { 5050 return true; 5051} 5052 5053// Is a (classpath) directory empty? 5054bool os::dir_is_empty(const char* path) { 5055 DIR *dir = NULL; 5056 struct dirent *ptr; 5057 5058 dir = opendir(path); 5059 if (dir == NULL) return true; 5060 5061 // Scan the directory 5062 bool result = true; 5063 char buf[sizeof(struct dirent) + MAX_PATH]; 5064 while (result && (ptr = ::readdir(dir)) != NULL) { 5065 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5066 result = false; 5067 } 5068 } 5069 closedir(dir); 5070 return result; 5071} 5072 5073// This code originates from JDK's sysOpen and open64_w 5074// from src/solaris/hpi/src/system_md.c 5075 5076int os::open(const char *path, int oflag, int mode) { 5077 if (strlen(path) > MAX_PATH - 1) { 5078 errno = ENAMETOOLONG; 5079 return -1; 5080 } 5081 5082 // All file descriptors that are opened in the Java process and not 5083 // specifically destined for a subprocess should have the close-on-exec 5084 // flag set. If we don't set it, then careless 3rd party native code 5085 // might fork and exec without closing all appropriate file descriptors 5086 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5087 // turn might: 5088 // 5089 // - cause end-of-file to fail to be detected on some file 5090 // descriptors, resulting in mysterious hangs, or 5091 // 5092 // - might cause an fopen in the subprocess to fail on a system 5093 // suffering from bug 1085341. 5094 // 5095 // (Yes, the default setting of the close-on-exec flag is a Unix 5096 // design flaw) 5097 // 5098 // See: 5099 // 1085341: 32-bit stdio routines should support file descriptors >255 5100 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5101 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5102 // 5103 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5104 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5105 // because it saves a system call and removes a small window where the flag 5106 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5107 // and we fall back to using FD_CLOEXEC (see below). 5108#ifdef O_CLOEXEC 5109 oflag |= O_CLOEXEC; 5110#endif 5111 5112 int fd = ::open64(path, oflag, mode); 5113 if (fd == -1) return -1; 5114 5115 //If the open succeeded, the file might still be a directory 5116 { 5117 struct stat64 buf64; 5118 int ret = ::fstat64(fd, &buf64); 5119 int st_mode = buf64.st_mode; 5120 5121 if (ret != -1) { 5122 if ((st_mode & S_IFMT) == S_IFDIR) { 5123 errno = EISDIR; 5124 ::close(fd); 5125 return -1; 5126 } 5127 } else { 5128 ::close(fd); 5129 return -1; 5130 } 5131 } 5132 5133#ifdef FD_CLOEXEC 5134 // Validate that the use of the O_CLOEXEC flag on open above worked. 5135 // With recent kernels, we will perform this check exactly once. 5136 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5137 if (!O_CLOEXEC_is_known_to_work) { 5138 int flags = ::fcntl(fd, F_GETFD); 5139 if (flags != -1) { 5140 if ((flags & FD_CLOEXEC) != 0) 5141 O_CLOEXEC_is_known_to_work = 1; 5142 else 5143 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5144 } 5145 } 5146#endif 5147 5148 return fd; 5149} 5150 5151 5152// create binary file, rewriting existing file if required 5153int os::create_binary_file(const char* path, bool rewrite_existing) { 5154 int oflags = O_WRONLY | O_CREAT; 5155 if (!rewrite_existing) { 5156 oflags |= O_EXCL; 5157 } 5158 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5159} 5160 5161// return current position of file pointer 5162jlong os::current_file_offset(int fd) { 5163 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5164} 5165 5166// move file pointer to the specified offset 5167jlong os::seek_to_file_offset(int fd, jlong offset) { 5168 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5169} 5170 5171// This code originates from JDK's sysAvailable 5172// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5173 5174int os::available(int fd, jlong *bytes) { 5175 jlong cur, end; 5176 int mode; 5177 struct stat64 buf64; 5178 5179 if (::fstat64(fd, &buf64) >= 0) { 5180 mode = buf64.st_mode; 5181 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5182 // XXX: is the following call interruptible? If so, this might 5183 // need to go through the INTERRUPT_IO() wrapper as for other 5184 // blocking, interruptible calls in this file. 5185 int n; 5186 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5187 *bytes = n; 5188 return 1; 5189 } 5190 } 5191 } 5192 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5193 return 0; 5194 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5195 return 0; 5196 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5197 return 0; 5198 } 5199 *bytes = end - cur; 5200 return 1; 5201} 5202 5203// Map a block of memory. 