1 /* 2 * Copyright (c) 1999, 2017, 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 "oops/oop.inline.hpp" 39 #include "os_linux.inline.hpp" 40 #include "os_share_linux.hpp" 41 #include "prims/jniFastGetField.hpp" 42 #include "prims/jvm.h" 43 #include "prims/jvm_misc.hpp" 44 #include "runtime/arguments.hpp" 45 #include "runtime/atomic.hpp" 46 #include "runtime/extendedPC.hpp" 47 #include "runtime/globals.hpp" 48 #include "runtime/interfaceSupport.hpp" 49 #include "runtime/init.hpp" 50 #include "runtime/java.hpp" 51 #include "runtime/javaCalls.hpp" 52 #include "runtime/mutexLocker.hpp" 53 #include "runtime/objectMonitor.hpp" 54 #include "runtime/orderAccess.inline.hpp" 55 #include "runtime/osThread.hpp" 56 #include "runtime/perfMemory.hpp" 57 #include "runtime/sharedRuntime.hpp" 58 #include "runtime/statSampler.hpp" 59 #include "runtime/stubRoutines.hpp" 60 #include "runtime/thread.inline.hpp" 61 #include "runtime/threadCritical.hpp" 62 #include "runtime/threadSMR.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/align.hpp" 69 #include "utilities/decoder.hpp" 70 #include "utilities/defaultStream.hpp" 71 #include "utilities/events.hpp" 72 #include "utilities/elfFile.hpp" 73 #include "utilities/growableArray.hpp" 74 #include "utilities/macros.hpp" 75 #include "utilities/vmError.hpp" 76 77 // put OS-includes here 78 # include <sys/types.h> 79 # include <sys/mman.h> 80 # include <sys/stat.h> 81 # include <sys/select.h> 82 # include <pthread.h> 83 # include <signal.h> 84 # include <errno.h> 85 # include <dlfcn.h> 86 # include <stdio.h> 87 # include <unistd.h> 88 # include <sys/resource.h> 89 # include <pthread.h> 90 # include <sys/stat.h> 91 # include <sys/time.h> 92 # include <sys/times.h> 93 # include <sys/utsname.h> 94 # include <sys/socket.h> 95 # include <sys/wait.h> 96 # include <pwd.h> 97 # include <poll.h> 98 # include <semaphore.h> 99 # include <fcntl.h> 100 # include <string.h> 101 # include <syscall.h> 102 # include <sys/sysinfo.h> 103 # include <gnu/libc-version.h> 104 # include <sys/ipc.h> 105 # include <sys/shm.h> 106 # include <link.h> 107 # include <stdint.h> 108 # include <inttypes.h> 109 # include <sys/ioctl.h> 110 111 #ifndef _GNU_SOURCE 112 #define _GNU_SOURCE 113 #include <sched.h> 114 #undef _GNU_SOURCE 115 #else 116 #include <sched.h> 117 #endif 118 119 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 120 // getrusage() is prepared to handle the associated failure. 121 #ifndef RUSAGE_THREAD 122 #define RUSAGE_THREAD (1) /* only the calling thread */ 123 #endif 124 125 #define MAX_PATH (2 * K) 126 127 #define MAX_SECS 100000000 128 129 // for timer info max values which include all bits 130 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 131 132 #define LARGEPAGES_BIT (1 << 6) 133 //////////////////////////////////////////////////////////////////////////////// 134 // global variables 135 julong os::Linux::_physical_memory = 0; 136 137 address os::Linux::_initial_thread_stack_bottom = NULL; 138 uintptr_t os::Linux::_initial_thread_stack_size = 0; 139 140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 142 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL; 143 Mutex* os::Linux::_createThread_lock = NULL; 144 pthread_t os::Linux::_main_thread; 145 int os::Linux::_page_size = -1; 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; 150 151 static jlong initial_time_count=0; 152 153 static int clock_tics_per_sec = 100; 154 155 // For diagnostics to print a message once. see run_periodic_checks 156 static sigset_t check_signal_done; 157 static bool check_signals = true; 158 159 // Signal number used to suspend/resume a thread 160 161 // do not use any signal number less than SIGSEGV, see 4355769 162 static int SR_signum = SIGUSR2; 163 sigset_t SR_sigset; 164 165 // utility functions 166 167 static int SR_initialize(); 168 169 julong os::available_memory() { 170 return Linux::available_memory(); 171 } 172 173 julong os::Linux::available_memory() { 174 // values in struct sysinfo are "unsigned long" 175 struct sysinfo si; 176 sysinfo(&si); 177 178 return (julong)si.freeram * si.mem_unit; 179 } 180 181 julong os::physical_memory() { 182 return Linux::physical_memory(); 183 } 184 185 // Return true if user is running as root. 186 187 bool os::have_special_privileges() { 188 static bool init = false; 189 static bool privileges = false; 190 if (!init) { 191 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 192 init = true; 193 } 194 return privileges; 195 } 196 197 198 #ifndef SYS_gettid 199 // i386: 224, ia64: 1105, amd64: 186, sparc 143 200 #ifdef __ia64__ 201 #define SYS_gettid 1105 202 #else 203 #ifdef __i386__ 204 #define SYS_gettid 224 205 #else 206 #ifdef __amd64__ 207 #define SYS_gettid 186 208 #else 209 #ifdef __sparc__ 210 #define SYS_gettid 143 211 #else 212 #error define gettid for the arch 213 #endif 214 #endif 215 #endif 216 #endif 217 #endif 218 219 220 // pid_t gettid() 221 // 222 // Returns the kernel thread id of the currently running thread. Kernel 223 // thread id is used to access /proc. 224 pid_t os::Linux::gettid() { 225 int rslt = syscall(SYS_gettid); 226 assert(rslt != -1, "must be."); // old linuxthreads implementation? 227 return (pid_t)rslt; 228 } 229 230 // Most versions of linux have a bug where the number of processors are 231 // determined by looking at the /proc file system. In a chroot environment, 232 // the system call returns 1. This causes the VM to act as if it is 233 // a single processor and elide locking (see is_MP() call). 234 static bool unsafe_chroot_detected = false; 235 static const char *unstable_chroot_error = "/proc file system not found.\n" 236 "Java may be unstable running multithreaded in a chroot " 237 "environment on Linux when /proc filesystem is not mounted."; 238 239 void os::Linux::initialize_system_info() { 240 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 241 if (processor_count() == 1) { 242 pid_t pid = os::Linux::gettid(); 243 char fname[32]; 244 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 245 FILE *fp = fopen(fname, "r"); 246 if (fp == NULL) { 247 unsafe_chroot_detected = true; 248 } else { 249 fclose(fp); 250 } 251 } 252 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 253 assert(processor_count() > 0, "linux error"); 254 } 255 256 void os::init_system_properties_values() { 257 // The next steps are taken in the product version: 258 // 259 // Obtain the JAVA_HOME value from the location of libjvm.so. 260 // This library should be located at: 261 // <JAVA_HOME>/lib/{client|server}/libjvm.so. 262 // 263 // If "/jre/lib/" appears at the right place in the path, then we 264 // assume libjvm.so is installed in a JDK and we use this path. 265 // 266 // Otherwise exit with message: "Could not create the Java virtual machine." 267 // 268 // The following extra steps are taken in the debugging version: 269 // 270 // If "/jre/lib/" does NOT appear at the right place in the path 271 // instead of exit check for $JAVA_HOME environment variable. 272 // 273 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 274 // then we append a fake suffix "hotspot/libjvm.so" to this path so 275 // it looks like libjvm.so is installed there 276 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 277 // 278 // Otherwise exit. 279 // 280 // Important note: if the location of libjvm.so changes this 281 // code needs to be changed accordingly. 282 283 // See ld(1): 284 // The linker uses the following search paths to locate required 285 // shared libraries: 286 // 1: ... 287 // ... 288 // 7: The default directories, normally /lib and /usr/lib. 289 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390) 290 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 291 #else 292 #define DEFAULT_LIBPATH "/lib:/usr/lib" 293 #endif 294 295 // Base path of extensions installed on the system. 296 #define SYS_EXT_DIR "/usr/java/packages" 297 #define EXTENSIONS_DIR "/lib/ext" 298 299 // Buffer that fits several sprintfs. 300 // Note that the space for the colon and the trailing null are provided 301 // by the nulls included by the sizeof operator. 302 const size_t bufsize = 303 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 304 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 305 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 306 307 // sysclasspath, java_home, dll_dir 308 { 309 char *pslash; 310 os::jvm_path(buf, bufsize); 311 312 // Found the full path to libjvm.so. 313 // Now cut the path to <java_home>/jre if we can. 314 pslash = strrchr(buf, '/'); 315 if (pslash != NULL) { 316 *pslash = '\0'; // Get rid of /libjvm.so. 317 } 318 pslash = strrchr(buf, '/'); 319 if (pslash != NULL) { 320 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 321 } 322 Arguments::set_dll_dir(buf); 323 324 if (pslash != NULL) { 325 pslash = strrchr(buf, '/'); 326 if (pslash != NULL) { 327 *pslash = '\0'; // Get rid of /lib. 328 } 329 } 330 Arguments::set_java_home(buf); 331 set_boot_path('/', ':'); 332 } 333 334 // Where to look for native libraries. 335 // 336 // Note: Due to a legacy implementation, most of the library path 337 // is set in the launcher. This was to accomodate linking restrictions 338 // on legacy Linux implementations (which are no longer supported). 339 // Eventually, all the library path setting will be done here. 340 // 341 // However, to prevent the proliferation of improperly built native 342 // libraries, the new path component /usr/java/packages is added here. 343 // Eventually, all the library path setting will be done here. 344 { 345 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 346 // should always exist (until the legacy problem cited above is 347 // addressed). 348 const char *v = ::getenv("LD_LIBRARY_PATH"); 349 const char *v_colon = ":"; 350 if (v == NULL) { v = ""; v_colon = ""; } 351 // That's +1 for the colon and +1 for the trailing '\0'. 352 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 353 strlen(v) + 1 + 354 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1, 355 mtInternal); 356 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon); 357 Arguments::set_library_path(ld_library_path); 358 FREE_C_HEAP_ARRAY(char, ld_library_path); 359 } 360 361 // Extensions directories. 362 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 363 Arguments::set_ext_dirs(buf); 364 365 FREE_C_HEAP_ARRAY(char, buf); 366 367 #undef DEFAULT_LIBPATH 368 #undef SYS_EXT_DIR 369 #undef EXTENSIONS_DIR 370 } 371 372 //////////////////////////////////////////////////////////////////////////////// 373 // breakpoint support 374 375 void os::breakpoint() { 376 BREAKPOINT; 377 } 378 379 extern "C" void breakpoint() { 380 // use debugger to set breakpoint here 381 } 382 383 //////////////////////////////////////////////////////////////////////////////// 384 // signal support 385 386 debug_only(static bool signal_sets_initialized = false); 387 static sigset_t unblocked_sigs, vm_sigs; 388 389 bool os::Linux::is_sig_ignored(int sig) { 390 struct sigaction oact; 391 sigaction(sig, (struct sigaction*)NULL, &oact); 392 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 393 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 394 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) { 395 return true; 396 } else { 397 return false; 398 } 399 } 400 401 void os::Linux::signal_sets_init() { 402 // Should also have an assertion stating we are still single-threaded. 403 assert(!signal_sets_initialized, "Already initialized"); 404 // Fill in signals that are necessarily unblocked for all threads in 405 // the VM. Currently, we unblock the following signals: 406 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 407 // by -Xrs (=ReduceSignalUsage)); 408 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 409 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 410 // the dispositions or masks wrt these signals. 411 // Programs embedding the VM that want to use the above signals for their 412 // own purposes must, at this time, use the "-Xrs" option to prevent 413 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 414 // (See bug 4345157, and other related bugs). 415 // In reality, though, unblocking these signals is really a nop, since 416 // these signals are not blocked by default. 417 sigemptyset(&unblocked_sigs); 418 sigaddset(&unblocked_sigs, SIGILL); 419 sigaddset(&unblocked_sigs, SIGSEGV); 420 sigaddset(&unblocked_sigs, SIGBUS); 421 sigaddset(&unblocked_sigs, SIGFPE); 422 #if defined(PPC64) 423 sigaddset(&unblocked_sigs, SIGTRAP); 424 #endif 425 sigaddset(&unblocked_sigs, SR_signum); 426 427 if (!ReduceSignalUsage) { 428 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 429 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 430 } 431 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 432 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 433 } 434 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 435 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 436 } 437 } 438 // Fill in signals that are blocked by all but the VM thread. 439 sigemptyset(&vm_sigs); 440 if (!ReduceSignalUsage) { 441 sigaddset(&vm_sigs, BREAK_SIGNAL); 442 } 443 debug_only(signal_sets_initialized = true); 444 445 } 446 447 // These are signals that are unblocked while a thread is running Java. 448 // (For some reason, they get blocked by default.) 449 sigset_t* os::Linux::unblocked_signals() { 450 assert(signal_sets_initialized, "Not initialized"); 451 return &unblocked_sigs; 452 } 453 454 // These are the signals that are blocked while a (non-VM) thread is 455 // running Java. Only the VM thread handles these signals. 456 sigset_t* os::Linux::vm_signals() { 457 assert(signal_sets_initialized, "Not initialized"); 458 return &vm_sigs; 459 } 460 461 void os::Linux::hotspot_sigmask(Thread* thread) { 462 463 //Save caller's signal mask before setting VM signal mask 464 sigset_t caller_sigmask; 465 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 466 467 OSThread* osthread = thread->osthread(); 468 osthread->set_caller_sigmask(caller_sigmask); 469 470 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 471 472 if (!ReduceSignalUsage) { 473 if (thread->is_VM_thread()) { 474 // Only the VM thread handles BREAK_SIGNAL ... 475 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 476 } else { 477 // ... all other threads block BREAK_SIGNAL 478 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 479 } 480 } 481 } 482 483 ////////////////////////////////////////////////////////////////////////////// 484 // detecting pthread library 485 486 void os::Linux::libpthread_init() { 487 // Save glibc and pthread version strings. 488 #if !defined(_CS_GNU_LIBC_VERSION) || \ 489 !defined(_CS_GNU_LIBPTHREAD_VERSION) 490 #error "glibc too old (< 2.3.2)" 491 #endif 492 493 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 494 assert(n > 0, "cannot retrieve glibc version"); 495 char *str = (char *)malloc(n, mtInternal); 496 confstr(_CS_GNU_LIBC_VERSION, str, n); 497 os::Linux::set_glibc_version(str); 498 499 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 500 assert(n > 0, "cannot retrieve pthread version"); 501 str = (char *)malloc(n, mtInternal); 502 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 503 os::Linux::set_libpthread_version(str); 504 } 505 506 ///////////////////////////////////////////////////////////////////////////// 507 // thread stack expansion 508 509 // os::Linux::manually_expand_stack() takes care of expanding the thread 510 // stack. Note that this is normally not needed: pthread stacks allocate 511 // thread stack using mmap() without MAP_NORESERVE, so the stack is already 512 // committed. Therefore it is not necessary to expand the stack manually. 513 // 514 // Manually expanding the stack was historically needed on LinuxThreads 515 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays 516 // it is kept to deal with very rare corner cases: 517 // 518 // For one, user may run the VM on an own implementation of threads 519 // whose stacks are - like the old LinuxThreads - implemented using 520 // mmap(MAP_GROWSDOWN). 521 // 522 // Also, this coding may be needed if the VM is running on the primordial 523 // thread. Normally we avoid running on the primordial thread; however, 524 // user may still invoke the VM on the primordial thread. 525 // 526 // The following historical comment describes the details about running 527 // on a thread stack allocated with mmap(MAP_GROWSDOWN): 528 529 530 // Force Linux kernel to expand current thread stack. If "bottom" is close 531 // to the stack guard, caller should block all signals. 532 // 533 // MAP_GROWSDOWN: 534 // A special mmap() flag that is used to implement thread stacks. It tells 535 // kernel that the memory region should extend downwards when needed. This 536 // allows early versions of LinuxThreads to only mmap the first few pages 537 // when creating a new thread. Linux kernel will automatically expand thread 538 // stack as needed (on page faults). 539 // 540 // However, because the memory region of a MAP_GROWSDOWN stack can grow on 541 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 542 // region, it's hard to tell if the fault is due to a legitimate stack 543 // access or because of reading/writing non-exist memory (e.g. buffer 544 // overrun). As a rule, if the fault happens below current stack pointer, 545 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the 546 // application (see Linux kernel fault.c). 547 // 548 // This Linux feature can cause SIGSEGV when VM bangs thread stack for 549 // stack overflow detection. 550 // 551 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 552 // not use MAP_GROWSDOWN. 553 // 554 // To get around the problem and allow stack banging on Linux, we need to 555 // manually expand thread stack after receiving the SIGSEGV. 556 // 557 // There are two ways to expand thread stack to address "bottom", we used 558 // both of them in JVM before 1.5: 559 // 1. adjust stack pointer first so that it is below "bottom", and then 560 // touch "bottom" 561 // 2. mmap() the page in question 562 // 563 // Now alternate signal stack is gone, it's harder to use 2. For instance, 564 // if current sp is already near the lower end of page 101, and we need to 565 // call mmap() to map page 100, it is possible that part of the mmap() frame 566 // will be placed in page 100. When page 100 is mapped, it is zero-filled. 567 // That will destroy the mmap() frame and cause VM to crash. 568 // 569 // The following code works by adjusting sp first, then accessing the "bottom" 570 // page to force a page fault. Linux kernel will then automatically expand the 571 // stack mapping. 572 // 573 // _expand_stack_to() assumes its frame size is less than page size, which 574 // should always be true if the function is not inlined. 575 576 static void NOINLINE _expand_stack_to(address bottom) { 577 address sp; 578 size_t size; 579 volatile char *p; 580 581 // Adjust bottom to point to the largest address within the same page, it 582 // gives us a one-page buffer if alloca() allocates slightly more memory. 583 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size()); 584 bottom += os::Linux::page_size() - 1; 585 586 // sp might be slightly above current stack pointer; if that's the case, we 587 // will alloca() a little more space than necessary, which is OK. Don't use 588 // os::current_stack_pointer(), as its result can be slightly below current 589 // stack pointer, causing us to not alloca enough to reach "bottom". 590 sp = (address)&sp; 591 592 if (sp > bottom) { 593 size = sp - bottom; 594 p = (volatile char *)alloca(size); 595 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 596 p[0] = '\0'; 597 } 598 } 599 600 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 601 assert(t!=NULL, "just checking"); 602 assert(t->osthread()->expanding_stack(), "expand should be set"); 603 assert(t->stack_base() != NULL, "stack_base was not initialized"); 604 605 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) { 606 sigset_t mask_all, old_sigset; 607 sigfillset(&mask_all); 608 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 609 _expand_stack_to(addr); 610 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 611 return true; 612 } 613 return false; 614 } 615 616 ////////////////////////////////////////////////////////////////////////////// 617 // create new thread 618 619 // Thread start routine for all newly created threads 620 static void *thread_native_entry(Thread *thread) { 621 // Try to randomize the cache line index of hot stack frames. 622 // This helps when threads of the same stack traces evict each other's 623 // cache lines. The threads can be either from the same JVM instance, or 624 // from different JVM instances. The benefit is especially true for 625 // processors with hyperthreading technology. 626 static int counter = 0; 627 int pid = os::current_process_id(); 628 alloca(((pid ^ counter++) & 7) * 128); 629 630 thread->initialize_thread_current(); 631 632 OSThread* osthread = thread->osthread(); 633 Monitor* sync = osthread->startThread_lock(); 634 635 osthread->set_thread_id(os::current_thread_id()); 636 637 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 638 os::current_thread_id(), (uintx) pthread_self()); 639 640 if (UseNUMA) { 641 int lgrp_id = os::numa_get_group_id(); 642 if (lgrp_id != -1) { 643 thread->set_lgrp_id(lgrp_id); 644 } 645 } 646 // initialize signal mask for this thread 647 os::Linux::hotspot_sigmask(thread); 648 649 // initialize floating point control register 650 os::Linux::init_thread_fpu_state(); 651 652 // handshaking with parent thread 653 { 654 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 655 656 // notify parent thread 657 osthread->set_state(INITIALIZED); 658 sync->notify_all(); 659 660 // wait until os::start_thread() 661 while (osthread->get_state() == INITIALIZED) { 662 sync->wait(Mutex::_no_safepoint_check_flag); 663 } 664 } 665 666 // call one more level start routine 667 thread->run(); 668 669 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 670 os::current_thread_id(), (uintx) pthread_self()); 671 672 // If a thread has not deleted itself ("delete this") as part of its 673 // termination sequence, we have to ensure thread-local-storage is 674 // cleared before we actually terminate. No threads should ever be 675 // deleted asynchronously with respect to their termination. 676 if (Thread::current_or_null_safe() != NULL) { 677 assert(Thread::current_or_null_safe() == thread, "current thread is wrong"); 678 thread->clear_thread_current(); 679 } 680 681 return 0; 682 } 683 684 bool os::create_thread(Thread* thread, ThreadType thr_type, 685 size_t req_stack_size) { 686 assert(thread->osthread() == NULL, "caller responsible"); 687 688 // Allocate the OSThread object 689 OSThread* osthread = new OSThread(NULL, NULL); 690 if (osthread == NULL) { 691 return false; 692 } 693 694 // set the correct thread state 695 osthread->set_thread_type(thr_type); 696 697 // Initial state is ALLOCATED but not INITIALIZED 698 osthread->set_state(ALLOCATED); 699 700 thread->set_osthread(osthread); 701 702 // init thread attributes 703 pthread_attr_t attr; 704 pthread_attr_init(&attr); 705 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 706 707 // Calculate stack size if it's not specified by caller. 