1 /* 2 * Copyright (c) 1999, 2016, 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 "logging/log.hpp" 36 #include "memory/allocation.inline.hpp" 37 #include "memory/filemap.hpp" 38 #include "mutex_linux.inline.hpp" 39 #include "oops/oop.inline.hpp" 40 #include "os_linux.inline.hpp" 41 #include "os_share_linux.hpp" 42 #include "prims/jniFastGetField.hpp" 43 #include "prims/jvm.h" 44 #include "prims/jvm_misc.hpp" 45 #include "runtime/arguments.hpp" 46 #include "runtime/atomic.inline.hpp" 47 #include "runtime/extendedPC.hpp" 48 #include "runtime/globals.hpp" 49 #include "runtime/interfaceSupport.hpp" 50 #include "runtime/init.hpp" 51 #include "runtime/java.hpp" 52 #include "runtime/javaCalls.hpp" 53 #include "runtime/mutexLocker.hpp" 54 #include "runtime/objectMonitor.hpp" 55 #include "runtime/orderAccess.inline.hpp" 56 #include "runtime/osThread.hpp" 57 #include "runtime/perfMemory.hpp" 58 #include "runtime/sharedRuntime.hpp" 59 #include "runtime/statSampler.hpp" 60 #include "runtime/stubRoutines.hpp" 61 #include "runtime/thread.inline.hpp" 62 #include "runtime/threadCritical.hpp" 63 #include "runtime/timer.hpp" 64 #include "semaphore_posix.hpp" 65 #include "services/attachListener.hpp" 66 #include "services/memTracker.hpp" 67 #include "services/runtimeService.hpp" 68 #include "utilities/decoder.hpp" 69 #include "utilities/defaultStream.hpp" 70 #include "utilities/events.hpp" 71 #include "utilities/elfFile.hpp" 72 #include "utilities/growableArray.hpp" 73 #include "utilities/macros.hpp" 74 #include "utilities/vmError.hpp" 75 76 // put OS-includes here 77 # include <sys/types.h> 78 # include <sys/mman.h> 79 # include <sys/stat.h> 80 # include <sys/select.h> 81 # include <pthread.h> 82 # include <signal.h> 83 # include <errno.h> 84 # include <dlfcn.h> 85 # include <stdio.h> 86 # include <unistd.h> 87 # include <sys/resource.h> 88 # include <pthread.h> 89 # include <sys/stat.h> 90 # include <sys/time.h> 91 # include <sys/times.h> 92 # include <sys/utsname.h> 93 # include <sys/socket.h> 94 # include <sys/wait.h> 95 # include <pwd.h> 96 # include <poll.h> 97 # include <semaphore.h> 98 # include <fcntl.h> 99 # include <string.h> 100 # include <syscall.h> 101 # include <sys/sysinfo.h> 102 # include <gnu/libc-version.h> 103 # include <sys/ipc.h> 104 # include <sys/shm.h> 105 # include <link.h> 106 # include <stdint.h> 107 # include <inttypes.h> 108 # include <sys/ioctl.h> 109 110 #ifndef _GNU_SOURCE 111 #define _GNU_SOURCE 112 #include <sched.h> 113 #undef _GNU_SOURCE 114 #else 115 #include <sched.h> 116 #endif 117 118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 119 // getrusage() is prepared to handle the associated failure. 120 #ifndef RUSAGE_THREAD 121 #define RUSAGE_THREAD (1) /* only the calling thread */ 122 #endif 123 124 #define MAX_PATH (2 * K) 125 126 #define MAX_SECS 100000000 127 128 // for timer info max values which include all bits 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 130 131 #define LARGEPAGES_BIT (1 << 6) 132 //////////////////////////////////////////////////////////////////////////////// 133 // global variables 134 julong os::Linux::_physical_memory = 0; 135 136 address os::Linux::_initial_thread_stack_bottom = NULL; 137 uintptr_t os::Linux::_initial_thread_stack_size = 0; 138 139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL; 142 Mutex* os::Linux::_createThread_lock = NULL; 143 pthread_t os::Linux::_main_thread; 144 int os::Linux::_page_size = -1; 145 const int os::Linux::_vm_default_page_size = (8 * K); 146 bool os::Linux::_supports_fast_thread_cpu_time = false; 147 uint32_t os::Linux::_os_version = 0; 148 const char * os::Linux::_glibc_version = NULL; 149 const char * os::Linux::_libpthread_version = NULL; 150 pthread_condattr_t os::Linux::_condattr[1]; 151 152 static jlong initial_time_count=0; 153 154 static int clock_tics_per_sec = 100; 155 156 // For diagnostics to print a message once. see run_periodic_checks 157 static sigset_t check_signal_done; 158 static bool check_signals = true; 159 160 // Signal number used to suspend/resume a thread 161 162 // do not use any signal number less than SIGSEGV, see 4355769 163 static int SR_signum = SIGUSR2; 164 sigset_t SR_sigset; 165 166 // Declarations 167 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 168 169 // utility functions 170 171 static int SR_initialize(); 172 173 julong os::available_memory() { 174 return Linux::available_memory(); 175 } 176 177 julong os::Linux::available_memory() { 178 // values in struct sysinfo are "unsigned long" 179 struct sysinfo si; 180 sysinfo(&si); 181 182 return (julong)si.freeram * si.mem_unit; 183 } 184 185 julong os::physical_memory() { 186 return Linux::physical_memory(); 187 } 188 189 // Return true if user is running as root. 190 191 bool os::have_special_privileges() { 192 static bool init = false; 193 static bool privileges = false; 194 if (!init) { 195 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 196 init = true; 197 } 198 return privileges; 199 } 200 201 202 #ifndef SYS_gettid 203 // i386: 224, ia64: 1105, amd64: 186, sparc 143 204 #ifdef __ia64__ 205 #define SYS_gettid 1105 206 #else 207 #ifdef __i386__ 208 #define SYS_gettid 224 209 #else 210 #ifdef __amd64__ 211 #define SYS_gettid 186 212 #else 213 #ifdef __sparc__ 214 #define SYS_gettid 143 215 #else 216 #error define gettid for the arch 217 #endif 218 #endif 219 #endif 220 #endif 221 #endif 222 223 // Cpu architecture string 224 static char cpu_arch[] = HOTSPOT_LIB_ARCH; 225 226 227 // pid_t gettid() 228 // 229 // Returns the kernel thread id of the currently running thread. Kernel 230 // thread id is used to access /proc. 231 pid_t os::Linux::gettid() { 232 int rslt = syscall(SYS_gettid); 233 assert(rslt != -1, "must be."); // old linuxthreads implementation? 234 return (pid_t)rslt; 235 } 236 237 // Most versions of linux have a bug where the number of processors are 238 // determined by looking at the /proc file system. In a chroot environment, 239 // the system call returns 1. This causes the VM to act as if it is 240 // a single processor and elide locking (see is_MP() call). 241 static bool unsafe_chroot_detected = false; 242 static const char *unstable_chroot_error = "/proc file system not found.\n" 243 "Java may be unstable running multithreaded in a chroot " 244 "environment on Linux when /proc filesystem is not mounted."; 245 246 void os::Linux::initialize_system_info() { 247 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 248 if (processor_count() == 1) { 249 pid_t pid = os::Linux::gettid(); 250 char fname[32]; 251 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 252 FILE *fp = fopen(fname, "r"); 253 if (fp == NULL) { 254 unsafe_chroot_detected = true; 255 } else { 256 fclose(fp); 257 } 258 } 259 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 260 assert(processor_count() > 0, "linux error"); 261 } 262 263 void os::init_system_properties_values() { 264 // The next steps are taken in the product version: 265 // 266 // Obtain the JAVA_HOME value from the location of libjvm.so. 267 // This library should be located at: 268 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 269 // 270 // If "/jre/lib/" appears at the right place in the path, then we 271 // assume libjvm.so is installed in a JDK and we use this path. 272 // 273 // Otherwise exit with message: "Could not create the Java virtual machine." 274 // 275 // The following extra steps are taken in the debugging version: 276 // 277 // If "/jre/lib/" does NOT appear at the right place in the path 278 // instead of exit check for $JAVA_HOME environment variable. 279 // 280 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 281 // then we append a fake suffix "hotspot/libjvm.so" to this path so 282 // it looks like libjvm.so is installed there 283 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 284 // 285 // Otherwise exit. 286 // 287 // Important note: if the location of libjvm.so changes this 288 // code needs to be changed accordingly. 289 290 // See ld(1): 291 // The linker uses the following search paths to locate required 292 // shared libraries: 293 // 1: ... 294 // ... 295 // 7: The default directories, normally /lib and /usr/lib. 296 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 297 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 298 #else 299 #define DEFAULT_LIBPATH "/lib:/usr/lib" 300 #endif 301 302 // Base path of extensions installed on the system. 303 #define SYS_EXT_DIR "/usr/java/packages" 304 #define EXTENSIONS_DIR "/lib/ext" 305 306 // Buffer that fits several sprintfs. 307 // Note that the space for the colon and the trailing null are provided 308 // by the nulls included by the sizeof operator. 309 const size_t bufsize = 310 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 311 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 312 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 313 314 // sysclasspath, java_home, dll_dir 315 { 316 char *pslash; 317 os::jvm_path(buf, bufsize); 318 319 // Found the full path to libjvm.so. 320 // Now cut the path to <java_home>/jre if we can. 321 pslash = strrchr(buf, '/'); 322 if (pslash != NULL) { 323 *pslash = '\0'; // Get rid of /libjvm.so. 324 } 325 pslash = strrchr(buf, '/'); 326 if (pslash != NULL) { 327 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 328 } 329 Arguments::set_dll_dir(buf); 330 331 if (pslash != NULL) { 332 pslash = strrchr(buf, '/'); 333 if (pslash != NULL) { 334 *pslash = '\0'; // Get rid of /<arch>. 335 pslash = strrchr(buf, '/'); 336 if (pslash != NULL) { 337 *pslash = '\0'; // Get rid of /lib. 338 } 339 } 340 } 341 Arguments::set_java_home(buf); 342 set_boot_path('/', ':'); 343 } 344 345 // Where to look for native libraries. 346 // 347 // Note: Due to a legacy implementation, most of the library path 348 // is set in the launcher. This was to accomodate linking restrictions 349 // on legacy Linux implementations (which are no longer supported). 350 // Eventually, all the library path setting will be done here. 351 // 352 // However, to prevent the proliferation of improperly built native 353 // libraries, the new path component /usr/java/packages is added here. 354 // Eventually, all the library path setting will be done here. 355 { 356 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 357 // should always exist (until the legacy problem cited above is 358 // addressed). 359 const char *v = ::getenv("LD_LIBRARY_PATH"); 360 const char *v_colon = ":"; 361 if (v == NULL) { v = ""; v_colon = ""; } 362 // That's +1 for the colon and +1 for the trailing '\0'. 363 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 364 strlen(v) + 1 + 365 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1, 366 mtInternal); 367 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch); 368 Arguments::set_library_path(ld_library_path); 369 FREE_C_HEAP_ARRAY(char, ld_library_path); 370 } 371 372 // Extensions directories. 373 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 374 Arguments::set_ext_dirs(buf); 375 376 FREE_C_HEAP_ARRAY(char, buf); 377 378 #undef DEFAULT_LIBPATH 379 #undef SYS_EXT_DIR 380 #undef EXTENSIONS_DIR 381 } 382 383 //////////////////////////////////////////////////////////////////////////////// 384 // breakpoint support 385 386 void os::breakpoint() { 387 BREAKPOINT; 388 } 389 390 extern "C" void breakpoint() { 391 // use debugger to set breakpoint here 392 } 393 394 //////////////////////////////////////////////////////////////////////////////// 395 // signal support 396 397 debug_only(static bool signal_sets_initialized = false); 398 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 399 400 bool os::Linux::is_sig_ignored(int sig) { 401 struct sigaction oact; 402 sigaction(sig, (struct sigaction*)NULL, &oact); 403 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 404 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 405 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) { 406 return true; 407 } else { 408 return false; 409 } 410 } 411 412 void os::Linux::signal_sets_init() { 413 // Should also have an assertion stating we are still single-threaded. 414 assert(!signal_sets_initialized, "Already initialized"); 415 // Fill in signals that are necessarily unblocked for all threads in 416 // the VM. Currently, we unblock the following signals: 417 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 418 // by -Xrs (=ReduceSignalUsage)); 419 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 420 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 421 // the dispositions or masks wrt these signals. 422 // Programs embedding the VM that want to use the above signals for their 423 // own purposes must, at this time, use the "-Xrs" option to prevent 424 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 425 // (See bug 4345157, and other related bugs). 426 // In reality, though, unblocking these signals is really a nop, since 427 // these signals are not blocked by default. 428 sigemptyset(&unblocked_sigs); 429 sigemptyset(&allowdebug_blocked_sigs); 430 sigaddset(&unblocked_sigs, SIGILL); 431 sigaddset(&unblocked_sigs, SIGSEGV); 432 sigaddset(&unblocked_sigs, SIGBUS); 433 sigaddset(&unblocked_sigs, SIGFPE); 434 #if defined(PPC64) 435 sigaddset(&unblocked_sigs, SIGTRAP); 436 #endif 437 sigaddset(&unblocked_sigs, SR_signum); 438 439 if (!ReduceSignalUsage) { 440 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 441 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 442 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 443 } 444 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 445 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 446 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 447 } 448 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 449 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 450 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 451 } 452 } 453 // Fill in signals that are blocked by all but the VM thread. 454 sigemptyset(&vm_sigs); 455 if (!ReduceSignalUsage) { 456 sigaddset(&vm_sigs, BREAK_SIGNAL); 457 } 458 debug_only(signal_sets_initialized = true); 459 460 } 461 462 // These are signals that are unblocked while a thread is running Java. 463 // (For some reason, they get blocked by default.) 464 sigset_t* os::Linux::unblocked_signals() { 465 assert(signal_sets_initialized, "Not initialized"); 466 return &unblocked_sigs; 467 } 468 469 // These are the signals that are blocked while a (non-VM) thread is 470 // running Java. Only the VM thread handles these signals. 471 sigset_t* os::Linux::vm_signals() { 472 assert(signal_sets_initialized, "Not initialized"); 473 return &vm_sigs; 474 } 475 476 // These are signals that are blocked during cond_wait to allow debugger in 477 sigset_t* os::Linux::allowdebug_blocked_signals() { 478 assert(signal_sets_initialized, "Not initialized"); 479 return &allowdebug_blocked_sigs; 480 } 481 482 void os::Linux::hotspot_sigmask(Thread* thread) { 483 484 //Save caller's signal mask before setting VM signal mask 485 sigset_t caller_sigmask; 486 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 487 488 OSThread* osthread = thread->osthread(); 489 osthread->set_caller_sigmask(caller_sigmask); 490 491 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 492 493 if (!ReduceSignalUsage) { 494 if (thread->is_VM_thread()) { 495 // Only the VM thread handles BREAK_SIGNAL ... 496 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 497 } else { 498 // ... all other threads block BREAK_SIGNAL 499 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 500 } 501 } 502 } 503 504 ////////////////////////////////////////////////////////////////////////////// 505 // detecting pthread library 506 507 void os::Linux::libpthread_init() { 508 // Save glibc and pthread version strings. 509 #if !defined(_CS_GNU_LIBC_VERSION) || \ 510 !defined(_CS_GNU_LIBPTHREAD_VERSION) 511 #error "glibc too old (< 2.3.2)" 512 #endif 513 514 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 515 assert(n > 0, "cannot retrieve glibc version"); 516 char *str = (char *)malloc(n, mtInternal); 517 confstr(_CS_GNU_LIBC_VERSION, str, n); 518 os::Linux::set_glibc_version(str); 519 520 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 521 assert(n > 0, "cannot retrieve pthread version"); 522 str = (char *)malloc(n, mtInternal); 523 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 524 os::Linux::set_libpthread_version(str); 525 } 526 527 ///////////////////////////////////////////////////////////////////////////// 528 // thread stack expansion 529 530 // os::Linux::manually_expand_stack() takes care of expanding the thread 531 // stack. Note that this is normally not needed: pthread stacks allocate 532 // thread stack using mmap() without MAP_NORESERVE, so the stack is already 533 // committed. Therefore it is not necessary to expand the stack manually. 534 // 535 // Manually expanding the stack was historically needed on LinuxThreads 536 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays 537 // it is kept to deal with very rare corner cases: 538 // 539 // For one, user may run the VM on an own implementation of threads 540 // whose stacks are - like the old LinuxThreads - implemented using 541 // mmap(MAP_GROWSDOWN). 542 // 543 // Also, this coding may be needed if the VM is running on the primordial 544 // thread. Normally we avoid running on the primordial thread; however, 545 // user may still invoke the VM on the primordial thread. 546 // 547 // The following historical comment describes the details about running 548 // on a thread stack allocated with mmap(MAP_GROWSDOWN): 549 550 551 // Force Linux kernel to expand current thread stack. If "bottom" is close 552 // to the stack guard, caller should block all signals. 553 // 554 // MAP_GROWSDOWN: 555 // A special mmap() flag that is used to implement thread stacks. It tells 556 // kernel that the memory region should extend downwards when needed. This 557 // allows early versions of LinuxThreads to only mmap the first few pages 558 // when creating a new thread. Linux kernel will automatically expand thread 559 // stack as needed (on page faults). 560 // 561 // However, because the memory region of a MAP_GROWSDOWN stack can grow on 562 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 563 // region, it's hard to tell if the fault is due to a legitimate stack 564 // access or because of reading/writing non-exist memory (e.g. buffer 565 // overrun). As a rule, if the fault happens below current stack pointer, 566 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the 567 // application (see Linux kernel fault.c). 568 // 569 // This Linux feature can cause SIGSEGV when VM bangs thread stack for 570 // stack overflow detection. 571 // 572 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 573 // not use MAP_GROWSDOWN. 574 // 575 // To get around the problem and allow stack banging on Linux, we need to 576 // manually expand thread stack after receiving the SIGSEGV. 577 // 578 // There are two ways to expand thread stack to address "bottom", we used 579 // both of them in JVM before 1.5: 580 // 1. adjust stack pointer first so that it is below "bottom", and then 581 // touch "bottom" 582 // 2. mmap() the page in question 583 // 584 // Now alternate signal stack is gone, it's harder to use 2. For instance, 585 // if current sp is already near the lower end of page 101, and we need to 586 // call mmap() to map page 100, it is possible that part of the mmap() frame 587 // will be placed in page 100. When page 100 is mapped, it is zero-filled. 588 // That will destroy the mmap() frame and cause VM to crash. 589 // 590 // The following code works by adjusting sp first, then accessing the "bottom" 591 // page to force a page fault. Linux kernel will then automatically expand the 592 // stack mapping. 593 // 594 // _expand_stack_to() assumes its frame size is less than page size, which 595 // should always be true if the function is not inlined. 596 597 static void NOINLINE _expand_stack_to(address bottom) { 598 address sp; 599 size_t size; 600 volatile char *p; 601 602 // Adjust bottom to point to the largest address within the same page, it 603 // gives us a one-page buffer if alloca() allocates slightly more memory. 604 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 605 bottom += os::Linux::page_size() - 1; 606 607 // sp might be slightly above current stack pointer; if that's the case, we 608 // will alloca() a little more space than necessary, which is OK. Don't use 609 // os::current_stack_pointer(), as its result can be slightly below current 610 // stack pointer, causing us to not alloca enough to reach "bottom". 611 sp = (address)&sp; 612 613 if (sp > bottom) { 614 size = sp - bottom; 615 p = (volatile char *)alloca(size); 616 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 617 p[0] = '\0'; 618 } 619 } 620 621 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 622 assert(t!=NULL, "just checking"); 623 assert(t->osthread()->expanding_stack(), "expand should be set"); 624 assert(t->stack_base() != NULL, "stack_base was not initialized"); 625 626 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) { 627 sigset_t mask_all, old_sigset; 628 sigfillset(&mask_all); 629 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 630 _expand_stack_to(addr); 631 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 632 return true; 633 } 634 return false; 635 } 636 637 ////////////////////////////////////////////////////////////////////////////// 638 // create new thread 639 640 // Thread start routine for all newly created threads 641 static void *thread_native_entry(Thread *thread) { 642 // Try to randomize the cache line index of hot stack frames. 643 // This helps when threads of the same stack traces evict each other's 644 // cache lines. The threads can be either from the same JVM instance, or 645 // from different JVM instances. The benefit is especially true for 646 // processors with hyperthreading technology. 647 static int counter = 0; 648 int pid = os::current_process_id(); 649 alloca(((pid ^ counter++) & 7) * 128); 650 651 thread->initialize_thread_current(); 652 653 OSThread* osthread = thread->osthread(); 654 Monitor* sync = osthread->startThread_lock(); 655 656 osthread->set_thread_id(os::current_thread_id()); 657 658 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 659 os::current_thread_id(), (uintx) pthread_self()); 660 661 if (UseNUMA) { 662 int lgrp_id = os::numa_get_group_id(); 663 if (lgrp_id != -1) { 664 thread->set_lgrp_id(lgrp_id); 665 } 666 } 667 // initialize signal mask for this thread 668 os::Linux::hotspot_sigmask(thread); 669 670 // initialize floating point control register 671 os::Linux::init_thread_fpu_state(); 672 673 // handshaking with parent thread 674 { 675 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 676 677 // notify parent thread 678 osthread->set_state(INITIALIZED); 679 sync->notify_all(); 680 681 // wait until os::start_thread() 682 while (osthread->get_state() == INITIALIZED) { 683 sync->wait(Mutex::_no_safepoint_check_flag); 684 } 685 } 686 687 // call one more level start routine 688 thread->run(); 689 690 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 691 os::current_thread_id(), (uintx) pthread_self()); 692 693 // If a thread has not deleted itself ("delete this") as part of its 694 // termination sequence, we have to ensure thread-local-storage is 695 // cleared before we actually terminate. No threads should ever be 696 // deleted asynchronously with respect to their termination. 697 if (Thread::current_or_null_safe() != NULL) { 698 assert(Thread::current_or_null_safe() == thread, "current thread is wrong"); 699 thread->clear_thread_current(); 700 } 701 702 return 0; 703 } 704 705 bool os::create_thread(Thread* thread, ThreadType thr_type, 706 size_t stack_size) { 707 assert(thread->osthread() == NULL, "caller responsible"); 708 709 // Allocate the OSThread object 710 OSThread* osthread = new OSThread(NULL, NULL); 711 if (osthread == NULL) { 712 return false; 713 } 714 715 // set the correct thread state 716 osthread->set_thread_type(thr_type); 717 718 // Initial state is ALLOCATED but not INITIALIZED 719 osthread->set_state(ALLOCATED); 720 721 thread->set_osthread(osthread); 722 723 // init thread attributes 724 pthread_attr_t attr; 725 pthread_attr_init(&attr); 726 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 727 728 // stack size 729 // calculate stack size if it's not specified by caller 730 if (stack_size == 0) { 731 stack_size = os::Linux::default_stack_size(thr_type); 732 733 switch (thr_type) { 734 case os::java_thread: 735 // Java threads use ThreadStackSize which default value can be 736 // changed with the flag -Xss 737 assert(JavaThread::stack_size_at_create() > 0, "this should be set"); 738 stack_size = JavaThread::stack_size_at_create(); 739 break; 740 case os::compiler_thread: 741 if (CompilerThreadStackSize > 0) { 742 stack_size = (size_t)(CompilerThreadStackSize * K); 743 break; 744 } // else fall through: 745 // use VMThreadStackSize if CompilerThreadStackSize is not defined 746 case os::vm_thread: 747 case os::pgc_thread: 748 case os::cgc_thread: 749 case os::watcher_thread: 750 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 751 break; 752 } 753 } 754 755 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 756 pthread_attr_setstacksize(&attr, stack_size); 757 758 // glibc guard page 759 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 760 761 ThreadState state; 762 763 { 764 pthread_t tid; 765 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread); 766 767 char buf[64]; 768 if (ret == 0) { 769 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ", 770 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 771 } else { 772 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.", 773 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 774 } 775 776 pthread_attr_destroy(&attr); 777 778 if (ret != 0) { 779 // Need to clean up stuff we've allocated so far 780 thread->set_osthread(NULL); 781 delete osthread; 782 return false; 783 } 784 785 // Store pthread info into the OSThread 786 osthread->set_pthread_id(tid); 787 788 // Wait until child thread is either initialized or aborted 789 { 790 Monitor* sync_with_child = osthread->startThread_lock(); 791 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 792 while ((state = osthread->get_state()) == ALLOCATED) { 793 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 794 } 795 } 796 } 797 798 // Aborted due to thread limit being reached 799 if (state == ZOMBIE) { 800 thread->set_osthread(NULL); 801 delete osthread; 802 return false; 803 } 804 805 // The thread is returned suspended (in state INITIALIZED), 806 // and is started higher up in the call chain 807 assert(state == INITIALIZED, "race condition"); 808 return true; 809 } 810 811 ///////////////////////////////////////////////////////////////////////////// 812 // attach existing thread 813 814 // bootstrap the main thread 815 bool os::create_main_thread(JavaThread* thread) { 816 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 817 return create_attached_thread(thread); 818 } 819 820 bool os::create_attached_thread(JavaThread* thread) { 821 #ifdef ASSERT 822 thread->verify_not_published(); 823 #endif 824 825 // Allocate the OSThread object 826 OSThread* osthread = new OSThread(NULL, NULL); 827 828 if (osthread == NULL) { 829 return false; 830 } 831 832 // Store pthread info into the OSThread 833 osthread->set_thread_id(os::Linux::gettid()); 834 osthread->set_pthread_id(::pthread_self()); 835 836 // initialize floating point control register 837 os::Linux::init_thread_fpu_state(); 838 839 // Initial thread state is RUNNABLE 840 osthread->set_state(RUNNABLE); 841 842 thread->set_osthread(osthread); 843 844 if (UseNUMA) { 845 int lgrp_id = os::numa_get_group_id(); 846 if (lgrp_id != -1) { 847 thread->set_lgrp_id(lgrp_id); 848 } 849 } 850 851 if (os::Linux::is_initial_thread()) { 852 // If current thread is initial thread, its stack is mapped on demand, 853 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 854 // the entire stack region to avoid SEGV in stack banging. 