os_linux.cpp revision 5215:bb57d48691f5
1/* 2 * Copyright (c) 1999, 2013, 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_share_linux.hpp" 40#include "prims/jniFastGetField.hpp" 41#include "prims/jvm.h" 42#include "prims/jvm_misc.hpp" 43#include "runtime/arguments.hpp" 44#include "runtime/extendedPC.hpp" 45#include "runtime/globals.hpp" 46#include "runtime/interfaceSupport.hpp" 47#include "runtime/init.hpp" 48#include "runtime/java.hpp" 49#include "runtime/javaCalls.hpp" 50#include "runtime/mutexLocker.hpp" 51#include "runtime/objectMonitor.hpp" 52#include "runtime/osThread.hpp" 53#include "runtime/perfMemory.hpp" 54#include "runtime/sharedRuntime.hpp" 55#include "runtime/statSampler.hpp" 56#include "runtime/stubRoutines.hpp" 57#include "runtime/thread.inline.hpp" 58#include "runtime/threadCritical.hpp" 59#include "runtime/timer.hpp" 60#include "services/attachListener.hpp" 61#include "services/memTracker.hpp" 62#include "services/runtimeService.hpp" 63#include "utilities/decoder.hpp" 64#include "utilities/defaultStream.hpp" 65#include "utilities/events.hpp" 66#include "utilities/elfFile.hpp" 67#include "utilities/growableArray.hpp" 68#include "utilities/vmError.hpp" 69 70// put OS-includes here 71# include <sys/types.h> 72# include <sys/mman.h> 73# include <sys/stat.h> 74# include <sys/select.h> 75# include <pthread.h> 76# include <signal.h> 77# include <errno.h> 78# include <dlfcn.h> 79# include <stdio.h> 80# include <unistd.h> 81# include <sys/resource.h> 82# include <pthread.h> 83# include <sys/stat.h> 84# include <sys/time.h> 85# include <sys/times.h> 86# include <sys/utsname.h> 87# include <sys/socket.h> 88# include <sys/wait.h> 89# include <pwd.h> 90# include <poll.h> 91# include <semaphore.h> 92# include <fcntl.h> 93# include <string.h> 94# include <syscall.h> 95# include <sys/sysinfo.h> 96# include <gnu/libc-version.h> 97# include <sys/ipc.h> 98# include <sys/shm.h> 99# include <link.h> 100# include <stdint.h> 101# include <inttypes.h> 102# include <sys/ioctl.h> 103 104// if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 105// getrusage() is prepared to handle the associated failure. 106#ifndef RUSAGE_THREAD 107#define RUSAGE_THREAD (1) /* only the calling thread */ 108#endif 109 110#define MAX_PATH (2 * K) 111 112// for timer info max values which include all bits 113#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 114 115#define LARGEPAGES_BIT (1 << 6) 116//////////////////////////////////////////////////////////////////////////////// 117// global variables 118julong os::Linux::_physical_memory = 0; 119 120address os::Linux::_initial_thread_stack_bottom = NULL; 121uintptr_t os::Linux::_initial_thread_stack_size = 0; 122 123int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 124int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 125Mutex* os::Linux::_createThread_lock = NULL; 126pthread_t os::Linux::_main_thread; 127int os::Linux::_page_size = -1; 128const int os::Linux::_vm_default_page_size = (8 * K); 129bool os::Linux::_is_floating_stack = false; 130bool os::Linux::_is_NPTL = false; 131bool os::Linux::_supports_fast_thread_cpu_time = false; 132const char * os::Linux::_glibc_version = NULL; 133const char * os::Linux::_libpthread_version = NULL; 134 135static jlong initial_time_count=0; 136 137static int clock_tics_per_sec = 100; 138 139// For diagnostics to print a message once. see run_periodic_checks 140static sigset_t check_signal_done; 141static bool check_signals = true;; 142 143static pid_t _initial_pid = 0; 144 145/* Signal number used to suspend/resume a thread */ 146 147/* do not use any signal number less than SIGSEGV, see 4355769 */ 148static int SR_signum = SIGUSR2; 149sigset_t SR_sigset; 150 151/* Used to protect dlsym() calls */ 152static pthread_mutex_t dl_mutex; 153 154// Declarations 155static void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 156 157#ifdef JAVASE_EMBEDDED 158class MemNotifyThread: public Thread { 159 friend class VMStructs; 160 public: 161 virtual void run(); 162 163 private: 164 static MemNotifyThread* _memnotify_thread; 165 int _fd; 166 167 public: 168 169 // Constructor 170 MemNotifyThread(int fd); 171 172 // Tester 173 bool is_memnotify_thread() const { return true; } 174 175 // Printing 176 char* name() const { return (char*)"Linux MemNotify Thread"; } 177 178 // Returns the single instance of the MemNotifyThread 179 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; } 180 181 // Create and start the single instance of MemNotifyThread 182 static void start(); 183}; 184#endif // JAVASE_EMBEDDED 185 186// utility functions 187 188static int SR_initialize(); 189 190julong os::available_memory() { 191 return Linux::available_memory(); 192} 193 194julong os::Linux::available_memory() { 195 // values in struct sysinfo are "unsigned long" 196 struct sysinfo si; 197 sysinfo(&si); 198 199 return (julong)si.freeram * si.mem_unit; 200} 201 202julong os::physical_memory() { 203 return Linux::physical_memory(); 204} 205 206//////////////////////////////////////////////////////////////////////////////// 207// environment support 208 209bool os::getenv(const char* name, char* buf, int len) { 210 const char* val = ::getenv(name); 211 if (val != NULL && strlen(val) < (size_t)len) { 212 strcpy(buf, val); 213 return true; 214 } 215 if (len > 0) buf[0] = 0; // return a null string 216 return false; 217} 218 219 220// Return true if user is running as root. 221 222bool os::have_special_privileges() { 223 static bool init = false; 224 static bool privileges = false; 225 if (!init) { 226 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 227 init = true; 228 } 229 return privileges; 230} 231 232 233#ifndef SYS_gettid 234// i386: 224, ia64: 1105, amd64: 186, sparc 143 235#ifdef __ia64__ 236#define SYS_gettid 1105 237#elif __i386__ 238#define SYS_gettid 224 239#elif __amd64__ 240#define SYS_gettid 186 241#elif __sparc__ 242#define SYS_gettid 143 243#else 244#error define gettid for the arch 245#endif 246#endif 247 248// Cpu architecture string 249#if defined(ZERO) 250static char cpu_arch[] = ZERO_LIBARCH; 251#elif defined(IA64) 252static char cpu_arch[] = "ia64"; 253#elif defined(IA32) 254static char cpu_arch[] = "i386"; 255#elif defined(AMD64) 256static char cpu_arch[] = "amd64"; 257#elif defined(ARM) 258static char cpu_arch[] = "arm"; 259#elif defined(PPC) 260static char cpu_arch[] = "ppc"; 261#elif defined(SPARC) 262# ifdef _LP64 263static char cpu_arch[] = "sparcv9"; 264# else 265static char cpu_arch[] = "sparc"; 266# endif 267#else 268#error Add appropriate cpu_arch setting 269#endif 270 271 272// pid_t gettid() 273// 274// Returns the kernel thread id of the currently running thread. Kernel 275// thread id is used to access /proc. 276// 277// (Note that getpid() on LinuxThreads returns kernel thread id too; but 278// on NPTL, it returns the same pid for all threads, as required by POSIX.) 279// 280pid_t os::Linux::gettid() { 281 int rslt = syscall(SYS_gettid); 282 if (rslt == -1) { 283 // old kernel, no NPTL support 284 return getpid(); 285 } else { 286 return (pid_t)rslt; 287 } 288} 289 290// Most versions of linux have a bug where the number of processors are 291// determined by looking at the /proc file system. In a chroot environment, 292// the system call returns 1. This causes the VM to act as if it is 293// a single processor and elide locking (see is_MP() call). 294static bool unsafe_chroot_detected = false; 295static const char *unstable_chroot_error = "/proc file system not found.\n" 296 "Java may be unstable running multithreaded in a chroot " 297 "environment on Linux when /proc filesystem is not mounted."; 298 299void os::Linux::initialize_system_info() { 300 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 301 if (processor_count() == 1) { 302 pid_t pid = os::Linux::gettid(); 303 char fname[32]; 304 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 305 FILE *fp = fopen(fname, "r"); 306 if (fp == NULL) { 307 unsafe_chroot_detected = true; 308 } else { 309 fclose(fp); 310 } 311 } 312 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 313 assert(processor_count() > 0, "linux error"); 314} 315 316void os::init_system_properties_values() { 317// char arch[12]; 318// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 319 320 // The next steps are taken in the product version: 321 // 322 // Obtain the JAVA_HOME value from the location of libjvm.so. 323 // This library should be located at: 324 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 325 // 326 // If "/jre/lib/" appears at the right place in the path, then we 327 // assume libjvm.so is installed in a JDK and we use this path. 328 // 329 // Otherwise exit with message: "Could not create the Java virtual machine." 330 // 331 // The following extra steps are taken in the debugging version: 332 // 333 // If "/jre/lib/" does NOT appear at the right place in the path 334 // instead of exit check for $JAVA_HOME environment variable. 335 // 336 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 337 // then we append a fake suffix "hotspot/libjvm.so" to this path so 338 // it looks like libjvm.so is installed there 339 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 340 // 341 // Otherwise exit. 342 // 343 // Important note: if the location of libjvm.so changes this 344 // code needs to be changed accordingly. 345 346 // The next few definitions allow the code to be verbatim: 347#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal) 348#define getenv(n) ::getenv(n) 349 350/* 351 * See ld(1): 352 * The linker uses the following search paths to locate required 353 * shared libraries: 354 * 1: ... 355 * ... 356 * 7: The default directories, normally /lib and /usr/lib. 357 */ 358#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 359#define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 360#else 361#define DEFAULT_LIBPATH "/lib:/usr/lib" 362#endif 363 364#define EXTENSIONS_DIR "/lib/ext" 365#define ENDORSED_DIR "/lib/endorsed" 366#define REG_DIR "/usr/java/packages" 367 368 { 369 /* sysclasspath, java_home, dll_dir */ 370 { 371 char *home_path; 372 char *dll_path; 373 char *pslash; 374 char buf[MAXPATHLEN]; 375 os::jvm_path(buf, sizeof(buf)); 376 377 // Found the full path to libjvm.so. 378 // Now cut the path to <java_home>/jre if we can. 379 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 380 pslash = strrchr(buf, '/'); 381 if (pslash != NULL) 382 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 383 dll_path = malloc(strlen(buf) + 1); 384 if (dll_path == NULL) 385 return; 386 strcpy(dll_path, buf); 387 Arguments::set_dll_dir(dll_path); 388 389 if (pslash != NULL) { 390 pslash = strrchr(buf, '/'); 391 if (pslash != NULL) { 392 *pslash = '\0'; /* get rid of /<arch> */ 393 pslash = strrchr(buf, '/'); 394 if (pslash != NULL) 395 *pslash = '\0'; /* get rid of /lib */ 396 } 397 } 398 399 home_path = malloc(strlen(buf) + 1); 400 if (home_path == NULL) 401 return; 402 strcpy(home_path, buf); 403 Arguments::set_java_home(home_path); 404 405 if (!set_boot_path('/', ':')) 406 return; 407 } 408 409 /* 410 * Where to look for native libraries 411 * 412 * Note: Due to a legacy implementation, most of the library path 413 * is set in the launcher. This was to accomodate linking restrictions 414 * on legacy Linux implementations (which are no longer supported). 415 * Eventually, all the library path setting will be done here. 416 * 417 * However, to prevent the proliferation of improperly built native 418 * libraries, the new path component /usr/java/packages is added here. 419 * Eventually, all the library path setting will be done here. 420 */ 421 { 422 char *ld_library_path; 423 424 /* 425 * Construct the invariant part of ld_library_path. Note that the 426 * space for the colon and the trailing null are provided by the 427 * nulls included by the sizeof operator (so actually we allocate 428 * a byte more than necessary). 429 */ 430 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 431 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 432 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 433 434 /* 435 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 436 * should always exist (until the legacy problem cited above is 437 * addressed). 438 */ 439 char *v = getenv("LD_LIBRARY_PATH"); 440 if (v != NULL) { 441 char *t = ld_library_path; 442 /* That's +1 for the colon and +1 for the trailing '\0' */ 443 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 444 sprintf(ld_library_path, "%s:%s", v, t); 445 } 446 Arguments::set_library_path(ld_library_path); 447 } 448 449 /* 450 * Extensions directories. 451 * 452 * Note that the space for the colon and the trailing null are provided 453 * by the nulls included by the sizeof operator (so actually one byte more 454 * than necessary is allocated). 455 */ 456 { 457 char *buf = malloc(strlen(Arguments::get_java_home()) + 458 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 459 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 460 Arguments::get_java_home()); 461 Arguments::set_ext_dirs(buf); 462 } 463 464 /* Endorsed standards default directory. */ 465 { 466 char * buf; 467 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 468 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 469 Arguments::set_endorsed_dirs(buf); 470 } 471 } 472 473#undef malloc 474#undef getenv 475#undef EXTENSIONS_DIR 476#undef ENDORSED_DIR 477 478 // Done 479 return; 480} 481 482//////////////////////////////////////////////////////////////////////////////// 483// breakpoint support 484 485void os::breakpoint() { 486 BREAKPOINT; 487} 488 489extern "C" void breakpoint() { 490 // use debugger to set breakpoint here 491} 492 493//////////////////////////////////////////////////////////////////////////////// 494// signal support 495 496debug_only(static bool signal_sets_initialized = false); 497static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 498 499bool os::Linux::is_sig_ignored(int sig) { 500 struct sigaction oact; 501 sigaction(sig, (struct sigaction*)NULL, &oact); 502 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 503 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 504 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 505 return true; 506 else 507 return false; 508} 509 510void os::Linux::signal_sets_init() { 511 // Should also have an assertion stating we are still single-threaded. 512 assert(!signal_sets_initialized, "Already initialized"); 513 // Fill in signals that are necessarily unblocked for all threads in 514 // the VM. Currently, we unblock the following signals: 515 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 516 // by -Xrs (=ReduceSignalUsage)); 517 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 518 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 519 // the dispositions or masks wrt these signals. 520 // Programs embedding the VM that want to use the above signals for their 521 // own purposes must, at this time, use the "-Xrs" option to prevent 522 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 523 // (See bug 4345157, and other related bugs). 524 // In reality, though, unblocking these signals is really a nop, since 525 // these signals are not blocked by default. 526 sigemptyset(&unblocked_sigs); 527 sigemptyset(&allowdebug_blocked_sigs); 528 sigaddset(&unblocked_sigs, SIGILL); 529 sigaddset(&unblocked_sigs, SIGSEGV); 530 sigaddset(&unblocked_sigs, SIGBUS); 531 sigaddset(&unblocked_sigs, SIGFPE); 532 sigaddset(&unblocked_sigs, SR_signum); 533 534 if (!ReduceSignalUsage) { 535 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 536 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 537 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 538 } 539 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 540 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 541 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 542 } 543 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 544 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 545 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 546 } 547 } 548 // Fill in signals that are blocked by all but the VM thread. 549 sigemptyset(&vm_sigs); 550 if (!ReduceSignalUsage) 551 sigaddset(&vm_sigs, BREAK_SIGNAL); 552 debug_only(signal_sets_initialized = true); 553 554} 555 556// These are signals that are unblocked while a thread is running Java. 557// (For some reason, they get blocked by default.) 558sigset_t* os::Linux::unblocked_signals() { 559 assert(signal_sets_initialized, "Not initialized"); 560 return &unblocked_sigs; 561} 562 563// These are the signals that are blocked while a (non-VM) thread is 564// running Java. Only the VM thread handles these signals. 565sigset_t* os::Linux::vm_signals() { 566 assert(signal_sets_initialized, "Not initialized"); 567 return &vm_sigs; 568} 569 570// These are signals that are blocked during cond_wait to allow debugger in 571sigset_t* os::Linux::allowdebug_blocked_signals() { 572 assert(signal_sets_initialized, "Not initialized"); 573 return &allowdebug_blocked_sigs; 574} 575 576void os::Linux::hotspot_sigmask(Thread* thread) { 577 578 //Save caller's signal mask before setting VM signal mask 579 sigset_t caller_sigmask; 580 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 581 582 OSThread* osthread = thread->osthread(); 583 osthread->set_caller_sigmask(caller_sigmask); 584 585 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 586 587 if (!ReduceSignalUsage) { 588 if (thread->is_VM_thread()) { 589 // Only the VM thread handles BREAK_SIGNAL ... 590 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 591 } else { 592 // ... all other threads block BREAK_SIGNAL 593 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 594 } 595 } 596} 597 598////////////////////////////////////////////////////////////////////////////// 599// detecting pthread library 600 601void os::Linux::libpthread_init() { 602 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 603 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 604 // generic name for earlier versions. 605 // Define macros here so we can build HotSpot on old systems. 606# ifndef _CS_GNU_LIBC_VERSION 607# define _CS_GNU_LIBC_VERSION 2 608# endif 609# ifndef _CS_GNU_LIBPTHREAD_VERSION 610# define _CS_GNU_LIBPTHREAD_VERSION 3 611# endif 612 613 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 614 if (n > 0) { 615 char *str = (char *)malloc(n, mtInternal); 616 confstr(_CS_GNU_LIBC_VERSION, str, n); 617 os::Linux::set_glibc_version(str); 618 } else { 619 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 620 static char _gnu_libc_version[32]; 621 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 622 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 623 os::Linux::set_glibc_version(_gnu_libc_version); 624 } 625 626 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 627 if (n > 0) { 628 char *str = (char *)malloc(n, mtInternal); 629 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 630 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 631 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 632 // is the case. LinuxThreads has a hard limit on max number of threads. 633 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 634 // On the other hand, NPTL does not have such a limit, sysconf() 635 // will return -1 and errno is not changed. Check if it is really NPTL. 636 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 637 strstr(str, "NPTL") && 638 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 639 free(str); 640 os::Linux::set_libpthread_version("linuxthreads"); 641 } else { 642 os::Linux::set_libpthread_version(str); 643 } 644 } else { 645 // glibc before 2.3.2 only has LinuxThreads. 646 os::Linux::set_libpthread_version("linuxthreads"); 647 } 648 649 if (strstr(libpthread_version(), "NPTL")) { 650 os::Linux::set_is_NPTL(); 651 } else { 652 os::Linux::set_is_LinuxThreads(); 653 } 654 655 // LinuxThreads have two flavors: floating-stack mode, which allows variable 656 // stack size; and fixed-stack mode. NPTL is always floating-stack. 657 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 658 os::Linux::set_is_floating_stack(); 659 } 660} 661 662///////////////////////////////////////////////////////////////////////////// 663// thread stack 664 665// Force Linux kernel to expand current thread stack. If "bottom" is close 666// to the stack guard, caller should block all signals. 667// 668// MAP_GROWSDOWN: 669// A special mmap() flag that is used to implement thread stacks. It tells 670// kernel that the memory region should extend downwards when needed. This 671// allows early versions of LinuxThreads to only mmap the first few pages 672// when creating a new thread. Linux kernel will automatically expand thread 673// stack as needed (on page faults). 674// 675// However, because the memory region of a MAP_GROWSDOWN stack can grow on 676// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 677// region, it's hard to tell if the fault is due to a legitimate stack 678// access or because of reading/writing non-exist memory (e.g. buffer 679// overrun). As a rule, if the fault happens below current stack pointer, 680// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 681// application (see Linux kernel fault.c). 682// 683// This Linux feature can cause SIGSEGV when VM bangs thread stack for 684// stack overflow detection. 685// 686// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 687// not use this flag. However, the stack of initial thread is not created 688// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 689// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 690// and then attach the thread to JVM. 691// 692// To get around the problem and allow stack banging on Linux, we need to 693// manually expand thread stack after receiving the SIGSEGV. 694// 695// There are two ways to expand thread stack to address "bottom", we used 696// both of them in JVM before 1.5: 697// 1. adjust stack pointer first so that it is below "bottom", and then 698// touch "bottom" 699// 2. mmap() the page in question 700// 701// Now alternate signal stack is gone, it's harder to use 2. For instance, 702// if current sp is already near the lower end of page 101, and we need to 703// call mmap() to map page 100, it is possible that part of the mmap() frame 704// will be placed in page 100. When page 100 is mapped, it is zero-filled. 705// That will destroy the mmap() frame and cause VM to crash. 706// 707// The following code works by adjusting sp first, then accessing the "bottom" 708// page to force a page fault. Linux kernel will then automatically expand the 709// stack mapping. 710// 711// _expand_stack_to() assumes its frame size is less than page size, which 712// should always be true if the function is not inlined. 713 714#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 715#define NOINLINE 716#else 717#define NOINLINE __attribute__ ((noinline)) 718#endif 719 720static void _expand_stack_to(address bottom) NOINLINE; 721 722static void _expand_stack_to(address bottom) { 723 address sp; 724 size_t size; 725 volatile char *p; 726 727 // Adjust bottom to point to the largest address within the same page, it 728 // gives us a one-page buffer if alloca() allocates slightly more memory. 729 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 730 bottom += os::Linux::page_size() - 1; 731 732 // sp might be slightly above current stack pointer; if that's the case, we 733 // will alloca() a little more space than necessary, which is OK. Don't use 734 // os::current_stack_pointer(), as its result can be slightly below current 735 // stack pointer, causing us to not alloca enough to reach "bottom". 736 sp = (address)&sp; 737 738 if (sp > bottom) { 739 size = sp - bottom; 740 p = (volatile char *)alloca(size); 741 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 742 p[0] = '\0'; 743 } 744} 745 746bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 747 assert(t!=NULL, "just checking"); 748 assert(t->osthread()->expanding_stack(), "expand should be set"); 749 assert(t->stack_base() != NULL, "stack_base was not initialized"); 750 751 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 752 sigset_t mask_all, old_sigset; 753 sigfillset(&mask_all); 754 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 755 _expand_stack_to(addr); 756 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 757 return true; 758 } 759 return false; 760} 761 762////////////////////////////////////////////////////////////////////////////// 763// create new thread 764 765static address highest_vm_reserved_address(); 766 767// check if it's safe to start a new thread 768static bool _thread_safety_check(Thread* thread) { 769 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 770 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 771 // Heap is mmap'ed at lower end of memory space. Thread stacks are 772 // allocated (MAP_FIXED) from high address space. Every thread stack 773 // occupies a fixed size slot (usually 2Mbytes, but user can change 774 // it to other values if they rebuild LinuxThreads). 775 // 776 // Problem with MAP_FIXED is that mmap() can still succeed even part of 777 // the memory region has already been mmap'ed. That means if we have too 778 // many threads and/or very large heap, eventually thread stack will 779 // collide with heap. 780 // 781 // Here we try to prevent heap/stack collision by comparing current 782 // stack bottom with the highest address that has been mmap'ed by JVM 783 // plus a safety margin for memory maps created by native code. 784 // 785 // This feature can be disabled by setting ThreadSafetyMargin to 0 786 // 787 if (ThreadSafetyMargin > 0) { 788 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 789 790 // not safe if our stack extends below the safety margin 791 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 792 } else { 793 return true; 794 } 795 } else { 796 // Floating stack LinuxThreads or NPTL: 797 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 798 // there's not enough space left, pthread_create() will fail. If we come 799 // here, that means enough space has been reserved for stack. 