708 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size); 709 // In the Linux NPTL pthread implementation the guard size mechanism 710 // is not implemented properly. The posix standard requires adding 711 // the size of the guard pages to the stack size, instead Linux 712 // takes the space out of 'stacksize'. Thus we adapt the requested 713 // stack_size by the size of the guard pages to mimick proper 714 // behaviour. However, be careful not to end up with a size 715 // of zero due to overflow. Don't add the guard page in that case. 716 size_t guard_size = os::Linux::default_guard_size(thr_type); 717 if (stack_size <= SIZE_MAX - guard_size) { 718 stack_size += guard_size; 719 } 720 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned"); 721 722 int status = pthread_attr_setstacksize(&attr, stack_size); 723 assert_status(status == 0, status, "pthread_attr_setstacksize"); 724 725 // Configure glibc guard page. 726 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 727 728 ThreadState state; 729 730 { 731 pthread_t tid; 732 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread); 733 734 char buf[64]; 735 if (ret == 0) { 736 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ", 737 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 738 } else { 739 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.", 740 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 741 } 742 743 pthread_attr_destroy(&attr); 744 745 if (ret != 0) { 746 // Need to clean up stuff we've allocated so far 747 thread->set_osthread(NULL); 748 delete osthread; 749 return false; 750 } 751 752 // Store pthread info into the OSThread 753 osthread->set_pthread_id(tid); 754 755 // Wait until child thread is either initialized or aborted 756 { 757 Monitor* sync_with_child = osthread->startThread_lock(); 758 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 759 while ((state = osthread->get_state()) == ALLOCATED) { 760 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 761 } 762 } 763 } 764 765 // Aborted due to thread limit being reached 766 if (state == ZOMBIE) { 767 thread->set_osthread(NULL); 768 delete osthread; 769 return false; 770 } 771 772 // The thread is returned suspended (in state INITIALIZED), 773 // and is started higher up in the call chain 774 assert(state == INITIALIZED, "race condition"); 775 return true; 776 } 777 778 ///////////////////////////////////////////////////////////////////////////// 779 // attach existing thread 780 781 // bootstrap the main thread 782 bool os::create_main_thread(JavaThread* thread) { 783 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 784 return create_attached_thread(thread); 785 } 786 787 bool os::create_attached_thread(JavaThread* thread) { 788 #ifdef ASSERT 789 thread->verify_not_published(); 790 #endif 791 792 // Allocate the OSThread object 793 OSThread* osthread = new OSThread(NULL, NULL); 794 795 if (osthread == NULL) { 796 return false; 797 } 798 799 // Store pthread info into the OSThread 800 osthread->set_thread_id(os::Linux::gettid()); 801 osthread->set_pthread_id(::pthread_self()); 802 803 // initialize floating point control register 804 os::Linux::init_thread_fpu_state(); 805 806 // Initial thread state is RUNNABLE 807 osthread->set_state(RUNNABLE); 808 809 thread->set_osthread(osthread); 810 811 if (UseNUMA) { 812 int lgrp_id = os::numa_get_group_id(); 813 if (lgrp_id != -1) { 814 thread->set_lgrp_id(lgrp_id); 815 } 816 } 817 818 if (os::Linux::is_initial_thread()) { 819 // If current thread is initial thread, its stack is mapped on demand, 820 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 821 // the entire stack region to avoid SEGV in stack banging. 822 // It is also useful to get around the heap-stack-gap problem on SuSE 823 // kernel (see 4821821 for details). We first expand stack to the top 824 // of yellow zone, then enable stack yellow zone (order is significant, 825 // enabling yellow zone first will crash JVM on SuSE Linux), so there 826 // is no gap between the last two virtual memory regions. 827 828 JavaThread *jt = (JavaThread *)thread; 829 address addr = jt->stack_reserved_zone_base(); 830 assert(addr != NULL, "initialization problem?"); 831 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 832 833 osthread->set_expanding_stack(); 834 os::Linux::manually_expand_stack(jt, addr); 835 osthread->clear_expanding_stack(); 836 } 837 838 // initialize signal mask for this thread 839 // and save the caller's signal mask 840 os::Linux::hotspot_sigmask(thread); 841 842 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 843 os::current_thread_id(), (uintx) pthread_self()); 844 845 return true; 846 } 847 848 void os::pd_start_thread(Thread* thread) { 849 OSThread * osthread = thread->osthread(); 850 assert(osthread->get_state() != INITIALIZED, "just checking"); 851 Monitor* sync_with_child = osthread->startThread_lock(); 852 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 853 sync_with_child->notify(); 854 } 855 856 // Free Linux resources related to the OSThread 857 void os::free_thread(OSThread* osthread) { 858 assert(osthread != NULL, "osthread not set"); 859 860 // We are told to free resources of the argument thread, 861 // but we can only really operate on the current thread. 862 assert(Thread::current()->osthread() == osthread, 863 "os::free_thread but not current thread"); 864 865 #ifdef ASSERT 866 sigset_t current; 867 sigemptyset(¤t); 868 pthread_sigmask(SIG_SETMASK, NULL, ¤t); 869 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!"); 870 #endif 871 872 // Restore caller's signal mask 873 sigset_t sigmask = osthread->caller_sigmask(); 874 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 875 876 delete osthread; 877 } 878 879 ////////////////////////////////////////////////////////////////////////////// 880 // initial thread 881 882 // Check if current thread is the initial thread, similar to Solaris thr_main. 883 bool os::Linux::is_initial_thread(void) { 884 char dummy; 885 // If called before init complete, thread stack bottom will be null. 886 // Can be called if fatal error occurs before initialization. 887 if (initial_thread_stack_bottom() == NULL) return false; 888 assert(initial_thread_stack_bottom() != NULL && 889 initial_thread_stack_size() != 0, 890 "os::init did not locate initial thread's stack region"); 891 if ((address)&dummy >= initial_thread_stack_bottom() && 892 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) { 893 return true; 894 } else { 895 return false; 896 } 897 } 898 899 // Find the virtual memory area that contains addr 900 static bool find_vma(address addr, address* vma_low, address* vma_high) { 901 FILE *fp = fopen("/proc/self/maps", "r"); 902 if (fp) { 903 address low, high; 904 while (!feof(fp)) { 905 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 906 if (low <= addr && addr < high) { 907 if (vma_low) *vma_low = low; 908 if (vma_high) *vma_high = high; 909 fclose(fp); 910 return true; 911 } 912 } 913 for (;;) { 914 int ch = fgetc(fp); 915 if (ch == EOF || ch == (int)'\n') break; 916 } 917 } 918 fclose(fp); 919 } 920 return false; 921 } 922 923 // Locate initial thread stack. This special handling of initial thread stack 924 // is needed because pthread_getattr_np() on most (all?) Linux distros returns 925 // bogus value for the primordial process thread. While the launcher has created 926 // the VM in a new thread since JDK 6, we still have to allow for the use of the 927 // JNI invocation API from a primordial thread. 928 void os::Linux::capture_initial_stack(size_t max_size) { 929 930 // max_size is either 0 (which means accept OS default for thread stacks) or 931 // a user-specified value known to be at least the minimum needed. If we 932 // are actually on the primordial thread we can make it appear that we have a 933 // smaller max_size stack by inserting the guard pages at that location. But we 934 // cannot do anything to emulate a larger stack than what has been provided by 935 // the OS or threading library. In fact if we try to use a stack greater than 936 // what is set by rlimit then we will crash the hosting process. 937 938 // Maximum stack size is the easy part, get it from RLIMIT_STACK. 939 // If this is "unlimited" then it will be a huge value. 940 struct rlimit rlim; 941 getrlimit(RLIMIT_STACK, &rlim); 942 size_t stack_size = rlim.rlim_cur; 943 944 // 6308388: a bug in ld.so will relocate its own .data section to the 945 // lower end of primordial stack; reduce ulimit -s value a little bit 946 // so we won't install guard page on ld.so's data section. 947 stack_size -= 2 * page_size(); 948 949 // Try to figure out where the stack base (top) is. This is harder. 950 // 951 // When an application is started, glibc saves the initial stack pointer in 952 // a global variable "__libc_stack_end", which is then used by system 953 // libraries. __libc_stack_end should be pretty close to stack top. The 954 // variable is available since the very early days. However, because it is 955 // a private interface, it could disappear in the future. 956 // 957 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 958 // to __libc_stack_end, it is very close to stack top, but isn't the real 959 // stack top. Note that /proc may not exist if VM is running as a chroot 960 // program, so reading /proc/<pid>/stat could fail. Also the contents of 961 // /proc/<pid>/stat could change in the future (though unlikely). 962 // 963 // We try __libc_stack_end first. If that doesn't work, look for 964 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 965 // as a hint, which should work well in most cases. 966 967 uintptr_t stack_start; 968 969 // try __libc_stack_end first 970 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 971 if (p && *p) { 972 stack_start = *p; 973 } else { 974 // see if we can get the start_stack field from /proc/self/stat 975 FILE *fp; 976 int pid; 977 char state; 978 int ppid; 979 int pgrp; 980 int session; 981 int nr; 982 int tpgrp; 983 unsigned long flags; 984 unsigned long minflt; 985 unsigned long cminflt; 986 unsigned long majflt; 987 unsigned long cmajflt; 988 unsigned long utime; 989 unsigned long stime; 990 long cutime; 991 long cstime; 992 long prio; 993 long nice; 994 long junk; 995 long it_real; 996 uintptr_t start; 997 uintptr_t vsize; 998 intptr_t rss; 999 uintptr_t rsslim; 1000 uintptr_t scodes; 1001 uintptr_t ecode; 1002 int i; 1003 1004 // Figure what the primordial thread stack base is. Code is inspired 1005 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1006 // followed by command name surrounded by parentheses, state, etc. 1007 char stat[2048]; 1008 int statlen; 1009 1010 fp = fopen("/proc/self/stat", "r"); 1011 if (fp) { 1012 statlen = fread(stat, 1, 2047, fp); 1013 stat[statlen] = '\0'; 1014 fclose(fp); 1015 1016 // Skip pid and the command string. Note that we could be dealing with 1017 // weird command names, e.g. user could decide to rename java launcher 1018 // to "java 1.4.2 :)", then the stat file would look like 1019 // 1234 (java 1.4.2 :)) R ... ... 1020 // We don't really need to know the command string, just find the last 1021 // occurrence of ")" and then start parsing from there. See bug 4726580. 1022 char * s = strrchr(stat, ')'); 1023 1024 i = 0; 1025 if (s) { 1026 // Skip blank chars 1027 do { s++; } while (s && isspace(*s)); 1028 1029 #define _UFM UINTX_FORMAT 1030 #define _DFM INTX_FORMAT 1031 1032 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1033 // 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 1034 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, 1035 &state, // 3 %c 1036 &ppid, // 4 %d 1037 &pgrp, // 5 %d 1038 &session, // 6 %d 1039 &nr, // 7 %d 1040 &tpgrp, // 8 %d 1041 &flags, // 9 %lu 1042 &minflt, // 10 %lu 1043 &cminflt, // 11 %lu 1044 &majflt, // 12 %lu 1045 &cmajflt, // 13 %lu 1046 &utime, // 14 %lu 1047 &stime, // 15 %lu 1048 &cutime, // 16 %ld 1049 &cstime, // 17 %ld 1050 &prio, // 18 %ld 1051 &nice, // 19 %ld 1052 &junk, // 20 %ld 1053 &it_real, // 21 %ld 1054 &start, // 22 UINTX_FORMAT 1055 &vsize, // 23 UINTX_FORMAT 1056 &rss, // 24 INTX_FORMAT 1057 &rsslim, // 25 UINTX_FORMAT 1058 &scodes, // 26 UINTX_FORMAT 1059 &ecode, // 27 UINTX_FORMAT 1060 &stack_start); // 28 UINTX_FORMAT 1061 } 1062 1063 #undef _UFM 1064 #undef _DFM 1065 1066 if (i != 28 - 2) { 1067 assert(false, "Bad conversion from /proc/self/stat"); 1068 // product mode - assume we are the initial thread, good luck in the 1069 // embedded case. 1070 warning("Can't detect initial thread stack location - bad conversion"); 1071 stack_start = (uintptr_t) &rlim; 1072 } 1073 } else { 1074 // For some reason we can't open /proc/self/stat (for example, running on 1075 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1076 // most cases, so don't abort: 1077 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1078 stack_start = (uintptr_t) &rlim; 1079 } 1080 } 1081 1082 // Now we have a pointer (stack_start) very close to the stack top, the 1083 // next thing to do is to figure out the exact location of stack top. We 1084 // can find out the virtual memory area that contains stack_start by 1085 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1086 // and its upper limit is the real stack top. (again, this would fail if 1087 // running inside chroot, because /proc may not exist.) 1088 1089 uintptr_t stack_top; 1090 address low, high; 1091 if (find_vma((address)stack_start, &low, &high)) { 1092 // success, "high" is the true stack top. (ignore "low", because initial 1093 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1094 stack_top = (uintptr_t)high; 1095 } else { 1096 // failed, likely because /proc/self/maps does not exist 1097 warning("Can't detect initial thread stack location - find_vma failed"); 1098 // best effort: stack_start is normally within a few pages below the real 1099 // stack top, use it as stack top, and reduce stack size so we won't put 1100 // guard page outside stack. 1101 stack_top = stack_start; 1102 stack_size -= 16 * page_size(); 1103 } 1104 1105 // stack_top could be partially down the page so align it 1106 stack_top = align_up(stack_top, page_size()); 1107 1108 // Allowed stack value is minimum of max_size and what we derived from rlimit 1109 if (max_size > 0) { 1110 _initial_thread_stack_size = MIN2(max_size, stack_size); 1111 } else { 1112 // Accept the rlimit max, but if stack is unlimited then it will be huge, so 1113 // clamp it at 8MB as we do on Solaris 1114 _initial_thread_stack_size = MIN2(stack_size, 8*M); 1115 } 1116 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size()); 1117 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1118 1119 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!"); 1120 1121 if (log_is_enabled(Info, os, thread)) { 1122 // See if we seem to be on primordial process thread 1123 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) && 1124 uintptr_t(&rlim) < stack_top; 1125 1126 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: " 1127 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT, 1128 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K, 1129 stack_top, intptr_t(_initial_thread_stack_bottom)); 1130 } 1131 } 1132 1133 //////////////////////////////////////////////////////////////////////////////// 1134 // time support 1135 1136 // Time since start-up in seconds to a fine granularity. 1137 // Used by VMSelfDestructTimer and the MemProfiler. 1138 double os::elapsedTime() { 1139 1140 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1141 } 1142 1143 jlong os::elapsed_counter() { 1144 return javaTimeNanos() - initial_time_count; 1145 } 1146 1147 jlong os::elapsed_frequency() { 1148 return NANOSECS_PER_SEC; // nanosecond resolution 1149 } 1150 1151 bool os::supports_vtime() { return true; } 1152 bool os::enable_vtime() { return false; } 1153 bool os::vtime_enabled() { return false; } 1154 1155 double os::elapsedVTime() { 1156 struct rusage usage; 1157 int retval = getrusage(RUSAGE_THREAD, &usage); 1158 if (retval == 0) { 1159 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); 1160 } else { 1161 // better than nothing, but not much 1162 return elapsedTime(); 1163 } 1164 } 1165 1166 jlong os::javaTimeMillis() { 1167 timeval time; 1168 int status = gettimeofday(&time, NULL); 1169 assert(status != -1, "linux error"); 1170 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1171 } 1172 1173 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) { 1174 timeval time; 1175 int status = gettimeofday(&time, NULL); 1176 assert(status != -1, "linux error"); 1177 seconds = jlong(time.tv_sec); 1178 nanos = jlong(time.tv_usec) * 1000; 1179 } 1180 1181 1182 #ifndef CLOCK_MONOTONIC 1183 #define CLOCK_MONOTONIC (1) 1184 #endif 1185 1186 void os::Linux::clock_init() { 1187 // we do dlopen's in this particular order due to bug in linux 1188 // dynamical loader (see 6348968) leading to crash on exit 1189 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1190 if (handle == NULL) { 1191 handle = dlopen("librt.so", RTLD_LAZY); 1192 } 1193 1194 if (handle) { 1195 int (*clock_getres_func)(clockid_t, struct timespec*) = 1196 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1197 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1198 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1199 if (clock_getres_func && clock_gettime_func) { 1200 // See if monotonic clock is supported by the kernel. Note that some 1201 // early implementations simply return kernel jiffies (updated every 1202 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1203 // for nano time (though the monotonic property is still nice to have). 1204 // It's fixed in newer kernels, however clock_getres() still returns 1205 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1206 // resolution for now. Hopefully as people move to new kernels, this 1207 // won't be a problem. 1208 struct timespec res; 1209 struct timespec tp; 1210 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1211 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1212 // yes, monotonic clock is supported 1213 _clock_gettime = clock_gettime_func; 1214 return; 1215 } else { 1216 // close librt if there is no monotonic clock 1217 dlclose(handle); 1218 } 1219 } 1220 } 1221 warning("No monotonic clock was available - timed services may " \ 1222 "be adversely affected if the time-of-day clock changes"); 1223 } 1224 1225 #ifndef SYS_clock_getres 1226 #if defined(X86) || defined(PPC64) || defined(S390) 1227 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261) 1228 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1229 #else 1230 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1231 #define sys_clock_getres(x,y) -1 1232 #endif 1233 #else 1234 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1235 #endif 1236 1237 void os::Linux::fast_thread_clock_init() { 1238 if (!UseLinuxPosixThreadCPUClocks) { 1239 return; 1240 } 1241 clockid_t clockid; 1242 struct timespec tp; 1243 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1244 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1245 1246 // Switch to using fast clocks for thread cpu time if 1247 // the sys_clock_getres() returns 0 error code. 1248 // Note, that some kernels may support the current thread 1249 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1250 // returned by the pthread_getcpuclockid(). 1251 // If the fast Posix clocks are supported then the sys_clock_getres() 1252 // must return at least tp.tv_sec == 0 which means a resolution 1253 // better than 1 sec. This is extra check for reliability. 1254 1255 if (pthread_getcpuclockid_func && 1256 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1257 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1258 _supports_fast_thread_cpu_time = true; 1259 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1260 } 1261 } 1262 1263 jlong os::javaTimeNanos() { 1264 if (os::supports_monotonic_clock()) { 1265 struct timespec tp; 1266 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1267 assert(status == 0, "gettime error"); 1268 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1269 return result; 1270 } else { 1271 timeval time; 1272 int status = gettimeofday(&time, NULL); 1273 assert(status != -1, "linux error"); 1274 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1275 return 1000 * usecs; 1276 } 1277 } 1278 1279 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1280 if (os::supports_monotonic_clock()) { 1281 info_ptr->max_value = ALL_64_BITS; 1282 1283 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1284 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1285 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1286 } else { 1287 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1288 info_ptr->max_value = ALL_64_BITS; 1289 1290 // gettimeofday is a real time clock so it skips 1291 info_ptr->may_skip_backward = true; 1292 info_ptr->may_skip_forward = true; 1293 } 1294 1295 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1296 } 1297 1298 // Return the real, user, and system times in seconds from an 1299 // arbitrary fixed point in the past. 1300 bool os::getTimesSecs(double* process_real_time, 1301 double* process_user_time, 1302 double* process_system_time) { 1303 struct tms ticks; 1304 clock_t real_ticks = times(&ticks); 1305 1306 if (real_ticks == (clock_t) (-1)) { 1307 return false; 1308 } else { 1309 double ticks_per_second = (double) clock_tics_per_sec; 1310 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1311 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1312 *process_real_time = ((double) real_ticks) / ticks_per_second; 1313 1314 return true; 1315 } 1316 } 1317 1318 1319 char * os::local_time_string(char *buf, size_t buflen) { 1320 struct tm t; 1321 time_t long_time; 1322 time(&long_time); 1323 localtime_r(&long_time, &t); 1324 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1325 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1326 t.tm_hour, t.tm_min, t.tm_sec); 1327 return buf; 1328 } 1329 1330 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1331 return localtime_r(clock, res); 1332 } 1333 1334 //////////////////////////////////////////////////////////////////////////////// 1335 // runtime exit support 1336 1337 // Note: os::shutdown() might be called very early during initialization, or 1338 // called from signal handler. Before adding something to os::shutdown(), make 1339 // sure it is async-safe and can handle partially initialized VM. 1340 void os::shutdown() { 1341 1342 // allow PerfMemory to attempt cleanup of any persistent resources 1343 perfMemory_exit(); 1344 1345 // needs to remove object in file system 1346 AttachListener::abort(); 1347 1348 // flush buffered output, finish log files 1349 ostream_abort(); 1350 1351 // Check for abort hook 1352 abort_hook_t abort_hook = Arguments::abort_hook(); 1353 if (abort_hook != NULL) { 1354 abort_hook(); 1355 } 1356 1357 } 1358 1359 // Note: os::abort() might be called very early during initialization, or 1360 // called from signal handler. Before adding something to os::abort(), make 1361 // sure it is async-safe and can handle partially initialized VM. 1362 void os::abort(bool dump_core, void* siginfo, const void* context) { 1363 os::shutdown(); 1364 if (dump_core) { 1365 #ifndef PRODUCT 1366 fdStream out(defaultStream::output_fd()); 1367 out.print_raw("Current thread is "); 1368 char buf[16]; 1369 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1370 out.print_raw_cr(buf); 1371 out.print_raw_cr("Dumping core ..."); 1372 #endif 1373 ::abort(); // dump core 1374 } 1375 1376 ::exit(1); 1377 } 1378 1379 // Die immediately, no exit hook, no abort hook, no cleanup. 