855 // It is also useful to get around the heap-stack-gap problem on SuSE 856 // kernel (see 4821821 for details). We first expand stack to the top 857 // of yellow zone, then enable stack yellow zone (order is significant, 858 // enabling yellow zone first will crash JVM on SuSE Linux), so there 859 // is no gap between the last two virtual memory regions. 860 861 JavaThread *jt = (JavaThread *)thread; 862 address addr = jt->stack_reserved_zone_base(); 863 assert(addr != NULL, "initialization problem?"); 864 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 865 866 osthread->set_expanding_stack(); 867 os::Linux::manually_expand_stack(jt, addr); 868 osthread->clear_expanding_stack(); 869 } 870 871 // initialize signal mask for this thread 872 // and save the caller's signal mask 873 os::Linux::hotspot_sigmask(thread); 874 875 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 876 os::current_thread_id(), (uintx) pthread_self()); 877 878 return true; 879 } 880 881 void os::pd_start_thread(Thread* thread) { 882 OSThread * osthread = thread->osthread(); 883 assert(osthread->get_state() != INITIALIZED, "just checking"); 884 Monitor* sync_with_child = osthread->startThread_lock(); 885 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 886 sync_with_child->notify(); 887 } 888 889 // Free Linux resources related to the OSThread 890 void os::free_thread(OSThread* osthread) { 891 assert(osthread != NULL, "osthread not set"); 892 893 // We are told to free resources of the argument thread, 894 // but we can only really operate on the current thread. 895 assert(Thread::current()->osthread() == osthread, 896 "os::free_thread but not current thread"); 897 898 #ifdef ASSERT 899 sigset_t current; 900 sigemptyset(¤t); 901 pthread_sigmask(SIG_SETMASK, NULL, ¤t); 902 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!"); 903 #endif 904 905 // Restore caller's signal mask 906 sigset_t sigmask = osthread->caller_sigmask(); 907 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 908 909 delete osthread; 910 } 911 912 ////////////////////////////////////////////////////////////////////////////// 913 // initial thread 914 915 // Check if current thread is the initial thread, similar to Solaris thr_main. 916 bool os::Linux::is_initial_thread(void) { 917 char dummy; 918 // If called before init complete, thread stack bottom will be null. 919 // Can be called if fatal error occurs before initialization. 920 if (initial_thread_stack_bottom() == NULL) return false; 921 assert(initial_thread_stack_bottom() != NULL && 922 initial_thread_stack_size() != 0, 923 "os::init did not locate initial thread's stack region"); 924 if ((address)&dummy >= initial_thread_stack_bottom() && 925 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) { 926 return true; 927 } else { 928 return false; 929 } 930 } 931 932 // Find the virtual memory area that contains addr 933 static bool find_vma(address addr, address* vma_low, address* vma_high) { 934 FILE *fp = fopen("/proc/self/maps", "r"); 935 if (fp) { 936 address low, high; 937 while (!feof(fp)) { 938 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 939 if (low <= addr && addr < high) { 940 if (vma_low) *vma_low = low; 941 if (vma_high) *vma_high = high; 942 fclose(fp); 943 return true; 944 } 945 } 946 for (;;) { 947 int ch = fgetc(fp); 948 if (ch == EOF || ch == (int)'\n') break; 949 } 950 } 951 fclose(fp); 952 } 953 return false; 954 } 955 956 // Locate initial thread stack. This special handling of initial thread stack 957 // is needed because pthread_getattr_np() on most (all?) Linux distros returns 958 // bogus value for initial thread. 959 void os::Linux::capture_initial_stack(size_t max_size) { 960 // stack size is the easy part, get it from RLIMIT_STACK 961 size_t stack_size; 962 struct rlimit rlim; 963 getrlimit(RLIMIT_STACK, &rlim); 964 stack_size = rlim.rlim_cur; 965 966 // 6308388: a bug in ld.so will relocate its own .data section to the 967 // lower end of primordial stack; reduce ulimit -s value a little bit 968 // so we won't install guard page on ld.so's data section. 969 stack_size -= 2 * page_size(); 970 971 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 972 // 7.1, in both cases we will get 2G in return value. 973 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 974 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 975 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 976 // in case other parts in glibc still assumes 2M max stack size. 977 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 978 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 979 if (stack_size > 2 * K * K IA64_ONLY(*2)) { 980 stack_size = 2 * K * K IA64_ONLY(*2); 981 } 982 // Try to figure out where the stack base (top) is. This is harder. 983 // 984 // When an application is started, glibc saves the initial stack pointer in 985 // a global variable "__libc_stack_end", which is then used by system 986 // libraries. __libc_stack_end should be pretty close to stack top. The 987 // variable is available since the very early days. However, because it is 988 // a private interface, it could disappear in the future. 989 // 990 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 991 // to __libc_stack_end, it is very close to stack top, but isn't the real 992 // stack top. Note that /proc may not exist if VM is running as a chroot 993 // program, so reading /proc/<pid>/stat could fail. Also the contents of 994 // /proc/<pid>/stat could change in the future (though unlikely). 995 // 996 // We try __libc_stack_end first. If that doesn't work, look for 997 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 998 // as a hint, which should work well in most cases. 999 1000 uintptr_t stack_start; 1001 1002 // try __libc_stack_end first 1003 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1004 if (p && *p) { 1005 stack_start = *p; 1006 } else { 1007 // see if we can get the start_stack field from /proc/self/stat 1008 FILE *fp; 1009 int pid; 1010 char state; 1011 int ppid; 1012 int pgrp; 1013 int session; 1014 int nr; 1015 int tpgrp; 1016 unsigned long flags; 1017 unsigned long minflt; 1018 unsigned long cminflt; 1019 unsigned long majflt; 1020 unsigned long cmajflt; 1021 unsigned long utime; 1022 unsigned long stime; 1023 long cutime; 1024 long cstime; 1025 long prio; 1026 long nice; 1027 long junk; 1028 long it_real; 1029 uintptr_t start; 1030 uintptr_t vsize; 1031 intptr_t rss; 1032 uintptr_t rsslim; 1033 uintptr_t scodes; 1034 uintptr_t ecode; 1035 int i; 1036 1037 // Figure what the primordial thread stack base is. Code is inspired 1038 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1039 // followed by command name surrounded by parentheses, state, etc. 1040 char stat[2048]; 1041 int statlen; 1042 1043 fp = fopen("/proc/self/stat", "r"); 1044 if (fp) { 1045 statlen = fread(stat, 1, 2047, fp); 1046 stat[statlen] = '\0'; 1047 fclose(fp); 1048 1049 // Skip pid and the command string. Note that we could be dealing with 1050 // weird command names, e.g. user could decide to rename java launcher 1051 // to "java 1.4.2 :)", then the stat file would look like 1052 // 1234 (java 1.4.2 :)) R ... ... 1053 // We don't really need to know the command string, just find the last 1054 // occurrence of ")" and then start parsing from there. See bug 4726580. 1055 char * s = strrchr(stat, ')'); 1056 1057 i = 0; 1058 if (s) { 1059 // Skip blank chars 1060 do { s++; } while (s && isspace(*s)); 1061 1062 #define _UFM UINTX_FORMAT 1063 #define _DFM INTX_FORMAT 1064 1065 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1066 // 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 1067 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, 1068 &state, // 3 %c 1069 &ppid, // 4 %d 1070 &pgrp, // 5 %d 1071 &session, // 6 %d 1072 &nr, // 7 %d 1073 &tpgrp, // 8 %d 1074 &flags, // 9 %lu 1075 &minflt, // 10 %lu 1076 &cminflt, // 11 %lu 1077 &majflt, // 12 %lu 1078 &cmajflt, // 13 %lu 1079 &utime, // 14 %lu 1080 &stime, // 15 %lu 1081 &cutime, // 16 %ld 1082 &cstime, // 17 %ld 1083 &prio, // 18 %ld 1084 &nice, // 19 %ld 1085 &junk, // 20 %ld 1086 &it_real, // 21 %ld 1087 &start, // 22 UINTX_FORMAT 1088 &vsize, // 23 UINTX_FORMAT 1089 &rss, // 24 INTX_FORMAT 1090 &rsslim, // 25 UINTX_FORMAT 1091 &scodes, // 26 UINTX_FORMAT 1092 &ecode, // 27 UINTX_FORMAT 1093 &stack_start); // 28 UINTX_FORMAT 1094 } 1095 1096 #undef _UFM 1097 #undef _DFM 1098 1099 if (i != 28 - 2) { 1100 assert(false, "Bad conversion from /proc/self/stat"); 1101 // product mode - assume we are the initial thread, good luck in the 1102 // embedded case. 1103 warning("Can't detect initial thread stack location - bad conversion"); 1104 stack_start = (uintptr_t) &rlim; 1105 } 1106 } else { 1107 // For some reason we can't open /proc/self/stat (for example, running on 1108 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1109 // most cases, so don't abort: 1110 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1111 stack_start = (uintptr_t) &rlim; 1112 } 1113 } 1114 1115 // Now we have a pointer (stack_start) very close to the stack top, the 1116 // next thing to do is to figure out the exact location of stack top. We 1117 // can find out the virtual memory area that contains stack_start by 1118 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1119 // and its upper limit is the real stack top. (again, this would fail if 1120 // running inside chroot, because /proc may not exist.) 1121 1122 uintptr_t stack_top; 1123 address low, high; 1124 if (find_vma((address)stack_start, &low, &high)) { 1125 // success, "high" is the true stack top. (ignore "low", because initial 1126 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1127 stack_top = (uintptr_t)high; 1128 } else { 1129 // failed, likely because /proc/self/maps does not exist 1130 warning("Can't detect initial thread stack location - find_vma failed"); 1131 // best effort: stack_start is normally within a few pages below the real 1132 // stack top, use it as stack top, and reduce stack size so we won't put 1133 // guard page outside stack. 1134 stack_top = stack_start; 1135 stack_size -= 16 * page_size(); 1136 } 1137 1138 // stack_top could be partially down the page so align it 1139 stack_top = align_size_up(stack_top, page_size()); 1140 1141 if (max_size && stack_size > max_size) { 1142 _initial_thread_stack_size = max_size; 1143 } else { 1144 _initial_thread_stack_size = stack_size; 1145 } 1146 1147 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1148 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1149 } 1150 1151 //////////////////////////////////////////////////////////////////////////////// 1152 // time support 1153 1154 // Time since start-up in seconds to a fine granularity. 1155 // Used by VMSelfDestructTimer and the MemProfiler. 1156 double os::elapsedTime() { 1157 1158 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1159 } 1160 1161 jlong os::elapsed_counter() { 1162 return javaTimeNanos() - initial_time_count; 1163 } 1164 1165 jlong os::elapsed_frequency() { 1166 return NANOSECS_PER_SEC; // nanosecond resolution 1167 } 1168 1169 bool os::supports_vtime() { return true; } 1170 bool os::enable_vtime() { return false; } 1171 bool os::vtime_enabled() { return false; } 1172 1173 double os::elapsedVTime() { 1174 struct rusage usage; 1175 int retval = getrusage(RUSAGE_THREAD, &usage); 1176 if (retval == 0) { 1177 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); 1178 } else { 1179 // better than nothing, but not much 1180 return elapsedTime(); 1181 } 1182 } 1183 1184 jlong os::javaTimeMillis() { 1185 timeval time; 1186 int status = gettimeofday(&time, NULL); 1187 assert(status != -1, "linux error"); 1188 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1189 } 1190 1191 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) { 1192 timeval time; 1193 int status = gettimeofday(&time, NULL); 1194 assert(status != -1, "linux error"); 1195 seconds = jlong(time.tv_sec); 1196 nanos = jlong(time.tv_usec) * 1000; 1197 } 1198 1199 1200 #ifndef CLOCK_MONOTONIC 1201 #define CLOCK_MONOTONIC (1) 1202 #endif 1203 1204 void os::Linux::clock_init() { 1205 // we do dlopen's in this particular order due to bug in linux 1206 // dynamical loader (see 6348968) leading to crash on exit 1207 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1208 if (handle == NULL) { 1209 handle = dlopen("librt.so", RTLD_LAZY); 1210 } 1211 1212 if (handle) { 1213 int (*clock_getres_func)(clockid_t, struct timespec*) = 1214 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1215 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1216 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1217 if (clock_getres_func && clock_gettime_func) { 1218 // See if monotonic clock is supported by the kernel. Note that some 1219 // early implementations simply return kernel jiffies (updated every 1220 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1221 // for nano time (though the monotonic property is still nice to have). 1222 // It's fixed in newer kernels, however clock_getres() still returns 1223 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1224 // resolution for now. Hopefully as people move to new kernels, this 1225 // won't be a problem. 1226 struct timespec res; 1227 struct timespec tp; 1228 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1229 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1230 // yes, monotonic clock is supported 1231 _clock_gettime = clock_gettime_func; 1232 return; 1233 } else { 1234 // close librt if there is no monotonic clock 1235 dlclose(handle); 1236 } 1237 } 1238 } 1239 warning("No monotonic clock was available - timed services may " \ 1240 "be adversely affected if the time-of-day clock changes"); 1241 } 1242 1243 #ifndef SYS_clock_getres 1244 #if defined(IA32) || defined(AMD64) 1245 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1246 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1247 #else 1248 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1249 #define sys_clock_getres(x,y) -1 1250 #endif 1251 #else 1252 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1253 #endif 1254 1255 void os::Linux::fast_thread_clock_init() { 1256 if (!UseLinuxPosixThreadCPUClocks) { 1257 return; 1258 } 1259 clockid_t clockid; 1260 struct timespec tp; 1261 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1262 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1263 1264 // Switch to using fast clocks for thread cpu time if 1265 // the sys_clock_getres() returns 0 error code. 1266 // Note, that some kernels may support the current thread 1267 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1268 // returned by the pthread_getcpuclockid(). 1269 // If the fast Posix clocks are supported then the sys_clock_getres() 1270 // must return at least tp.tv_sec == 0 which means a resolution 1271 // better than 1 sec. This is extra check for reliability. 1272 1273 if (pthread_getcpuclockid_func && 1274 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1275 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1276 _supports_fast_thread_cpu_time = true; 1277 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1278 } 1279 } 1280 1281 jlong os::javaTimeNanos() { 1282 if (os::supports_monotonic_clock()) { 1283 struct timespec tp; 1284 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1285 assert(status == 0, "gettime error"); 1286 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1287 return result; 1288 } else { 1289 timeval time; 1290 int status = gettimeofday(&time, NULL); 1291 assert(status != -1, "linux error"); 1292 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1293 return 1000 * usecs; 1294 } 1295 } 1296 1297 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1298 if (os::supports_monotonic_clock()) { 1299 info_ptr->max_value = ALL_64_BITS; 1300 1301 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1302 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1303 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1304 } else { 1305 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1306 info_ptr->max_value = ALL_64_BITS; 1307 1308 // gettimeofday is a real time clock so it skips 1309 info_ptr->may_skip_backward = true; 1310 info_ptr->may_skip_forward = true; 1311 } 1312 1313 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1314 } 1315 1316 // Return the real, user, and system times in seconds from an 1317 // arbitrary fixed point in the past. 1318 bool os::getTimesSecs(double* process_real_time, 1319 double* process_user_time, 1320 double* process_system_time) { 1321 struct tms ticks; 1322 clock_t real_ticks = times(&ticks); 1323 1324 if (real_ticks == (clock_t) (-1)) { 1325 return false; 1326 } else { 1327 double ticks_per_second = (double) clock_tics_per_sec; 1328 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1329 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1330 *process_real_time = ((double) real_ticks) / ticks_per_second; 1331 1332 return true; 1333 } 1334 } 1335 1336 1337 char * os::local_time_string(char *buf, size_t buflen) { 1338 struct tm t; 1339 time_t long_time; 1340 time(&long_time); 1341 localtime_r(&long_time, &t); 1342 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1343 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1344 t.tm_hour, t.tm_min, t.tm_sec); 1345 return buf; 1346 } 1347 1348 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1349 return localtime_r(clock, res); 1350 } 1351 1352 //////////////////////////////////////////////////////////////////////////////// 1353 // runtime exit support 1354 1355 // Note: os::shutdown() might be called very early during initialization, or 1356 // called from signal handler. Before adding something to os::shutdown(), make 1357 // sure it is async-safe and can handle partially initialized VM. 1358 void os::shutdown() { 1359 1360 // allow PerfMemory to attempt cleanup of any persistent resources 1361 perfMemory_exit(); 1362 1363 // needs to remove object in file system 1364 AttachListener::abort(); 1365 1366 // flush buffered output, finish log files 1367 ostream_abort(); 1368 1369 // Check for abort hook 1370 abort_hook_t abort_hook = Arguments::abort_hook(); 1371 if (abort_hook != NULL) { 1372 abort_hook(); 1373 } 1374 1375 } 1376 1377 // Note: os::abort() might be called very early during initialization, or 1378 // called from signal handler. Before adding something to os::abort(), make 1379 // sure it is async-safe and can handle partially initialized VM. 1380 void os::abort(bool dump_core, void* siginfo, const void* context) { 1381 os::shutdown(); 1382 if (dump_core) { 1383 #ifndef PRODUCT 1384 fdStream out(defaultStream::output_fd()); 1385 out.print_raw("Current thread is "); 1386 char buf[16]; 1387 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1388 out.print_raw_cr(buf); 1389 out.print_raw_cr("Dumping core ..."); 1390 #endif 1391 ::abort(); // dump core 1392 } 1393 1394 ::exit(1); 1395 } 1396 1397 // Die immediately, no exit hook, no abort hook, no cleanup. 1398 void os::die() { 1399 ::abort(); 1400 } 1401 1402 1403 // This method is a copy of JDK's sysGetLastErrorString 1404 // from src/solaris/hpi/src/system_md.c 1405 1406 size_t os::lasterror(char *buf, size_t len) { 1407 if (errno == 0) return 0; 1408 1409 const char *s = os::strerror(errno); 1410 size_t n = ::strlen(s); 1411 if (n >= len) { 1412 n = len - 1; 1413 } 1414 ::strncpy(buf, s, n); 1415 buf[n] = '\0'; 1416 return n; 1417 } 1418 1419 // thread_id is kernel thread id (similar to Solaris LWP id) 1420 intx os::current_thread_id() { return os::Linux::gettid(); } 1421 int os::current_process_id() { 1422 return ::getpid(); 1423 } 1424 1425 // DLL functions 1426 1427 const char* os::dll_file_extension() { return ".so"; } 1428 1429 // This must be hard coded because it's the system's temporary 1430 // directory not the java application's temp directory, ala java.io.tmpdir. 1431 const char* os::get_temp_directory() { return "/tmp"; } 1432 1433 static bool file_exists(const char* filename) { 1434 struct stat statbuf; 1435 if (filename == NULL || strlen(filename) == 0) { 1436 return false; 1437 } 1438 return os::stat(filename, &statbuf) == 0; 1439 } 1440 1441 bool os::dll_build_name(char* buffer, size_t buflen, 1442 const char* pname, const char* fname) { 1443 bool retval = false; 1444 // Copied from libhpi 1445 const size_t pnamelen = pname ? strlen(pname) : 0; 1446 1447 // Return error on buffer overflow. 1448 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1449 return retval; 1450 } 1451 1452 if (pnamelen == 0) { 1453 snprintf(buffer, buflen, "lib%s.so", fname); 1454 retval = true; 1455 } else if (strchr(pname, *os::path_separator()) != NULL) { 1456 int n; 1457 char** pelements = split_path(pname, &n); 1458 if (pelements == NULL) { 1459 return false; 1460 } 1461 for (int i = 0; i < n; i++) { 1462 // Really shouldn't be NULL, but check can't hurt 1463 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1464 continue; // skip the empty path values 1465 } 1466 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1467 if (file_exists(buffer)) { 1468 retval = true; 1469 break; 1470 } 1471 } 1472 // release the storage 1473 for (int i = 0; i < n; i++) { 1474 if (pelements[i] != NULL) { 1475 FREE_C_HEAP_ARRAY(char, pelements[i]); 1476 } 1477 } 1478 if (pelements != NULL) { 1479 FREE_C_HEAP_ARRAY(char*, pelements); 1480 } 1481 } else { 1482 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1483 retval = true; 1484 } 1485 return retval; 1486 } 1487 1488 // check if addr is inside libjvm.so 1489 bool os::address_is_in_vm(address addr) { 1490 static address libjvm_base_addr; 1491 Dl_info dlinfo; 1492 1493 if (libjvm_base_addr == NULL) { 1494 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1495 libjvm_base_addr = (address)dlinfo.dli_fbase; 1496 } 1497 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1498 } 1499 1500 if (dladdr((void *)addr, &dlinfo) != 0) { 1501 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1502 } 1503 1504 return false; 1505 } 1506 1507 bool os::dll_address_to_function_name(address addr, char *buf, 1508 int buflen, int *offset, 1509 bool demangle) { 1510 // buf is not optional, but offset is optional 1511 assert(buf != NULL, "sanity check"); 1512 1513 Dl_info dlinfo; 1514 1515 if (dladdr((void*)addr, &dlinfo) != 0) { 1516 // see if we have a matching symbol 1517 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1518 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) { 1519 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1520 } 1521 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1522 return true; 1523 } 1524 // no matching symbol so try for just file info 1525 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1526 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1527 buf, buflen, offset, dlinfo.dli_fname, demangle)) { 1528 return true; 1529 } 1530 } 1531 } 1532 1533 buf[0] = '\0'; 1534 if (offset != NULL) *offset = -1; 1535 return false; 1536 } 1537 1538 struct _address_to_library_name { 1539 address addr; // input : memory address 1540 size_t buflen; // size of fname 1541 char* fname; // output: library name 1542 address base; // library base addr 1543 }; 1544 1545 static int address_to_library_name_callback(struct dl_phdr_info *info, 1546 size_t size, void *data) { 1547 int i; 1548 bool found = false; 1549 address libbase = NULL; 1550 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1551 1552 // iterate through all loadable segments 1553 for (i = 0; i < info->dlpi_phnum; i++) { 1554 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1555 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1556 // base address of a library is the lowest address of its loaded 1557 // segments. 1558 if (libbase == NULL || libbase > segbase) { 1559 libbase = segbase; 1560 } 1561 // see if 'addr' is within current segment 1562 if (segbase <= d->addr && 1563 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1564 found = true; 1565 } 1566 } 1567 } 1568 1569 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1570 // so dll_address_to_library_name() can fall through to use dladdr() which 1571 // can figure out executable name from argv[0]. 