os_linux.cpp revision 5177:d8e99408faad
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 2770void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2771 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2772} 2773 2774bool os::numa_topology_changed() { return false; } 2775 2776size_t os::numa_get_groups_num() { 2777 int max_node = Linux::numa_max_node(); 2778 return max_node > 0 ? max_node + 1 : 1; 2779} 2780 2781int os::numa_get_group_id() { 2782 int cpu_id = Linux::sched_getcpu(); 2783 if (cpu_id != -1) { 2784 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2785 if (lgrp_id != -1) { 2786 return lgrp_id; 2787 } 2788 } 2789 return 0; 2790} 2791 2792size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2793 for (size_t i = 0; i < size; i++) { 2794 ids[i] = i; 2795 } 2796 return size; 2797} 2798 2799bool os::get_page_info(char *start, page_info* info) { 2800 return false; 2801} 2802 2803char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2804 return end; 2805} 2806 2807 2808int os::Linux::sched_getcpu_syscall(void) { 2809 unsigned int cpu; 2810 int retval = -1; 2811 2812#if defined(IA32) 2813# ifndef SYS_getcpu 2814# define SYS_getcpu 318 2815# endif 2816 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2817#elif defined(AMD64) 2818// Unfortunately we have to bring all these macros here from vsyscall.h 2819// to be able to compile on old linuxes. 2820# define __NR_vgetcpu 2 2821# define VSYSCALL_START (-10UL << 20) 2822# define VSYSCALL_SIZE 1024 2823# define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2824 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2825 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2826 retval = vgetcpu(&cpu, NULL, NULL); 2827#endif 2828 2829 return (retval == -1) ? retval : cpu; 2830} 2831 2832// Something to do with the numa-aware allocator needs these symbols 2833extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2834extern "C" JNIEXPORT void numa_error(char *where) { } 2835extern "C" JNIEXPORT int fork1() { return fork(); } 2836 2837 2838// If we are running with libnuma version > 2, then we should 2839// be trying to use symbols with versions 1.1 2840// If we are running with earlier version, which did not have symbol versions, 2841// we should use the base version. 2842void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2843 void *f = dlvsym(handle, name, "libnuma_1.1"); 2844 if (f == NULL) { 2845 f = dlsym(handle, name); 2846 } 2847 return f; 2848} 2849 2850bool os::Linux::libnuma_init() { 2851 // sched_getcpu() should be in libc. 2852 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2853 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2854 2855 // If it's not, try a direct syscall. 2856 if (sched_getcpu() == -1) 2857 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall)); 2858 2859 if (sched_getcpu() != -1) { // Does it work? 2860 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2861 if (handle != NULL) { 2862 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2863 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2864 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2865 libnuma_dlsym(handle, "numa_max_node"))); 2866 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2867 libnuma_dlsym(handle, "numa_available"))); 2868 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2869 libnuma_dlsym(handle, "numa_tonode_memory"))); 2870 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2871 libnuma_dlsym(handle, "numa_interleave_memory"))); 2872 2873 2874 if (numa_available() != -1) { 2875 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2876 // Create a cpu -> node mapping 2877 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2878 rebuild_cpu_to_node_map(); 2879 return true; 2880 } 2881 } 2882 } 2883 return false; 2884} 2885 2886// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2887// The table is later used in get_node_by_cpu(). 2888void os::Linux::rebuild_cpu_to_node_map() { 2889 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2890 // in libnuma (possible values are starting from 16, 2891 // and continuing up with every other power of 2, but less 2892 // than the maximum number of CPUs supported by kernel), and 2893 // is a subject to change (in libnuma version 2 the requirements 2894 // are more reasonable) we'll just hardcode the number they use 2895 // in the library. 2896 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2897 2898 size_t cpu_num = os::active_processor_count(); 2899 size_t cpu_map_size = NCPUS / BitsPerCLong; 2900 size_t cpu_map_valid_size = 2901 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2902 2903 cpu_to_node()->clear(); 2904 cpu_to_node()->at_grow(cpu_num - 1); 2905 size_t node_num = numa_get_groups_num(); 2906 2907 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2908 for (size_t i = 0; i < node_num; i++) { 2909 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2910 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2911 if (cpu_map[j] != 0) { 2912 for (size_t k = 0; k < BitsPerCLong; k++) { 2913 if (cpu_map[j] & (1UL << k)) { 2914 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2915 } 2916 } 2917 } 2918 } 2919 } 2920 } 2921 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2922} 2923 2924int os::Linux::get_node_by_cpu(int cpu_id) { 2925 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2926 return cpu_to_node()->at(cpu_id); 2927 } 2928 return -1; 2929} 2930 2931GrowableArray<int>* os::Linux::_cpu_to_node; 2932os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2933os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2934os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2935os::Linux::numa_available_func_t os::Linux::_numa_available; 2936os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2937os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2938unsigned long* os::Linux::_numa_all_nodes; 2939 2940bool os::pd_uncommit_memory(char* addr, size_t size) { 2941 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2942 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2943 return res != (uintptr_t) MAP_FAILED; 2944} 2945 2946static 2947address get_stack_commited_bottom(address bottom, size_t size) { 2948 address nbot = bottom; 2949 address ntop = bottom + size; 2950 2951 size_t page_sz = os::vm_page_size(); 2952 unsigned pages = size / page_sz; 2953 2954 unsigned char vec[1]; 2955 unsigned imin = 1, imax = pages + 1, imid; 2956 int mincore_return_value; 2957 2958 while (imin < imax) { 2959 imid = (imax + imin) / 2; 2960 nbot = ntop - (imid * page_sz); 2961 2962 // Use a trick with mincore to check whether the page is mapped or not. 2963 // mincore sets vec to 1 if page resides in memory and to 0 if page 2964 // is swapped output but if page we are asking for is unmapped 2965 // it returns -1,ENOMEM 2966 mincore_return_value = mincore(nbot, page_sz, vec); 2967 2968 if (mincore_return_value == -1) { 2969 // Page is not mapped go up 2970 // to find first mapped page 2971 if (errno != EAGAIN) { 2972 assert(errno == ENOMEM, "Unexpected mincore errno"); 2973 imax = imid; 2974 } 2975 } else { 2976 // Page is mapped go down 2977 // to find first not mapped page 2978 imin = imid + 1; 2979 } 2980 } 2981 2982 nbot = nbot + page_sz; 2983 2984 // Adjust stack bottom one page up if last checked page is not mapped 2985 if (mincore_return_value == -1) { 2986 nbot = nbot + page_sz; 2987 } 2988 2989 return nbot; 2990} 2991 2992 2993// Linux uses a growable mapping for the stack, and if the mapping for 2994// the stack guard pages is not removed when we detach a thread the 2995// stack cannot grow beyond the pages where the stack guard was 2996// mapped. If at some point later in the process the stack expands to 2997// that point, the Linux kernel cannot expand the stack any further 2998// because the guard pages are in the way, and a segfault occurs. 2999// 3000// However, it's essential not to split the stack region by unmapping 3001// a region (leaving a hole) that's already part of the stack mapping, 3002// so if the stack mapping has already grown beyond the guard pages at 3003// the time we create them, we have to truncate the stack mapping. 3004// So, we need to know the extent of the stack mapping when 3005// create_stack_guard_pages() is called. 3006 3007// We only need this for stacks that are growable: at the time of 3008// writing thread stacks don't use growable mappings (i.e. those 3009// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3010// only applies to the main thread. 3011 3012// If the (growable) stack mapping already extends beyond the point 3013// where we're going to put our guard pages, truncate the mapping at 3014// that point by munmap()ping it. This ensures that when we later 3015// munmap() the guard pages we don't leave a hole in the stack 3016// mapping. This only affects the main/initial thread 3017 3018bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3019 3020 if (os::Linux::is_initial_thread()) { 3021 // As we manually grow stack up to bottom inside create_attached_thread(), 3022 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3023 // we don't need to do anything special. 3024 // Check it first, before calling heavy function. 3025 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3026 unsigned char vec[1]; 3027 3028 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3029 // Fallback to slow path on all errors, including EAGAIN 3030 stack_extent = (uintptr_t) get_stack_commited_bottom( 3031 os::Linux::initial_thread_stack_bottom(), 3032 (size_t)addr - stack_extent); 3033 } 3034 3035 if (stack_extent < (uintptr_t)addr) { 3036 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3037 } 3038 } 3039 3040 return os::commit_memory(addr, size, !ExecMem); 3041} 3042 3043// If this is a growable mapping, remove the guard pages entirely by 3044// munmap()ping them. If not, just call uncommit_memory(). This only 3045// affects the main/initial thread, but guard against future OS changes 3046// It's safe to always unmap guard pages for initial thread because we 3047// always place it right after end of the mapped region 3048 3049bool os::remove_stack_guard_pages(char* addr, size_t size) { 3050 uintptr_t stack_extent, stack_base; 3051 3052 if (os::Linux::is_initial_thread()) { 3053 return ::munmap(addr, size) == 0; 3054 } 3055 3056 return os::uncommit_memory(addr, size); 3057} 3058 3059static address _highest_vm_reserved_address = NULL; 3060 3061// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3062// at 'requested_addr'. If there are existing memory mappings at the same 3063// location, however, they will be overwritten. If 'fixed' is false, 3064// 'requested_addr' is only treated as a hint, the return value may or 3065// may not start from the requested address. Unlike Linux mmap(), this 3066// function returns NULL to indicate failure. 3067static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3068 char * addr; 3069 int flags; 3070 3071 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3072 if (fixed) { 3073 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3074 flags |= MAP_FIXED; 3075 } 3076 3077 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3078 // touch an uncommitted page. Otherwise, the read/write might 3079 // succeed if we have enough swap space to back the physical page. 3080 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3081 flags, -1, 0); 3082 3083 if (addr != MAP_FAILED) { 3084 // anon_mmap() should only get called during VM initialization, 3085 // don't need lock (actually we can skip locking even it can be called 3086 // from multiple threads, because _highest_vm_reserved_address is just a 3087 // hint about the upper limit of non-stack memory regions.) 3088 if ((address)addr + bytes > _highest_vm_reserved_address) { 3089 _highest_vm_reserved_address = (address)addr + bytes; 3090 } 3091 } 3092 3093 return addr == MAP_FAILED ? NULL : addr; 3094} 3095 3096// Don't update _highest_vm_reserved_address, because there might be memory 3097// regions above addr + size. If so, releasing a memory region only creates 3098// a hole in the address space, it doesn't help prevent heap-stack collision. 3099// 3100static int anon_munmap(char * addr, size_t size) { 3101 return ::munmap(addr, size) == 0; 3102} 3103 3104char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3105 size_t alignment_hint) { 3106 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3107} 3108 3109bool os::pd_release_memory(char* addr, size_t size) { 3110 return anon_munmap(addr, size); 3111} 3112 3113static address highest_vm_reserved_address() { 3114 return _highest_vm_reserved_address; 3115} 3116 3117static bool linux_mprotect(char* addr, size_t size, int prot) { 3118 // Linux wants the mprotect address argument to be page aligned. 3119 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3120 3121 // According to SUSv3, mprotect() should only be used with mappings 3122 // established by mmap(), and mmap() always maps whole pages. Unaligned 3123 // 'addr' likely indicates problem in the VM (e.g. trying to change 3124 // protection of malloc'ed or statically allocated memory). Check the 3125 // caller if you hit this assert. 3126 assert(addr == bottom, "sanity check"); 3127 3128 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3129 return ::mprotect(bottom, size, prot) == 0; 3130} 3131 3132// Set protections specified 3133bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3134 bool is_committed) { 3135 unsigned int p = 0; 3136 switch (prot) { 3137 case MEM_PROT_NONE: p = PROT_NONE; break; 3138 case MEM_PROT_READ: p = PROT_READ; break; 3139 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3140 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3141 default: 3142 ShouldNotReachHere(); 3143 } 3144 // is_committed is unused. 