1380 void os::die() { 1381 ::abort(); 1382 } 1383 1384 1385 // This method is a copy of JDK's sysGetLastErrorString 1386 // from src/solaris/hpi/src/system_md.c 1387 1388 size_t os::lasterror(char *buf, size_t len) { 1389 if (errno == 0) return 0; 1390 1391 const char *s = os::strerror(errno); 1392 size_t n = ::strlen(s); 1393 if (n >= len) { 1394 n = len - 1; 1395 } 1396 ::strncpy(buf, s, n); 1397 buf[n] = '\0'; 1398 return n; 1399 } 1400 1401 // thread_id is kernel thread id (similar to Solaris LWP id) 1402 intx os::current_thread_id() { return os::Linux::gettid(); } 1403 int os::current_process_id() { 1404 return ::getpid(); 1405 } 1406 1407 // DLL functions 1408 1409 const char* os::dll_file_extension() { return ".so"; } 1410 1411 // This must be hard coded because it's the system's temporary 1412 // directory not the java application's temp directory, ala java.io.tmpdir. 1413 const char* os::get_temp_directory() { return "/tmp"; } 1414 1415 static bool file_exists(const char* filename) { 1416 struct stat statbuf; 1417 if (filename == NULL || strlen(filename) == 0) { 1418 return false; 1419 } 1420 return os::stat(filename, &statbuf) == 0; 1421 } 1422 1423 // check if addr is inside libjvm.so 1424 bool os::address_is_in_vm(address addr) { 1425 static address libjvm_base_addr; 1426 Dl_info dlinfo; 1427 1428 if (libjvm_base_addr == NULL) { 1429 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1430 libjvm_base_addr = (address)dlinfo.dli_fbase; 1431 } 1432 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1433 } 1434 1435 if (dladdr((void *)addr, &dlinfo) != 0) { 1436 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1437 } 1438 1439 return false; 1440 } 1441 1442 bool os::dll_address_to_function_name(address addr, char *buf, 1443 int buflen, int *offset, 1444 bool demangle) { 1445 // buf is not optional, but offset is optional 1446 assert(buf != NULL, "sanity check"); 1447 1448 Dl_info dlinfo; 1449 1450 if (dladdr((void*)addr, &dlinfo) != 0) { 1451 // see if we have a matching symbol 1452 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1453 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) { 1454 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1455 } 1456 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1457 return true; 1458 } 1459 // no matching symbol so try for just file info 1460 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1461 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1462 buf, buflen, offset, dlinfo.dli_fname, demangle)) { 1463 return true; 1464 } 1465 } 1466 } 1467 1468 buf[0] = '\0'; 1469 if (offset != NULL) *offset = -1; 1470 return false; 1471 } 1472 1473 struct _address_to_library_name { 1474 address addr; // input : memory address 1475 size_t buflen; // size of fname 1476 char* fname; // output: library name 1477 address base; // library base addr 1478 }; 1479 1480 static int address_to_library_name_callback(struct dl_phdr_info *info, 1481 size_t size, void *data) { 1482 int i; 1483 bool found = false; 1484 address libbase = NULL; 1485 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1486 1487 // iterate through all loadable segments 1488 for (i = 0; i < info->dlpi_phnum; i++) { 1489 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1490 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1491 // base address of a library is the lowest address of its loaded 1492 // segments. 1493 if (libbase == NULL || libbase > segbase) { 1494 libbase = segbase; 1495 } 1496 // see if 'addr' is within current segment 1497 if (segbase <= d->addr && 1498 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1499 found = true; 1500 } 1501 } 1502 } 1503 1504 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1505 // so dll_address_to_library_name() can fall through to use dladdr() which 1506 // can figure out executable name from argv[0]. 1507 if (found && info->dlpi_name && info->dlpi_name[0]) { 1508 d->base = libbase; 1509 if (d->fname) { 1510 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1511 } 1512 return 1; 1513 } 1514 return 0; 1515 } 1516 1517 bool os::dll_address_to_library_name(address addr, char* buf, 1518 int buflen, int* offset) { 1519 // buf is not optional, but offset is optional 1520 assert(buf != NULL, "sanity check"); 1521 1522 Dl_info dlinfo; 1523 struct _address_to_library_name data; 1524 1525 // There is a bug in old glibc dladdr() implementation that it could resolve 1526 // to wrong library name if the .so file has a base address != NULL. Here 1527 // we iterate through the program headers of all loaded libraries to find 1528 // out which library 'addr' really belongs to. This workaround can be 1529 // removed once the minimum requirement for glibc is moved to 2.3.x. 1530 data.addr = addr; 1531 data.fname = buf; 1532 data.buflen = buflen; 1533 data.base = NULL; 1534 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1535 1536 if (rslt) { 1537 // buf already contains library name 1538 if (offset) *offset = addr - data.base; 1539 return true; 1540 } 1541 if (dladdr((void*)addr, &dlinfo) != 0) { 1542 if (dlinfo.dli_fname != NULL) { 1543 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1544 } 1545 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1546 *offset = addr - (address)dlinfo.dli_fbase; 1547 } 1548 return true; 1549 } 1550 1551 buf[0] = '\0'; 1552 if (offset) *offset = -1; 1553 return false; 1554 } 1555 1556 // Loads .dll/.so and 1557 // in case of error it checks if .dll/.so was built for the 1558 // same architecture as Hotspot is running on 1559 1560 1561 // Remember the stack's state. The Linux dynamic linker will change 1562 // the stack to 'executable' at most once, so we must safepoint only once. 1563 bool os::Linux::_stack_is_executable = false; 1564 1565 // VM operation that loads a library. This is necessary if stack protection 1566 // of the Java stacks can be lost during loading the library. If we 1567 // do not stop the Java threads, they can stack overflow before the stacks 1568 // are protected again. 1569 class VM_LinuxDllLoad: public VM_Operation { 1570 private: 1571 const char *_filename; 1572 char *_ebuf; 1573 int _ebuflen; 1574 void *_lib; 1575 public: 1576 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1577 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1578 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1579 void doit() { 1580 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1581 os::Linux::_stack_is_executable = true; 1582 } 1583 void* loaded_library() { return _lib; } 1584 }; 1585 1586 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1587 void * result = NULL; 1588 bool load_attempted = false; 1589 1590 // Check whether the library to load might change execution rights 1591 // of the stack. If they are changed, the protection of the stack 1592 // guard pages will be lost. We need a safepoint to fix this. 1593 // 1594 // See Linux man page execstack(8) for more info. 1595 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1596 if (!ElfFile::specifies_noexecstack(filename)) { 1597 if (!is_init_completed()) { 1598 os::Linux::_stack_is_executable = true; 1599 // This is OK - No Java threads have been created yet, and hence no 1600 // stack guard pages to fix. 1601 // 1602 // This should happen only when you are building JDK7 using a very 1603 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1604 // 1605 // Dynamic loader will make all stacks executable after 1606 // this function returns, and will not do that again. 1607 #ifdef ASSERT 1608 ThreadsListHandle tlh; 1609 assert(tlh.length() == 0, "no Java threads should exist yet."); 1610 #endif 1611 } else { 1612 warning("You have loaded library %s which might have disabled stack guard. " 1613 "The VM will try to fix the stack guard now.\n" 1614 "It's highly recommended that you fix the library with " 1615 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1616 filename); 1617 1618 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1619 JavaThread *jt = JavaThread::current(); 1620 if (jt->thread_state() != _thread_in_native) { 1621 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1622 // that requires ExecStack. Cannot enter safe point. Let's give up. 1623 warning("Unable to fix stack guard. Giving up."); 1624 } else { 1625 if (!LoadExecStackDllInVMThread) { 1626 // This is for the case where the DLL has an static 1627 // constructor function that executes JNI code. We cannot 1628 // load such DLLs in the VMThread. 1629 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1630 } 1631 1632 ThreadInVMfromNative tiv(jt); 1633 debug_only(VMNativeEntryWrapper vew;) 1634 1635 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1636 VMThread::execute(&op); 1637 if (LoadExecStackDllInVMThread) { 1638 result = op.loaded_library(); 1639 } 1640 load_attempted = true; 1641 } 1642 } 1643 } 1644 } 1645 1646 if (!load_attempted) { 1647 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1648 } 1649 1650 if (result != NULL) { 1651 // Successful loading 1652 return result; 1653 } 1654 1655 Elf32_Ehdr elf_head; 1656 int diag_msg_max_length=ebuflen-strlen(ebuf); 1657 char* diag_msg_buf=ebuf+strlen(ebuf); 1658 1659 if (diag_msg_max_length==0) { 1660 // No more space in ebuf for additional diagnostics message 1661 return NULL; 1662 } 1663 1664 1665 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1666 1667 if (file_descriptor < 0) { 1668 // Can't open library, report dlerror() message 1669 return NULL; 1670 } 1671 1672 bool failed_to_read_elf_head= 1673 (sizeof(elf_head)!= 1674 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1675 1676 ::close(file_descriptor); 1677 if (failed_to_read_elf_head) { 1678 // file i/o error - report dlerror() msg 1679 return NULL; 1680 } 1681 1682 typedef struct { 1683 Elf32_Half code; // Actual value as defined in elf.h 1684 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1685 unsigned char elf_class; // 32 or 64 bit 1686 unsigned char endianess; // MSB or LSB 1687 char* name; // String representation 1688 } arch_t; 1689 1690 #ifndef EM_486 1691 #define EM_486 6 /* Intel 80486 */ 1692 #endif 1693 #ifndef EM_AARCH64 1694 #define EM_AARCH64 183 /* ARM AARCH64 */ 1695 #endif 1696 1697 static const arch_t arch_array[]={ 1698 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1699 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1700 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1701 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1702 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1703 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1704 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1705 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1706 #if defined(VM_LITTLE_ENDIAN) 1707 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"}, 1708 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"}, 1709 #else 1710 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1711 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"}, 1712 #endif 1713 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1714 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1715 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1716 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1717 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1718 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1719 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1720 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1721 }; 1722 1723 #if (defined IA32) 1724 static Elf32_Half running_arch_code=EM_386; 1725 #elif (defined AMD64) 1726 static Elf32_Half running_arch_code=EM_X86_64; 1727 #elif (defined IA64) 1728 static Elf32_Half running_arch_code=EM_IA_64; 1729 #elif (defined __sparc) && (defined _LP64) 1730 static Elf32_Half running_arch_code=EM_SPARCV9; 1731 #elif (defined __sparc) && (!defined _LP64) 1732 static Elf32_Half running_arch_code=EM_SPARC; 1733 #elif (defined __powerpc64__) 1734 static Elf32_Half running_arch_code=EM_PPC64; 1735 #elif (defined __powerpc__) 1736 static Elf32_Half running_arch_code=EM_PPC; 1737 #elif (defined AARCH64) 1738 static Elf32_Half running_arch_code=EM_AARCH64; 1739 #elif (defined ARM) 1740 static Elf32_Half running_arch_code=EM_ARM; 1741 #elif (defined S390) 1742 static Elf32_Half running_arch_code=EM_S390; 1743 #elif (defined ALPHA) 1744 static Elf32_Half running_arch_code=EM_ALPHA; 1745 #elif (defined MIPSEL) 1746 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1747 #elif (defined PARISC) 1748 static Elf32_Half running_arch_code=EM_PARISC; 1749 #elif (defined MIPS) 1750 static Elf32_Half running_arch_code=EM_MIPS; 1751 #elif (defined M68K) 1752 static Elf32_Half running_arch_code=EM_68K; 1753 #elif (defined SH) 1754 static Elf32_Half running_arch_code=EM_SH; 1755 #else 1756 #error Method os::dll_load requires that one of following is defined:\ 1757 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc 1758 #endif 1759 1760 // Identify compatability class for VM's architecture and library's architecture 1761 // Obtain string descriptions for architectures 1762 1763 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1764 int running_arch_index=-1; 1765 1766 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1767 if (running_arch_code == arch_array[i].code) { 1768 running_arch_index = i; 1769 } 1770 if (lib_arch.code == arch_array[i].code) { 1771 lib_arch.compat_class = arch_array[i].compat_class; 1772 lib_arch.name = arch_array[i].name; 1773 } 1774 } 1775 1776 assert(running_arch_index != -1, 1777 "Didn't find running architecture code (running_arch_code) in arch_array"); 1778 if (running_arch_index == -1) { 1779 // Even though running architecture detection failed 1780 // we may still continue with reporting dlerror() message 1781 return NULL; 1782 } 1783 1784 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1785 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1786 return NULL; 1787 } 1788 1789 #ifndef S390 1790 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1791 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1792 return NULL; 1793 } 1794 #endif // !S390 1795 1796 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1797 if (lib_arch.name!=NULL) { 1798 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1799 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1800 lib_arch.name, arch_array[running_arch_index].name); 1801 } else { 1802 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1803 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1804 lib_arch.code, 1805 arch_array[running_arch_index].name); 1806 } 1807 } 1808 1809 return NULL; 1810 } 1811 1812 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 1813 int ebuflen) { 1814 void * result = ::dlopen(filename, RTLD_LAZY); 1815 if (result == NULL) { 1816 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 1817 ebuf[ebuflen-1] = '\0'; 1818 } 1819 return result; 1820 } 1821 1822 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 1823 int ebuflen) { 1824 void * result = NULL; 1825 if (LoadExecStackDllInVMThread) { 1826 result = dlopen_helper(filename, ebuf, ebuflen); 1827 } 1828 1829 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 1830 // library that requires an executable stack, or which does not have this 1831 // stack attribute set, dlopen changes the stack attribute to executable. The 1832 // read protection of the guard pages gets lost. 1833 // 1834 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 1835 // may have been queued at the same time. 1836 1837 if (!_stack_is_executable) { 1838 ThreadsListHandle tlh; 1839 JavaThreadIterator jti(tlh.list()); 1840 for (JavaThread *jt = jti.first(); jt != NULL; jt = jti.next()) { 1841 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 1842 jt->stack_guards_enabled()) { // No pending stack overflow exceptions 1843 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) { 1844 warning("Attempt to reguard stack yellow zone failed."); 1845 } 1846 } 1847 } 1848 } 1849 1850 return result; 1851 } 1852 1853 void* os::dll_lookup(void* handle, const char* name) { 1854 void* res = dlsym(handle, name); 1855 return res; 1856 } 1857 1858 void* os::get_default_process_handle() { 1859 return (void*)::dlopen(NULL, RTLD_LAZY); 1860 } 1861 1862 static bool _print_ascii_file(const char* filename, outputStream* st) { 1863 int fd = ::open(filename, O_RDONLY); 1864 if (fd == -1) { 1865 return false; 1866 } 1867 1868 char buf[33]; 1869 int bytes; 1870 buf[32] = '\0'; 1871 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) { 1872 st->print_raw(buf, bytes); 1873 } 1874 1875 ::close(fd); 1876 1877 return true; 1878 } 1879 1880 void os::print_dll_info(outputStream *st) { 1881 st->print_cr("Dynamic libraries:"); 1882 1883 char fname[32]; 1884 pid_t pid = os::Linux::gettid(); 1885 1886 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1887 1888 if (!_print_ascii_file(fname, st)) { 1889 st->print("Can not get library information for pid = %d\n", pid); 1890 } 1891 } 1892 1893 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 1894 FILE *procmapsFile = NULL; 1895 1896 // Open the procfs maps file for the current process 1897 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 1898 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 1899 char line[PATH_MAX + 100]; 1900 1901 // Read line by line from 'file' 1902 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 1903 u8 base, top, offset, inode; 1904 char permissions[5]; 1905 char device[6]; 1906 char name[PATH_MAX + 1]; 1907 1908 // Parse fields from line 1909 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s", 1910 &base, &top, permissions, &offset, device, &inode, name); 1911 1912 // Filter by device id '00:00' so that we only get file system mapped files. 1913 if (strcmp(device, "00:00") != 0) { 1914 1915 // Call callback with the fields of interest 1916 if(callback(name, (address)base, (address)top, param)) { 1917 // Oops abort, callback aborted 1918 fclose(procmapsFile); 1919 return 1; 1920 } 1921 } 1922 } 1923 fclose(procmapsFile); 1924 } 1925 return 0; 1926 } 1927 1928 void os::print_os_info_brief(outputStream* st) { 1929 os::Linux::print_distro_info(st); 1930 1931 os::Posix::print_uname_info(st); 1932 1933 os::Linux::print_libversion_info(st); 1934 1935 } 1936 1937 void os::print_os_info(outputStream* st) { 1938 st->print("OS:"); 1939 1940 os::Linux::print_distro_info(st); 1941 1942 os::Posix::print_uname_info(st); 1943 1944 // Print warning if unsafe chroot environment detected 1945 if (unsafe_chroot_detected) { 1946 st->print("WARNING!! "); 1947 st->print_cr("%s", unstable_chroot_error); 1948 } 1949 1950 os::Linux::print_libversion_info(st); 1951 1952 os::Posix::print_rlimit_info(st); 1953 1954 os::Posix::print_load_average(st); 1955 1956 os::Linux::print_full_memory_info(st); 1957 } 1958 1959 // Try to identify popular distros. 1960 // Most Linux distributions have a /etc/XXX-release file, which contains 1961 // the OS version string. Newer Linux distributions have a /etc/lsb-release 1962 // file that also contains the OS version string. Some have more than one 1963 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 1964 // /etc/redhat-release.), so the order is important. 1965 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 1966 // their own specific XXX-release file as well as a redhat-release file. 1967 // Because of this the XXX-release file needs to be searched for before the 1968 // redhat-release file. 1969 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the 1970 // search for redhat-release / SuSE-release needs to be before lsb-release. 1971 // Since the lsb-release file is the new standard it needs to be searched 1972 // before the older style release files. 1973 // Searching system-release (Red Hat) and os-release (other Linuxes) are a 1974 // next to last resort. The os-release file is a new standard that contains 1975 // distribution information and the system-release file seems to be an old 1976 // standard that has been replaced by the lsb-release and os-release files. 1977 // Searching for the debian_version file is the last resort. It contains 1978 // an informative string like "6.0.6" or "wheezy/sid". Because of this 1979 // "Debian " is printed before the contents of the debian_version file. 1980 1981 const char* distro_files[] = { 1982 "/etc/oracle-release", 1983 "/etc/mandriva-release", 1984 "/etc/mandrake-release", 1985 "/etc/sun-release", 1986 "/etc/redhat-release", 1987 "/etc/SuSE-release", 1988 "/etc/lsb-release", 1989 "/etc/turbolinux-release", 1990 "/etc/gentoo-release", 1991 "/etc/ltib-release", 1992 "/etc/angstrom-version", 1993 "/etc/system-release", 1994 "/etc/os-release", 1995 NULL }; 1996 1997 void os::Linux::print_distro_info(outputStream* st) { 1998 for (int i = 0;; i++) { 1999 const char* file = distro_files[i]; 2000 if (file == NULL) { 2001 break; // done 2002 } 2003 // If file prints, we found it. 2004 if (_print_ascii_file(file, st)) { 2005 return; 2006 } 2007 } 2008 2009 if (file_exists("/etc/debian_version")) { 2010 st->print("Debian "); 2011 _print_ascii_file("/etc/debian_version", st); 2012 } else { 2013 st->print("Linux"); 2014 } 2015 st->cr(); 2016 } 2017 2018 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) { 2019 char buf[256]; 2020 while (fgets(buf, sizeof(buf), fp)) { 2021 // Edit out extra stuff in expected format 2022 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) { 2023 char* ptr = strstr(buf, "\""); // the name is in quotes 2024 if (ptr != NULL) { 2025 ptr++; // go beyond first quote 2026 char* nl = strchr(ptr, '\"'); 2027 if (nl != NULL) *nl = '\0'; 2028 strncpy(distro, ptr, length); 2029 } else { 2030 ptr = strstr(buf, "="); 2031 ptr++; // go beyond equals then 2032 char* nl = strchr(ptr, '\n'); 2033 if (nl != NULL) *nl = '\0'; 2034 strncpy(distro, ptr, length); 2035 } 2036 return; 2037 } else if (get_first_line) { 2038 char* nl = strchr(buf, '\n'); 2039 if (nl != NULL) *nl = '\0'; 2040 strncpy(distro, buf, length); 2041 return; 2042 } 2043 } 2044 // print last line and close 2045 char* nl = strchr(buf, '\n'); 2046 if (nl != NULL) *nl = '\0'; 2047 strncpy(distro, buf, length); 2048 } 2049 2050 static void parse_os_info(char* distro, size_t length, const char* file) { 2051 FILE* fp = fopen(file, "r"); 2052 if (fp != NULL) { 2053 // if suse format, print out first line 2054 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0); 2055 parse_os_info_helper(fp, distro, length, get_first_line); 2056 fclose(fp); 2057 } 2058 } 2059 2060 void os::get_summary_os_info(char* buf, size_t buflen) { 2061 for (int i = 0;; i++) { 2062 const char* file = distro_files[i]; 2063 if (file == NULL) { 2064 break; // ran out of distro_files 2065 } 2066 if (file_exists(file)) { 2067 parse_os_info(buf, buflen, file); 2068 return; 2069 } 2070 } 2071 // special case for debian 2072 if (file_exists("/etc/debian_version")) { 2073 strncpy(buf, "Debian ", buflen); 2074 parse_os_info(&buf[7], buflen-7, "/etc/debian_version"); 2075 } else { 2076 strncpy(buf, "Linux", buflen); 2077 } 2078 } 2079 2080 void os::Linux::print_libversion_info(outputStream* st) { 2081 // libc, pthread 2082 st->print("libc:"); 2083 st->print("%s ", os::Linux::glibc_version()); 2084 st->print("%s ", os::Linux::libpthread_version()); 2085 st->cr(); 2086 } 2087 2088 void os::Linux::print_full_memory_info(outputStream* st) { 2089 st->print("\n/proc/meminfo:\n"); 2090 _print_ascii_file("/proc/meminfo", st); 2091 st->cr(); 2092 } 2093 2094 void os::print_memory_info(outputStream* st) { 2095 2096 st->print("Memory:"); 2097 st->print(" %dk page", os::vm_page_size()>>10); 2098 2099 // values in struct sysinfo are "unsigned long" 2100 struct sysinfo si; 2101 sysinfo(&si); 2102 2103 st->print(", physical " UINT64_FORMAT "k", 2104 os::physical_memory() >> 10); 2105 st->print("(" UINT64_FORMAT "k free)", 2106 os::available_memory() >> 10); 2107 st->print(", swap " UINT64_FORMAT "k", 2108 ((jlong)si.totalswap * si.mem_unit) >> 10); 2109 st->print("(" UINT64_FORMAT "k free)", 2110 ((jlong)si.freeswap * si.mem_unit) >> 10); 2111 st->cr(); 2112 } 2113 2114 // Print the first "model name" line and the first "flags" line 2115 // that we find and nothing more. We assume "model name" comes 2116 // before "flags" so if we find a second "model name", then the 2117 // "flags" field is considered missing. 