1572 if (found && info->dlpi_name && info->dlpi_name[0]) { 1573 d->base = libbase; 1574 if (d->fname) { 1575 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1576 } 1577 return 1; 1578 } 1579 return 0; 1580 } 1581 1582 bool os::dll_address_to_library_name(address addr, char* buf, 1583 int buflen, int* offset) { 1584 // buf is not optional, but offset is optional 1585 assert(buf != NULL, "sanity check"); 1586 1587 Dl_info dlinfo; 1588 struct _address_to_library_name data; 1589 1590 // There is a bug in old glibc dladdr() implementation that it could resolve 1591 // to wrong library name if the .so file has a base address != NULL. Here 1592 // we iterate through the program headers of all loaded libraries to find 1593 // out which library 'addr' really belongs to. This workaround can be 1594 // removed once the minimum requirement for glibc is moved to 2.3.x. 1595 data.addr = addr; 1596 data.fname = buf; 1597 data.buflen = buflen; 1598 data.base = NULL; 1599 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1600 1601 if (rslt) { 1602 // buf already contains library name 1603 if (offset) *offset = addr - data.base; 1604 return true; 1605 } 1606 if (dladdr((void*)addr, &dlinfo) != 0) { 1607 if (dlinfo.dli_fname != NULL) { 1608 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1609 } 1610 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1611 *offset = addr - (address)dlinfo.dli_fbase; 1612 } 1613 return true; 1614 } 1615 1616 buf[0] = '\0'; 1617 if (offset) *offset = -1; 1618 return false; 1619 } 1620 1621 // Loads .dll/.so and 1622 // in case of error it checks if .dll/.so was built for the 1623 // same architecture as Hotspot is running on 1624 1625 1626 // Remember the stack's state. The Linux dynamic linker will change 1627 // the stack to 'executable' at most once, so we must safepoint only once. 1628 bool os::Linux::_stack_is_executable = false; 1629 1630 // VM operation that loads a library. This is necessary if stack protection 1631 // of the Java stacks can be lost during loading the library. If we 1632 // do not stop the Java threads, they can stack overflow before the stacks 1633 // are protected again. 1634 class VM_LinuxDllLoad: public VM_Operation { 1635 private: 1636 const char *_filename; 1637 char *_ebuf; 1638 int _ebuflen; 1639 void *_lib; 1640 public: 1641 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1642 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1643 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1644 void doit() { 1645 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1646 os::Linux::_stack_is_executable = true; 1647 } 1648 void* loaded_library() { return _lib; } 1649 }; 1650 1651 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1652 void * result = NULL; 1653 bool load_attempted = false; 1654 1655 // Check whether the library to load might change execution rights 1656 // of the stack. If they are changed, the protection of the stack 1657 // guard pages will be lost. We need a safepoint to fix this. 1658 // 1659 // See Linux man page execstack(8) for more info. 1660 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1661 ElfFile ef(filename); 1662 if (!ef.specifies_noexecstack()) { 1663 if (!is_init_completed()) { 1664 os::Linux::_stack_is_executable = true; 1665 // This is OK - No Java threads have been created yet, and hence no 1666 // stack guard pages to fix. 1667 // 1668 // This should happen only when you are building JDK7 using a very 1669 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1670 // 1671 // Dynamic loader will make all stacks executable after 1672 // this function returns, and will not do that again. 1673 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1674 } else { 1675 warning("You have loaded library %s which might have disabled stack guard. " 1676 "The VM will try to fix the stack guard now.\n" 1677 "It's highly recommended that you fix the library with " 1678 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1679 filename); 1680 1681 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1682 JavaThread *jt = JavaThread::current(); 1683 if (jt->thread_state() != _thread_in_native) { 1684 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1685 // that requires ExecStack. Cannot enter safe point. Let's give up. 1686 warning("Unable to fix stack guard. Giving up."); 1687 } else { 1688 if (!LoadExecStackDllInVMThread) { 1689 // This is for the case where the DLL has an static 1690 // constructor function that executes JNI code. We cannot 1691 // load such DLLs in the VMThread. 1692 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1693 } 1694 1695 ThreadInVMfromNative tiv(jt); 1696 debug_only(VMNativeEntryWrapper vew;) 1697 1698 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1699 VMThread::execute(&op); 1700 if (LoadExecStackDllInVMThread) { 1701 result = op.loaded_library(); 1702 } 1703 load_attempted = true; 1704 } 1705 } 1706 } 1707 } 1708 1709 if (!load_attempted) { 1710 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1711 } 1712 1713 if (result != NULL) { 1714 // Successful loading 1715 return result; 1716 } 1717 1718 Elf32_Ehdr elf_head; 1719 int diag_msg_max_length=ebuflen-strlen(ebuf); 1720 char* diag_msg_buf=ebuf+strlen(ebuf); 1721 1722 if (diag_msg_max_length==0) { 1723 // No more space in ebuf for additional diagnostics message 1724 return NULL; 1725 } 1726 1727 1728 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1729 1730 if (file_descriptor < 0) { 1731 // Can't open library, report dlerror() message 1732 return NULL; 1733 } 1734 1735 bool failed_to_read_elf_head= 1736 (sizeof(elf_head)!= 1737 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1738 1739 ::close(file_descriptor); 1740 if (failed_to_read_elf_head) { 1741 // file i/o error - report dlerror() msg 1742 return NULL; 1743 } 1744 1745 typedef struct { 1746 Elf32_Half code; // Actual value as defined in elf.h 1747 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1748 char elf_class; // 32 or 64 bit 1749 char endianess; // MSB or LSB 1750 char* name; // String representation 1751 } arch_t; 1752 1753 #ifndef EM_486 1754 #define EM_486 6 /* Intel 80486 */ 1755 #endif 1756 #ifndef EM_AARCH64 1757 #define EM_AARCH64 183 /* ARM AARCH64 */ 1758 #endif 1759 1760 static const arch_t arch_array[]={ 1761 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1762 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1763 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1764 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1765 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1766 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1767 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1768 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1769 #if defined(VM_LITTLE_ENDIAN) 1770 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"}, 1771 #else 1772 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"}, 1773 #endif 1774 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1775 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1776 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1777 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1778 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1779 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1780 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1781 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1782 }; 1783 1784 #if (defined IA32) 1785 static Elf32_Half running_arch_code=EM_386; 1786 #elif (defined AMD64) 1787 static Elf32_Half running_arch_code=EM_X86_64; 1788 #elif (defined IA64) 1789 static Elf32_Half running_arch_code=EM_IA_64; 1790 #elif (defined __sparc) && (defined _LP64) 1791 static Elf32_Half running_arch_code=EM_SPARCV9; 1792 #elif (defined __sparc) && (!defined _LP64) 1793 static Elf32_Half running_arch_code=EM_SPARC; 1794 #elif (defined __powerpc64__) 1795 static Elf32_Half running_arch_code=EM_PPC64; 1796 #elif (defined __powerpc__) 1797 static Elf32_Half running_arch_code=EM_PPC; 1798 #elif (defined ARM) 1799 static Elf32_Half running_arch_code=EM_ARM; 1800 #elif (defined S390) 1801 static Elf32_Half running_arch_code=EM_S390; 1802 #elif (defined ALPHA) 1803 static Elf32_Half running_arch_code=EM_ALPHA; 1804 #elif (defined MIPSEL) 1805 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1806 #elif (defined PARISC) 1807 static Elf32_Half running_arch_code=EM_PARISC; 1808 #elif (defined MIPS) 1809 static Elf32_Half running_arch_code=EM_MIPS; 1810 #elif (defined M68K) 1811 static Elf32_Half running_arch_code=EM_68K; 1812 #elif (defined AARCH64) 1813 static Elf32_Half running_arch_code=EM_AARCH64; 1814 #else 1815 #error Method os::dll_load requires that one of following is defined:\ 1816 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64 1817 #endif 1818 1819 // Identify compatability class for VM's architecture and library's architecture 1820 // Obtain string descriptions for architectures 1821 1822 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1823 int running_arch_index=-1; 1824 1825 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1826 if (running_arch_code == arch_array[i].code) { 1827 running_arch_index = i; 1828 } 1829 if (lib_arch.code == arch_array[i].code) { 1830 lib_arch.compat_class = arch_array[i].compat_class; 1831 lib_arch.name = arch_array[i].name; 1832 } 1833 } 1834 1835 assert(running_arch_index != -1, 1836 "Didn't find running architecture code (running_arch_code) in arch_array"); 1837 if (running_arch_index == -1) { 1838 // Even though running architecture detection failed 1839 // we may still continue with reporting dlerror() message 1840 return NULL; 1841 } 1842 1843 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1844 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1845 return NULL; 1846 } 1847 1848 #ifndef S390 1849 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1850 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1851 return NULL; 1852 } 1853 #endif // !S390 1854 1855 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1856 if (lib_arch.name!=NULL) { 1857 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1858 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1859 lib_arch.name, arch_array[running_arch_index].name); 1860 } else { 1861 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1862 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1863 lib_arch.code, 1864 arch_array[running_arch_index].name); 1865 } 1866 } 1867 1868 return NULL; 1869 } 1870 1871 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 1872 int ebuflen) { 1873 void * result = ::dlopen(filename, RTLD_LAZY); 1874 if (result == NULL) { 1875 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 1876 ebuf[ebuflen-1] = '\0'; 1877 } 1878 return result; 1879 } 1880 1881 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 1882 int ebuflen) { 1883 void * result = NULL; 1884 if (LoadExecStackDllInVMThread) { 1885 result = dlopen_helper(filename, ebuf, ebuflen); 1886 } 1887 1888 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 1889 // library that requires an executable stack, or which does not have this 1890 // stack attribute set, dlopen changes the stack attribute to executable. The 1891 // read protection of the guard pages gets lost. 1892 // 1893 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 1894 // may have been queued at the same time. 1895 1896 if (!_stack_is_executable) { 1897 JavaThread *jt = Threads::first(); 1898 1899 while (jt) { 1900 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 1901 jt->stack_guards_enabled()) { // No pending stack overflow exceptions 1902 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) { 1903 warning("Attempt to reguard stack yellow zone failed."); 1904 } 1905 } 1906 jt = jt->next(); 1907 } 1908 } 1909 1910 return result; 1911 } 1912 1913 void* os::dll_lookup(void* handle, const char* name) { 1914 void* res = dlsym(handle, name); 1915 return res; 1916 } 1917 1918 void* os::get_default_process_handle() { 1919 return (void*)::dlopen(NULL, RTLD_LAZY); 1920 } 1921 1922 static bool _print_ascii_file(const char* filename, outputStream* st) { 1923 int fd = ::open(filename, O_RDONLY); 1924 if (fd == -1) { 1925 return false; 1926 } 1927 1928 char buf[32]; 1929 int bytes; 1930 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 1931 st->print_raw(buf, bytes); 1932 } 1933 1934 ::close(fd); 1935 1936 return true; 1937 } 1938 1939 void os::print_dll_info(outputStream *st) { 1940 st->print_cr("Dynamic libraries:"); 1941 1942 char fname[32]; 1943 pid_t pid = os::Linux::gettid(); 1944 1945 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1946 1947 if (!_print_ascii_file(fname, st)) { 1948 st->print("Can not get library information for pid = %d\n", pid); 1949 } 1950 } 1951 1952 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 1953 FILE *procmapsFile = NULL; 1954 1955 // Open the procfs maps file for the current process 1956 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 1957 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 1958 char line[PATH_MAX + 100]; 1959 1960 // Read line by line from 'file' 1961 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 1962 u8 base, top, offset, inode; 1963 char permissions[5]; 1964 char device[6]; 1965 char name[PATH_MAX + 1]; 1966 1967 // Parse fields from line 1968 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s", 1969 &base, &top, permissions, &offset, device, &inode, name); 1970 1971 // Filter by device id '00:00' so that we only get file system mapped files. 1972 if (strcmp(device, "00:00") != 0) { 1973 1974 // Call callback with the fields of interest 1975 if(callback(name, (address)base, (address)top, param)) { 1976 // Oops abort, callback aborted 1977 fclose(procmapsFile); 1978 return 1; 1979 } 1980 } 1981 } 1982 fclose(procmapsFile); 1983 } 1984 return 0; 1985 } 1986 1987 void os::print_os_info_brief(outputStream* st) { 1988 os::Linux::print_distro_info(st); 1989 1990 os::Posix::print_uname_info(st); 1991 1992 os::Linux::print_libversion_info(st); 1993 1994 } 1995 1996 void os::print_os_info(outputStream* st) { 1997 st->print("OS:"); 1998 1999 os::Linux::print_distro_info(st); 2000 2001 os::Posix::print_uname_info(st); 2002 2003 // Print warning if unsafe chroot environment detected 2004 if (unsafe_chroot_detected) { 2005 st->print("WARNING!! "); 2006 st->print_cr("%s", unstable_chroot_error); 2007 } 2008 2009 os::Linux::print_libversion_info(st); 2010 2011 os::Posix::print_rlimit_info(st); 2012 2013 os::Posix::print_load_average(st); 2014 2015 os::Linux::print_full_memory_info(st); 2016 } 2017 2018 // Try to identify popular distros. 2019 // Most Linux distributions have a /etc/XXX-release file, which contains 2020 // the OS version string. Newer Linux distributions have a /etc/lsb-release 2021 // file that also contains the OS version string. Some have more than one 2022 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2023 // /etc/redhat-release.), so the order is important. 2024 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2025 // their own specific XXX-release file as well as a redhat-release file. 2026 // Because of this the XXX-release file needs to be searched for before the 2027 // redhat-release file. 2028 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the 2029 // search for redhat-release / SuSE-release needs to be before lsb-release. 2030 // Since the lsb-release file is the new standard it needs to be searched 2031 // before the older style release files. 2032 // Searching system-release (Red Hat) and os-release (other Linuxes) are a 2033 // next to last resort. The os-release file is a new standard that contains 2034 // distribution information and the system-release file seems to be an old 2035 // standard that has been replaced by the lsb-release and os-release files. 2036 // Searching for the debian_version file is the last resort. It contains 2037 // an informative string like "6.0.6" or "wheezy/sid". Because of this 2038 // "Debian " is printed before the contents of the debian_version file. 2039 2040 const char* distro_files[] = { 2041 "/etc/oracle-release", 2042 "/etc/mandriva-release", 2043 "/etc/mandrake-release", 2044 "/etc/sun-release", 2045 "/etc/redhat-release", 2046 "/etc/SuSE-release", 2047 "/etc/lsb-release", 2048 "/etc/turbolinux-release", 2049 "/etc/gentoo-release", 2050 "/etc/ltib-release", 2051 "/etc/angstrom-version", 2052 "/etc/system-release", 2053 "/etc/os-release", 2054 NULL }; 2055 2056 void os::Linux::print_distro_info(outputStream* st) { 2057 for (int i = 0;; i++) { 2058 const char* file = distro_files[i]; 2059 if (file == NULL) { 2060 break; // done 2061 } 2062 // If file prints, we found it. 2063 if (_print_ascii_file(file, st)) { 2064 return; 2065 } 2066 } 2067 2068 if (file_exists("/etc/debian_version")) { 2069 st->print("Debian "); 2070 _print_ascii_file("/etc/debian_version", st); 2071 } else { 2072 st->print("Linux"); 2073 } 2074 st->cr(); 2075 } 2076 2077 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) { 2078 char buf[256]; 2079 while (fgets(buf, sizeof(buf), fp)) { 2080 // Edit out extra stuff in expected format 2081 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) { 2082 char* ptr = strstr(buf, "\""); // the name is in quotes 2083 if (ptr != NULL) { 2084 ptr++; // go beyond first quote 2085 char* nl = strchr(ptr, '\"'); 2086 if (nl != NULL) *nl = '\0'; 2087 strncpy(distro, ptr, length); 2088 } else { 2089 ptr = strstr(buf, "="); 2090 ptr++; // go beyond equals then 2091 char* nl = strchr(ptr, '\n'); 2092 if (nl != NULL) *nl = '\0'; 2093 strncpy(distro, ptr, length); 2094 } 2095 return; 2096 } else if (get_first_line) { 2097 char* nl = strchr(buf, '\n'); 2098 if (nl != NULL) *nl = '\0'; 2099 strncpy(distro, buf, length); 2100 return; 2101 } 2102 } 2103 // print last line and close 2104 char* nl = strchr(buf, '\n'); 2105 if (nl != NULL) *nl = '\0'; 2106 strncpy(distro, buf, length); 2107 } 2108 2109 static void parse_os_info(char* distro, size_t length, const char* file) { 2110 FILE* fp = fopen(file, "r"); 2111 if (fp != NULL) { 2112 // if suse format, print out first line 2113 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0); 2114 parse_os_info_helper(fp, distro, length, get_first_line); 2115 fclose(fp); 2116 } 2117 } 2118 2119 void os::get_summary_os_info(char* buf, size_t buflen) { 2120 for (int i = 0;; i++) { 2121 const char* file = distro_files[i]; 2122 if (file == NULL) { 2123 break; // ran out of distro_files 2124 } 2125 if (file_exists(file)) { 2126 parse_os_info(buf, buflen, file); 2127 return; 2128 } 2129 } 2130 // special case for debian 2131 if (file_exists("/etc/debian_version")) { 2132 strncpy(buf, "Debian ", buflen); 2133 parse_os_info(&buf[7], buflen-7, "/etc/debian_version"); 2134 } else { 2135 strncpy(buf, "Linux", buflen); 2136 } 2137 } 2138 2139 void os::Linux::print_libversion_info(outputStream* st) { 2140 // libc, pthread 2141 st->print("libc:"); 2142 st->print("%s ", os::Linux::glibc_version()); 2143 st->print("%s ", os::Linux::libpthread_version()); 2144 st->cr(); 2145 } 2146 2147 void os::Linux::print_full_memory_info(outputStream* st) { 2148 st->print("\n/proc/meminfo:\n"); 2149 _print_ascii_file("/proc/meminfo", st); 2150 st->cr(); 2151 } 2152 2153 void os::print_memory_info(outputStream* st) { 2154 2155 st->print("Memory:"); 2156 st->print(" %dk page", os::vm_page_size()>>10); 2157 2158 // values in struct sysinfo are "unsigned long" 2159 struct sysinfo si; 2160 sysinfo(&si); 2161 2162 st->print(", physical " UINT64_FORMAT "k", 2163 os::physical_memory() >> 10); 2164 st->print("(" UINT64_FORMAT "k free)", 2165 os::available_memory() >> 10); 2166 st->print(", swap " UINT64_FORMAT "k", 2167 ((jlong)si.totalswap * si.mem_unit) >> 10); 2168 st->print("(" UINT64_FORMAT "k free)", 2169 ((jlong)si.freeswap * si.mem_unit) >> 10); 2170 st->cr(); 2171 } 2172 2173 // Print the first "model name" line and the first "flags" line 2174 // that we find and nothing more. We assume "model name" comes 2175 // before "flags" so if we find a second "model name", then the 2176 // "flags" field is considered missing. 2177 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) { 2178 #if defined(IA32) || defined(AMD64) 2179 // Other platforms have less repetitive cpuinfo files 2180 FILE *fp = fopen("/proc/cpuinfo", "r"); 2181 if (fp) { 2182 while (!feof(fp)) { 2183 if (fgets(buf, buflen, fp)) { 2184 // Assume model name comes before flags 2185 bool model_name_printed = false; 2186 if (strstr(buf, "model name") != NULL) { 2187 if (!model_name_printed) { 2188 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n"); 2189 st->print_raw(buf); 2190 model_name_printed = true; 2191 } else { 2192 // model name printed but not flags? Odd, just return 2193 fclose(fp); 2194 return true; 2195 } 2196 } 2197 // print the flags line too 2198 if (strstr(buf, "flags") != NULL) { 2199 st->print_raw(buf); 2200 fclose(fp); 2201 return true; 2202 } 2203 } 2204 } 2205 fclose(fp); 2206 } 2207 #endif // x86 platforms 2208 return false; 2209 } 2210 2211 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) { 2212 // Only print the model name if the platform provides this as a summary 2213 if (!print_model_name_and_flags(st, buf, buflen)) { 2214 st->print("\n/proc/cpuinfo:\n"); 2215 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2216 st->print_cr(" <Not Available>"); 2217 } 2218 } 2219 } 2220 2221 #if defined(AMD64) || defined(IA32) || defined(X32) 2222 const char* search_string = "model name"; 2223 #elif defined(SPARC) 2224 const char* search_string = "cpu"; 2225 #elif defined(PPC64) 2226 const char* search_string = "cpu"; 2227 #else 2228 const char* search_string = "Processor"; 2229 #endif 2230 2231 // Parses the cpuinfo file for string representing the model name. 2232 void os::get_summary_cpu_info(char* cpuinfo, size_t length) { 2233 FILE* fp = fopen("/proc/cpuinfo", "r"); 2234 if (fp != NULL) { 2235 while (!feof(fp)) { 2236 char buf[256]; 2237 if (fgets(buf, sizeof(buf), fp)) { 2238 char* start = strstr(buf, search_string); 2239 if (start != NULL) { 2240 char *ptr = start + strlen(search_string); 2241 char *end = buf + strlen(buf); 2242 while (ptr != end) { 2243 // skip whitespace and colon for the rest of the name. 2244 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') { 2245 break; 2246 } 2247 ptr++; 2248 } 2249 if (ptr != end) { 2250 // reasonable string, get rid of newline and keep the rest 2251 char* nl = strchr(buf, '\n'); 2252 if (nl != NULL) *nl = '\0'; 2253 strncpy(cpuinfo, ptr, length); 2254 fclose(fp); 2255 return; 2256 } 2257 } 2258 } 2259 } 2260 fclose(fp); 2261 } 2262 // cpuinfo not found or parsing failed, just print generic string. The entire 2263 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86) 2264 #if defined(AMD64) 2265 strncpy(cpuinfo, "x86_64", length); 2266 #elif defined(IA32) 2267 strncpy(cpuinfo, "x86_32", length); 2268 #elif defined(IA64) 2269 strncpy(cpuinfo, "IA64", length); 2270 #elif defined(SPARC) 2271 strncpy(cpuinfo, "sparcv9", length); 2272 #elif defined(AARCH64) 2273 strncpy(cpuinfo, "AArch64", length); 2274 #elif defined(ARM) 2275 strncpy(cpuinfo, "ARM", length); 2276 #elif defined(PPC) 2277 strncpy(cpuinfo, "PPC64", length); 2278 #elif defined(ZERO_LIBARCH) 2279 strncpy(cpuinfo, ZERO_LIBARCH, length); 2280 #else 2281 strncpy(cpuinfo, "unknown", length); 2282 #endif 2283 } 2284 2285 static void print_signal_handler(outputStream* st, int sig, 2286 char* buf, size_t buflen); 2287 2288 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2289 st->print_cr("Signal Handlers:"); 2290 print_signal_handler(st, SIGSEGV, buf, buflen); 2291 print_signal_handler(st, SIGBUS , buf, buflen); 2292 print_signal_handler(st, SIGFPE , buf, buflen); 2293 print_signal_handler(st, SIGPIPE, buf, buflen); 2294 print_signal_handler(st, SIGXFSZ, buf, buflen); 2295 print_signal_handler(st, SIGILL , buf, buflen); 2296 print_signal_handler(st, SR_signum, buf, buflen); 2297 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2298 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2299 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2300 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2301 #if defined(PPC64) 2302 print_signal_handler(st, SIGTRAP, buf, buflen); 2303 #endif 2304 } 2305 2306 static char saved_jvm_path[MAXPATHLEN] = {0}; 2307 2308 // Find the full path to the current module, libjvm.so 2309 void os::jvm_path(char *buf, jint buflen) { 2310 // Error checking. 2311 if (buflen < MAXPATHLEN) { 2312 assert(false, "must use a large-enough buffer"); 2313 buf[0] = '\0'; 2314 return; 2315 } 2316 // Lazy resolve the path to current module. 2317 if (saved_jvm_path[0] != 0) { 2318 strcpy(buf, saved_jvm_path); 2319 return; 2320 } 2321 2322 char dli_fname[MAXPATHLEN]; 2323 bool ret = dll_address_to_library_name( 2324 CAST_FROM_FN_PTR(address, os::jvm_path), 2325 dli_fname, sizeof(dli_fname), NULL); 2326 assert(ret, "cannot locate libjvm"); 2327 char *rp = NULL; 2328 if (ret && dli_fname[0] != '\0') { 2329 rp = realpath(dli_fname, buf); 2330 } 2331 if (rp == NULL) { 2332 return; 2333 } 2334 2335 if (Arguments::sun_java_launcher_is_altjvm()) { 2336 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2337 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2338 // If "/jre/lib/" appears at the right place in the string, then 2339 // assume we are installed in a JDK and we're done. Otherwise, check 2340 // for a JAVA_HOME environment variable and fix up the path so it 2341 // looks like libjvm.so is installed there (append a fake suffix 2342 // hotspot/libjvm.so). 2343 const char *p = buf + strlen(buf) - 1; 2344 for (int count = 0; p > buf && count < 5; ++count) { 2345 for (--p; p > buf && *p != '/'; --p) 2346 /* empty */ ; 2347 } 2348 2349 if (strncmp(p, "/jre/lib/", 9) != 0) { 2350 // Look for JAVA_HOME in the environment. 2351 char* java_home_var = ::getenv("JAVA_HOME"); 2352 if (java_home_var != NULL && java_home_var[0] != 0) { 2353 char* jrelib_p; 2354 int len; 2355 2356 // Check the current module name "libjvm.so". 2357 p = strrchr(buf, '/'); 2358 if (p == NULL) { 2359 return; 2360 } 2361 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2362 2363 rp = realpath(java_home_var, buf); 2364 if (rp == NULL) { 2365 return; 2366 } 2367 2368 // determine if this is a legacy image or modules image 2369 // modules image doesn't have "jre" subdirectory 2370 len = strlen(buf); 2371 assert(len < buflen, "Ran out of buffer room"); 2372 jrelib_p = buf + len; 2373 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2374 if (0 != access(buf, F_OK)) { 2375 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2376 } 2377 2378 if (0 == access(buf, F_OK)) { 2379 // Use current module name "libjvm.so" 2380 len = strlen(buf); 2381 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2382 } else { 2383 // Go back to path of .so 2384 rp = realpath(dli_fname, buf); 2385 if (rp == NULL) { 2386 return; 2387 } 2388 } 2389 } 2390 } 2391 } 2392 2393 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2394 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2395 } 2396 2397 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2398 // no prefix required, not even "_" 2399 } 2400 2401 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2402 // no suffix required 2403 } 2404 2405 //////////////////////////////////////////////////////////////////////////////// 2406 // sun.misc.Signal support 2407 2408 static volatile jint sigint_count = 0; 2409 2410 static void UserHandler(int sig, void *siginfo, void *context) { 2411 // 4511530 - sem_post is serialized and handled by the manager thread. When 2412 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2413 // don't want to flood the manager thread with sem_post requests. 