5204char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5205 char *addr, size_t bytes, bool read_only, 5206 bool allow_exec) { 5207 int prot; 5208 int flags = MAP_PRIVATE; 5209 5210 if (read_only) { 5211 prot = PROT_READ; 5212 } else { 5213 prot = PROT_READ | PROT_WRITE; 5214 } 5215 5216 if (allow_exec) { 5217 prot |= PROT_EXEC; 5218 } 5219 5220 if (addr != NULL) { 5221 flags |= MAP_FIXED; 5222 } 5223 5224 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5225 fd, file_offset); 5226 if (mapped_address == MAP_FAILED) { 5227 return NULL; 5228 } 5229 return mapped_address; 5230} 5231 5232 5233// Remap a block of memory. 5234char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5235 char *addr, size_t bytes, bool read_only, 5236 bool allow_exec) { 5237 // same as map_memory() on this OS 5238 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5239 allow_exec); 5240} 5241 5242 5243// Unmap a block of memory. 5244bool os::pd_unmap_memory(char* addr, size_t bytes) { 5245 return munmap(addr, bytes) == 0; 5246} 5247 5248static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5249 5250static clockid_t thread_cpu_clockid(Thread* thread) { 5251 pthread_t tid = thread->osthread()->pthread_id(); 5252 clockid_t clockid; 5253 5254 // Get thread clockid 5255 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5256 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5257 return clockid; 5258} 5259 5260// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5261// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5262// of a thread. 5263// 5264// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5265// the fast estimate available on the platform. 5266 5267jlong os::current_thread_cpu_time() { 5268 if (os::Linux::supports_fast_thread_cpu_time()) { 5269 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5270 } else { 5271 // return user + sys since the cost is the same 5272 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5273 } 5274} 5275 5276jlong os::thread_cpu_time(Thread* thread) { 5277 // consistent with what current_thread_cpu_time() returns 5278 if (os::Linux::supports_fast_thread_cpu_time()) { 5279 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5280 } else { 5281 return slow_thread_cpu_time(thread, true /* user + sys */); 5282 } 5283} 5284 5285jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5286 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5287 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5288 } else { 5289 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5290 } 5291} 5292 5293jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5294 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5295 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5296 } else { 5297 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5298 } 5299} 5300 5301// -1 on error. 5302static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5303 pid_t tid = thread->osthread()->thread_id(); 5304 char *s; 5305 char stat[2048]; 5306 int statlen; 5307 char proc_name[64]; 5308 int count; 5309 long sys_time, user_time; 5310 char cdummy; 5311 int idummy; 5312 long ldummy; 5313 FILE *fp; 5314 5315 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5316 fp = fopen(proc_name, "r"); 5317 if (fp == NULL) return -1; 5318 statlen = fread(stat, 1, 2047, fp); 5319 stat[statlen] = '\0'; 5320 fclose(fp); 5321 5322 // Skip pid and the command string. Note that we could be dealing with 5323 // weird command names, e.g. user could decide to rename java launcher 5324 // to "java 1.4.2 :)", then the stat file would look like 5325 // 1234 (java 1.4.2 :)) R ... ... 5326 // We don't really need to know the command string, just find the last 5327 // occurrence of ")" and then start parsing from there. See bug 4726580. 5328 s = strrchr(stat, ')'); 5329 if (s == NULL) return -1; 5330 5331 // Skip blank chars 5332 do { s++; } while (s && isspace(*s)); 5333 5334 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5335 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5336 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5337 &user_time, &sys_time); 5338 if (count != 13) return -1; 5339 if (user_sys_cpu_time) { 5340 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5341 } else { 5342 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5343 } 5344} 5345 5346void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5347 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5348 info_ptr->may_skip_backward = false; // elapsed time not wall time 5349 info_ptr->may_skip_forward = false; // elapsed time not wall time 5350 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5351} 5352 5353void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5354 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5355 info_ptr->may_skip_backward = false; // elapsed time not wall time 5356 info_ptr->may_skip_forward = false; // elapsed time not wall time 5357 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5358} 5359 5360bool os::is_thread_cpu_time_supported() { 5361 return true; 5362} 5363 5364// System loadavg support. Returns -1 if load average cannot be obtained. 