800 return true; 801 } 802} 803 804// Thread start routine for all newly created threads 805static void *java_start(Thread *thread) { 806 // Try to randomize the cache line index of hot stack frames. 807 // This helps when threads of the same stack traces evict each other's 808 // cache lines. The threads can be either from the same JVM instance, or 809 // from different JVM instances. The benefit is especially true for 810 // processors with hyperthreading technology. 811 static int counter = 0; 812 int pid = os::current_process_id(); 813 alloca(((pid ^ counter++) & 7) * 128); 814 815 ThreadLocalStorage::set_thread(thread); 816 817 OSThread* osthread = thread->osthread(); 818 Monitor* sync = osthread->startThread_lock(); 819 820 // non floating stack LinuxThreads needs extra check, see above 821 if (!_thread_safety_check(thread)) { 822 // notify parent thread 823 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 824 osthread->set_state(ZOMBIE); 825 sync->notify_all(); 826 return NULL; 827 } 828 829 // thread_id is kernel thread id (similar to Solaris LWP id) 830 osthread->set_thread_id(os::Linux::gettid()); 831 832 if (UseNUMA) { 833 int lgrp_id = os::numa_get_group_id(); 834 if (lgrp_id != -1) { 835 thread->set_lgrp_id(lgrp_id); 836 } 837 } 838 // initialize signal mask for this thread 839 os::Linux::hotspot_sigmask(thread); 840 841 // initialize floating point control register 842 os::Linux::init_thread_fpu_state(); 843 844 // handshaking with parent thread 845 { 846 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 847 848 // notify parent thread 849 osthread->set_state(INITIALIZED); 850 sync->notify_all(); 851 852 // wait until os::start_thread() 853 while (osthread->get_state() == INITIALIZED) { 854 sync->wait(Mutex::_no_safepoint_check_flag); 855 } 856 } 857 858 // call one more level start routine 859 thread->run(); 860 861 return 0; 862} 863 864bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 865 assert(thread->osthread() == NULL, "caller responsible"); 866 867 // Allocate the OSThread object 868 OSThread* osthread = new OSThread(NULL, NULL); 869 if (osthread == NULL) { 870 return false; 871 } 872 873 // set the correct thread state 874 osthread->set_thread_type(thr_type); 875 876 // Initial state is ALLOCATED but not INITIALIZED 877 osthread->set_state(ALLOCATED); 878 879 thread->set_osthread(osthread); 880 881 // init thread attributes 882 pthread_attr_t attr; 883 pthread_attr_init(&attr); 884 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 885 886 // stack size 887 if (os::Linux::supports_variable_stack_size()) { 888 // calculate stack size if it's not specified by caller 889 if (stack_size == 0) { 890 stack_size = os::Linux::default_stack_size(thr_type); 891 892 switch (thr_type) { 893 case os::java_thread: 894 // Java threads use ThreadStackSize which default value can be 895 // changed with the flag -Xss 896 assert (JavaThread::stack_size_at_create() > 0, "this should be set"); 897 stack_size = JavaThread::stack_size_at_create(); 898 break; 899 case os::compiler_thread: 900 if (CompilerThreadStackSize > 0) { 901 stack_size = (size_t)(CompilerThreadStackSize * K); 902 break; 903 } // else fall through: 904 // use VMThreadStackSize if CompilerThreadStackSize is not defined 905 case os::vm_thread: 906 case os::pgc_thread: 907 case os::cgc_thread: 908 case os::watcher_thread: 909 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 910 break; 911 } 912 } 913 914 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 915 pthread_attr_setstacksize(&attr, stack_size); 916 } else { 917 // let pthread_create() pick the default value. 918 } 919 920 // glibc guard page 921 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 922 923 ThreadState state; 924 925 { 926 // Serialize thread creation if we are running with fixed stack LinuxThreads 927 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 928 if (lock) { 929 os::Linux::createThread_lock()->lock_without_safepoint_check(); 930 } 931 932 pthread_t tid; 933 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 934 935 pthread_attr_destroy(&attr); 936 937 if (ret != 0) { 938 if (PrintMiscellaneous && (Verbose || WizardMode)) { 939 perror("pthread_create()"); 940 } 941 // Need to clean up stuff we've allocated so far 942 thread->set_osthread(NULL); 943 delete osthread; 944 if (lock) os::Linux::createThread_lock()->unlock(); 945 return false; 946 } 947 948 // Store pthread info into the OSThread 949 osthread->set_pthread_id(tid); 950 951 // Wait until child thread is either initialized or aborted 952 { 953 Monitor* sync_with_child = osthread->startThread_lock(); 954 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 955 while ((state = osthread->get_state()) == ALLOCATED) { 956 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 957 } 958 } 959 960 if (lock) { 961 os::Linux::createThread_lock()->unlock(); 962 } 963 } 964 965 // Aborted due to thread limit being reached 966 if (state == ZOMBIE) { 967 thread->set_osthread(NULL); 968 delete osthread; 969 return false; 970 } 971 972 // The thread is returned suspended (in state INITIALIZED), 973 // and is started higher up in the call chain 974 assert(state == INITIALIZED, "race condition"); 975 return true; 976} 977 978///////////////////////////////////////////////////////////////////////////// 979// attach existing thread 980 981// bootstrap the main thread 982bool os::create_main_thread(JavaThread* thread) { 983 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 984 return create_attached_thread(thread); 985} 986 987bool os::create_attached_thread(JavaThread* thread) { 988#ifdef ASSERT 989 thread->verify_not_published(); 990#endif 991 992 // Allocate the OSThread object 993 OSThread* osthread = new OSThread(NULL, NULL); 994 995 if (osthread == NULL) { 996 return false; 997 } 998 999 // Store pthread info into the OSThread 1000 osthread->set_thread_id(os::Linux::gettid()); 1001 osthread->set_pthread_id(::pthread_self()); 1002 1003 // initialize floating point control register 1004 os::Linux::init_thread_fpu_state(); 1005 1006 // Initial thread state is RUNNABLE 1007 osthread->set_state(RUNNABLE); 1008 1009 thread->set_osthread(osthread); 1010 1011 if (UseNUMA) { 1012 int lgrp_id = os::numa_get_group_id(); 1013 if (lgrp_id != -1) { 1014 thread->set_lgrp_id(lgrp_id); 1015 } 1016 } 1017 1018 if (os::Linux::is_initial_thread()) { 1019 // If current thread is initial thread, its stack is mapped on demand, 1020 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 1021 // the entire stack region to avoid SEGV in stack banging. 1022 // It is also useful to get around the heap-stack-gap problem on SuSE 1023 // kernel (see 4821821 for details). We first expand stack to the top 1024 // of yellow zone, then enable stack yellow zone (order is significant, 1025 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1026 // is no gap between the last two virtual memory regions. 1027 1028 JavaThread *jt = (JavaThread *)thread; 1029 address addr = jt->stack_yellow_zone_base(); 1030 assert(addr != NULL, "initialization problem?"); 1031 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1032 1033 osthread->set_expanding_stack(); 1034 os::Linux::manually_expand_stack(jt, addr); 1035 osthread->clear_expanding_stack(); 1036 } 1037 1038 // initialize signal mask for this thread 1039 // and save the caller's signal mask 1040 os::Linux::hotspot_sigmask(thread); 1041 1042 return true; 1043} 1044 1045void os::pd_start_thread(Thread* thread) { 1046 OSThread * osthread = thread->osthread(); 1047 assert(osthread->get_state() != INITIALIZED, "just checking"); 1048 Monitor* sync_with_child = osthread->startThread_lock(); 1049 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1050 sync_with_child->notify(); 1051} 1052 1053// Free Linux resources related to the OSThread 1054void os::free_thread(OSThread* osthread) { 1055 assert(osthread != NULL, "osthread not set"); 1056 1057 if (Thread::current()->osthread() == osthread) { 1058 // Restore caller's signal mask 1059 sigset_t sigmask = osthread->caller_sigmask(); 1060 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1061 } 1062 1063 delete osthread; 1064} 1065 1066////////////////////////////////////////////////////////////////////////////// 1067// thread local storage 1068 1069int os::allocate_thread_local_storage() { 1070 pthread_key_t key; 1071 int rslt = pthread_key_create(&key, NULL); 1072 assert(rslt == 0, "cannot allocate thread local storage"); 1073 return (int)key; 1074} 1075 1076// Note: This is currently not used by VM, as we don't destroy TLS key 1077// on VM exit. 1078void os::free_thread_local_storage(int index) { 1079 int rslt = pthread_key_delete((pthread_key_t)index); 1080 assert(rslt == 0, "invalid index"); 1081} 1082 1083void os::thread_local_storage_at_put(int index, void* value) { 1084 int rslt = pthread_setspecific((pthread_key_t)index, value); 1085 assert(rslt == 0, "pthread_setspecific failed"); 1086} 1087 1088extern "C" Thread* get_thread() { 1089 return ThreadLocalStorage::thread(); 1090} 1091 1092////////////////////////////////////////////////////////////////////////////// 1093// initial thread 1094 1095// Check if current thread is the initial thread, similar to Solaris thr_main. 1096bool os::Linux::is_initial_thread(void) { 1097 char dummy; 1098 // If called before init complete, thread stack bottom will be null. 1099 // Can be called if fatal error occurs before initialization. 1100 if (initial_thread_stack_bottom() == NULL) return false; 1101 assert(initial_thread_stack_bottom() != NULL && 1102 initial_thread_stack_size() != 0, 1103 "os::init did not locate initial thread's stack region"); 1104 if ((address)&dummy >= initial_thread_stack_bottom() && 1105 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1106 return true; 1107 else return false; 1108} 1109 1110// Find the virtual memory area that contains addr 1111static bool find_vma(address addr, address* vma_low, address* vma_high) { 1112 FILE *fp = fopen("/proc/self/maps", "r"); 1113 if (fp) { 1114 address low, high; 1115 while (!feof(fp)) { 1116 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1117 if (low <= addr && addr < high) { 1118 if (vma_low) *vma_low = low; 1119 if (vma_high) *vma_high = high; 1120 fclose (fp); 1121 return true; 1122 } 1123 } 1124 for (;;) { 1125 int ch = fgetc(fp); 1126 if (ch == EOF || ch == (int)'\n') break; 1127 } 1128 } 1129 fclose(fp); 1130 } 1131 return false; 1132} 1133 1134// Locate initial thread stack. This special handling of initial thread stack 1135// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1136// bogus value for initial thread. 1137void os::Linux::capture_initial_stack(size_t max_size) { 1138 // stack size is the easy part, get it from RLIMIT_STACK 1139 size_t stack_size; 1140 struct rlimit rlim; 1141 getrlimit(RLIMIT_STACK, &rlim); 1142 stack_size = rlim.rlim_cur; 1143 1144 // 6308388: a bug in ld.so will relocate its own .data section to the 1145 // lower end of primordial stack; reduce ulimit -s value a little bit 1146 // so we won't install guard page on ld.so's data section. 1147 stack_size -= 2 * page_size(); 1148 1149 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1150 // 7.1, in both cases we will get 2G in return value. 1151 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1152 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1153 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1154 // in case other parts in glibc still assumes 2M max stack size. 1155 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1156 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1157 if (stack_size > 2 * K * K IA64_ONLY(*2)) 1158 stack_size = 2 * K * K IA64_ONLY(*2); 1159 // Try to figure out where the stack base (top) is. This is harder. 1160 // 1161 // When an application is started, glibc saves the initial stack pointer in 1162 // a global variable "__libc_stack_end", which is then used by system 1163 // libraries. __libc_stack_end should be pretty close to stack top. The 1164 // variable is available since the very early days. However, because it is 1165 // a private interface, it could disappear in the future. 1166 // 1167 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1168 // to __libc_stack_end, it is very close to stack top, but isn't the real 1169 // stack top. Note that /proc may not exist if VM is running as a chroot 1170 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1171 // /proc/<pid>/stat could change in the future (though unlikely). 1172 // 1173 // We try __libc_stack_end first. If that doesn't work, look for 1174 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1175 // as a hint, which should work well in most cases. 1176 1177 uintptr_t stack_start; 1178 1179 // try __libc_stack_end first 1180 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1181 if (p && *p) { 1182 stack_start = *p; 1183 } else { 1184 // see if we can get the start_stack field from /proc/self/stat 1185 FILE *fp; 1186 int pid; 1187 char state; 1188 int ppid; 1189 int pgrp; 1190 int session; 1191 int nr; 1192 int tpgrp; 1193 unsigned long flags; 1194 unsigned long minflt; 1195 unsigned long cminflt; 1196 unsigned long majflt; 1197 unsigned long cmajflt; 1198 unsigned long utime; 1199 unsigned long stime; 1200 long cutime; 1201 long cstime; 1202 long prio; 1203 long nice; 1204 long junk; 1205 long it_real; 1206 uintptr_t start; 1207 uintptr_t vsize; 1208 intptr_t rss; 1209 uintptr_t rsslim; 1210 uintptr_t scodes; 1211 uintptr_t ecode; 1212 int i; 1213 1214 // Figure what the primordial thread stack base is. Code is inspired 1215 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1216 // followed by command name surrounded by parentheses, state, etc. 1217 char stat[2048]; 1218 int statlen; 1219 1220 fp = fopen("/proc/self/stat", "r"); 1221 if (fp) { 1222 statlen = fread(stat, 1, 2047, fp); 1223 stat[statlen] = '\0'; 1224 fclose(fp); 1225 1226 // Skip pid and the command string. Note that we could be dealing with 1227 // weird command names, e.g. user could decide to rename java launcher 1228 // to "java 1.4.2 :)", then the stat file would look like 1229 // 1234 (java 1.4.2 :)) R ... ... 1230 // We don't really need to know the command string, just find the last 1231 // occurrence of ")" and then start parsing from there. See bug 4726580. 1232 char * s = strrchr(stat, ')'); 1233 1234 i = 0; 1235 if (s) { 1236 // Skip blank chars 1237 do s++; while (isspace(*s)); 1238 1239#define _UFM UINTX_FORMAT 1240#define _DFM INTX_FORMAT 1241 1242 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1243 /* 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 */ 1244 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, 1245 &state, /* 3 %c */ 1246 &ppid, /* 4 %d */ 1247 &pgrp, /* 5 %d */ 1248 &session, /* 6 %d */ 1249 &nr, /* 7 %d */ 1250 &tpgrp, /* 8 %d */ 1251 &flags, /* 9 %lu */ 1252 &minflt, /* 10 %lu */ 1253 &cminflt, /* 11 %lu */ 1254 &majflt, /* 12 %lu */ 1255 &cmajflt, /* 13 %lu */ 1256 &utime, /* 14 %lu */ 1257 &stime, /* 15 %lu */ 1258 &cutime, /* 16 %ld */ 1259 &cstime, /* 17 %ld */ 1260 &prio, /* 18 %ld */ 1261 &nice, /* 19 %ld */ 1262 &junk, /* 20 %ld */ 1263 &it_real, /* 21 %ld */ 1264 &start, /* 22 UINTX_FORMAT */ 1265 &vsize, /* 23 UINTX_FORMAT */ 1266 &rss, /* 24 INTX_FORMAT */ 1267 &rsslim, /* 25 UINTX_FORMAT */ 1268 &scodes, /* 26 UINTX_FORMAT */ 1269 &ecode, /* 27 UINTX_FORMAT */ 1270 &stack_start); /* 28 UINTX_FORMAT */ 1271 } 1272 1273#undef _UFM 1274#undef _DFM 1275 1276 if (i != 28 - 2) { 1277 assert(false, "Bad conversion from /proc/self/stat"); 1278 // product mode - assume we are the initial thread, good luck in the 1279 // embedded case. 1280 warning("Can't detect initial thread stack location - bad conversion"); 1281 stack_start = (uintptr_t) &rlim; 1282 } 1283 } else { 1284 // For some reason we can't open /proc/self/stat (for example, running on 1285 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1286 // most cases, so don't abort: 1287 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1288 stack_start = (uintptr_t) &rlim; 1289 } 1290 } 1291 1292 // Now we have a pointer (stack_start) very close to the stack top, the 1293 // next thing to do is to figure out the exact location of stack top. We 1294 // can find out the virtual memory area that contains stack_start by 1295 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1296 // and its upper limit is the real stack top. (again, this would fail if 1297 // running inside chroot, because /proc may not exist.) 1298 1299 uintptr_t stack_top; 1300 address low, high; 1301 if (find_vma((address)stack_start, &low, &high)) { 1302 // success, "high" is the true stack top. (ignore "low", because initial 1303 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1304 stack_top = (uintptr_t)high; 1305 } else { 1306 // failed, likely because /proc/self/maps does not exist 1307 warning("Can't detect initial thread stack location - find_vma failed"); 1308 // best effort: stack_start is normally within a few pages below the real 1309 // stack top, use it as stack top, and reduce stack size so we won't put 1310 // guard page outside stack. 1311 stack_top = stack_start; 1312 stack_size -= 16 * page_size(); 1313 } 1314 1315 // stack_top could be partially down the page so align it 1316 stack_top = align_size_up(stack_top, page_size()); 1317 1318 if (max_size && stack_size > max_size) { 1319 _initial_thread_stack_size = max_size; 1320 } else { 1321 _initial_thread_stack_size = stack_size; 1322 } 1323 1324 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1325 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1326} 1327 1328//////////////////////////////////////////////////////////////////////////////// 1329// time support 1330 1331// Time since start-up in seconds to a fine granularity. 1332// Used by VMSelfDestructTimer and the MemProfiler. 1333double os::elapsedTime() { 1334 1335 return (double)(os::elapsed_counter()) * 0.000001; 1336} 1337 1338jlong os::elapsed_counter() { 1339 timeval time; 1340 int status = gettimeofday(&time, NULL); 1341 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1342} 1343 1344jlong os::elapsed_frequency() { 1345 return (1000 * 1000); 1346} 1347 1348bool os::supports_vtime() { return true; } 1349bool os::enable_vtime() { return false; } 1350bool os::vtime_enabled() { return false; } 1351 1352double os::elapsedVTime() { 1353 struct rusage usage; 1354 int retval = getrusage(RUSAGE_THREAD, &usage); 1355 if (retval == 0) { 1356 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); 1357 } else { 1358 // better than nothing, but not much 1359 return elapsedTime(); 1360 } 1361} 1362 1363jlong os::javaTimeMillis() { 1364 timeval time; 1365 int status = gettimeofday(&time, NULL); 1366 assert(status != -1, "linux error"); 1367 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1368} 1369 1370#ifndef CLOCK_MONOTONIC 1371#define CLOCK_MONOTONIC (1) 1372#endif 1373 1374void os::Linux::clock_init() { 1375 // we do dlopen's in this particular order due to bug in linux 1376 // dynamical loader (see 6348968) leading to crash on exit 1377 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1378 if (handle == NULL) { 1379 handle = dlopen("librt.so", RTLD_LAZY); 1380 } 1381 1382 if (handle) { 1383 int (*clock_getres_func)(clockid_t, struct timespec*) = 1384 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1385 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1386 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1387 if (clock_getres_func && clock_gettime_func) { 1388 // See if monotonic clock is supported by the kernel. Note that some 1389 // early implementations simply return kernel jiffies (updated every 1390 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1391 // for nano time (though the monotonic property is still nice to have). 1392 // It's fixed in newer kernels, however clock_getres() still returns 1393 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1394 // resolution for now. Hopefully as people move to new kernels, this 1395 // won't be a problem. 1396 struct timespec res; 1397 struct timespec tp; 1398 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1399 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1400 // yes, monotonic clock is supported 1401 _clock_gettime = clock_gettime_func; 1402 } else { 1403 // close librt if there is no monotonic clock 1404 dlclose(handle); 1405 } 1406 } 1407 } 1408} 1409 1410#ifndef SYS_clock_getres 1411 1412#if defined(IA32) || defined(AMD64) 1413#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1414#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1415#else 1416#warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1417#define sys_clock_getres(x,y) -1 1418#endif 1419 1420#else 1421#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1422#endif 1423 1424void os::Linux::fast_thread_clock_init() { 1425 if (!UseLinuxPosixThreadCPUClocks) { 1426 return; 1427 } 1428 clockid_t clockid; 1429 struct timespec tp; 1430 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1431 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1432 1433 // Switch to using fast clocks for thread cpu time if 1434 // the sys_clock_getres() returns 0 error code. 1435 // Note, that some kernels may support the current thread 1436 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1437 // returned by the pthread_getcpuclockid(). 1438 // If the fast Posix clocks are supported then the sys_clock_getres() 1439 // must return at least tp.tv_sec == 0 which means a resolution 1440 // better than 1 sec. This is extra check for reliability. 1441 1442 if(pthread_getcpuclockid_func && 1443 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1444 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1445 1446 _supports_fast_thread_cpu_time = true; 1447 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1448 } 1449} 1450 1451jlong os::javaTimeNanos() { 1452 if (Linux::supports_monotonic_clock()) { 1453 struct timespec tp; 1454 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1455 assert(status == 0, "gettime error"); 1456 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1457 return result; 1458 } else { 1459 timeval time; 1460 int status = gettimeofday(&time, NULL); 1461 assert(status != -1, "linux error"); 1462 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1463 return 1000 * usecs; 1464 } 1465} 1466 1467void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1468 if (Linux::supports_monotonic_clock()) { 1469 info_ptr->max_value = ALL_64_BITS; 1470 1471 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1472 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1473 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1474 } else { 1475 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1476 info_ptr->max_value = ALL_64_BITS; 1477 1478 // gettimeofday is a real time clock so it skips 1479 info_ptr->may_skip_backward = true; 1480 info_ptr->may_skip_forward = true; 1481 } 1482 1483 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1484} 1485 1486// Return the real, user, and system times in seconds from an 1487// arbitrary fixed point in the past. 1488bool os::getTimesSecs(double* process_real_time, 1489 double* process_user_time, 1490 double* process_system_time) { 1491 struct tms ticks; 1492 clock_t real_ticks = times(&ticks); 1493 1494 if (real_ticks == (clock_t) (-1)) { 1495 return false; 1496 } else { 1497 double ticks_per_second = (double) clock_tics_per_sec; 1498 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1499 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1500 *process_real_time = ((double) real_ticks) / ticks_per_second; 1501 1502 return true; 1503 } 1504} 1505 1506 1507char * os::local_time_string(char *buf, size_t buflen) { 1508 struct tm t; 1509 time_t long_time; 1510 time(&long_time); 1511 localtime_r(&long_time, &t); 1512 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1513 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1514 t.tm_hour, t.tm_min, t.tm_sec); 1515 return buf; 1516} 1517 1518struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1519 return localtime_r(clock, res); 1520} 1521 1522//////////////////////////////////////////////////////////////////////////////// 1523// runtime exit support 1524 1525// Note: os::shutdown() might be called very early during initialization, or 1526// called from signal handler. Before adding something to os::shutdown(), make 1527// sure it is async-safe and can handle partially initialized VM. 1528void os::shutdown() { 1529 1530 // allow PerfMemory to attempt cleanup of any persistent resources 1531 perfMemory_exit(); 1532 1533 // needs to remove object in file system 1534 AttachListener::abort(); 1535 1536 // flush buffered output, finish log files 1537 ostream_abort(); 1538 1539 // Check for abort hook 1540 abort_hook_t abort_hook = Arguments::abort_hook(); 1541 if (abort_hook != NULL) { 1542 abort_hook(); 1543 } 1544 1545} 1546 1547// Note: os::abort() might be called very early during initialization, or 1548// called from signal handler. Before adding something to os::abort(), make 1549// sure it is async-safe and can handle partially initialized VM. 1550void os::abort(bool dump_core) { 1551 os::shutdown(); 1552 if (dump_core) { 1553#ifndef PRODUCT 1554 fdStream out(defaultStream::output_fd()); 1555 out.print_raw("Current thread is "); 1556 char buf[16]; 1557 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1558 out.print_raw_cr(buf); 1559 out.print_raw_cr("Dumping core ..."); 1560#endif 1561 ::abort(); // dump core 1562 } 1563 1564 ::exit(1); 1565} 1566 1567// Die immediately, no exit hook, no abort hook, no cleanup. 