3145 return linux_mprotect(addr, bytes, p); 3146} 3147 3148bool os::guard_memory(char* addr, size_t size) { 3149 return linux_mprotect(addr, size, PROT_NONE); 3150} 3151 3152bool os::unguard_memory(char* addr, size_t size) { 3153 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3154} 3155 3156bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) { 3157 bool result = false; 3158 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3159 MAP_ANONYMOUS|MAP_PRIVATE, 3160 -1, 0); 3161 if (p != MAP_FAILED) { 3162 void *aligned_p = align_ptr_up(p, page_size); 3163 3164 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3165 3166 munmap(p, page_size * 2); 3167 } 3168 3169 if (warn && !result) { 3170 warning("TransparentHugePages is not supported by the operating system."); 3171 } 3172 3173 return result; 3174} 3175 3176bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3177 bool result = false; 3178 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3179 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3180 -1, 0); 3181 3182 if (p != MAP_FAILED) { 3183 // We don't know if this really is a huge page or not. 3184 FILE *fp = fopen("/proc/self/maps", "r"); 3185 if (fp) { 3186 while (!feof(fp)) { 3187 char chars[257]; 3188 long x = 0; 3189 if (fgets(chars, sizeof(chars), fp)) { 3190 if (sscanf(chars, "%lx-%*x", &x) == 1 3191 && x == (long)p) { 3192 if (strstr (chars, "hugepage")) { 3193 result = true; 3194 break; 3195 } 3196 } 3197 } 3198 } 3199 fclose(fp); 3200 } 3201 munmap(p, page_size); 3202 } 3203 3204 if (warn && !result) { 3205 warning("HugeTLBFS is not supported by the operating system."); 3206 } 3207 3208 return result; 3209} 3210 3211/* 3212* Set the coredump_filter bits to include largepages in core dump (bit 6) 3213* 3214* From the coredump_filter documentation: 3215* 3216* - (bit 0) anonymous private memory 3217* - (bit 1) anonymous shared memory 3218* - (bit 2) file-backed private memory 3219* - (bit 3) file-backed shared memory 3220* - (bit 4) ELF header pages in file-backed private memory areas (it is 3221* effective only if the bit 2 is cleared) 3222* - (bit 5) hugetlb private memory 3223* - (bit 6) hugetlb shared memory 3224*/ 3225static void set_coredump_filter(void) { 3226 FILE *f; 3227 long cdm; 3228 3229 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3230 return; 3231 } 3232 3233 if (fscanf(f, "%lx", &cdm) != 1) { 3234 fclose(f); 3235 return; 3236 } 3237 3238 rewind(f); 3239 3240 if ((cdm & LARGEPAGES_BIT) == 0) { 3241 cdm |= LARGEPAGES_BIT; 3242 fprintf(f, "%#lx", cdm); 3243 } 3244 3245 fclose(f); 3246} 3247 3248// Large page support 3249 3250static size_t _large_page_size = 0; 3251 3252size_t os::Linux::find_large_page_size() { 3253 size_t large_page_size = 0; 3254 3255 // large_page_size on Linux is used to round up heap size. x86 uses either 3256 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3257 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3258 // page as large as 256M. 3259 // 3260 // Here we try to figure out page size by parsing /proc/meminfo and looking 3261 // for a line with the following format: 3262 // Hugepagesize: 2048 kB 3263 // 3264 // If we can't determine the value (e.g. /proc is not mounted, or the text 3265 // format has been changed), we'll use the largest page size supported by 3266 // the processor. 3267 3268#ifndef ZERO 3269 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3270 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3271#endif // ZERO 3272 3273 FILE *fp = fopen("/proc/meminfo", "r"); 3274 if (fp) { 3275 while (!feof(fp)) { 3276 int x = 0; 3277 char buf[16]; 3278 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3279 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3280 large_page_size = x * K; 3281 break; 3282 } 3283 } else { 3284 // skip to next line 3285 for (;;) { 3286 int ch = fgetc(fp); 3287 if (ch == EOF || ch == (int)'\n') break; 3288 } 3289 } 3290 } 3291 fclose(fp); 3292 } 3293 3294 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3295 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3296 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3297 proper_unit_for_byte_size(large_page_size)); 3298 } 3299 3300 return large_page_size; 3301} 3302 3303size_t os::Linux::setup_large_page_size() { 3304 _large_page_size = Linux::find_large_page_size(); 3305 const size_t default_page_size = (size_t)Linux::page_size(); 3306 if (_large_page_size > default_page_size) { 3307 _page_sizes[0] = _large_page_size; 3308 _page_sizes[1] = default_page_size; 3309 _page_sizes[2] = 0; 3310 } 3311 3312 return _large_page_size; 3313} 3314 3315bool os::Linux::setup_large_page_type(size_t page_size) { 3316 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3317 FLAG_IS_DEFAULT(UseSHM) && 3318 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3319 // If UseLargePages is specified on the command line try all methods, 3320 // if it's default, then try only UseTransparentHugePages. 3321 if (FLAG_IS_DEFAULT(UseLargePages)) { 3322 UseTransparentHugePages = true; 3323 } else { 3324 UseHugeTLBFS = UseTransparentHugePages = UseSHM = true; 3325 } 3326 } 3327 3328 if (UseTransparentHugePages) { 3329 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3330 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3331 UseHugeTLBFS = false; 3332 UseSHM = false; 3333 return true; 3334 } 3335 UseTransparentHugePages = false; 3336 } 3337 3338 if (UseHugeTLBFS) { 3339 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3340 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3341 UseSHM = false; 3342 return true; 3343 } 3344 UseHugeTLBFS = false; 3345 } 3346 3347 return UseSHM; 3348} 3349 3350void os::large_page_init() { 3351 if (!UseLargePages) { 3352 UseHugeTLBFS = false; 3353 UseTransparentHugePages = false; 3354 UseSHM = false; 3355 return; 3356 } 3357 3358 size_t large_page_size = Linux::setup_large_page_size(); 3359 UseLargePages = Linux::setup_large_page_type(large_page_size); 3360 3361 set_coredump_filter(); 3362} 3363 3364#ifndef SHM_HUGETLB 3365#define SHM_HUGETLB 04000 3366#endif 3367 3368char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3369 // "exec" is passed in but not used. Creating the shared image for 3370 // the code cache doesn't have an SHM_X executable permission to check. 3371 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3372 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3373 3374 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3375 return NULL; // Fallback to small pages. 3376 } 3377 3378 key_t key = IPC_PRIVATE; 3379 char *addr; 3380 3381 bool warn_on_failure = UseLargePages && 3382 (!FLAG_IS_DEFAULT(UseLargePages) || 3383 !FLAG_IS_DEFAULT(UseSHM) || 3384 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 3385 ); 3386 char msg[128]; 3387 3388 // Create a large shared memory region to attach to based on size. 3389 // Currently, size is the total size of the heap 3390 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3391 if (shmid == -1) { 3392 // Possible reasons for shmget failure: 3393 // 1. shmmax is too small for Java heap. 3394 // > check shmmax value: cat /proc/sys/kernel/shmmax 3395 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3396 // 2. not enough large page memory. 3397 // > check available large pages: cat /proc/meminfo 3398 // > increase amount of large pages: 3399 // echo new_value > /proc/sys/vm/nr_hugepages 3400 // Note 1: different Linux may use different name for this property, 3401 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3402 // Note 2: it's possible there's enough physical memory available but 3403 // they are so fragmented after a long run that they can't 3404 // coalesce into large pages. Try to reserve large pages when 3405 // the system is still "fresh". 3406 if (warn_on_failure) { 3407 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3408 warning(msg); 3409 } 3410 return NULL; 3411 } 3412 3413 // attach to the region 3414 addr = (char*)shmat(shmid, req_addr, 0); 3415 int err = errno; 3416 3417 // Remove shmid. If shmat() is successful, the actual shared memory segment 3418 // will be deleted when it's detached by shmdt() or when the process 3419 // terminates. If shmat() is not successful this will remove the shared 3420 // segment immediately. 3421 shmctl(shmid, IPC_RMID, NULL); 3422 3423 if ((intptr_t)addr == -1) { 3424 if (warn_on_failure) { 3425 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3426 warning(msg); 3427 } 3428 return NULL; 3429 } 3430 3431 return addr; 3432} 3433 3434static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) { 3435 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3436 3437 bool warn_on_failure = UseLargePages && 3438 (!FLAG_IS_DEFAULT(UseLargePages) || 3439 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3440 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3441 3442 if (warn_on_failure) { 3443 char msg[128]; 3444 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3445 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3446 warning(msg); 3447 } 3448} 3449 3450char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) { 3451 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3452 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3453 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3454 3455 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3456 char* addr = (char*)::mmap(req_addr, bytes, prot, 3457 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3458 -1, 0); 3459 3460 if (addr == MAP_FAILED) { 3461 warn_on_large_pages_failure(req_addr, bytes, errno); 3462 return NULL; 3463 } 3464 3465 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3466 3467 return addr; 3468} 3469 3470char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3471 size_t large_page_size = os::large_page_size(); 3472 3473 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3474 3475 // Allocate small pages. 3476 3477 char* start; 3478 if (req_addr != NULL) { 3479 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3480 assert(is_size_aligned(bytes, alignment), "Must be"); 3481 start = os::reserve_memory(bytes, req_addr); 3482 assert(start == NULL || start == req_addr, "Must be"); 3483 } else { 3484 start = os::reserve_memory_aligned(bytes, alignment); 3485 } 3486 3487 if (start == NULL) { 3488 return NULL; 3489 } 3490 3491 assert(is_ptr_aligned(start, alignment), "Must be"); 3492 3493 // os::reserve_memory_special will record this memory area. 3494 // Need to release it here to prevent overlapping reservations. 3495 MemTracker::record_virtual_memory_release((address)start, bytes); 3496 3497 char* end = start + bytes; 3498 3499 // Find the regions of the allocated chunk that can be promoted to large pages. 3500 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3501 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3502 3503 size_t lp_bytes = lp_end - lp_start; 3504 3505 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3506 3507 if (lp_bytes == 0) { 3508 // The mapped region doesn't even span the start and the end of a large page. 3509 // Fall back to allocate a non-special area. 3510 ::munmap(start, end - start); 3511 return NULL; 3512 } 3513 3514 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3515 3516 3517 void* result; 3518 3519 if (start != lp_start) { 3520 result = ::mmap(start, lp_start - start, prot, 3521 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3522 -1, 0); 3523 if (result == MAP_FAILED) { 3524 ::munmap(lp_start, end - lp_start); 3525 return NULL; 3526 } 3527 } 3528 3529 result = ::mmap(lp_start, lp_bytes, prot, 3530 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3531 -1, 0); 3532 if (result == MAP_FAILED) { 3533 warn_on_large_pages_failure(req_addr, bytes, errno); 3534 // If the mmap above fails, the large pages region will be unmapped and we 3535 // have regions before and after with small pages. Release these regions. 3536 // 3537 // | mapped | unmapped | mapped | 3538 // ^ ^ ^ ^ 3539 // start lp_start lp_end end 3540 // 3541 ::munmap(start, lp_start - start); 3542 ::munmap(lp_end, end - lp_end); 3543 return NULL; 3544 } 3545 3546 if (lp_end != end) { 3547 result = ::mmap(lp_end, end - lp_end, prot, 3548 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3549 -1, 0); 3550 if (result == MAP_FAILED) { 3551 ::munmap(start, lp_end - start); 3552 return NULL; 3553 } 3554 } 3555 3556 return start; 3557} 3558 3559char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3560 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3561 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3562 assert(is_power_of_2(alignment), "Must be"); 3563 assert(is_power_of_2(os::large_page_size()), "Must be"); 3564 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3565 3566 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3567 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3568 } else { 3569 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3570 } 3571} 3572 3573char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3574 assert(UseLargePages, "only for large pages"); 3575 3576 char* addr; 3577 if (UseSHM) { 3578 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3579 } else { 3580 assert(UseHugeTLBFS, "must be"); 3581 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3582 } 3583 3584 if (addr != NULL) { 3585 if (UseNUMAInterleaving) { 3586 numa_make_global(addr, bytes); 3587 } 3588 3589 // The memory is committed 3590 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC); 3591 } 3592 3593 return addr; 3594} 3595 3596bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3597 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3598 return shmdt(base) == 0; 3599} 3600 3601bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3602 return pd_release_memory(base, bytes); 3603} 3604 3605bool os::release_memory_special(char* base, size_t bytes) { 3606 assert(UseLargePages, "only for large pages"); 3607 3608 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3609 3610 bool res; 3611 if (UseSHM) { 3612 res = os::Linux::release_memory_special_shm(base, bytes); 3613 } else { 3614 assert(UseHugeTLBFS, "must be"); 3615 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3616 } 3617 3618 if (res) { 3619 tkr.record((address)base, bytes); 3620 } else { 3621 tkr.discard(); 3622 } 3623 3624 return res; 3625} 3626 3627size_t os::large_page_size() { 3628 return _large_page_size; 3629} 3630 3631// With SysV SHM the entire memory region must be allocated as shared 3632// memory. 3633// HugeTLBFS allows application to commit large page memory on demand. 3634// However, when committing memory with HugeTLBFS fails, the region 3635// that was supposed to be committed will lose the old reservation 3636// and allow other threads to steal that memory region. Because of this 3637// behavior we can't commit HugeTLBFS memory. 