5365// Linux doesn't yet have a (official) notion of processor sets, 5366// so just return the system wide load average. 5367int os::loadavg(double loadavg[], int nelem) { 5368 return ::getloadavg(loadavg, nelem); 5369} 5370 5371void os::pause() { 5372 char filename[MAX_PATH]; 5373 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5374 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5375 } else { 5376 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5377 } 5378 5379 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5380 if (fd != -1) { 5381 struct stat buf; 5382 ::close(fd); 5383 while (::stat(filename, &buf) == 0) { 5384 (void)::poll(NULL, 0, 100); 5385 } 5386 } else { 5387 jio_fprintf(stderr, 5388 "Could not open pause file '%s', continuing immediately.\n", filename); 5389 } 5390} 5391 5392 5393// Refer to the comments in os_solaris.cpp park-unpark. The next two 5394// comment paragraphs are worth repeating here: 5395// 5396// Assumption: 5397// Only one parker can exist on an event, which is why we allocate 5398// them per-thread. Multiple unparkers can coexist. 5399// 5400// _Event serves as a restricted-range semaphore. 5401// -1 : thread is blocked, i.e. there is a waiter 5402// 0 : neutral: thread is running or ready, 5403// could have been signaled after a wait started 5404// 1 : signaled - thread is running or ready 5405// 5406// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5407// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5408// For specifics regarding the bug see GLIBC BUGID 261237 : 5409// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5410// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5411// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5412// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5413// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5414// and monitorenter when we're using 1-0 locking. All those operations may result in 5415// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5416// of libpthread avoids the problem, but isn't practical. 5417// 5418// Possible remedies: 5419// 5420// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5421// This is palliative and probabilistic, however. If the thread is preempted 5422// between the call to compute_abstime() and pthread_cond_timedwait(), more 5423// than the minimum period may have passed, and the abstime may be stale (in the 5424// past) resultin in a hang. Using this technique reduces the odds of a hang 5425// but the JVM is still vulnerable, particularly on heavily loaded systems. 5426// 5427// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5428// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5429// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5430// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5431// thread. 5432// 5433// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5434// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5435// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5436// This also works well. In fact it avoids kernel-level scalability impediments 5437// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5438// timers in a graceful fashion. 5439// 5440// 4. When the abstime value is in the past it appears that control returns 5441// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5442// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5443// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5444// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5445// It may be possible to avoid reinitialization by checking the return 5446// value from pthread_cond_timedwait(). In addition to reinitializing the 5447// condvar we must establish the invariant that cond_signal() is only called 5448// within critical sections protected by the adjunct mutex. This prevents 5449// cond_signal() from "seeing" a condvar that's in the midst of being 5450// reinitialized or that is corrupt. Sadly, this invariant obviates the 5451// desirable signal-after-unlock optimization that avoids futile context switching. 