2414 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2415 return; 2416 } 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 2427 void* os::user_handler() { 2428 return CAST_FROM_FN_PTR(void*, UserHandler); 2429 } 2430 2431 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) { 2432 struct timespec ts; 2433 // Semaphore's are always associated with CLOCK_REALTIME 2434 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2435 // see unpackTime for discussion on overflow checking 2436 if (sec >= MAX_SECS) { 2437 ts.tv_sec += MAX_SECS; 2438 ts.tv_nsec = 0; 2439 } else { 2440 ts.tv_sec += sec; 2441 ts.tv_nsec += nsec; 2442 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2443 ts.tv_nsec -= NANOSECS_PER_SEC; 2444 ++ts.tv_sec; // note: this must be <= max_secs 2445 } 2446 } 2447 2448 return ts; 2449 } 2450 2451 extern "C" { 2452 typedef void (*sa_handler_t)(int); 2453 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2454 } 2455 2456 void* os::signal(int signal_number, void* handler) { 2457 struct sigaction sigAct, oldSigAct; 2458 2459 sigfillset(&(sigAct.sa_mask)); 2460 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2461 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2462 2463 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2464 // -1 means registration failed 2465 return (void *)-1; 2466 } 2467 2468 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2469 } 2470 2471 void os::signal_raise(int signal_number) { 2472 ::raise(signal_number); 2473 } 2474 2475 // The following code is moved from os.cpp for making this 2476 // code platform specific, which it is by its very nature. 2477 2478 // Will be modified when max signal is changed to be dynamic 2479 int os::sigexitnum_pd() { 2480 return NSIG; 2481 } 2482 2483 // a counter for each possible signal value 2484 static volatile jint pending_signals[NSIG+1] = { 0 }; 2485 2486 // Linux(POSIX) specific hand shaking semaphore. 2487 static sem_t sig_sem; 2488 static PosixSemaphore sr_semaphore; 2489 2490 void os::signal_init_pd() { 2491 // Initialize signal structures 2492 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2493 2494 // Initialize signal semaphore 2495 ::sem_init(&sig_sem, 0, 0); 2496 } 2497 2498 void os::signal_notify(int sig) { 2499 Atomic::inc(&pending_signals[sig]); 2500 ::sem_post(&sig_sem); 2501 } 2502 2503 static int check_pending_signals(bool wait) { 2504 Atomic::store(0, &sigint_count); 2505 for (;;) { 2506 for (int i = 0; i < NSIG + 1; i++) { 2507 jint n = pending_signals[i]; 2508 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2509 return i; 2510 } 2511 } 2512 if (!wait) { 2513 return -1; 2514 } 2515 JavaThread *thread = JavaThread::current(); 2516 ThreadBlockInVM tbivm(thread); 2517 2518 bool threadIsSuspended; 2519 do { 2520 thread->set_suspend_equivalent(); 2521 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2522 ::sem_wait(&sig_sem); 2523 2524 // were we externally suspended while we were waiting? 2525 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2526 if (threadIsSuspended) { 2527 // The semaphore has been incremented, but while we were waiting 2528 // another thread suspended us. We don't want to continue running 2529 // while suspended because that would surprise the thread that 2530 // suspended us. 2531 ::sem_post(&sig_sem); 2532 2533 thread->java_suspend_self(); 2534 } 2535 } while (threadIsSuspended); 2536 } 2537 } 2538 2539 int os::signal_lookup() { 2540 return check_pending_signals(false); 2541 } 2542 2543 int os::signal_wait() { 2544 return check_pending_signals(true); 2545 } 2546 2547 //////////////////////////////////////////////////////////////////////////////// 2548 // Virtual Memory 2549 2550 int os::vm_page_size() { 2551 // Seems redundant as all get out 2552 assert(os::Linux::page_size() != -1, "must call os::init"); 2553 return os::Linux::page_size(); 2554 } 2555 2556 // Solaris allocates memory by pages. 2557 int os::vm_allocation_granularity() { 2558 assert(os::Linux::page_size() != -1, "must call os::init"); 2559 return os::Linux::page_size(); 2560 } 2561 2562 // Rationale behind this function: 2563 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2564 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2565 // samples for JITted code. Here we create private executable mapping over the code cache 2566 // and then we can use standard (well, almost, as mapping can change) way to provide 2567 // info for the reporting script by storing timestamp and location of symbol 2568 void linux_wrap_code(char* base, size_t size) { 2569 static volatile jint cnt = 0; 2570 2571 if (!UseOprofile) { 2572 return; 2573 } 2574 2575 char buf[PATH_MAX+1]; 2576 int num = Atomic::add(1, &cnt); 2577 2578 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2579 os::get_temp_directory(), os::current_process_id(), num); 2580 unlink(buf); 2581 2582 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2583 2584 if (fd != -1) { 2585 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2586 if (rv != (off_t)-1) { 2587 if (::write(fd, "", 1) == 1) { 2588 mmap(base, size, 2589 PROT_READ|PROT_WRITE|PROT_EXEC, 2590 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2591 } 2592 } 2593 ::close(fd); 2594 unlink(buf); 2595 } 2596 } 2597 2598 static bool recoverable_mmap_error(int err) { 2599 // See if the error is one we can let the caller handle. This 2600 // list of errno values comes from JBS-6843484. I can't find a 2601 // Linux man page that documents this specific set of errno 2602 // values so while this list currently matches Solaris, it may 2603 // change as we gain experience with this failure mode. 2604 switch (err) { 2605 case EBADF: 2606 case EINVAL: 2607 case ENOTSUP: 2608 // let the caller deal with these errors 2609 return true; 2610 2611 default: 2612 // Any remaining errors on this OS can cause our reserved mapping 2613 // to be lost. That can cause confusion where different data 2614 // structures think they have the same memory mapped. The worst 2615 // scenario is if both the VM and a library think they have the 2616 // same memory mapped. 2617 return false; 2618 } 2619 } 2620 2621 static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2622 int err) { 2623 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2624 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec, 2625 os::strerror(err), err); 2626 } 2627 2628 static void warn_fail_commit_memory(char* addr, size_t size, 2629 size_t alignment_hint, bool exec, 2630 int err) { 2631 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2632 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, 2633 alignment_hint, exec, os::strerror(err), err); 2634 } 2635 2636 // NOTE: Linux kernel does not really reserve the pages for us. 2637 // All it does is to check if there are enough free pages 2638 // left at the time of mmap(). This could be a potential 2639 // problem. 2640 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2641 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2642 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2643 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2644 if (res != (uintptr_t) MAP_FAILED) { 2645 if (UseNUMAInterleaving) { 2646 numa_make_global(addr, size); 2647 } 2648 return 0; 2649 } 2650 2651 int err = errno; // save errno from mmap() call above 2652 2653 if (!recoverable_mmap_error(err)) { 2654 warn_fail_commit_memory(addr, size, exec, err); 2655 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2656 } 2657 2658 return err; 2659 } 2660 2661 bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2662 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2663 } 2664 2665 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2666 const char* mesg) { 2667 assert(mesg != NULL, "mesg must be specified"); 2668 int err = os::Linux::commit_memory_impl(addr, size, exec); 2669 if (err != 0) { 2670 // the caller wants all commit errors to exit with the specified mesg: 2671 warn_fail_commit_memory(addr, size, exec, err); 2672 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2673 } 2674 } 2675 2676 // Define MAP_HUGETLB here so we can build HotSpot on old systems. 2677 #ifndef MAP_HUGETLB 2678 #define MAP_HUGETLB 0x40000 2679 #endif 2680 2681 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2682 #ifndef MADV_HUGEPAGE 2683 #define MADV_HUGEPAGE 14 2684 #endif 2685 2686 int os::Linux::commit_memory_impl(char* addr, size_t size, 2687 size_t alignment_hint, bool exec) { 2688 int err = os::Linux::commit_memory_impl(addr, size, exec); 2689 if (err == 0) { 2690 realign_memory(addr, size, alignment_hint); 2691 } 2692 return err; 2693 } 2694 2695 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2696 bool exec) { 2697 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2698 } 2699 2700 void os::pd_commit_memory_or_exit(char* addr, size_t size, 2701 size_t alignment_hint, bool exec, 2702 const char* mesg) { 2703 assert(mesg != NULL, "mesg must be specified"); 2704 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2705 if (err != 0) { 2706 // the caller wants all commit errors to exit with the specified mesg: 2707 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2708 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2709 } 2710 } 2711 2712 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2713 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2714 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2715 // be supported or the memory may already be backed by huge pages. 2716 ::madvise(addr, bytes, MADV_HUGEPAGE); 2717 } 2718 } 2719 2720 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2721 // This method works by doing an mmap over an existing mmaping and effectively discarding 2722 // the existing pages. However it won't work for SHM-based large pages that cannot be 2723 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2724 // small pages on top of the SHM segment. This method always works for small pages, so we 2725 // allow that in any case. 2726 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2727 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2728 } 2729 } 2730 2731 void os::numa_make_global(char *addr, size_t bytes) { 2732 Linux::numa_interleave_memory(addr, bytes); 2733 } 2734 2735 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2736 // bind policy to MPOL_PREFERRED for the current thread. 2737 #define USE_MPOL_PREFERRED 0 2738 2739 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2740 // To make NUMA and large pages more robust when both enabled, we need to ease 2741 // the requirements on where the memory should be allocated. MPOL_BIND is the 2742 // default policy and it will force memory to be allocated on the specified 2743 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2744 // the specified node, but will not force it. Using this policy will prevent 2745 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2746 // free large pages. 2747 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2748 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2749 } 2750 2751 bool os::numa_topology_changed() { return false; } 2752 2753 size_t os::numa_get_groups_num() { 2754 int max_node = Linux::numa_max_node(); 2755 return max_node > 0 ? max_node + 1 : 1; 2756 } 2757 2758 int os::numa_get_group_id() { 2759 int cpu_id = Linux::sched_getcpu(); 2760 if (cpu_id != -1) { 2761 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2762 if (lgrp_id != -1) { 2763 return lgrp_id; 2764 } 2765 } 2766 return 0; 2767 } 2768 2769 size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2770 for (size_t i = 0; i < size; i++) { 2771 ids[i] = i; 2772 } 2773 return size; 2774 } 2775 2776 bool os::get_page_info(char *start, page_info* info) { 2777 return false; 2778 } 2779 2780 char *os::scan_pages(char *start, char* end, page_info* page_expected, 2781 page_info* page_found) { 2782 return end; 2783 } 2784 2785 2786 int os::Linux::sched_getcpu_syscall(void) { 2787 unsigned int cpu = 0; 2788 int retval = -1; 2789 2790 #if defined(IA32) 2791 #ifndef SYS_getcpu 2792 #define SYS_getcpu 318 2793 #endif 2794 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2795 #elif defined(AMD64) 2796 // Unfortunately we have to bring all these macros here from vsyscall.h 2797 // to be able to compile on old linuxes. 2798 #define __NR_vgetcpu 2 2799 #define VSYSCALL_START (-10UL << 20) 2800 #define VSYSCALL_SIZE 1024 2801 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2802 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2803 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2804 retval = vgetcpu(&cpu, NULL, NULL); 2805 #endif 2806 2807 return (retval == -1) ? retval : cpu; 2808 } 2809 2810 // Something to do with the numa-aware allocator needs these symbols 2811 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2812 extern "C" JNIEXPORT void numa_error(char *where) { } 2813 2814 2815 // If we are running with libnuma version > 2, then we should 2816 // be trying to use symbols with versions 1.1 2817 // If we are running with earlier version, which did not have symbol versions, 2818 // we should use the base version. 2819 void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2820 void *f = dlvsym(handle, name, "libnuma_1.1"); 2821 if (f == NULL) { 2822 f = dlsym(handle, name); 2823 } 2824 return f; 2825 } 2826 2827 bool os::Linux::libnuma_init() { 2828 // sched_getcpu() should be in libc. 2829 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2830 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2831 2832 // If it's not, try a direct syscall. 2833 if (sched_getcpu() == -1) { 2834 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2835 (void*)&sched_getcpu_syscall)); 2836 } 2837 2838 if (sched_getcpu() != -1) { // Does it work? 2839 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2840 if (handle != NULL) { 2841 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2842 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2843 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2844 libnuma_dlsym(handle, "numa_max_node"))); 2845 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2846 libnuma_dlsym(handle, "numa_available"))); 2847 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2848 libnuma_dlsym(handle, "numa_tonode_memory"))); 2849 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2850 libnuma_dlsym(handle, "numa_interleave_memory"))); 2851 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2852 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2853 2854 2855 if (numa_available() != -1) { 2856 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2857 // Create a cpu -> node mapping 2858 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2859 rebuild_cpu_to_node_map(); 2860 return true; 2861 } 2862 } 2863 } 2864 return false; 2865 } 2866 2867 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2868 // The table is later used in get_node_by_cpu(). 2869 void os::Linux::rebuild_cpu_to_node_map() { 2870 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2871 // in libnuma (possible values are starting from 16, 2872 // and continuing up with every other power of 2, but less 2873 // than the maximum number of CPUs supported by kernel), and 2874 // is a subject to change (in libnuma version 2 the requirements 2875 // are more reasonable) we'll just hardcode the number they use 2876 // in the library. 2877 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2878 2879 size_t cpu_num = os::active_processor_count(); 2880 size_t cpu_map_size = NCPUS / BitsPerCLong; 2881 size_t cpu_map_valid_size = 2882 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2883 2884 cpu_to_node()->clear(); 2885 cpu_to_node()->at_grow(cpu_num - 1); 2886 size_t node_num = numa_get_groups_num(); 2887 2888 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2889 for (size_t i = 0; i < node_num; i++) { 2890 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2891 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2892 if (cpu_map[j] != 0) { 2893 for (size_t k = 0; k < BitsPerCLong; k++) { 2894 if (cpu_map[j] & (1UL << k)) { 2895 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2896 } 2897 } 2898 } 2899 } 2900 } 2901 } 2902 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2903 } 2904 2905 int os::Linux::get_node_by_cpu(int cpu_id) { 2906 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2907 return cpu_to_node()->at(cpu_id); 2908 } 2909 return -1; 2910 } 2911 2912 GrowableArray<int>* os::Linux::_cpu_to_node; 2913 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2914 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2915 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2916 os::Linux::numa_available_func_t os::Linux::_numa_available; 2917 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2918 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2919 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2920 unsigned long* os::Linux::_numa_all_nodes; 2921 2922 bool os::pd_uncommit_memory(char* addr, size_t size) { 2923 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2924 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2925 return res != (uintptr_t) MAP_FAILED; 2926 } 2927 2928 static address get_stack_commited_bottom(address bottom, size_t size) { 2929 address nbot = bottom; 2930 address ntop = bottom + size; 2931 2932 size_t page_sz = os::vm_page_size(); 2933 unsigned pages = size / page_sz; 2934 2935 unsigned char vec[1]; 2936 unsigned imin = 1, imax = pages + 1, imid; 2937 int mincore_return_value = 0; 2938 2939 assert(imin <= imax, "Unexpected page size"); 2940 2941 while (imin < imax) { 2942 imid = (imax + imin) / 2; 2943 nbot = ntop - (imid * page_sz); 2944 2945 // Use a trick with mincore to check whether the page is mapped or not. 2946 // mincore sets vec to 1 if page resides in memory and to 0 if page 2947 // is swapped output but if page we are asking for is unmapped 2948 // it returns -1,ENOMEM 2949 mincore_return_value = mincore(nbot, page_sz, vec); 2950 2951 if (mincore_return_value == -1) { 2952 // Page is not mapped go up 2953 // to find first mapped page 2954 if (errno != EAGAIN) { 2955 assert(errno == ENOMEM, "Unexpected mincore errno"); 2956 imax = imid; 2957 } 2958 } else { 2959 // Page is mapped go down 2960 // to find first not mapped page 2961 imin = imid + 1; 2962 } 2963 } 2964 2965 nbot = nbot + page_sz; 2966 2967 // Adjust stack bottom one page up if last checked page is not mapped 2968 if (mincore_return_value == -1) { 2969 nbot = nbot + page_sz; 2970 } 2971 2972 return nbot; 2973 } 2974 2975 2976 // Linux uses a growable mapping for the stack, and if the mapping for 2977 // the stack guard pages is not removed when we detach a thread the 2978 // stack cannot grow beyond the pages where the stack guard was 2979 // mapped. If at some point later in the process the stack expands to 2980 // that point, the Linux kernel cannot expand the stack any further 2981 // because the guard pages are in the way, and a segfault occurs. 2982 // 2983 // However, it's essential not to split the stack region by unmapping 2984 // a region (leaving a hole) that's already part of the stack mapping, 2985 // so if the stack mapping has already grown beyond the guard pages at 2986 // the time we create them, we have to truncate the stack mapping. 2987 // So, we need to know the extent of the stack mapping when 2988 // create_stack_guard_pages() is called. 2989 2990 // We only need this for stacks that are growable: at the time of 2991 // writing thread stacks don't use growable mappings (i.e. those 2992 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 2993 // only applies to the main thread. 2994 2995 // If the (growable) stack mapping already extends beyond the point 2996 // where we're going to put our guard pages, truncate the mapping at 2997 // that point by munmap()ping it. This ensures that when we later 2998 // munmap() the guard pages we don't leave a hole in the stack 2999 // mapping. This only affects the main/initial thread 3000 3001 bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3002 if (os::Linux::is_initial_thread()) { 3003 // As we manually grow stack up to bottom inside create_attached_thread(), 3004 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3005 // we don't need to do anything special. 3006 // Check it first, before calling heavy function. 3007 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3008 unsigned char vec[1]; 3009 3010 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3011 // Fallback to slow path on all errors, including EAGAIN 3012 stack_extent = (uintptr_t) get_stack_commited_bottom( 3013 os::Linux::initial_thread_stack_bottom(), 3014 (size_t)addr - stack_extent); 3015 } 3016 3017 if (stack_extent < (uintptr_t)addr) { 3018 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3019 } 3020 } 3021 3022 return os::commit_memory(addr, size, !ExecMem); 3023 } 3024 3025 // If this is a growable mapping, remove the guard pages entirely by 3026 // munmap()ping them. If not, just call uncommit_memory(). This only 3027 // affects the main/initial thread, but guard against future OS changes 3028 // It's safe to always unmap guard pages for initial thread because we 3029 // always place it right after end of the mapped region 3030 3031 bool os::remove_stack_guard_pages(char* addr, size_t size) { 3032 uintptr_t stack_extent, stack_base; 3033 3034 if (os::Linux::is_initial_thread()) { 3035 return ::munmap(addr, size) == 0; 3036 } 3037 3038 return os::uncommit_memory(addr, size); 3039 } 3040 3041 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3042 // at 'requested_addr'. If there are existing memory mappings at the same 3043 // location, however, they will be overwritten. If 'fixed' is false, 3044 // 'requested_addr' is only treated as a hint, the return value may or 3045 // may not start from the requested address. Unlike Linux mmap(), this 3046 // function returns NULL to indicate failure. 3047 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3048 char * addr; 3049 int flags; 3050 3051 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3052 if (fixed) { 3053 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3054 flags |= MAP_FIXED; 3055 } 3056 3057 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3058 // touch an uncommitted page. Otherwise, the read/write might 3059 // succeed if we have enough swap space to back the physical page. 3060 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3061 flags, -1, 0); 3062 3063 return addr == MAP_FAILED ? NULL : addr; 3064 } 3065 3066 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address 3067 // (req_addr != NULL) or with a given alignment. 3068 // - bytes shall be a multiple of alignment. 3069 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3070 // - alignment sets the alignment at which memory shall be allocated. 3071 // It must be a multiple of allocation granularity. 3072 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3073 // req_addr or NULL. 3074 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) { 3075 3076 size_t extra_size = bytes; 3077 if (req_addr == NULL && alignment > 0) { 3078 extra_size += alignment; 3079 } 3080 3081 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE, 3082 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 3083 -1, 0); 3084 if (start == MAP_FAILED) { 3085 start = NULL; 3086 } else { 3087 if (req_addr != NULL) { 3088 if (start != req_addr) { 3089 ::munmap(start, extra_size); 3090 start = NULL; 3091 } 3092 } else { 3093 char* const start_aligned = (char*) align_ptr_up(start, alignment); 3094 char* const end_aligned = start_aligned + bytes; 3095 char* const end = start + extra_size; 3096 if (start_aligned > start) { 3097 ::munmap(start, start_aligned - start); 3098 } 3099 if (end_aligned < end) { 3100 ::munmap(end_aligned, end - end_aligned); 3101 } 3102 start = start_aligned; 3103 } 3104 } 3105 return start; 3106 } 3107 3108 static int anon_munmap(char * addr, size_t size) { 3109 return ::munmap(addr, size) == 0; 3110 } 3111 3112 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3113 size_t alignment_hint) { 3114 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3115 } 3116 3117 bool os::pd_release_memory(char* addr, size_t size) { 3118 return anon_munmap(addr, size); 3119 } 3120 3121 static bool linux_mprotect(char* addr, size_t size, int prot) { 3122 // Linux wants the mprotect address argument to be page aligned. 3123 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3124 3125 // According to SUSv3, mprotect() should only be used with mappings 3126 // established by mmap(), and mmap() always maps whole pages. Unaligned 3127 // 'addr' likely indicates problem in the VM (e.g. trying to change 3128 // protection of malloc'ed or statically allocated memory). Check the 3129 // caller if you hit this assert. 3130 assert(addr == bottom, "sanity check"); 3131 3132 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3133 return ::mprotect(bottom, size, prot) == 0; 3134 } 3135 3136 // Set protections specified 3137 bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3138 bool is_committed) { 3139 unsigned int p = 0; 3140 switch (prot) { 3141 case MEM_PROT_NONE: p = PROT_NONE; break; 3142 case MEM_PROT_READ: p = PROT_READ; break; 3143 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3144 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3145 default: 3146 ShouldNotReachHere(); 3147 } 3148 // is_committed is unused. 