3638bool os::can_commit_large_page_memory() { 3639 return UseTransparentHugePages; 3640} 3641 3642bool os::can_execute_large_page_memory() { 3643 return UseTransparentHugePages || UseHugeTLBFS; 3644} 3645 3646// Reserve memory at an arbitrary address, only if that area is 3647// available (and not reserved for something else). 3648 3649char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3650 const int max_tries = 10; 3651 char* base[max_tries]; 3652 size_t size[max_tries]; 3653 const size_t gap = 0x000000; 3654 3655 // Assert only that the size is a multiple of the page size, since 3656 // that's all that mmap requires, and since that's all we really know 3657 // about at this low abstraction level. If we need higher alignment, 3658 // we can either pass an alignment to this method or verify alignment 3659 // in one of the methods further up the call chain. See bug 5044738. 3660 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3661 3662 // Repeatedly allocate blocks until the block is allocated at the 3663 // right spot. Give up after max_tries. Note that reserve_memory() will 3664 // automatically update _highest_vm_reserved_address if the call is 3665 // successful. The variable tracks the highest memory address every reserved 3666 // by JVM. It is used to detect heap-stack collision if running with 3667 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3668 // space than needed, it could confuse the collision detecting code. To 3669 // solve the problem, save current _highest_vm_reserved_address and 3670 // calculate the correct value before return. 3671 address old_highest = _highest_vm_reserved_address; 3672 3673 // Linux mmap allows caller to pass an address as hint; give it a try first, 3674 // if kernel honors the hint then we can return immediately. 3675 char * addr = anon_mmap(requested_addr, bytes, false); 3676 if (addr == requested_addr) { 3677 return requested_addr; 3678 } 3679 3680 if (addr != NULL) { 3681 // mmap() is successful but it fails to reserve at the requested address 3682 anon_munmap(addr, bytes); 3683 } 3684 3685 int i; 3686 for (i = 0; i < max_tries; ++i) { 3687 base[i] = reserve_memory(bytes); 3688 3689 if (base[i] != NULL) { 3690 // Is this the block we wanted? 3691 if (base[i] == requested_addr) { 3692 size[i] = bytes; 3693 break; 3694 } 3695 3696 // Does this overlap the block we wanted? Give back the overlapped 3697 // parts and try again. 3698 3699 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3700 if (top_overlap >= 0 && top_overlap < bytes) { 3701 unmap_memory(base[i], top_overlap); 3702 base[i] += top_overlap; 3703 size[i] = bytes - top_overlap; 3704 } else { 3705 size_t bottom_overlap = base[i] + bytes - requested_addr; 3706 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3707 unmap_memory(requested_addr, bottom_overlap); 3708 size[i] = bytes - bottom_overlap; 3709 } else { 3710 size[i] = bytes; 3711 } 3712 } 3713 } 3714 } 3715 3716 // Give back the unused reserved pieces. 3717 3718 for (int j = 0; j < i; ++j) { 3719 if (base[j] != NULL) { 3720 unmap_memory(base[j], size[j]); 3721 } 3722 } 3723 3724 if (i < max_tries) { 3725 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3726 return requested_addr; 3727 } else { 3728 _highest_vm_reserved_address = old_highest; 3729 return NULL; 3730 } 3731} 3732 3733size_t os::read(int fd, void *buf, unsigned int nBytes) { 3734 return ::read(fd, buf, nBytes); 3735} 3736 3737// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 3738// Solaris uses poll(), linux uses park(). 3739// Poll() is likely a better choice, assuming that Thread.interrupt() 3740// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 3741// SIGSEGV, see 4355769. 3742 3743int os::sleep(Thread* thread, jlong millis, bool interruptible) { 3744 assert(thread == Thread::current(), "thread consistency check"); 3745 3746 ParkEvent * const slp = thread->_SleepEvent ; 3747 slp->reset() ; 3748 OrderAccess::fence() ; 3749 3750 if (interruptible) { 3751 jlong prevtime = javaTimeNanos(); 3752 3753 for (;;) { 3754 if (os::is_interrupted(thread, true)) { 3755 return OS_INTRPT; 3756 } 3757 3758 jlong newtime = javaTimeNanos(); 3759 3760 if (newtime - prevtime < 0) { 3761 // time moving backwards, should only happen if no monotonic clock 3762 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3763 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3764 } else { 3765 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3766 } 3767 3768 if(millis <= 0) { 3769 return OS_OK; 3770 } 3771 3772 prevtime = newtime; 3773 3774 { 3775 assert(thread->is_Java_thread(), "sanity check"); 3776 JavaThread *jt = (JavaThread *) thread; 3777 ThreadBlockInVM tbivm(jt); 3778 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 3779 3780 jt->set_suspend_equivalent(); 3781 // cleared by handle_special_suspend_equivalent_condition() or 3782 // java_suspend_self() via check_and_wait_while_suspended() 3783 3784 slp->park(millis); 3785 3786 // were we externally suspended while we were waiting? 3787 jt->check_and_wait_while_suspended(); 3788 } 3789 } 3790 } else { 3791 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 3792 jlong prevtime = javaTimeNanos(); 3793 3794 for (;;) { 3795 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 3796 // the 1st iteration ... 3797 jlong newtime = javaTimeNanos(); 3798 3799 if (newtime - prevtime < 0) { 3800 // time moving backwards, should only happen if no monotonic clock 3801 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3802 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3803 } else { 3804 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3805 } 3806 3807 if(millis <= 0) break ; 3808 3809 prevtime = newtime; 3810 slp->park(millis); 3811 } 3812 return OS_OK ; 3813 } 3814} 3815 3816int os::naked_sleep() { 3817 // %% make the sleep time an integer flag. for now use 1 millisec. 3818 return os::sleep(Thread::current(), 1, false); 3819} 3820 3821// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3822void os::infinite_sleep() { 3823 while (true) { // sleep forever ... 3824 ::sleep(100); // ... 100 seconds at a time 3825 } 3826} 3827 3828// Used to convert frequent JVM_Yield() to nops 3829bool os::dont_yield() { 3830 return DontYieldALot; 3831} 3832 3833void os::yield() { 3834 sched_yield(); 3835} 3836 3837os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3838 3839void os::yield_all(int attempts) { 3840 // Yields to all threads, including threads with lower priorities 3841 // Threads on Linux are all with same priority. The Solaris style 3842 // os::yield_all() with nanosleep(1ms) is not necessary. 3843 sched_yield(); 3844} 3845 3846// Called from the tight loops to possibly influence time-sharing heuristics 3847void os::loop_breaker(int attempts) { 3848 os::yield_all(attempts); 3849} 3850 3851//////////////////////////////////////////////////////////////////////////////// 3852// thread priority support 3853 3854// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3855// only supports dynamic priority, static priority must be zero. For real-time 3856// applications, Linux supports SCHED_RR which allows static priority (1-99). 3857// However, for large multi-threaded applications, SCHED_RR is not only slower 3858// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3859// of 5 runs - Sep 2005). 3860// 3861// The following code actually changes the niceness of kernel-thread/LWP. It 3862// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3863// not the entire user process, and user level threads are 1:1 mapped to kernel 3864// threads. It has always been the case, but could change in the future. For 3865// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3866// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3867 3868int os::java_to_os_priority[CriticalPriority + 1] = { 3869 19, // 0 Entry should never be used 3870 3871 4, // 1 MinPriority 3872 3, // 2 3873 2, // 3 3874 3875 1, // 4 3876 0, // 5 NormPriority 3877 -1, // 6 3878 3879 -2, // 7 3880 -3, // 8 3881 -4, // 9 NearMaxPriority 3882 3883 -5, // 10 MaxPriority 3884 3885 -5 // 11 CriticalPriority 3886}; 3887 3888static int prio_init() { 3889 if (ThreadPriorityPolicy == 1) { 3890 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3891 // if effective uid is not root. Perhaps, a more elegant way of doing 3892 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3893 if (geteuid() != 0) { 3894 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3895 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3896 } 3897 ThreadPriorityPolicy = 0; 3898 } 3899 } 3900 if (UseCriticalJavaThreadPriority) { 3901 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3902 } 3903 return 0; 3904} 3905 3906OSReturn os::set_native_priority(Thread* thread, int newpri) { 3907 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3908 3909 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3910 return (ret == 0) ? OS_OK : OS_ERR; 3911} 3912 3913OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3914 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3915 *priority_ptr = java_to_os_priority[NormPriority]; 3916 return OS_OK; 3917 } 3918 3919 errno = 0; 3920 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3921 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3922} 3923 3924// Hint to the underlying OS that a task switch would not be good. 3925// Void return because it's a hint and can fail. 3926void os::hint_no_preempt() {} 3927 3928//////////////////////////////////////////////////////////////////////////////// 3929// suspend/resume support 3930 3931// the low-level signal-based suspend/resume support is a remnant from the 3932// old VM-suspension that used to be for java-suspension, safepoints etc, 3933// within hotspot. Now there is a single use-case for this: 3934// - calling get_thread_pc() on the VMThread by the flat-profiler task 3935// that runs in the watcher thread. 3936// The remaining code is greatly simplified from the more general suspension 3937// code that used to be used. 3938// 3939// The protocol is quite simple: 3940// - suspend: 3941// - sends a signal to the target thread 3942// - polls the suspend state of the osthread using a yield loop 3943// - target thread signal handler (SR_handler) sets suspend state 3944// and blocks in sigsuspend until continued 3945// - resume: 3946// - sets target osthread state to continue 3947// - sends signal to end the sigsuspend loop in the SR_handler 3948// 3949// Note that the SR_lock plays no role in this suspend/resume protocol. 3950// 3951 3952static void resume_clear_context(OSThread *osthread) { 3953 osthread->set_ucontext(NULL); 3954 osthread->set_siginfo(NULL); 3955} 3956 3957static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3958 osthread->set_ucontext(context); 3959 osthread->set_siginfo(siginfo); 3960} 3961 3962// 3963// Handler function invoked when a thread's execution is suspended or 3964// resumed. We have to be careful that only async-safe functions are 3965// called here (Note: most pthread functions are not async safe and 3966// should be avoided.) 3967// 3968// Note: sigwait() is a more natural fit than sigsuspend() from an 3969// interface point of view, but sigwait() prevents the signal hander 3970// from being run. libpthread would get very confused by not having 3971// its signal handlers run and prevents sigwait()'s use with the 3972// mutex granting granting signal. 3973// 3974// Currently only ever called on the VMThread and JavaThreads (PC sampling) 3975// 3976static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3977 // Save and restore errno to avoid confusing native code with EINTR 3978 // after sigsuspend. 3979 int old_errno = errno; 3980 3981 Thread* thread = Thread::current(); 3982 OSThread* osthread = thread->osthread(); 3983 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3984 3985 os::SuspendResume::State current = osthread->sr.state(); 3986 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3987 suspend_save_context(osthread, siginfo, context); 3988 3989 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 3990 os::SuspendResume::State state = osthread->sr.suspended(); 3991 if (state == os::SuspendResume::SR_SUSPENDED) { 3992 sigset_t suspend_set; // signals for sigsuspend() 3993 3994 // get current set of blocked signals and unblock resume signal 3995 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3996 sigdelset(&suspend_set, SR_signum); 3997 3998 sr_semaphore.signal(); 3999 // wait here until we are resumed 4000 while (1) { 4001 sigsuspend(&suspend_set); 4002 4003 os::SuspendResume::State result = osthread->sr.running(); 4004 if (result == os::SuspendResume::SR_RUNNING) { 4005 sr_semaphore.signal(); 4006 break; 4007 } 4008 } 4009 4010 } else if (state == os::SuspendResume::SR_RUNNING) { 4011 // request was cancelled, continue 4012 } else { 4013 ShouldNotReachHere(); 4014 } 4015 4016 resume_clear_context(osthread); 4017 } else if (current == os::SuspendResume::SR_RUNNING) { 4018 // request was cancelled, continue 4019 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4020 // ignore 4021 } else { 4022 // ignore 4023 } 4024 4025 errno = old_errno; 4026} 4027 4028 4029static int SR_initialize() { 4030 struct sigaction act; 4031 char *s; 4032 /* Get signal number to use for suspend/resume */ 4033 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4034 int sig = ::strtol(s, 0, 10); 4035 if (sig > 0 || sig < _NSIG) { 4036 SR_signum = sig; 4037 } 4038 } 4039 4040 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4041 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4042 4043 sigemptyset(&SR_sigset); 4044 sigaddset(&SR_sigset, SR_signum); 4045 4046 /* Set up signal handler for suspend/resume */ 4047 act.sa_flags = SA_RESTART|SA_SIGINFO; 4048 act.sa_handler = (void (*)(int)) SR_handler; 4049 4050 // SR_signum is blocked by default. 