1568void os::die() { 1569 // _exit() on LinuxThreads only kills current thread 1570 ::abort(); 1571} 1572 1573// unused on linux for now. 1574void os::set_error_file(const char *logfile) {} 1575 1576 1577// This method is a copy of JDK's sysGetLastErrorString 1578// from src/solaris/hpi/src/system_md.c 1579 1580size_t os::lasterror(char *buf, size_t len) { 1581 1582 if (errno == 0) return 0; 1583 1584 const char *s = ::strerror(errno); 1585 size_t n = ::strlen(s); 1586 if (n >= len) { 1587 n = len - 1; 1588 } 1589 ::strncpy(buf, s, n); 1590 buf[n] = '\0'; 1591 return n; 1592} 1593 1594intx os::current_thread_id() { return (intx)pthread_self(); } 1595int os::current_process_id() { 1596 1597 // Under the old linux thread library, linux gives each thread 1598 // its own process id. Because of this each thread will return 1599 // a different pid if this method were to return the result 1600 // of getpid(2). Linux provides no api that returns the pid 1601 // of the launcher thread for the vm. This implementation 1602 // returns a unique pid, the pid of the launcher thread 1603 // that starts the vm 'process'. 1604 1605 // Under the NPTL, getpid() returns the same pid as the 1606 // launcher thread rather than a unique pid per thread. 1607 // Use gettid() if you want the old pre NPTL behaviour. 1608 1609 // if you are looking for the result of a call to getpid() that 1610 // returns a unique pid for the calling thread, then look at the 1611 // OSThread::thread_id() method in osThread_linux.hpp file 1612 1613 return (int)(_initial_pid ? _initial_pid : getpid()); 1614} 1615 1616// DLL functions 1617 1618const char* os::dll_file_extension() { return ".so"; } 1619 1620// This must be hard coded because it's the system's temporary 1621// directory not the java application's temp directory, ala java.io.tmpdir. 1622const char* os::get_temp_directory() { return "/tmp"; } 1623 1624static bool file_exists(const char* filename) { 1625 struct stat statbuf; 1626 if (filename == NULL || strlen(filename) == 0) { 1627 return false; 1628 } 1629 return os::stat(filename, &statbuf) == 0; 1630} 1631 1632bool os::dll_build_name(char* buffer, size_t buflen, 1633 const char* pname, const char* fname) { 1634 bool retval = false; 1635 // Copied from libhpi 1636 const size_t pnamelen = pname ? strlen(pname) : 0; 1637 1638 // Return error on buffer overflow. 1639 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1640 return retval; 1641 } 1642 1643 if (pnamelen == 0) { 1644 snprintf(buffer, buflen, "lib%s.so", fname); 1645 retval = true; 1646 } else if (strchr(pname, *os::path_separator()) != NULL) { 1647 int n; 1648 char** pelements = split_path(pname, &n); 1649 if (pelements == NULL) { 1650 return false; 1651 } 1652 for (int i = 0 ; i < n ; i++) { 1653 // Really shouldn't be NULL, but check can't hurt 1654 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1655 continue; // skip the empty path values 1656 } 1657 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1658 if (file_exists(buffer)) { 1659 retval = true; 1660 break; 1661 } 1662 } 1663 // release the storage 1664 for (int i = 0 ; i < n ; i++) { 1665 if (pelements[i] != NULL) { 1666 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1667 } 1668 } 1669 if (pelements != NULL) { 1670 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1671 } 1672 } else { 1673 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1674 retval = true; 1675 } 1676 return retval; 1677} 1678 1679// check if addr is inside libjvm.so 1680bool os::address_is_in_vm(address addr) { 1681 static address libjvm_base_addr; 1682 Dl_info dlinfo; 1683 1684 if (libjvm_base_addr == NULL) { 1685 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1686 libjvm_base_addr = (address)dlinfo.dli_fbase; 1687 } 1688 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1689 } 1690 1691 if (dladdr((void *)addr, &dlinfo) != 0) { 1692 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1693 } 1694 1695 return false; 1696} 1697 1698bool os::dll_address_to_function_name(address addr, char *buf, 1699 int buflen, int *offset) { 1700 // buf is not optional, but offset is optional 1701 assert(buf != NULL, "sanity check"); 1702 1703 Dl_info dlinfo; 1704 1705 if (dladdr((void*)addr, &dlinfo) != 0) { 1706 // see if we have a matching symbol 1707 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1708 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1709 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1710 } 1711 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1712 return true; 1713 } 1714 // no matching symbol so try for just file info 1715 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1716 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1717 buf, buflen, offset, dlinfo.dli_fname)) { 1718 return true; 1719 } 1720 } 1721 } 1722 1723 buf[0] = '\0'; 1724 if (offset != NULL) *offset = -1; 1725 return false; 1726} 1727 1728struct _address_to_library_name { 1729 address addr; // input : memory address 1730 size_t buflen; // size of fname 1731 char* fname; // output: library name 1732 address base; // library base addr 1733}; 1734 1735static int address_to_library_name_callback(struct dl_phdr_info *info, 1736 size_t size, void *data) { 1737 int i; 1738 bool found = false; 1739 address libbase = NULL; 1740 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1741 1742 // iterate through all loadable segments 1743 for (i = 0; i < info->dlpi_phnum; i++) { 1744 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1745 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1746 // base address of a library is the lowest address of its loaded 1747 // segments. 1748 if (libbase == NULL || libbase > segbase) { 1749 libbase = segbase; 1750 } 1751 // see if 'addr' is within current segment 1752 if (segbase <= d->addr && 1753 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1754 found = true; 1755 } 1756 } 1757 } 1758 1759 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1760 // so dll_address_to_library_name() can fall through to use dladdr() which 1761 // can figure out executable name from argv[0]. 1762 if (found && info->dlpi_name && info->dlpi_name[0]) { 1763 d->base = libbase; 1764 if (d->fname) { 1765 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1766 } 1767 return 1; 1768 } 1769 return 0; 1770} 1771 1772bool os::dll_address_to_library_name(address addr, char* buf, 1773 int buflen, int* offset) { 1774 // buf is not optional, but offset is optional 1775 assert(buf != NULL, "sanity check"); 1776 1777 Dl_info dlinfo; 1778 struct _address_to_library_name data; 1779 1780 // There is a bug in old glibc dladdr() implementation that it could resolve 1781 // to wrong library name if the .so file has a base address != NULL. Here 1782 // we iterate through the program headers of all loaded libraries to find 1783 // out which library 'addr' really belongs to. This workaround can be 1784 // removed once the minimum requirement for glibc is moved to 2.3.x. 1785 data.addr = addr; 1786 data.fname = buf; 1787 data.buflen = buflen; 1788 data.base = NULL; 1789 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1790 1791 if (rslt) { 1792 // buf already contains library name 1793 if (offset) *offset = addr - data.base; 1794 return true; 1795 } 1796 if (dladdr((void*)addr, &dlinfo) != 0) { 1797 if (dlinfo.dli_fname != NULL) { 1798 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1799 } 1800 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1801 *offset = addr - (address)dlinfo.dli_fbase; 1802 } 1803 return true; 1804 } 1805 1806 buf[0] = '\0'; 1807 if (offset) *offset = -1; 1808 return false; 1809} 1810 1811 // Loads .dll/.so and 1812 // in case of error it checks if .dll/.so was built for the 1813 // same architecture as Hotspot is running on 1814 1815 1816// Remember the stack's state. The Linux dynamic linker will change 1817// the stack to 'executable' at most once, so we must safepoint only once. 1818bool os::Linux::_stack_is_executable = false; 1819 1820// VM operation that loads a library. This is necessary if stack protection 1821// of the Java stacks can be lost during loading the library. If we 1822// do not stop the Java threads, they can stack overflow before the stacks 1823// are protected again. 1824class VM_LinuxDllLoad: public VM_Operation { 1825 private: 1826 const char *_filename; 1827 char *_ebuf; 1828 int _ebuflen; 1829 void *_lib; 1830 public: 1831 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1832 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1833 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1834 void doit() { 1835 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1836 os::Linux::_stack_is_executable = true; 1837 } 1838 void* loaded_library() { return _lib; } 1839}; 1840 1841void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1842{ 1843 void * result = NULL; 1844 bool load_attempted = false; 1845 1846 // Check whether the library to load might change execution rights 1847 // of the stack. If they are changed, the protection of the stack 1848 // guard pages will be lost. We need a safepoint to fix this. 1849 // 1850 // See Linux man page execstack(8) for more info. 1851 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1852 ElfFile ef(filename); 1853 if (!ef.specifies_noexecstack()) { 1854 if (!is_init_completed()) { 1855 os::Linux::_stack_is_executable = true; 1856 // This is OK - No Java threads have been created yet, and hence no 1857 // stack guard pages to fix. 1858 // 1859 // This should happen only when you are building JDK7 using a very 1860 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1861 // 1862 // Dynamic loader will make all stacks executable after 1863 // this function returns, and will not do that again. 1864 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1865 } else { 1866 warning("You have loaded library %s which might have disabled stack guard. " 1867 "The VM will try to fix the stack guard now.\n" 1868 "It's highly recommended that you fix the library with " 1869 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1870 filename); 1871 1872 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1873 JavaThread *jt = JavaThread::current(); 1874 if (jt->thread_state() != _thread_in_native) { 1875 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1876 // that requires ExecStack. Cannot enter safe point. Let's give up. 1877 warning("Unable to fix stack guard. Giving up."); 1878 } else { 1879 if (!LoadExecStackDllInVMThread) { 1880 // This is for the case where the DLL has an static 1881 // constructor function that executes JNI code. We cannot 1882 // load such DLLs in the VMThread. 1883 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1884 } 1885 1886 ThreadInVMfromNative tiv(jt); 1887 debug_only(VMNativeEntryWrapper vew;) 1888 1889 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1890 VMThread::execute(&op); 1891 if (LoadExecStackDllInVMThread) { 1892 result = op.loaded_library(); 1893 } 1894 load_attempted = true; 1895 } 1896 } 1897 } 1898 } 1899 1900 if (!load_attempted) { 1901 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1902 } 1903 1904 if (result != NULL) { 1905 // Successful loading 1906 return result; 1907 } 1908 1909 Elf32_Ehdr elf_head; 1910 int diag_msg_max_length=ebuflen-strlen(ebuf); 1911 char* diag_msg_buf=ebuf+strlen(ebuf); 1912 1913 if (diag_msg_max_length==0) { 1914 // No more space in ebuf for additional diagnostics message 1915 return NULL; 1916 } 1917 1918 1919 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1920 1921 if (file_descriptor < 0) { 1922 // Can't open library, report dlerror() message 1923 return NULL; 1924 } 1925 1926 bool failed_to_read_elf_head= 1927 (sizeof(elf_head)!= 1928 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1929 1930 ::close(file_descriptor); 1931 if (failed_to_read_elf_head) { 1932 // file i/o error - report dlerror() msg 1933 return NULL; 1934 } 1935 1936 typedef struct { 1937 Elf32_Half code; // Actual value as defined in elf.h 1938 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1939 char elf_class; // 32 or 64 bit 1940 char endianess; // MSB or LSB 1941 char* name; // String representation 1942 } arch_t; 1943 1944 #ifndef EM_486 1945 #define EM_486 6 /* Intel 80486 */ 1946 #endif 1947 1948 static const arch_t arch_array[]={ 1949 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1950 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1951 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1952 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1953 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1954 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1955 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1956 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1957 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1958 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1959 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1960 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1961 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1962 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1963 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1964 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1965 }; 1966 1967 #if (defined IA32) 1968 static Elf32_Half running_arch_code=EM_386; 1969 #elif (defined AMD64) 1970 static Elf32_Half running_arch_code=EM_X86_64; 1971 #elif (defined IA64) 1972 static Elf32_Half running_arch_code=EM_IA_64; 1973 #elif (defined __sparc) && (defined _LP64) 1974 static Elf32_Half running_arch_code=EM_SPARCV9; 1975 #elif (defined __sparc) && (!defined _LP64) 1976 static Elf32_Half running_arch_code=EM_SPARC; 1977 #elif (defined __powerpc64__) 1978 static Elf32_Half running_arch_code=EM_PPC64; 1979 #elif (defined __powerpc__) 1980 static Elf32_Half running_arch_code=EM_PPC; 1981 #elif (defined ARM) 1982 static Elf32_Half running_arch_code=EM_ARM; 1983 #elif (defined S390) 1984 static Elf32_Half running_arch_code=EM_S390; 1985 #elif (defined ALPHA) 1986 static Elf32_Half running_arch_code=EM_ALPHA; 1987 #elif (defined MIPSEL) 1988 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1989 #elif (defined PARISC) 1990 static Elf32_Half running_arch_code=EM_PARISC; 1991 #elif (defined MIPS) 1992 static Elf32_Half running_arch_code=EM_MIPS; 1993 #elif (defined M68K) 1994 static Elf32_Half running_arch_code=EM_68K; 1995 #else 1996 #error Method os::dll_load requires that one of following is defined:\ 1997 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 1998 #endif 1999 2000 // Identify compatability class for VM's architecture and library's architecture 2001 // Obtain string descriptions for architectures 2002 2003 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 2004 int running_arch_index=-1; 2005 2006 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 2007 if (running_arch_code == arch_array[i].code) { 2008 running_arch_index = i; 2009 } 2010 if (lib_arch.code == arch_array[i].code) { 2011 lib_arch.compat_class = arch_array[i].compat_class; 2012 lib_arch.name = arch_array[i].name; 2013 } 2014 } 2015 2016 assert(running_arch_index != -1, 2017 "Didn't find running architecture code (running_arch_code) in arch_array"); 2018 if (running_arch_index == -1) { 2019 // Even though running architecture detection failed 2020 // we may still continue with reporting dlerror() message 2021 return NULL; 2022 } 2023 2024 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 2025 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 2026 return NULL; 2027 } 2028 2029#ifndef S390 2030 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 2031 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 2032 return NULL; 2033 } 2034#endif // !S390 2035 2036 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2037 if ( lib_arch.name!=NULL ) { 2038 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2039 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2040 lib_arch.name, arch_array[running_arch_index].name); 2041 } else { 2042 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2043 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2044 lib_arch.code, 2045 arch_array[running_arch_index].name); 2046 } 2047 } 2048 2049 return NULL; 2050} 2051 2052void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) { 2053 void * result = ::dlopen(filename, RTLD_LAZY); 2054 if (result == NULL) { 2055 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2056 ebuf[ebuflen-1] = '\0'; 2057 } 2058 return result; 2059} 2060 2061void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) { 2062 void * result = NULL; 2063 if (LoadExecStackDllInVMThread) { 2064 result = dlopen_helper(filename, ebuf, ebuflen); 2065 } 2066 2067 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2068 // library that requires an executable stack, or which does not have this 2069 // stack attribute set, dlopen changes the stack attribute to executable. The 2070 // read protection of the guard pages gets lost. 2071 // 2072 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2073 // may have been queued at the same time. 2074 2075 if (!_stack_is_executable) { 2076 JavaThread *jt = Threads::first(); 2077 2078 while (jt) { 2079 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2080 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2081 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2082 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2083 warning("Attempt to reguard stack yellow zone failed."); 2084 } 2085 } 2086 jt = jt->next(); 2087 } 2088 } 2089 2090 return result; 2091} 2092 2093/* 2094 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2095 * chances are you might want to run the generated bits against glibc-2.0 2096 * libdl.so, so always use locking for any version of glibc. 2097 */ 2098void* os::dll_lookup(void* handle, const char* name) { 2099 pthread_mutex_lock(&dl_mutex); 2100 void* res = dlsym(handle, name); 2101 pthread_mutex_unlock(&dl_mutex); 2102 return res; 2103} 2104 2105 2106static bool _print_ascii_file(const char* filename, outputStream* st) { 2107 int fd = ::open(filename, O_RDONLY); 2108 if (fd == -1) { 2109 return false; 2110 } 2111 2112 char buf[32]; 2113 int bytes; 2114 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2115 st->print_raw(buf, bytes); 2116 } 2117 2118 ::close(fd); 2119 2120 return true; 2121} 2122 2123void os::print_dll_info(outputStream *st) { 2124 st->print_cr("Dynamic libraries:"); 2125 2126 char fname[32]; 2127 pid_t pid = os::Linux::gettid(); 2128 2129 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2130 2131 if (!_print_ascii_file(fname, st)) { 2132 st->print("Can not get library information for pid = %d\n", pid); 2133 } 2134} 2135 2136void os::print_os_info_brief(outputStream* st) { 2137 os::Linux::print_distro_info(st); 2138 2139 os::Posix::print_uname_info(st); 2140 2141 os::Linux::print_libversion_info(st); 2142 2143} 2144 2145void os::print_os_info(outputStream* st) { 2146 st->print("OS:"); 2147 2148 os::Linux::print_distro_info(st); 2149 2150 os::Posix::print_uname_info(st); 2151 2152 // Print warning if unsafe chroot environment detected 2153 if (unsafe_chroot_detected) { 2154 st->print("WARNING!! "); 2155 st->print_cr(unstable_chroot_error); 2156 } 2157 2158 os::Linux::print_libversion_info(st); 2159 2160 os::Posix::print_rlimit_info(st); 2161 2162 os::Posix::print_load_average(st); 2163 2164 os::Linux::print_full_memory_info(st); 2165} 2166 2167// Try to identify popular distros. 2168// Most Linux distributions have /etc/XXX-release file, which contains 2169// the OS version string. Some have more than one /etc/XXX-release file 2170// (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 2171// so the order is important. 2172void os::Linux::print_distro_info(outputStream* st) { 2173 if (!_print_ascii_file("/etc/mandrake-release", st) && 2174 !_print_ascii_file("/etc/sun-release", st) && 2175 !_print_ascii_file("/etc/redhat-release", st) && 2176 !_print_ascii_file("/etc/SuSE-release", st) && 2177 !_print_ascii_file("/etc/turbolinux-release", st) && 2178 !_print_ascii_file("/etc/gentoo-release", st) && 2179 !_print_ascii_file("/etc/debian_version", st) && 2180 !_print_ascii_file("/etc/ltib-release", st) && 2181 !_print_ascii_file("/etc/angstrom-version", st)) { 2182 st->print("Linux"); 2183 } 2184 st->cr(); 2185} 2186 2187void os::Linux::print_libversion_info(outputStream* st) { 2188 // libc, pthread 2189 st->print("libc:"); 2190 st->print(os::Linux::glibc_version()); st->print(" "); 2191 st->print(os::Linux::libpthread_version()); st->print(" "); 2192 if (os::Linux::is_LinuxThreads()) { 2193 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2194 } 2195 st->cr(); 2196} 2197 2198void os::Linux::print_full_memory_info(outputStream* st) { 2199 st->print("\n/proc/meminfo:\n"); 2200 _print_ascii_file("/proc/meminfo", st); 2201 st->cr(); 2202} 2203 2204void os::print_memory_info(outputStream* st) { 2205 2206 st->print("Memory:"); 2207 st->print(" %dk page", os::vm_page_size()>>10); 2208 2209 // values in struct sysinfo are "unsigned long" 2210 struct sysinfo si; 2211 sysinfo(&si); 2212 2213 st->print(", physical " UINT64_FORMAT "k", 2214 os::physical_memory() >> 10); 2215 st->print("(" UINT64_FORMAT "k free)", 2216 os::available_memory() >> 10); 2217 st->print(", swap " UINT64_FORMAT "k", 2218 ((jlong)si.totalswap * si.mem_unit) >> 10); 2219 st->print("(" UINT64_FORMAT "k free)", 2220 ((jlong)si.freeswap * si.mem_unit) >> 10); 2221 st->cr(); 2222} 2223 2224void os::pd_print_cpu_info(outputStream* st) { 2225 st->print("\n/proc/cpuinfo:\n"); 2226 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2227 st->print(" <Not Available>"); 2228 } 2229 st->cr(); 2230} 2231 2232// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 2233// but they're the same for all the linux arch that we support 2234// and they're the same for solaris but there's no common place to put this. 2235const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 2236 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 2237 "ILL_COPROC", "ILL_BADSTK" }; 2238 2239const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 2240 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 2241 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 2242 2243const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 2244 2245const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 2246 2247void os::print_siginfo(outputStream* st, void* siginfo) { 2248 st->print("siginfo:"); 2249 2250 const int buflen = 100; 2251 char buf[buflen]; 2252 siginfo_t *si = (siginfo_t*)siginfo; 2253 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 2254 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 2255 st->print("si_errno=%s", buf); 2256 } else { 2257 st->print("si_errno=%d", si->si_errno); 2258 } 2259 const int c = si->si_code; 2260 assert(c > 0, "unexpected si_code"); 2261 switch (si->si_signo) { 2262 case SIGILL: 2263 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 2264 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2265 break; 2266 case SIGFPE: 2267 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 2268 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2269 break; 2270 case SIGSEGV: 2271 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 2272 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2273 break; 2274 case SIGBUS: 2275 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 2276 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2277 break; 2278 default: 2279 st->print(", si_code=%d", si->si_code); 2280 // no si_addr 2281 } 2282 2283 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2284 UseSharedSpaces) { 2285 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2286 if (mapinfo->is_in_shared_space(si->si_addr)) { 2287 st->print("\n\nError accessing class data sharing archive." \ 2288 " Mapped file inaccessible during execution, " \ 2289 " possible disk/network problem."); 2290 } 2291 } 2292 st->cr(); 2293} 2294 2295 2296static void print_signal_handler(outputStream* st, int sig, 2297 char* buf, size_t buflen); 2298 2299void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2300 st->print_cr("Signal Handlers:"); 2301 print_signal_handler(st, SIGSEGV, buf, buflen); 2302 print_signal_handler(st, SIGBUS , buf, buflen); 2303 print_signal_handler(st, SIGFPE , buf, buflen); 2304 print_signal_handler(st, SIGPIPE, buf, buflen); 2305 print_signal_handler(st, SIGXFSZ, buf, buflen); 2306 print_signal_handler(st, SIGILL , buf, buflen); 2307 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2308 print_signal_handler(st, SR_signum, buf, buflen); 2309 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2310 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2311 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2312 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2313} 2314 2315static char saved_jvm_path[MAXPATHLEN] = {0}; 2316 2317// Find the full path to the current module, libjvm.so 2318void os::jvm_path(char *buf, jint buflen) { 2319 // Error checking. 2320 if (buflen < MAXPATHLEN) { 2321 assert(false, "must use a large-enough buffer"); 2322 buf[0] = '\0'; 2323 return; 2324 } 2325 // Lazy resolve the path to current module. 