2118 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) { 2119 #if defined(IA32) || defined(AMD64) 2120 // Other platforms have less repetitive cpuinfo files 2121 FILE *fp = fopen("/proc/cpuinfo", "r"); 2122 if (fp) { 2123 while (!feof(fp)) { 2124 if (fgets(buf, buflen, fp)) { 2125 // Assume model name comes before flags 2126 bool model_name_printed = false; 2127 if (strstr(buf, "model name") != NULL) { 2128 if (!model_name_printed) { 2129 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n"); 2130 st->print_raw(buf); 2131 model_name_printed = true; 2132 } else { 2133 // model name printed but not flags? Odd, just return 2134 fclose(fp); 2135 return true; 2136 } 2137 } 2138 // print the flags line too 2139 if (strstr(buf, "flags") != NULL) { 2140 st->print_raw(buf); 2141 fclose(fp); 2142 return true; 2143 } 2144 } 2145 } 2146 fclose(fp); 2147 } 2148 #endif // x86 platforms 2149 return false; 2150 } 2151 2152 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) { 2153 // Only print the model name if the platform provides this as a summary 2154 if (!print_model_name_and_flags(st, buf, buflen)) { 2155 st->print("\n/proc/cpuinfo:\n"); 2156 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2157 st->print_cr(" <Not Available>"); 2158 } 2159 } 2160 } 2161 2162 #if defined(AMD64) || defined(IA32) || defined(X32) 2163 const char* search_string = "model name"; 2164 #elif defined(M68K) 2165 const char* search_string = "CPU"; 2166 #elif defined(PPC64) 2167 const char* search_string = "cpu"; 2168 #elif defined(S390) 2169 const char* search_string = "processor"; 2170 #elif defined(SPARC) 2171 const char* search_string = "cpu"; 2172 #else 2173 const char* search_string = "Processor"; 2174 #endif 2175 2176 // Parses the cpuinfo file for string representing the model name. 2177 void os::get_summary_cpu_info(char* cpuinfo, size_t length) { 2178 FILE* fp = fopen("/proc/cpuinfo", "r"); 2179 if (fp != NULL) { 2180 while (!feof(fp)) { 2181 char buf[256]; 2182 if (fgets(buf, sizeof(buf), fp)) { 2183 char* start = strstr(buf, search_string); 2184 if (start != NULL) { 2185 char *ptr = start + strlen(search_string); 2186 char *end = buf + strlen(buf); 2187 while (ptr != end) { 2188 // skip whitespace and colon for the rest of the name. 2189 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') { 2190 break; 2191 } 2192 ptr++; 2193 } 2194 if (ptr != end) { 2195 // reasonable string, get rid of newline and keep the rest 2196 char* nl = strchr(buf, '\n'); 2197 if (nl != NULL) *nl = '\0'; 2198 strncpy(cpuinfo, ptr, length); 2199 fclose(fp); 2200 return; 2201 } 2202 } 2203 } 2204 } 2205 fclose(fp); 2206 } 2207 // cpuinfo not found or parsing failed, just print generic string. The entire 2208 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86) 2209 #if defined(AARCH64) 2210 strncpy(cpuinfo, "AArch64", length); 2211 #elif defined(AMD64) 2212 strncpy(cpuinfo, "x86_64", length); 2213 #elif defined(ARM) // Order wrt. AARCH64 is relevant! 2214 strncpy(cpuinfo, "ARM", length); 2215 #elif defined(IA32) 2216 strncpy(cpuinfo, "x86_32", length); 2217 #elif defined(IA64) 2218 strncpy(cpuinfo, "IA64", length); 2219 #elif defined(PPC) 2220 strncpy(cpuinfo, "PPC64", length); 2221 #elif defined(S390) 2222 strncpy(cpuinfo, "S390", length); 2223 #elif defined(SPARC) 2224 strncpy(cpuinfo, "sparcv9", length); 2225 #elif defined(ZERO_LIBARCH) 2226 strncpy(cpuinfo, ZERO_LIBARCH, length); 2227 #else 2228 strncpy(cpuinfo, "unknown", length); 2229 #endif 2230 } 2231 2232 static void print_signal_handler(outputStream* st, int sig, 2233 char* buf, size_t buflen); 2234 2235 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2236 st->print_cr("Signal Handlers:"); 2237 print_signal_handler(st, SIGSEGV, buf, buflen); 2238 print_signal_handler(st, SIGBUS , buf, buflen); 2239 print_signal_handler(st, SIGFPE , buf, buflen); 2240 print_signal_handler(st, SIGPIPE, buf, buflen); 2241 print_signal_handler(st, SIGXFSZ, buf, buflen); 2242 print_signal_handler(st, SIGILL , buf, buflen); 2243 print_signal_handler(st, SR_signum, buf, buflen); 2244 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2245 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2246 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2247 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2248 #if defined(PPC64) 2249 print_signal_handler(st, SIGTRAP, buf, buflen); 2250 #endif 2251 } 2252 2253 static char saved_jvm_path[MAXPATHLEN] = {0}; 2254 2255 // Find the full path to the current module, libjvm.so 2256 void os::jvm_path(char *buf, jint buflen) { 2257 // Error checking. 2258 if (buflen < MAXPATHLEN) { 2259 assert(false, "must use a large-enough buffer"); 2260 buf[0] = '\0'; 2261 return; 2262 } 2263 // Lazy resolve the path to current module. 2264 if (saved_jvm_path[0] != 0) { 2265 strcpy(buf, saved_jvm_path); 2266 return; 2267 } 2268 2269 char dli_fname[MAXPATHLEN]; 2270 bool ret = dll_address_to_library_name( 2271 CAST_FROM_FN_PTR(address, os::jvm_path), 2272 dli_fname, sizeof(dli_fname), NULL); 2273 assert(ret, "cannot locate libjvm"); 2274 char *rp = NULL; 2275 if (ret && dli_fname[0] != '\0') { 2276 rp = os::Posix::realpath(dli_fname, buf, buflen); 2277 } 2278 if (rp == NULL) { 2279 return; 2280 } 2281 2282 if (Arguments::sun_java_launcher_is_altjvm()) { 2283 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2284 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so". 2285 // If "/jre/lib/" appears at the right place in the string, then 2286 // assume we are installed in a JDK and we're done. Otherwise, check 2287 // for a JAVA_HOME environment variable and fix up the path so it 2288 // looks like libjvm.so is installed there (append a fake suffix 2289 // hotspot/libjvm.so). 2290 const char *p = buf + strlen(buf) - 1; 2291 for (int count = 0; p > buf && count < 5; ++count) { 2292 for (--p; p > buf && *p != '/'; --p) 2293 /* empty */ ; 2294 } 2295 2296 if (strncmp(p, "/jre/lib/", 9) != 0) { 2297 // Look for JAVA_HOME in the environment. 2298 char* java_home_var = ::getenv("JAVA_HOME"); 2299 if (java_home_var != NULL && java_home_var[0] != 0) { 2300 char* jrelib_p; 2301 int len; 2302 2303 // Check the current module name "libjvm.so". 2304 p = strrchr(buf, '/'); 2305 if (p == NULL) { 2306 return; 2307 } 2308 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2309 2310 rp = os::Posix::realpath(java_home_var, buf, buflen); 2311 if (rp == NULL) { 2312 return; 2313 } 2314 2315 // determine if this is a legacy image or modules image 2316 // modules image doesn't have "jre" subdirectory 2317 len = strlen(buf); 2318 assert(len < buflen, "Ran out of buffer room"); 2319 jrelib_p = buf + len; 2320 snprintf(jrelib_p, buflen-len, "/jre/lib"); 2321 if (0 != access(buf, F_OK)) { 2322 snprintf(jrelib_p, buflen-len, "/lib"); 2323 } 2324 2325 if (0 == access(buf, F_OK)) { 2326 // Use current module name "libjvm.so" 2327 len = strlen(buf); 2328 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2329 } else { 2330 // Go back to path of .so 2331 rp = os::Posix::realpath(dli_fname, buf, buflen); 2332 if (rp == NULL) { 2333 return; 2334 } 2335 } 2336 } 2337 } 2338 } 2339 2340 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2341 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2342 } 2343 2344 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2345 // no prefix required, not even "_" 2346 } 2347 2348 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2349 // no suffix required 2350 } 2351 2352 //////////////////////////////////////////////////////////////////////////////// 2353 // sun.misc.Signal support 2354 2355 static volatile jint sigint_count = 0; 2356 2357 static void UserHandler(int sig, void *siginfo, void *context) { 2358 // 4511530 - sem_post is serialized and handled by the manager thread. When 2359 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2360 // don't want to flood the manager thread with sem_post requests. 2361 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2362 return; 2363 } 2364 2365 // Ctrl-C is pressed during error reporting, likely because the error 2366 // handler fails to abort. Let VM die immediately. 2367 if (sig == SIGINT && VMError::is_error_reported()) { 2368 os::die(); 2369 } 2370 2371 os::signal_notify(sig); 2372 } 2373 2374 void* os::user_handler() { 2375 return CAST_FROM_FN_PTR(void*, UserHandler); 2376 } 2377 2378 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) { 2379 struct timespec ts; 2380 // Semaphore's are always associated with CLOCK_REALTIME 2381 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2382 // see unpackTime for discussion on overflow checking 2383 if (sec >= MAX_SECS) { 2384 ts.tv_sec += MAX_SECS; 2385 ts.tv_nsec = 0; 2386 } else { 2387 ts.tv_sec += sec; 2388 ts.tv_nsec += nsec; 2389 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2390 ts.tv_nsec -= NANOSECS_PER_SEC; 2391 ++ts.tv_sec; // note: this must be <= max_secs 2392 } 2393 } 2394 2395 return ts; 2396 } 2397 2398 extern "C" { 2399 typedef void (*sa_handler_t)(int); 2400 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2401 } 2402 2403 void* os::signal(int signal_number, void* handler) { 2404 struct sigaction sigAct, oldSigAct; 2405 2406 sigfillset(&(sigAct.sa_mask)); 2407 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2408 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2409 2410 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2411 // -1 means registration failed 2412 return (void *)-1; 2413 } 2414 2415 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2416 } 2417 2418 void os::signal_raise(int signal_number) { 2419 ::raise(signal_number); 2420 } 2421 2422 // The following code is moved from os.cpp for making this 2423 // code platform specific, which it is by its very nature. 2424 2425 // Will be modified when max signal is changed to be dynamic 2426 int os::sigexitnum_pd() { 2427 return NSIG; 2428 } 2429 2430 // a counter for each possible signal value 2431 static volatile jint pending_signals[NSIG+1] = { 0 }; 2432 2433 // Linux(POSIX) specific hand shaking semaphore. 2434 static sem_t sig_sem; 2435 static PosixSemaphore sr_semaphore; 2436 2437 void os::signal_init_pd() { 2438 // Initialize signal structures 2439 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2440 2441 // Initialize signal semaphore 2442 ::sem_init(&sig_sem, 0, 0); 2443 } 2444 2445 void os::signal_notify(int sig) { 2446 Atomic::inc(&pending_signals[sig]); 2447 ::sem_post(&sig_sem); 2448 } 2449 2450 static int check_pending_signals(bool wait) { 2451 Atomic::store(0, &sigint_count); 2452 for (;;) { 2453 for (int i = 0; i < NSIG + 1; i++) { 2454 jint n = pending_signals[i]; 2455 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2456 return i; 2457 } 2458 } 2459 if (!wait) { 2460 return -1; 2461 } 2462 JavaThread *thread = JavaThread::current(); 2463 ThreadBlockInVM tbivm(thread); 2464 2465 bool threadIsSuspended; 2466 do { 2467 thread->set_suspend_equivalent(); 2468 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2469 ::sem_wait(&sig_sem); 2470 2471 // were we externally suspended while we were waiting? 2472 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2473 if (threadIsSuspended) { 2474 // The semaphore has been incremented, but while we were waiting 2475 // another thread suspended us. We don't want to continue running 2476 // while suspended because that would surprise the thread that 2477 // suspended us. 2478 ::sem_post(&sig_sem); 2479 2480 thread->java_suspend_self(); 2481 } 2482 } while (threadIsSuspended); 2483 } 2484 } 2485 2486 int os::signal_lookup() { 2487 return check_pending_signals(false); 2488 } 2489 2490 int os::signal_wait() { 2491 return check_pending_signals(true); 2492 } 2493 2494 //////////////////////////////////////////////////////////////////////////////// 2495 // Virtual Memory 2496 2497 int os::vm_page_size() { 2498 // Seems redundant as all get out 2499 assert(os::Linux::page_size() != -1, "must call os::init"); 2500 return os::Linux::page_size(); 2501 } 2502 2503 // Solaris allocates memory by pages. 2504 int os::vm_allocation_granularity() { 2505 assert(os::Linux::page_size() != -1, "must call os::init"); 2506 return os::Linux::page_size(); 2507 } 2508 2509 // Rationale behind this function: 2510 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2511 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2512 // samples for JITted code. Here we create private executable mapping over the code cache 2513 // and then we can use standard (well, almost, as mapping can change) way to provide 2514 // info for the reporting script by storing timestamp and location of symbol 2515 void linux_wrap_code(char* base, size_t size) { 2516 static volatile jint cnt = 0; 2517 2518 if (!UseOprofile) { 2519 return; 2520 } 2521 2522 char buf[PATH_MAX+1]; 2523 int num = Atomic::add(1, &cnt); 2524 2525 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2526 os::get_temp_directory(), os::current_process_id(), num); 2527 unlink(buf); 2528 2529 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2530 2531 if (fd != -1) { 2532 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2533 if (rv != (off_t)-1) { 2534 if (::write(fd, "", 1) == 1) { 2535 mmap(base, size, 2536 PROT_READ|PROT_WRITE|PROT_EXEC, 2537 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2538 } 2539 } 2540 ::close(fd); 2541 unlink(buf); 2542 } 2543 } 2544 2545 static bool recoverable_mmap_error(int err) { 2546 // See if the error is one we can let the caller handle. This 2547 // list of errno values comes from JBS-6843484. I can't find a 2548 // Linux man page that documents this specific set of errno 2549 // values so while this list currently matches Solaris, it may 2550 // change as we gain experience with this failure mode. 2551 switch (err) { 2552 case EBADF: 2553 case EINVAL: 2554 case ENOTSUP: 2555 // let the caller deal with these errors 2556 return true; 2557 2558 default: 2559 // Any remaining errors on this OS can cause our reserved mapping 2560 // to be lost. That can cause confusion where different data 2561 // structures think they have the same memory mapped. The worst 2562 // scenario is if both the VM and a library think they have the 2563 // same memory mapped. 2564 return false; 2565 } 2566 } 2567 2568 static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2569 int err) { 2570 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2571 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec, 2572 os::strerror(err), err); 2573 } 2574 2575 static void warn_fail_commit_memory(char* addr, size_t size, 2576 size_t alignment_hint, bool exec, 2577 int err) { 2578 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2579 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, 2580 alignment_hint, exec, os::strerror(err), err); 2581 } 2582 2583 // NOTE: Linux kernel does not really reserve the pages for us. 2584 // All it does is to check if there are enough free pages 2585 // left at the time of mmap(). This could be a potential 2586 // problem. 2587 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2588 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2589 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2590 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2591 if (res != (uintptr_t) MAP_FAILED) { 2592 if (UseNUMAInterleaving) { 2593 numa_make_global(addr, size); 2594 } 2595 return 0; 2596 } 2597 2598 int err = errno; // save errno from mmap() call above 2599 2600 if (!recoverable_mmap_error(err)) { 2601 warn_fail_commit_memory(addr, size, exec, err); 2602 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2603 } 2604 2605 return err; 2606 } 2607 2608 bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2609 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2610 } 2611 2612 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2613 const char* mesg) { 2614 assert(mesg != NULL, "mesg must be specified"); 2615 int err = os::Linux::commit_memory_impl(addr, size, exec); 2616 if (err != 0) { 2617 // the caller wants all commit errors to exit with the specified mesg: 2618 warn_fail_commit_memory(addr, size, exec, err); 2619 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2620 } 2621 } 2622 2623 // Define MAP_HUGETLB here so we can build HotSpot on old systems. 2624 #ifndef MAP_HUGETLB 2625 #define MAP_HUGETLB 0x40000 2626 #endif 2627 2628 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2629 #ifndef MADV_HUGEPAGE 2630 #define MADV_HUGEPAGE 14 2631 #endif 2632 2633 int os::Linux::commit_memory_impl(char* addr, size_t size, 2634 size_t alignment_hint, bool exec) { 2635 int err = os::Linux::commit_memory_impl(addr, size, exec); 2636 if (err == 0) { 2637 realign_memory(addr, size, alignment_hint); 2638 } 2639 return err; 2640 } 2641 2642 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2643 bool exec) { 2644 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2645 } 2646 2647 void os::pd_commit_memory_or_exit(char* addr, size_t size, 2648 size_t alignment_hint, bool exec, 2649 const char* mesg) { 2650 assert(mesg != NULL, "mesg must be specified"); 2651 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2652 if (err != 0) { 2653 // the caller wants all commit errors to exit with the specified mesg: 2654 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2655 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2656 } 2657 } 2658 2659 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2660 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2661 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2662 // be supported or the memory may already be backed by huge pages. 2663 ::madvise(addr, bytes, MADV_HUGEPAGE); 2664 } 2665 } 2666 2667 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2668 // This method works by doing an mmap over an existing mmaping and effectively discarding 2669 // the existing pages. However it won't work for SHM-based large pages that cannot be 2670 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2671 // small pages on top of the SHM segment. This method always works for small pages, so we 2672 // allow that in any case. 2673 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2674 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2675 } 2676 } 2677 2678 void os::numa_make_global(char *addr, size_t bytes) { 2679 Linux::numa_interleave_memory(addr, bytes); 2680 } 2681 2682 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2683 // bind policy to MPOL_PREFERRED for the current thread. 2684 #define USE_MPOL_PREFERRED 0 2685 2686 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2687 // To make NUMA and large pages more robust when both enabled, we need to ease 2688 // the requirements on where the memory should be allocated. MPOL_BIND is the 2689 // default policy and it will force memory to be allocated on the specified 2690 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2691 // the specified node, but will not force it. Using this policy will prevent 2692 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2693 // free large pages. 2694 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2695 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2696 } 2697 2698 bool os::numa_topology_changed() { return false; } 2699 2700 size_t os::numa_get_groups_num() { 2701 // Return just the number of nodes in which it's possible to allocate memory 2702 // (in numa terminology, configured nodes). 2703 return Linux::numa_num_configured_nodes(); 2704 } 2705 2706 int os::numa_get_group_id() { 2707 int cpu_id = Linux::sched_getcpu(); 2708 if (cpu_id != -1) { 2709 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2710 if (lgrp_id != -1) { 2711 return lgrp_id; 2712 } 2713 } 2714 return 0; 2715 } 2716 2717 int os::Linux::get_existing_num_nodes() { 2718 size_t node; 2719 size_t highest_node_number = Linux::numa_max_node(); 2720 int num_nodes = 0; 2721 2722 // Get the total number of nodes in the system including nodes without memory. 2723 for (node = 0; node <= highest_node_number; node++) { 2724 if (isnode_in_existing_nodes(node)) { 2725 num_nodes++; 2726 } 2727 } 2728 return num_nodes; 2729 } 2730 2731 size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2732 size_t highest_node_number = Linux::numa_max_node(); 2733 size_t i = 0; 2734 2735 // Map all node ids in which is possible to allocate memory. Also nodes are 2736 // not always consecutively available, i.e. available from 0 to the highest 2737 // node number. 2738 for (size_t node = 0; node <= highest_node_number; node++) { 2739 if (Linux::isnode_in_configured_nodes(node)) { 2740 ids[i++] = node; 2741 } 2742 } 2743 return i; 2744 } 2745 2746 bool os::get_page_info(char *start, page_info* info) { 2747 return false; 2748 } 2749 2750 char *os::scan_pages(char *start, char* end, page_info* page_expected, 2751 page_info* page_found) { 2752 return end; 2753 } 2754 2755 2756 int os::Linux::sched_getcpu_syscall(void) { 2757 unsigned int cpu = 0; 2758 int retval = -1; 2759 2760 #if defined(IA32) 2761 #ifndef SYS_getcpu 2762 #define SYS_getcpu 318 2763 #endif 2764 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2765 #elif defined(AMD64) 2766 // Unfortunately we have to bring all these macros here from vsyscall.h 2767 // to be able to compile on old linuxes. 2768 #define __NR_vgetcpu 2 2769 #define VSYSCALL_START (-10UL << 20) 2770 #define VSYSCALL_SIZE 1024 2771 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2772 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2773 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2774 retval = vgetcpu(&cpu, NULL, NULL); 2775 #endif 2776 2777 return (retval == -1) ? retval : cpu; 2778 } 2779 2780 void os::Linux::sched_getcpu_init() { 2781 // sched_getcpu() should be in libc. 2782 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2783 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2784 2785 // If it's not, try a direct syscall. 2786 if (sched_getcpu() == -1) { 2787 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2788 (void*)&sched_getcpu_syscall)); 2789 } 2790 } 2791 2792 // Something to do with the numa-aware allocator needs these symbols 2793 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2794 extern "C" JNIEXPORT void numa_error(char *where) { } 2795 2796 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails 2797 // load symbol from base version instead. 