4051 // 4528190 - We also need to block pthread restart signal (32 on all 4052 // supported Linux platforms). Note that LinuxThreads need to block 4053 // this signal for all threads to work properly. So we don't have 4054 // to use hard-coded signal number when setting up the mask. 4055 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4056 4057 if (sigaction(SR_signum, &act, 0) == -1) { 4058 return -1; 4059 } 4060 4061 // Save signal flag 4062 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4063 return 0; 4064} 4065 4066static int sr_notify(OSThread* osthread) { 4067 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4068 assert_status(status == 0, status, "pthread_kill"); 4069 return status; 4070} 4071 4072// "Randomly" selected value for how long we want to spin 4073// before bailing out on suspending a thread, also how often 4074// we send a signal to a thread we want to resume 4075static const int RANDOMLY_LARGE_INTEGER = 1000000; 4076static const int RANDOMLY_LARGE_INTEGER2 = 100; 4077 4078// returns true on success and false on error - really an error is fatal 4079// but this seems the normal response to library errors 4080static bool do_suspend(OSThread* osthread) { 4081 assert(osthread->sr.is_running(), "thread should be running"); 4082 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4083 4084 // mark as suspended and send signal 4085 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4086 // failed to switch, state wasn't running? 4087 ShouldNotReachHere(); 4088 return false; 4089 } 4090 4091 if (sr_notify(osthread) != 0) { 4092 ShouldNotReachHere(); 4093 } 4094 4095 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4096 while (true) { 4097 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4098 break; 4099 } else { 4100 // timeout 4101 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4102 if (cancelled == os::SuspendResume::SR_RUNNING) { 4103 return false; 4104 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4105 // make sure that we consume the signal on the semaphore as well 4106 sr_semaphore.wait(); 4107 break; 4108 } else { 4109 ShouldNotReachHere(); 4110 return false; 4111 } 4112 } 4113 } 4114 4115 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4116 return true; 4117} 4118 4119static void do_resume(OSThread* osthread) { 4120 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4121 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4122 4123 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4124 // failed to switch to WAKEUP_REQUEST 4125 ShouldNotReachHere(); 4126 return; 4127 } 4128 4129 while (true) { 4130 if (sr_notify(osthread) == 0) { 4131 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4132 if (osthread->sr.is_running()) { 4133 return; 4134 } 4135 } 4136 } else { 4137 ShouldNotReachHere(); 4138 } 4139 } 4140 4141 guarantee(osthread->sr.is_running(), "Must be running!"); 4142} 4143 4144//////////////////////////////////////////////////////////////////////////////// 4145// interrupt support 4146 4147void os::interrupt(Thread* thread) { 4148 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 4149 "possibility of dangling Thread pointer"); 4150 4151 OSThread* osthread = thread->osthread(); 4152 4153 if (!osthread->interrupted()) { 4154 osthread->set_interrupted(true); 4155 // More than one thread can get here with the same value of osthread, 4156 // resulting in multiple notifications. We do, however, want the store 4157 // to interrupted() to be visible to other threads before we execute unpark(). 4158 OrderAccess::fence(); 4159 ParkEvent * const slp = thread->_SleepEvent ; 4160 if (slp != NULL) slp->unpark() ; 4161 } 4162 4163 // For JSR166. Unpark even if interrupt status already was set 4164 if (thread->is_Java_thread()) 4165 ((JavaThread*)thread)->parker()->unpark(); 4166 4167 ParkEvent * ev = thread->_ParkEvent ; 4168 if (ev != NULL) ev->unpark() ; 4169 4170} 4171 4172bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 4173 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 4174 "possibility of dangling Thread pointer"); 4175 4176 OSThread* osthread = thread->osthread(); 4177 4178 bool interrupted = osthread->interrupted(); 4179 4180 if (interrupted && clear_interrupted) { 4181 osthread->set_interrupted(false); 4182 // consider thread->_SleepEvent->reset() ... optional optimization 4183 } 4184 4185 return interrupted; 4186} 4187 4188/////////////////////////////////////////////////////////////////////////////////// 4189// signal handling (except suspend/resume) 4190 4191// This routine may be used by user applications as a "hook" to catch signals. 4192// The user-defined signal handler must pass unrecognized signals to this 4193// routine, and if it returns true (non-zero), then the signal handler must 4194// return immediately. If the flag "abort_if_unrecognized" is true, then this 4195// routine will never retun false (zero), but instead will execute a VM panic 4196// routine kill the process. 4197// 4198// If this routine returns false, it is OK to call it again. This allows 4199// the user-defined signal handler to perform checks either before or after 4200// the VM performs its own checks. Naturally, the user code would be making 4201// a serious error if it tried to handle an exception (such as a null check 4202// or breakpoint) that the VM was generating for its own correct operation. 4203// 4204// This routine may recognize any of the following kinds of signals: 4205// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4206// It should be consulted by handlers for any of those signals. 4207// 4208// The caller of this routine must pass in the three arguments supplied 4209// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4210// field of the structure passed to sigaction(). This routine assumes that 4211// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4212// 4213// Note that the VM will print warnings if it detects conflicting signal 4214// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4215// 4216extern "C" JNIEXPORT int 4217JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 4218 void* ucontext, int abort_if_unrecognized); 4219 4220void signalHandler(int sig, siginfo_t* info, void* uc) { 4221 assert(info != NULL && uc != NULL, "it must be old kernel"); 4222 int orig_errno = errno; // Preserve errno value over signal handler. 4223 JVM_handle_linux_signal(sig, info, uc, true); 4224 errno = orig_errno; 4225} 4226 4227 4228// This boolean allows users to forward their own non-matching signals 4229// to JVM_handle_linux_signal, harmlessly. 4230bool os::Linux::signal_handlers_are_installed = false; 4231 4232// For signal-chaining 4233struct sigaction os::Linux::sigact[MAXSIGNUM]; 4234unsigned int os::Linux::sigs = 0; 4235bool os::Linux::libjsig_is_loaded = false; 4236typedef struct sigaction *(*get_signal_t)(int); 4237get_signal_t os::Linux::get_signal_action = NULL; 4238 4239struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4240 struct sigaction *actp = NULL; 4241 4242 if (libjsig_is_loaded) { 4243 // Retrieve the old signal handler from libjsig 4244 actp = (*get_signal_action)(sig); 4245 } 4246 if (actp == NULL) { 4247 // Retrieve the preinstalled signal handler from jvm 4248 actp = get_preinstalled_handler(sig); 4249 } 4250 4251 return actp; 4252} 4253 4254static bool call_chained_handler(struct sigaction *actp, int sig, 4255 siginfo_t *siginfo, void *context) { 4256 // Call the old signal handler 4257 if (actp->sa_handler == SIG_DFL) { 4258 // It's more reasonable to let jvm treat it as an unexpected exception 4259 // instead of taking the default action. 4260 return false; 4261 } else if (actp->sa_handler != SIG_IGN) { 4262 if ((actp->sa_flags & SA_NODEFER) == 0) { 4263 // automaticlly block the signal 4264 sigaddset(&(actp->sa_mask), sig); 4265 } 4266 4267 sa_handler_t hand; 4268 sa_sigaction_t sa; 4269 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4270 // retrieve the chained handler 4271 if (siginfo_flag_set) { 4272 sa = actp->sa_sigaction; 4273 } else { 4274 hand = actp->sa_handler; 4275 } 4276 4277 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4278 actp->sa_handler = SIG_DFL; 4279 } 4280 4281 // try to honor the signal mask 4282 sigset_t oset; 4283 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4284 4285 // call into the chained handler 4286 if (siginfo_flag_set) { 4287 (*sa)(sig, siginfo, context); 4288 } else { 4289 (*hand)(sig); 4290 } 4291 4292 // restore the signal mask 4293 pthread_sigmask(SIG_SETMASK, &oset, 0); 4294 } 4295 // Tell jvm's signal handler the signal is taken care of. 4296 return true; 4297} 4298 4299bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4300 bool chained = false; 4301 // signal-chaining 4302 if (UseSignalChaining) { 4303 struct sigaction *actp = get_chained_signal_action(sig); 4304 if (actp != NULL) { 4305 chained = call_chained_handler(actp, sig, siginfo, context); 4306 } 4307 } 4308 return chained; 4309} 4310 4311struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4312 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 4313 return &sigact[sig]; 4314 } 4315 return NULL; 4316} 4317 4318void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4319 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4320 sigact[sig] = oldAct; 4321 sigs |= (unsigned int)1 << sig; 4322} 4323 4324// for diagnostic 4325int os::Linux::sigflags[MAXSIGNUM]; 4326 4327int os::Linux::get_our_sigflags(int sig) { 4328 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4329 return sigflags[sig]; 4330} 4331 4332void os::Linux::set_our_sigflags(int sig, int flags) { 4333 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4334 sigflags[sig] = flags; 4335} 4336 4337void os::Linux::set_signal_handler(int sig, bool set_installed) { 4338 // Check for overwrite. 4339 struct sigaction oldAct; 4340 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4341 4342 void* oldhand = oldAct.sa_sigaction 4343 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4344 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4345 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4346 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4347 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4348 if (AllowUserSignalHandlers || !set_installed) { 4349 // Do not overwrite; user takes responsibility to forward to us. 4350 return; 4351 } else if (UseSignalChaining) { 4352 // save the old handler in jvm 4353 save_preinstalled_handler(sig, oldAct); 4354 // libjsig also interposes the sigaction() call below and saves the 4355 // old sigaction on it own. 4356 } else { 4357 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4358 "%#lx for signal %d.", (long)oldhand, sig)); 4359 } 4360 } 4361 4362 struct sigaction sigAct; 4363 sigfillset(&(sigAct.sa_mask)); 4364 sigAct.sa_handler = SIG_DFL; 4365 if (!set_installed) { 4366 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4367 } else { 4368 sigAct.sa_sigaction = signalHandler; 4369 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4370 } 4371 // Save flags, which are set by ours 4372 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4373 sigflags[sig] = sigAct.sa_flags; 4374 4375 int ret = sigaction(sig, &sigAct, &oldAct); 4376 assert(ret == 0, "check"); 4377 4378 void* oldhand2 = oldAct.sa_sigaction 4379 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4380 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4381 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4382} 4383 4384// install signal handlers for signals that HotSpot needs to 4385// handle in order to support Java-level exception handling. 4386 4387void os::Linux::install_signal_handlers() { 4388 if (!signal_handlers_are_installed) { 4389 signal_handlers_are_installed = true; 4390 4391 // signal-chaining 4392 typedef void (*signal_setting_t)(); 4393 signal_setting_t begin_signal_setting = NULL; 4394 signal_setting_t end_signal_setting = NULL; 4395 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4396 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4397 if (begin_signal_setting != NULL) { 4398 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4399 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4400 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4401 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4402 libjsig_is_loaded = true; 4403 assert(UseSignalChaining, "should enable signal-chaining"); 4404 } 4405 if (libjsig_is_loaded) { 4406 // Tell libjsig jvm is setting signal handlers 4407 (*begin_signal_setting)(); 4408 } 4409 4410 set_signal_handler(SIGSEGV, true); 4411 set_signal_handler(SIGPIPE, true); 4412 set_signal_handler(SIGBUS, true); 4413 set_signal_handler(SIGILL, true); 4414 set_signal_handler(SIGFPE, true); 4415 set_signal_handler(SIGXFSZ, true); 4416 4417 if (libjsig_is_loaded) { 4418 // Tell libjsig jvm finishes setting signal handlers 4419 (*end_signal_setting)(); 4420 } 4421 4422 // We don't activate signal checker if libjsig is in place, we trust ourselves 4423 // and if UserSignalHandler is installed all bets are off. 4424 // Log that signal checking is off only if -verbose:jni is specified. 4425 if (CheckJNICalls) { 4426 if (libjsig_is_loaded) { 4427 if (PrintJNIResolving) { 4428 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4429 } 4430 check_signals = false; 4431 } 4432 if (AllowUserSignalHandlers) { 4433 if (PrintJNIResolving) { 4434 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4435 } 4436 check_signals = false; 4437 } 4438 } 4439 } 4440} 4441 4442// This is the fastest way to get thread cpu time on Linux. 4443// Returns cpu time (user+sys) for any thread, not only for current. 4444// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4445// It might work on 2.6.10+ with a special kernel/glibc patch. 4446// For reference, please, see IEEE Std 1003.1-2004: 4447// http://www.unix.org/single_unix_specification 4448 4449jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4450 struct timespec tp; 4451 int rc = os::Linux::clock_gettime(clockid, &tp); 4452 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4453 4454 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4455} 4456 4457///// 4458// glibc on Linux platform uses non-documented flag 4459// to indicate, that some special sort of signal 4460// trampoline is used. 4461// We will never set this flag, and we should 4462// ignore this flag in our diagnostic 4463#ifdef SIGNIFICANT_SIGNAL_MASK 4464#undef SIGNIFICANT_SIGNAL_MASK 4465#endif 4466#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4467 4468static const char* get_signal_handler_name(address handler, 4469 char* buf, int buflen) { 4470 int offset; 4471 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4472 if (found) { 4473 // skip directory names 4474 const char *p1, *p2; 4475 p1 = buf; 4476 size_t len = strlen(os::file_separator()); 4477 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4478 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4479 } else { 4480 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4481 } 4482 return buf; 4483} 4484 4485static void print_signal_handler(outputStream* st, int sig, 4486 char* buf, size_t buflen) { 4487 struct sigaction sa; 4488 4489 sigaction(sig, NULL, &sa); 4490 4491 // See comment for SIGNIFICANT_SIGNAL_MASK define 4492 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4493 4494 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4495 4496 address handler = (sa.sa_flags & SA_SIGINFO) 4497 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4498 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4499 4500 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4501 st->print("SIG_DFL"); 4502 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4503 st->print("SIG_IGN"); 4504 } else { 4505 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4506 } 4507 4508 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 4509 4510 address rh = VMError::get_resetted_sighandler(sig); 4511 // May be, handler was resetted by VMError? 