5452// 5453// I'm also concerned that some versions of NTPL might allocate an auxilliary 5454// structure when a condvar is used or initialized. cond_destroy() would 5455// release the helper structure. Our reinitialize-after-timedwait fix 5456// put excessive stress on malloc/free and locks protecting the c-heap. 5457// 5458// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5459// It may be possible to refine (4) by checking the kernel and NTPL verisons 5460// and only enabling the work-around for vulnerable environments. 5461 5462// utility to compute the abstime argument to timedwait: 5463// millis is the relative timeout time 5464// abstime will be the absolute timeout time 5465// TODO: replace compute_abstime() with unpackTime() 5466 5467static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5468 if (millis < 0) millis = 0; 5469 5470 jlong seconds = millis / 1000; 5471 millis %= 1000; 5472 if (seconds > 50000000) { // see man cond_timedwait(3T) 5473 seconds = 50000000; 5474 } 5475 5476 if (os::supports_monotonic_clock()) { 5477 struct timespec now; 5478 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5479 assert_status(status == 0, status, "clock_gettime"); 5480 abstime->tv_sec = now.tv_sec + seconds; 5481 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5482 if (nanos >= NANOSECS_PER_SEC) { 5483 abstime->tv_sec += 1; 5484 nanos -= NANOSECS_PER_SEC; 5485 } 5486 abstime->tv_nsec = nanos; 5487 } else { 5488 struct timeval now; 5489 int status = gettimeofday(&now, NULL); 5490 assert(status == 0, "gettimeofday"); 5491 abstime->tv_sec = now.tv_sec + seconds; 5492 long usec = now.tv_usec + millis * 1000; 5493 if (usec >= 1000000) { 5494 abstime->tv_sec += 1; 5495 usec -= 1000000; 5496 } 5497 abstime->tv_nsec = usec * 1000; 5498 } 5499 return abstime; 5500} 5501 5502void os::PlatformEvent::park() { // AKA "down()" 5503 // Transitions for _Event: 5504 // -1 => -1 : illegal 5505 // 1 => 0 : pass - return immediately 5506 // 0 => -1 : block; then set _Event to 0 before returning 5507 5508 // Invariant: Only the thread associated with the Event/PlatformEvent 5509 // may call park(). 5510 // TODO: assert that _Assoc != NULL or _Assoc == Self 5511 assert(_nParked == 0, "invariant"); 5512 5513 int v; 5514 for (;;) { 5515 v = _Event; 5516 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5517 } 5518 guarantee(v >= 0, "invariant"); 5519 if (v == 0) { 5520 // Do this the hard way by blocking ... 5521 int status = pthread_mutex_lock(_mutex); 5522 assert_status(status == 0, status, "mutex_lock"); 5523 guarantee(_nParked == 0, "invariant"); 5524 ++_nParked; 5525 while (_Event < 0) { 5526 status = pthread_cond_wait(_cond, _mutex); 5527 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5528 // Treat this the same as if the wait was interrupted 5529 if (status == ETIME) { status = EINTR; } 5530 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5531 } 5532 --_nParked; 5533 5534 _Event = 0; 5535 status = pthread_mutex_unlock(_mutex); 5536 assert_status(status == 0, status, "mutex_unlock"); 5537 // Paranoia to ensure our locked and lock-free paths interact 5538 // correctly with each other. 5539 OrderAccess::fence(); 5540 } 5541 guarantee(_Event >= 0, "invariant"); 5542} 5543 5544int os::PlatformEvent::park(jlong millis) { 5545 // Transitions for _Event: 5546 // -1 => -1 : illegal 5547 // 1 => 0 : pass - return immediately 5548 // 0 => -1 : block; then set _Event to 0 before returning 5549 5550 guarantee(_nParked == 0, "invariant"); 5551 5552 int v; 5553 for (;;) { 5554 v = _Event; 5555 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5556 } 5557 guarantee(v >= 0, "invariant"); 5558 if (v != 0) return OS_OK; 5559 5560 // We do this the hard way, by blocking the thread. 5561 // Consider enforcing a minimum timeout value. 5562 struct timespec abst; 5563 compute_abstime(&abst, millis); 5564 5565 int ret = OS_TIMEOUT; 5566 int status = pthread_mutex_lock(_mutex); 5567 assert_status(status == 0, status, "mutex_lock"); 5568 guarantee(_nParked == 0, "invariant"); 5569 ++_nParked; 5570 5571 // Object.wait(timo) will return because of 5572 // (a) notification 5573 // (b) timeout 5574 // (c) thread.interrupt 5575 // 5576 // Thread.interrupt and object.notify{All} both call Event::set. 5577 // That is, we treat thread.interrupt as a special case of notification. 5578 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false. 5579 // We assume all ETIME returns are valid. 5580 // 5581 // TODO: properly differentiate simultaneous notify+interrupt. 5582 // In that case, we should propagate the notify to another waiter. 5583 5584 while (_Event < 0) { 5585 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5586 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5587 pthread_cond_destroy(_cond); 5588 pthread_cond_init(_cond, os::Linux::condAttr()); 5589 } 5590 assert_status(status == 0 || status == EINTR || 5591 status == ETIME || status == ETIMEDOUT, 5592 status, "cond_timedwait"); 5593 if (!FilterSpuriousWakeups) break; // previous semantics 5594 if (status == ETIME || status == ETIMEDOUT) break; 5595 // We consume and ignore EINTR and spurious wakeups. 