2326 if (saved_jvm_path[0] != 0) { 2327 strcpy(buf, saved_jvm_path); 2328 return; 2329 } 2330 2331 char dli_fname[MAXPATHLEN]; 2332 bool ret = dll_address_to_library_name( 2333 CAST_FROM_FN_PTR(address, os::jvm_path), 2334 dli_fname, sizeof(dli_fname), NULL); 2335 assert(ret, "cannot locate libjvm"); 2336 char *rp = NULL; 2337 if (ret && dli_fname[0] != '\0') { 2338 rp = realpath(dli_fname, buf); 2339 } 2340 if (rp == NULL) 2341 return; 2342 2343 if (Arguments::created_by_gamma_launcher()) { 2344 // Support for the gamma launcher. Typical value for buf is 2345 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 2346 // the right place in the string, then assume we are installed in a JDK and 2347 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2348 // up the path so it looks like libjvm.so is installed there (append a 2349 // fake suffix hotspot/libjvm.so). 2350 const char *p = buf + strlen(buf) - 1; 2351 for (int count = 0; p > buf && count < 5; ++count) { 2352 for (--p; p > buf && *p != '/'; --p) 2353 /* empty */ ; 2354 } 2355 2356 if (strncmp(p, "/jre/lib/", 9) != 0) { 2357 // Look for JAVA_HOME in the environment. 2358 char* java_home_var = ::getenv("JAVA_HOME"); 2359 if (java_home_var != NULL && java_home_var[0] != 0) { 2360 char* jrelib_p; 2361 int len; 2362 2363 // Check the current module name "libjvm.so". 2364 p = strrchr(buf, '/'); 2365 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2366 2367 rp = realpath(java_home_var, buf); 2368 if (rp == NULL) 2369 return; 2370 2371 // determine if this is a legacy image or modules image 2372 // modules image doesn't have "jre" subdirectory 2373 len = strlen(buf); 2374 jrelib_p = buf + len; 2375 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2376 if (0 != access(buf, F_OK)) { 2377 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2378 } 2379 2380 if (0 == access(buf, F_OK)) { 2381 // Use current module name "libjvm.so" 2382 len = strlen(buf); 2383 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2384 } else { 2385 // Go back to path of .so 2386 rp = realpath(dli_fname, buf); 2387 if (rp == NULL) 2388 return; 2389 } 2390 } 2391 } 2392 } 2393 2394 strcpy(saved_jvm_path, buf); 2395} 2396 2397void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2398 // no prefix required, not even "_" 2399} 2400 2401void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2402 // no suffix required 2403} 2404 2405//////////////////////////////////////////////////////////////////////////////// 2406// sun.misc.Signal support 2407 2408static volatile jint sigint_count = 0; 2409 2410static void 2411UserHandler(int sig, void *siginfo, void *context) { 2412 // 4511530 - sem_post is serialized and handled by the manager thread. When 2413 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2414 // don't want to flood the manager thread with sem_post requests. 2415 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2416 return; 2417 2418 // Ctrl-C is pressed during error reporting, likely because the error 2419 // handler fails to abort. Let VM die immediately. 2420 if (sig == SIGINT && is_error_reported()) { 2421 os::die(); 2422 } 2423 2424 os::signal_notify(sig); 2425} 2426 2427void* os::user_handler() { 2428 return CAST_FROM_FN_PTR(void*, UserHandler); 2429} 2430 2431class Semaphore : public StackObj { 2432 public: 2433 Semaphore(); 2434 ~Semaphore(); 2435 void signal(); 2436 void wait(); 2437 bool trywait(); 2438 bool timedwait(unsigned int sec, int nsec); 2439 private: 2440 sem_t _semaphore; 2441}; 2442 2443 2444Semaphore::Semaphore() { 2445 sem_init(&_semaphore, 0, 0); 2446} 2447 2448Semaphore::~Semaphore() { 2449 sem_destroy(&_semaphore); 2450} 2451 2452void Semaphore::signal() { 2453 sem_post(&_semaphore); 2454} 2455 2456void Semaphore::wait() { 2457 sem_wait(&_semaphore); 2458} 2459 2460bool Semaphore::trywait() { 2461 return sem_trywait(&_semaphore) == 0; 2462} 2463 2464bool Semaphore::timedwait(unsigned int sec, int nsec) { 2465 struct timespec ts; 2466 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec); 2467 2468 while (1) { 2469 int result = sem_timedwait(&_semaphore, &ts); 2470 if (result == 0) { 2471 return true; 2472 } else if (errno == EINTR) { 2473 continue; 2474 } else if (errno == ETIMEDOUT) { 2475 return false; 2476 } else { 2477 return false; 2478 } 2479 } 2480} 2481 2482extern "C" { 2483 typedef void (*sa_handler_t)(int); 2484 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2485} 2486 2487void* os::signal(int signal_number, void* handler) { 2488 struct sigaction sigAct, oldSigAct; 2489 2490 sigfillset(&(sigAct.sa_mask)); 2491 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2492 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2493 2494 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2495 // -1 means registration failed 2496 return (void *)-1; 2497 } 2498 2499 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2500} 2501 2502void os::signal_raise(int signal_number) { 2503 ::raise(signal_number); 2504} 2505 2506/* 2507 * The following code is moved from os.cpp for making this 2508 * code platform specific, which it is by its very nature. 2509 */ 2510 2511// Will be modified when max signal is changed to be dynamic 2512int os::sigexitnum_pd() { 2513 return NSIG; 2514} 2515 2516// a counter for each possible signal value 2517static volatile jint pending_signals[NSIG+1] = { 0 }; 2518 2519// Linux(POSIX) specific hand shaking semaphore. 2520static sem_t sig_sem; 2521static Semaphore sr_semaphore; 2522 2523void os::signal_init_pd() { 2524 // Initialize signal structures 2525 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2526 2527 // Initialize signal semaphore 2528 ::sem_init(&sig_sem, 0, 0); 2529} 2530 2531void os::signal_notify(int sig) { 2532 Atomic::inc(&pending_signals[sig]); 2533 ::sem_post(&sig_sem); 2534} 2535 2536static int check_pending_signals(bool wait) { 2537 Atomic::store(0, &sigint_count); 2538 for (;;) { 2539 for (int i = 0; i < NSIG + 1; i++) { 2540 jint n = pending_signals[i]; 2541 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2542 return i; 2543 } 2544 } 2545 if (!wait) { 2546 return -1; 2547 } 2548 JavaThread *thread = JavaThread::current(); 2549 ThreadBlockInVM tbivm(thread); 2550 2551 bool threadIsSuspended; 2552 do { 2553 thread->set_suspend_equivalent(); 2554 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2555 ::sem_wait(&sig_sem); 2556 2557 // were we externally suspended while we were waiting? 2558 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2559 if (threadIsSuspended) { 2560 // 2561 // The semaphore has been incremented, but while we were waiting 2562 // another thread suspended us. We don't want to continue running 2563 // while suspended because that would surprise the thread that 2564 // suspended us. 2565 // 2566 ::sem_post(&sig_sem); 2567 2568 thread->java_suspend_self(); 2569 } 2570 } while (threadIsSuspended); 2571 } 2572} 2573 2574int os::signal_lookup() { 2575 return check_pending_signals(false); 2576} 2577 2578int os::signal_wait() { 2579 return check_pending_signals(true); 2580} 2581 2582//////////////////////////////////////////////////////////////////////////////// 2583// Virtual Memory 2584 2585int os::vm_page_size() { 2586 // Seems redundant as all get out 2587 assert(os::Linux::page_size() != -1, "must call os::init"); 2588 return os::Linux::page_size(); 2589} 2590 2591// Solaris allocates memory by pages. 2592int os::vm_allocation_granularity() { 2593 assert(os::Linux::page_size() != -1, "must call os::init"); 2594 return os::Linux::page_size(); 2595} 2596 2597// Rationale behind this function: 2598// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2599// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2600// samples for JITted code. Here we create private executable mapping over the code cache 2601// and then we can use standard (well, almost, as mapping can change) way to provide 2602// info for the reporting script by storing timestamp and location of symbol 2603void linux_wrap_code(char* base, size_t size) { 2604 static volatile jint cnt = 0; 2605 2606 if (!UseOprofile) { 2607 return; 2608 } 2609 2610 char buf[PATH_MAX+1]; 2611 int num = Atomic::add(1, &cnt); 2612 2613 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2614 os::get_temp_directory(), os::current_process_id(), num); 2615 unlink(buf); 2616 2617 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2618 2619 if (fd != -1) { 2620 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2621 if (rv != (off_t)-1) { 2622 if (::write(fd, "", 1) == 1) { 2623 mmap(base, size, 2624 PROT_READ|PROT_WRITE|PROT_EXEC, 2625 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2626 } 2627 } 2628 ::close(fd); 2629 unlink(buf); 2630 } 2631} 2632 2633static bool recoverable_mmap_error(int err) { 2634 // See if the error is one we can let the caller handle. This 2635 // list of errno values comes from JBS-6843484. I can't find a 2636 // Linux man page that documents this specific set of errno 2637 // values so while this list currently matches Solaris, it may 2638 // change as we gain experience with this failure mode. 2639 switch (err) { 2640 case EBADF: 2641 case EINVAL: 2642 case ENOTSUP: 2643 // let the caller deal with these errors 2644 return true; 2645 2646 default: 2647 // Any remaining errors on this OS can cause our reserved mapping 2648 // to be lost. That can cause confusion where different data 2649 // structures think they have the same memory mapped. The worst 2650 // scenario is if both the VM and a library think they have the 2651 // same memory mapped. 2652 return false; 2653 } 2654} 2655 2656static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2657 int err) { 2658 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2659 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2660 strerror(err), err); 2661} 2662 2663static void warn_fail_commit_memory(char* addr, size_t size, 2664 size_t alignment_hint, bool exec, 2665 int err) { 2666 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2667 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2668 alignment_hint, exec, strerror(err), err); 2669} 2670 2671// NOTE: Linux kernel does not really reserve the pages for us. 2672// All it does is to check if there are enough free pages 2673// left at the time of mmap(). This could be a potential 2674// problem. 2675int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2676 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2677 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2678 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2679 if (res != (uintptr_t) MAP_FAILED) { 2680 if (UseNUMAInterleaving) { 2681 numa_make_global(addr, size); 2682 } 2683 return 0; 2684 } 2685 2686 int err = errno; // save errno from mmap() call above 2687 2688 if (!recoverable_mmap_error(err)) { 2689 warn_fail_commit_memory(addr, size, exec, err); 2690 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2691 } 2692 2693 return err; 2694} 2695 2696bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2697 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2698} 2699 2700void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2701 const char* mesg) { 2702 assert(mesg != NULL, "mesg must be specified"); 2703 int err = os::Linux::commit_memory_impl(addr, size, exec); 2704 if (err != 0) { 2705 // the caller wants all commit errors to exit with the specified mesg: 2706 warn_fail_commit_memory(addr, size, exec, err); 2707 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2708 } 2709} 2710 2711// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2712#ifndef MAP_HUGETLB 2713#define MAP_HUGETLB 0x40000 2714#endif 2715 2716// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2717#ifndef MADV_HUGEPAGE 2718#define MADV_HUGEPAGE 14 2719#endif 2720 2721int os::Linux::commit_memory_impl(char* addr, size_t size, 2722 size_t alignment_hint, bool exec) { 2723 int err = os::Linux::commit_memory_impl(addr, size, exec); 2724 if (err == 0) { 2725 realign_memory(addr, size, alignment_hint); 2726 } 2727 return err; 2728} 2729 2730bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2731 bool exec) { 2732 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2733} 2734 2735void os::pd_commit_memory_or_exit(char* addr, size_t size, 2736 size_t alignment_hint, bool exec, 2737 const char* mesg) { 2738 assert(mesg != NULL, "mesg must be specified"); 2739 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2740 if (err != 0) { 2741 // the caller wants all commit errors to exit with the specified mesg: 2742 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2743 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2744 } 2745} 2746 2747void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2748 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2749 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2750 // be supported or the memory may already be backed by huge pages. 2751 ::madvise(addr, bytes, MADV_HUGEPAGE); 2752 } 2753} 2754 2755void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2756 // This method works by doing an mmap over an existing mmaping and effectively discarding 2757 // the existing pages. However it won't work for SHM-based large pages that cannot be 2758 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2759 // small pages on top of the SHM segment. This method always works for small pages, so we 2760 // allow that in any case. 2761 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2762 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2763 } 2764} 2765 2766void os::numa_make_global(char *addr, size_t bytes) { 2767 Linux::numa_interleave_memory(addr, bytes); 2768} 2769 2770// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2771// bind policy to MPOL_PREFERRED for the current thread. 2772#define USE_MPOL_PREFERRED 0 2773 2774void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2775 // To make NUMA and large pages more robust when both enabled, we need to ease 2776 // the requirements on where the memory should be allocated. MPOL_BIND is the 2777 // default policy and it will force memory to be allocated on the specified 2778 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2779 // the specified node, but will not force it. Using this policy will prevent 2780 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2781 // free large pages. 2782 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2783 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2784} 2785 2786bool os::numa_topology_changed() { return false; } 2787 2788size_t os::numa_get_groups_num() { 2789 int max_node = Linux::numa_max_node(); 2790 return max_node > 0 ? max_node + 1 : 1; 2791} 2792 2793int os::numa_get_group_id() { 2794 int cpu_id = Linux::sched_getcpu(); 2795 if (cpu_id != -1) { 2796 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2797 if (lgrp_id != -1) { 2798 return lgrp_id; 2799 } 2800 } 2801 return 0; 2802} 2803 2804size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2805 for (size_t i = 0; i < size; i++) { 2806 ids[i] = i; 2807 } 2808 return size; 2809} 2810 2811bool os::get_page_info(char *start, page_info* info) { 2812 return false; 2813} 2814 2815char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2816 return end; 2817} 2818 2819 2820int os::Linux::sched_getcpu_syscall(void) { 2821 unsigned int cpu; 2822 int retval = -1; 2823 2824#if defined(IA32) 2825# ifndef SYS_getcpu 2826# define SYS_getcpu 318 2827# endif 2828 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2829#elif defined(AMD64) 2830// Unfortunately we have to bring all these macros here from vsyscall.h 2831// to be able to compile on old linuxes. 2832# define __NR_vgetcpu 2 2833# define VSYSCALL_START (-10UL << 20) 2834# define VSYSCALL_SIZE 1024 2835# define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2836 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2837 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2838 retval = vgetcpu(&cpu, NULL, NULL); 2839#endif 2840 2841 return (retval == -1) ? retval : cpu; 2842} 2843 2844// Something to do with the numa-aware allocator needs these symbols 2845extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2846extern "C" JNIEXPORT void numa_error(char *where) { } 2847extern "C" JNIEXPORT int fork1() { return fork(); } 2848 2849 2850// If we are running with libnuma version > 2, then we should 2851// be trying to use symbols with versions 1.1 2852// If we are running with earlier version, which did not have symbol versions, 2853// we should use the base version. 2854void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2855 void *f = dlvsym(handle, name, "libnuma_1.1"); 2856 if (f == NULL) { 2857 f = dlsym(handle, name); 2858 } 2859 return f; 2860} 2861 2862bool os::Linux::libnuma_init() { 2863 // sched_getcpu() should be in libc. 2864 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2865 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2866 2867 // If it's not, try a direct syscall. 2868 if (sched_getcpu() == -1) 2869 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall)); 2870 2871 if (sched_getcpu() != -1) { // Does it work? 2872 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2873 if (handle != NULL) { 2874 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2875 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2876 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2877 libnuma_dlsym(handle, "numa_max_node"))); 2878 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2879 libnuma_dlsym(handle, "numa_available"))); 2880 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2881 libnuma_dlsym(handle, "numa_tonode_memory"))); 2882 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2883 libnuma_dlsym(handle, "numa_interleave_memory"))); 2884 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2885 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2886 2887 2888 if (numa_available() != -1) { 2889 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2890 // Create a cpu -> node mapping 2891 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2892 rebuild_cpu_to_node_map(); 2893 return true; 2894 } 2895 } 2896 } 2897 return false; 2898} 2899 2900// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2901// The table is later used in get_node_by_cpu(). 2902void os::Linux::rebuild_cpu_to_node_map() { 2903 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2904 // in libnuma (possible values are starting from 16, 2905 // and continuing up with every other power of 2, but less 2906 // than the maximum number of CPUs supported by kernel), and 2907 // is a subject to change (in libnuma version 2 the requirements 2908 // are more reasonable) we'll just hardcode the number they use 2909 // in the library. 2910 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2911 2912 size_t cpu_num = os::active_processor_count(); 2913 size_t cpu_map_size = NCPUS / BitsPerCLong; 2914 size_t cpu_map_valid_size = 2915 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2916 2917 cpu_to_node()->clear(); 2918 cpu_to_node()->at_grow(cpu_num - 1); 2919 size_t node_num = numa_get_groups_num(); 2920 2921 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2922 for (size_t i = 0; i < node_num; i++) { 2923 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2924 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2925 if (cpu_map[j] != 0) { 2926 for (size_t k = 0; k < BitsPerCLong; k++) { 2927 if (cpu_map[j] & (1UL << k)) { 2928 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2929 } 2930 } 2931 } 2932 } 2933 } 2934 } 2935 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2936} 2937 2938int os::Linux::get_node_by_cpu(int cpu_id) { 2939 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2940 return cpu_to_node()->at(cpu_id); 2941 } 2942 return -1; 2943} 2944 2945GrowableArray<int>* os::Linux::_cpu_to_node; 2946os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2947os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2948os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2949os::Linux::numa_available_func_t os::Linux::_numa_available; 2950os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2951os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2952os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2953unsigned long* os::Linux::_numa_all_nodes; 2954 2955bool os::pd_uncommit_memory(char* addr, size_t size) { 2956 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2957 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2958 return res != (uintptr_t) MAP_FAILED; 2959} 2960 2961static 2962address get_stack_commited_bottom(address bottom, size_t size) { 2963 address nbot = bottom; 2964 address ntop = bottom + size; 2965 2966 size_t page_sz = os::vm_page_size(); 2967 unsigned pages = size / page_sz; 2968 2969 unsigned char vec[1]; 2970 unsigned imin = 1, imax = pages + 1, imid; 2971 int mincore_return_value; 2972 2973 while (imin < imax) { 2974 imid = (imax + imin) / 2; 2975 nbot = ntop - (imid * page_sz); 2976 2977 // Use a trick with mincore to check whether the page is mapped or not. 2978 // mincore sets vec to 1 if page resides in memory and to 0 if page 2979 // is swapped output but if page we are asking for is unmapped 2980 // it returns -1,ENOMEM 2981 mincore_return_value = mincore(nbot, page_sz, vec); 2982 2983 if (mincore_return_value == -1) { 2984 // Page is not mapped go up 2985 // to find first mapped page 2986 if (errno != EAGAIN) { 2987 assert(errno == ENOMEM, "Unexpected mincore errno"); 2988 imax = imid; 2989 } 2990 } else { 2991 // Page is mapped go down 2992 // to find first not mapped page 2993 imin = imid + 1; 2994 } 2995 } 2996 2997 nbot = nbot + page_sz; 2998 2999 // Adjust stack bottom one page up if last checked page is not mapped 3000 if (mincore_return_value == -1) { 3001 nbot = nbot + page_sz; 3002 } 3003 3004 return nbot; 3005} 3006 3007 3008// Linux uses a growable mapping for the stack, and if the mapping for 3009// the stack guard pages is not removed when we detach a thread the 3010// stack cannot grow beyond the pages where the stack guard was 3011// mapped. If at some point later in the process the stack expands to 3012// that point, the Linux kernel cannot expand the stack any further 3013// because the guard pages are in the way, and a segfault occurs. 3014// 3015// However, it's essential not to split the stack region by unmapping 3016// a region (leaving a hole) that's already part of the stack mapping, 3017// so if the stack mapping has already grown beyond the guard pages at 3018// the time we create them, we have to truncate the stack mapping. 3019// So, we need to know the extent of the stack mapping when 3020// create_stack_guard_pages() is called. 3021 3022// We only need this for stacks that are growable: at the time of 3023// writing thread stacks don't use growable mappings (i.e. those 3024// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3025// only applies to the main thread. 3026 3027// If the (growable) stack mapping already extends beyond the point 3028// where we're going to put our guard pages, truncate the mapping at 3029// that point by munmap()ping it. This ensures that when we later 3030// munmap() the guard pages we don't leave a hole in the stack 3031// mapping. This only affects the main/initial thread 3032 3033bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3034 3035 if (os::Linux::is_initial_thread()) { 3036 // As we manually grow stack up to bottom inside create_attached_thread(), 3037 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3038 // we don't need to do anything special. 3039 // Check it first, before calling heavy function. 3040 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3041 unsigned char vec[1]; 3042 3043 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3044 // Fallback to slow path on all errors, including EAGAIN 3045 stack_extent = (uintptr_t) get_stack_commited_bottom( 3046 os::Linux::initial_thread_stack_bottom(), 3047 (size_t)addr - stack_extent); 3048 } 3049 3050 if (stack_extent < (uintptr_t)addr) { 3051 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3052 } 3053 } 3054 3055 return os::commit_memory(addr, size, !ExecMem); 3056} 3057 3058// If this is a growable mapping, remove the guard pages entirely by 3059// munmap()ping them. If not, just call uncommit_memory(). This only 3060// affects the main/initial thread, but guard against future OS changes 3061// It's safe to always unmap guard pages for initial thread because we 3062// always place it right after end of the mapped region 3063 3064bool os::remove_stack_guard_pages(char* addr, size_t size) { 3065 uintptr_t stack_extent, stack_base; 3066 3067 if (os::Linux::is_initial_thread()) { 3068 return ::munmap(addr, size) == 0; 3069 } 3070 3071 return os::uncommit_memory(addr, size); 3072} 3073 3074static address _highest_vm_reserved_address = NULL; 3075 3076// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3077// at 'requested_addr'. If there are existing memory mappings at the same 3078// location, however, they will be overwritten. If 'fixed' is false, 3079// 'requested_addr' is only treated as a hint, the return value may or 3080// may not start from the requested address. Unlike Linux mmap(), this 3081// function returns NULL to indicate failure. 3082static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3083 char * addr; 3084 int flags; 3085 3086 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3087 if (fixed) { 3088 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3089 flags |= MAP_FIXED; 3090 } 3091 3092 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3093 // touch an uncommitted page. Otherwise, the read/write might 3094 // succeed if we have enough swap space to back the physical page. 3095 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3096 flags, -1, 0); 3097 3098 if (addr != MAP_FAILED) { 3099 // anon_mmap() should only get called during VM initialization, 3100 // don't need lock (actually we can skip locking even it can be called 3101 // from multiple threads, because _highest_vm_reserved_address is just a 3102 // hint about the upper limit of non-stack memory regions.) 