3149 return linux_mprotect(addr, bytes, p); 3150 } 3151 3152 bool os::guard_memory(char* addr, size_t size) { 3153 return linux_mprotect(addr, size, PROT_NONE); 3154 } 3155 3156 bool os::unguard_memory(char* addr, size_t size) { 3157 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3158 } 3159 3160 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3161 size_t page_size) { 3162 bool result = false; 3163 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3164 MAP_ANONYMOUS|MAP_PRIVATE, 3165 -1, 0); 3166 if (p != MAP_FAILED) { 3167 void *aligned_p = align_ptr_up(p, page_size); 3168 3169 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3170 3171 munmap(p, page_size * 2); 3172 } 3173 3174 if (warn && !result) { 3175 warning("TransparentHugePages is not supported by the operating system."); 3176 } 3177 3178 return result; 3179 } 3180 3181 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3182 bool result = false; 3183 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3184 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3185 -1, 0); 3186 3187 if (p != MAP_FAILED) { 3188 // We don't know if this really is a huge page or not. 3189 FILE *fp = fopen("/proc/self/maps", "r"); 3190 if (fp) { 3191 while (!feof(fp)) { 3192 char chars[257]; 3193 long x = 0; 3194 if (fgets(chars, sizeof(chars), fp)) { 3195 if (sscanf(chars, "%lx-%*x", &x) == 1 3196 && x == (long)p) { 3197 if (strstr (chars, "hugepage")) { 3198 result = true; 3199 break; 3200 } 3201 } 3202 } 3203 } 3204 fclose(fp); 3205 } 3206 munmap(p, page_size); 3207 } 3208 3209 if (warn && !result) { 3210 warning("HugeTLBFS is not supported by the operating system."); 3211 } 3212 3213 return result; 3214 } 3215 3216 // Set the coredump_filter bits to include largepages in core dump (bit 6) 3217 // 3218 // From the coredump_filter documentation: 3219 // 3220 // - (bit 0) anonymous private memory 3221 // - (bit 1) anonymous shared memory 3222 // - (bit 2) file-backed private memory 3223 // - (bit 3) file-backed shared memory 3224 // - (bit 4) ELF header pages in file-backed private memory areas (it is 3225 // effective only if the bit 2 is cleared) 3226 // - (bit 5) hugetlb private memory 3227 // - (bit 6) hugetlb shared memory 3228 // 3229 static void set_coredump_filter(void) { 3230 FILE *f; 3231 long cdm; 3232 3233 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3234 return; 3235 } 3236 3237 if (fscanf(f, "%lx", &cdm) != 1) { 3238 fclose(f); 3239 return; 3240 } 3241 3242 rewind(f); 3243 3244 if ((cdm & LARGEPAGES_BIT) == 0) { 3245 cdm |= LARGEPAGES_BIT; 3246 fprintf(f, "%#lx", cdm); 3247 } 3248 3249 fclose(f); 3250 } 3251 3252 // Large page support 3253 3254 static size_t _large_page_size = 0; 3255 3256 size_t os::Linux::find_large_page_size() { 3257 size_t large_page_size = 0; 3258 3259 // large_page_size on Linux is used to round up heap size. x86 uses either 3260 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3261 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3262 // page as large as 256M. 3263 // 3264 // Here we try to figure out page size by parsing /proc/meminfo and looking 3265 // for a line with the following format: 3266 // Hugepagesize: 2048 kB 3267 // 3268 // If we can't determine the value (e.g. /proc is not mounted, or the text 3269 // format has been changed), we'll use the largest page size supported by 3270 // the processor. 3271 3272 #ifndef ZERO 3273 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3274 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M); 3275 #endif // ZERO 3276 3277 FILE *fp = fopen("/proc/meminfo", "r"); 3278 if (fp) { 3279 while (!feof(fp)) { 3280 int x = 0; 3281 char buf[16]; 3282 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3283 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3284 large_page_size = x * K; 3285 break; 3286 } 3287 } else { 3288 // skip to next line 3289 for (;;) { 3290 int ch = fgetc(fp); 3291 if (ch == EOF || ch == (int)'\n') break; 3292 } 3293 } 3294 } 3295 fclose(fp); 3296 } 3297 3298 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3299 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3300 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3301 proper_unit_for_byte_size(large_page_size)); 3302 } 3303 3304 return large_page_size; 3305 } 3306 3307 size_t os::Linux::setup_large_page_size() { 3308 _large_page_size = Linux::find_large_page_size(); 3309 const size_t default_page_size = (size_t)Linux::page_size(); 3310 if (_large_page_size > default_page_size) { 3311 _page_sizes[0] = _large_page_size; 3312 _page_sizes[1] = default_page_size; 3313 _page_sizes[2] = 0; 3314 } 3315 3316 return _large_page_size; 3317 } 3318 3319 bool os::Linux::setup_large_page_type(size_t page_size) { 3320 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3321 FLAG_IS_DEFAULT(UseSHM) && 3322 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3323 3324 // The type of large pages has not been specified by the user. 3325 3326 // Try UseHugeTLBFS and then UseSHM. 3327 UseHugeTLBFS = UseSHM = true; 3328 3329 // Don't try UseTransparentHugePages since there are known 3330 // performance issues with it turned on. This might change in the future. 3331 UseTransparentHugePages = false; 3332 } 3333 3334 if (UseTransparentHugePages) { 3335 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3336 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3337 UseHugeTLBFS = false; 3338 UseSHM = false; 3339 return true; 3340 } 3341 UseTransparentHugePages = false; 3342 } 3343 3344 if (UseHugeTLBFS) { 3345 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3346 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3347 UseSHM = false; 3348 return true; 3349 } 3350 UseHugeTLBFS = false; 3351 } 3352 3353 return UseSHM; 3354 } 3355 3356 void os::large_page_init() { 3357 if (!UseLargePages && 3358 !UseTransparentHugePages && 3359 !UseHugeTLBFS && 3360 !UseSHM) { 3361 // Not using large pages. 3362 return; 3363 } 3364 3365 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3366 // The user explicitly turned off large pages. 3367 // Ignore the rest of the large pages flags. 3368 UseTransparentHugePages = false; 3369 UseHugeTLBFS = false; 3370 UseSHM = false; 3371 return; 3372 } 3373 3374 size_t large_page_size = Linux::setup_large_page_size(); 3375 UseLargePages = Linux::setup_large_page_type(large_page_size); 3376 3377 set_coredump_filter(); 3378 } 3379 3380 #ifndef SHM_HUGETLB 3381 #define SHM_HUGETLB 04000 3382 #endif 3383 3384 #define shm_warning_format(format, ...) \ 3385 do { \ 3386 if (UseLargePages && \ 3387 (!FLAG_IS_DEFAULT(UseLargePages) || \ 3388 !FLAG_IS_DEFAULT(UseSHM) || \ 3389 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \ 3390 warning(format, __VA_ARGS__); \ 3391 } \ 3392 } while (0) 3393 3394 #define shm_warning(str) shm_warning_format("%s", str) 3395 3396 #define shm_warning_with_errno(str) \ 3397 do { \ 3398 int err = errno; \ 3399 shm_warning_format(str " (error = %d)", err); \ 3400 } while (0) 3401 3402 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) { 3403 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment"); 3404 3405 if (!is_size_aligned(alignment, SHMLBA)) { 3406 assert(false, "Code below assumes that alignment is at least SHMLBA aligned"); 3407 return NULL; 3408 } 3409 3410 // To ensure that we get 'alignment' aligned memory from shmat, 3411 // we pre-reserve aligned virtual memory and then attach to that. 3412 3413 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL); 3414 if (pre_reserved_addr == NULL) { 3415 // Couldn't pre-reserve aligned memory. 3416 shm_warning("Failed to pre-reserve aligned memory for shmat."); 3417 return NULL; 3418 } 3419 3420 // SHM_REMAP is needed to allow shmat to map over an existing mapping. 3421 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP); 3422 3423 if ((intptr_t)addr == -1) { 3424 int err = errno; 3425 shm_warning_with_errno("Failed to attach shared memory."); 3426 3427 assert(err != EACCES, "Unexpected error"); 3428 assert(err != EIDRM, "Unexpected error"); 3429 assert(err != EINVAL, "Unexpected error"); 3430 3431 // Since we don't know if the kernel unmapped the pre-reserved memory area 3432 // we can't unmap it, since that would potentially unmap memory that was 3433 // mapped from other threads. 3434 return NULL; 3435 } 3436 3437 return addr; 3438 } 3439 3440 static char* shmat_at_address(int shmid, char* req_addr) { 3441 if (!is_ptr_aligned(req_addr, SHMLBA)) { 3442 assert(false, "Requested address needs to be SHMLBA aligned"); 3443 return NULL; 3444 } 3445 3446 char* addr = (char*)shmat(shmid, req_addr, 0); 3447 3448 if ((intptr_t)addr == -1) { 3449 shm_warning_with_errno("Failed to attach shared memory."); 3450 return NULL; 3451 } 3452 3453 return addr; 3454 } 3455 3456 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) { 3457 // If a req_addr has been provided, we assume that the caller has already aligned the address. 3458 if (req_addr != NULL) { 3459 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size"); 3460 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment"); 3461 return shmat_at_address(shmid, req_addr); 3462 } 3463 3464 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically 3465 // return large page size aligned memory addresses when req_addr == NULL. 3466 // However, if the alignment is larger than the large page size, we have 3467 // to manually ensure that the memory returned is 'alignment' aligned. 3468 if (alignment > os::large_page_size()) { 3469 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size"); 3470 return shmat_with_alignment(shmid, bytes, alignment); 3471 } else { 3472 return shmat_at_address(shmid, NULL); 3473 } 3474 } 3475 3476 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3477 char* req_addr, bool exec) { 3478 // "exec" is passed in but not used. Creating the shared image for 3479 // the code cache doesn't have an SHM_X executable permission to check. 3480 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3481 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3482 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address"); 3483 3484 if (!is_size_aligned(bytes, os::large_page_size())) { 3485 return NULL; // Fallback to small pages. 3486 } 3487 3488 // Create a large shared memory region to attach to based on size. 3489 // Currently, size is the total size of the heap. 3490 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3491 if (shmid == -1) { 3492 // Possible reasons for shmget failure: 3493 // 1. shmmax is too small for Java heap. 3494 // > check shmmax value: cat /proc/sys/kernel/shmmax 3495 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3496 // 2. not enough large page memory. 3497 // > check available large pages: cat /proc/meminfo 3498 // > increase amount of large pages: 3499 // echo new_value > /proc/sys/vm/nr_hugepages 3500 // Note 1: different Linux may use different name for this property, 3501 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3502 // Note 2: it's possible there's enough physical memory available but 3503 // they are so fragmented after a long run that they can't 3504 // coalesce into large pages. Try to reserve large pages when 3505 // the system is still "fresh". 3506 shm_warning_with_errno("Failed to reserve shared memory."); 3507 return NULL; 3508 } 3509 3510 // Attach to the region. 3511 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr); 3512 3513 // Remove shmid. If shmat() is successful, the actual shared memory segment 3514 // will be deleted when it's detached by shmdt() or when the process 3515 // terminates. If shmat() is not successful this will remove the shared 3516 // segment immediately. 3517 shmctl(shmid, IPC_RMID, NULL); 3518 3519 return addr; 3520 } 3521 3522 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3523 int error) { 3524 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3525 3526 bool warn_on_failure = UseLargePages && 3527 (!FLAG_IS_DEFAULT(UseLargePages) || 3528 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3529 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3530 3531 if (warn_on_failure) { 3532 char msg[128]; 3533 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3534 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3535 warning("%s", msg); 3536 } 3537 } 3538 3539 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3540 char* req_addr, 3541 bool exec) { 3542 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3543 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3544 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3545 3546 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3547 char* addr = (char*)::mmap(req_addr, bytes, prot, 3548 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3549 -1, 0); 3550 3551 if (addr == MAP_FAILED) { 3552 warn_on_large_pages_failure(req_addr, bytes, errno); 3553 return NULL; 3554 } 3555 3556 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3557 3558 return addr; 3559 } 3560 3561 // Helper function to create a temp file in the given directory 3562 int os::create_tmpfile(const char* dir, size_t size, bool exec) { 3563 3564 char name_template[] = "/jvmheap.XXXXXX"; 3565 3566 char *fullname = (char*)alloca(strlen(dir) + sizeof(name_template)); 3567 (void)strcpy(fullname, dir); 3568 (void)strcat(fullname, name_template); 3569 3570 sigset_t set, oldset; 3571 sigfillset(&set); 3572 3573 // block all signals while we do the file operation 3574 (void)sigprocmask(SIG_BLOCK, &set, &oldset); 3575 3576 // set the file creation mask 3577 mode_t new_mask = exec ? (S_IRUSR | S_IWUSR | S_IXUSR) : (S_IRUSR | S_IWUSR); 3578 mode_t prev_umask = umask(new_mask); 3579 3580 // create a new file 3581 int fd = mkstemp(fullname); 3582 3583 // reset the file creation mask 3584 umask(prev_umask); 3585 3586 if (fd < 0) { 3587 warning("Could not create file for heap"); 3588 return -1; 3589 } 3590 3591 // delete the name from the filesystem. When 'fd' is closed, the file (and space) will be deleted 3592 (void)unlink(fullname); 3593 3594 // reset the signal mask 3595 (void)sigprocmask(SIG_SETMASK, &oldset, NULL); 3596 3597 // allocate space for the file 3598 if ((errno = posix_fallocate(fd, 0, (off_t)size)) != 0) { 3599 warning("Could not allocate sufficient disk space for heap"); 3600 return -1; 3601 } 3602 3603 return fd; 3604 } 3605 3606 // Map the given address range to a temporary file created at the specified directory. 3607 // The address range must already be reserved for guaranteed success. If it not reserved, their could be an error while mapping leading to JVM shutdown 3608 bool os::map_memory_to_file(char* base, size_t size, const char* backingFileDir) { 3609 3610 int fd = os::create_tmpfile(backingFileDir, size, false); 3611 if (fd == -1) { 3612 vm_exit_during_initialization(err_msg("Could not create temporary file in %s for object heap", backingFileDir)); 3613 return false; 3614 } 3615 int prot = PROT_READ | PROT_WRITE; 3616 char* addr = (char*)mmap(base, size, prot, MAP_SHARED | MAP_FIXED, fd, 0); 3617 3618 if (addr == MAP_FAILED || addr != base) { 3619 close(fd); 3620 vm_exit_during_initialization(err_msg("Error in mapping object heap at the given filesystem dir %s", backingFileDir)); 3621 return false; 3622 } 3623 3624 return true; 3625 } 3626 3627 // Reserve memory using mmap(MAP_HUGETLB). 3628 // - bytes shall be a multiple of alignment. 3629 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3630 // - alignment sets the alignment at which memory shall be allocated. 3631 // It must be a multiple of allocation granularity. 3632 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3633 // req_addr or NULL. 3634 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3635 size_t alignment, 3636 char* req_addr, 3637 bool exec) { 3638 size_t large_page_size = os::large_page_size(); 3639 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3640 3641 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3642 assert(is_size_aligned(bytes, alignment), "Must be"); 3643 3644 // First reserve - but not commit - the address range in small pages. 3645 char* const start = anon_mmap_aligned(bytes, alignment, req_addr); 3646 3647 if (start == NULL) { 3648 return NULL; 3649 } 3650 3651 assert(is_ptr_aligned(start, alignment), "Must be"); 3652 3653 char* end = start + bytes; 3654 3655 // Find the regions of the allocated chunk that can be promoted to large pages. 3656 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3657 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3658 3659 size_t lp_bytes = lp_end - lp_start; 3660 3661 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3662 3663 if (lp_bytes == 0) { 3664 // The mapped region doesn't even span the start and the end of a large page. 3665 // Fall back to allocate a non-special area. 3666 ::munmap(start, end - start); 3667 return NULL; 3668 } 3669 3670 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3671 3672 void* result; 3673 3674 // Commit small-paged leading area. 3675 if (start != lp_start) { 3676 result = ::mmap(start, lp_start - start, prot, 3677 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3678 -1, 0); 3679 if (result == MAP_FAILED) { 3680 ::munmap(lp_start, end - lp_start); 3681 return NULL; 3682 } 3683 } 3684 3685 // Commit large-paged area. 3686 result = ::mmap(lp_start, lp_bytes, prot, 3687 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3688 -1, 0); 3689 if (result == MAP_FAILED) { 3690 warn_on_large_pages_failure(lp_start, lp_bytes, errno); 3691 // If the mmap above fails, the large pages region will be unmapped and we 3692 // have regions before and after with small pages. Release these regions. 3693 // 3694 // | mapped | unmapped | mapped | 3695 // ^ ^ ^ ^ 3696 // start lp_start lp_end end 3697 // 3698 ::munmap(start, lp_start - start); 3699 ::munmap(lp_end, end - lp_end); 3700 return NULL; 3701 } 3702 3703 // Commit small-paged trailing area. 3704 if (lp_end != end) { 3705 result = ::mmap(lp_end, end - lp_end, prot, 3706 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3707 -1, 0); 3708 if (result == MAP_FAILED) { 3709 ::munmap(start, lp_end - start); 3710 return NULL; 3711 } 3712 } 3713 3714 return start; 3715 } 3716 3717 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3718 size_t alignment, 3719 char* req_addr, 3720 bool exec) { 3721 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3722 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3723 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be"); 3724 assert(is_power_of_2(os::large_page_size()), "Must be"); 3725 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3726 3727 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3728 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3729 } else { 3730 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3731 } 3732 } 3733 3734 char* os::reserve_memory_special(size_t bytes, size_t alignment, 3735 char* req_addr, bool exec) { 3736 assert(UseLargePages, "only for large pages"); 3737 3738 char* addr; 3739 if (UseSHM) { 3740 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3741 } else { 3742 assert(UseHugeTLBFS, "must be"); 3743 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3744 } 3745 3746 if (addr != NULL) { 3747 if (UseNUMAInterleaving) { 3748 numa_make_global(addr, bytes); 3749 } 3750 3751 // The memory is committed 3752 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3753 } 3754 3755 return addr; 3756 } 3757 3758 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3759 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3760 return shmdt(base) == 0; 3761 } 3762 3763 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3764 return pd_release_memory(base, bytes); 3765 } 3766 3767 bool os::release_memory_special(char* base, size_t bytes) { 3768 bool res; 3769 if (MemTracker::tracking_level() > NMT_minimal) { 3770 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3771 res = os::Linux::release_memory_special_impl(base, bytes); 3772 if (res) { 3773 tkr.record((address)base, bytes); 3774 } 3775 3776 } else { 3777 res = os::Linux::release_memory_special_impl(base, bytes); 3778 } 3779 return res; 3780 } 3781 3782 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3783 assert(UseLargePages, "only for large pages"); 3784 bool res; 3785 3786 if (UseSHM) { 3787 res = os::Linux::release_memory_special_shm(base, bytes); 3788 } else { 3789 assert(UseHugeTLBFS, "must be"); 3790 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3791 } 3792 return res; 3793 } 3794 3795 size_t os::large_page_size() { 3796 return _large_page_size; 3797 } 3798 3799 // With SysV SHM the entire memory region must be allocated as shared 3800 // memory. 3801 // HugeTLBFS allows application to commit large page memory on demand. 3802 // However, when committing memory with HugeTLBFS fails, the region 3803 // that was supposed to be committed will lose the old reservation 3804 // and allow other threads to steal that memory region. Because of this 3805 // behavior we can't commit HugeTLBFS memory. 3806 bool os::can_commit_large_page_memory() { 3807 return UseTransparentHugePages; 3808 } 3809 3810 bool os::can_execute_large_page_memory() { 3811 return UseTransparentHugePages || UseHugeTLBFS; 3812 } 3813 3814 // Reserve memory at an arbitrary address, only if that area is 3815 // available (and not reserved for something else). 3816 3817 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3818 const int max_tries = 10; 3819 char* base[max_tries]; 3820 size_t size[max_tries]; 3821 const size_t gap = 0x000000; 3822 3823 // Assert only that the size is a multiple of the page size, since 3824 // that's all that mmap requires, and since that's all we really know 3825 // about at this low abstraction level. If we need higher alignment, 3826 // we can either pass an alignment to this method or verify alignment 3827 // in one of the methods further up the call chain. See bug 5044738. 3828 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3829 3830 // Repeatedly allocate blocks until the block is allocated at the 3831 // right spot. 3832 3833 // Linux mmap allows caller to pass an address as hint; give it a try first, 3834 // if kernel honors the hint then we can return immediately. 3835 char * addr = anon_mmap(requested_addr, bytes, false); 3836 if (addr == requested_addr) { 3837 return requested_addr; 3838 } 3839 3840 if (addr != NULL) { 3841 // mmap() is successful but it fails to reserve at the requested address 3842 anon_munmap(addr, bytes); 3843 } 3844 3845 int i; 3846 for (i = 0; i < max_tries; ++i) { 3847 base[i] = reserve_memory(bytes); 3848 3849 if (base[i] != NULL) { 3850 // Is this the block we wanted? 3851 if (base[i] == requested_addr) { 3852 size[i] = bytes; 3853 break; 3854 } 3855 3856 // Does this overlap the block we wanted? Give back the overlapped 3857 // parts and try again. 3858 3859 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3860 if (top_overlap >= 0 && (size_t)top_overlap < bytes) { 3861 unmap_memory(base[i], top_overlap); 3862 base[i] += top_overlap; 3863 size[i] = bytes - top_overlap; 3864 } else { 3865 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr; 3866 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) { 3867 unmap_memory(requested_addr, bottom_overlap); 3868 size[i] = bytes - bottom_overlap; 3869 } else { 3870 size[i] = bytes; 3871 } 3872 } 3873 } 3874 } 3875 3876 // Give back the unused reserved pieces. 3877 3878 for (int j = 0; j < i; ++j) { 3879 if (base[j] != NULL) { 3880 unmap_memory(base[j], size[j]); 3881 } 3882 } 3883 3884 if (i < max_tries) { 3885 return requested_addr; 3886 } else { 3887 return NULL; 3888 } 3889 } 3890 3891 size_t os::read(int fd, void *buf, unsigned int nBytes) { 3892 return ::read(fd, buf, nBytes); 3893 } 3894 3895 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) { 3896 return ::pread(fd, buf, nBytes, offset); 3897 } 3898 3899 // Short sleep, direct OS call. 3900 // 3901 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3902 // sched_yield(2) will actually give up the CPU: 3903 // 3904 // * Alone on this pariticular CPU, keeps running. 3905 // * Before the introduction of "skip_buddy" with "compat_yield" disabled 3906 // (pre 2.6.39). 3907 // 3908 // So calling this with 0 is an alternative. 3909 // 3910 void os::naked_short_sleep(jlong ms) { 3911 struct timespec req; 3912 3913 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3914 req.tv_sec = 0; 3915 if (ms > 0) { 3916 req.tv_nsec = (ms % 1000) * 1000000; 3917 } else { 3918 req.tv_nsec = 1; 3919 } 3920 3921 nanosleep(&req, NULL); 3922 3923 return; 3924 } 3925 3926 // Sleep forever; naked call to OS-specific sleep; use with CAUTION 3927 void os::infinite_sleep() { 3928 while (true) { // sleep forever ... 3929 ::sleep(100); // ... 100 seconds at a time 3930 } 3931 } 3932 3933 // Used to convert frequent JVM_Yield() to nops 3934 bool os::dont_yield() { 3935 return DontYieldALot; 3936 } 3937 3938 void os::naked_yield() { 3939 sched_yield(); 3940 } 3941 3942 //////////////////////////////////////////////////////////////////////////////// 3943 // thread priority support 3944 3945 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3946 // only supports dynamic priority, static priority must be zero. For real-time 3947 // applications, Linux supports SCHED_RR which allows static priority (1-99). 3948 // However, for large multi-threaded applications, SCHED_RR is not only slower 3949 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3950 // of 5 runs - Sep 2005). 3951 // 3952 // The following code actually changes the niceness of kernel-thread/LWP. It 3953 // has an assumption that setpriority() only modifies one kernel-thread/LWP, 3954 // not the entire user process, and user level threads are 1:1 mapped to kernel 3955 // threads. It has always been the case, but could change in the future. For 3956 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3957 // It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3958 3959 int os::java_to_os_priority[CriticalPriority + 1] = { 3960 19, // 0 Entry should never be used 3961 3962 4, // 1 MinPriority 3963 3, // 2 3964 2, // 3 3965 3966 1, // 4 3967 0, // 5 NormPriority 3968 -1, // 6 3969 3970 -2, // 7 3971 -3, // 8 3972 -4, // 9 NearMaxPriority 3973 3974 -5, // 10 MaxPriority 3975 3976 -5 // 11 CriticalPriority 3977 }; 3978 3979 static int prio_init() { 3980 if (ThreadPriorityPolicy == 1) { 3981 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3982 // if effective uid is not root. Perhaps, a more elegant way of doing 3983 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3984 if (geteuid() != 0) { 3985 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3986 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3987 } 3988 ThreadPriorityPolicy = 0; 3989 } 3990 } 3991 if (UseCriticalJavaThreadPriority) { 3992 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3993 } 3994 return 0; 3995 } 3996 3997 OSReturn os::set_native_priority(Thread* thread, int newpri) { 3998 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 3999 4000 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 4001 return (ret == 0) ? OS_OK : OS_ERR; 4002 } 4003 4004 OSReturn os::get_native_priority(const Thread* const thread, 4005 int *priority_ptr) { 4006 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 4007 *priority_ptr = java_to_os_priority[NormPriority]; 4008 return OS_OK; 4009 } 4010 4011 errno = 0; 4012 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 4013 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 4014 } 4015 4016 // Hint to the underlying OS that a task switch would not be good. 4017 // Void return because it's a hint and can fail. 4018 void os::hint_no_preempt() {} 4019 4020 //////////////////////////////////////////////////////////////////////////////// 4021 // suspend/resume support 4022 4023 // the low-level signal-based suspend/resume support is a remnant from the 4024 // old VM-suspension that used to be for java-suspension, safepoints etc, 4025 // within hotspot. Now there is a single use-case for this: 4026 // - calling get_thread_pc() on the VMThread by the flat-profiler task 4027 // that runs in the watcher thread. 