4512 if(rh != NULL) { 4513 handler = rh; 4514 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4515 } 4516 4517 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 4518 4519 // Check: is it our handler? 4520 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4521 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4522 // It is our signal handler 4523 // check for flags, reset system-used one! 4524 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4525 st->print( 4526 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4527 os::Linux::get_our_sigflags(sig)); 4528 } 4529 } 4530 st->cr(); 4531} 4532 4533 4534#define DO_SIGNAL_CHECK(sig) \ 4535 if (!sigismember(&check_signal_done, sig)) \ 4536 os::Linux::check_signal_handler(sig) 4537 4538// This method is a periodic task to check for misbehaving JNI applications 4539// under CheckJNI, we can add any periodic checks here 4540 4541void os::run_periodic_checks() { 4542 4543 if (check_signals == false) return; 4544 4545 // SEGV and BUS if overridden could potentially prevent 4546 // generation of hs*.log in the event of a crash, debugging 4547 // such a case can be very challenging, so we absolutely 4548 // check the following for a good measure: 4549 DO_SIGNAL_CHECK(SIGSEGV); 4550 DO_SIGNAL_CHECK(SIGILL); 4551 DO_SIGNAL_CHECK(SIGFPE); 4552 DO_SIGNAL_CHECK(SIGBUS); 4553 DO_SIGNAL_CHECK(SIGPIPE); 4554 DO_SIGNAL_CHECK(SIGXFSZ); 4555 4556 4557 // ReduceSignalUsage allows the user to override these handlers 4558 // see comments at the very top and jvm_solaris.h 4559 if (!ReduceSignalUsage) { 4560 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4561 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4562 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4563 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4564 } 4565 4566 DO_SIGNAL_CHECK(SR_signum); 4567 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4568} 4569 4570typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4571 4572static os_sigaction_t os_sigaction = NULL; 4573 4574void os::Linux::check_signal_handler(int sig) { 4575 char buf[O_BUFLEN]; 4576 address jvmHandler = NULL; 4577 4578 4579 struct sigaction act; 4580 if (os_sigaction == NULL) { 4581 // only trust the default sigaction, in case it has been interposed 4582 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4583 if (os_sigaction == NULL) return; 4584 } 4585 4586 os_sigaction(sig, (struct sigaction*)NULL, &act); 4587 4588 4589 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4590 4591 address thisHandler = (act.sa_flags & SA_SIGINFO) 4592 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4593 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 4594 4595 4596 switch(sig) { 4597 case SIGSEGV: 4598 case SIGBUS: 4599 case SIGFPE: 4600 case SIGPIPE: 4601 case SIGILL: 4602 case SIGXFSZ: 4603 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4604 break; 4605 4606 case SHUTDOWN1_SIGNAL: 4607 case SHUTDOWN2_SIGNAL: 4608 case SHUTDOWN3_SIGNAL: 4609 case BREAK_SIGNAL: 4610 jvmHandler = (address)user_handler(); 4611 break; 4612 4613 case INTERRUPT_SIGNAL: 4614 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4615 break; 4616 4617 default: 4618 if (sig == SR_signum) { 4619 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4620 } else { 4621 return; 4622 } 4623 break; 4624 } 4625 4626 if (thisHandler != jvmHandler) { 4627 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4628 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4629 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4630 // No need to check this sig any longer 4631 sigaddset(&check_signal_done, sig); 4632 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4633 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4634 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4635 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4636 // No need to check this sig any longer 4637 sigaddset(&check_signal_done, sig); 4638 } 4639 4640 // Dump all the signal 4641 if (sigismember(&check_signal_done, sig)) { 4642 print_signal_handlers(tty, buf, O_BUFLEN); 4643 } 4644} 4645 4646extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 4647 4648extern bool signal_name(int signo, char* buf, size_t len); 4649 4650const char* os::exception_name(int exception_code, char* buf, size_t size) { 4651 if (0 < exception_code && exception_code <= SIGRTMAX) { 4652 // signal 4653 if (!signal_name(exception_code, buf, size)) { 4654 jio_snprintf(buf, size, "SIG%d", exception_code); 4655 } 4656 return buf; 4657 } else { 4658 return NULL; 4659 } 4660} 4661 4662// this is called _before_ the most of global arguments have been parsed 4663void os::init(void) { 4664 char dummy; /* used to get a guess on initial stack address */ 4665// first_hrtime = gethrtime(); 4666 4667 // With LinuxThreads the JavaMain thread pid (primordial thread) 4668 // is different than the pid of the java launcher thread. 4669 // So, on Linux, the launcher thread pid is passed to the VM 4670 // via the sun.java.launcher.pid property. 4671 // Use this property instead of getpid() if it was correctly passed. 4672 // See bug 6351349. 4673 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4674 4675 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4676 4677 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4678 4679 init_random(1234567); 4680 4681 ThreadCritical::initialize(); 4682 4683 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4684 if (Linux::page_size() == -1) { 4685 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4686 strerror(errno))); 4687 } 4688 init_page_sizes((size_t) Linux::page_size()); 4689 4690 Linux::initialize_system_info(); 4691 4692 // main_thread points to the aboriginal thread 4693 Linux::_main_thread = pthread_self(); 4694 4695 Linux::clock_init(); 4696 initial_time_count = os::elapsed_counter(); 4697 pthread_mutex_init(&dl_mutex, NULL); 4698 4699 // If the pagesize of the VM is greater than 8K determine the appropriate 4700 // number of initial guard pages. The user can change this with the 4701 // command line arguments, if needed. 4702 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4703 StackYellowPages = 1; 4704 StackRedPages = 1; 4705 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4706 } 4707} 4708 4709// To install functions for atexit system call 4710extern "C" { 4711 static void perfMemory_exit_helper() { 4712 perfMemory_exit(); 4713 } 4714} 4715 4716// this is called _after_ the global arguments have been parsed 4717jint os::init_2(void) 4718{ 4719 Linux::fast_thread_clock_init(); 4720 4721 // Allocate a single page and mark it as readable for safepoint polling 4722 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4723 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 4724 4725 os::set_polling_page( polling_page ); 4726 4727#ifndef PRODUCT 4728 if(Verbose && PrintMiscellaneous) 4729 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 4730#endif 4731 4732 if (!UseMembar) { 4733 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4734 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4735 os::set_memory_serialize_page( mem_serialize_page ); 4736 4737#ifndef PRODUCT 4738 if(Verbose && PrintMiscellaneous) 4739 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 4740#endif 4741 } 4742 4743 os::large_page_init(); 4744 4745 // initialize suspend/resume support - must do this before signal_sets_init() 4746 if (SR_initialize() != 0) { 4747 perror("SR_initialize failed"); 4748 return JNI_ERR; 4749 } 4750 4751 Linux::signal_sets_init(); 4752 Linux::install_signal_handlers(); 4753 4754 // Check minimum allowable stack size for thread creation and to initialize 4755 // the java system classes, including StackOverflowError - depends on page 4756 // size. Add a page for compiler2 recursion in main thread. 4757 // Add in 2*BytesPerWord times page size to account for VM stack during 4758 // class initialization depending on 32 or 64 bit VM. 4759 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4760 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4761 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4762 4763 size_t threadStackSizeInBytes = ThreadStackSize * K; 4764 if (threadStackSizeInBytes != 0 && 4765 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4766 tty->print_cr("\nThe stack size specified is too small, " 4767 "Specify at least %dk", 4768 os::Linux::min_stack_allowed/ K); 4769 return JNI_ERR; 4770 } 4771 4772 // Make the stack size a multiple of the page size so that 4773 // the yellow/red zones can be guarded. 4774 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4775 vm_page_size())); 4776 4777 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4778 4779 Linux::libpthread_init(); 4780 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4781 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4782 Linux::glibc_version(), Linux::libpthread_version(), 4783 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4784 } 4785 4786 if (UseNUMA) { 4787 if (!Linux::libnuma_init()) { 4788 UseNUMA = false; 4789 } else { 4790 if ((Linux::numa_max_node() < 1)) { 4791 // There's only one node(they start from 0), disable NUMA. 4792 UseNUMA = false; 4793 } 4794 } 4795 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4796 // we can make the adaptive lgrp chunk resizing work. If the user specified 4797 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4798 // disable adaptive resizing. 4799 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4800 if (FLAG_IS_DEFAULT(UseNUMA)) { 4801 UseNUMA = false; 4802 } else { 4803 if (FLAG_IS_DEFAULT(UseLargePages) && 4804 FLAG_IS_DEFAULT(UseSHM) && 4805 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4806 UseLargePages = false; 4807 } else { 4808 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing"); 4809 UseAdaptiveSizePolicy = false; 4810 UseAdaptiveNUMAChunkSizing = false; 4811 } 4812 } 4813 } 4814 if (!UseNUMA && ForceNUMA) { 4815 UseNUMA = true; 4816 } 4817 } 4818 4819 if (MaxFDLimit) { 4820 // set the number of file descriptors to max. print out error 4821 // if getrlimit/setrlimit fails but continue regardless. 4822 struct rlimit nbr_files; 4823 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4824 if (status != 0) { 4825 if (PrintMiscellaneous && (Verbose || WizardMode)) 4826 perror("os::init_2 getrlimit failed"); 4827 } else { 4828 nbr_files.rlim_cur = nbr_files.rlim_max; 4829 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4830 if (status != 0) { 4831 if (PrintMiscellaneous && (Verbose || WizardMode)) 4832 perror("os::init_2 setrlimit failed"); 4833 } 4834 } 4835 } 4836 4837 // Initialize lock used to serialize thread creation (see os::create_thread) 4838 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4839 4840 // at-exit methods are called in the reverse order of their registration. 4841 // atexit functions are called on return from main or as a result of a 4842 // call to exit(3C). There can be only 32 of these functions registered 4843 // and atexit() does not set errno. 4844 4845 if (PerfAllowAtExitRegistration) { 4846 // only register atexit functions if PerfAllowAtExitRegistration is set. 4847 // atexit functions can be delayed until process exit time, which 4848 // can be problematic for embedded VM situations. Embedded VMs should 4849 // call DestroyJavaVM() to assure that VM resources are released. 4850 4851 // note: perfMemory_exit_helper atexit function may be removed in 4852 // the future if the appropriate cleanup code can be added to the 4853 // VM_Exit VMOperation's doit method. 4854 if (atexit(perfMemory_exit_helper) != 0) { 4855 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 4856 } 4857 } 4858 4859 // initialize thread priority policy 4860 prio_init(); 4861 4862 return JNI_OK; 4863} 4864 4865// this is called at the end of vm_initialization 4866void os::init_3(void) 4867{ 4868#ifdef JAVASE_EMBEDDED 4869 // Start the MemNotifyThread 4870 if (LowMemoryProtection) { 4871 MemNotifyThread::start(); 4872 } 4873 return; 4874#endif 4875} 4876 4877// Mark the polling page as unreadable 4878void os::make_polling_page_unreadable(void) { 4879 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4880 fatal("Could not disable polling page"); 4881}; 4882 4883// Mark the polling page as readable 4884void os::make_polling_page_readable(void) { 4885 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4886 fatal("Could not enable polling page"); 4887 } 4888}; 4889 4890int os::active_processor_count() { 4891 // Linux doesn't yet have a (official) notion of processor sets, 4892 // so just return the number of online processors. 4893 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4894 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4895 return online_cpus; 4896} 4897 4898void os::set_native_thread_name(const char *name) { 4899 // Not yet implemented. 4900 return; 4901} 4902 4903bool os::distribute_processes(uint length, uint* distribution) { 4904 // Not yet implemented. 4905 return false; 4906} 4907 4908bool os::bind_to_processor(uint processor_id) { 4909 // Not yet implemented. 