5596 } 5597 --_nParked; 5598 if (_Event >= 0) { 5599 ret = OS_OK; 5600 } 5601 _Event = 0; 5602 status = pthread_mutex_unlock(_mutex); 5603 assert_status(status == 0, status, "mutex_unlock"); 5604 assert(_nParked == 0, "invariant"); 5605 // Paranoia to ensure our locked and lock-free paths interact 5606 // correctly with each other. 5607 OrderAccess::fence(); 5608 return ret; 5609} 5610 5611void os::PlatformEvent::unpark() { 5612 // Transitions for _Event: 5613 // 0 => 1 : just return 5614 // 1 => 1 : just return 5615 // -1 => either 0 or 1; must signal target thread 5616 // That is, we can safely transition _Event from -1 to either 5617 // 0 or 1. 5618 // See also: "Semaphores in Plan 9" by Mullender & Cox 5619 // 5620 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5621 // that it will take two back-to-back park() calls for the owning 5622 // thread to block. This has the benefit of forcing a spurious return 5623 // from the first park() call after an unpark() call which will help 5624 // shake out uses of park() and unpark() without condition variables. 5625 5626 if (Atomic::xchg(1, &_Event) >= 0) return; 5627 5628 // Wait for the thread associated with the event to vacate 5629 int status = pthread_mutex_lock(_mutex); 5630 assert_status(status == 0, status, "mutex_lock"); 5631 int AnyWaiters = _nParked; 5632 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5633 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5634 AnyWaiters = 0; 5635 pthread_cond_signal(_cond); 5636 } 5637 status = pthread_mutex_unlock(_mutex); 5638 assert_status(status == 0, status, "mutex_unlock"); 5639 if (AnyWaiters != 0) { 5640 // Note that we signal() *after* dropping the lock for "immortal" Events. 5641 // This is safe and avoids a common class of futile wakeups. In rare 5642 // circumstances this can cause a thread to return prematurely from 5643 // cond_{timed}wait() but the spurious wakeup is benign and the victim 5644 // will simply re-test the condition and re-park itself. 5645 // This provides particular benefit if the underlying platform does not 5646 // provide wait morphing. 5647 status = pthread_cond_signal(_cond); 5648 assert_status(status == 0, status, "cond_signal"); 5649 } 5650} 5651 5652 5653// JSR166 5654// ------------------------------------------------------- 5655 5656// The solaris and linux implementations of park/unpark are fairly 5657// conservative for now, but can be improved. They currently use a 5658// mutex/condvar pair, plus a a count. 5659// Park decrements count if > 0, else does a condvar wait. Unpark 5660// sets count to 1 and signals condvar. Only one thread ever waits 5661// on the condvar. Contention seen when trying to park implies that someone 5662// is unparking you, so don't wait. And spurious returns are fine, so there 5663// is no need to track notifications. 5664 5665// This code is common to linux and solaris and will be moved to a 5666// common place in dolphin. 5667// 5668// The passed in time value is either a relative time in nanoseconds 5669// or an absolute time in milliseconds. Either way it has to be unpacked 5670// into suitable seconds and nanoseconds components and stored in the 5671// given timespec structure. 5672// Given time is a 64-bit value and the time_t used in the timespec is only 5673// a signed-32-bit value (except on 64-bit Linux) we have to watch for 5674// overflow if times way in the future are given. Further on Solaris versions 5675// prior to 10 there is a restriction (see cond_timedwait) that the specified 5676// number of seconds, in abstime, is less than current_time + 100,000,000. 5677// As it will be 28 years before "now + 100000000" will overflow we can 5678// ignore overflow and just impose a hard-limit on seconds using the value 5679// of "now + 100,000,000". This places a limit on the timeout of about 3.17 5680// years from "now". 