3103 if ((address)addr + bytes > _highest_vm_reserved_address) { 3104 _highest_vm_reserved_address = (address)addr + bytes; 3105 } 3106 } 3107 3108 return addr == MAP_FAILED ? NULL : addr; 3109} 3110 3111// Don't update _highest_vm_reserved_address, because there might be memory 3112// regions above addr + size. If so, releasing a memory region only creates 3113// a hole in the address space, it doesn't help prevent heap-stack collision. 3114// 3115static int anon_munmap(char * addr, size_t size) { 3116 return ::munmap(addr, size) == 0; 3117} 3118 3119char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3120 size_t alignment_hint) { 3121 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3122} 3123 3124bool os::pd_release_memory(char* addr, size_t size) { 3125 return anon_munmap(addr, size); 3126} 3127 3128static address highest_vm_reserved_address() { 3129 return _highest_vm_reserved_address; 3130} 3131 3132static bool linux_mprotect(char* addr, size_t size, int prot) { 3133 // Linux wants the mprotect address argument to be page aligned. 3134 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3135 3136 // According to SUSv3, mprotect() should only be used with mappings 3137 // established by mmap(), and mmap() always maps whole pages. Unaligned 3138 // 'addr' likely indicates problem in the VM (e.g. trying to change 3139 // protection of malloc'ed or statically allocated memory). Check the 3140 // caller if you hit this assert. 3141 assert(addr == bottom, "sanity check"); 3142 3143 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3144 return ::mprotect(bottom, size, prot) == 0; 3145} 3146 3147// Set protections specified 3148bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3149 bool is_committed) { 3150 unsigned int p = 0; 3151 switch (prot) { 3152 case MEM_PROT_NONE: p = PROT_NONE; break; 3153 case MEM_PROT_READ: p = PROT_READ; break; 3154 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3155 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3156 default: 3157 ShouldNotReachHere(); 3158 } 3159 // is_committed is unused. 3160 return linux_mprotect(addr, bytes, p); 3161} 3162 3163bool os::guard_memory(char* addr, size_t size) { 3164 return linux_mprotect(addr, size, PROT_NONE); 3165} 3166 3167bool os::unguard_memory(char* addr, size_t size) { 3168 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3169} 3170 3171bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) { 3172 bool result = false; 3173 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3174 MAP_ANONYMOUS|MAP_PRIVATE, 3175 -1, 0); 3176 if (p != MAP_FAILED) { 3177 void *aligned_p = align_ptr_up(p, page_size); 3178 3179 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3180 3181 munmap(p, page_size * 2); 3182 } 3183 3184 if (warn && !result) { 3185 warning("TransparentHugePages is not supported by the operating system."); 3186 } 3187 3188 return result; 3189} 3190 3191bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3192 bool result = false; 3193 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3194 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3195 -1, 0); 3196 3197 if (p != MAP_FAILED) { 3198 // We don't know if this really is a huge page or not. 3199 FILE *fp = fopen("/proc/self/maps", "r"); 3200 if (fp) { 3201 while (!feof(fp)) { 3202 char chars[257]; 3203 long x = 0; 3204 if (fgets(chars, sizeof(chars), fp)) { 3205 if (sscanf(chars, "%lx-%*x", &x) == 1 3206 && x == (long)p) { 3207 if (strstr (chars, "hugepage")) { 3208 result = true; 3209 break; 3210 } 3211 } 3212 } 3213 } 3214 fclose(fp); 3215 } 3216 munmap(p, page_size); 3217 } 3218 3219 if (warn && !result) { 3220 warning("HugeTLBFS is not supported by the operating system."); 3221 } 3222 3223 return result; 3224} 3225 3226/* 3227* Set the coredump_filter bits to include largepages in core dump (bit 6) 3228* 3229* From the coredump_filter documentation: 3230* 3231* - (bit 0) anonymous private memory 3232* - (bit 1) anonymous shared memory 3233* - (bit 2) file-backed private memory 3234* - (bit 3) file-backed shared memory 3235* - (bit 4) ELF header pages in file-backed private memory areas (it is 3236* effective only if the bit 2 is cleared) 3237* - (bit 5) hugetlb private memory 3238* - (bit 6) hugetlb shared memory 3239*/ 3240static void set_coredump_filter(void) { 3241 FILE *f; 3242 long cdm; 3243 3244 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3245 return; 3246 } 3247 3248 if (fscanf(f, "%lx", &cdm) != 1) { 3249 fclose(f); 3250 return; 3251 } 3252 3253 rewind(f); 3254 3255 if ((cdm & LARGEPAGES_BIT) == 0) { 3256 cdm |= LARGEPAGES_BIT; 3257 fprintf(f, "%#lx", cdm); 3258 } 3259 3260 fclose(f); 3261} 3262 3263// Large page support 3264 3265static size_t _large_page_size = 0; 3266 3267size_t os::Linux::find_large_page_size() { 3268 size_t large_page_size = 0; 3269 3270 // large_page_size on Linux is used to round up heap size. x86 uses either 3271 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3272 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3273 // page as large as 256M. 3274 // 3275 // Here we try to figure out page size by parsing /proc/meminfo and looking 3276 // for a line with the following format: 3277 // Hugepagesize: 2048 kB 3278 // 3279 // If we can't determine the value (e.g. /proc is not mounted, or the text 3280 // format has been changed), we'll use the largest page size supported by 3281 // the processor. 3282 3283#ifndef ZERO 3284 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3285 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3286#endif // ZERO 3287 3288 FILE *fp = fopen("/proc/meminfo", "r"); 3289 if (fp) { 3290 while (!feof(fp)) { 3291 int x = 0; 3292 char buf[16]; 3293 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3294 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3295 large_page_size = x * K; 3296 break; 3297 } 3298 } else { 3299 // skip to next line 3300 for (;;) { 3301 int ch = fgetc(fp); 3302 if (ch == EOF || ch == (int)'\n') break; 3303 } 3304 } 3305 } 3306 fclose(fp); 3307 } 3308 3309 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3310 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3311 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3312 proper_unit_for_byte_size(large_page_size)); 3313 } 3314 3315 return large_page_size; 3316} 3317 3318size_t os::Linux::setup_large_page_size() { 3319 _large_page_size = Linux::find_large_page_size(); 3320 const size_t default_page_size = (size_t)Linux::page_size(); 3321 if (_large_page_size > default_page_size) { 3322 _page_sizes[0] = _large_page_size; 3323 _page_sizes[1] = default_page_size; 3324 _page_sizes[2] = 0; 3325 } 3326 3327 return _large_page_size; 3328} 3329 3330bool os::Linux::setup_large_page_type(size_t page_size) { 3331 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3332 FLAG_IS_DEFAULT(UseSHM) && 3333 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3334 // If UseLargePages is specified on the command line try all methods, 3335 // if it's default, then try only UseTransparentHugePages. 3336 if (FLAG_IS_DEFAULT(UseLargePages)) { 3337 UseTransparentHugePages = true; 3338 } else { 3339 UseHugeTLBFS = UseTransparentHugePages = UseSHM = true; 3340 } 3341 } 3342 3343 if (UseTransparentHugePages) { 3344 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3345 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3346 UseHugeTLBFS = false; 3347 UseSHM = false; 3348 return true; 3349 } 3350 UseTransparentHugePages = false; 3351 } 3352 3353 if (UseHugeTLBFS) { 3354 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3355 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3356 UseSHM = false; 3357 return true; 3358 } 3359 UseHugeTLBFS = false; 3360 } 3361 3362 return UseSHM; 3363} 3364 3365void os::large_page_init() { 3366 if (!UseLargePages) { 3367 UseHugeTLBFS = false; 3368 UseTransparentHugePages = false; 3369 UseSHM = false; 3370 return; 3371 } 3372 3373 size_t large_page_size = Linux::setup_large_page_size(); 3374 UseLargePages = Linux::setup_large_page_type(large_page_size); 3375 3376 set_coredump_filter(); 3377} 3378 3379#ifndef SHM_HUGETLB 3380#define SHM_HUGETLB 04000 3381#endif 3382 3383char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3384 // "exec" is passed in but not used. Creating the shared image for 3385 // the code cache doesn't have an SHM_X executable permission to check. 3386 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3387 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3388 3389 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3390 return NULL; // Fallback to small pages. 3391 } 3392 3393 key_t key = IPC_PRIVATE; 3394 char *addr; 3395 3396 bool warn_on_failure = UseLargePages && 3397 (!FLAG_IS_DEFAULT(UseLargePages) || 3398 !FLAG_IS_DEFAULT(UseSHM) || 3399 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 3400 ); 3401 char msg[128]; 3402 3403 // Create a large shared memory region to attach to based on size. 3404 // Currently, size is the total size of the heap 3405 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3406 if (shmid == -1) { 3407 // Possible reasons for shmget failure: 3408 // 1. shmmax is too small for Java heap. 3409 // > check shmmax value: cat /proc/sys/kernel/shmmax 3410 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3411 // 2. not enough large page memory. 3412 // > check available large pages: cat /proc/meminfo 3413 // > increase amount of large pages: 3414 // echo new_value > /proc/sys/vm/nr_hugepages 3415 // Note 1: different Linux may use different name for this property, 3416 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3417 // Note 2: it's possible there's enough physical memory available but 3418 // they are so fragmented after a long run that they can't 3419 // coalesce into large pages. Try to reserve large pages when 3420 // the system is still "fresh". 3421 if (warn_on_failure) { 3422 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3423 warning(msg); 3424 } 3425 return NULL; 3426 } 3427 3428 // attach to the region 3429 addr = (char*)shmat(shmid, req_addr, 0); 3430 int err = errno; 3431 3432 // Remove shmid. If shmat() is successful, the actual shared memory segment 3433 // will be deleted when it's detached by shmdt() or when the process 3434 // terminates. If shmat() is not successful this will remove the shared 3435 // segment immediately. 3436 shmctl(shmid, IPC_RMID, NULL); 3437 3438 if ((intptr_t)addr == -1) { 3439 if (warn_on_failure) { 3440 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3441 warning(msg); 3442 } 3443 return NULL; 3444 } 3445 3446 return addr; 3447} 3448 3449static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) { 3450 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3451 3452 bool warn_on_failure = UseLargePages && 3453 (!FLAG_IS_DEFAULT(UseLargePages) || 3454 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3455 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3456 3457 if (warn_on_failure) { 3458 char msg[128]; 3459 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3460 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3461 warning(msg); 3462 } 3463} 3464 3465char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) { 3466 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3467 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3468 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3469 3470 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3471 char* addr = (char*)::mmap(req_addr, bytes, prot, 3472 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3473 -1, 0); 3474 3475 if (addr == MAP_FAILED) { 3476 warn_on_large_pages_failure(req_addr, bytes, errno); 3477 return NULL; 3478 } 3479 3480 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3481 3482 return addr; 3483} 3484 3485char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3486 size_t large_page_size = os::large_page_size(); 3487 3488 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3489 3490 // Allocate small pages. 3491 3492 char* start; 3493 if (req_addr != NULL) { 3494 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3495 assert(is_size_aligned(bytes, alignment), "Must be"); 3496 start = os::reserve_memory(bytes, req_addr); 3497 assert(start == NULL || start == req_addr, "Must be"); 3498 } else { 3499 start = os::reserve_memory_aligned(bytes, alignment); 3500 } 3501 3502 if (start == NULL) { 3503 return NULL; 3504 } 3505 3506 assert(is_ptr_aligned(start, alignment), "Must be"); 3507 3508 // os::reserve_memory_special will record this memory area. 3509 // Need to release it here to prevent overlapping reservations. 3510 MemTracker::record_virtual_memory_release((address)start, bytes); 3511 3512 char* end = start + bytes; 3513 3514 // Find the regions of the allocated chunk that can be promoted to large pages. 3515 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3516 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3517 3518 size_t lp_bytes = lp_end - lp_start; 3519 3520 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3521 3522 if (lp_bytes == 0) { 3523 // The mapped region doesn't even span the start and the end of a large page. 3524 // Fall back to allocate a non-special area. 3525 ::munmap(start, end - start); 3526 return NULL; 3527 } 3528 3529 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3530 3531 3532 void* result; 3533 3534 if (start != lp_start) { 3535 result = ::mmap(start, lp_start - start, prot, 3536 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3537 -1, 0); 3538 if (result == MAP_FAILED) { 3539 ::munmap(lp_start, end - lp_start); 3540 return NULL; 3541 } 3542 } 3543 3544 result = ::mmap(lp_start, lp_bytes, prot, 3545 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3546 -1, 0); 3547 if (result == MAP_FAILED) { 3548 warn_on_large_pages_failure(req_addr, bytes, errno); 3549 // If the mmap above fails, the large pages region will be unmapped and we 3550 // have regions before and after with small pages. Release these regions. 3551 // 3552 // | mapped | unmapped | mapped | 3553 // ^ ^ ^ ^ 3554 // start lp_start lp_end end 3555 // 3556 ::munmap(start, lp_start - start); 3557 ::munmap(lp_end, end - lp_end); 3558 return NULL; 3559 } 3560 3561 if (lp_end != end) { 3562 result = ::mmap(lp_end, end - lp_end, prot, 3563 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3564 -1, 0); 3565 if (result == MAP_FAILED) { 3566 ::munmap(start, lp_end - start); 3567 return NULL; 3568 } 3569 } 3570 3571 return start; 3572} 3573 3574char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3575 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3576 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3577 assert(is_power_of_2(alignment), "Must be"); 3578 assert(is_power_of_2(os::large_page_size()), "Must be"); 3579 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3580 3581 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3582 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3583 } else { 3584 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3585 } 3586} 3587 3588char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3589 assert(UseLargePages, "only for large pages"); 3590 3591 char* addr; 3592 if (UseSHM) { 3593 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3594 } else { 3595 assert(UseHugeTLBFS, "must be"); 3596 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3597 } 3598 3599 if (addr != NULL) { 3600 if (UseNUMAInterleaving) { 3601 numa_make_global(addr, bytes); 3602 } 3603 3604 // The memory is committed 3605 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC); 3606 } 3607 3608 return addr; 3609} 3610 3611bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3612 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3613 return shmdt(base) == 0; 3614} 3615 3616bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3617 return pd_release_memory(base, bytes); 3618} 3619 3620bool os::release_memory_special(char* base, size_t bytes) { 3621 assert(UseLargePages, "only for large pages"); 3622 3623 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3624 3625 bool res; 3626 if (UseSHM) { 3627 res = os::Linux::release_memory_special_shm(base, bytes); 3628 } else { 3629 assert(UseHugeTLBFS, "must be"); 3630 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3631 } 3632 3633 if (res) { 3634 tkr.record((address)base, bytes); 3635 } else { 3636 tkr.discard(); 3637 } 3638 3639 return res; 3640} 3641 3642size_t os::large_page_size() { 3643 return _large_page_size; 3644} 3645 3646// With SysV SHM the entire memory region must be allocated as shared 3647// memory. 3648// HugeTLBFS allows application to commit large page memory on demand. 3649// However, when committing memory with HugeTLBFS fails, the region 3650// that was supposed to be committed will lose the old reservation 3651// and allow other threads to steal that memory region. Because of this 3652// behavior we can't commit HugeTLBFS memory. 3653bool os::can_commit_large_page_memory() { 3654 return UseTransparentHugePages; 3655} 3656 3657bool os::can_execute_large_page_memory() { 3658 return UseTransparentHugePages || UseHugeTLBFS; 3659} 3660 3661// Reserve memory at an arbitrary address, only if that area is 3662// available (and not reserved for something else). 3663 3664char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3665 const int max_tries = 10; 3666 char* base[max_tries]; 3667 size_t size[max_tries]; 3668 const size_t gap = 0x000000; 3669 3670 // Assert only that the size is a multiple of the page size, since 3671 // that's all that mmap requires, and since that's all we really know 3672 // about at this low abstraction level. If we need higher alignment, 3673 // we can either pass an alignment to this method or verify alignment 3674 // in one of the methods further up the call chain. See bug 5044738. 3675 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3676 3677 // Repeatedly allocate blocks until the block is allocated at the 3678 // right spot. Give up after max_tries. Note that reserve_memory() will 3679 // automatically update _highest_vm_reserved_address if the call is 3680 // successful. The variable tracks the highest memory address every reserved 3681 // by JVM. It is used to detect heap-stack collision if running with 3682 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3683 // space than needed, it could confuse the collision detecting code. To 3684 // solve the problem, save current _highest_vm_reserved_address and 3685 // calculate the correct value before return. 3686 address old_highest = _highest_vm_reserved_address; 3687 3688 // Linux mmap allows caller to pass an address as hint; give it a try first, 3689 // if kernel honors the hint then we can return immediately. 3690 char * addr = anon_mmap(requested_addr, bytes, false); 3691 if (addr == requested_addr) { 3692 return requested_addr; 3693 } 3694 3695 if (addr != NULL) { 3696 // mmap() is successful but it fails to reserve at the requested address 3697 anon_munmap(addr, bytes); 3698 } 3699 3700 int i; 3701 for (i = 0; i < max_tries; ++i) { 3702 base[i] = reserve_memory(bytes); 3703 3704 if (base[i] != NULL) { 3705 // Is this the block we wanted? 3706 if (base[i] == requested_addr) { 3707 size[i] = bytes; 3708 break; 3709 } 3710 3711 // Does this overlap the block we wanted? Give back the overlapped 3712 // parts and try again. 3713 3714 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3715 if (top_overlap >= 0 && top_overlap < bytes) { 3716 unmap_memory(base[i], top_overlap); 3717 base[i] += top_overlap; 3718 size[i] = bytes - top_overlap; 3719 } else { 3720 size_t bottom_overlap = base[i] + bytes - requested_addr; 3721 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3722 unmap_memory(requested_addr, bottom_overlap); 3723 size[i] = bytes - bottom_overlap; 3724 } else { 3725 size[i] = bytes; 3726 } 3727 } 3728 } 3729 } 3730 3731 // Give back the unused reserved pieces. 3732 3733 for (int j = 0; j < i; ++j) { 3734 if (base[j] != NULL) { 3735 unmap_memory(base[j], size[j]); 3736 } 3737 } 3738 3739 if (i < max_tries) { 3740 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3741 return requested_addr; 3742 } else { 3743 _highest_vm_reserved_address = old_highest; 3744 return NULL; 3745 } 3746} 3747 3748size_t os::read(int fd, void *buf, unsigned int nBytes) { 3749 return ::read(fd, buf, nBytes); 3750} 3751 3752// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 3753// Solaris uses poll(), linux uses park(). 3754// Poll() is likely a better choice, assuming that Thread.interrupt() 3755// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 3756// SIGSEGV, see 4355769. 3757 3758int os::sleep(Thread* thread, jlong millis, bool interruptible) { 3759 assert(thread == Thread::current(), "thread consistency check"); 3760 3761 ParkEvent * const slp = thread->_SleepEvent ; 3762 slp->reset() ; 3763 OrderAccess::fence() ; 3764 3765 if (interruptible) { 3766 jlong prevtime = javaTimeNanos(); 3767 3768 for (;;) { 3769 if (os::is_interrupted(thread, true)) { 3770 return OS_INTRPT; 3771 } 3772 3773 jlong newtime = javaTimeNanos(); 3774 3775 if (newtime - prevtime < 0) { 3776 // time moving backwards, should only happen if no monotonic clock 3777 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3778 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3779 } else { 3780 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3781 } 3782 3783 if(millis <= 0) { 3784 return OS_OK; 3785 } 3786 3787 prevtime = newtime; 3788 3789 { 3790 assert(thread->is_Java_thread(), "sanity check"); 3791 JavaThread *jt = (JavaThread *) thread; 3792 ThreadBlockInVM tbivm(jt); 3793 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 3794 3795 jt->set_suspend_equivalent(); 3796 // cleared by handle_special_suspend_equivalent_condition() or 3797 // java_suspend_self() via check_and_wait_while_suspended() 3798 3799 slp->park(millis); 3800 3801 // were we externally suspended while we were waiting? 3802 jt->check_and_wait_while_suspended(); 3803 } 3804 } 3805 } else { 3806 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 3807 jlong prevtime = javaTimeNanos(); 3808 3809 for (;;) { 3810 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 3811 // the 1st iteration ... 3812 jlong newtime = javaTimeNanos(); 3813 3814 if (newtime - prevtime < 0) { 3815 // time moving backwards, should only happen if no monotonic clock 3816 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3817 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3818 } else { 3819 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3820 } 3821 3822 if(millis <= 0) break ; 3823 3824 prevtime = newtime; 3825 slp->park(millis); 3826 } 3827 return OS_OK ; 3828 } 3829} 3830 3831int os::naked_sleep() { 3832 // %% make the sleep time an integer flag. for now use 1 millisec. 3833 return os::sleep(Thread::current(), 1, false); 3834} 3835 3836// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3837void os::infinite_sleep() { 3838 while (true) { // sleep forever ... 3839 ::sleep(100); // ... 100 seconds at a time 3840 } 3841} 3842 3843// Used to convert frequent JVM_Yield() to nops 3844bool os::dont_yield() { 3845 return DontYieldALot; 3846} 3847 3848void os::yield() { 3849 sched_yield(); 3850} 3851 3852os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3853 3854void os::yield_all(int attempts) { 3855 // Yields to all threads, including threads with lower priorities 3856 // Threads on Linux are all with same priority. The Solaris style 3857 // os::yield_all() with nanosleep(1ms) is not necessary. 3858 sched_yield(); 3859} 3860 3861// Called from the tight loops to possibly influence time-sharing heuristics 3862void os::loop_breaker(int attempts) { 3863 os::yield_all(attempts); 3864} 3865 3866//////////////////////////////////////////////////////////////////////////////// 3867// thread priority support 3868 3869// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3870// only supports dynamic priority, static priority must be zero. For real-time 3871// applications, Linux supports SCHED_RR which allows static priority (1-99). 3872// However, for large multi-threaded applications, SCHED_RR is not only slower 3873// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3874// of 5 runs - Sep 2005). 3875// 3876// The following code actually changes the niceness of kernel-thread/LWP. It 3877// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3878// not the entire user process, and user level threads are 1:1 mapped to kernel 3879// threads. It has always been the case, but could change in the future. For 3880// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3881// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3882 3883int os::java_to_os_priority[CriticalPriority + 1] = { 3884 19, // 0 Entry should never be used 3885 3886 4, // 1 MinPriority 3887 3, // 2 3888 2, // 3 3889 3890 1, // 4 3891 0, // 5 NormPriority 3892 -1, // 6 3893 3894 -2, // 7 3895 -3, // 8 3896 -4, // 9 NearMaxPriority 3897 3898 -5, // 10 MaxPriority 3899 3900 -5 // 11 CriticalPriority 3901}; 3902 3903static int prio_init() { 3904 if (ThreadPriorityPolicy == 1) { 3905 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3906 // if effective uid is not root. Perhaps, a more elegant way of doing 3907 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3908 if (geteuid() != 0) { 3909 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3910 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3911 } 3912 ThreadPriorityPolicy = 0; 3913 } 3914 } 3915 if (UseCriticalJavaThreadPriority) { 3916 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3917 } 3918 return 0; 3919} 3920 3921OSReturn os::set_native_priority(Thread* thread, int newpri) { 3922 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3923 3924 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3925 return (ret == 0) ? OS_OK : OS_ERR; 3926} 3927 3928OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3929 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3930 *priority_ptr = java_to_os_priority[NormPriority]; 3931 return OS_OK; 3932 } 3933 3934 errno = 0; 3935 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3936 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3937} 3938 3939// Hint to the underlying OS that a task switch would not be good. 3940// Void return because it's a hint and can fail. 3941void os::hint_no_preempt() {} 3942 3943//////////////////////////////////////////////////////////////////////////////// 3944// suspend/resume support 3945 3946// the low-level signal-based suspend/resume support is a remnant from the 3947// old VM-suspension that used to be for java-suspension, safepoints etc, 3948// within hotspot. Now there is a single use-case for this: 3949// - calling get_thread_pc() on the VMThread by the flat-profiler task 3950// that runs in the watcher thread. 