4910 return false; 4911} 4912 4913/// 4914 4915void os::SuspendedThreadTask::internal_do_task() { 4916 if (do_suspend(_thread->osthread())) { 4917 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4918 do_task(context); 4919 do_resume(_thread->osthread()); 4920 } 4921} 4922 4923class PcFetcher : public os::SuspendedThreadTask { 4924public: 4925 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4926 ExtendedPC result(); 4927protected: 4928 void do_task(const os::SuspendedThreadTaskContext& context); 4929private: 4930 ExtendedPC _epc; 4931}; 4932 4933ExtendedPC PcFetcher::result() { 4934 guarantee(is_done(), "task is not done yet."); 4935 return _epc; 4936} 4937 4938void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4939 Thread* thread = context.thread(); 4940 OSThread* osthread = thread->osthread(); 4941 if (osthread->ucontext() != NULL) { 4942 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4943 } else { 4944 // NULL context is unexpected, double-check this is the VMThread 4945 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4946 } 4947} 4948 4949// Suspends the target using the signal mechanism and then grabs the PC before 4950// resuming the target. Used by the flat-profiler only 4951ExtendedPC os::get_thread_pc(Thread* thread) { 4952 // Make sure that it is called by the watcher for the VMThread 4953 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4954 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4955 4956 PcFetcher fetcher(thread); 4957 fetcher.run(); 4958 return fetcher.result(); 4959} 4960 4961int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4962{ 4963 if (is_NPTL()) { 4964 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4965 } else { 4966 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4967 // word back to default 64bit precision if condvar is signaled. Java 4968 // wants 53bit precision. Save and restore current value. 4969 int fpu = get_fpu_control_word(); 4970 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4971 set_fpu_control_word(fpu); 4972 return status; 4973 } 4974} 4975 4976//////////////////////////////////////////////////////////////////////////////// 4977// debug support 4978 4979bool os::find(address addr, outputStream* st) { 4980 Dl_info dlinfo; 4981 memset(&dlinfo, 0, sizeof(dlinfo)); 4982 if (dladdr(addr, &dlinfo) != 0) { 4983 st->print(PTR_FORMAT ": ", addr); 4984 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 4985 st->print("%s+%#x", dlinfo.dli_sname, 4986 addr - (intptr_t)dlinfo.dli_saddr); 4987 } else if (dlinfo.dli_fbase != NULL) { 4988 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4989 } else { 4990 st->print("<absolute address>"); 4991 } 4992 if (dlinfo.dli_fname != NULL) { 4993 st->print(" in %s", dlinfo.dli_fname); 4994 } 4995 if (dlinfo.dli_fbase != NULL) { 4996 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4997 } 4998 st->cr(); 4999 5000 if (Verbose) { 5001 // decode some bytes around the PC 5002 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5003 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5004 address lowest = (address) dlinfo.dli_sname; 5005 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5006 if (begin < lowest) begin = lowest; 5007 Dl_info dlinfo2; 5008 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5009 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 5010 end = (address) dlinfo2.dli_saddr; 5011 Disassembler::decode(begin, end, st); 5012 } 5013 return true; 5014 } 5015 return false; 5016} 5017 5018//////////////////////////////////////////////////////////////////////////////// 5019// misc 5020 5021// This does not do anything on Linux. This is basically a hook for being 5022// able to use structured exception handling (thread-local exception filters) 5023// on, e.g., Win32. 5024void 5025os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 5026 JavaCallArguments* args, Thread* thread) { 5027 f(value, method, args, thread); 5028} 5029 5030void os::print_statistics() { 5031} 5032 5033int os::message_box(const char* title, const char* message) { 5034 int i; 5035 fdStream err(defaultStream::error_fd()); 5036 for (i = 0; i < 78; i++) err.print_raw("="); 5037 err.cr(); 5038 err.print_raw_cr(title); 5039 for (i = 0; i < 78; i++) err.print_raw("-"); 5040 err.cr(); 5041 err.print_raw_cr(message); 5042 for (i = 0; i < 78; i++) err.print_raw("="); 5043 err.cr(); 5044 5045 char buf[16]; 5046 // Prevent process from exiting upon "read error" without consuming all CPU 5047 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5048 5049 return buf[0] == 'y' || buf[0] == 'Y'; 5050} 5051 5052int os::stat(const char *path, struct stat *sbuf) { 5053 char pathbuf[MAX_PATH]; 5054 if (strlen(path) > MAX_PATH - 1) { 5055 errno = ENAMETOOLONG; 5056 return -1; 5057 } 5058 os::native_path(strcpy(pathbuf, path)); 5059 return ::stat(pathbuf, sbuf); 5060} 5061 5062bool os::check_heap(bool force) { 5063 return true; 5064} 5065 5066int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 5067 return ::vsnprintf(buf, count, format, args); 5068} 5069 5070// Is a (classpath) directory empty? 5071bool os::dir_is_empty(const char* path) { 5072 DIR *dir = NULL; 5073 struct dirent *ptr; 5074 5075 dir = opendir(path); 5076 if (dir == NULL) return true; 5077 5078 /* Scan the directory */ 5079 bool result = true; 5080 char buf[sizeof(struct dirent) + MAX_PATH]; 5081 while (result && (ptr = ::readdir(dir)) != NULL) { 5082 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5083 result = false; 5084 } 5085 } 5086 closedir(dir); 5087 return result; 5088} 5089 5090// This code originates from JDK's sysOpen and open64_w 5091// from src/solaris/hpi/src/system_md.c 5092 5093#ifndef O_DELETE 5094#define O_DELETE 0x10000 5095#endif 5096 5097// Open a file. Unlink the file immediately after open returns 5098// if the specified oflag has the O_DELETE flag set. 5099// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 5100 5101int os::open(const char *path, int oflag, int mode) { 5102 5103 if (strlen(path) > MAX_PATH - 1) { 5104 errno = ENAMETOOLONG; 5105 return -1; 5106 } 5107 int fd; 5108 int o_delete = (oflag & O_DELETE); 5109 oflag = oflag & ~O_DELETE; 5110 5111 fd = ::open64(path, oflag, mode); 5112 if (fd == -1) return -1; 5113 5114 //If the open succeeded, the file might still be a directory 5115 { 5116 struct stat64 buf64; 5117 int ret = ::fstat64(fd, &buf64); 5118 int st_mode = buf64.st_mode; 5119 5120 if (ret != -1) { 5121 if ((st_mode & S_IFMT) == S_IFDIR) { 5122 errno = EISDIR; 5123 ::close(fd); 5124 return -1; 5125 } 5126 } else { 5127 ::close(fd); 5128 return -1; 5129 } 5130 } 5131 5132 /* 5133 * All file descriptors that are opened in the JVM and not 5134 * specifically destined for a subprocess should have the 5135 * close-on-exec flag set. If we don't set it, then careless 3rd 5136 * party native code might fork and exec without closing all 5137 * appropriate file descriptors (e.g. as we do in closeDescriptors in 5138 * UNIXProcess.c), and this in turn might: 5139 * 5140 * - cause end-of-file to fail to be detected on some file 5141 * descriptors, resulting in mysterious hangs, or 5142 * 5143 * - might cause an fopen in the subprocess to fail on a system 5144 * suffering from bug 1085341. 5145 * 5146 * (Yes, the default setting of the close-on-exec flag is a Unix 5147 * design flaw) 5148 * 5149 * See: 5150 * 1085341: 32-bit stdio routines should support file descriptors >255 5151 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5152 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5153 */ 5154#ifdef FD_CLOEXEC 5155 { 5156 int flags = ::fcntl(fd, F_GETFD); 5157 if (flags != -1) 5158 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5159 } 5160#endif 5161 5162 if (o_delete != 0) { 5163 ::unlink(path); 5164 } 5165 return fd; 5166} 5167 5168 5169// create binary file, rewriting existing file if required 5170int os::create_binary_file(const char* path, bool rewrite_existing) { 5171 int oflags = O_WRONLY | O_CREAT; 5172 if (!rewrite_existing) { 5173 oflags |= O_EXCL; 5174 } 5175 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5176} 5177 5178// return current position of file pointer 5179jlong os::current_file_offset(int fd) { 5180 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5181} 5182 5183// move file pointer to the specified offset 5184jlong os::seek_to_file_offset(int fd, jlong offset) { 5185 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5186} 5187 5188// This code originates from JDK's sysAvailable 5189// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5190 5191int os::available(int fd, jlong *bytes) { 5192 jlong cur, end; 5193 int mode; 5194 struct stat64 buf64; 5195 5196 if (::fstat64(fd, &buf64) >= 0) { 5197 mode = buf64.st_mode; 5198 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5199 /* 5200 * XXX: is the following call interruptible? If so, this might 5201 * need to go through the INTERRUPT_IO() wrapper as for other 5202 * blocking, interruptible calls in this file. 5203 */ 5204 int n; 5205 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5206 *bytes = n; 5207 return 1; 5208 } 5209 } 5210 } 5211 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5212 return 0; 5213 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5214 return 0; 5215 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5216 return 0; 5217 } 5218 *bytes = end - cur; 5219 return 1; 5220} 5221 5222int os::socket_available(int fd, jint *pbytes) { 5223 // Linux doc says EINTR not returned, unlike Solaris 5224 int ret = ::ioctl(fd, FIONREAD, pbytes); 5225 5226 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 5227 // is expected to return 0 on failure and 1 on success to the jdk. 5228 return (ret < 0) ? 0 : 1; 5229} 5230 5231// Map a block of memory. 5232char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5233 char *addr, size_t bytes, bool read_only, 5234 bool allow_exec) { 5235 int prot; 5236 int flags = MAP_PRIVATE; 5237 5238 if (read_only) { 5239 prot = PROT_READ; 5240 } else { 5241 prot = PROT_READ | PROT_WRITE; 5242 } 5243 5244 if (allow_exec) { 5245 prot |= PROT_EXEC; 5246 } 5247 5248 if (addr != NULL) { 5249 flags |= MAP_FIXED; 5250 } 5251 5252 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5253 fd, file_offset); 5254 if (mapped_address == MAP_FAILED) { 5255 return NULL; 5256 } 5257 return mapped_address; 5258} 5259 5260 5261// Remap a block of memory. 5262char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5263 char *addr, size_t bytes, bool read_only, 5264 bool allow_exec) { 5265 // same as map_memory() on this OS 5266 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5267 allow_exec); 5268} 5269 5270 5271// Unmap a block of memory. 5272bool os::pd_unmap_memory(char* addr, size_t bytes) { 5273 return munmap(addr, bytes) == 0; 5274} 5275 5276static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5277 5278static clockid_t thread_cpu_clockid(Thread* thread) { 5279 pthread_t tid = thread->osthread()->pthread_id(); 5280 clockid_t clockid; 5281 5282 // Get thread clockid 5283 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5284 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5285 return clockid; 5286} 5287 5288// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5289// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5290// of a thread. 5291// 5292// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5293// the fast estimate available on the platform. 5294 5295jlong os::current_thread_cpu_time() { 5296 if (os::Linux::supports_fast_thread_cpu_time()) { 5297 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5298 } else { 5299 // return user + sys since the cost is the same 5300 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5301 } 5302} 5303 5304jlong os::thread_cpu_time(Thread* thread) { 5305 // consistent with what current_thread_cpu_time() returns 5306 if (os::Linux::supports_fast_thread_cpu_time()) { 5307 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5308 } else { 5309 return slow_thread_cpu_time(thread, true /* user + sys */); 5310 } 5311} 5312 5313jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5314 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5315 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5316 } else { 5317 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5318 } 5319} 5320 5321jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5322 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5323 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5324 } else { 5325 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5326 } 5327} 5328 5329// 5330// -1 on error. 5331// 5332 5333static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5334 static bool proc_task_unchecked = true; 5335 static const char *proc_stat_path = "/proc/%d/stat"; 5336 pid_t tid = thread->osthread()->thread_id(); 5337 char *s; 5338 char stat[2048]; 5339 int statlen; 5340 char proc_name[64]; 5341 int count; 5342 long sys_time, user_time; 5343 char cdummy; 5344 int idummy; 5345 long ldummy; 5346 FILE *fp; 5347 5348 // The /proc/<tid>/stat aggregates per-process usage on 5349 // new Linux kernels 2.6+ where NPTL is supported. 5350 // The /proc/self/task/<tid>/stat still has the per-thread usage. 5351 // See bug 6328462. 5352 // There possibly can be cases where there is no directory 5353 // /proc/self/task, so we check its availability. 5354 if (proc_task_unchecked && os::Linux::is_NPTL()) { 5355 // This is executed only once 5356 proc_task_unchecked = false; 5357 fp = fopen("/proc/self/task", "r"); 5358 if (fp != NULL) { 5359 proc_stat_path = "/proc/self/task/%d/stat"; 5360 fclose(fp); 5361 } 5362 } 5363 5364 sprintf(proc_name, proc_stat_path, tid); 5365 fp = fopen(proc_name, "r"); 5366 if ( fp == NULL ) return -1; 5367 statlen = fread(stat, 1, 2047, fp); 5368 stat[statlen] = '\0'; 5369 fclose(fp); 5370 5371 // Skip pid and the command string. Note that we could be dealing with 5372 // weird command names, e.g. user could decide to rename java launcher 5373 // to "java 1.4.2 :)", then the stat file would look like 5374 // 1234 (java 1.4.2 :)) R ... ... 5375 // We don't really need to know the command string, just find the last 5376 // occurrence of ")" and then start parsing from there. See bug 4726580. 