5681 5682static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5683 assert(time > 0, "convertTime"); 5684 time_t max_secs = 0; 5685 5686 if (!os::supports_monotonic_clock() || isAbsolute) { 5687 struct timeval now; 5688 int status = gettimeofday(&now, NULL); 5689 assert(status == 0, "gettimeofday"); 5690 5691 max_secs = now.tv_sec + MAX_SECS; 5692 5693 if (isAbsolute) { 5694 jlong secs = time / 1000; 5695 if (secs > max_secs) { 5696 absTime->tv_sec = max_secs; 5697 } else { 5698 absTime->tv_sec = secs; 5699 } 5700 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5701 } else { 5702 jlong secs = time / NANOSECS_PER_SEC; 5703 if (secs >= MAX_SECS) { 5704 absTime->tv_sec = max_secs; 5705 absTime->tv_nsec = 0; 5706 } else { 5707 absTime->tv_sec = now.tv_sec + secs; 5708 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5709 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5710 absTime->tv_nsec -= NANOSECS_PER_SEC; 5711 ++absTime->tv_sec; // note: this must be <= max_secs 5712 } 5713 } 5714 } 5715 } else { 5716 // must be relative using monotonic clock 5717 struct timespec now; 5718 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5719 assert_status(status == 0, status, "clock_gettime"); 5720 max_secs = now.tv_sec + MAX_SECS; 5721 jlong secs = time / NANOSECS_PER_SEC; 5722 if (secs >= MAX_SECS) { 5723 absTime->tv_sec = max_secs; 5724 absTime->tv_nsec = 0; 5725 } else { 5726 absTime->tv_sec = now.tv_sec + secs; 5727 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5728 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5729 absTime->tv_nsec -= NANOSECS_PER_SEC; 5730 ++absTime->tv_sec; // note: this must be <= max_secs 5731 } 5732 } 5733 } 5734 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5735 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5736 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5737 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5738} 5739 5740void Parker::park(bool isAbsolute, jlong time) { 5741 // Ideally we'd do something useful while spinning, such 5742 // as calling unpackTime(). 5743 5744 // Optional fast-path check: 5745 // Return immediately if a permit is available. 5746 // We depend on Atomic::xchg() having full barrier semantics 5747 // since we are doing a lock-free update to _counter. 5748 if (Atomic::xchg(0, &_counter) > 0) return; 5749 5750 Thread* thread = Thread::current(); 5751 assert(thread->is_Java_thread(), "Must be JavaThread"); 5752 JavaThread *jt = (JavaThread *)thread; 5753 5754 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5755 // Check interrupt before trying to wait 5756 if (Thread::is_interrupted(thread, false)) { 5757 return; 5758 } 5759 5760 // Next, demultiplex/decode time arguments 5761 timespec absTime; 5762 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all 5763 return; 5764 } 5765 if (time > 0) { 5766 unpackTime(&absTime, isAbsolute, time); 5767 } 5768 5769 5770 // Enter safepoint region 5771 // Beware of deadlocks such as 6317397. 5772 // The per-thread Parker:: mutex is a classic leaf-lock. 5773 // In particular a thread must never block on the Threads_lock while 5774 // holding the Parker:: mutex. If safepoints are pending both the 5775 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5776 ThreadBlockInVM tbivm(jt); 5777 5778 // Don't wait if cannot get lock since interference arises from 5779 // unblocking. Also. check interrupt before trying wait 5780 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5781 return; 5782 } 5783 5784 int status; 5785 if (_counter > 0) { // no wait needed 5786 _counter = 0; 5787 status = pthread_mutex_unlock(_mutex); 5788 assert(status == 0, "invariant"); 5789 // Paranoia to ensure our locked and lock-free paths interact 5790 // correctly with each other and Java-level accesses. 5791 OrderAccess::fence(); 5792 return; 5793 } 5794 5795#ifdef ASSERT 5796 // Don't catch signals while blocked; let the running threads have the signals. 5797 // (This allows a debugger to break into the running thread.) 5798 sigset_t oldsigs; 5799 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5800 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5801#endif 5802 5803 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5804 jt->set_suspend_equivalent(); 5805 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5806 5807 assert(_cur_index == -1, "invariant"); 5808 if (time == 0) { 5809 _cur_index = REL_INDEX; // arbitrary choice when not timed 5810 status = pthread_cond_wait(&_cond[_cur_index], _mutex); 5811 } else { 5812 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5813 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime); 5814 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5815 pthread_cond_destroy(&_cond[_cur_index]); 5816 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr()); 5817 } 5818 } 5819 _cur_index = -1; 5820 assert_status(status == 0 || status == EINTR || 5821 status == ETIME || status == ETIMEDOUT, 5822 status, "cond_timedwait"); 5823 5824#ifdef ASSERT 5825 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5826#endif 5827 5828 _counter = 0; 5829 status = pthread_mutex_unlock(_mutex); 5830 assert_status(status == 0, status, "invariant"); 5831 // Paranoia to ensure our locked and lock-free paths interact 5832 // correctly with each other and Java-level accesses. 