2798 void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2799 void *f = dlvsym(handle, name, "libnuma_1.1"); 2800 if (f == NULL) { 2801 f = dlsym(handle, name); 2802 } 2803 return f; 2804 } 2805 2806 // Handle request to load libnuma symbol version 1.2 (API v2) only. 2807 // Return NULL if the symbol is not defined in this particular version. 2808 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) { 2809 return dlvsym(handle, name, "libnuma_1.2"); 2810 } 2811 2812 bool os::Linux::libnuma_init() { 2813 if (sched_getcpu() != -1) { // Requires sched_getcpu() support 2814 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2815 if (handle != NULL) { 2816 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2817 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2818 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2819 libnuma_dlsym(handle, "numa_max_node"))); 2820 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t, 2821 libnuma_dlsym(handle, "numa_num_configured_nodes"))); 2822 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2823 libnuma_dlsym(handle, "numa_available"))); 2824 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2825 libnuma_dlsym(handle, "numa_tonode_memory"))); 2826 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2827 libnuma_dlsym(handle, "numa_interleave_memory"))); 2828 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t, 2829 libnuma_v2_dlsym(handle, "numa_interleave_memory"))); 2830 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2831 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2832 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t, 2833 libnuma_dlsym(handle, "numa_bitmask_isbitset"))); 2834 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t, 2835 libnuma_dlsym(handle, "numa_distance"))); 2836 2837 if (numa_available() != -1) { 2838 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2839 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr")); 2840 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr")); 2841 // Create an index -> node mapping, since nodes are not always consecutive 2842 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2843 rebuild_nindex_to_node_map(); 2844 // Create a cpu -> node mapping 2845 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2846 rebuild_cpu_to_node_map(); 2847 return true; 2848 } 2849 } 2850 } 2851 return false; 2852 } 2853 2854 size_t os::Linux::default_guard_size(os::ThreadType thr_type) { 2855 // Creating guard page is very expensive. Java thread has HotSpot 2856 // guard pages, only enable glibc guard page for non-Java threads. 2857 // (Remember: compiler thread is a Java thread, too!) 2858 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size()); 2859 } 2860 2861 void os::Linux::rebuild_nindex_to_node_map() { 2862 int highest_node_number = Linux::numa_max_node(); 2863 2864 nindex_to_node()->clear(); 2865 for (int node = 0; node <= highest_node_number; node++) { 2866 if (Linux::isnode_in_existing_nodes(node)) { 2867 nindex_to_node()->append(node); 2868 } 2869 } 2870 } 2871 2872 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2873 // The table is later used in get_node_by_cpu(). 2874 void os::Linux::rebuild_cpu_to_node_map() { 2875 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2876 // in libnuma (possible values are starting from 16, 2877 // and continuing up with every other power of 2, but less 2878 // than the maximum number of CPUs supported by kernel), and 2879 // is a subject to change (in libnuma version 2 the requirements 2880 // are more reasonable) we'll just hardcode the number they use 2881 // in the library. 2882 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2883 2884 size_t cpu_num = processor_count(); 2885 size_t cpu_map_size = NCPUS / BitsPerCLong; 2886 size_t cpu_map_valid_size = 2887 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2888 2889 cpu_to_node()->clear(); 2890 cpu_to_node()->at_grow(cpu_num - 1); 2891 2892 size_t node_num = get_existing_num_nodes(); 2893 2894 int distance = 0; 2895 int closest_distance = INT_MAX; 2896 int closest_node = 0; 2897 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2898 for (size_t i = 0; i < node_num; i++) { 2899 // Check if node is configured (not a memory-less node). If it is not, find 2900 // the closest configured node. 2901 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) { 2902 closest_distance = INT_MAX; 2903 // Check distance from all remaining nodes in the system. Ignore distance 2904 // from itself and from another non-configured node. 2905 for (size_t m = 0; m < node_num; m++) { 2906 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) { 2907 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m)); 2908 // If a closest node is found, update. There is always at least one 2909 // configured node in the system so there is always at least one node 2910 // close. 2911 if (distance != 0 && distance < closest_distance) { 2912 closest_distance = distance; 2913 closest_node = nindex_to_node()->at(m); 2914 } 2915 } 2916 } 2917 } else { 2918 // Current node is already a configured node. 2919 closest_node = nindex_to_node()->at(i); 2920 } 2921 2922 // Get cpus from the original node and map them to the closest node. If node 2923 // is a configured node (not a memory-less node), then original node and 2924 // closest node are the same. 2925 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2926 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2927 if (cpu_map[j] != 0) { 2928 for (size_t k = 0; k < BitsPerCLong; k++) { 2929 if (cpu_map[j] & (1UL << k)) { 2930 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node); 2931 } 2932 } 2933 } 2934 } 2935 } 2936 } 2937 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2938 } 2939 2940 int os::Linux::get_node_by_cpu(int cpu_id) { 2941 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2942 return cpu_to_node()->at(cpu_id); 2943 } 2944 return -1; 2945 } 2946 2947 GrowableArray<int>* os::Linux::_cpu_to_node; 2948 GrowableArray<int>* os::Linux::_nindex_to_node; 2949 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2950 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2951 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2952 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes; 2953 os::Linux::numa_available_func_t os::Linux::_numa_available; 2954 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2955 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2956 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2; 2957 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2958 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset; 2959 os::Linux::numa_distance_func_t os::Linux::_numa_distance; 2960 unsigned long* os::Linux::_numa_all_nodes; 2961 struct bitmask* os::Linux::_numa_all_nodes_ptr; 2962 struct bitmask* os::Linux::_numa_nodes_ptr; 2963 2964 bool os::pd_uncommit_memory(char* addr, size_t size) { 2965 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2966 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2967 return res != (uintptr_t) MAP_FAILED; 2968 } 2969 2970 static address get_stack_commited_bottom(address bottom, size_t size) { 2971 address nbot = bottom; 2972 address ntop = bottom + size; 2973 2974 size_t page_sz = os::vm_page_size(); 2975 unsigned pages = size / page_sz; 2976 2977 unsigned char vec[1]; 2978 unsigned imin = 1, imax = pages + 1, imid; 2979 int mincore_return_value = 0; 2980 2981 assert(imin <= imax, "Unexpected page size"); 2982 2983 while (imin < imax) { 2984 imid = (imax + imin) / 2; 2985 nbot = ntop - (imid * page_sz); 2986 2987 // Use a trick with mincore to check whether the page is mapped or not. 2988 // mincore sets vec to 1 if page resides in memory and to 0 if page 2989 // is swapped output but if page we are asking for is unmapped 2990 // it returns -1,ENOMEM 2991 mincore_return_value = mincore(nbot, page_sz, vec); 2992 2993 if (mincore_return_value == -1) { 2994 // Page is not mapped go up 2995 // to find first mapped page 2996 if (errno != EAGAIN) { 2997 assert(errno == ENOMEM, "Unexpected mincore errno"); 2998 imax = imid; 2999 } 3000 } else { 3001 // Page is mapped go down 3002 // to find first not mapped page 3003 imin = imid + 1; 3004 } 3005 } 3006 3007 nbot = nbot + page_sz; 3008 3009 // Adjust stack bottom one page up if last checked page is not mapped 3010 if (mincore_return_value == -1) { 3011 nbot = nbot + page_sz; 3012 } 3013 3014 return nbot; 3015 } 3016 3017 3018 // Linux uses a growable mapping for the stack, and if the mapping for 3019 // the stack guard pages is not removed when we detach a thread the 3020 // stack cannot grow beyond the pages where the stack guard was 3021 // mapped. If at some point later in the process the stack expands to 3022 // that point, the Linux kernel cannot expand the stack any further 3023 // because the guard pages are in the way, and a segfault occurs. 3024 // 3025 // However, it's essential not to split the stack region by unmapping 3026 // a region (leaving a hole) that's already part of the stack mapping, 3027 // so if the stack mapping has already grown beyond the guard pages at 3028 // the time we create them, we have to truncate the stack mapping. 3029 // So, we need to know the extent of the stack mapping when 3030 // create_stack_guard_pages() is called. 3031 3032 // We only need this for stacks that are growable: at the time of 3033 // writing thread stacks don't use growable mappings (i.e. those 3034 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3035 // only applies to the main thread. 3036 3037 // If the (growable) stack mapping already extends beyond the point 3038 // where we're going to put our guard pages, truncate the mapping at 3039 // that point by munmap()ping it. This ensures that when we later 3040 // munmap() the guard pages we don't leave a hole in the stack 3041 // mapping. This only affects the main/initial thread 3042 3043 bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3044 if (os::Linux::is_initial_thread()) { 3045 // As we manually grow stack up to bottom inside create_attached_thread(), 3046 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3047 // we don't need to do anything special. 3048 // Check it first, before calling heavy function. 3049 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3050 unsigned char vec[1]; 3051 3052 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3053 // Fallback to slow path on all errors, including EAGAIN 3054 stack_extent = (uintptr_t) get_stack_commited_bottom( 3055 os::Linux::initial_thread_stack_bottom(), 3056 (size_t)addr - stack_extent); 3057 } 3058 3059 if (stack_extent < (uintptr_t)addr) { 3060 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3061 } 3062 } 3063 3064 return os::commit_memory(addr, size, !ExecMem); 3065 } 3066 3067 // If this is a growable mapping, remove the guard pages entirely by 3068 // munmap()ping them. If not, just call uncommit_memory(). This only 3069 // affects the main/initial thread, but guard against future OS changes 3070 // It's safe to always unmap guard pages for initial thread because we 3071 // always place it right after end of the mapped region 3072 3073 bool os::remove_stack_guard_pages(char* addr, size_t size) { 3074 uintptr_t stack_extent, stack_base; 3075 3076 if (os::Linux::is_initial_thread()) { 3077 return ::munmap(addr, size) == 0; 3078 } 3079 3080 return os::uncommit_memory(addr, size); 3081 } 3082 3083 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3084 // at 'requested_addr'. If there are existing memory mappings at the same 3085 // location, however, they will be overwritten. If 'fixed' is false, 3086 // 'requested_addr' is only treated as a hint, the return value may or 3087 // may not start from the requested address. Unlike Linux mmap(), this 3088 // function returns NULL to indicate failure. 3089 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3090 char * addr; 3091 int flags; 3092 3093 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3094 if (fixed) { 3095 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3096 flags |= MAP_FIXED; 3097 } 3098 3099 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3100 // touch an uncommitted page. Otherwise, the read/write might 3101 // succeed if we have enough swap space to back the physical page. 3102 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3103 flags, -1, 0); 3104 3105 return addr == MAP_FAILED ? NULL : addr; 3106 } 3107 3108 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address 3109 // (req_addr != NULL) or with a given alignment. 3110 // - bytes shall be a multiple of alignment. 3111 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3112 // - alignment sets the alignment at which memory shall be allocated. 3113 // It must be a multiple of allocation granularity. 3114 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3115 // req_addr or NULL. 3116 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) { 3117 3118 size_t extra_size = bytes; 3119 if (req_addr == NULL && alignment > 0) { 3120 extra_size += alignment; 3121 } 3122 3123 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE, 3124 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 3125 -1, 0); 3126 if (start == MAP_FAILED) { 3127 start = NULL; 3128 } else { 3129 if (req_addr != NULL) { 3130 if (start != req_addr) { 3131 ::munmap(start, extra_size); 3132 start = NULL; 3133 } 3134 } else { 3135 char* const start_aligned = align_up(start, alignment); 3136 char* const end_aligned = start_aligned + bytes; 3137 char* const end = start + extra_size; 3138 if (start_aligned > start) { 3139 ::munmap(start, start_aligned - start); 3140 } 3141 if (end_aligned < end) { 3142 ::munmap(end_aligned, end - end_aligned); 3143 } 3144 start = start_aligned; 3145 } 3146 } 3147 return start; 3148 } 3149 3150 static int anon_munmap(char * addr, size_t size) { 3151 return ::munmap(addr, size) == 0; 3152 } 3153 3154 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3155 size_t alignment_hint) { 3156 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3157 } 3158 3159 bool os::pd_release_memory(char* addr, size_t size) { 3160 return anon_munmap(addr, size); 3161 } 3162 3163 static bool linux_mprotect(char* addr, size_t size, int prot) { 3164 // Linux wants the mprotect address argument to be page aligned. 3165 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size()); 3166 3167 // According to SUSv3, mprotect() should only be used with mappings 3168 // established by mmap(), and mmap() always maps whole pages. Unaligned 3169 // 'addr' likely indicates problem in the VM (e.g. trying to change 3170 // protection of malloc'ed or statically allocated memory). Check the 3171 // caller if you hit this assert. 3172 assert(addr == bottom, "sanity check"); 3173 3174 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3175 return ::mprotect(bottom, size, prot) == 0; 3176 } 3177 3178 // Set protections specified 3179 bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3180 bool is_committed) { 3181 unsigned int p = 0; 3182 switch (prot) { 3183 case MEM_PROT_NONE: p = PROT_NONE; break; 3184 case MEM_PROT_READ: p = PROT_READ; break; 3185 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3186 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3187 default: 3188 ShouldNotReachHere(); 3189 } 3190 // is_committed is unused. 3191 return linux_mprotect(addr, bytes, p); 3192 } 3193 3194 bool os::guard_memory(char* addr, size_t size) { 3195 return linux_mprotect(addr, size, PROT_NONE); 3196 } 3197 3198 bool os::unguard_memory(char* addr, size_t size) { 3199 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3200 } 3201 3202 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3203 size_t page_size) { 3204 bool result = false; 3205 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3206 MAP_ANONYMOUS|MAP_PRIVATE, 3207 -1, 0); 3208 if (p != MAP_FAILED) { 3209 void *aligned_p = align_up(p, page_size); 3210 3211 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3212 3213 munmap(p, page_size * 2); 3214 } 3215 3216 if (warn && !result) { 3217 warning("TransparentHugePages is not supported by the operating system."); 3218 } 3219 3220 return result; 3221 } 3222 3223 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3224 bool result = false; 3225 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3226 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3227 -1, 0); 3228 3229 if (p != MAP_FAILED) { 3230 // We don't know if this really is a huge page or not. 3231 FILE *fp = fopen("/proc/self/maps", "r"); 3232 if (fp) { 3233 while (!feof(fp)) { 3234 char chars[257]; 3235 long x = 0; 3236 if (fgets(chars, sizeof(chars), fp)) { 3237 if (sscanf(chars, "%lx-%*x", &x) == 1 3238 && x == (long)p) { 3239 if (strstr (chars, "hugepage")) { 3240 result = true; 3241 break; 3242 } 3243 } 3244 } 3245 } 3246 fclose(fp); 3247 } 3248 munmap(p, page_size); 3249 } 3250 3251 if (warn && !result) { 3252 warning("HugeTLBFS is not supported by the operating system."); 3253 } 3254 3255 return result; 3256 } 3257 3258 // Set the coredump_filter bits to include largepages in core dump (bit 6) 3259 // 3260 // From the coredump_filter documentation: 3261 // 3262 // - (bit 0) anonymous private memory 3263 // - (bit 1) anonymous shared memory 3264 // - (bit 2) file-backed private memory 3265 // - (bit 3) file-backed shared memory 3266 // - (bit 4) ELF header pages in file-backed private memory areas (it is 3267 // effective only if the bit 2 is cleared) 3268 // - (bit 5) hugetlb private memory 3269 // - (bit 6) hugetlb shared memory 3270 // 3271 static void set_coredump_filter(void) { 3272 FILE *f; 3273 long cdm; 3274 3275 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3276 return; 3277 } 3278 3279 if (fscanf(f, "%lx", &cdm) != 1) { 3280 fclose(f); 3281 return; 3282 } 3283 3284 rewind(f); 3285 3286 if ((cdm & LARGEPAGES_BIT) == 0) { 3287 cdm |= LARGEPAGES_BIT; 3288 fprintf(f, "%#lx", cdm); 3289 } 3290 3291 fclose(f); 3292 } 3293 3294 // Large page support 3295 3296 static size_t _large_page_size = 0; 3297 3298 size_t os::Linux::find_large_page_size() { 3299 size_t large_page_size = 0; 3300 3301 // large_page_size on Linux is used to round up heap size. x86 uses either 3302 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3303 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3304 // page as large as 256M. 3305 // 3306 // Here we try to figure out page size by parsing /proc/meminfo and looking 3307 // for a line with the following format: 3308 // Hugepagesize: 2048 kB 3309 // 3310 // If we can't determine the value (e.g. /proc is not mounted, or the text 3311 // format has been changed), we'll use the largest page size supported by 3312 // the processor. 3313 3314 #ifndef ZERO 3315 large_page_size = 3316 AARCH64_ONLY(2 * M) 3317 AMD64_ONLY(2 * M) 3318 ARM32_ONLY(2 * M) 3319 IA32_ONLY(4 * M) 3320 IA64_ONLY(256 * M) 3321 PPC_ONLY(4 * M) 3322 S390_ONLY(1 * M) 3323 SPARC_ONLY(4 * M); 3324 #endif // ZERO 3325 3326 FILE *fp = fopen("/proc/meminfo", "r"); 3327 if (fp) { 3328 while (!feof(fp)) { 3329 int x = 0; 3330 char buf[16]; 3331 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3332 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3333 large_page_size = x * K; 3334 break; 3335 } 3336 } else { 3337 // skip to next line 3338 for (;;) { 3339 int ch = fgetc(fp); 3340 if (ch == EOF || ch == (int)'\n') break; 3341 } 3342 } 3343 } 3344 fclose(fp); 3345 } 3346 3347 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3348 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3349 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3350 proper_unit_for_byte_size(large_page_size)); 3351 } 3352 3353 return large_page_size; 3354 } 3355 3356 size_t os::Linux::setup_large_page_size() { 3357 _large_page_size = Linux::find_large_page_size(); 3358 const size_t default_page_size = (size_t)Linux::page_size(); 3359 if (_large_page_size > default_page_size) { 3360 _page_sizes[0] = _large_page_size; 3361 _page_sizes[1] = default_page_size; 3362 _page_sizes[2] = 0; 3363 } 3364 3365 return _large_page_size; 3366 } 3367 3368 bool os::Linux::setup_large_page_type(size_t page_size) { 3369 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3370 FLAG_IS_DEFAULT(UseSHM) && 3371 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3372 3373 // The type of large pages has not been specified by the user. 3374 3375 // Try UseHugeTLBFS and then UseSHM. 3376 UseHugeTLBFS = UseSHM = true; 3377 3378 // Don't try UseTransparentHugePages since there are known 3379 // performance issues with it turned on. This might change in the future. 3380 UseTransparentHugePages = false; 3381 } 3382 3383 if (UseTransparentHugePages) { 3384 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3385 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3386 UseHugeTLBFS = false; 3387 UseSHM = false; 3388 return true; 3389 } 3390 UseTransparentHugePages = false; 3391 } 3392 3393 if (UseHugeTLBFS) { 3394 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3395 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3396 UseSHM = false; 3397 return true; 3398 } 3399 UseHugeTLBFS = false; 3400 } 3401 3402 return UseSHM; 3403 } 3404 3405 void os::large_page_init() { 3406 if (!UseLargePages && 3407 !UseTransparentHugePages && 3408 !UseHugeTLBFS && 3409 !UseSHM) { 3410 // Not using large pages. 3411 return; 3412 } 3413 3414 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3415 // The user explicitly turned off large pages. 3416 // Ignore the rest of the large pages flags. 3417 UseTransparentHugePages = false; 3418 UseHugeTLBFS = false; 3419 UseSHM = false; 3420 return; 3421 } 3422 3423 size_t large_page_size = Linux::setup_large_page_size(); 3424 UseLargePages = Linux::setup_large_page_type(large_page_size); 3425 3426 set_coredump_filter(); 3427 } 3428 3429 #ifndef SHM_HUGETLB 3430 #define SHM_HUGETLB 04000 3431 #endif 3432 3433 #define shm_warning_format(format, ...) \ 3434 do { \ 3435 if (UseLargePages && \ 3436 (!FLAG_IS_DEFAULT(UseLargePages) || \ 3437 !FLAG_IS_DEFAULT(UseSHM) || \ 3438 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \ 3439 warning(format, __VA_ARGS__); \ 3440 } \ 3441 } while (0) 3442 3443 #define shm_warning(str) shm_warning_format("%s", str) 3444 3445 #define shm_warning_with_errno(str) \ 3446 do { \ 3447 int err = errno; \ 3448 shm_warning_format(str " (error = %d)", err); \ 3449 } while (0) 3450 3451 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) { 3452 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment"); 3453 3454 if (!is_aligned(alignment, SHMLBA)) { 3455 assert(false, "Code below assumes that alignment is at least SHMLBA aligned"); 3456 return NULL; 3457 } 3458 3459 // To ensure that we get 'alignment' aligned memory from shmat, 3460 // we pre-reserve aligned virtual memory and then attach to that. 3461 3462 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL); 3463 if (pre_reserved_addr == NULL) { 3464 // Couldn't pre-reserve aligned memory. 3465 shm_warning("Failed to pre-reserve aligned memory for shmat."); 3466 return NULL; 3467 } 3468 3469 // SHM_REMAP is needed to allow shmat to map over an existing mapping. 3470 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP); 3471 3472 if ((intptr_t)addr == -1) { 3473 int err = errno; 3474 shm_warning_with_errno("Failed to attach shared memory."); 3475 3476 assert(err != EACCES, "Unexpected error"); 3477 assert(err != EIDRM, "Unexpected error"); 3478 assert(err != EINVAL, "Unexpected error"); 3479 3480 // Since we don't know if the kernel unmapped the pre-reserved memory area 3481 // we can't unmap it, since that would potentially unmap memory that was 3482 // mapped from other threads. 