3951// The remaining code is greatly simplified from the more general suspension 3952// code that used to be used. 3953// 3954// The protocol is quite simple: 3955// - suspend: 3956// - sends a signal to the target thread 3957// - polls the suspend state of the osthread using a yield loop 3958// - target thread signal handler (SR_handler) sets suspend state 3959// and blocks in sigsuspend until continued 3960// - resume: 3961// - sets target osthread state to continue 3962// - sends signal to end the sigsuspend loop in the SR_handler 3963// 3964// Note that the SR_lock plays no role in this suspend/resume protocol. 3965// 3966 3967static void resume_clear_context(OSThread *osthread) { 3968 osthread->set_ucontext(NULL); 3969 osthread->set_siginfo(NULL); 3970} 3971 3972static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3973 osthread->set_ucontext(context); 3974 osthread->set_siginfo(siginfo); 3975} 3976 3977// 3978// Handler function invoked when a thread's execution is suspended or 3979// resumed. We have to be careful that only async-safe functions are 3980// called here (Note: most pthread functions are not async safe and 3981// should be avoided.) 3982// 3983// Note: sigwait() is a more natural fit than sigsuspend() from an 3984// interface point of view, but sigwait() prevents the signal hander 3985// from being run. libpthread would get very confused by not having 3986// its signal handlers run and prevents sigwait()'s use with the 3987// mutex granting granting signal. 3988// 3989// Currently only ever called on the VMThread and JavaThreads (PC sampling) 3990// 3991static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3992 // Save and restore errno to avoid confusing native code with EINTR 3993 // after sigsuspend. 3994 int old_errno = errno; 3995 3996 Thread* thread = Thread::current(); 3997 OSThread* osthread = thread->osthread(); 3998 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3999 4000 os::SuspendResume::State current = osthread->sr.state(); 4001 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 4002 suspend_save_context(osthread, siginfo, context); 4003 4004 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4005 os::SuspendResume::State state = osthread->sr.suspended(); 4006 if (state == os::SuspendResume::SR_SUSPENDED) { 4007 sigset_t suspend_set; // signals for sigsuspend() 4008 4009 // get current set of blocked signals and unblock resume signal 4010 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4011 sigdelset(&suspend_set, SR_signum); 4012 4013 sr_semaphore.signal(); 4014 // wait here until we are resumed 4015 while (1) { 4016 sigsuspend(&suspend_set); 4017 4018 os::SuspendResume::State result = osthread->sr.running(); 4019 if (result == os::SuspendResume::SR_RUNNING) { 4020 sr_semaphore.signal(); 4021 break; 4022 } 4023 } 4024 4025 } else if (state == os::SuspendResume::SR_RUNNING) { 4026 // request was cancelled, continue 4027 } else { 4028 ShouldNotReachHere(); 4029 } 4030 4031 resume_clear_context(osthread); 4032 } else if (current == os::SuspendResume::SR_RUNNING) { 4033 // request was cancelled, continue 4034 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4035 // ignore 4036 } else { 4037 // ignore 4038 } 4039 4040 errno = old_errno; 4041} 4042 4043 4044static int SR_initialize() { 4045 struct sigaction act; 4046 char *s; 4047 /* Get signal number to use for suspend/resume */ 4048 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4049 int sig = ::strtol(s, 0, 10); 4050 if (sig > 0 || sig < _NSIG) { 4051 SR_signum = sig; 4052 } 4053 } 4054 4055 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4056 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4057 4058 sigemptyset(&SR_sigset); 4059 sigaddset(&SR_sigset, SR_signum); 4060 4061 /* Set up signal handler for suspend/resume */ 4062 act.sa_flags = SA_RESTART|SA_SIGINFO; 4063 act.sa_handler = (void (*)(int)) SR_handler; 4064 4065 // SR_signum is blocked by default. 4066 // 4528190 - We also need to block pthread restart signal (32 on all 4067 // supported Linux platforms). Note that LinuxThreads need to block 4068 // this signal for all threads to work properly. So we don't have 4069 // to use hard-coded signal number when setting up the mask. 4070 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4071 4072 if (sigaction(SR_signum, &act, 0) == -1) { 4073 return -1; 4074 } 4075 4076 // Save signal flag 4077 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4078 return 0; 4079} 4080 4081static int sr_notify(OSThread* osthread) { 4082 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4083 assert_status(status == 0, status, "pthread_kill"); 4084 return status; 4085} 4086 4087// "Randomly" selected value for how long we want to spin 4088// before bailing out on suspending a thread, also how often 4089// we send a signal to a thread we want to resume 4090static const int RANDOMLY_LARGE_INTEGER = 1000000; 4091static const int RANDOMLY_LARGE_INTEGER2 = 100; 4092 4093// returns true on success and false on error - really an error is fatal 4094// but this seems the normal response to library errors 4095static bool do_suspend(OSThread* osthread) { 4096 assert(osthread->sr.is_running(), "thread should be running"); 4097 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4098 4099 // mark as suspended and send signal 4100 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4101 // failed to switch, state wasn't running? 4102 ShouldNotReachHere(); 4103 return false; 4104 } 4105 4106 if (sr_notify(osthread) != 0) { 4107 ShouldNotReachHere(); 4108 } 4109 4110 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4111 while (true) { 4112 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4113 break; 4114 } else { 4115 // timeout 4116 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4117 if (cancelled == os::SuspendResume::SR_RUNNING) { 4118 return false; 4119 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4120 // make sure that we consume the signal on the semaphore as well 4121 sr_semaphore.wait(); 4122 break; 4123 } else { 4124 ShouldNotReachHere(); 4125 return false; 4126 } 4127 } 4128 } 4129 4130 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4131 return true; 4132} 4133 4134static void do_resume(OSThread* osthread) { 4135 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4136 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4137 4138 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4139 // failed to switch to WAKEUP_REQUEST 4140 ShouldNotReachHere(); 4141 return; 4142 } 4143 4144 while (true) { 4145 if (sr_notify(osthread) == 0) { 4146 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4147 if (osthread->sr.is_running()) { 4148 return; 4149 } 4150 } 4151 } else { 4152 ShouldNotReachHere(); 4153 } 4154 } 4155 4156 guarantee(osthread->sr.is_running(), "Must be running!"); 4157} 4158 4159//////////////////////////////////////////////////////////////////////////////// 4160// interrupt support 4161 4162void os::interrupt(Thread* thread) { 4163 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 4164 "possibility of dangling Thread pointer"); 4165 4166 OSThread* osthread = thread->osthread(); 4167 4168 if (!osthread->interrupted()) { 4169 osthread->set_interrupted(true); 4170 // More than one thread can get here with the same value of osthread, 4171 // resulting in multiple notifications. We do, however, want the store 4172 // to interrupted() to be visible to other threads before we execute unpark(). 4173 OrderAccess::fence(); 4174 ParkEvent * const slp = thread->_SleepEvent ; 4175 if (slp != NULL) slp->unpark() ; 4176 } 4177 4178 // For JSR166. Unpark even if interrupt status already was set 4179 if (thread->is_Java_thread()) 4180 ((JavaThread*)thread)->parker()->unpark(); 4181 4182 ParkEvent * ev = thread->_ParkEvent ; 4183 if (ev != NULL) ev->unpark() ; 4184 4185} 4186 4187bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 4188 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 4189 "possibility of dangling Thread pointer"); 4190 4191 OSThread* osthread = thread->osthread(); 4192 4193 bool interrupted = osthread->interrupted(); 4194 4195 if (interrupted && clear_interrupted) { 4196 osthread->set_interrupted(false); 4197 // consider thread->_SleepEvent->reset() ... optional optimization 4198 } 4199 4200 return interrupted; 4201} 4202 4203/////////////////////////////////////////////////////////////////////////////////// 4204// signal handling (except suspend/resume) 4205 4206// This routine may be used by user applications as a "hook" to catch signals. 4207// The user-defined signal handler must pass unrecognized signals to this 4208// routine, and if it returns true (non-zero), then the signal handler must 4209// return immediately. If the flag "abort_if_unrecognized" is true, then this 4210// routine will never retun false (zero), but instead will execute a VM panic 4211// routine kill the process. 4212// 4213// If this routine returns false, it is OK to call it again. This allows 4214// the user-defined signal handler to perform checks either before or after 4215// the VM performs its own checks. Naturally, the user code would be making 4216// a serious error if it tried to handle an exception (such as a null check 4217// or breakpoint) that the VM was generating for its own correct operation. 4218// 4219// This routine may recognize any of the following kinds of signals: 4220// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4221// It should be consulted by handlers for any of those signals. 4222// 4223// The caller of this routine must pass in the three arguments supplied 4224// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4225// field of the structure passed to sigaction(). This routine assumes that 4226// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4227// 4228// Note that the VM will print warnings if it detects conflicting signal 4229// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4230// 4231extern "C" JNIEXPORT int 4232JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 4233 void* ucontext, int abort_if_unrecognized); 4234 4235void signalHandler(int sig, siginfo_t* info, void* uc) { 4236 assert(info != NULL && uc != NULL, "it must be old kernel"); 4237 int orig_errno = errno; // Preserve errno value over signal handler. 4238 JVM_handle_linux_signal(sig, info, uc, true); 4239 errno = orig_errno; 4240} 4241 4242 4243// This boolean allows users to forward their own non-matching signals 4244// to JVM_handle_linux_signal, harmlessly. 4245bool os::Linux::signal_handlers_are_installed = false; 4246 4247// For signal-chaining 4248struct sigaction os::Linux::sigact[MAXSIGNUM]; 4249unsigned int os::Linux::sigs = 0; 4250bool os::Linux::libjsig_is_loaded = false; 4251typedef struct sigaction *(*get_signal_t)(int); 4252get_signal_t os::Linux::get_signal_action = NULL; 4253 4254struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4255 struct sigaction *actp = NULL; 4256 4257 if (libjsig_is_loaded) { 4258 // Retrieve the old signal handler from libjsig 4259 actp = (*get_signal_action)(sig); 4260 } 4261 if (actp == NULL) { 4262 // Retrieve the preinstalled signal handler from jvm 4263 actp = get_preinstalled_handler(sig); 4264 } 4265 4266 return actp; 4267} 4268 4269static bool call_chained_handler(struct sigaction *actp, int sig, 4270 siginfo_t *siginfo, void *context) { 4271 // Call the old signal handler 4272 if (actp->sa_handler == SIG_DFL) { 4273 // It's more reasonable to let jvm treat it as an unexpected exception 4274 // instead of taking the default action. 4275 return false; 4276 } else if (actp->sa_handler != SIG_IGN) { 4277 if ((actp->sa_flags & SA_NODEFER) == 0) { 4278 // automaticlly block the signal 4279 sigaddset(&(actp->sa_mask), sig); 4280 } 4281 4282 sa_handler_t hand; 4283 sa_sigaction_t sa; 4284 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4285 // retrieve the chained handler 4286 if (siginfo_flag_set) { 4287 sa = actp->sa_sigaction; 4288 } else { 4289 hand = actp->sa_handler; 4290 } 4291 4292 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4293 actp->sa_handler = SIG_DFL; 4294 } 4295 4296 // try to honor the signal mask 4297 sigset_t oset; 4298 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4299 4300 // call into the chained handler 4301 if (siginfo_flag_set) { 4302 (*sa)(sig, siginfo, context); 4303 } else { 4304 (*hand)(sig); 4305 } 4306 4307 // restore the signal mask 4308 pthread_sigmask(SIG_SETMASK, &oset, 0); 4309 } 4310 // Tell jvm's signal handler the signal is taken care of. 4311 return true; 4312} 4313 4314bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4315 bool chained = false; 4316 // signal-chaining 4317 if (UseSignalChaining) { 4318 struct sigaction *actp = get_chained_signal_action(sig); 4319 if (actp != NULL) { 4320 chained = call_chained_handler(actp, sig, siginfo, context); 4321 } 4322 } 4323 return chained; 4324} 4325 4326struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4327 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 4328 return &sigact[sig]; 4329 } 4330 return NULL; 4331} 4332 4333void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4334 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4335 sigact[sig] = oldAct; 4336 sigs |= (unsigned int)1 << sig; 4337} 4338 4339// for diagnostic 4340int os::Linux::sigflags[MAXSIGNUM]; 4341 4342int os::Linux::get_our_sigflags(int sig) { 4343 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4344 return sigflags[sig]; 4345} 4346 4347void os::Linux::set_our_sigflags(int sig, int flags) { 4348 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4349 sigflags[sig] = flags; 4350} 4351 4352void os::Linux::set_signal_handler(int sig, bool set_installed) { 4353 // Check for overwrite. 4354 struct sigaction oldAct; 4355 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4356 4357 void* oldhand = oldAct.sa_sigaction 4358 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4359 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4360 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4361 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4362 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4363 if (AllowUserSignalHandlers || !set_installed) { 4364 // Do not overwrite; user takes responsibility to forward to us. 4365 return; 4366 } else if (UseSignalChaining) { 4367 // save the old handler in jvm 4368 save_preinstalled_handler(sig, oldAct); 4369 // libjsig also interposes the sigaction() call below and saves the 4370 // old sigaction on it own. 4371 } else { 4372 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4373 "%#lx for signal %d.", (long)oldhand, sig)); 4374 } 4375 } 4376 4377 struct sigaction sigAct; 4378 sigfillset(&(sigAct.sa_mask)); 4379 sigAct.sa_handler = SIG_DFL; 4380 if (!set_installed) { 4381 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4382 } else { 4383 sigAct.sa_sigaction = signalHandler; 4384 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4385 } 4386 // Save flags, which are set by ours 4387 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4388 sigflags[sig] = sigAct.sa_flags; 4389 4390 int ret = sigaction(sig, &sigAct, &oldAct); 4391 assert(ret == 0, "check"); 4392 4393 void* oldhand2 = oldAct.sa_sigaction 4394 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4395 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4396 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4397} 4398 4399// install signal handlers for signals that HotSpot needs to 4400// handle in order to support Java-level exception handling. 4401 4402void os::Linux::install_signal_handlers() { 4403 if (!signal_handlers_are_installed) { 4404 signal_handlers_are_installed = true; 4405 4406 // signal-chaining 4407 typedef void (*signal_setting_t)(); 4408 signal_setting_t begin_signal_setting = NULL; 4409 signal_setting_t end_signal_setting = NULL; 4410 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4411 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4412 if (begin_signal_setting != NULL) { 4413 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4414 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4415 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4416 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4417 libjsig_is_loaded = true; 4418 assert(UseSignalChaining, "should enable signal-chaining"); 4419 } 4420 if (libjsig_is_loaded) { 4421 // Tell libjsig jvm is setting signal handlers 4422 (*begin_signal_setting)(); 4423 } 4424 4425 set_signal_handler(SIGSEGV, true); 4426 set_signal_handler(SIGPIPE, true); 4427 set_signal_handler(SIGBUS, true); 4428 set_signal_handler(SIGILL, true); 4429 set_signal_handler(SIGFPE, true); 4430 set_signal_handler(SIGXFSZ, true); 4431 4432 if (libjsig_is_loaded) { 4433 // Tell libjsig jvm finishes setting signal handlers 4434 (*end_signal_setting)(); 4435 } 4436 4437 // We don't activate signal checker if libjsig is in place, we trust ourselves 4438 // and if UserSignalHandler is installed all bets are off. 4439 // Log that signal checking is off only if -verbose:jni is specified. 4440 if (CheckJNICalls) { 4441 if (libjsig_is_loaded) { 4442 if (PrintJNIResolving) { 4443 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4444 } 4445 check_signals = false; 4446 } 4447 if (AllowUserSignalHandlers) { 4448 if (PrintJNIResolving) { 4449 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4450 } 4451 check_signals = false; 4452 } 4453 } 4454 } 4455} 4456 4457// This is the fastest way to get thread cpu time on Linux. 4458// Returns cpu time (user+sys) for any thread, not only for current. 4459// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4460// It might work on 2.6.10+ with a special kernel/glibc patch. 4461// For reference, please, see IEEE Std 1003.1-2004: 4462// http://www.unix.org/single_unix_specification 4463 4464jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4465 struct timespec tp; 4466 int rc = os::Linux::clock_gettime(clockid, &tp); 4467 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4468 4469 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4470} 4471 4472///// 4473// glibc on Linux platform uses non-documented flag 4474// to indicate, that some special sort of signal 4475// trampoline is used. 4476// We will never set this flag, and we should 4477// ignore this flag in our diagnostic 4478#ifdef SIGNIFICANT_SIGNAL_MASK 4479#undef SIGNIFICANT_SIGNAL_MASK 4480#endif 4481#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4482 4483static const char* get_signal_handler_name(address handler, 4484 char* buf, int buflen) { 4485 int offset; 4486 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4487 if (found) { 4488 // skip directory names 4489 const char *p1, *p2; 4490 p1 = buf; 4491 size_t len = strlen(os::file_separator()); 4492 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4493 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4494 } else { 4495 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4496 } 4497 return buf; 4498} 4499 4500static void print_signal_handler(outputStream* st, int sig, 4501 char* buf, size_t buflen) { 4502 struct sigaction sa; 4503 4504 sigaction(sig, NULL, &sa); 4505 4506 // See comment for SIGNIFICANT_SIGNAL_MASK define 4507 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4508 4509 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4510 4511 address handler = (sa.sa_flags & SA_SIGINFO) 4512 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4513 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4514 4515 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4516 st->print("SIG_DFL"); 4517 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4518 st->print("SIG_IGN"); 4519 } else { 4520 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4521 } 4522 4523 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 4524 4525 address rh = VMError::get_resetted_sighandler(sig); 4526 // May be, handler was resetted by VMError? 4527 if(rh != NULL) { 4528 handler = rh; 4529 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4530 } 4531 4532 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 4533 4534 // Check: is it our handler? 4535 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4536 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4537 // It is our signal handler 4538 // check for flags, reset system-used one! 4539 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4540 st->print( 4541 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4542 os::Linux::get_our_sigflags(sig)); 4543 } 4544 } 4545 st->cr(); 4546} 4547 4548 4549#define DO_SIGNAL_CHECK(sig) \ 4550 if (!sigismember(&check_signal_done, sig)) \ 4551 os::Linux::check_signal_handler(sig) 4552 4553// This method is a periodic task to check for misbehaving JNI applications 4554// under CheckJNI, we can add any periodic checks here 4555 4556void os::run_periodic_checks() { 4557 4558 if (check_signals == false) return; 4559 4560 // SEGV and BUS if overridden could potentially prevent 4561 // generation of hs*.log in the event of a crash, debugging 4562 // such a case can be very challenging, so we absolutely 4563 // check the following for a good measure: 4564 DO_SIGNAL_CHECK(SIGSEGV); 4565 DO_SIGNAL_CHECK(SIGILL); 4566 DO_SIGNAL_CHECK(SIGFPE); 4567 DO_SIGNAL_CHECK(SIGBUS); 4568 DO_SIGNAL_CHECK(SIGPIPE); 4569 DO_SIGNAL_CHECK(SIGXFSZ); 4570 4571 4572 // ReduceSignalUsage allows the user to override these handlers 4573 // see comments at the very top and jvm_solaris.h 4574 if (!ReduceSignalUsage) { 4575 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4576 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4577 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4578 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4579 } 4580 4581 DO_SIGNAL_CHECK(SR_signum); 4582 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4583} 4584 4585typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4586 4587static os_sigaction_t os_sigaction = NULL; 4588 4589void os::Linux::check_signal_handler(int sig) { 4590 char buf[O_BUFLEN]; 4591 address jvmHandler = NULL; 4592 4593 4594 struct sigaction act; 4595 if (os_sigaction == NULL) { 4596 // only trust the default sigaction, in case it has been interposed 4597 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4598 if (os_sigaction == NULL) return; 4599 } 4600 4601 os_sigaction(sig, (struct sigaction*)NULL, &act); 4602 4603 4604 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4605 4606 address thisHandler = (act.sa_flags & SA_SIGINFO) 4607 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4608 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 4609 4610 4611 switch(sig) { 4612 case SIGSEGV: 4613 case SIGBUS: 4614 case SIGFPE: 4615 case SIGPIPE: 4616 case SIGILL: 4617 case SIGXFSZ: 4618 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4619 break; 4620 4621 case SHUTDOWN1_SIGNAL: 4622 case SHUTDOWN2_SIGNAL: 4623 case SHUTDOWN3_SIGNAL: 4624 case BREAK_SIGNAL: 4625 jvmHandler = (address)user_handler(); 4626 break; 4627 4628 case INTERRUPT_SIGNAL: 4629 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4630 break; 4631 4632 default: 4633 if (sig == SR_signum) { 4634 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4635 } else { 4636 return; 4637 } 4638 break; 4639 } 4640 4641 if (thisHandler != jvmHandler) { 4642 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4643 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4644 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4645 // No need to check this sig any longer 4646 sigaddset(&check_signal_done, sig); 4647 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4648 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4649 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4650 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4651 // No need to check this sig any longer 4652 sigaddset(&check_signal_done, sig); 4653 } 4654 4655 // Dump all the signal 4656 if (sigismember(&check_signal_done, sig)) { 4657 print_signal_handlers(tty, buf, O_BUFLEN); 4658 } 4659} 4660 4661extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 4662 4663extern bool signal_name(int signo, char* buf, size_t len); 4664 4665const char* os::exception_name(int exception_code, char* buf, size_t size) { 4666 if (0 < exception_code && exception_code <= SIGRTMAX) { 4667 // signal 4668 if (!signal_name(exception_code, buf, size)) { 4669 jio_snprintf(buf, size, "SIG%d", exception_code); 4670 } 4671 return buf; 4672 } else { 4673 return NULL; 4674 } 4675} 4676 4677// this is called _before_ the most of global arguments have been parsed 4678void os::init(void) { 4679 char dummy; /* used to get a guess on initial stack address */ 4680// first_hrtime = gethrtime(); 4681 4682 // With LinuxThreads the JavaMain thread pid (primordial thread) 4683 // is different than the pid of the java launcher thread. 