5833 OrderAccess::fence(); 5834 5835 // If externally suspended while waiting, re-suspend 5836 if (jt->handle_special_suspend_equivalent_condition()) { 5837 jt->java_suspend_self(); 5838 } 5839} 5840 5841void Parker::unpark() { 5842 int status = pthread_mutex_lock(_mutex); 5843 assert(status == 0, "invariant"); 5844 const int s = _counter; 5845 _counter = 1; 5846 if (s < 1) { 5847 // thread might be parked 5848 if (_cur_index != -1) { 5849 // thread is definitely parked 5850 if (WorkAroundNPTLTimedWaitHang) { 5851 status = pthread_cond_signal(&_cond[_cur_index]); 5852 assert(status == 0, "invariant"); 5853 status = pthread_mutex_unlock(_mutex); 5854 assert(status == 0, "invariant"); 5855 } else { 5856 status = pthread_mutex_unlock(_mutex); 5857 assert(status == 0, "invariant"); 5858 status = pthread_cond_signal(&_cond[_cur_index]); 5859 assert(status == 0, "invariant"); 5860 } 5861 } else { 5862 pthread_mutex_unlock(_mutex); 5863 assert(status == 0, "invariant"); 5864 } 5865 } else { 5866 pthread_mutex_unlock(_mutex); 5867 assert(status == 0, "invariant"); 5868 } 5869} 5870 5871 5872extern char** environ; 5873 5874#ifndef __NR_fork 5875 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) AARCH64_ONLY(1079) 5876#endif 5877 5878#ifndef __NR_execve 5879 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) AARCH64_ONLY(221) 5880#endif 5881 5882// Run the specified command in a separate process. Return its exit value, 5883// or -1 on failure (e.g. can't fork a new process). 5884// Unlike system(), this function can be called from signal handler. It 5885// doesn't block SIGINT et al. 5886int os::fork_and_exec(char* cmd) { 5887 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5888 5889 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5890 // pthread_atfork handlers and reset pthread library. All we need is a 5891 // separate process to execve. Make a direct syscall to fork process. 5892 // On IA64 there's no fork syscall, we have to use fork() and hope for 5893 // the best... 5894 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5895 IA64_ONLY(fork();) 5896 5897 if (pid < 0) { 5898 // fork failed 5899 return -1; 5900 5901 } else if (pid == 0) { 5902 // child process 5903 5904 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5905 // first to kill every thread on the thread list. Because this list is 5906 // not reset by fork() (see notes above), execve() will instead kill 5907 // every thread in the parent process. We know this is the only thread 5908 // in the new process, so make a system call directly. 5909 // IA64 should use normal execve() from glibc to match the glibc fork() 5910 // above. 5911 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5912 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5913 5914 // execve failed 5915 _exit(-1); 5916 5917 } else { 5918 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5919 // care about the actual exit code, for now. 5920 5921 int status; 5922 5923 // Wait for the child process to exit. This returns immediately if 5924 // the child has already exited. */ 5925 while (waitpid(pid, &status, 0) < 0) { 5926 switch (errno) { 5927 case ECHILD: return 0; 5928 case EINTR: break; 5929 default: return -1; 5930 } 5931 } 5932 5933 if (WIFEXITED(status)) { 5934 // The child exited normally; get its exit code. 5935 return WEXITSTATUS(status); 5936 } else if (WIFSIGNALED(status)) { 5937 // The child exited because of a signal 5938 // The best value to return is 0x80 + signal number, 5939 // because that is what all Unix shells do, and because 5940 // it allows callers to distinguish between process exit and 5941 // process death by signal. 5942 return 0x80 + WTERMSIG(status); 5943 } else { 5944 // Unknown exit code; pass it through 5945 return status; 5946 } 5947 } 5948} 5949 5950// is_headless_jre() 5951// 5952// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5953// in order to report if we are running in a headless jre 5954// 5955// Since JDK8 xawt/libmawt.so was moved into the same directory 5956// as libawt.so, and renamed libawt_xawt.so 5957// 5958bool os::is_headless_jre() { 5959 struct stat statbuf; 5960 char buf[MAXPATHLEN]; 5961 char libmawtpath[MAXPATHLEN]; 5962 const char *xawtstr = "/xawt/libmawt.so"; 5963 const char *new_xawtstr = "/libawt_xawt.so"; 5964 char *p; 5965 5966 // Get path to libjvm.so 5967 os::jvm_path(buf, sizeof(buf)); 5968 5969 // Get rid of libjvm.so 5970 p = strrchr(buf, '/'); 5971 if (p == NULL) { 5972 return false; 5973 } else { 5974 *p = '\0'; 5975 } 5976 5977 // Get rid of client or server 5978 p = strrchr(buf, '/'); 5979 if (p == NULL) { 5980 return false; 5981 } else { 5982 *p = '\0'; 5983 } 5984 5985 // check xawt/libmawt.so 