4028 // The remaining code is greatly simplified from the more general suspension 4029 // code that used to be used. 4030 // 4031 // The protocol is quite simple: 4032 // - suspend: 4033 // - sends a signal to the target thread 4034 // - polls the suspend state of the osthread using a yield loop 4035 // - target thread signal handler (SR_handler) sets suspend state 4036 // and blocks in sigsuspend until continued 4037 // - resume: 4038 // - sets target osthread state to continue 4039 // - sends signal to end the sigsuspend loop in the SR_handler 4040 // 4041 // Note that the SR_lock plays no role in this suspend/resume protocol. 4042 4043 static void resume_clear_context(OSThread *osthread) { 4044 osthread->set_ucontext(NULL); 4045 osthread->set_siginfo(NULL); 4046 } 4047 4048 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 4049 ucontext_t* context) { 4050 osthread->set_ucontext(context); 4051 osthread->set_siginfo(siginfo); 4052 } 4053 4054 // Handler function invoked when a thread's execution is suspended or 4055 // resumed. We have to be careful that only async-safe functions are 4056 // called here (Note: most pthread functions are not async safe and 4057 // should be avoided.) 4058 // 4059 // Note: sigwait() is a more natural fit than sigsuspend() from an 4060 // interface point of view, but sigwait() prevents the signal hander 4061 // from being run. libpthread would get very confused by not having 4062 // its signal handlers run and prevents sigwait()'s use with the 4063 // mutex granting granting signal. 4064 // 4065 // Currently only ever called on the VMThread and JavaThreads (PC sampling) 4066 // 4067 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 4068 // Save and restore errno to avoid confusing native code with EINTR 4069 // after sigsuspend. 4070 int old_errno = errno; 4071 4072 Thread* thread = Thread::current_or_null_safe(); 4073 assert(thread != NULL, "Missing current thread in SR_handler"); 4074 OSThread* osthread = thread->osthread(); 4075 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 4076 4077 os::SuspendResume::State current = osthread->sr.state(); 4078 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 4079 suspend_save_context(osthread, siginfo, context); 4080 4081 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4082 os::SuspendResume::State state = osthread->sr.suspended(); 4083 if (state == os::SuspendResume::SR_SUSPENDED) { 4084 sigset_t suspend_set; // signals for sigsuspend() 4085 sigemptyset(&suspend_set); 4086 // get current set of blocked signals and unblock resume signal 4087 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4088 sigdelset(&suspend_set, SR_signum); 4089 4090 sr_semaphore.signal(); 4091 // wait here until we are resumed 4092 while (1) { 4093 sigsuspend(&suspend_set); 4094 4095 os::SuspendResume::State result = osthread->sr.running(); 4096 if (result == os::SuspendResume::SR_RUNNING) { 4097 sr_semaphore.signal(); 4098 break; 4099 } 4100 } 4101 4102 } else if (state == os::SuspendResume::SR_RUNNING) { 4103 // request was cancelled, continue 4104 } else { 4105 ShouldNotReachHere(); 4106 } 4107 4108 resume_clear_context(osthread); 4109 } else if (current == os::SuspendResume::SR_RUNNING) { 4110 // request was cancelled, continue 4111 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4112 // ignore 4113 } else { 4114 // ignore 4115 } 4116 4117 errno = old_errno; 4118 } 4119 4120 static int SR_initialize() { 4121 struct sigaction act; 4122 char *s; 4123 4124 // Get signal number to use for suspend/resume 4125 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4126 int sig = ::strtol(s, 0, 10); 4127 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769. 4128 sig < NSIG) { // Must be legal signal and fit into sigflags[]. 4129 SR_signum = sig; 4130 } else { 4131 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.", 4132 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum); 4133 } 4134 } 4135 4136 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4137 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4138 4139 sigemptyset(&SR_sigset); 4140 sigaddset(&SR_sigset, SR_signum); 4141 4142 // Set up signal handler for suspend/resume 4143 act.sa_flags = SA_RESTART|SA_SIGINFO; 4144 act.sa_handler = (void (*)(int)) SR_handler; 4145 4146 // SR_signum is blocked by default. 4147 // 4528190 - We also need to block pthread restart signal (32 on all 4148 // supported Linux platforms). Note that LinuxThreads need to block 4149 // this signal for all threads to work properly. So we don't have 4150 // to use hard-coded signal number when setting up the mask. 4151 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4152 4153 if (sigaction(SR_signum, &act, 0) == -1) { 4154 return -1; 4155 } 4156 4157 // Save signal flag 4158 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4159 return 0; 4160 } 4161 4162 static int sr_notify(OSThread* osthread) { 4163 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4164 assert_status(status == 0, status, "pthread_kill"); 4165 return status; 4166 } 4167 4168 // "Randomly" selected value for how long we want to spin 4169 // before bailing out on suspending a thread, also how often 4170 // we send a signal to a thread we want to resume 4171 static const int RANDOMLY_LARGE_INTEGER = 1000000; 4172 static const int RANDOMLY_LARGE_INTEGER2 = 100; 4173 4174 // returns true on success and false on error - really an error is fatal 4175 // but this seems the normal response to library errors 4176 static bool do_suspend(OSThread* osthread) { 4177 assert(osthread->sr.is_running(), "thread should be running"); 4178 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4179 4180 // mark as suspended and send signal 4181 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4182 // failed to switch, state wasn't running? 4183 ShouldNotReachHere(); 4184 return false; 4185 } 4186 4187 if (sr_notify(osthread) != 0) { 4188 ShouldNotReachHere(); 4189 } 4190 4191 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4192 while (true) { 4193 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4194 break; 4195 } else { 4196 // timeout 4197 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4198 if (cancelled == os::SuspendResume::SR_RUNNING) { 4199 return false; 4200 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4201 // make sure that we consume the signal on the semaphore as well 4202 sr_semaphore.wait(); 4203 break; 4204 } else { 4205 ShouldNotReachHere(); 4206 return false; 4207 } 4208 } 4209 } 4210 4211 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4212 return true; 4213 } 4214 4215 static void do_resume(OSThread* osthread) { 4216 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4217 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4218 4219 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4220 // failed to switch to WAKEUP_REQUEST 4221 ShouldNotReachHere(); 4222 return; 4223 } 4224 4225 while (true) { 4226 if (sr_notify(osthread) == 0) { 4227 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4228 if (osthread->sr.is_running()) { 4229 return; 4230 } 4231 } 4232 } else { 4233 ShouldNotReachHere(); 4234 } 4235 } 4236 4237 guarantee(osthread->sr.is_running(), "Must be running!"); 4238 } 4239 4240 /////////////////////////////////////////////////////////////////////////////////// 4241 // signal handling (except suspend/resume) 4242 4243 // This routine may be used by user applications as a "hook" to catch signals. 4244 // The user-defined signal handler must pass unrecognized signals to this 4245 // routine, and if it returns true (non-zero), then the signal handler must 4246 // return immediately. If the flag "abort_if_unrecognized" is true, then this 4247 // routine will never retun false (zero), but instead will execute a VM panic 4248 // routine kill the process. 4249 // 4250 // If this routine returns false, it is OK to call it again. This allows 4251 // the user-defined signal handler to perform checks either before or after 4252 // the VM performs its own checks. Naturally, the user code would be making 4253 // a serious error if it tried to handle an exception (such as a null check 4254 // or breakpoint) that the VM was generating for its own correct operation. 4255 // 4256 // This routine may recognize any of the following kinds of signals: 4257 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4258 // It should be consulted by handlers for any of those signals. 4259 // 4260 // The caller of this routine must pass in the three arguments supplied 4261 // to the function referred to in the "sa_sigaction" (not the "sa_handler") 4262 // field of the structure passed to sigaction(). This routine assumes that 4263 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4264 // 4265 // Note that the VM will print warnings if it detects conflicting signal 4266 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4267 // 4268 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4269 siginfo_t* siginfo, 4270 void* ucontext, 4271 int abort_if_unrecognized); 4272 4273 void signalHandler(int sig, siginfo_t* info, void* uc) { 4274 assert(info != NULL && uc != NULL, "it must be old kernel"); 4275 int orig_errno = errno; // Preserve errno value over signal handler. 4276 JVM_handle_linux_signal(sig, info, uc, true); 4277 errno = orig_errno; 4278 } 4279 4280 4281 // This boolean allows users to forward their own non-matching signals 4282 // to JVM_handle_linux_signal, harmlessly. 4283 bool os::Linux::signal_handlers_are_installed = false; 4284 4285 // For signal-chaining 4286 struct sigaction sigact[NSIG]; 4287 uint64_t sigs = 0; 4288 #if (64 < NSIG-1) 4289 #error "Not all signals can be encoded in sigs. Adapt its type!" 4290 #endif 4291 bool os::Linux::libjsig_is_loaded = false; 4292 typedef struct sigaction *(*get_signal_t)(int); 4293 get_signal_t os::Linux::get_signal_action = NULL; 4294 4295 struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4296 struct sigaction *actp = NULL; 4297 4298 if (libjsig_is_loaded) { 4299 // Retrieve the old signal handler from libjsig 4300 actp = (*get_signal_action)(sig); 4301 } 4302 if (actp == NULL) { 4303 // Retrieve the preinstalled signal handler from jvm 4304 actp = get_preinstalled_handler(sig); 4305 } 4306 4307 return actp; 4308 } 4309 4310 static bool call_chained_handler(struct sigaction *actp, int sig, 4311 siginfo_t *siginfo, void *context) { 4312 // Call the old signal handler 4313 if (actp->sa_handler == SIG_DFL) { 4314 // It's more reasonable to let jvm treat it as an unexpected exception 4315 // instead of taking the default action. 4316 return false; 4317 } else if (actp->sa_handler != SIG_IGN) { 4318 if ((actp->sa_flags & SA_NODEFER) == 0) { 4319 // automaticlly block the signal 4320 sigaddset(&(actp->sa_mask), sig); 4321 } 4322 4323 sa_handler_t hand = NULL; 4324 sa_sigaction_t sa = NULL; 4325 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4326 // retrieve the chained handler 4327 if (siginfo_flag_set) { 4328 sa = actp->sa_sigaction; 4329 } else { 4330 hand = actp->sa_handler; 4331 } 4332 4333 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4334 actp->sa_handler = SIG_DFL; 4335 } 4336 4337 // try to honor the signal mask 4338 sigset_t oset; 4339 sigemptyset(&oset); 4340 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4341 4342 // call into the chained handler 4343 if (siginfo_flag_set) { 4344 (*sa)(sig, siginfo, context); 4345 } else { 4346 (*hand)(sig); 4347 } 4348 4349 // restore the signal mask 4350 pthread_sigmask(SIG_SETMASK, &oset, NULL); 4351 } 4352 // Tell jvm's signal handler the signal is taken care of. 4353 return true; 4354 } 4355 4356 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4357 bool chained = false; 4358 // signal-chaining 4359 if (UseSignalChaining) { 4360 struct sigaction *actp = get_chained_signal_action(sig); 4361 if (actp != NULL) { 4362 chained = call_chained_handler(actp, sig, siginfo, context); 4363 } 4364 } 4365 return chained; 4366 } 4367 4368 struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4369 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) { 4370 return &sigact[sig]; 4371 } 4372 return NULL; 4373 } 4374 4375 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4376 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4377 sigact[sig] = oldAct; 4378 sigs |= (uint64_t)1 << (sig-1); 4379 } 4380 4381 // for diagnostic 4382 int sigflags[NSIG]; 4383 4384 int os::Linux::get_our_sigflags(int sig) { 4385 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4386 return sigflags[sig]; 4387 } 4388 4389 void os::Linux::set_our_sigflags(int sig, int flags) { 4390 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4391 if (sig > 0 && sig < NSIG) { 4392 sigflags[sig] = flags; 4393 } 4394 } 4395 4396 void os::Linux::set_signal_handler(int sig, bool set_installed) { 4397 // Check for overwrite. 4398 struct sigaction oldAct; 4399 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4400 4401 void* oldhand = oldAct.sa_sigaction 4402 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4403 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4404 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4405 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4406 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4407 if (AllowUserSignalHandlers || !set_installed) { 4408 // Do not overwrite; user takes responsibility to forward to us. 4409 return; 4410 } else if (UseSignalChaining) { 4411 // save the old handler in jvm 4412 save_preinstalled_handler(sig, oldAct); 4413 // libjsig also interposes the sigaction() call below and saves the 4414 // old sigaction on it own. 4415 } else { 4416 fatal("Encountered unexpected pre-existing sigaction handler " 4417 "%#lx for signal %d.", (long)oldhand, sig); 4418 } 4419 } 4420 4421 struct sigaction sigAct; 4422 sigfillset(&(sigAct.sa_mask)); 4423 sigAct.sa_handler = SIG_DFL; 4424 if (!set_installed) { 4425 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4426 } else { 4427 sigAct.sa_sigaction = signalHandler; 4428 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4429 } 4430 // Save flags, which are set by ours 4431 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4432 sigflags[sig] = sigAct.sa_flags; 4433 4434 int ret = sigaction(sig, &sigAct, &oldAct); 4435 assert(ret == 0, "check"); 4436 4437 void* oldhand2 = oldAct.sa_sigaction 4438 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4439 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4440 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4441 } 4442 4443 // install signal handlers for signals that HotSpot needs to 4444 // handle in order to support Java-level exception handling. 4445 4446 void os::Linux::install_signal_handlers() { 4447 if (!signal_handlers_are_installed) { 4448 signal_handlers_are_installed = true; 4449 4450 // signal-chaining 4451 typedef void (*signal_setting_t)(); 4452 signal_setting_t begin_signal_setting = NULL; 4453 signal_setting_t end_signal_setting = NULL; 4454 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4455 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4456 if (begin_signal_setting != NULL) { 4457 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4458 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4459 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4460 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4461 libjsig_is_loaded = true; 4462 assert(UseSignalChaining, "should enable signal-chaining"); 4463 } 4464 if (libjsig_is_loaded) { 4465 // Tell libjsig jvm is setting signal handlers 4466 (*begin_signal_setting)(); 4467 } 4468 4469 set_signal_handler(SIGSEGV, true); 4470 set_signal_handler(SIGPIPE, true); 4471 set_signal_handler(SIGBUS, true); 4472 set_signal_handler(SIGILL, true); 4473 set_signal_handler(SIGFPE, true); 4474 #if defined(PPC64) 4475 set_signal_handler(SIGTRAP, true); 4476 #endif 4477 set_signal_handler(SIGXFSZ, true); 4478 4479 if (libjsig_is_loaded) { 4480 // Tell libjsig jvm finishes setting signal handlers 4481 (*end_signal_setting)(); 4482 } 4483 4484 // We don't activate signal checker if libjsig is in place, we trust ourselves 4485 // and if UserSignalHandler is installed all bets are off. 4486 // Log that signal checking is off only if -verbose:jni is specified. 4487 if (CheckJNICalls) { 4488 if (libjsig_is_loaded) { 4489 if (PrintJNIResolving) { 4490 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4491 } 4492 check_signals = false; 4493 } 4494 if (AllowUserSignalHandlers) { 4495 if (PrintJNIResolving) { 4496 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4497 } 4498 check_signals = false; 4499 } 4500 } 4501 } 4502 } 4503 4504 // This is the fastest way to get thread cpu time on Linux. 4505 // Returns cpu time (user+sys) for any thread, not only for current. 4506 // POSIX compliant clocks are implemented in the kernels 2.6.16+. 4507 // It might work on 2.6.10+ with a special kernel/glibc patch. 4508 // For reference, please, see IEEE Std 1003.1-2004: 4509 // http://www.unix.org/single_unix_specification 4510 4511 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4512 struct timespec tp; 4513 int rc = os::Linux::clock_gettime(clockid, &tp); 4514 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4515 4516 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4517 } 4518 4519 void os::Linux::initialize_os_info() { 4520 assert(_os_version == 0, "OS info already initialized"); 4521 4522 struct utsname _uname; 4523 4524 uint32_t major; 4525 uint32_t minor; 4526 uint32_t fix; 4527 4528 int rc; 4529 4530 // Kernel version is unknown if 4531 // verification below fails. 4532 _os_version = 0x01000000; 4533 4534 rc = uname(&_uname); 4535 if (rc != -1) { 4536 4537 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix); 4538 if (rc == 3) { 4539 4540 if (major < 256 && minor < 256 && fix < 256) { 4541 // Kernel version format is as expected, 4542 // set it overriding unknown state. 4543 _os_version = (major << 16) | 4544 (minor << 8 ) | 4545 (fix << 0 ) ; 4546 } 4547 } 4548 } 4549 } 4550 4551 uint32_t os::Linux::os_version() { 4552 assert(_os_version != 0, "not initialized"); 4553 return _os_version & 0x00FFFFFF; 4554 } 4555 4556 bool os::Linux::os_version_is_known() { 4557 assert(_os_version != 0, "not initialized"); 4558 return _os_version & 0x01000000 ? false : true; 4559 } 4560 4561 ///// 4562 // glibc on Linux platform uses non-documented flag 4563 // to indicate, that some special sort of signal 4564 // trampoline is used. 4565 // We will never set this flag, and we should 4566 // ignore this flag in our diagnostic 4567 #ifdef SIGNIFICANT_SIGNAL_MASK 4568 #undef SIGNIFICANT_SIGNAL_MASK 4569 #endif 4570 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4571 4572 static const char* get_signal_handler_name(address handler, 4573 char* buf, int buflen) { 4574 int offset = 0; 4575 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4576 if (found) { 4577 // skip directory names 4578 const char *p1, *p2; 4579 p1 = buf; 4580 size_t len = strlen(os::file_separator()); 4581 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4582 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4583 } else { 4584 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4585 } 4586 return buf; 4587 } 4588 4589 static void print_signal_handler(outputStream* st, int sig, 4590 char* buf, size_t buflen) { 4591 struct sigaction sa; 4592 4593 sigaction(sig, NULL, &sa); 4594 4595 // See comment for SIGNIFICANT_SIGNAL_MASK define 4596 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4597 4598 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4599 4600 address handler = (sa.sa_flags & SA_SIGINFO) 4601 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4602 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4603 4604 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4605 st->print("SIG_DFL"); 4606 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4607 st->print("SIG_IGN"); 4608 } else { 4609 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4610 } 4611 4612 st->print(", sa_mask[0]="); 4613 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4614 4615 address rh = VMError::get_resetted_sighandler(sig); 4616 // May be, handler was resetted by VMError? 4617 if (rh != NULL) { 4618 handler = rh; 4619 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4620 } 4621 4622 st->print(", sa_flags="); 4623 os::Posix::print_sa_flags(st, sa.sa_flags); 4624 4625 // Check: is it our handler? 4626 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4627 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4628 // It is our signal handler 4629 // check for flags, reset system-used one! 4630 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4631 st->print( 4632 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4633 os::Linux::get_our_sigflags(sig)); 4634 } 4635 } 4636 st->cr(); 4637 } 4638 4639 4640 #define DO_SIGNAL_CHECK(sig) \ 4641 do { \ 4642 if (!sigismember(&check_signal_done, sig)) { \ 4643 os::Linux::check_signal_handler(sig); \ 4644 } \ 4645 } while (0) 4646 4647 // This method is a periodic task to check for misbehaving JNI applications 4648 // under CheckJNI, we can add any periodic checks here 4649 4650 void os::run_periodic_checks() { 4651 if (check_signals == false) return; 4652 4653 // SEGV and BUS if overridden could potentially prevent 4654 // generation of hs*.log in the event of a crash, debugging 4655 // such a case can be very challenging, so we absolutely 4656 // check the following for a good measure: 4657 DO_SIGNAL_CHECK(SIGSEGV); 4658 DO_SIGNAL_CHECK(SIGILL); 4659 DO_SIGNAL_CHECK(SIGFPE); 4660 DO_SIGNAL_CHECK(SIGBUS); 4661 DO_SIGNAL_CHECK(SIGPIPE); 4662 DO_SIGNAL_CHECK(SIGXFSZ); 4663 #if defined(PPC64) 4664 DO_SIGNAL_CHECK(SIGTRAP); 4665 #endif 4666 4667 // ReduceSignalUsage allows the user to override these handlers 4668 // see comments at the very top and jvm_solaris.h 4669 if (!ReduceSignalUsage) { 4670 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4671 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4672 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4673 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4674 } 4675 4676 DO_SIGNAL_CHECK(SR_signum); 4677 } 4678 4679 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4680 4681 static os_sigaction_t os_sigaction = NULL; 4682 4683 void os::Linux::check_signal_handler(int sig) { 4684 char buf[O_BUFLEN]; 4685 address jvmHandler = NULL; 4686 4687 4688 struct sigaction act; 4689 if (os_sigaction == NULL) { 4690 // only trust the default sigaction, in case it has been interposed 4691 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4692 if (os_sigaction == NULL) return; 4693 } 4694 4695 os_sigaction(sig, (struct sigaction*)NULL, &act); 4696 4697 4698 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4699 4700 address thisHandler = (act.sa_flags & SA_SIGINFO) 4701 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4702 : CAST_FROM_FN_PTR(address, act.sa_handler); 4703 4704 4705 switch (sig) { 4706 case SIGSEGV: 4707 case SIGBUS: 4708 case SIGFPE: 4709 case SIGPIPE: 4710 case SIGILL: 4711 case SIGXFSZ: 4712 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4713 break; 4714 4715 case SHUTDOWN1_SIGNAL: 4716 case SHUTDOWN2_SIGNAL: 4717 case SHUTDOWN3_SIGNAL: 4718 case BREAK_SIGNAL: 4719 jvmHandler = (address)user_handler(); 4720 break; 4721 4722 default: 4723 if (sig == SR_signum) { 4724 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4725 } else { 4726 return; 4727 } 4728 break; 4729 } 4730 4731 if (thisHandler != jvmHandler) { 4732 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4733 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4734 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4735 // No need to check this sig any longer 4736 sigaddset(&check_signal_done, sig); 4737 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4738 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4739 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4740 exception_name(sig, buf, O_BUFLEN)); 4741 } 4742 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4743 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4744 tty->print("expected:"); 4745 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig)); 4746 tty->cr(); 4747 tty->print(" found:"); 4748 os::Posix::print_sa_flags(tty, act.sa_flags); 4749 tty->cr(); 4750 // No need to check this sig any longer 4751 sigaddset(&check_signal_done, sig); 4752 } 4753 4754 // Dump all the signal 4755 if (sigismember(&check_signal_done, sig)) { 4756 print_signal_handlers(tty, buf, O_BUFLEN); 4757 } 4758 } 4759 4760 extern void report_error(char* file_name, int line_no, char* title, 4761 char* format, ...); 4762 4763 // this is called _before_ the most of global arguments have been parsed 4764 void os::init(void) { 4765 char dummy; // used to get a guess on initial stack address 4766 // first_hrtime = gethrtime(); 4767 4768 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4769 4770 init_random(1234567); 4771 4772 ThreadCritical::initialize(); 4773 4774 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4775 if (Linux::page_size() == -1) { 4776 fatal("os_linux.cpp: os::init: sysconf failed (%s)", 4777 os::strerror(errno)); 4778 } 4779 init_page_sizes((size_t) Linux::page_size()); 4780 4781 Linux::initialize_system_info(); 4782 4783 Linux::initialize_os_info(); 4784 4785 // main_thread points to the aboriginal thread 4786 Linux::_main_thread = pthread_self(); 4787 4788 Linux::clock_init(); 4789 initial_time_count = javaTimeNanos(); 4790 4791 // pthread_condattr initialization for monotonic clock 4792 int status; 4793 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4794 if ((status = pthread_condattr_init(_condattr)) != 0) { 4795 fatal("pthread_condattr_init: %s", os::strerror(status)); 4796 } 4797 // Only set the clock if CLOCK_MONOTONIC is available 4798 if (os::supports_monotonic_clock()) { 4799 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4800 if (status == EINVAL) { 4801 warning("Unable to use monotonic clock with relative timed-waits" \ 4802 " - changes to the time-of-day clock may have adverse affects"); 4803 } else { 4804 fatal("pthread_condattr_setclock: %s", os::strerror(status)); 4805 } 4806 } 4807 } 4808 // else it defaults to CLOCK_REALTIME 4809 4810 // retrieve entry point for pthread_setname_np 4811 Linux::_pthread_setname_np = 4812 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4813 4814 } 4815 4816 // To install functions for atexit system call 4817 extern "C" { 4818 static void perfMemory_exit_helper() { 4819 perfMemory_exit(); 4820 } 4821 } 4822 4823 // this is called _after_ the global arguments have been parsed 4824 jint os::init_2(void) { 4825 Linux::fast_thread_clock_init(); 4826 4827 // Allocate a single page and mark it as readable for safepoint polling 4828 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4829 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page"); 4830 4831 os::set_polling_page(polling_page); 4832 log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page)); 4833 4834 if (!UseMembar) { 4835 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4836 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4837 os::set_memory_serialize_page(mem_serialize_page); 4838 log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page)); 4839 } 4840 4841 // initialize suspend/resume support - must do this before signal_sets_init() 4842 if (SR_initialize() != 0) { 4843 perror("SR_initialize failed"); 4844 return JNI_ERR; 4845 } 4846 4847 Linux::signal_sets_init(); 4848 Linux::install_signal_handlers(); 4849 4850 // Check minimum allowable stack size for thread creation and to initialize 4851 // the java system classes, including StackOverflowError - depends on page 4852 // size. Add a page for compiler2 recursion in main thread. 