3483 return NULL; 3484 } 3485 3486 return addr; 3487 } 3488 3489 static char* shmat_at_address(int shmid, char* req_addr) { 3490 if (!is_aligned(req_addr, SHMLBA)) { 3491 assert(false, "Requested address needs to be SHMLBA aligned"); 3492 return NULL; 3493 } 3494 3495 char* addr = (char*)shmat(shmid, req_addr, 0); 3496 3497 if ((intptr_t)addr == -1) { 3498 shm_warning_with_errno("Failed to attach shared memory."); 3499 return NULL; 3500 } 3501 3502 return addr; 3503 } 3504 3505 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) { 3506 // If a req_addr has been provided, we assume that the caller has already aligned the address. 3507 if (req_addr != NULL) { 3508 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size"); 3509 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment"); 3510 return shmat_at_address(shmid, req_addr); 3511 } 3512 3513 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically 3514 // return large page size aligned memory addresses when req_addr == NULL. 3515 // However, if the alignment is larger than the large page size, we have 3516 // to manually ensure that the memory returned is 'alignment' aligned. 3517 if (alignment > os::large_page_size()) { 3518 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size"); 3519 return shmat_with_alignment(shmid, bytes, alignment); 3520 } else { 3521 return shmat_at_address(shmid, NULL); 3522 } 3523 } 3524 3525 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3526 char* req_addr, bool exec) { 3527 // "exec" is passed in but not used. Creating the shared image for 3528 // the code cache doesn't have an SHM_X executable permission to check. 3529 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3530 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3531 assert(is_aligned(req_addr, alignment), "Unaligned address"); 3532 3533 if (!is_aligned(bytes, os::large_page_size())) { 3534 return NULL; // Fallback to small pages. 3535 } 3536 3537 // Create a large shared memory region to attach to based on size. 3538 // Currently, size is the total size of the heap. 3539 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3540 if (shmid == -1) { 3541 // Possible reasons for shmget failure: 3542 // 1. shmmax is too small for Java heap. 3543 // > check shmmax value: cat /proc/sys/kernel/shmmax 3544 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3545 // 2. not enough large page memory. 3546 // > check available large pages: cat /proc/meminfo 3547 // > increase amount of large pages: 3548 // echo new_value > /proc/sys/vm/nr_hugepages 3549 // Note 1: different Linux may use different name for this property, 3550 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3551 // Note 2: it's possible there's enough physical memory available but 3552 // they are so fragmented after a long run that they can't 3553 // coalesce into large pages. Try to reserve large pages when 3554 // the system is still "fresh". 3555 shm_warning_with_errno("Failed to reserve shared memory."); 3556 return NULL; 3557 } 3558 3559 // Attach to the region. 3560 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr); 3561 3562 // Remove shmid. If shmat() is successful, the actual shared memory segment 3563 // will be deleted when it's detached by shmdt() or when the process 3564 // terminates. If shmat() is not successful this will remove the shared 3565 // segment immediately. 3566 shmctl(shmid, IPC_RMID, NULL); 3567 3568 return addr; 3569 } 3570 3571 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3572 int error) { 3573 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3574 3575 bool warn_on_failure = UseLargePages && 3576 (!FLAG_IS_DEFAULT(UseLargePages) || 3577 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3578 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3579 3580 if (warn_on_failure) { 3581 char msg[128]; 3582 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3583 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3584 warning("%s", msg); 3585 } 3586 } 3587 3588 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3589 char* req_addr, 3590 bool exec) { 3591 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3592 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size"); 3593 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3594 3595 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3596 char* addr = (char*)::mmap(req_addr, bytes, prot, 3597 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3598 -1, 0); 3599 3600 if (addr == MAP_FAILED) { 3601 warn_on_large_pages_failure(req_addr, bytes, errno); 3602 return NULL; 3603 } 3604 3605 assert(is_aligned(addr, os::large_page_size()), "Must be"); 3606 3607 return addr; 3608 } 3609 3610 // Reserve memory using mmap(MAP_HUGETLB). 3611 // - bytes shall be a multiple of alignment. 3612 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3613 // - alignment sets the alignment at which memory shall be allocated. 3614 // It must be a multiple of allocation granularity. 3615 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3616 // req_addr or NULL. 3617 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3618 size_t alignment, 3619 char* req_addr, 3620 bool exec) { 3621 size_t large_page_size = os::large_page_size(); 3622 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3623 3624 assert(is_aligned(req_addr, alignment), "Must be"); 3625 assert(is_aligned(bytes, alignment), "Must be"); 3626 3627 // First reserve - but not commit - the address range in small pages. 3628 char* const start = anon_mmap_aligned(bytes, alignment, req_addr); 3629 3630 if (start == NULL) { 3631 return NULL; 3632 } 3633 3634 assert(is_aligned(start, alignment), "Must be"); 3635 3636 char* end = start + bytes; 3637 3638 // Find the regions of the allocated chunk that can be promoted to large pages. 3639 char* lp_start = align_up(start, large_page_size); 3640 char* lp_end = align_down(end, large_page_size); 3641 3642 size_t lp_bytes = lp_end - lp_start; 3643 3644 assert(is_aligned(lp_bytes, large_page_size), "Must be"); 3645 3646 if (lp_bytes == 0) { 3647 // The mapped region doesn't even span the start and the end of a large page. 3648 // Fall back to allocate a non-special area. 3649 ::munmap(start, end - start); 3650 return NULL; 3651 } 3652 3653 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3654 3655 void* result; 3656 3657 // Commit small-paged leading area. 3658 if (start != lp_start) { 3659 result = ::mmap(start, lp_start - start, prot, 3660 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3661 -1, 0); 3662 if (result == MAP_FAILED) { 3663 ::munmap(lp_start, end - lp_start); 3664 return NULL; 3665 } 3666 } 3667 3668 // Commit large-paged area. 3669 result = ::mmap(lp_start, lp_bytes, prot, 3670 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3671 -1, 0); 3672 if (result == MAP_FAILED) { 3673 warn_on_large_pages_failure(lp_start, lp_bytes, errno); 3674 // If the mmap above fails, the large pages region will be unmapped and we 3675 // have regions before and after with small pages. Release these regions. 3676 // 3677 // | mapped | unmapped | mapped | 3678 // ^ ^ ^ ^ 3679 // start lp_start lp_end end 3680 // 3681 ::munmap(start, lp_start - start); 3682 ::munmap(lp_end, end - lp_end); 3683 return NULL; 3684 } 3685 3686 // Commit small-paged trailing area. 3687 if (lp_end != end) { 3688 result = ::mmap(lp_end, end - lp_end, prot, 3689 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3690 -1, 0); 3691 if (result == MAP_FAILED) { 3692 ::munmap(start, lp_end - start); 3693 return NULL; 3694 } 3695 } 3696 3697 return start; 3698 } 3699 3700 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3701 size_t alignment, 3702 char* req_addr, 3703 bool exec) { 3704 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3705 assert(is_aligned(req_addr, alignment), "Must be"); 3706 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be"); 3707 assert(is_power_of_2(os::large_page_size()), "Must be"); 3708 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3709 3710 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3711 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3712 } else { 3713 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3714 } 3715 } 3716 3717 char* os::reserve_memory_special(size_t bytes, size_t alignment, 3718 char* req_addr, bool exec) { 3719 assert(UseLargePages, "only for large pages"); 3720 3721 char* addr; 3722 if (UseSHM) { 3723 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3724 } else { 3725 assert(UseHugeTLBFS, "must be"); 3726 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3727 } 3728 3729 if (addr != NULL) { 3730 if (UseNUMAInterleaving) { 3731 numa_make_global(addr, bytes); 3732 } 3733 3734 // The memory is committed 3735 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3736 } 3737 3738 return addr; 3739 } 3740 3741 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3742 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3743 return shmdt(base) == 0; 3744 } 3745 3746 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3747 return pd_release_memory(base, bytes); 3748 } 3749 3750 bool os::release_memory_special(char* base, size_t bytes) { 3751 bool res; 3752 if (MemTracker::tracking_level() > NMT_minimal) { 3753 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3754 res = os::Linux::release_memory_special_impl(base, bytes); 3755 if (res) { 3756 tkr.record((address)base, bytes); 3757 } 3758 3759 } else { 3760 res = os::Linux::release_memory_special_impl(base, bytes); 3761 } 3762 return res; 3763 } 3764 3765 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3766 assert(UseLargePages, "only for large pages"); 3767 bool res; 3768 3769 if (UseSHM) { 3770 res = os::Linux::release_memory_special_shm(base, bytes); 3771 } else { 3772 assert(UseHugeTLBFS, "must be"); 3773 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3774 } 3775 return res; 3776 } 3777 3778 size_t os::large_page_size() { 3779 return _large_page_size; 3780 } 3781 3782 // With SysV SHM the entire memory region must be allocated as shared 3783 // memory. 3784 // HugeTLBFS allows application to commit large page memory on demand. 3785 // However, when committing memory with HugeTLBFS fails, the region 3786 // that was supposed to be committed will lose the old reservation 3787 // and allow other threads to steal that memory region. Because of this 3788 // behavior we can't commit HugeTLBFS memory. 3789 bool os::can_commit_large_page_memory() { 3790 return UseTransparentHugePages; 3791 } 3792 3793 bool os::can_execute_large_page_memory() { 3794 return UseTransparentHugePages || UseHugeTLBFS; 3795 } 3796 3797 // Reserve memory at an arbitrary address, only if that area is 3798 // available (and not reserved for something else). 3799 3800 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3801 const int max_tries = 10; 3802 char* base[max_tries]; 3803 size_t size[max_tries]; 3804 const size_t gap = 0x000000; 3805 3806 // Assert only that the size is a multiple of the page size, since 3807 // that's all that mmap requires, and since that's all we really know 3808 // about at this low abstraction level. If we need higher alignment, 3809 // we can either pass an alignment to this method or verify alignment 3810 // in one of the methods further up the call chain. See bug 5044738. 3811 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3812 3813 // Repeatedly allocate blocks until the block is allocated at the 3814 // right spot. 3815 3816 // Linux mmap allows caller to pass an address as hint; give it a try first, 3817 // if kernel honors the hint then we can return immediately. 3818 char * addr = anon_mmap(requested_addr, bytes, false); 3819 if (addr == requested_addr) { 3820 return requested_addr; 3821 } 3822 3823 if (addr != NULL) { 3824 // mmap() is successful but it fails to reserve at the requested address 3825 anon_munmap(addr, bytes); 3826 } 3827 3828 int i; 3829 for (i = 0; i < max_tries; ++i) { 3830 base[i] = reserve_memory(bytes); 3831 3832 if (base[i] != NULL) { 3833 // Is this the block we wanted? 3834 if (base[i] == requested_addr) { 3835 size[i] = bytes; 3836 break; 3837 } 3838 3839 // Does this overlap the block we wanted? Give back the overlapped 3840 // parts and try again. 3841 3842 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3843 if (top_overlap >= 0 && (size_t)top_overlap < bytes) { 3844 unmap_memory(base[i], top_overlap); 3845 base[i] += top_overlap; 3846 size[i] = bytes - top_overlap; 3847 } else { 3848 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr; 3849 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) { 3850 unmap_memory(requested_addr, bottom_overlap); 3851 size[i] = bytes - bottom_overlap; 3852 } else { 3853 size[i] = bytes; 3854 } 3855 } 3856 } 3857 } 3858 3859 // Give back the unused reserved pieces. 3860 3861 for (int j = 0; j < i; ++j) { 3862 if (base[j] != NULL) { 3863 unmap_memory(base[j], size[j]); 3864 } 3865 } 3866 3867 if (i < max_tries) { 3868 return requested_addr; 3869 } else { 3870 return NULL; 3871 } 3872 } 3873 3874 size_t os::read(int fd, void *buf, unsigned int nBytes) { 3875 return ::read(fd, buf, nBytes); 3876 } 3877 3878 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) { 3879 return ::pread(fd, buf, nBytes, offset); 3880 } 3881 3882 // Short sleep, direct OS call. 3883 // 3884 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3885 // sched_yield(2) will actually give up the CPU: 3886 // 3887 // * Alone on this pariticular CPU, keeps running. 3888 // * Before the introduction of "skip_buddy" with "compat_yield" disabled 3889 // (pre 2.6.39). 3890 // 3891 // So calling this with 0 is an alternative. 3892 // 3893 void os::naked_short_sleep(jlong ms) { 3894 struct timespec req; 3895 3896 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3897 req.tv_sec = 0; 3898 if (ms > 0) { 3899 req.tv_nsec = (ms % 1000) * 1000000; 3900 } else { 3901 req.tv_nsec = 1; 3902 } 3903 3904 nanosleep(&req, NULL); 3905 3906 return; 3907 } 3908 3909 // Sleep forever; naked call to OS-specific sleep; use with CAUTION 3910 void os::infinite_sleep() { 3911 while (true) { // sleep forever ... 3912 ::sleep(100); // ... 100 seconds at a time 3913 } 3914 } 3915 3916 // Used to convert frequent JVM_Yield() to nops 3917 bool os::dont_yield() { 3918 return DontYieldALot; 3919 } 3920 3921 void os::naked_yield() { 3922 sched_yield(); 3923 } 3924 3925 //////////////////////////////////////////////////////////////////////////////// 3926 // thread priority support 3927 3928 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3929 // only supports dynamic priority, static priority must be zero. For real-time 3930 // applications, Linux supports SCHED_RR which allows static priority (1-99). 3931 // However, for large multi-threaded applications, SCHED_RR is not only slower 3932 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3933 // of 5 runs - Sep 2005). 3934 // 3935 // The following code actually changes the niceness of kernel-thread/LWP. It 3936 // has an assumption that setpriority() only modifies one kernel-thread/LWP, 3937 // not the entire user process, and user level threads are 1:1 mapped to kernel 3938 // threads. It has always been the case, but could change in the future. For 3939 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3940 // It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3941 3942 int os::java_to_os_priority[CriticalPriority + 1] = { 3943 19, // 0 Entry should never be used 3944 3945 4, // 1 MinPriority 3946 3, // 2 3947 2, // 3 3948 3949 1, // 4 3950 0, // 5 NormPriority 3951 -1, // 6 3952 3953 -2, // 7 3954 -3, // 8 3955 -4, // 9 NearMaxPriority 3956 3957 -5, // 10 MaxPriority 3958 3959 -5 // 11 CriticalPriority 3960 }; 3961 3962 static int prio_init() { 3963 if (ThreadPriorityPolicy == 1) { 3964 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3965 // if effective uid is not root. Perhaps, a more elegant way of doing 3966 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3967 if (geteuid() != 0) { 3968 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3969 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3970 } 3971 ThreadPriorityPolicy = 0; 3972 } 3973 } 3974 if (UseCriticalJavaThreadPriority) { 3975 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3976 } 3977 return 0; 3978 } 3979 3980 OSReturn os::set_native_priority(Thread* thread, int newpri) { 3981 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 3982 3983 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3984 return (ret == 0) ? OS_OK : OS_ERR; 3985 } 3986 3987 OSReturn os::get_native_priority(const Thread* const thread, 3988 int *priority_ptr) { 3989 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 3990 *priority_ptr = java_to_os_priority[NormPriority]; 3991 return OS_OK; 3992 } 3993 3994 errno = 0; 3995 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3996 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3997 } 3998 3999 // Hint to the underlying OS that a task switch would not be good. 4000 // Void return because it's a hint and can fail. 4001 void os::hint_no_preempt() {} 4002 4003 //////////////////////////////////////////////////////////////////////////////// 4004 // suspend/resume support 4005 4006 // The low-level signal-based suspend/resume support is a remnant from the 4007 // old VM-suspension that used to be for java-suspension, safepoints etc, 4008 // within hotspot. Currently used by JFR's OSThreadSampler 4009 // 4010 // The remaining code is greatly simplified from the more general suspension 4011 // code that used to be used. 4012 // 4013 // The protocol is quite simple: 4014 // - suspend: 4015 // - sends a signal to the target thread 4016 // - polls the suspend state of the osthread using a yield loop 4017 // - target thread signal handler (SR_handler) sets suspend state 4018 // and blocks in sigsuspend until continued 4019 // - resume: 4020 // - sets target osthread state to continue 4021 // - sends signal to end the sigsuspend loop in the SR_handler 4022 // 4023 // Note that the SR_lock plays no role in this suspend/resume protocol, 4024 // but is checked for NULL in SR_handler as a thread termination indicator. 4025 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs. 4026 // 4027 // Note that resume_clear_context() and suspend_save_context() are needed 4028 // by SR_handler(), so that fetch_frame_from_ucontext() works, 4029 // which in part is used by: 4030 // - Forte Analyzer: AsyncGetCallTrace() 4031 // - StackBanging: get_frame_at_stack_banging_point() 4032 4033 static void resume_clear_context(OSThread *osthread) { 4034 osthread->set_ucontext(NULL); 4035 osthread->set_siginfo(NULL); 4036 } 4037 4038 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 4039 ucontext_t* context) { 4040 osthread->set_ucontext(context); 4041 osthread->set_siginfo(siginfo); 4042 } 4043 4044 // Handler function invoked when a thread's execution is suspended or 4045 // resumed. We have to be careful that only async-safe functions are 4046 // called here (Note: most pthread functions are not async safe and 4047 // should be avoided.) 4048 // 4049 // Note: sigwait() is a more natural fit than sigsuspend() from an 4050 // interface point of view, but sigwait() prevents the signal hander 4051 // from being run. libpthread would get very confused by not having 4052 // its signal handlers run and prevents sigwait()'s use with the 4053 // mutex granting granting signal. 4054 // 4055 // Currently only ever called on the VMThread and JavaThreads (PC sampling) 4056 // 4057 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 4058 // Save and restore errno to avoid confusing native code with EINTR 4059 // after sigsuspend. 4060 int old_errno = errno; 4061 4062 Thread* thread = Thread::current_or_null_safe(); 4063 assert(thread != NULL, "Missing current thread in SR_handler"); 4064 4065 // On some systems we have seen signal delivery get "stuck" until the signal 4066 // mask is changed as part of thread termination. Check that the current thread 4067 // has not already terminated (via SR_lock()) - else the following assertion 4068 // will fail because the thread is no longer a JavaThread as the ~JavaThread 4069 // destructor has completed. 4070 4071 if (thread->SR_lock() == NULL) { 4072 return; 4073 } 4074 4075 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 4076 4077 OSThread* osthread = thread->osthread(); 4078 4079 os::SuspendResume::State current = osthread->sr.state(); 4080 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 4081 suspend_save_context(osthread, siginfo, context); 4082 4083 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4084 os::SuspendResume::State state = osthread->sr.suspended(); 4085 if (state == os::SuspendResume::SR_SUSPENDED) { 4086 sigset_t suspend_set; // signals for sigsuspend() 4087 sigemptyset(&suspend_set); 4088 // get current set of blocked signals and unblock resume signal 4089 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4090 sigdelset(&suspend_set, SR_signum); 4091 4092 sr_semaphore.signal(); 4093 // wait here until we are resumed 4094 while (1) { 4095 sigsuspend(&suspend_set); 4096 4097 os::SuspendResume::State result = osthread->sr.running(); 4098 if (result == os::SuspendResume::SR_RUNNING) { 4099 sr_semaphore.signal(); 4100 break; 4101 } 4102 } 4103 4104 } else if (state == os::SuspendResume::SR_RUNNING) { 4105 // request was cancelled, continue 4106 } else { 4107 ShouldNotReachHere(); 4108 } 4109 4110 resume_clear_context(osthread); 4111 } else if (current == os::SuspendResume::SR_RUNNING) { 4112 // request was cancelled, continue 4113 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4114 // ignore 4115 } else { 4116 // ignore 4117 } 4118 4119 errno = old_errno; 4120 } 4121 4122 static int SR_initialize() { 4123 struct sigaction act; 4124 char *s; 4125 4126 // Get signal number to use for suspend/resume 4127 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4128 int sig = ::strtol(s, 0, 10); 4129 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769. 4130 sig < NSIG) { // Must be legal signal and fit into sigflags[]. 4131 SR_signum = sig; 4132 } else { 4133 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.", 4134 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum); 4135 } 4136 } 4137 4138 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4139 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4140 4141 sigemptyset(&SR_sigset); 4142 sigaddset(&SR_sigset, SR_signum); 4143 4144 // Set up signal handler for suspend/resume 4145 act.sa_flags = SA_RESTART|SA_SIGINFO; 4146 act.sa_handler = (void (*)(int)) SR_handler; 4147 4148 // SR_signum is blocked by default. 