5986 strcpy(libmawtpath, buf); 5987 strcat(libmawtpath, xawtstr); 5988 if (::stat(libmawtpath, &statbuf) == 0) return false; 5989 5990 // check libawt_xawt.so 5991 strcpy(libmawtpath, buf); 5992 strcat(libmawtpath, new_xawtstr); 5993 if (::stat(libmawtpath, &statbuf) == 0) return false; 5994 5995 return true; 5996} 5997 5998// Get the default path to the core file 5999// Returns the length of the string 6000int os::get_core_path(char* buffer, size_t bufferSize) { 6001 const char* p = get_current_directory(buffer, bufferSize); 6002 6003 if (p == NULL) { 6004 assert(p != NULL, "failed to get current directory"); 6005 return 0; 6006 } 6007 6008 return strlen(buffer); 6009} 6010 6011/////////////// Unit tests /////////////// 6012 6013#ifndef PRODUCT 6014 6015#define test_log(...) \ 6016 do { \ 6017 if (VerboseInternalVMTests) { \ 6018 tty->print_cr(__VA_ARGS__); \ 6019 tty->flush(); \ 6020 } \ 6021 } while (false) 6022 6023class TestReserveMemorySpecial : AllStatic { 6024 public: 6025 static void small_page_write(void* addr, size_t size) { 6026 size_t page_size = os::vm_page_size(); 6027 6028 char* end = (char*)addr + size; 6029 for (char* p = (char*)addr; p < end; p += page_size) { 6030 *p = 1; 6031 } 6032 } 6033 6034 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6035 if (!UseHugeTLBFS) { 6036 return; 6037 } 6038 6039 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6040 6041 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6042 6043 if (addr != NULL) { 6044 small_page_write(addr, size); 6045 6046 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6047 } 6048 } 6049 6050 static void test_reserve_memory_special_huge_tlbfs_only() { 6051 if (!UseHugeTLBFS) { 6052 return; 6053 } 6054 6055 size_t lp = os::large_page_size(); 6056 6057 for (size_t size = lp; size <= lp * 10; size += lp) { 6058 test_reserve_memory_special_huge_tlbfs_only(size); 6059 } 6060 } 6061 6062 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) { 6063 if (!UseHugeTLBFS) { 6064 return; 6065 } 6066 6067 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")", 6068 size, alignment); 6069 6070 assert(size >= os::large_page_size(), "Incorrect input to test"); 6071 6072 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6073 6074 if (addr != NULL) { 6075 small_page_write(addr, size); 6076 6077 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6078 } 6079 } 6080 6081 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) { 6082 size_t lp = os::large_page_size(); 6083 size_t ag = os::vm_allocation_granularity(); 6084 6085 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6086 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment); 6087 } 6088 } 6089 6090 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6091 size_t lp = os::large_page_size(); 6092 size_t ag = os::vm_allocation_granularity(); 6093 6094 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp); 6095 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag); 6096 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2); 6097 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2); 6098 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag); 6099 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag); 6100 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2); 6101 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10); 6102 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2); 6103 } 6104 6105 static void test_reserve_memory_special_huge_tlbfs() { 6106 if (!UseHugeTLBFS) { 6107 return; 6108 } 6109 6110 test_reserve_memory_special_huge_tlbfs_only(); 6111 test_reserve_memory_special_huge_tlbfs_mixed(); 6112 } 6113 6114 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6115 if (!UseSHM) { 6116 return; 6117 } 6118 6119 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6120 6121 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6122 6123 if (addr != NULL) { 6124 assert(is_ptr_aligned(addr, alignment), "Check"); 6125 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6126 6127 small_page_write(addr, size); 6128 6129 os::Linux::release_memory_special_shm(addr, size); 6130 } 6131 } 6132 6133 static void test_reserve_memory_special_shm() { 6134 size_t lp = os::large_page_size(); 6135 size_t ag = os::vm_allocation_granularity(); 6136 6137 for (size_t size = ag; size < lp * 3; size += ag) { 6138 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6139 test_reserve_memory_special_shm(size, alignment); 6140 } 6141 } 6142 } 6143 6144 static void test() { 6145 test_reserve_memory_special_huge_tlbfs(); 6146 test_reserve_memory_special_shm(); 6147 } 6148}; 6149 6150void TestReserveMemorySpecial_test() { 6151 TestReserveMemorySpecial::test(); 6152} 6153 6154#endif 6155