4853 // Add in 2*BytesPerWord times page size to account for VM stack during 4854 // class initialization depending on 32 or 64 bit VM. 4855 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4856 JavaThread::stack_guard_zone_size() + 4857 JavaThread::stack_shadow_zone_size() + 4858 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4859 4860 size_t threadStackSizeInBytes = ThreadStackSize * K; 4861 if (threadStackSizeInBytes != 0 && 4862 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4863 tty->print_cr("\nThe stack size specified is too small, " 4864 "Specify at least " SIZE_FORMAT "k", 4865 os::Linux::min_stack_allowed/ K); 4866 return JNI_ERR; 4867 } 4868 4869 // Make the stack size a multiple of the page size so that 4870 // the yellow/red zones can be guarded. 4871 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4872 vm_page_size())); 4873 4874 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4875 4876 #if defined(IA32) 4877 workaround_expand_exec_shield_cs_limit(); 4878 #endif 4879 4880 Linux::libpthread_init(); 4881 log_info(os)("HotSpot is running with %s, %s", 4882 Linux::glibc_version(), Linux::libpthread_version()); 4883 4884 if (UseNUMA) { 4885 if (!Linux::libnuma_init()) { 4886 UseNUMA = false; 4887 } else { 4888 if ((Linux::numa_max_node() < 1)) { 4889 // There's only one node(they start from 0), disable NUMA. 4890 UseNUMA = false; 4891 } 4892 } 4893 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4894 // we can make the adaptive lgrp chunk resizing work. If the user specified 4895 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4896 // disable adaptive resizing. 4897 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4898 if (FLAG_IS_DEFAULT(UseNUMA)) { 4899 UseNUMA = false; 4900 } else { 4901 if (FLAG_IS_DEFAULT(UseLargePages) && 4902 FLAG_IS_DEFAULT(UseSHM) && 4903 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4904 UseLargePages = false; 4905 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) { 4906 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)"); 4907 UseAdaptiveSizePolicy = false; 4908 UseAdaptiveNUMAChunkSizing = false; 4909 } 4910 } 4911 } 4912 if (!UseNUMA && ForceNUMA) { 4913 UseNUMA = true; 4914 } 4915 } 4916 4917 if (MaxFDLimit) { 4918 // set the number of file descriptors to max. print out error 4919 // if getrlimit/setrlimit fails but continue regardless. 4920 struct rlimit nbr_files; 4921 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4922 if (status != 0) { 4923 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno)); 4924 } else { 4925 nbr_files.rlim_cur = nbr_files.rlim_max; 4926 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4927 if (status != 0) { 4928 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno)); 4929 } 4930 } 4931 } 4932 4933 // Initialize lock used to serialize thread creation (see os::create_thread) 4934 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4935 4936 // at-exit methods are called in the reverse order of their registration. 4937 // atexit functions are called on return from main or as a result of a 4938 // call to exit(3C). There can be only 32 of these functions registered 4939 // and atexit() does not set errno. 4940 4941 if (PerfAllowAtExitRegistration) { 4942 // only register atexit functions if PerfAllowAtExitRegistration is set. 4943 // atexit functions can be delayed until process exit time, which 4944 // can be problematic for embedded VM situations. Embedded VMs should 4945 // call DestroyJavaVM() to assure that VM resources are released. 4946 4947 // note: perfMemory_exit_helper atexit function may be removed in 4948 // the future if the appropriate cleanup code can be added to the 4949 // VM_Exit VMOperation's doit method. 4950 if (atexit(perfMemory_exit_helper) != 0) { 4951 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4952 } 4953 } 4954 4955 // initialize thread priority policy 4956 prio_init(); 4957 4958 return JNI_OK; 4959 } 4960 4961 // Mark the polling page as unreadable 4962 void os::make_polling_page_unreadable(void) { 4963 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 4964 fatal("Could not disable polling page"); 4965 } 4966 } 4967 4968 // Mark the polling page as readable 4969 void os::make_polling_page_readable(void) { 4970 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4971 fatal("Could not enable polling page"); 4972 } 4973 } 4974 4975 // older glibc versions don't have this macro (which expands to 4976 // an optimized bit-counting function) so we have to roll our own 4977 #ifndef CPU_COUNT 4978 4979 static int _cpu_count(const cpu_set_t* cpus) { 4980 int count = 0; 4981 // only look up to the number of configured processors 4982 for (int i = 0; i < os::processor_count(); i++) { 4983 if (CPU_ISSET(i, cpus)) { 4984 count++; 4985 } 4986 } 4987 return count; 4988 } 4989 4990 #define CPU_COUNT(cpus) _cpu_count(cpus) 4991 4992 #endif // CPU_COUNT 4993 4994 // Get the current number of available processors for this process. 4995 // This value can change at any time during a process's lifetime. 4996 // sched_getaffinity gives an accurate answer as it accounts for cpusets. 4997 // If it appears there may be more than 1024 processors then we do a 4998 // dynamic check - see 6515172 for details. 4999 // If anything goes wrong we fallback to returning the number of online 5000 // processors - which can be greater than the number available to the process. 5001 int os::active_processor_count() { 5002 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors 5003 cpu_set_t* cpus_p = &cpus; 5004 int cpus_size = sizeof(cpu_set_t); 5005 5006 int configured_cpus = processor_count(); // upper bound on available cpus 5007 int cpu_count = 0; 5008 5009 // old build platforms may not support dynamic cpu sets 5010 #ifdef CPU_ALLOC 5011 5012 // To enable easy testing of the dynamic path on different platforms we 5013 // introduce a diagnostic flag: UseCpuAllocPath 5014 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) { 5015 // kernel may use a mask bigger than cpu_set_t 5016 log_trace(os)("active_processor_count: using dynamic path %s" 5017 "- configured processors: %d", 5018 UseCpuAllocPath ? "(forced) " : "", 5019 configured_cpus); 5020 cpus_p = CPU_ALLOC(configured_cpus); 5021 if (cpus_p != NULL) { 5022 cpus_size = CPU_ALLOC_SIZE(configured_cpus); 5023 // zero it just to be safe 5024 CPU_ZERO_S(cpus_size, cpus_p); 5025 } 5026 else { 5027 // failed to allocate so fallback to online cpus 5028 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 5029 log_trace(os)("active_processor_count: " 5030 "CPU_ALLOC failed (%s) - using " 5031 "online processor count: %d", 5032 os::strerror(errno), online_cpus); 5033 return online_cpus; 5034 } 5035 } 5036 else { 5037 log_trace(os)("active_processor_count: using static path - configured processors: %d", 5038 configured_cpus); 5039 } 5040 #else // CPU_ALLOC 5041 // these stubs won't be executed 5042 #define CPU_COUNT_S(size, cpus) -1 5043 #define CPU_FREE(cpus) 5044 5045 log_trace(os)("active_processor_count: only static path available - configured processors: %d", 5046 configured_cpus); 5047 #endif // CPU_ALLOC 5048 5049 // pid 0 means the current thread - which we have to assume represents the process 5050 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) { 5051 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5052 cpu_count = CPU_COUNT_S(cpus_size, cpus_p); 5053 } 5054 else { 5055 cpu_count = CPU_COUNT(cpus_p); 5056 } 5057 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count); 5058 } 5059 else { 5060 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN); 5061 warning("sched_getaffinity failed (%s)- using online processor count (%d) " 5062 "which may exceed available processors", os::strerror(errno), cpu_count); 5063 } 5064 5065 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5066 CPU_FREE(cpus_p); 5067 } 5068 5069 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check"); 5070 return cpu_count; 5071 } 5072 5073 void os::set_native_thread_name(const char *name) { 5074 if (Linux::_pthread_setname_np) { 5075 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 5076 snprintf(buf, sizeof(buf), "%s", name); 5077 buf[sizeof(buf) - 1] = '\0'; 5078 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 5079 // ERANGE should not happen; all other errors should just be ignored. 5080 assert(rc != ERANGE, "pthread_setname_np failed"); 5081 } 5082 } 5083 5084 bool os::distribute_processes(uint length, uint* distribution) { 5085 // Not yet implemented. 5086 return false; 5087 } 5088 5089 bool os::bind_to_processor(uint processor_id) { 5090 // Not yet implemented. 5091 return false; 5092 } 5093 5094 /// 5095 5096 void os::SuspendedThreadTask::internal_do_task() { 5097 if (do_suspend(_thread->osthread())) { 5098 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 5099 do_task(context); 5100 do_resume(_thread->osthread()); 5101 } 5102 } 5103 5104 class PcFetcher : public os::SuspendedThreadTask { 5105 public: 5106 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 5107 ExtendedPC result(); 5108 protected: 5109 void do_task(const os::SuspendedThreadTaskContext& context); 5110 private: 5111 ExtendedPC _epc; 5112 }; 5113 5114 ExtendedPC PcFetcher::result() { 5115 guarantee(is_done(), "task is not done yet."); 5116 return _epc; 5117 } 5118 5119 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 5120 Thread* thread = context.thread(); 5121 OSThread* osthread = thread->osthread(); 5122 if (osthread->ucontext() != NULL) { 5123 _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext()); 5124 } else { 5125 // NULL context is unexpected, double-check this is the VMThread 5126 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 5127 } 5128 } 5129 5130 // Suspends the target using the signal mechanism and then grabs the PC before 5131 // resuming the target. Used by the flat-profiler only 5132 ExtendedPC os::get_thread_pc(Thread* thread) { 5133 // Make sure that it is called by the watcher for the VMThread 5134 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 5135 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 5136 5137 PcFetcher fetcher(thread); 5138 fetcher.run(); 5139 return fetcher.result(); 5140 } 5141 5142 //////////////////////////////////////////////////////////////////////////////// 5143 // debug support 5144 5145 bool os::find(address addr, outputStream* st) { 5146 Dl_info dlinfo; 5147 memset(&dlinfo, 0, sizeof(dlinfo)); 5148 if (dladdr(addr, &dlinfo) != 0) { 5149 st->print(PTR_FORMAT ": ", p2i(addr)); 5150 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5151 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname, 5152 p2i(addr) - p2i(dlinfo.dli_saddr)); 5153 } else if (dlinfo.dli_fbase != NULL) { 5154 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase)); 5155 } else { 5156 st->print("<absolute address>"); 5157 } 5158 if (dlinfo.dli_fname != NULL) { 5159 st->print(" in %s", dlinfo.dli_fname); 5160 } 5161 if (dlinfo.dli_fbase != NULL) { 5162 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase)); 5163 } 5164 st->cr(); 5165 5166 if (Verbose) { 5167 // decode some bytes around the PC 5168 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5169 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5170 address lowest = (address) dlinfo.dli_sname; 5171 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5172 if (begin < lowest) begin = lowest; 5173 Dl_info dlinfo2; 5174 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5175 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 5176 end = (address) dlinfo2.dli_saddr; 5177 } 5178 Disassembler::decode(begin, end, st); 5179 } 5180 return true; 5181 } 5182 return false; 5183 } 5184 5185 //////////////////////////////////////////////////////////////////////////////// 5186 // misc 5187 5188 // This does not do anything on Linux. This is basically a hook for being 5189 // able to use structured exception handling (thread-local exception filters) 5190 // on, e.g., Win32. 5191 void 5192 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method, 5193 JavaCallArguments* args, Thread* thread) { 5194 f(value, method, args, thread); 5195 } 5196 5197 void os::print_statistics() { 5198 } 5199 5200 bool os::message_box(const char* title, const char* message) { 5201 int i; 5202 fdStream err(defaultStream::error_fd()); 5203 for (i = 0; i < 78; i++) err.print_raw("="); 5204 err.cr(); 5205 err.print_raw_cr(title); 5206 for (i = 0; i < 78; i++) err.print_raw("-"); 5207 err.cr(); 5208 err.print_raw_cr(message); 5209 for (i = 0; i < 78; i++) err.print_raw("="); 5210 err.cr(); 5211 5212 char buf[16]; 5213 // Prevent process from exiting upon "read error" without consuming all CPU 5214 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5215 5216 return buf[0] == 'y' || buf[0] == 'Y'; 5217 } 5218 5219 int os::stat(const char *path, struct stat *sbuf) { 5220 char pathbuf[MAX_PATH]; 5221 if (strlen(path) > MAX_PATH - 1) { 5222 errno = ENAMETOOLONG; 5223 return -1; 5224 } 5225 os::native_path(strcpy(pathbuf, path)); 5226 return ::stat(pathbuf, sbuf); 5227 } 5228 5229 bool os::check_heap(bool force) { 5230 return true; 5231 } 5232 5233 // Is a (classpath) directory empty? 5234 bool os::dir_is_empty(const char* path) { 5235 DIR *dir = NULL; 5236 struct dirent *ptr; 5237 5238 dir = opendir(path); 5239 if (dir == NULL) return true; 5240 5241 // Scan the directory 5242 bool result = true; 5243 char buf[sizeof(struct dirent) + MAX_PATH]; 5244 while (result && (ptr = ::readdir(dir)) != NULL) { 5245 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5246 result = false; 5247 } 5248 } 5249 closedir(dir); 5250 return result; 5251 } 5252 5253 // This code originates from JDK's sysOpen and open64_w 5254 // from src/solaris/hpi/src/system_md.c 5255 5256 int os::open(const char *path, int oflag, int mode) { 5257 if (strlen(path) > MAX_PATH - 1) { 5258 errno = ENAMETOOLONG; 5259 return -1; 5260 } 5261 5262 // All file descriptors that are opened in the Java process and not 5263 // specifically destined for a subprocess should have the close-on-exec 5264 // flag set. If we don't set it, then careless 3rd party native code 5265 // might fork and exec without closing all appropriate file descriptors 5266 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5267 // turn might: 5268 // 5269 // - cause end-of-file to fail to be detected on some file 5270 // descriptors, resulting in mysterious hangs, or 5271 // 5272 // - might cause an fopen in the subprocess to fail on a system 5273 // suffering from bug 1085341. 5274 // 5275 // (Yes, the default setting of the close-on-exec flag is a Unix 5276 // design flaw) 5277 // 5278 // See: 5279 // 1085341: 32-bit stdio routines should support file descriptors >255 5280 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5281 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5282 // 5283 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5284 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5285 // because it saves a system call and removes a small window where the flag 5286 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5287 // and we fall back to using FD_CLOEXEC (see below). 5288 #ifdef O_CLOEXEC 5289 oflag |= O_CLOEXEC; 5290 #endif 5291 5292 int fd = ::open64(path, oflag, mode); 5293 if (fd == -1) return -1; 5294 5295 //If the open succeeded, the file might still be a directory 5296 { 5297 struct stat64 buf64; 5298 int ret = ::fstat64(fd, &buf64); 5299 int st_mode = buf64.st_mode; 5300 5301 if (ret != -1) { 5302 if ((st_mode & S_IFMT) == S_IFDIR) { 5303 errno = EISDIR; 5304 ::close(fd); 5305 return -1; 5306 } 5307 } else { 5308 ::close(fd); 5309 return -1; 5310 } 5311 } 5312 5313 #ifdef FD_CLOEXEC 5314 // Validate that the use of the O_CLOEXEC flag on open above worked. 5315 // With recent kernels, we will perform this check exactly once. 5316 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5317 if (!O_CLOEXEC_is_known_to_work) { 5318 int flags = ::fcntl(fd, F_GETFD); 5319 if (flags != -1) { 5320 if ((flags & FD_CLOEXEC) != 0) 5321 O_CLOEXEC_is_known_to_work = 1; 5322 else 5323 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5324 } 5325 } 5326 #endif 5327 5328 return fd; 5329 } 5330 5331 5332 // create binary file, rewriting existing file if required 5333 int os::create_binary_file(const char* path, bool rewrite_existing) { 5334 int oflags = O_WRONLY | O_CREAT; 5335 if (!rewrite_existing) { 5336 oflags |= O_EXCL; 5337 } 5338 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5339 } 5340 5341 // return current position of file pointer 5342 jlong os::current_file_offset(int fd) { 5343 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5344 } 5345 5346 // move file pointer to the specified offset 5347 jlong os::seek_to_file_offset(int fd, jlong offset) { 5348 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5349 } 5350 5351 // This code originates from JDK's sysAvailable 5352 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5353 5354 int os::available(int fd, jlong *bytes) { 5355 jlong cur, end; 5356 int mode; 5357 struct stat64 buf64; 5358 5359 if (::fstat64(fd, &buf64) >= 0) { 5360 mode = buf64.st_mode; 5361 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5362 int n; 5363 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5364 *bytes = n; 5365 return 1; 5366 } 5367 } 5368 } 5369 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5370 return 0; 5371 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5372 return 0; 5373 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5374 return 0; 5375 } 5376 *bytes = end - cur; 5377 return 1; 5378 } 5379 5380 // Map a block of memory. 5381 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5382 char *addr, size_t bytes, bool read_only, 5383 bool allow_exec) { 5384 int prot; 5385 int flags = MAP_PRIVATE; 5386 5387 if (read_only) { 5388 prot = PROT_READ; 5389 } else { 5390 prot = PROT_READ | PROT_WRITE; 5391 } 5392 5393 if (allow_exec) { 5394 prot |= PROT_EXEC; 5395 } 5396 5397 if (addr != NULL) { 5398 flags |= MAP_FIXED; 5399 } 5400 5401 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5402 fd, file_offset); 5403 if (mapped_address == MAP_FAILED) { 5404 return NULL; 5405 } 5406 return mapped_address; 5407 } 5408 5409 5410 // Remap a block of memory. 5411 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5412 char *addr, size_t bytes, bool read_only, 5413 bool allow_exec) { 5414 // same as map_memory() on this OS 5415 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5416 allow_exec); 5417 } 5418 5419 5420 // Unmap a block of memory. 5421 bool os::pd_unmap_memory(char* addr, size_t bytes) { 5422 return munmap(addr, bytes) == 0; 5423 } 5424 5425 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5426 5427 static clockid_t thread_cpu_clockid(Thread* thread) { 5428 pthread_t tid = thread->osthread()->pthread_id(); 5429 clockid_t clockid; 5430 5431 // Get thread clockid 5432 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5433 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5434 return clockid; 5435 } 5436 5437 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5438 // are used by JVM M&M and JVMTI to get user+sys or user CPU time 5439 // of a thread. 5440 // 5441 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5442 // the fast estimate available on the platform. 5443 5444 jlong os::current_thread_cpu_time() { 5445 if (os::Linux::supports_fast_thread_cpu_time()) { 5446 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5447 } else { 5448 // return user + sys since the cost is the same 5449 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5450 } 5451 } 5452 5453 jlong os::thread_cpu_time(Thread* thread) { 5454 // consistent with what current_thread_cpu_time() returns 5455 if (os::Linux::supports_fast_thread_cpu_time()) { 5456 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5457 } else { 5458 return slow_thread_cpu_time(thread, true /* user + sys */); 5459 } 5460 } 5461 5462 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5463 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5464 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5465 } else { 5466 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5467 } 5468 } 5469 5470 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5471 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5472 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5473 } else { 5474 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5475 } 5476 } 5477 5478 // -1 on error. 5479 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5480 pid_t tid = thread->osthread()->thread_id(); 5481 char *s; 5482 char stat[2048]; 5483 int statlen; 5484 char proc_name[64]; 5485 int count; 5486 long sys_time, user_time; 5487 char cdummy; 5488 int idummy; 5489 long ldummy; 5490 FILE *fp; 5491 5492 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5493 fp = fopen(proc_name, "r"); 5494 if (fp == NULL) return -1; 5495 statlen = fread(stat, 1, 2047, fp); 5496 stat[statlen] = '\0'; 5497 fclose(fp); 5498 5499 // Skip pid and the command string. Note that we could be dealing with 5500 // weird command names, e.g. user could decide to rename java launcher 5501 // to "java 1.4.2 :)", then the stat file would look like 5502 // 1234 (java 1.4.2 :)) R ... ... 5503 // We don't really need to know the command string, just find the last 5504 // occurrence of ")" and then start parsing from there. See bug 4726580. 5505 s = strrchr(stat, ')'); 5506 if (s == NULL) return -1; 5507 5508 // Skip blank chars 5509 do { s++; } while (s && isspace(*s)); 5510 5511 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5512 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5513 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5514 &user_time, &sys_time); 5515 if (count != 13) return -1; 5516 if (user_sys_cpu_time) { 5517 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5518 } else { 5519 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5520 } 5521 } 5522 5523 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5524 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5525 info_ptr->may_skip_backward = false; // elapsed time not wall time 5526 info_ptr->may_skip_forward = false; // elapsed time not wall time 5527 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5528 } 5529 5530 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5531 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5532 info_ptr->may_skip_backward = false; // elapsed time not wall time 5533 info_ptr->may_skip_forward = false; // elapsed time not wall time 5534 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5535 } 5536 5537 bool os::is_thread_cpu_time_supported() { 5538 return true; 5539 } 5540 5541 // System loadavg support. Returns -1 if load average cannot be obtained. 5542 // Linux doesn't yet have a (official) notion of processor sets, 5543 // so just return the system wide load average. 5544 int os::loadavg(double loadavg[], int nelem) { 5545 return ::getloadavg(loadavg, nelem); 5546 } 5547 5548 void os::pause() { 5549 char filename[MAX_PATH]; 5550 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5551 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile); 5552 } else { 5553 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5554 } 5555 5556 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5557 if (fd != -1) { 5558 struct stat buf; 5559 ::close(fd); 5560 while (::stat(filename, &buf) == 0) { 5561 (void)::poll(NULL, 0, 100); 5562 } 5563 } else { 5564 jio_fprintf(stderr, 5565 "Could not open pause file '%s', continuing immediately.\n", filename); 5566 } 5567 } 5568 5569 5570 // Refer to the comments in os_solaris.cpp park-unpark. The next two 5571 // comment paragraphs are worth repeating here: 5572 // 5573 // Assumption: 5574 // Only one parker can exist on an event, which is why we allocate 5575 // them per-thread. Multiple unparkers can coexist. 5576 // 5577 // _Event serves as a restricted-range semaphore. 