4684 // So, on Linux, the launcher thread pid is passed to the VM 4685 // via the sun.java.launcher.pid property. 4686 // Use this property instead of getpid() if it was correctly passed. 4687 // See bug 6351349. 4688 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4689 4690 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4691 4692 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4693 4694 init_random(1234567); 4695 4696 ThreadCritical::initialize(); 4697 4698 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4699 if (Linux::page_size() == -1) { 4700 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4701 strerror(errno))); 4702 } 4703 init_page_sizes((size_t) Linux::page_size()); 4704 4705 Linux::initialize_system_info(); 4706 4707 // main_thread points to the aboriginal thread 4708 Linux::_main_thread = pthread_self(); 4709 4710 Linux::clock_init(); 4711 initial_time_count = os::elapsed_counter(); 4712 pthread_mutex_init(&dl_mutex, NULL); 4713 4714 // If the pagesize of the VM is greater than 8K determine the appropriate 4715 // number of initial guard pages. The user can change this with the 4716 // command line arguments, if needed. 4717 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4718 StackYellowPages = 1; 4719 StackRedPages = 1; 4720 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4721 } 4722} 4723 4724// To install functions for atexit system call 4725extern "C" { 4726 static void perfMemory_exit_helper() { 4727 perfMemory_exit(); 4728 } 4729} 4730 4731// this is called _after_ the global arguments have been parsed 4732jint os::init_2(void) 4733{ 4734 Linux::fast_thread_clock_init(); 4735 4736 // Allocate a single page and mark it as readable for safepoint polling 4737 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4738 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 4739 4740 os::set_polling_page( polling_page ); 4741 4742#ifndef PRODUCT 4743 if(Verbose && PrintMiscellaneous) 4744 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 4745#endif 4746 4747 if (!UseMembar) { 4748 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4749 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4750 os::set_memory_serialize_page( mem_serialize_page ); 4751 4752#ifndef PRODUCT 4753 if(Verbose && PrintMiscellaneous) 4754 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 4755#endif 4756 } 4757 4758 os::large_page_init(); 4759 4760 // initialize suspend/resume support - must do this before signal_sets_init() 4761 if (SR_initialize() != 0) { 4762 perror("SR_initialize failed"); 4763 return JNI_ERR; 4764 } 4765 4766 Linux::signal_sets_init(); 4767 Linux::install_signal_handlers(); 4768 4769 // Check minimum allowable stack size for thread creation and to initialize 4770 // the java system classes, including StackOverflowError - depends on page 4771 // size. Add a page for compiler2 recursion in main thread. 4772 // Add in 2*BytesPerWord times page size to account for VM stack during 4773 // class initialization depending on 32 or 64 bit VM. 4774 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4775 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4776 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4777 4778 size_t threadStackSizeInBytes = ThreadStackSize * K; 4779 if (threadStackSizeInBytes != 0 && 4780 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4781 tty->print_cr("\nThe stack size specified is too small, " 4782 "Specify at least %dk", 4783 os::Linux::min_stack_allowed/ K); 4784 return JNI_ERR; 4785 } 4786 4787 // Make the stack size a multiple of the page size so that 4788 // the yellow/red zones can be guarded. 4789 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4790 vm_page_size())); 4791 4792 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4793 4794 Linux::libpthread_init(); 4795 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4796 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4797 Linux::glibc_version(), Linux::libpthread_version(), 4798 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4799 } 4800 4801 if (UseNUMA) { 4802 if (!Linux::libnuma_init()) { 4803 UseNUMA = false; 4804 } else { 4805 if ((Linux::numa_max_node() < 1)) { 4806 // There's only one node(they start from 0), disable NUMA. 4807 UseNUMA = false; 4808 } 4809 } 4810 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4811 // we can make the adaptive lgrp chunk resizing work. If the user specified 4812 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4813 // disable adaptive resizing. 4814 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4815 if (FLAG_IS_DEFAULT(UseNUMA)) { 4816 UseNUMA = false; 4817 } else { 4818 if (FLAG_IS_DEFAULT(UseLargePages) && 4819 FLAG_IS_DEFAULT(UseSHM) && 4820 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4821 UseLargePages = false; 4822 } else { 4823 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing"); 4824 UseAdaptiveSizePolicy = false; 4825 UseAdaptiveNUMAChunkSizing = false; 4826 } 4827 } 4828 } 4829 if (!UseNUMA && ForceNUMA) { 4830 UseNUMA = true; 4831 } 4832 } 4833 4834 if (MaxFDLimit) { 4835 // set the number of file descriptors to max. print out error 4836 // if getrlimit/setrlimit fails but continue regardless. 4837 struct rlimit nbr_files; 4838 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4839 if (status != 0) { 4840 if (PrintMiscellaneous && (Verbose || WizardMode)) 4841 perror("os::init_2 getrlimit failed"); 4842 } else { 4843 nbr_files.rlim_cur = nbr_files.rlim_max; 4844 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4845 if (status != 0) { 4846 if (PrintMiscellaneous && (Verbose || WizardMode)) 4847 perror("os::init_2 setrlimit failed"); 4848 } 4849 } 4850 } 4851 4852 // Initialize lock used to serialize thread creation (see os::create_thread) 4853 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4854 4855 // at-exit methods are called in the reverse order of their registration. 4856 // atexit functions are called on return from main or as a result of a 4857 // call to exit(3C). There can be only 32 of these functions registered 4858 // and atexit() does not set errno. 4859 4860 if (PerfAllowAtExitRegistration) { 4861 // only register atexit functions if PerfAllowAtExitRegistration is set. 4862 // atexit functions can be delayed until process exit time, which 4863 // can be problematic for embedded VM situations. Embedded VMs should 4864 // call DestroyJavaVM() to assure that VM resources are released. 4865 4866 // note: perfMemory_exit_helper atexit function may be removed in 4867 // the future if the appropriate cleanup code can be added to the 4868 // VM_Exit VMOperation's doit method. 4869 if (atexit(perfMemory_exit_helper) != 0) { 4870 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 4871 } 4872 } 4873 4874 // initialize thread priority policy 4875 prio_init(); 4876 4877 return JNI_OK; 4878} 4879 4880// this is called at the end of vm_initialization 4881void os::init_3(void) 4882{ 4883#ifdef JAVASE_EMBEDDED 4884 // Start the MemNotifyThread 4885 if (LowMemoryProtection) { 4886 MemNotifyThread::start(); 4887 } 4888 return; 4889#endif 4890} 4891 4892// Mark the polling page as unreadable 4893void os::make_polling_page_unreadable(void) { 4894 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4895 fatal("Could not disable polling page"); 4896}; 4897 4898// Mark the polling page as readable 4899void os::make_polling_page_readable(void) { 4900 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4901 fatal("Could not enable polling page"); 4902 } 4903}; 4904 4905int os::active_processor_count() { 4906 // Linux doesn't yet have a (official) notion of processor sets, 4907 // so just return the number of online processors. 4908 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4909 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4910 return online_cpus; 4911} 4912 4913void os::set_native_thread_name(const char *name) { 4914 // Not yet implemented. 4915 return; 4916} 4917 4918bool os::distribute_processes(uint length, uint* distribution) { 4919 // Not yet implemented. 4920 return false; 4921} 4922 4923bool os::bind_to_processor(uint processor_id) { 4924 // Not yet implemented. 4925 return false; 4926} 4927 4928/// 4929 4930void os::SuspendedThreadTask::internal_do_task() { 4931 if (do_suspend(_thread->osthread())) { 4932 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4933 do_task(context); 4934 do_resume(_thread->osthread()); 4935 } 4936} 4937 4938class PcFetcher : public os::SuspendedThreadTask { 4939public: 4940 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4941 ExtendedPC result(); 4942protected: 4943 void do_task(const os::SuspendedThreadTaskContext& context); 4944private: 4945 ExtendedPC _epc; 4946}; 4947 4948ExtendedPC PcFetcher::result() { 4949 guarantee(is_done(), "task is not done yet."); 4950 return _epc; 4951} 4952 4953void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4954 Thread* thread = context.thread(); 4955 OSThread* osthread = thread->osthread(); 4956 if (osthread->ucontext() != NULL) { 4957 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4958 } else { 4959 // NULL context is unexpected, double-check this is the VMThread 4960 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4961 } 4962} 4963 4964// Suspends the target using the signal mechanism and then grabs the PC before 4965// resuming the target. Used by the flat-profiler only 4966ExtendedPC os::get_thread_pc(Thread* thread) { 4967 // Make sure that it is called by the watcher for the VMThread 4968 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4969 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4970 4971 PcFetcher fetcher(thread); 4972 fetcher.run(); 4973 return fetcher.result(); 4974} 4975 4976int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4977{ 4978 if (is_NPTL()) { 4979 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4980 } else { 4981 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4982 // word back to default 64bit precision if condvar is signaled. Java 4983 // wants 53bit precision. Save and restore current value. 4984 int fpu = get_fpu_control_word(); 4985 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4986 set_fpu_control_word(fpu); 4987 return status; 4988 } 4989} 4990 4991//////////////////////////////////////////////////////////////////////////////// 4992// debug support 4993 4994bool os::find(address addr, outputStream* st) { 4995 Dl_info dlinfo; 4996 memset(&dlinfo, 0, sizeof(dlinfo)); 4997 if (dladdr(addr, &dlinfo) != 0) { 4998 st->print(PTR_FORMAT ": ", addr); 4999 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5000 st->print("%s+%#x", dlinfo.dli_sname, 5001 addr - (intptr_t)dlinfo.dli_saddr); 5002 } else if (dlinfo.dli_fbase != NULL) { 5003 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 5004 } else { 5005 st->print("<absolute address>"); 5006 } 5007 if (dlinfo.dli_fname != NULL) { 5008 st->print(" in %s", dlinfo.dli_fname); 5009 } 5010 if (dlinfo.dli_fbase != NULL) { 5011 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 5012 } 5013 st->cr(); 5014 5015 if (Verbose) { 5016 // decode some bytes around the PC 5017 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5018 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5019 address lowest = (address) dlinfo.dli_sname; 5020 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5021 if (begin < lowest) begin = lowest; 5022 Dl_info dlinfo2; 5023 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5024 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 5025 end = (address) dlinfo2.dli_saddr; 5026 Disassembler::decode(begin, end, st); 5027 } 5028 return true; 5029 } 5030 return false; 5031} 5032 5033//////////////////////////////////////////////////////////////////////////////// 5034// misc 5035 5036// This does not do anything on Linux. This is basically a hook for being 5037// able to use structured exception handling (thread-local exception filters) 5038// on, e.g., Win32. 5039void 5040os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 5041 JavaCallArguments* args, Thread* thread) { 5042 f(value, method, args, thread); 5043} 5044 5045void os::print_statistics() { 5046} 5047 5048int os::message_box(const char* title, const char* message) { 5049 int i; 5050 fdStream err(defaultStream::error_fd()); 5051 for (i = 0; i < 78; i++) err.print_raw("="); 5052 err.cr(); 5053 err.print_raw_cr(title); 5054 for (i = 0; i < 78; i++) err.print_raw("-"); 5055 err.cr(); 5056 err.print_raw_cr(message); 5057 for (i = 0; i < 78; i++) err.print_raw("="); 5058 err.cr(); 5059 5060 char buf[16]; 5061 // Prevent process from exiting upon "read error" without consuming all CPU 5062 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5063 5064 return buf[0] == 'y' || buf[0] == 'Y'; 5065} 5066 5067int os::stat(const char *path, struct stat *sbuf) { 5068 char pathbuf[MAX_PATH]; 5069 if (strlen(path) > MAX_PATH - 1) { 5070 errno = ENAMETOOLONG; 5071 return -1; 5072 } 5073 os::native_path(strcpy(pathbuf, path)); 5074 return ::stat(pathbuf, sbuf); 5075} 5076 5077bool os::check_heap(bool force) { 5078 return true; 5079} 5080 5081int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 5082 return ::vsnprintf(buf, count, format, args); 5083} 5084 5085// Is a (classpath) directory empty? 5086bool os::dir_is_empty(const char* path) { 5087 DIR *dir = NULL; 5088 struct dirent *ptr; 5089 5090 dir = opendir(path); 5091 if (dir == NULL) return true; 5092 5093 /* Scan the directory */ 5094 bool result = true; 5095 char buf[sizeof(struct dirent) + MAX_PATH]; 5096 while (result && (ptr = ::readdir(dir)) != NULL) { 5097 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5098 result = false; 5099 } 5100 } 5101 closedir(dir); 5102 return result; 5103} 5104 5105// This code originates from JDK's sysOpen and open64_w 5106// from src/solaris/hpi/src/system_md.c 5107 5108#ifndef O_DELETE 5109#define O_DELETE 0x10000 5110#endif 5111 5112// Open a file. Unlink the file immediately after open returns 5113// if the specified oflag has the O_DELETE flag set. 5114// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 5115 5116int os::open(const char *path, int oflag, int mode) { 5117 5118 if (strlen(path) > MAX_PATH - 1) { 5119 errno = ENAMETOOLONG; 5120 return -1; 5121 } 5122 int fd; 5123 int o_delete = (oflag & O_DELETE); 5124 oflag = oflag & ~O_DELETE; 5125 5126 fd = ::open64(path, oflag, mode); 5127 if (fd == -1) return -1; 5128 5129 //If the open succeeded, the file might still be a directory 5130 { 5131 struct stat64 buf64; 5132 int ret = ::fstat64(fd, &buf64); 5133 int st_mode = buf64.st_mode; 5134 5135 if (ret != -1) { 5136 if ((st_mode & S_IFMT) == S_IFDIR) { 5137 errno = EISDIR; 5138 ::close(fd); 5139 return -1; 5140 } 5141 } else { 5142 ::close(fd); 5143 return -1; 5144 } 5145 } 5146 5147 /* 5148 * All file descriptors that are opened in the JVM and not 5149 * specifically destined for a subprocess should have the 5150 * close-on-exec flag set. If we don't set it, then careless 3rd 5151 * party native code might fork and exec without closing all 5152 * appropriate file descriptors (e.g. as we do in closeDescriptors in 5153 * UNIXProcess.c), and this in turn might: 5154 * 5155 * - cause end-of-file to fail to be detected on some file 5156 * descriptors, resulting in mysterious hangs, or 5157 * 5158 * - might cause an fopen in the subprocess to fail on a system 5159 * suffering from bug 1085341. 5160 * 5161 * (Yes, the default setting of the close-on-exec flag is a Unix 5162 * design flaw) 5163 * 5164 * See: 5165 * 1085341: 32-bit stdio routines should support file descriptors >255 5166 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5167 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5168 */ 5169#ifdef FD_CLOEXEC 5170 { 5171 int flags = ::fcntl(fd, F_GETFD); 5172 if (flags != -1) 5173 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5174 } 5175#endif 5176 5177 if (o_delete != 0) { 5178 ::unlink(path); 5179 } 5180 return fd; 5181} 5182 5183 5184// create binary file, rewriting existing file if required 5185int os::create_binary_file(const char* path, bool rewrite_existing) { 5186 int oflags = O_WRONLY | O_CREAT; 5187 if (!rewrite_existing) { 5188 oflags |= O_EXCL; 5189 } 5190 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5191} 5192 5193// return current position of file pointer 5194jlong os::current_file_offset(int fd) { 5195 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5196} 5197 5198// move file pointer to the specified offset 5199jlong os::seek_to_file_offset(int fd, jlong offset) { 5200 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5201} 5202 5203// This code originates from JDK's sysAvailable 5204// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5205 5206int os::available(int fd, jlong *bytes) { 5207 jlong cur, end; 5208 int mode; 5209 struct stat64 buf64; 5210 5211 if (::fstat64(fd, &buf64) >= 0) { 5212 mode = buf64.st_mode; 5213 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5214 /* 5215 * XXX: is the following call interruptible? If so, this might 5216 * need to go through the INTERRUPT_IO() wrapper as for other 5217 * blocking, interruptible calls in this file. 5218 */ 5219 int n; 5220 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5221 *bytes = n; 5222 return 1; 5223 } 5224 } 5225 } 5226 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5227 return 0; 5228 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5229 return 0; 5230 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5231 return 0; 5232 } 5233 *bytes = end - cur; 5234 return 1; 5235} 5236 5237int os::socket_available(int fd, jint *pbytes) { 5238 // Linux doc says EINTR not returned, unlike Solaris 5239 int ret = ::ioctl(fd, FIONREAD, pbytes); 5240 5241 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 5242 // is expected to return 0 on failure and 1 on success to the jdk. 5243 return (ret < 0) ? 0 : 1; 5244} 5245 5246// Map a block of memory. 5247char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5248 char *addr, size_t bytes, bool read_only, 5249 bool allow_exec) { 5250 int prot; 5251 int flags = MAP_PRIVATE; 5252 5253 if (read_only) { 5254 prot = PROT_READ; 5255 } else { 5256 prot = PROT_READ | PROT_WRITE; 5257 } 5258 5259 if (allow_exec) { 5260 prot |= PROT_EXEC; 5261 } 5262 5263 if (addr != NULL) { 5264 flags |= MAP_FIXED; 5265 } 5266 5267 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5268 fd, file_offset); 5269 if (mapped_address == MAP_FAILED) { 5270 return NULL; 5271 } 5272 return mapped_address; 5273} 5274 5275 5276// Remap a block of memory. 5277char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5278 char *addr, size_t bytes, bool read_only, 5279 bool allow_exec) { 5280 // same as map_memory() on this OS 5281 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5282 allow_exec); 5283} 5284 5285 5286// Unmap a block of memory. 5287bool os::pd_unmap_memory(char* addr, size_t bytes) { 5288 return munmap(addr, bytes) == 0; 5289} 5290 5291static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5292 5293static clockid_t thread_cpu_clockid(Thread* thread) { 5294 pthread_t tid = thread->osthread()->pthread_id(); 5295 clockid_t clockid; 5296 5297 // Get thread clockid 5298 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5299 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5300 return clockid; 5301} 5302 5303// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5304// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5305// of a thread. 5306// 5307// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5308// the fast estimate available on the platform. 5309 5310jlong os::current_thread_cpu_time() { 5311 if (os::Linux::supports_fast_thread_cpu_time()) { 5312 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5313 } else { 5314 // return user + sys since the cost is the same 5315 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5316 } 5317} 5318 5319jlong os::thread_cpu_time(Thread* thread) { 5320 // consistent with what current_thread_cpu_time() returns 5321 if (os::Linux::supports_fast_thread_cpu_time()) { 5322 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5323 } else { 5324 return slow_thread_cpu_time(thread, true /* user + sys */); 5325 } 5326} 5327 5328jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5329 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5330 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5331 } else { 5332 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5333 } 5334} 5335 5336jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5337 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5338 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5339 } else { 5340 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5341 } 5342} 5343 5344// 5345// -1 on error. 5346// 5347 5348static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5349 static bool proc_task_unchecked = true; 5350 static const char *proc_stat_path = "/proc/%d/stat"; 5351 pid_t tid = thread->osthread()->thread_id(); 5352 char *s; 5353 char stat[2048]; 5354 int statlen; 5355 char proc_name[64]; 5356 int count; 5357 long sys_time, user_time; 5358 char cdummy; 5359 int idummy; 5360 long ldummy; 5361 FILE *fp; 5362 5363 // The /proc/<tid>/stat aggregates per-process usage on 5364 // new Linux kernels 2.6+ where NPTL is supported. 5365 // The /proc/self/task/<tid>/stat still has the per-thread usage. 5366 // See bug 6328462. 5367 // There possibly can be cases where there is no directory 5368 // /proc/self/task, so we check its availability. 5369 if (proc_task_unchecked && os::Linux::is_NPTL()) { 5370 // This is executed only once 5371 proc_task_unchecked = false; 5372 fp = fopen("/proc/self/task", "r"); 5373 if (fp != NULL) { 5374 proc_stat_path = "/proc/self/task/%d/stat"; 5375 fclose(fp); 5376 } 5377 } 5378 5379 sprintf(proc_name, proc_stat_path, tid); 5380 fp = fopen(proc_name, "r"); 5381 if ( fp == NULL ) return -1; 5382 statlen = fread(stat, 1, 2047, fp); 5383 stat[statlen] = '\0'; 5384 fclose(fp); 5385 5386 // Skip pid and the command string. Note that we could be dealing with 5387 // weird command names, e.g. user could decide to rename java launcher 5388 // to "java 1.4.2 :)", then the stat file would look like 5389 // 1234 (java 1.4.2 :)) R ... ... 5390 // We don't really need to know the command string, just find the last 5391 // occurrence of ")" and then start parsing from there. See bug 4726580. 5392 s = strrchr(stat, ')'); 5393 if (s == NULL ) return -1; 5394 5395 // Skip blank chars 5396 do s++; while (isspace(*s)); 5397 5398 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5399 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5400 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5401 &user_time, &sys_time); 5402 if ( count != 13 ) return -1; 5403 if (user_sys_cpu_time) { 5404 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5405 } else { 5406 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5407 } 5408} 5409 5410void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5411 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5412 info_ptr->may_skip_backward = false; // elapsed time not wall time 5413 info_ptr->may_skip_forward = false; // elapsed time not wall time 5414 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5415} 5416 5417void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5418 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5419 info_ptr->may_skip_backward = false; // elapsed time not wall time 5420 info_ptr->may_skip_forward = false; // elapsed time not wall time 5421 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5422} 5423 5424bool os::is_thread_cpu_time_supported() { 5425 return true; 5426} 5427 5428// System loadavg support. Returns -1 if load average cannot be obtained. 5429// Linux doesn't yet have a (official) notion of processor sets, 5430// so just return the system wide load average. 5431int os::loadavg(double loadavg[], int nelem) { 5432 return ::getloadavg(loadavg, nelem); 5433} 5434 5435void os::pause() { 5436 char filename[MAX_PATH]; 5437 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5438 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5439 } else { 5440 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5441 } 5442 5443 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5444 if (fd != -1) { 5445 struct stat buf; 5446 ::close(fd); 5447 while (::stat(filename, &buf) == 0) { 5448 (void)::poll(NULL, 0, 100); 5449 } 5450 } else { 5451 jio_fprintf(stderr, 5452 "Could not open pause file '%s', continuing immediately.\n", filename); 5453 } 5454} 5455 5456 5457// Refer to the comments in os_solaris.cpp park-unpark. 5458// 5459// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5460// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5461// For specifics regarding the bug see GLIBC BUGID 261237 : 5462// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5463// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5464// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5465// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5466// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5467// and monitorenter when we're using 1-0 locking. All those operations may result in 5468// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5469// of libpthread avoids the problem, but isn't practical. 5470// 5471// Possible remedies: 5472// 5473// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5474// This is palliative and probabilistic, however. If the thread is preempted 5475// between the call to compute_abstime() and pthread_cond_timedwait(), more 5476// than the minimum period may have passed, and the abstime may be stale (in the 5477// past) resultin in a hang. Using this technique reduces the odds of a hang 5478// but the JVM is still vulnerable, particularly on heavily loaded systems. 