4149 // 4528190 - We also need to block pthread restart signal (32 on all 4150 // supported Linux platforms). Note that LinuxThreads need to block 4151 // this signal for all threads to work properly. So we don't have 4152 // to use hard-coded signal number when setting up the mask. 4153 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4154 4155 if (sigaction(SR_signum, &act, 0) == -1) { 4156 return -1; 4157 } 4158 4159 // Save signal flag 4160 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4161 return 0; 4162 } 4163 4164 static int sr_notify(OSThread* osthread) { 4165 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4166 assert_status(status == 0, status, "pthread_kill"); 4167 return status; 4168 } 4169 4170 // "Randomly" selected value for how long we want to spin 4171 // before bailing out on suspending a thread, also how often 4172 // we send a signal to a thread we want to resume 4173 static const int RANDOMLY_LARGE_INTEGER = 1000000; 4174 static const int RANDOMLY_LARGE_INTEGER2 = 100; 4175 4176 // returns true on success and false on error - really an error is fatal 4177 // but this seems the normal response to library errors 4178 static bool do_suspend(OSThread* osthread) { 4179 assert(osthread->sr.is_running(), "thread should be running"); 4180 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4181 4182 // mark as suspended and send signal 4183 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4184 // failed to switch, state wasn't running? 4185 ShouldNotReachHere(); 4186 return false; 4187 } 4188 4189 if (sr_notify(osthread) != 0) { 4190 ShouldNotReachHere(); 4191 } 4192 4193 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4194 while (true) { 4195 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4196 break; 4197 } else { 4198 // timeout 4199 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4200 if (cancelled == os::SuspendResume::SR_RUNNING) { 4201 return false; 4202 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4203 // make sure that we consume the signal on the semaphore as well 4204 sr_semaphore.wait(); 4205 break; 4206 } else { 4207 ShouldNotReachHere(); 4208 return false; 4209 } 4210 } 4211 } 4212 4213 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4214 return true; 4215 } 4216 4217 static void do_resume(OSThread* osthread) { 4218 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4219 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4220 4221 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4222 // failed to switch to WAKEUP_REQUEST 4223 ShouldNotReachHere(); 4224 return; 4225 } 4226 4227 while (true) { 4228 if (sr_notify(osthread) == 0) { 4229 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4230 if (osthread->sr.is_running()) { 4231 return; 4232 } 4233 } 4234 } else { 4235 ShouldNotReachHere(); 4236 } 4237 } 4238 4239 guarantee(osthread->sr.is_running(), "Must be running!"); 4240 } 4241 4242 /////////////////////////////////////////////////////////////////////////////////// 4243 // signal handling (except suspend/resume) 4244 4245 // This routine may be used by user applications as a "hook" to catch signals. 4246 // The user-defined signal handler must pass unrecognized signals to this 4247 // routine, and if it returns true (non-zero), then the signal handler must 4248 // return immediately. If the flag "abort_if_unrecognized" is true, then this 4249 // routine will never retun false (zero), but instead will execute a VM panic 4250 // routine kill the process. 4251 // 4252 // If this routine returns false, it is OK to call it again. This allows 4253 // the user-defined signal handler to perform checks either before or after 4254 // the VM performs its own checks. Naturally, the user code would be making 4255 // a serious error if it tried to handle an exception (such as a null check 4256 // or breakpoint) that the VM was generating for its own correct operation. 4257 // 4258 // This routine may recognize any of the following kinds of signals: 4259 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4260 // It should be consulted by handlers for any of those signals. 4261 // 4262 // The caller of this routine must pass in the three arguments supplied 4263 // to the function referred to in the "sa_sigaction" (not the "sa_handler") 4264 // field of the structure passed to sigaction(). This routine assumes that 4265 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4266 // 4267 // Note that the VM will print warnings if it detects conflicting signal 4268 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4269 // 4270 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4271 siginfo_t* siginfo, 4272 void* ucontext, 4273 int abort_if_unrecognized); 4274 4275 void signalHandler(int sig, siginfo_t* info, void* uc) { 4276 assert(info != NULL && uc != NULL, "it must be old kernel"); 4277 int orig_errno = errno; // Preserve errno value over signal handler. 4278 JVM_handle_linux_signal(sig, info, uc, true); 4279 errno = orig_errno; 4280 } 4281 4282 4283 // This boolean allows users to forward their own non-matching signals 4284 // to JVM_handle_linux_signal, harmlessly. 4285 bool os::Linux::signal_handlers_are_installed = false; 4286 4287 // For signal-chaining 4288 struct sigaction sigact[NSIG]; 4289 uint64_t sigs = 0; 4290 #if (64 < NSIG-1) 4291 #error "Not all signals can be encoded in sigs. Adapt its type!" 4292 #endif 4293 bool os::Linux::libjsig_is_loaded = false; 4294 typedef struct sigaction *(*get_signal_t)(int); 4295 get_signal_t os::Linux::get_signal_action = NULL; 4296 4297 struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4298 struct sigaction *actp = NULL; 4299 4300 if (libjsig_is_loaded) { 4301 // Retrieve the old signal handler from libjsig 4302 actp = (*get_signal_action)(sig); 4303 } 4304 if (actp == NULL) { 4305 // Retrieve the preinstalled signal handler from jvm 4306 actp = get_preinstalled_handler(sig); 4307 } 4308 4309 return actp; 4310 } 4311 4312 static bool call_chained_handler(struct sigaction *actp, int sig, 4313 siginfo_t *siginfo, void *context) { 4314 // Call the old signal handler 4315 if (actp->sa_handler == SIG_DFL) { 4316 // It's more reasonable to let jvm treat it as an unexpected exception 4317 // instead of taking the default action. 4318 return false; 4319 } else if (actp->sa_handler != SIG_IGN) { 4320 if ((actp->sa_flags & SA_NODEFER) == 0) { 4321 // automaticlly block the signal 4322 sigaddset(&(actp->sa_mask), sig); 4323 } 4324 4325 sa_handler_t hand = NULL; 4326 sa_sigaction_t sa = NULL; 4327 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4328 // retrieve the chained handler 4329 if (siginfo_flag_set) { 4330 sa = actp->sa_sigaction; 4331 } else { 4332 hand = actp->sa_handler; 4333 } 4334 4335 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4336 actp->sa_handler = SIG_DFL; 4337 } 4338 4339 // try to honor the signal mask 4340 sigset_t oset; 4341 sigemptyset(&oset); 4342 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4343 4344 // call into the chained handler 4345 if (siginfo_flag_set) { 4346 (*sa)(sig, siginfo, context); 4347 } else { 4348 (*hand)(sig); 4349 } 4350 4351 // restore the signal mask 4352 pthread_sigmask(SIG_SETMASK, &oset, NULL); 4353 } 4354 // Tell jvm's signal handler the signal is taken care of. 4355 return true; 4356 } 4357 4358 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4359 bool chained = false; 4360 // signal-chaining 4361 if (UseSignalChaining) { 4362 struct sigaction *actp = get_chained_signal_action(sig); 4363 if (actp != NULL) { 4364 chained = call_chained_handler(actp, sig, siginfo, context); 4365 } 4366 } 4367 return chained; 4368 } 4369 4370 struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4371 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) { 4372 return &sigact[sig]; 4373 } 4374 return NULL; 4375 } 4376 4377 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4378 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4379 sigact[sig] = oldAct; 4380 sigs |= (uint64_t)1 << (sig-1); 4381 } 4382 4383 // for diagnostic 4384 int sigflags[NSIG]; 4385 4386 int os::Linux::get_our_sigflags(int sig) { 4387 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4388 return sigflags[sig]; 4389 } 4390 4391 void os::Linux::set_our_sigflags(int sig, int flags) { 4392 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4393 if (sig > 0 && sig < NSIG) { 4394 sigflags[sig] = flags; 4395 } 4396 } 4397 4398 void os::Linux::set_signal_handler(int sig, bool set_installed) { 4399 // Check for overwrite. 4400 struct sigaction oldAct; 4401 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4402 4403 void* oldhand = oldAct.sa_sigaction 4404 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4405 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4406 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4407 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4408 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4409 if (AllowUserSignalHandlers || !set_installed) { 4410 // Do not overwrite; user takes responsibility to forward to us. 4411 return; 4412 } else if (UseSignalChaining) { 4413 // save the old handler in jvm 4414 save_preinstalled_handler(sig, oldAct); 4415 // libjsig also interposes the sigaction() call below and saves the 4416 // old sigaction on it own. 4417 } else { 4418 fatal("Encountered unexpected pre-existing sigaction handler " 4419 "%#lx for signal %d.", (long)oldhand, sig); 4420 } 4421 } 4422 4423 struct sigaction sigAct; 4424 sigfillset(&(sigAct.sa_mask)); 4425 sigAct.sa_handler = SIG_DFL; 4426 if (!set_installed) { 4427 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4428 } else { 4429 sigAct.sa_sigaction = signalHandler; 4430 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4431 } 4432 // Save flags, which are set by ours 4433 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4434 sigflags[sig] = sigAct.sa_flags; 4435 4436 int ret = sigaction(sig, &sigAct, &oldAct); 4437 assert(ret == 0, "check"); 4438 4439 void* oldhand2 = oldAct.sa_sigaction 4440 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4441 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4442 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4443 } 4444 4445 // install signal handlers for signals that HotSpot needs to 4446 // handle in order to support Java-level exception handling. 4447 4448 void os::Linux::install_signal_handlers() { 4449 if (!signal_handlers_are_installed) { 4450 signal_handlers_are_installed = true; 4451 4452 // signal-chaining 4453 typedef void (*signal_setting_t)(); 4454 signal_setting_t begin_signal_setting = NULL; 4455 signal_setting_t end_signal_setting = NULL; 4456 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4457 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4458 if (begin_signal_setting != NULL) { 4459 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4460 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4461 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4462 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4463 libjsig_is_loaded = true; 4464 assert(UseSignalChaining, "should enable signal-chaining"); 4465 } 4466 if (libjsig_is_loaded) { 4467 // Tell libjsig jvm is setting signal handlers 4468 (*begin_signal_setting)(); 4469 } 4470 4471 set_signal_handler(SIGSEGV, true); 4472 set_signal_handler(SIGPIPE, true); 4473 set_signal_handler(SIGBUS, true); 4474 set_signal_handler(SIGILL, true); 4475 set_signal_handler(SIGFPE, true); 4476 #if defined(PPC64) 4477 set_signal_handler(SIGTRAP, true); 4478 #endif 4479 set_signal_handler(SIGXFSZ, true); 4480 4481 if (libjsig_is_loaded) { 4482 // Tell libjsig jvm finishes setting signal handlers 4483 (*end_signal_setting)(); 4484 } 4485 4486 // We don't activate signal checker if libjsig is in place, we trust ourselves 4487 // and if UserSignalHandler is installed all bets are off. 4488 // Log that signal checking is off only if -verbose:jni is specified. 4489 if (CheckJNICalls) { 4490 if (libjsig_is_loaded) { 4491 if (PrintJNIResolving) { 4492 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4493 } 4494 check_signals = false; 4495 } 4496 if (AllowUserSignalHandlers) { 4497 if (PrintJNIResolving) { 4498 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4499 } 4500 check_signals = false; 4501 } 4502 } 4503 } 4504 } 4505 4506 // This is the fastest way to get thread cpu time on Linux. 4507 // Returns cpu time (user+sys) for any thread, not only for current. 4508 // POSIX compliant clocks are implemented in the kernels 2.6.16+. 4509 // It might work on 2.6.10+ with a special kernel/glibc patch. 4510 // For reference, please, see IEEE Std 1003.1-2004: 4511 // http://www.unix.org/single_unix_specification 4512 4513 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4514 struct timespec tp; 4515 int rc = os::Linux::clock_gettime(clockid, &tp); 4516 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4517 4518 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4519 } 4520 4521 void os::Linux::initialize_os_info() { 4522 assert(_os_version == 0, "OS info already initialized"); 4523 4524 struct utsname _uname; 4525 4526 uint32_t major; 4527 uint32_t minor; 4528 uint32_t fix; 4529 4530 int rc; 4531 4532 // Kernel version is unknown if 4533 // verification below fails. 4534 _os_version = 0x01000000; 4535 4536 rc = uname(&_uname); 4537 if (rc != -1) { 4538 4539 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix); 4540 if (rc == 3) { 4541 4542 if (major < 256 && minor < 256 && fix < 256) { 4543 // Kernel version format is as expected, 4544 // set it overriding unknown state. 4545 _os_version = (major << 16) | 4546 (minor << 8 ) | 4547 (fix << 0 ) ; 4548 } 4549 } 4550 } 4551 } 4552 4553 uint32_t os::Linux::os_version() { 4554 assert(_os_version != 0, "not initialized"); 4555 return _os_version & 0x00FFFFFF; 4556 } 4557 4558 bool os::Linux::os_version_is_known() { 4559 assert(_os_version != 0, "not initialized"); 4560 return _os_version & 0x01000000 ? false : true; 4561 } 4562 4563 ///// 4564 // glibc on Linux platform uses non-documented flag 4565 // to indicate, that some special sort of signal 4566 // trampoline is used. 4567 // We will never set this flag, and we should 4568 // ignore this flag in our diagnostic 4569 #ifdef SIGNIFICANT_SIGNAL_MASK 4570 #undef SIGNIFICANT_SIGNAL_MASK 4571 #endif 4572 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4573 4574 static const char* get_signal_handler_name(address handler, 4575 char* buf, int buflen) { 4576 int offset = 0; 4577 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4578 if (found) { 4579 // skip directory names 4580 const char *p1, *p2; 4581 p1 = buf; 4582 size_t len = strlen(os::file_separator()); 4583 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4584 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4585 } else { 4586 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4587 } 4588 return buf; 4589 } 4590 4591 static void print_signal_handler(outputStream* st, int sig, 4592 char* buf, size_t buflen) { 4593 struct sigaction sa; 4594 4595 sigaction(sig, NULL, &sa); 4596 4597 // See comment for SIGNIFICANT_SIGNAL_MASK define 4598 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4599 4600 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4601 4602 address handler = (sa.sa_flags & SA_SIGINFO) 4603 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4604 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4605 4606 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4607 st->print("SIG_DFL"); 4608 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4609 st->print("SIG_IGN"); 4610 } else { 4611 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4612 } 4613 4614 st->print(", sa_mask[0]="); 4615 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4616 4617 address rh = VMError::get_resetted_sighandler(sig); 4618 // May be, handler was resetted by VMError? 4619 if (rh != NULL) { 4620 handler = rh; 4621 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4622 } 4623 4624 st->print(", sa_flags="); 4625 os::Posix::print_sa_flags(st, sa.sa_flags); 4626 4627 // Check: is it our handler? 4628 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4629 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4630 // It is our signal handler 4631 // check for flags, reset system-used one! 4632 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4633 st->print( 4634 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4635 os::Linux::get_our_sigflags(sig)); 4636 } 4637 } 4638 st->cr(); 4639 } 4640 4641 4642 #define DO_SIGNAL_CHECK(sig) \ 4643 do { \ 4644 if (!sigismember(&check_signal_done, sig)) { \ 4645 os::Linux::check_signal_handler(sig); \ 4646 } \ 4647 } while (0) 4648 4649 // This method is a periodic task to check for misbehaving JNI applications 4650 // under CheckJNI, we can add any periodic checks here 4651 4652 void os::run_periodic_checks() { 4653 if (check_signals == false) return; 4654 4655 // SEGV and BUS if overridden could potentially prevent 4656 // generation of hs*.log in the event of a crash, debugging 4657 // such a case can be very challenging, so we absolutely 4658 // check the following for a good measure: 4659 DO_SIGNAL_CHECK(SIGSEGV); 4660 DO_SIGNAL_CHECK(SIGILL); 4661 DO_SIGNAL_CHECK(SIGFPE); 4662 DO_SIGNAL_CHECK(SIGBUS); 4663 DO_SIGNAL_CHECK(SIGPIPE); 4664 DO_SIGNAL_CHECK(SIGXFSZ); 4665 #if defined(PPC64) 4666 DO_SIGNAL_CHECK(SIGTRAP); 4667 #endif 4668 4669 // ReduceSignalUsage allows the user to override these handlers 4670 // see comments at the very top and jvm_solaris.h 4671 if (!ReduceSignalUsage) { 4672 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4673 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4674 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4675 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4676 } 4677 4678 DO_SIGNAL_CHECK(SR_signum); 4679 } 4680 4681 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4682 4683 static os_sigaction_t os_sigaction = NULL; 4684 4685 void os::Linux::check_signal_handler(int sig) { 4686 char buf[O_BUFLEN]; 4687 address jvmHandler = NULL; 4688 4689 4690 struct sigaction act; 4691 if (os_sigaction == NULL) { 4692 // only trust the default sigaction, in case it has been interposed 4693 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4694 if (os_sigaction == NULL) return; 4695 } 4696 4697 os_sigaction(sig, (struct sigaction*)NULL, &act); 4698 4699 4700 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4701 4702 address thisHandler = (act.sa_flags & SA_SIGINFO) 4703 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4704 : CAST_FROM_FN_PTR(address, act.sa_handler); 4705 4706 4707 switch (sig) { 4708 case SIGSEGV: 4709 case SIGBUS: 4710 case SIGFPE: 4711 case SIGPIPE: 4712 case SIGILL: 4713 case SIGXFSZ: 4714 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4715 break; 4716 4717 case SHUTDOWN1_SIGNAL: 4718 case SHUTDOWN2_SIGNAL: 4719 case SHUTDOWN3_SIGNAL: 4720 case BREAK_SIGNAL: 4721 jvmHandler = (address)user_handler(); 4722 break; 4723 4724 default: 4725 if (sig == SR_signum) { 4726 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4727 } else { 4728 return; 4729 } 4730 break; 4731 } 4732 4733 if (thisHandler != jvmHandler) { 4734 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4735 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4736 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4737 // No need to check this sig any longer 4738 sigaddset(&check_signal_done, sig); 4739 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4740 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4741 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4742 exception_name(sig, buf, O_BUFLEN)); 4743 } 4744 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4745 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4746 tty->print("expected:"); 4747 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig)); 4748 tty->cr(); 4749 tty->print(" found:"); 4750 os::Posix::print_sa_flags(tty, act.sa_flags); 4751 tty->cr(); 4752 // No need to check this sig any longer 4753 sigaddset(&check_signal_done, sig); 4754 } 4755 4756 // Dump all the signal 4757 if (sigismember(&check_signal_done, sig)) { 4758 print_signal_handlers(tty, buf, O_BUFLEN); 4759 } 4760 } 4761 4762 extern void report_error(char* file_name, int line_no, char* title, 4763 char* format, ...); 4764 4765 // this is called _before_ the most of global arguments have been parsed 4766 void os::init(void) { 4767 char dummy; // used to get a guess on initial stack address 4768 // first_hrtime = gethrtime(); 4769 4770 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4771 4772 init_random(1234567); 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 // retrieve entry point for pthread_setname_np 4792 Linux::_pthread_setname_np = 4793 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4794 4795 os::Posix::init(); 4796 } 4797 4798 // To install functions for atexit system call 4799 extern "C" { 4800 static void perfMemory_exit_helper() { 4801 perfMemory_exit(); 4802 } 4803 } 4804 4805 // this is called _after_ the global arguments have been parsed 4806 jint os::init_2(void) { 4807 4808 os::Posix::init_2(); 4809 4810 Linux::fast_thread_clock_init(); 4811 4812 // Allocate a single page and mark it as readable for safepoint polling 4813 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4814 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page"); 4815 4816 os::set_polling_page(polling_page); 4817 log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page)); 4818 4819 if (!UseMembar) { 4820 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4821 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4822 os::set_memory_serialize_page(mem_serialize_page); 4823 log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page)); 4824 } 4825 4826 // initialize suspend/resume support - must do this before signal_sets_init() 4827 if (SR_initialize() != 0) { 4828 perror("SR_initialize failed"); 4829 return JNI_ERR; 4830 } 4831 4832 Linux::signal_sets_init(); 4833 Linux::install_signal_handlers(); 4834 4835 // Check and sets minimum stack sizes against command line options 4836 if (Posix::set_minimum_stack_sizes() == JNI_ERR) { 4837 return JNI_ERR; 4838 } 4839 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4840 4841 #if defined(IA32) 4842 workaround_expand_exec_shield_cs_limit(); 4843 #endif 4844 4845 Linux::libpthread_init(); 4846 Linux::sched_getcpu_init(); 4847 log_info(os)("HotSpot is running with %s, %s", 4848 Linux::glibc_version(), Linux::libpthread_version()); 4849 4850 if (UseNUMA) { 4851 if (!Linux::libnuma_init()) { 4852 UseNUMA = false; 4853 } else { 4854 if ((Linux::numa_max_node() < 1)) { 4855 // There's only one node(they start from 0), disable NUMA. 4856 UseNUMA = false; 4857 } 4858 } 4859 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4860 // we can make the adaptive lgrp chunk resizing work. If the user specified 4861 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4862 // disable adaptive resizing. 4863 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4864 if (FLAG_IS_DEFAULT(UseNUMA)) { 4865 UseNUMA = false; 4866 } else { 4867 if (FLAG_IS_DEFAULT(UseLargePages) && 4868 FLAG_IS_DEFAULT(UseSHM) && 4869 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4870 UseLargePages = false; 4871 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) { 4872 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)"); 4873 UseAdaptiveSizePolicy = false; 4874 UseAdaptiveNUMAChunkSizing = false; 4875 } 4876 } 4877 } 4878 if (!UseNUMA && ForceNUMA) { 4879 UseNUMA = true; 4880 } 4881 } 4882 4883 if (MaxFDLimit) { 4884 // set the number of file descriptors to max. print out error 4885 // if getrlimit/setrlimit fails but continue regardless. 