5578 // -1 : thread is blocked, i.e. there is a waiter 5579 // 0 : neutral: thread is running or ready, 5580 // could have been signaled after a wait started 5581 // 1 : signaled - thread is running or ready 5582 // 5583 5584 // utility to compute the abstime argument to timedwait: 5585 // millis is the relative timeout time 5586 // abstime will be the absolute timeout time 5587 // TODO: replace compute_abstime() with unpackTime() 5588 5589 static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5590 if (millis < 0) millis = 0; 5591 5592 jlong seconds = millis / 1000; 5593 millis %= 1000; 5594 if (seconds > 50000000) { // see man cond_timedwait(3T) 5595 seconds = 50000000; 5596 } 5597 5598 if (os::supports_monotonic_clock()) { 5599 struct timespec now; 5600 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5601 assert_status(status == 0, status, "clock_gettime"); 5602 abstime->tv_sec = now.tv_sec + seconds; 5603 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5604 if (nanos >= NANOSECS_PER_SEC) { 5605 abstime->tv_sec += 1; 5606 nanos -= NANOSECS_PER_SEC; 5607 } 5608 abstime->tv_nsec = nanos; 5609 } else { 5610 struct timeval now; 5611 int status = gettimeofday(&now, NULL); 5612 assert(status == 0, "gettimeofday"); 5613 abstime->tv_sec = now.tv_sec + seconds; 5614 long usec = now.tv_usec + millis * 1000; 5615 if (usec >= 1000000) { 5616 abstime->tv_sec += 1; 5617 usec -= 1000000; 5618 } 5619 abstime->tv_nsec = usec * 1000; 5620 } 5621 return abstime; 5622 } 5623 5624 void os::PlatformEvent::park() { // AKA "down()" 5625 // Transitions for _Event: 5626 // -1 => -1 : illegal 5627 // 1 => 0 : pass - return immediately 5628 // 0 => -1 : block; then set _Event to 0 before returning 5629 5630 // Invariant: Only the thread associated with the Event/PlatformEvent 5631 // may call park(). 5632 // TODO: assert that _Assoc != NULL or _Assoc == Self 5633 assert(_nParked == 0, "invariant"); 5634 5635 int v; 5636 for (;;) { 5637 v = _Event; 5638 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5639 } 5640 guarantee(v >= 0, "invariant"); 5641 if (v == 0) { 5642 // Do this the hard way by blocking ... 5643 int status = pthread_mutex_lock(_mutex); 5644 assert_status(status == 0, status, "mutex_lock"); 5645 guarantee(_nParked == 0, "invariant"); 5646 ++_nParked; 5647 while (_Event < 0) { 5648 status = pthread_cond_wait(_cond, _mutex); 5649 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5650 // Treat this the same as if the wait was interrupted 5651 if (status == ETIME) { status = EINTR; } 5652 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5653 } 5654 --_nParked; 5655 5656 _Event = 0; 5657 status = pthread_mutex_unlock(_mutex); 5658 assert_status(status == 0, status, "mutex_unlock"); 5659 // Paranoia to ensure our locked and lock-free paths interact 5660 // correctly with each other. 5661 OrderAccess::fence(); 5662 } 5663 guarantee(_Event >= 0, "invariant"); 5664 } 5665 5666 int os::PlatformEvent::park(jlong millis) { 5667 // Transitions for _Event: 5668 // -1 => -1 : illegal 5669 // 1 => 0 : pass - return immediately 5670 // 0 => -1 : block; then set _Event to 0 before returning 5671 5672 guarantee(_nParked == 0, "invariant"); 5673 5674 int v; 5675 for (;;) { 5676 v = _Event; 5677 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5678 } 5679 guarantee(v >= 0, "invariant"); 5680 if (v != 0) return OS_OK; 5681 5682 // We do this the hard way, by blocking the thread. 5683 // Consider enforcing a minimum timeout value. 5684 struct timespec abst; 5685 compute_abstime(&abst, millis); 5686 5687 int ret = OS_TIMEOUT; 5688 int status = pthread_mutex_lock(_mutex); 5689 assert_status(status == 0, status, "mutex_lock"); 5690 guarantee(_nParked == 0, "invariant"); 5691 ++_nParked; 5692 5693 // Object.wait(timo) will return because of 5694 // (a) notification 5695 // (b) timeout 5696 // (c) thread.interrupt 5697 // 5698 // Thread.interrupt and object.notify{All} both call Event::set. 5699 // That is, we treat thread.interrupt as a special case of notification. 5700 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false. 5701 // We assume all ETIME returns are valid. 5702 // 5703 // TODO: properly differentiate simultaneous notify+interrupt. 5704 // In that case, we should propagate the notify to another waiter. 5705 5706 while (_Event < 0) { 5707 status = pthread_cond_timedwait(_cond, _mutex, &abst); 5708 assert_status(status == 0 || status == EINTR || 5709 status == ETIME || status == ETIMEDOUT, 5710 status, "cond_timedwait"); 5711 if (!FilterSpuriousWakeups) break; // previous semantics 5712 if (status == ETIME || status == ETIMEDOUT) break; 5713 // We consume and ignore EINTR and spurious wakeups. 5714 } 5715 --_nParked; 5716 if (_Event >= 0) { 5717 ret = OS_OK; 5718 } 5719 _Event = 0; 5720 status = pthread_mutex_unlock(_mutex); 5721 assert_status(status == 0, status, "mutex_unlock"); 5722 assert(_nParked == 0, "invariant"); 5723 // Paranoia to ensure our locked and lock-free paths interact 5724 // correctly with each other. 5725 OrderAccess::fence(); 5726 return ret; 5727 } 5728 5729 void os::PlatformEvent::unpark() { 5730 // Transitions for _Event: 5731 // 0 => 1 : just return 5732 // 1 => 1 : just return 5733 // -1 => either 0 or 1; must signal target thread 5734 // That is, we can safely transition _Event from -1 to either 5735 // 0 or 1. 5736 // See also: "Semaphores in Plan 9" by Mullender & Cox 5737 // 5738 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5739 // that it will take two back-to-back park() calls for the owning 5740 // thread to block. This has the benefit of forcing a spurious return 5741 // from the first park() call after an unpark() call which will help 5742 // shake out uses of park() and unpark() without condition variables. 5743 5744 if (Atomic::xchg(1, &_Event) >= 0) return; 5745 5746 // Wait for the thread associated with the event to vacate 5747 int status = pthread_mutex_lock(_mutex); 5748 assert_status(status == 0, status, "mutex_lock"); 5749 int AnyWaiters = _nParked; 5750 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5751 status = pthread_mutex_unlock(_mutex); 5752 assert_status(status == 0, status, "mutex_unlock"); 5753 if (AnyWaiters != 0) { 5754 // Note that we signal() *after* dropping the lock for "immortal" Events. 5755 // This is safe and avoids a common class of futile wakeups. In rare 5756 // circumstances this can cause a thread to return prematurely from 5757 // cond_{timed}wait() but the spurious wakeup is benign and the victim 5758 // will simply re-test the condition and re-park itself. 5759 // This provides particular benefit if the underlying platform does not 5760 // provide wait morphing. 5761 status = pthread_cond_signal(_cond); 5762 assert_status(status == 0, status, "cond_signal"); 5763 } 5764 } 5765 5766 5767 // JSR166 5768 // ------------------------------------------------------- 5769 5770 // The solaris and linux implementations of park/unpark are fairly 5771 // conservative for now, but can be improved. They currently use a 5772 // mutex/condvar pair, plus a a count. 5773 // Park decrements count if > 0, else does a condvar wait. Unpark 5774 // sets count to 1 and signals condvar. Only one thread ever waits 5775 // on the condvar. Contention seen when trying to park implies that someone 5776 // is unparking you, so don't wait. And spurious returns are fine, so there 5777 // is no need to track notifications. 5778 5779 // This code is common to linux and solaris and will be moved to a 5780 // common place in dolphin. 5781 // 5782 // The passed in time value is either a relative time in nanoseconds 5783 // or an absolute time in milliseconds. Either way it has to be unpacked 5784 // into suitable seconds and nanoseconds components and stored in the 5785 // given timespec structure. 5786 // Given time is a 64-bit value and the time_t used in the timespec is only 5787 // a signed-32-bit value (except on 64-bit Linux) we have to watch for 5788 // overflow if times way in the future are given. Further on Solaris versions 5789 // prior to 10 there is a restriction (see cond_timedwait) that the specified 5790 // number of seconds, in abstime, is less than current_time + 100,000,000. 5791 // As it will be 28 years before "now + 100000000" will overflow we can 5792 // ignore overflow and just impose a hard-limit on seconds using the value 5793 // of "now + 100,000,000". This places a limit on the timeout of about 3.17 5794 // years from "now". 5795 5796 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5797 assert(time > 0, "convertTime"); 5798 time_t max_secs = 0; 5799 5800 if (!os::supports_monotonic_clock() || isAbsolute) { 5801 struct timeval now; 5802 int status = gettimeofday(&now, NULL); 5803 assert(status == 0, "gettimeofday"); 5804 5805 max_secs = now.tv_sec + MAX_SECS; 5806 5807 if (isAbsolute) { 5808 jlong secs = time / 1000; 5809 if (secs > max_secs) { 5810 absTime->tv_sec = max_secs; 5811 } else { 5812 absTime->tv_sec = secs; 5813 } 5814 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5815 } else { 5816 jlong secs = time / NANOSECS_PER_SEC; 5817 if (secs >= MAX_SECS) { 5818 absTime->tv_sec = max_secs; 5819 absTime->tv_nsec = 0; 5820 } else { 5821 absTime->tv_sec = now.tv_sec + secs; 5822 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5823 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5824 absTime->tv_nsec -= NANOSECS_PER_SEC; 5825 ++absTime->tv_sec; // note: this must be <= max_secs 5826 } 5827 } 5828 } 5829 } else { 5830 // must be relative using monotonic clock 5831 struct timespec now; 5832 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5833 assert_status(status == 0, status, "clock_gettime"); 5834 max_secs = now.tv_sec + MAX_SECS; 5835 jlong secs = time / NANOSECS_PER_SEC; 5836 if (secs >= MAX_SECS) { 5837 absTime->tv_sec = max_secs; 5838 absTime->tv_nsec = 0; 5839 } else { 5840 absTime->tv_sec = now.tv_sec + secs; 5841 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5842 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5843 absTime->tv_nsec -= NANOSECS_PER_SEC; 5844 ++absTime->tv_sec; // note: this must be <= max_secs 5845 } 5846 } 5847 } 5848 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5849 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5850 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5851 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5852 } 5853 5854 void Parker::park(bool isAbsolute, jlong time) { 5855 // Ideally we'd do something useful while spinning, such 5856 // as calling unpackTime(). 5857 5858 // Optional fast-path check: 5859 // Return immediately if a permit is available. 5860 // We depend on Atomic::xchg() having full barrier semantics 5861 // since we are doing a lock-free update to _counter. 5862 if (Atomic::xchg(0, &_counter) > 0) return; 5863 5864 Thread* thread = Thread::current(); 5865 assert(thread->is_Java_thread(), "Must be JavaThread"); 5866 JavaThread *jt = (JavaThread *)thread; 5867 5868 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5869 // Check interrupt before trying to wait 5870 if (Thread::is_interrupted(thread, false)) { 5871 return; 5872 } 5873 5874 // Next, demultiplex/decode time arguments 5875 timespec absTime; 5876 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all 5877 return; 5878 } 5879 if (time > 0) { 5880 unpackTime(&absTime, isAbsolute, time); 5881 } 5882 5883 5884 // Enter safepoint region 5885 // Beware of deadlocks such as 6317397. 5886 // The per-thread Parker:: mutex is a classic leaf-lock. 5887 // In particular a thread must never block on the Threads_lock while 5888 // holding the Parker:: mutex. If safepoints are pending both the 5889 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5890 ThreadBlockInVM tbivm(jt); 5891 5892 // Don't wait if cannot get lock since interference arises from 5893 // unblocking. Also. check interrupt before trying wait 5894 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5895 return; 5896 } 5897 5898 int status; 5899 if (_counter > 0) { // no wait needed 5900 _counter = 0; 5901 status = pthread_mutex_unlock(_mutex); 5902 assert_status(status == 0, status, "invariant"); 5903 // Paranoia to ensure our locked and lock-free paths interact 5904 // correctly with each other and Java-level accesses. 5905 OrderAccess::fence(); 5906 return; 5907 } 5908 5909 #ifdef ASSERT 5910 // Don't catch signals while blocked; let the running threads have the signals. 5911 // (This allows a debugger to break into the running thread.) 5912 sigset_t oldsigs; 5913 sigemptyset(&oldsigs); 5914 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5915 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5916 #endif 5917 5918 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5919 jt->set_suspend_equivalent(); 5920 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5921 5922 assert(_cur_index == -1, "invariant"); 5923 if (time == 0) { 5924 _cur_index = REL_INDEX; // arbitrary choice when not timed 5925 status = pthread_cond_wait(&_cond[_cur_index], _mutex); 5926 } else { 5927 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5928 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime); 5929 } 5930 _cur_index = -1; 5931 assert_status(status == 0 || status == EINTR || 5932 status == ETIME || status == ETIMEDOUT, 5933 status, "cond_timedwait"); 5934 5935 #ifdef ASSERT 5936 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5937 #endif 5938 5939 _counter = 0; 5940 status = pthread_mutex_unlock(_mutex); 5941 assert_status(status == 0, status, "invariant"); 5942 // Paranoia to ensure our locked and lock-free paths interact 5943 // correctly with each other and Java-level accesses. 5944 OrderAccess::fence(); 5945 5946 // If externally suspended while waiting, re-suspend 5947 if (jt->handle_special_suspend_equivalent_condition()) { 5948 jt->java_suspend_self(); 5949 } 5950 } 5951 5952 void Parker::unpark() { 5953 int status = pthread_mutex_lock(_mutex); 5954 assert_status(status == 0, status, "invariant"); 5955 const int s = _counter; 5956 _counter = 1; 5957 // must capture correct index before unlocking 5958 int index = _cur_index; 5959 status = pthread_mutex_unlock(_mutex); 5960 assert_status(status == 0, status, "invariant"); 5961 if (s < 1 && index != -1) { 5962 // thread is definitely parked 5963 status = pthread_cond_signal(&_cond[index]); 5964 assert_status(status == 0, status, "invariant"); 5965 } 5966 } 5967 5968 5969 extern char** environ; 5970 5971 // Run the specified command in a separate process. Return its exit value, 5972 // or -1 on failure (e.g. can't fork a new process). 5973 // Unlike system(), this function can be called from signal handler. It 5974 // doesn't block SIGINT et al. 5975 int os::fork_and_exec(char* cmd) { 5976 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5977 5978 pid_t pid = fork(); 5979 5980 if (pid < 0) { 5981 // fork failed 5982 return -1; 5983 5984 } else if (pid == 0) { 5985 // child process 5986 5987 execve("/bin/sh", (char* const*)argv, environ); 5988 5989 // execve failed 5990 _exit(-1); 5991 5992 } else { 5993 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5994 // care about the actual exit code, for now. 5995 5996 int status; 5997 5998 // Wait for the child process to exit. This returns immediately if 5999 // the child has already exited. */ 6000 while (waitpid(pid, &status, 0) < 0) { 6001 switch (errno) { 6002 case ECHILD: return 0; 6003 case EINTR: break; 6004 default: return -1; 6005 } 6006 } 6007 6008 if (WIFEXITED(status)) { 6009 // The child exited normally; get its exit code. 6010 return WEXITSTATUS(status); 6011 } else if (WIFSIGNALED(status)) { 6012 // The child exited because of a signal 6013 // The best value to return is 0x80 + signal number, 6014 // because that is what all Unix shells do, and because 6015 // it allows callers to distinguish between process exit and 6016 // process death by signal. 6017 return 0x80 + WTERMSIG(status); 6018 } else { 6019 // Unknown exit code; pass it through 6020 return status; 6021 } 6022 } 6023 } 6024 6025 // is_headless_jre() 6026 // 6027 // Test for the existence of xawt/libmawt.so or libawt_xawt.so 6028 // in order to report if we are running in a headless jre 6029 // 6030 // Since JDK8 xawt/libmawt.so was moved into the same directory 6031 // as libawt.so, and renamed libawt_xawt.so 6032 // 6033 bool os::is_headless_jre() { 6034 struct stat statbuf; 6035 char buf[MAXPATHLEN]; 6036 char libmawtpath[MAXPATHLEN]; 6037 const char *xawtstr = "/xawt/libmawt.so"; 6038 const char *new_xawtstr = "/libawt_xawt.so"; 6039 char *p; 6040 6041 // Get path to libjvm.so 6042 os::jvm_path(buf, sizeof(buf)); 6043 6044 // Get rid of libjvm.so 6045 p = strrchr(buf, '/'); 6046 if (p == NULL) { 6047 return false; 6048 } else { 6049 *p = '\0'; 6050 } 6051 6052 // Get rid of client or server 6053 p = strrchr(buf, '/'); 6054 if (p == NULL) { 6055 return false; 6056 } else { 6057 *p = '\0'; 6058 } 6059 6060 // check xawt/libmawt.so 6061 strcpy(libmawtpath, buf); 6062 strcat(libmawtpath, xawtstr); 6063 if (::stat(libmawtpath, &statbuf) == 0) return false; 6064 6065 // check libawt_xawt.so 6066 strcpy(libmawtpath, buf); 6067 strcat(libmawtpath, new_xawtstr); 6068 if (::stat(libmawtpath, &statbuf) == 0) return false; 6069 6070 return true; 6071 } 6072 6073 // Get the default path to the core file 6074 // Returns the length of the string 6075 int os::get_core_path(char* buffer, size_t bufferSize) { 6076 /* 6077 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 6078 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 6079 */ 6080 const int core_pattern_len = 129; 6081 char core_pattern[core_pattern_len] = {0}; 6082 6083 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 6084 if (core_pattern_file == -1) { 6085 return -1; 6086 } 6087 6088 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 6089 ::close(core_pattern_file); 6090 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') { 6091 return -1; 6092 } 6093 if (core_pattern[ret-1] == '\n') { 6094 core_pattern[ret-1] = '\0'; 6095 } else { 6096 core_pattern[ret] = '\0'; 6097 } 6098 6099 char *pid_pos = strstr(core_pattern, "%p"); 6100 int written; 6101 6102 if (core_pattern[0] == '/') { 6103 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern); 6104 } else { 6105 char cwd[PATH_MAX]; 6106 6107 const char* p = get_current_directory(cwd, PATH_MAX); 6108 if (p == NULL) { 6109 return -1; 6110 } 6111 6112 if (core_pattern[0] == '|') { 6113 written = jio_snprintf(buffer, bufferSize, 6114 "\"%s\" (or dumping to %s/core.%d)", 6115 &core_pattern[1], p, current_process_id()); 6116 } else { 6117 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 6118 } 6119 } 6120 6121 if (written < 0) { 6122 return -1; 6123 } 6124 6125 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 6126 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 6127 6128 if (core_uses_pid_file != -1) { 6129 char core_uses_pid = 0; 6130 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 6131 ::close(core_uses_pid_file); 6132 6133 if (core_uses_pid == '1') { 6134 jio_snprintf(buffer + written, bufferSize - written, 6135 ".%d", current_process_id()); 6136 } 6137 } 6138 } 6139 6140 return strlen(buffer); 6141 } 6142 6143 bool os::start_debugging(char *buf, int buflen) { 6144 int len = (int)strlen(buf); 6145 char *p = &buf[len]; 6146 6147 jio_snprintf(p, buflen-len, 6148 "\n\n" 6149 "Do you want to debug the problem?\n\n" 6150 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n" 6151 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n" 6152 "Otherwise, press RETURN to abort...", 6153 os::current_process_id(), os::current_process_id(), 6154 os::current_thread_id(), os::current_thread_id()); 6155 6156 bool yes = os::message_box("Unexpected Error", buf); 6157 6158 if (yes) { 6159 // yes, user asked VM to launch debugger 6160 jio_snprintf(buf, sizeof(buf), "gdb /proc/%d/exe %d", 6161 os::current_process_id(), os::current_process_id()); 6162 6163 os::fork_and_exec(buf); 6164 yes = false; 6165 } 6166 return yes; 6167 } 6168 6169 static inline struct timespec get_mtime(const char* filename) { 6170 struct stat st; 6171 int ret = os::stat(filename, &st); 6172 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno)); 6173 return st.st_mtim; 6174 } 6175 6176 int os::compare_file_modified_times(const char* file1, const char* file2) { 6177 struct timespec filetime1 = get_mtime(file1); 6178 struct timespec filetime2 = get_mtime(file2); 6179 int diff = filetime1.tv_sec - filetime2.tv_sec; 6180 if (diff == 0) { 6181 return filetime1.tv_nsec - filetime2.tv_nsec; 6182 } 6183 return diff; 6184 } 6185 6186 /////////////// Unit tests /////////////// 6187 6188 #ifndef PRODUCT 6189 6190 #define test_log(...) \ 6191 do { \ 6192 if (VerboseInternalVMTests) { \ 6193 tty->print_cr(__VA_ARGS__); \ 6194 tty->flush(); \ 6195 } \ 6196 } while (false) 6197 6198 class TestReserveMemorySpecial : AllStatic { 6199 public: 6200 static void small_page_write(void* addr, size_t size) { 6201 size_t page_size = os::vm_page_size(); 6202 6203 char* end = (char*)addr + size; 6204 for (char* p = (char*)addr; p < end; p += page_size) { 6205 *p = 1; 6206 } 6207 } 6208 6209 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6210 if (!UseHugeTLBFS) { 6211 return; 6212 } 6213 6214 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6215 6216 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6217 6218 if (addr != NULL) { 6219 small_page_write(addr, size); 6220 6221 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6222 } 6223 } 6224 6225 static void test_reserve_memory_special_huge_tlbfs_only() { 6226 if (!UseHugeTLBFS) { 6227 return; 6228 } 6229 6230 size_t lp = os::large_page_size(); 6231 6232 for (size_t size = lp; size <= lp * 10; size += lp) { 6233 test_reserve_memory_special_huge_tlbfs_only(size); 6234 } 6235 } 6236 6237 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6238 size_t lp = os::large_page_size(); 6239 size_t ag = os::vm_allocation_granularity(); 6240 6241 // sizes to test 6242 const size_t sizes[] = { 6243 lp, lp + ag, lp + lp / 2, lp * 2, 6244 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2, 6245 lp * 10, lp * 10 + lp / 2 6246 }; 6247 const int num_sizes = sizeof(sizes) / sizeof(size_t); 6248 6249 // For each size/alignment combination, we test three scenarios: 6250 // 1) with req_addr == NULL 6251 // 2) with a non-null req_addr at which we expect to successfully allocate 6252 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we 6253 // expect the allocation to either fail or to ignore req_addr 6254 6255 // Pre-allocate two areas; they shall be as large as the largest allocation 6256 // and aligned to the largest alignment we will be testing. 6257 const size_t mapping_size = sizes[num_sizes - 1] * 2; 6258 char* const mapping1 = (char*) ::mmap(NULL, mapping_size, 6259 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6260 -1, 0); 6261 assert(mapping1 != MAP_FAILED, "should work"); 6262 6263 char* const mapping2 = (char*) ::mmap(NULL, mapping_size, 6264 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6265 -1, 0); 6266 assert(mapping2 != MAP_FAILED, "should work"); 6267 6268 // Unmap the first mapping, but leave the second mapping intact: the first 6269 // mapping will serve as a value for a "good" req_addr (case 2). The second 6270 // mapping, still intact, as "bad" req_addr (case 3). 6271 ::munmap(mapping1, mapping_size); 6272 6273 // Case 1 6274 test_log("%s, req_addr NULL:", __FUNCTION__); 6275 test_log("size align result"); 6276 6277 for (int i = 0; i < num_sizes; i++) { 6278 const size_t size = sizes[i]; 6279 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6280 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6281 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s", 6282 size, alignment, p2i(p), (p != NULL ? "" : "(failed)")); 6283 if (p != NULL) { 6284 assert(is_ptr_aligned(p, alignment), "must be"); 6285 small_page_write(p, size); 6286 os::Linux::release_memory_special_huge_tlbfs(p, size); 6287 } 6288 } 6289 } 6290 6291 // Case 2 6292 test_log("%s, req_addr non-NULL:", __FUNCTION__); 6293 test_log("size align req_addr result"); 6294 6295 for (int i = 0; i < num_sizes; i++) { 6296 const size_t size = sizes[i]; 6297 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6298 char* const req_addr = (char*) align_ptr_up(mapping1, alignment); 6299 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6300 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6301 size, alignment, p2i(req_addr), p2i(p), 6302 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)"))); 6303 if (p != NULL) { 6304 assert(p == req_addr, "must be"); 6305 small_page_write(p, size); 6306 os::Linux::release_memory_special_huge_tlbfs(p, size); 6307 } 6308 } 6309 } 6310 6311 // Case 3 6312 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__); 6313 test_log("size align req_addr result"); 6314 6315 for (int i = 0; i < num_sizes; i++) { 6316 const size_t size = sizes[i]; 6317 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6318 char* const req_addr = (char*) align_ptr_up(mapping2, alignment); 6319 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6320 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6321 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)"))); 6322 // as the area around req_addr contains already existing mappings, the API should always 6323 // return NULL (as per contract, it cannot return another address) 6324 assert(p == NULL, "must be"); 6325 } 6326 } 6327 6328 ::munmap(mapping2, mapping_size); 6329 6330 } 6331 6332 static void test_reserve_memory_special_huge_tlbfs() { 6333 if (!UseHugeTLBFS) { 6334 return; 6335 } 6336 6337 test_reserve_memory_special_huge_tlbfs_only(); 6338 test_reserve_memory_special_huge_tlbfs_mixed(); 6339 } 6340 6341 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6342 if (!UseSHM) { 6343 return; 6344 } 6345 6346 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6347 6348 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6349 6350 if (addr != NULL) { 6351 assert(is_ptr_aligned(addr, alignment), "Check"); 6352 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6353 6354 small_page_write(addr, size); 6355 6356 os::Linux::release_memory_special_shm(addr, size); 6357 } 6358 } 6359 6360 static void test_reserve_memory_special_shm() { 6361 size_t lp = os::large_page_size(); 6362 size_t ag = os::vm_allocation_granularity(); 6363 6364 for (size_t size = ag; size < lp * 3; size += ag) { 6365 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6366 test_reserve_memory_special_shm(size, alignment); 6367 } 6368 } 6369 } 6370 6371 static void test() { 6372 test_reserve_memory_special_huge_tlbfs(); 6373 test_reserve_memory_special_shm(); 6374 } 6375 }; 6376 6377 void TestReserveMemorySpecial_test() { 6378 TestReserveMemorySpecial::test(); 6379 } 6380 6381 #endif