5479// 5480// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5481// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5482// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5483// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5484// thread. 5485// 5486// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5487// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5488// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5489// This also works well. In fact it avoids kernel-level scalability impediments 5490// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5491// timers in a graceful fashion. 5492// 5493// 4. When the abstime value is in the past it appears that control returns 5494// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5495// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5496// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5497// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5498// It may be possible to avoid reinitialization by checking the return 5499// value from pthread_cond_timedwait(). In addition to reinitializing the 5500// condvar we must establish the invariant that cond_signal() is only called 5501// within critical sections protected by the adjunct mutex. This prevents 5502// cond_signal() from "seeing" a condvar that's in the midst of being 5503// reinitialized or that is corrupt. Sadly, this invariant obviates the 5504// desirable signal-after-unlock optimization that avoids futile context switching. 5505// 5506// I'm also concerned that some versions of NTPL might allocate an auxilliary 5507// structure when a condvar is used or initialized. cond_destroy() would 5508// release the helper structure. Our reinitialize-after-timedwait fix 5509// put excessive stress on malloc/free and locks protecting the c-heap. 5510// 5511// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5512// It may be possible to refine (4) by checking the kernel and NTPL verisons 5513// and only enabling the work-around for vulnerable environments. 5514 5515// utility to compute the abstime argument to timedwait: 5516// millis is the relative timeout time 5517// abstime will be the absolute timeout time 5518// TODO: replace compute_abstime() with unpackTime() 5519 5520static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5521 if (millis < 0) millis = 0; 5522 struct timeval now; 5523 int status = gettimeofday(&now, NULL); 5524 assert(status == 0, "gettimeofday"); 5525 jlong seconds = millis / 1000; 5526 millis %= 1000; 5527 if (seconds > 50000000) { // see man cond_timedwait(3T) 5528 seconds = 50000000; 5529 } 5530 abstime->tv_sec = now.tv_sec + seconds; 5531 long usec = now.tv_usec + millis * 1000; 5532 if (usec >= 1000000) { 5533 abstime->tv_sec += 1; 5534 usec -= 1000000; 5535 } 5536 abstime->tv_nsec = usec * 1000; 5537 return abstime; 5538} 5539 5540 5541// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 5542// Conceptually TryPark() should be equivalent to park(0). 5543 5544int os::PlatformEvent::TryPark() { 5545 for (;;) { 5546 const int v = _Event ; 5547 guarantee ((v == 0) || (v == 1), "invariant") ; 5548 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 5549 } 5550} 5551 5552void os::PlatformEvent::park() { // AKA "down()" 5553 // Invariant: Only the thread associated with the Event/PlatformEvent 5554 // may call park(). 5555 // TODO: assert that _Assoc != NULL or _Assoc == Self 5556 int v ; 5557 for (;;) { 5558 v = _Event ; 5559 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5560 } 5561 guarantee (v >= 0, "invariant") ; 5562 if (v == 0) { 5563 // Do this the hard way by blocking ... 5564 int status = pthread_mutex_lock(_mutex); 5565 assert_status(status == 0, status, "mutex_lock"); 5566 guarantee (_nParked == 0, "invariant") ; 5567 ++ _nParked ; 5568 while (_Event < 0) { 5569 status = pthread_cond_wait(_cond, _mutex); 5570 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5571 // Treat this the same as if the wait was interrupted 5572 if (status == ETIME) { status = EINTR; } 5573 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5574 } 5575 -- _nParked ; 5576 5577 _Event = 0 ; 5578 status = pthread_mutex_unlock(_mutex); 5579 assert_status(status == 0, status, "mutex_unlock"); 5580 // Paranoia to ensure our locked and lock-free paths interact 5581 // correctly with each other. 5582 OrderAccess::fence(); 5583 } 5584 guarantee (_Event >= 0, "invariant") ; 5585} 5586 5587int os::PlatformEvent::park(jlong millis) { 5588 guarantee (_nParked == 0, "invariant") ; 5589 5590 int v ; 5591 for (;;) { 5592 v = _Event ; 5593 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5594 } 5595 guarantee (v >= 0, "invariant") ; 5596 if (v != 0) return OS_OK ; 5597 5598 // We do this the hard way, by blocking the thread. 5599 // Consider enforcing a minimum timeout value. 5600 struct timespec abst; 5601 compute_abstime(&abst, millis); 5602 5603 int ret = OS_TIMEOUT; 5604 int status = pthread_mutex_lock(_mutex); 5605 assert_status(status == 0, status, "mutex_lock"); 5606 guarantee (_nParked == 0, "invariant") ; 5607 ++_nParked ; 5608 5609 // Object.wait(timo) will return because of 5610 // (a) notification 5611 // (b) timeout 5612 // (c) thread.interrupt 5613 // 5614 // Thread.interrupt and object.notify{All} both call Event::set. 5615 // That is, we treat thread.interrupt as a special case of notification. 5616 // The underlying Solaris implementation, cond_timedwait, admits 5617 // spurious/premature wakeups, but the JLS/JVM spec prevents the 5618 // JVM from making those visible to Java code. As such, we must 5619 // filter out spurious wakeups. We assume all ETIME returns are valid. 5620 // 5621 // TODO: properly differentiate simultaneous notify+interrupt. 5622 // In that case, we should propagate the notify to another waiter. 5623 5624 while (_Event < 0) { 5625 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5626 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5627 pthread_cond_destroy (_cond); 5628 pthread_cond_init (_cond, NULL) ; 5629 } 5630 assert_status(status == 0 || status == EINTR || 5631 status == ETIME || status == ETIMEDOUT, 5632 status, "cond_timedwait"); 5633 if (!FilterSpuriousWakeups) break ; // previous semantics 5634 if (status == ETIME || status == ETIMEDOUT) break ; 5635 // We consume and ignore EINTR and spurious wakeups. 5636 } 5637 --_nParked ; 5638 if (_Event >= 0) { 5639 ret = OS_OK; 5640 } 5641 _Event = 0 ; 5642 status = pthread_mutex_unlock(_mutex); 5643 assert_status(status == 0, status, "mutex_unlock"); 5644 assert (_nParked == 0, "invariant") ; 5645 // Paranoia to ensure our locked and lock-free paths interact 5646 // correctly with each other. 5647 OrderAccess::fence(); 5648 return ret; 5649} 5650 5651void os::PlatformEvent::unpark() { 5652 // Transitions for _Event: 5653 // 0 :=> 1 5654 // 1 :=> 1 5655 // -1 :=> either 0 or 1; must signal target thread 5656 // That is, we can safely transition _Event from -1 to either 5657 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back 5658 // unpark() calls. 5659 // See also: "Semaphores in Plan 9" by Mullender & Cox 5660 // 5661 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5662 // that it will take two back-to-back park() calls for the owning 5663 // thread to block. This has the benefit of forcing a spurious return 5664 // from the first park() call after an unpark() call which will help 5665 // shake out uses of park() and unpark() without condition variables. 5666 5667 if (Atomic::xchg(1, &_Event) >= 0) return; 5668 5669 // Wait for the thread associated with the event to vacate 5670 int status = pthread_mutex_lock(_mutex); 5671 assert_status(status == 0, status, "mutex_lock"); 5672 int AnyWaiters = _nParked; 5673 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5674 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5675 AnyWaiters = 0; 5676 pthread_cond_signal(_cond); 5677 } 5678 status = pthread_mutex_unlock(_mutex); 5679 assert_status(status == 0, status, "mutex_unlock"); 5680 if (AnyWaiters != 0) { 5681 status = pthread_cond_signal(_cond); 5682 assert_status(status == 0, status, "cond_signal"); 5683 } 5684 5685 // Note that we signal() _after dropping the lock for "immortal" Events. 5686 // This is safe and avoids a common class of futile wakeups. In rare 5687 // circumstances this can cause a thread to return prematurely from 5688 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5689 // simply re-test the condition and re-park itself. 5690} 5691 5692 5693// JSR166 5694// ------------------------------------------------------- 5695 5696/* 5697 * The solaris and linux implementations of park/unpark are fairly 5698 * conservative for now, but can be improved. They currently use a 5699 * mutex/condvar pair, plus a a count. 5700 * Park decrements count if > 0, else does a condvar wait. Unpark 5701 * sets count to 1 and signals condvar. Only one thread ever waits 5702 * on the condvar. Contention seen when trying to park implies that someone 5703 * is unparking you, so don't wait. And spurious returns are fine, so there 5704 * is no need to track notifications. 5705 */ 5706 5707#define MAX_SECS 100000000 5708/* 5709 * This code is common to linux and solaris and will be moved to a 5710 * common place in dolphin. 5711 * 5712 * The passed in time value is either a relative time in nanoseconds 5713 * or an absolute time in milliseconds. Either way it has to be unpacked 5714 * into suitable seconds and nanoseconds components and stored in the 5715 * given timespec structure. 5716 * Given time is a 64-bit value and the time_t used in the timespec is only 5717 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 5718 * overflow if times way in the future are given. Further on Solaris versions 5719 * prior to 10 there is a restriction (see cond_timedwait) that the specified 5720 * number of seconds, in abstime, is less than current_time + 100,000,000. 5721 * As it will be 28 years before "now + 100000000" will overflow we can 5722 * ignore overflow and just impose a hard-limit on seconds using the value 5723 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 5724 * years from "now". 5725 */ 5726 5727static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5728 assert (time > 0, "convertTime"); 5729 5730 struct timeval now; 5731 int status = gettimeofday(&now, NULL); 5732 assert(status == 0, "gettimeofday"); 5733 5734 time_t max_secs = now.tv_sec + MAX_SECS; 5735 5736 if (isAbsolute) { 5737 jlong secs = time / 1000; 5738 if (secs > max_secs) { 5739 absTime->tv_sec = max_secs; 5740 } 5741 else { 5742 absTime->tv_sec = secs; 5743 } 5744 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5745 } 5746 else { 5747 jlong secs = time / NANOSECS_PER_SEC; 5748 if (secs >= MAX_SECS) { 5749 absTime->tv_sec = max_secs; 5750 absTime->tv_nsec = 0; 5751 } 5752 else { 5753 absTime->tv_sec = now.tv_sec + secs; 5754 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5755 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5756 absTime->tv_nsec -= NANOSECS_PER_SEC; 5757 ++absTime->tv_sec; // note: this must be <= max_secs 5758 } 5759 } 5760 } 5761 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5762 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5763 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5764 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5765} 5766 5767void Parker::park(bool isAbsolute, jlong time) { 5768 // Ideally we'd do something useful while spinning, such 5769 // as calling unpackTime(). 5770 5771 // Optional fast-path check: 5772 // Return immediately if a permit is available. 5773 // We depend on Atomic::xchg() having full barrier semantics 5774 // since we are doing a lock-free update to _counter. 5775 if (Atomic::xchg(0, &_counter) > 0) return; 5776 5777 Thread* thread = Thread::current(); 5778 assert(thread->is_Java_thread(), "Must be JavaThread"); 5779 JavaThread *jt = (JavaThread *)thread; 5780 5781 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5782 // Check interrupt before trying to wait 5783 if (Thread::is_interrupted(thread, false)) { 5784 return; 5785 } 5786 5787 // Next, demultiplex/decode time arguments 5788 timespec absTime; 5789 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5790 return; 5791 } 5792 if (time > 0) { 5793 unpackTime(&absTime, isAbsolute, time); 5794 } 5795 5796 5797 // Enter safepoint region 5798 // Beware of deadlocks such as 6317397. 5799 // The per-thread Parker:: mutex is a classic leaf-lock. 5800 // In particular a thread must never block on the Threads_lock while 5801 // holding the Parker:: mutex. If safepoints are pending both the 5802 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5803 ThreadBlockInVM tbivm(jt); 5804 5805 // Don't wait if cannot get lock since interference arises from 5806 // unblocking. Also. check interrupt before trying wait 5807 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5808 return; 5809 } 5810 5811 int status ; 5812 if (_counter > 0) { // no wait needed 5813 _counter = 0; 5814 status = pthread_mutex_unlock(_mutex); 5815 assert (status == 0, "invariant") ; 5816 // Paranoia to ensure our locked and lock-free paths interact 5817 // correctly with each other and Java-level accesses. 5818 OrderAccess::fence(); 5819 return; 5820 } 5821 5822#ifdef ASSERT 5823 // Don't catch signals while blocked; let the running threads have the signals. 5824 // (This allows a debugger to break into the running thread.) 5825 sigset_t oldsigs; 5826 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5827 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5828#endif 5829 5830 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5831 jt->set_suspend_equivalent(); 5832 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5833 5834 if (time == 0) { 5835 status = pthread_cond_wait (_cond, _mutex) ; 5836 } else { 5837 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 5838 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5839 pthread_cond_destroy (_cond) ; 5840 pthread_cond_init (_cond, NULL); 5841 } 5842 } 5843 assert_status(status == 0 || status == EINTR || 5844 status == ETIME || status == ETIMEDOUT, 5845 status, "cond_timedwait"); 5846 5847#ifdef ASSERT 5848 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5849#endif 5850 5851 _counter = 0 ; 5852 status = pthread_mutex_unlock(_mutex) ; 5853 assert_status(status == 0, status, "invariant") ; 5854 // Paranoia to ensure our locked and lock-free paths interact 5855 // correctly with each other and Java-level accesses. 5856 OrderAccess::fence(); 5857 5858 // If externally suspended while waiting, re-suspend 5859 if (jt->handle_special_suspend_equivalent_condition()) { 5860 jt->java_suspend_self(); 5861 } 5862} 5863 5864void Parker::unpark() { 5865 int s, status ; 5866 status = pthread_mutex_lock(_mutex); 5867 assert (status == 0, "invariant") ; 5868 s = _counter; 5869 _counter = 1; 5870 if (s < 1) { 5871 if (WorkAroundNPTLTimedWaitHang) { 5872 status = pthread_cond_signal (_cond) ; 5873 assert (status == 0, "invariant") ; 5874 status = pthread_mutex_unlock(_mutex); 5875 assert (status == 0, "invariant") ; 5876 } else { 5877 status = pthread_mutex_unlock(_mutex); 5878 assert (status == 0, "invariant") ; 5879 status = pthread_cond_signal (_cond) ; 5880 assert (status == 0, "invariant") ; 5881 } 5882 } else { 5883 pthread_mutex_unlock(_mutex); 5884 assert (status == 0, "invariant") ; 5885 } 5886} 5887 5888 5889extern char** environ; 5890 5891#ifndef __NR_fork 5892#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5893#endif 5894 5895#ifndef __NR_execve 5896#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5897#endif 5898 5899// Run the specified command in a separate process. Return its exit value, 5900// or -1 on failure (e.g. can't fork a new process). 5901// Unlike system(), this function can be called from signal handler. It 5902// doesn't block SIGINT et al. 5903int os::fork_and_exec(char* cmd) { 5904 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5905 5906 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5907 // pthread_atfork handlers and reset pthread library. All we need is a 5908 // separate process to execve. Make a direct syscall to fork process. 5909 // On IA64 there's no fork syscall, we have to use fork() and hope for 5910 // the best... 5911 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5912 IA64_ONLY(fork();) 5913 5914 if (pid < 0) { 5915 // fork failed 5916 return -1; 5917 5918 } else if (pid == 0) { 5919 // child process 5920 5921 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5922 // first to kill every thread on the thread list. Because this list is 5923 // not reset by fork() (see notes above), execve() will instead kill 5924 // every thread in the parent process. We know this is the only thread 5925 // in the new process, so make a system call directly. 5926 // IA64 should use normal execve() from glibc to match the glibc fork() 5927 // above. 5928 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5929 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5930 5931 // execve failed 5932 _exit(-1); 5933 5934 } else { 5935 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5936 // care about the actual exit code, for now. 5937 5938 int status; 5939 5940 // Wait for the child process to exit. This returns immediately if 5941 // the child has already exited. */ 5942 while (waitpid(pid, &status, 0) < 0) { 5943 switch (errno) { 5944 case ECHILD: return 0; 5945 case EINTR: break; 5946 default: return -1; 5947 } 5948 } 5949 5950 if (WIFEXITED(status)) { 5951 // The child exited normally; get its exit code. 5952 return WEXITSTATUS(status); 5953 } else if (WIFSIGNALED(status)) { 5954 // The child exited because of a signal 5955 // The best value to return is 0x80 + signal number, 5956 // because that is what all Unix shells do, and because 5957 // it allows callers to distinguish between process exit and 5958 // process death by signal. 5959 return 0x80 + WTERMSIG(status); 5960 } else { 5961 // Unknown exit code; pass it through 5962 return status; 5963 } 5964 } 5965} 5966 5967// is_headless_jre() 5968// 5969// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5970// in order to report if we are running in a headless jre 5971// 5972// Since JDK8 xawt/libmawt.so was moved into the same directory 5973// as libawt.so, and renamed libawt_xawt.so 5974// 5975bool os::is_headless_jre() { 5976 struct stat statbuf; 5977 char buf[MAXPATHLEN]; 5978 char libmawtpath[MAXPATHLEN]; 5979 const char *xawtstr = "/xawt/libmawt.so"; 5980 const char *new_xawtstr = "/libawt_xawt.so"; 5981 char *p; 5982 5983 // Get path to libjvm.so 5984 os::jvm_path(buf, sizeof(buf)); 5985 5986 // Get rid of libjvm.so 5987 p = strrchr(buf, '/'); 5988 if (p == NULL) return false; 5989 else *p = '\0'; 5990 5991 // Get rid of client or server 5992 p = strrchr(buf, '/'); 5993 if (p == NULL) return false; 5994 else *p = '\0'; 5995 5996 // check xawt/libmawt.so 5997 strcpy(libmawtpath, buf); 5998 strcat(libmawtpath, xawtstr); 5999 if (::stat(libmawtpath, &statbuf) == 0) return false; 6000 6001 // check libawt_xawt.so 6002 strcpy(libmawtpath, buf); 6003 strcat(libmawtpath, new_xawtstr); 6004 if (::stat(libmawtpath, &statbuf) == 0) return false; 6005 6006 return true; 6007} 6008 6009// Get the default path to the core file 6010// Returns the length of the string 6011int os::get_core_path(char* buffer, size_t bufferSize) { 6012 const char* p = get_current_directory(buffer, bufferSize); 6013 6014 if (p == NULL) { 6015 assert(p != NULL, "failed to get current directory"); 6016 return 0; 6017 } 6018 6019 return strlen(buffer); 6020} 6021 6022#ifdef JAVASE_EMBEDDED 6023// 6024// A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 6025// 6026MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 6027 6028// ctor 6029// 6030MemNotifyThread::MemNotifyThread(int fd): Thread() { 6031 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 6032 _fd = fd; 6033 6034 if (os::create_thread(this, os::os_thread)) { 6035 _memnotify_thread = this; 6036 os::set_priority(this, NearMaxPriority); 6037 os::start_thread(this); 6038 } 6039} 6040 6041// Where all the work gets done 6042// 6043void MemNotifyThread::run() { 6044 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 6045 6046 // Set up the select arguments 6047 fd_set rfds; 6048 if (_fd != -1) { 6049 FD_ZERO(&rfds); 6050 FD_SET(_fd, &rfds); 6051 } 6052 6053 // Now wait for the mem_notify device to wake up 6054 while (1) { 6055 // Wait for the mem_notify device to signal us.. 6056 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 6057 if (rc == -1) { 6058 perror("select!\n"); 6059 break; 6060 } else if (rc) { 6061 //ssize_t free_before = os::available_memory(); 6062 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 6063 6064 // The kernel is telling us there is not much memory left... 6065 // try to do something about that 6066 6067 // If we are not already in a GC, try one. 6068 if (!Universe::heap()->is_gc_active()) { 6069 Universe::heap()->collect(GCCause::_allocation_failure); 6070 6071 //ssize_t free_after = os::available_memory(); 6072 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 6073 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 6074 } 6075 // We might want to do something like the following if we find the GC's are not helping... 6076 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 6077 } 6078 } 6079} 6080 6081// 6082// See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 6083// 6084void MemNotifyThread::start() { 6085 int fd; 6086 fd = open ("/dev/mem_notify", O_RDONLY, 0); 6087 if (fd < 0) { 6088 return; 6089 } 6090 6091 if (memnotify_thread() == NULL) { 6092 new MemNotifyThread(fd); 6093 } 6094} 6095 6096#endif // JAVASE_EMBEDDED 6097 6098 6099/////////////// Unit tests /////////////// 6100 6101#ifndef PRODUCT 6102 6103#define test_log(...) \ 6104 do {\ 6105 if (VerboseInternalVMTests) { \ 6106 tty->print_cr(__VA_ARGS__); \ 6107 tty->flush(); \ 6108 }\ 6109 } while (false) 6110 6111class TestReserveMemorySpecial : AllStatic { 6112 public: 6113 static void small_page_write(void* addr, size_t size) { 6114 size_t page_size = os::vm_page_size(); 6115 6116 char* end = (char*)addr + size; 6117 for (char* p = (char*)addr; p < end; p += page_size) { 6118 *p = 1; 6119 } 6120 } 6121 6122 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6123 if (!UseHugeTLBFS) { 6124 return; 6125 } 6126 6127 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6128 6129 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6130 6131 if (addr != NULL) { 6132 small_page_write(addr, size); 6133 6134 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6135 } 6136 } 6137 6138 static void test_reserve_memory_special_huge_tlbfs_only() { 6139 if (!UseHugeTLBFS) { 6140 return; 6141 } 6142 6143 size_t lp = os::large_page_size(); 6144 6145 for (size_t size = lp; size <= lp * 10; size += lp) { 6146 test_reserve_memory_special_huge_tlbfs_only(size); 6147 } 6148 } 6149 6150 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) { 6151 if (!UseHugeTLBFS) { 6152 return; 6153 } 6154 6155 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")", 6156 size, alignment); 6157 6158 assert(size >= os::large_page_size(), "Incorrect input to test"); 6159 6160 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6161 6162 if (addr != NULL) { 6163 small_page_write(addr, size); 6164 6165 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6166 } 6167 } 6168 6169 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) { 6170 size_t lp = os::large_page_size(); 6171 size_t ag = os::vm_allocation_granularity(); 6172 6173 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6174 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment); 6175 } 6176 } 6177 6178 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6179 size_t lp = os::large_page_size(); 6180 size_t ag = os::vm_allocation_granularity(); 6181 6182 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp); 6183 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag); 6184 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2); 6185 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2); 6186 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag); 6187 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag); 6188 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2); 6189 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10); 6190 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2); 6191 } 6192 6193 static void test_reserve_memory_special_huge_tlbfs() { 6194 if (!UseHugeTLBFS) { 6195 return; 6196 } 6197 6198 test_reserve_memory_special_huge_tlbfs_only(); 6199 test_reserve_memory_special_huge_tlbfs_mixed(); 6200 } 6201 6202 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6203 if (!UseSHM) { 6204 return; 6205 } 6206 6207 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6208 6209 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6210 6211 if (addr != NULL) { 6212 assert(is_ptr_aligned(addr, alignment), "Check"); 6213 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6214 6215 small_page_write(addr, size); 6216 6217 os::Linux::release_memory_special_shm(addr, size); 6218 } 6219 } 6220 6221 static void test_reserve_memory_special_shm() { 6222 size_t lp = os::large_page_size(); 6223 size_t ag = os::vm_allocation_granularity(); 6224 6225 for (size_t size = ag; size < lp * 3; size += ag) { 6226 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6227 test_reserve_memory_special_shm(size, alignment); 6228 } 6229 } 6230 } 6231 6232 static void test() { 6233 test_reserve_memory_special_huge_tlbfs(); 6234 test_reserve_memory_special_shm(); 6235 } 6236}; 6237 6238void TestReserveMemorySpecial_test() { 6239 TestReserveMemorySpecial::test(); 6240} 6241 6242#endif 6243