5377 s = strrchr(stat, ')'); 5378 if (s == NULL ) return -1; 5379 5380 // Skip blank chars 5381 do s++; while (isspace(*s)); 5382 5383 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5384 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5385 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5386 &user_time, &sys_time); 5387 if ( count != 13 ) return -1; 5388 if (user_sys_cpu_time) { 5389 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5390 } else { 5391 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5392 } 5393} 5394 5395void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5396 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5397 info_ptr->may_skip_backward = false; // elapsed time not wall time 5398 info_ptr->may_skip_forward = false; // elapsed time not wall time 5399 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5400} 5401 5402void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5403 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5404 info_ptr->may_skip_backward = false; // elapsed time not wall time 5405 info_ptr->may_skip_forward = false; // elapsed time not wall time 5406 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5407} 5408 5409bool os::is_thread_cpu_time_supported() { 5410 return true; 5411} 5412 5413// System loadavg support. Returns -1 if load average cannot be obtained. 5414// Linux doesn't yet have a (official) notion of processor sets, 5415// so just return the system wide load average. 5416int os::loadavg(double loadavg[], int nelem) { 5417 return ::getloadavg(loadavg, nelem); 5418} 5419 5420void os::pause() { 5421 char filename[MAX_PATH]; 5422 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5423 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5424 } else { 5425 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5426 } 5427 5428 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5429 if (fd != -1) { 5430 struct stat buf; 5431 ::close(fd); 5432 while (::stat(filename, &buf) == 0) { 5433 (void)::poll(NULL, 0, 100); 5434 } 5435 } else { 5436 jio_fprintf(stderr, 5437 "Could not open pause file '%s', continuing immediately.\n", filename); 5438 } 5439} 5440 5441 5442// Refer to the comments in os_solaris.cpp park-unpark. 5443// 5444// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5445// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5446// For specifics regarding the bug see GLIBC BUGID 261237 : 5447// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5448// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5449// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5450// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5451// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5452// and monitorenter when we're using 1-0 locking. All those operations may result in 5453// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5454// of libpthread avoids the problem, but isn't practical. 5455// 5456// Possible remedies: 5457// 5458// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5459// This is palliative and probabilistic, however. If the thread is preempted 5460// between the call to compute_abstime() and pthread_cond_timedwait(), more 5461// than the minimum period may have passed, and the abstime may be stale (in the 5462// past) resultin in a hang. Using this technique reduces the odds of a hang 5463// but the JVM is still vulnerable, particularly on heavily loaded systems. 5464// 5465// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5466// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5467// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5468// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5469// thread. 5470// 5471// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5472// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5473// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5474// This also works well. In fact it avoids kernel-level scalability impediments 5475// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5476// timers in a graceful fashion. 5477// 5478// 4. When the abstime value is in the past it appears that control returns 5479// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5480// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5481// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5482// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5483// It may be possible to avoid reinitialization by checking the return 5484// value from pthread_cond_timedwait(). In addition to reinitializing the 5485// condvar we must establish the invariant that cond_signal() is only called 5486// within critical sections protected by the adjunct mutex. This prevents 5487// cond_signal() from "seeing" a condvar that's in the midst of being 5488// reinitialized or that is corrupt. Sadly, this invariant obviates the 5489// desirable signal-after-unlock optimization that avoids futile context switching. 5490// 5491// I'm also concerned that some versions of NTPL might allocate an auxilliary 5492// structure when a condvar is used or initialized. cond_destroy() would 5493// release the helper structure. Our reinitialize-after-timedwait fix 5494// put excessive stress on malloc/free and locks protecting the c-heap. 5495// 5496// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5497// It may be possible to refine (4) by checking the kernel and NTPL verisons 5498// and only enabling the work-around for vulnerable environments. 5499 5500// utility to compute the abstime argument to timedwait: 5501// millis is the relative timeout time 5502// abstime will be the absolute timeout time 5503// TODO: replace compute_abstime() with unpackTime() 5504 5505static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5506 if (millis < 0) millis = 0; 5507 struct timeval now; 5508 int status = gettimeofday(&now, NULL); 5509 assert(status == 0, "gettimeofday"); 5510 jlong seconds = millis / 1000; 5511 millis %= 1000; 5512 if (seconds > 50000000) { // see man cond_timedwait(3T) 5513 seconds = 50000000; 5514 } 5515 abstime->tv_sec = now.tv_sec + seconds; 5516 long usec = now.tv_usec + millis * 1000; 5517 if (usec >= 1000000) { 5518 abstime->tv_sec += 1; 5519 usec -= 1000000; 5520 } 5521 abstime->tv_nsec = usec * 1000; 5522 return abstime; 5523} 5524 5525 5526// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 5527// Conceptually TryPark() should be equivalent to park(0). 5528 5529int os::PlatformEvent::TryPark() { 5530 for (;;) { 5531 const int v = _Event ; 5532 guarantee ((v == 0) || (v == 1), "invariant") ; 5533 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 5534 } 5535} 5536 5537void os::PlatformEvent::park() { // AKA "down()" 5538 // Invariant: Only the thread associated with the Event/PlatformEvent 5539 // may call park(). 5540 // TODO: assert that _Assoc != NULL or _Assoc == Self 5541 int v ; 5542 for (;;) { 5543 v = _Event ; 5544 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5545 } 5546 guarantee (v >= 0, "invariant") ; 5547 if (v == 0) { 5548 // Do this the hard way by blocking ... 5549 int status = pthread_mutex_lock(_mutex); 5550 assert_status(status == 0, status, "mutex_lock"); 5551 guarantee (_nParked == 0, "invariant") ; 5552 ++ _nParked ; 5553 while (_Event < 0) { 5554 status = pthread_cond_wait(_cond, _mutex); 5555 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5556 // Treat this the same as if the wait was interrupted 5557 if (status == ETIME) { status = EINTR; } 5558 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5559 } 5560 -- _nParked ; 5561 5562 _Event = 0 ; 5563 status = pthread_mutex_unlock(_mutex); 5564 assert_status(status == 0, status, "mutex_unlock"); 5565 // Paranoia to ensure our locked and lock-free paths interact 5566 // correctly with each other. 5567 OrderAccess::fence(); 5568 } 5569 guarantee (_Event >= 0, "invariant") ; 5570} 5571 5572int os::PlatformEvent::park(jlong millis) { 5573 guarantee (_nParked == 0, "invariant") ; 5574 5575 int v ; 5576 for (;;) { 5577 v = _Event ; 5578 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5579 } 5580 guarantee (v >= 0, "invariant") ; 5581 if (v != 0) return OS_OK ; 5582 5583 // We do this the hard way, by blocking the thread. 5584 // Consider enforcing a minimum timeout value. 5585 struct timespec abst; 5586 compute_abstime(&abst, millis); 5587 5588 int ret = OS_TIMEOUT; 5589 int status = pthread_mutex_lock(_mutex); 5590 assert_status(status == 0, status, "mutex_lock"); 5591 guarantee (_nParked == 0, "invariant") ; 5592 ++_nParked ; 5593 5594 // Object.wait(timo) will return because of 5595 // (a) notification 5596 // (b) timeout 5597 // (c) thread.interrupt 5598 // 5599 // Thread.interrupt and object.notify{All} both call Event::set. 5600 // That is, we treat thread.interrupt as a special case of notification. 5601 // The underlying Solaris implementation, cond_timedwait, admits 5602 // spurious/premature wakeups, but the JLS/JVM spec prevents the 5603 // JVM from making those visible to Java code. As such, we must 5604 // filter out spurious wakeups. We assume all ETIME returns are valid. 5605 // 5606 // TODO: properly differentiate simultaneous notify+interrupt. 5607 // In that case, we should propagate the notify to another waiter. 5608 5609 while (_Event < 0) { 5610 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5611 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5612 pthread_cond_destroy (_cond); 5613 pthread_cond_init (_cond, NULL) ; 5614 } 5615 assert_status(status == 0 || status == EINTR || 5616 status == ETIME || status == ETIMEDOUT, 5617 status, "cond_timedwait"); 5618 if (!FilterSpuriousWakeups) break ; // previous semantics 5619 if (status == ETIME || status == ETIMEDOUT) break ; 5620 // We consume and ignore EINTR and spurious wakeups. 5621 } 5622 --_nParked ; 5623 if (_Event >= 0) { 5624 ret = OS_OK; 5625 } 5626 _Event = 0 ; 5627 status = pthread_mutex_unlock(_mutex); 5628 assert_status(status == 0, status, "mutex_unlock"); 5629 assert (_nParked == 0, "invariant") ; 5630 // Paranoia to ensure our locked and lock-free paths interact 5631 // correctly with each other. 5632 OrderAccess::fence(); 5633 return ret; 5634} 5635 5636void os::PlatformEvent::unpark() { 5637 // Transitions for _Event: 5638 // 0 :=> 1 5639 // 1 :=> 1 5640 // -1 :=> either 0 or 1; must signal target thread 5641 // That is, we can safely transition _Event from -1 to either 5642 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back 5643 // unpark() calls. 5644 // See also: "Semaphores in Plan 9" by Mullender & Cox 5645 // 5646 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5647 // that it will take two back-to-back park() calls for the owning 5648 // thread to block. This has the benefit of forcing a spurious return 5649 // from the first park() call after an unpark() call which will help 5650 // shake out uses of park() and unpark() without condition variables. 5651 5652 if (Atomic::xchg(1, &_Event) >= 0) return; 5653 5654 // Wait for the thread associated with the event to vacate 5655 int status = pthread_mutex_lock(_mutex); 5656 assert_status(status == 0, status, "mutex_lock"); 5657 int AnyWaiters = _nParked; 5658 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5659 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5660 AnyWaiters = 0; 5661 pthread_cond_signal(_cond); 5662 } 5663 status = pthread_mutex_unlock(_mutex); 5664 assert_status(status == 0, status, "mutex_unlock"); 5665 if (AnyWaiters != 0) { 5666 status = pthread_cond_signal(_cond); 5667 assert_status(status == 0, status, "cond_signal"); 5668 } 5669 5670 // Note that we signal() _after dropping the lock for "immortal" Events. 5671 // This is safe and avoids a common class of futile wakeups. In rare 5672 // circumstances this can cause a thread to return prematurely from 5673 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5674 // simply re-test the condition and re-park itself. 5675} 5676 5677 5678// JSR166 5679// ------------------------------------------------------- 5680 5681/* 5682 * The solaris and linux implementations of park/unpark are fairly 5683 * conservative for now, but can be improved. They currently use a 5684 * mutex/condvar pair, plus a a count. 5685 * Park decrements count if > 0, else does a condvar wait. Unpark 5686 * sets count to 1 and signals condvar. Only one thread ever waits 5687 * on the condvar. Contention seen when trying to park implies that someone 5688 * is unparking you, so don't wait. And spurious returns are fine, so there 5689 * is no need to track notifications. 5690 */ 5691 5692#define MAX_SECS 100000000 5693/* 5694 * This code is common to linux and solaris and will be moved to a 5695 * common place in dolphin. 5696 * 5697 * The passed in time value is either a relative time in nanoseconds 5698 * or an absolute time in milliseconds. Either way it has to be unpacked 5699 * into suitable seconds and nanoseconds components and stored in the 5700 * given timespec structure. 5701 * Given time is a 64-bit value and the time_t used in the timespec is only 5702 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 5703 * overflow if times way in the future are given. Further on Solaris versions 5704 * prior to 10 there is a restriction (see cond_timedwait) that the specified 5705 * number of seconds, in abstime, is less than current_time + 100,000,000. 5706 * As it will be 28 years before "now + 100000000" will overflow we can 5707 * ignore overflow and just impose a hard-limit on seconds using the value 5708 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 5709 * years from "now". 5710 */ 5711 5712static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5713 assert (time > 0, "convertTime"); 5714 5715 struct timeval now; 5716 int status = gettimeofday(&now, NULL); 5717 assert(status == 0, "gettimeofday"); 5718 5719 time_t max_secs = now.tv_sec + MAX_SECS; 5720 5721 if (isAbsolute) { 5722 jlong secs = time / 1000; 5723 if (secs > max_secs) { 5724 absTime->tv_sec = max_secs; 5725 } 5726 else { 5727 absTime->tv_sec = secs; 5728 } 5729 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5730 } 5731 else { 5732 jlong secs = time / NANOSECS_PER_SEC; 5733 if (secs >= MAX_SECS) { 5734 absTime->tv_sec = max_secs; 5735 absTime->tv_nsec = 0; 5736 } 5737 else { 5738 absTime->tv_sec = now.tv_sec + secs; 5739 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5740 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5741 absTime->tv_nsec -= NANOSECS_PER_SEC; 5742 ++absTime->tv_sec; // note: this must be <= max_secs 5743 } 5744 } 5745 } 5746 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5747 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5748 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5749 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5750} 5751 5752void Parker::park(bool isAbsolute, jlong time) { 5753 // Ideally we'd do something useful while spinning, such 5754 // as calling unpackTime(). 5755 5756 // Optional fast-path check: 5757 // Return immediately if a permit is available. 5758 // We depend on Atomic::xchg() having full barrier semantics 5759 // since we are doing a lock-free update to _counter. 5760 if (Atomic::xchg(0, &_counter) > 0) return; 5761 5762 Thread* thread = Thread::current(); 5763 assert(thread->is_Java_thread(), "Must be JavaThread"); 5764 JavaThread *jt = (JavaThread *)thread; 5765 5766 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5767 // Check interrupt before trying to wait 5768 if (Thread::is_interrupted(thread, false)) { 5769 return; 5770 } 5771 5772 // Next, demultiplex/decode time arguments 5773 timespec absTime; 5774 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5775 return; 5776 } 5777 if (time > 0) { 5778 unpackTime(&absTime, isAbsolute, time); 5779 } 5780 5781 5782 // Enter safepoint region 5783 // Beware of deadlocks such as 6317397. 5784 // The per-thread Parker:: mutex is a classic leaf-lock. 5785 // In particular a thread must never block on the Threads_lock while 5786 // holding the Parker:: mutex. If safepoints are pending both the 5787 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5788 ThreadBlockInVM tbivm(jt); 5789 5790 // Don't wait if cannot get lock since interference arises from 5791 // unblocking. Also. check interrupt before trying wait 5792 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5793 return; 5794 } 5795 5796 int status ; 5797 if (_counter > 0) { // no wait needed 5798 _counter = 0; 5799 status = pthread_mutex_unlock(_mutex); 5800 assert (status == 0, "invariant") ; 5801 // Paranoia to ensure our locked and lock-free paths interact 5802 // correctly with each other and Java-level accesses. 5803 OrderAccess::fence(); 5804 return; 5805 } 5806 5807#ifdef ASSERT 5808 // Don't catch signals while blocked; let the running threads have the signals. 5809 // (This allows a debugger to break into the running thread.) 5810 sigset_t oldsigs; 5811 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5812 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5813#endif 5814 5815 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5816 jt->set_suspend_equivalent(); 5817 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5818 5819 if (time == 0) { 5820 status = pthread_cond_wait (_cond, _mutex) ; 5821 } else { 5822 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 5823 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5824 pthread_cond_destroy (_cond) ; 5825 pthread_cond_init (_cond, NULL); 5826 } 5827 } 5828 assert_status(status == 0 || status == EINTR || 5829 status == ETIME || status == ETIMEDOUT, 5830 status, "cond_timedwait"); 5831 5832#ifdef ASSERT 5833 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5834#endif 5835 5836 _counter = 0 ; 5837 status = pthread_mutex_unlock(_mutex) ; 5838 assert_status(status == 0, status, "invariant") ; 5839 // Paranoia to ensure our locked and lock-free paths interact 5840 // correctly with each other and Java-level accesses. 5841 OrderAccess::fence(); 5842 5843 // If externally suspended while waiting, re-suspend 5844 if (jt->handle_special_suspend_equivalent_condition()) { 5845 jt->java_suspend_self(); 5846 } 5847} 5848 5849void Parker::unpark() { 5850 int s, status ; 5851 status = pthread_mutex_lock(_mutex); 5852 assert (status == 0, "invariant") ; 5853 s = _counter; 5854 _counter = 1; 5855 if (s < 1) { 5856 if (WorkAroundNPTLTimedWaitHang) { 5857 status = pthread_cond_signal (_cond) ; 5858 assert (status == 0, "invariant") ; 5859 status = pthread_mutex_unlock(_mutex); 5860 assert (status == 0, "invariant") ; 5861 } else { 5862 status = pthread_mutex_unlock(_mutex); 5863 assert (status == 0, "invariant") ; 5864 status = pthread_cond_signal (_cond) ; 5865 assert (status == 0, "invariant") ; 5866 } 5867 } else { 5868 pthread_mutex_unlock(_mutex); 5869 assert (status == 0, "invariant") ; 5870 } 5871} 5872 5873 5874extern char** environ; 5875 5876#ifndef __NR_fork 5877#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5878#endif 5879 5880#ifndef __NR_execve 5881#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5882#endif 5883 5884// Run the specified command in a separate process. Return its exit value, 5885// or -1 on failure (e.g. can't fork a new process). 5886// Unlike system(), this function can be called from signal handler. It 5887// doesn't block SIGINT et al. 5888int os::fork_and_exec(char* cmd) { 5889 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5890 5891 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5892 // pthread_atfork handlers and reset pthread library. All we need is a 5893 // separate process to execve. Make a direct syscall to fork process. 5894 // On IA64 there's no fork syscall, we have to use fork() and hope for 5895 // the best... 5896 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5897 IA64_ONLY(fork();) 5898 5899 if (pid < 0) { 5900 // fork failed 5901 return -1; 5902 5903 } else if (pid == 0) { 5904 // child process 5905 5906 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5907 // first to kill every thread on the thread list. Because this list is 5908 // not reset by fork() (see notes above), execve() will instead kill 5909 // every thread in the parent process. We know this is the only thread 5910 // in the new process, so make a system call directly. 5911 // IA64 should use normal execve() from glibc to match the glibc fork() 5912 // above. 5913 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5914 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5915 5916 // execve failed 5917 _exit(-1); 5918 5919 } else { 5920 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5921 // care about the actual exit code, for now. 5922 5923 int status; 5924 5925 // Wait for the child process to exit. This returns immediately if 5926 // the child has already exited. */ 5927 while (waitpid(pid, &status, 0) < 0) { 5928 switch (errno) { 5929 case ECHILD: return 0; 5930 case EINTR: break; 5931 default: return -1; 5932 } 5933 } 5934 5935 if (WIFEXITED(status)) { 5936 // The child exited normally; get its exit code. 5937 return WEXITSTATUS(status); 5938 } else if (WIFSIGNALED(status)) { 5939 // The child exited because of a signal 5940 // The best value to return is 0x80 + signal number, 5941 // because that is what all Unix shells do, and because 5942 // it allows callers to distinguish between process exit and 5943 // process death by signal. 5944 return 0x80 + WTERMSIG(status); 5945 } else { 5946 // Unknown exit code; pass it through 5947 return status; 5948 } 5949 } 5950} 5951 5952// is_headless_jre() 5953// 5954// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5955// in order to report if we are running in a headless jre 5956// 5957// Since JDK8 xawt/libmawt.so was moved into the same directory 5958// as libawt.so, and renamed libawt_xawt.so 5959// 5960bool os::is_headless_jre() { 5961 struct stat statbuf; 5962 char buf[MAXPATHLEN]; 5963 char libmawtpath[MAXPATHLEN]; 5964 const char *xawtstr = "/xawt/libmawt.so"; 5965 const char *new_xawtstr = "/libawt_xawt.so"; 5966 char *p; 5967 5968 // Get path to libjvm.so 5969 os::jvm_path(buf, sizeof(buf)); 5970 5971 // Get rid of libjvm.so 5972 p = strrchr(buf, '/'); 5973 if (p == NULL) return false; 5974 else *p = '\0'; 5975 5976 // Get rid of client or server 5977 p = strrchr(buf, '/'); 5978 if (p == NULL) return false; 5979 else *p = '\0'; 5980 5981 // check xawt/libmawt.so 5982 strcpy(libmawtpath, buf); 5983 strcat(libmawtpath, xawtstr); 5984 if (::stat(libmawtpath, &statbuf) == 0) return false; 5985 5986 // check libawt_xawt.so 5987 strcpy(libmawtpath, buf); 5988 strcat(libmawtpath, new_xawtstr); 5989 if (::stat(libmawtpath, &statbuf) == 0) return false; 5990 5991 return true; 5992} 5993 5994// Get the default path to the core file 5995// Returns the length of the string 5996int os::get_core_path(char* buffer, size_t bufferSize) { 5997 const char* p = get_current_directory(buffer, bufferSize); 5998 5999 if (p == NULL) { 6000 assert(p != NULL, "failed to get current directory"); 6001 return 0; 6002 } 6003 6004 return strlen(buffer); 6005} 6006 6007#ifdef JAVASE_EMBEDDED 6008// 6009// A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 6010// 6011MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 6012 6013// ctor 6014// 6015MemNotifyThread::MemNotifyThread(int fd): Thread() { 6016 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 6017 _fd = fd; 6018 6019 if (os::create_thread(this, os::os_thread)) { 6020 _memnotify_thread = this; 6021 os::set_priority(this, NearMaxPriority); 6022 os::start_thread(this); 6023 } 6024} 6025 6026// Where all the work gets done 6027// 6028void MemNotifyThread::run() { 6029 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 6030 6031 // Set up the select arguments 6032 fd_set rfds; 6033 if (_fd != -1) { 6034 FD_ZERO(&rfds); 6035 FD_SET(_fd, &rfds); 6036 } 6037 6038 // Now wait for the mem_notify device to wake up 6039 while (1) { 6040 // Wait for the mem_notify device to signal us.. 6041 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 6042 if (rc == -1) { 6043 perror("select!\n"); 6044 break; 6045 } else if (rc) { 6046 //ssize_t free_before = os::available_memory(); 6047 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 6048 6049 // The kernel is telling us there is not much memory left... 6050 // try to do something about that 6051 6052 // If we are not already in a GC, try one. 6053 if (!Universe::heap()->is_gc_active()) { 6054 Universe::heap()->collect(GCCause::_allocation_failure); 6055 6056 //ssize_t free_after = os::available_memory(); 6057 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 6058 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 6059 } 6060 // We might want to do something like the following if we find the GC's are not helping... 6061 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 6062 } 6063 } 6064} 6065 6066// 6067// See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 6068// 6069void MemNotifyThread::start() { 6070 int fd; 6071 fd = open ("/dev/mem_notify", O_RDONLY, 0); 6072 if (fd < 0) { 6073 return; 6074 } 6075 6076 if (memnotify_thread() == NULL) { 6077 new MemNotifyThread(fd); 6078 } 6079} 6080 6081#endif // JAVASE_EMBEDDED 6082 6083 6084/////////////// Unit tests /////////////// 6085 6086#ifndef PRODUCT 6087 6088#define test_log(...) \ 6089 do {\ 6090 if (VerboseInternalVMTests) { \ 6091 tty->print_cr(__VA_ARGS__); \ 6092 tty->flush(); \ 6093 }\ 6094 } while (false) 6095 6096class TestReserveMemorySpecial : AllStatic { 6097 public: 6098 static void small_page_write(void* addr, size_t size) { 6099 size_t page_size = os::vm_page_size(); 6100 6101 char* end = (char*)addr + size; 6102 for (char* p = (char*)addr; p < end; p += page_size) { 6103 *p = 1; 6104 } 6105 } 6106 6107 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6108 if (!UseHugeTLBFS) { 6109 return; 6110 } 6111 6112 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6113 6114 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6115 6116 if (addr != NULL) { 6117 small_page_write(addr, size); 6118 6119 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6120 } 6121 } 6122 6123 static void test_reserve_memory_special_huge_tlbfs_only() { 6124 if (!UseHugeTLBFS) { 6125 return; 6126 } 6127 6128 size_t lp = os::large_page_size(); 6129 6130 for (size_t size = lp; size <= lp * 10; size += lp) { 6131 test_reserve_memory_special_huge_tlbfs_only(size); 6132 } 6133 } 6134 6135 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) { 6136 if (!UseHugeTLBFS) { 6137 return; 6138 } 6139 6140 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")", 6141 size, alignment); 6142 6143 assert(size >= os::large_page_size(), "Incorrect input to test"); 6144 6145 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6146 6147 if (addr != NULL) { 6148 small_page_write(addr, size); 6149 6150 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6151 } 6152 } 6153 6154 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) { 6155 size_t lp = os::large_page_size(); 6156 size_t ag = os::vm_allocation_granularity(); 6157 6158 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6159 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment); 6160 } 6161 } 6162 6163 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6164 size_t lp = os::large_page_size(); 6165 size_t ag = os::vm_allocation_granularity(); 6166 6167 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp); 6168 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag); 6169 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2); 6170 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2); 6171 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag); 6172 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag); 6173 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2); 6174 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10); 6175 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2); 6176 } 6177 6178 static void test_reserve_memory_special_huge_tlbfs() { 6179 if (!UseHugeTLBFS) { 6180 return; 6181 } 6182 6183 test_reserve_memory_special_huge_tlbfs_only(); 6184 test_reserve_memory_special_huge_tlbfs_mixed(); 6185 } 6186 6187 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6188 if (!UseSHM) { 6189 return; 6190 } 6191 6192 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6193 6194 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6195 6196 if (addr != NULL) { 6197 assert(is_ptr_aligned(addr, alignment), "Check"); 6198 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6199 6200 small_page_write(addr, size); 6201 6202 os::Linux::release_memory_special_shm(addr, size); 6203 } 6204 } 6205 6206 static void test_reserve_memory_special_shm() { 6207 size_t lp = os::large_page_size(); 6208 size_t ag = os::vm_allocation_granularity(); 6209 6210 for (size_t size = ag; size < lp * 3; size += ag) { 6211 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6212 test_reserve_memory_special_shm(size, alignment); 6213 } 6214 } 6215 } 6216 6217 static void test() { 6218 test_reserve_memory_special_huge_tlbfs(); 6219 test_reserve_memory_special_shm(); 6220 } 6221}; 6222 6223void TestReserveMemorySpecial_test() { 6224 TestReserveMemorySpecial::test(); 6225} 6226 6227#endif 6228