4886 struct rlimit nbr_files; 4887 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4888 if (status != 0) { 4889 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno)); 4890 } else { 4891 nbr_files.rlim_cur = nbr_files.rlim_max; 4892 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4893 if (status != 0) { 4894 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno)); 4895 } 4896 } 4897 } 4898 4899 // Initialize lock used to serialize thread creation (see os::create_thread) 4900 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4901 4902 // at-exit methods are called in the reverse order of their registration. 4903 // atexit functions are called on return from main or as a result of a 4904 // call to exit(3C). There can be only 32 of these functions registered 4905 // and atexit() does not set errno. 4906 4907 if (PerfAllowAtExitRegistration) { 4908 // only register atexit functions if PerfAllowAtExitRegistration is set. 4909 // atexit functions can be delayed until process exit time, which 4910 // can be problematic for embedded VM situations. Embedded VMs should 4911 // call DestroyJavaVM() to assure that VM resources are released. 4912 4913 // note: perfMemory_exit_helper atexit function may be removed in 4914 // the future if the appropriate cleanup code can be added to the 4915 // VM_Exit VMOperation's doit method. 4916 if (atexit(perfMemory_exit_helper) != 0) { 4917 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4918 } 4919 } 4920 4921 // initialize thread priority policy 4922 prio_init(); 4923 4924 return JNI_OK; 4925 } 4926 4927 // Mark the polling page as unreadable 4928 void os::make_polling_page_unreadable(void) { 4929 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 4930 fatal("Could not disable polling page"); 4931 } 4932 } 4933 4934 // Mark the polling page as readable 4935 void os::make_polling_page_readable(void) { 4936 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4937 fatal("Could not enable polling page"); 4938 } 4939 } 4940 4941 // older glibc versions don't have this macro (which expands to 4942 // an optimized bit-counting function) so we have to roll our own 4943 #ifndef CPU_COUNT 4944 4945 static int _cpu_count(const cpu_set_t* cpus) { 4946 int count = 0; 4947 // only look up to the number of configured processors 4948 for (int i = 0; i < os::processor_count(); i++) { 4949 if (CPU_ISSET(i, cpus)) { 4950 count++; 4951 } 4952 } 4953 return count; 4954 } 4955 4956 #define CPU_COUNT(cpus) _cpu_count(cpus) 4957 4958 #endif // CPU_COUNT 4959 4960 // Get the current number of available processors for this process. 4961 // This value can change at any time during a process's lifetime. 4962 // sched_getaffinity gives an accurate answer as it accounts for cpusets. 4963 // If it appears there may be more than 1024 processors then we do a 4964 // dynamic check - see 6515172 for details. 4965 // If anything goes wrong we fallback to returning the number of online 4966 // processors - which can be greater than the number available to the process. 4967 int os::active_processor_count() { 4968 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors 4969 cpu_set_t* cpus_p = &cpus; 4970 int cpus_size = sizeof(cpu_set_t); 4971 4972 int configured_cpus = processor_count(); // upper bound on available cpus 4973 int cpu_count = 0; 4974 4975 // old build platforms may not support dynamic cpu sets 4976 #ifdef CPU_ALLOC 4977 4978 // To enable easy testing of the dynamic path on different platforms we 4979 // introduce a diagnostic flag: UseCpuAllocPath 4980 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) { 4981 // kernel may use a mask bigger than cpu_set_t 4982 log_trace(os)("active_processor_count: using dynamic path %s" 4983 "- configured processors: %d", 4984 UseCpuAllocPath ? "(forced) " : "", 4985 configured_cpus); 4986 cpus_p = CPU_ALLOC(configured_cpus); 4987 if (cpus_p != NULL) { 4988 cpus_size = CPU_ALLOC_SIZE(configured_cpus); 4989 // zero it just to be safe 4990 CPU_ZERO_S(cpus_size, cpus_p); 4991 } 4992 else { 4993 // failed to allocate so fallback to online cpus 4994 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4995 log_trace(os)("active_processor_count: " 4996 "CPU_ALLOC failed (%s) - using " 4997 "online processor count: %d", 4998 os::strerror(errno), online_cpus); 4999 return online_cpus; 5000 } 5001 } 5002 else { 5003 log_trace(os)("active_processor_count: using static path - configured processors: %d", 5004 configured_cpus); 5005 } 5006 #else // CPU_ALLOC 5007 // these stubs won't be executed 5008 #define CPU_COUNT_S(size, cpus) -1 5009 #define CPU_FREE(cpus) 5010 5011 log_trace(os)("active_processor_count: only static path available - configured processors: %d", 5012 configured_cpus); 5013 #endif // CPU_ALLOC 5014 5015 // pid 0 means the current thread - which we have to assume represents the process 5016 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) { 5017 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5018 cpu_count = CPU_COUNT_S(cpus_size, cpus_p); 5019 } 5020 else { 5021 cpu_count = CPU_COUNT(cpus_p); 5022 } 5023 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count); 5024 } 5025 else { 5026 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN); 5027 warning("sched_getaffinity failed (%s)- using online processor count (%d) " 5028 "which may exceed available processors", os::strerror(errno), cpu_count); 5029 } 5030 5031 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5032 CPU_FREE(cpus_p); 5033 } 5034 5035 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check"); 5036 return cpu_count; 5037 } 5038 5039 void os::set_native_thread_name(const char *name) { 5040 if (Linux::_pthread_setname_np) { 5041 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 5042 snprintf(buf, sizeof(buf), "%s", name); 5043 buf[sizeof(buf) - 1] = '\0'; 5044 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 5045 // ERANGE should not happen; all other errors should just be ignored. 5046 assert(rc != ERANGE, "pthread_setname_np failed"); 5047 } 5048 } 5049 5050 bool os::distribute_processes(uint length, uint* distribution) { 5051 // Not yet implemented. 5052 return false; 5053 } 5054 5055 bool os::bind_to_processor(uint processor_id) { 5056 // Not yet implemented. 5057 return false; 5058 } 5059 5060 /// 5061 5062 void os::SuspendedThreadTask::internal_do_task() { 5063 if (do_suspend(_thread->osthread())) { 5064 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 5065 do_task(context); 5066 do_resume(_thread->osthread()); 5067 } 5068 } 5069 5070 //////////////////////////////////////////////////////////////////////////////// 5071 // debug support 5072 5073 bool os::find(address addr, outputStream* st) { 5074 Dl_info dlinfo; 5075 memset(&dlinfo, 0, sizeof(dlinfo)); 5076 if (dladdr(addr, &dlinfo) != 0) { 5077 st->print(PTR_FORMAT ": ", p2i(addr)); 5078 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5079 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname, 5080 p2i(addr) - p2i(dlinfo.dli_saddr)); 5081 } else if (dlinfo.dli_fbase != NULL) { 5082 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase)); 5083 } else { 5084 st->print("<absolute address>"); 5085 } 5086 if (dlinfo.dli_fname != NULL) { 5087 st->print(" in %s", dlinfo.dli_fname); 5088 } 5089 if (dlinfo.dli_fbase != NULL) { 5090 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase)); 5091 } 5092 st->cr(); 5093 5094 if (Verbose) { 5095 // decode some bytes around the PC 5096 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5097 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5098 address lowest = (address) dlinfo.dli_sname; 5099 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5100 if (begin < lowest) begin = lowest; 5101 Dl_info dlinfo2; 5102 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5103 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 5104 end = (address) dlinfo2.dli_saddr; 5105 } 5106 Disassembler::decode(begin, end, st); 5107 } 5108 return true; 5109 } 5110 return false; 5111 } 5112 5113 //////////////////////////////////////////////////////////////////////////////// 5114 // misc 5115 5116 // This does not do anything on Linux. This is basically a hook for being 5117 // able to use structured exception handling (thread-local exception filters) 5118 // on, e.g., Win32. 5119 void 5120 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method, 5121 JavaCallArguments* args, Thread* thread) { 5122 f(value, method, args, thread); 5123 } 5124 5125 void os::print_statistics() { 5126 } 5127 5128 bool os::message_box(const char* title, const char* message) { 5129 int i; 5130 fdStream err(defaultStream::error_fd()); 5131 for (i = 0; i < 78; i++) err.print_raw("="); 5132 err.cr(); 5133 err.print_raw_cr(title); 5134 for (i = 0; i < 78; i++) err.print_raw("-"); 5135 err.cr(); 5136 err.print_raw_cr(message); 5137 for (i = 0; i < 78; i++) err.print_raw("="); 5138 err.cr(); 5139 5140 char buf[16]; 5141 // Prevent process from exiting upon "read error" without consuming all CPU 5142 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5143 5144 return buf[0] == 'y' || buf[0] == 'Y'; 5145 } 5146 5147 int os::stat(const char *path, struct stat *sbuf) { 5148 char pathbuf[MAX_PATH]; 5149 if (strlen(path) > MAX_PATH - 1) { 5150 errno = ENAMETOOLONG; 5151 return -1; 5152 } 5153 os::native_path(strcpy(pathbuf, path)); 5154 return ::stat(pathbuf, sbuf); 5155 } 5156 5157 // Is a (classpath) directory empty? 5158 bool os::dir_is_empty(const char* path) { 5159 DIR *dir = NULL; 5160 struct dirent *ptr; 5161 5162 dir = opendir(path); 5163 if (dir == NULL) return true; 5164 5165 // Scan the directory 5166 bool result = true; 5167 char buf[sizeof(struct dirent) + MAX_PATH]; 5168 while (result && (ptr = ::readdir(dir)) != NULL) { 5169 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5170 result = false; 5171 } 5172 } 5173 closedir(dir); 5174 return result; 5175 } 5176 5177 // This code originates from JDK's sysOpen and open64_w 5178 // from src/solaris/hpi/src/system_md.c 5179 5180 int os::open(const char *path, int oflag, int mode) { 5181 if (strlen(path) > MAX_PATH - 1) { 5182 errno = ENAMETOOLONG; 5183 return -1; 5184 } 5185 5186 // All file descriptors that are opened in the Java process and not 5187 // specifically destined for a subprocess should have the close-on-exec 5188 // flag set. If we don't set it, then careless 3rd party native code 5189 // might fork and exec without closing all appropriate file descriptors 5190 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5191 // turn might: 5192 // 5193 // - cause end-of-file to fail to be detected on some file 5194 // descriptors, resulting in mysterious hangs, or 5195 // 5196 // - might cause an fopen in the subprocess to fail on a system 5197 // suffering from bug 1085341. 5198 // 5199 // (Yes, the default setting of the close-on-exec flag is a Unix 5200 // design flaw) 5201 // 5202 // See: 5203 // 1085341: 32-bit stdio routines should support file descriptors >255 5204 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5205 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5206 // 5207 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5208 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5209 // because it saves a system call and removes a small window where the flag 5210 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5211 // and we fall back to using FD_CLOEXEC (see below). 5212 #ifdef O_CLOEXEC 5213 oflag |= O_CLOEXEC; 5214 #endif 5215 5216 int fd = ::open64(path, oflag, mode); 5217 if (fd == -1) return -1; 5218 5219 //If the open succeeded, the file might still be a directory 5220 { 5221 struct stat64 buf64; 5222 int ret = ::fstat64(fd, &buf64); 5223 int st_mode = buf64.st_mode; 5224 5225 if (ret != -1) { 5226 if ((st_mode & S_IFMT) == S_IFDIR) { 5227 errno = EISDIR; 5228 ::close(fd); 5229 return -1; 5230 } 5231 } else { 5232 ::close(fd); 5233 return -1; 5234 } 5235 } 5236 5237 #ifdef FD_CLOEXEC 5238 // Validate that the use of the O_CLOEXEC flag on open above worked. 5239 // With recent kernels, we will perform this check exactly once. 5240 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5241 if (!O_CLOEXEC_is_known_to_work) { 5242 int flags = ::fcntl(fd, F_GETFD); 5243 if (flags != -1) { 5244 if ((flags & FD_CLOEXEC) != 0) 5245 O_CLOEXEC_is_known_to_work = 1; 5246 else 5247 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5248 } 5249 } 5250 #endif 5251 5252 return fd; 5253 } 5254 5255 5256 // create binary file, rewriting existing file if required 5257 int os::create_binary_file(const char* path, bool rewrite_existing) { 5258 int oflags = O_WRONLY | O_CREAT; 5259 if (!rewrite_existing) { 5260 oflags |= O_EXCL; 5261 } 5262 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5263 } 5264 5265 // return current position of file pointer 5266 jlong os::current_file_offset(int fd) { 5267 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5268 } 5269 5270 // move file pointer to the specified offset 5271 jlong os::seek_to_file_offset(int fd, jlong offset) { 5272 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5273 } 5274 5275 // This code originates from JDK's sysAvailable 5276 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5277 5278 int os::available(int fd, jlong *bytes) { 5279 jlong cur, end; 5280 int mode; 5281 struct stat64 buf64; 5282 5283 if (::fstat64(fd, &buf64) >= 0) { 5284 mode = buf64.st_mode; 5285 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5286 int n; 5287 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5288 *bytes = n; 5289 return 1; 5290 } 5291 } 5292 } 5293 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5294 return 0; 5295 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5296 return 0; 5297 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5298 return 0; 5299 } 5300 *bytes = end - cur; 5301 return 1; 5302 } 5303 5304 // Map a block of memory. 5305 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5306 char *addr, size_t bytes, bool read_only, 5307 bool allow_exec) { 5308 int prot; 5309 int flags = MAP_PRIVATE; 5310 5311 if (read_only) { 5312 prot = PROT_READ; 5313 } else { 5314 prot = PROT_READ | PROT_WRITE; 5315 } 5316 5317 if (allow_exec) { 5318 prot |= PROT_EXEC; 5319 } 5320 5321 if (addr != NULL) { 5322 flags |= MAP_FIXED; 5323 } 5324 5325 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5326 fd, file_offset); 5327 if (mapped_address == MAP_FAILED) { 5328 return NULL; 5329 } 5330 return mapped_address; 5331 } 5332 5333 5334 // Remap a block of memory. 5335 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5336 char *addr, size_t bytes, bool read_only, 5337 bool allow_exec) { 5338 // same as map_memory() on this OS 5339 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5340 allow_exec); 5341 } 5342 5343 5344 // Unmap a block of memory. 5345 bool os::pd_unmap_memory(char* addr, size_t bytes) { 5346 return munmap(addr, bytes) == 0; 5347 } 5348 5349 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5350 5351 static clockid_t thread_cpu_clockid(Thread* thread) { 5352 pthread_t tid = thread->osthread()->pthread_id(); 5353 clockid_t clockid; 5354 5355 // Get thread clockid 5356 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5357 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5358 return clockid; 5359 } 5360 5361 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5362 // are used by JVM M&M and JVMTI to get user+sys or user CPU time 5363 // of a thread. 5364 // 5365 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5366 // the fast estimate available on the platform. 5367 5368 jlong os::current_thread_cpu_time() { 5369 if (os::Linux::supports_fast_thread_cpu_time()) { 5370 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5371 } else { 5372 // return user + sys since the cost is the same 5373 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5374 } 5375 } 5376 5377 jlong os::thread_cpu_time(Thread* thread) { 5378 // consistent with what current_thread_cpu_time() returns 5379 if (os::Linux::supports_fast_thread_cpu_time()) { 5380 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5381 } else { 5382 return slow_thread_cpu_time(thread, true /* user + sys */); 5383 } 5384 } 5385 5386 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5387 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5388 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5389 } else { 5390 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5391 } 5392 } 5393 5394 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5395 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5396 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5397 } else { 5398 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5399 } 5400 } 5401 5402 // -1 on error. 5403 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5404 pid_t tid = thread->osthread()->thread_id(); 5405 char *s; 5406 char stat[2048]; 5407 int statlen; 5408 char proc_name[64]; 5409 int count; 5410 long sys_time, user_time; 5411 char cdummy; 5412 int idummy; 5413 long ldummy; 5414 FILE *fp; 5415 5416 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5417 fp = fopen(proc_name, "r"); 5418 if (fp == NULL) return -1; 5419 statlen = fread(stat, 1, 2047, fp); 5420 stat[statlen] = '\0'; 5421 fclose(fp); 5422 5423 // Skip pid and the command string. Note that we could be dealing with 5424 // weird command names, e.g. user could decide to rename java launcher 5425 // to "java 1.4.2 :)", then the stat file would look like 5426 // 1234 (java 1.4.2 :)) R ... ... 5427 // We don't really need to know the command string, just find the last 5428 // occurrence of ")" and then start parsing from there. See bug 4726580. 5429 s = strrchr(stat, ')'); 5430 if (s == NULL) return -1; 5431 5432 // Skip blank chars 5433 do { s++; } while (s && isspace(*s)); 5434 5435 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5436 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5437 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5438 &user_time, &sys_time); 5439 if (count != 13) return -1; 5440 if (user_sys_cpu_time) { 5441 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5442 } else { 5443 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5444 } 5445 } 5446 5447 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5448 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5449 info_ptr->may_skip_backward = false; // elapsed time not wall time 5450 info_ptr->may_skip_forward = false; // elapsed time not wall time 5451 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5452 } 5453 5454 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5455 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5456 info_ptr->may_skip_backward = false; // elapsed time not wall time 5457 info_ptr->may_skip_forward = false; // elapsed time not wall time 5458 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5459 } 5460 5461 bool os::is_thread_cpu_time_supported() { 5462 return true; 5463 } 5464 5465 // System loadavg support. Returns -1 if load average cannot be obtained. 5466 // Linux doesn't yet have a (official) notion of processor sets, 5467 // so just return the system wide load average. 5468 int os::loadavg(double loadavg[], int nelem) { 5469 return ::getloadavg(loadavg, nelem); 5470 } 5471 5472 void os::pause() { 5473 char filename[MAX_PATH]; 5474 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5475 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile); 5476 } else { 5477 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5478 } 5479 5480 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5481 if (fd != -1) { 5482 struct stat buf; 5483 ::close(fd); 5484 while (::stat(filename, &buf) == 0) { 5485 (void)::poll(NULL, 0, 100); 5486 } 5487 } else { 5488 jio_fprintf(stderr, 5489 "Could not open pause file '%s', continuing immediately.\n", filename); 5490 } 5491 } 5492 5493 extern char** environ; 5494 5495 // Run the specified command in a separate process. Return its exit value, 5496 // or -1 on failure (e.g. can't fork a new process). 5497 // Unlike system(), this function can be called from signal handler. It 5498 // doesn't block SIGINT et al. 5499 int os::fork_and_exec(char* cmd) { 5500 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5501 5502 pid_t pid = fork(); 5503 5504 if (pid < 0) { 5505 // fork failed 5506 return -1; 5507 5508 } else if (pid == 0) { 5509 // child process 5510 5511 execve("/bin/sh", (char* const*)argv, environ); 5512 5513 // execve failed 5514 _exit(-1); 5515 5516 } else { 5517 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5518 // care about the actual exit code, for now. 5519 5520 int status; 5521 5522 // Wait for the child process to exit. This returns immediately if 5523 // the child has already exited. */ 5524 while (waitpid(pid, &status, 0) < 0) { 5525 switch (errno) { 5526 case ECHILD: return 0; 5527 case EINTR: break; 5528 default: return -1; 5529 } 5530 } 5531 5532 if (WIFEXITED(status)) { 5533 // The child exited normally; get its exit code. 5534 return WEXITSTATUS(status); 5535 } else if (WIFSIGNALED(status)) { 5536 // The child exited because of a signal 5537 // The best value to return is 0x80 + signal number, 5538 // because that is what all Unix shells do, and because 5539 // it allows callers to distinguish between process exit and 5540 // process death by signal. 5541 return 0x80 + WTERMSIG(status); 5542 } else { 5543 // Unknown exit code; pass it through 5544 return status; 5545 } 5546 } 5547 } 5548 5549 // is_headless_jre() 5550 // 5551 // Test for the existence of xawt/libmawt.so or libawt_xawt.so 5552 // in order to report if we are running in a headless jre 5553 // 5554 // Since JDK8 xawt/libmawt.so was moved into the same directory 5555 // as libawt.so, and renamed libawt_xawt.so 5556 // 5557 bool os::is_headless_jre() { 5558 struct stat statbuf; 5559 char buf[MAXPATHLEN]; 5560 char libmawtpath[MAXPATHLEN]; 5561 const char *xawtstr = "/xawt/libmawt.so"; 5562 const char *new_xawtstr = "/libawt_xawt.so"; 5563 char *p; 5564 5565 // Get path to libjvm.so 5566 os::jvm_path(buf, sizeof(buf)); 5567 5568 // Get rid of libjvm.so 5569 p = strrchr(buf, '/'); 5570 if (p == NULL) { 5571 return false; 5572 } else { 5573 *p = '\0'; 5574 } 5575 5576 // Get rid of client or server 5577 p = strrchr(buf, '/'); 5578 if (p == NULL) { 5579 return false; 5580 } else { 5581 *p = '\0'; 5582 } 5583 5584 // check xawt/libmawt.so 5585 strcpy(libmawtpath, buf); 5586 strcat(libmawtpath, xawtstr); 5587 if (::stat(libmawtpath, &statbuf) == 0) return false; 5588 5589 // check libawt_xawt.so 5590 strcpy(libmawtpath, buf); 5591 strcat(libmawtpath, new_xawtstr); 5592 if (::stat(libmawtpath, &statbuf) == 0) return false; 5593 5594 return true; 5595 } 5596 5597 // Get the default path to the core file 5598 // Returns the length of the string 5599 int os::get_core_path(char* buffer, size_t bufferSize) { 5600 /* 5601 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 5602 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 5603 */ 5604 const int core_pattern_len = 129; 5605 char core_pattern[core_pattern_len] = {0}; 5606 5607 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 5608 if (core_pattern_file == -1) { 5609 return -1; 5610 } 5611 5612 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 5613 ::close(core_pattern_file); 5614 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') { 5615 return -1; 5616 } 5617 if (core_pattern[ret-1] == '\n') { 5618 core_pattern[ret-1] = '\0'; 5619 } else { 5620 core_pattern[ret] = '\0'; 5621 } 5622 5623 char *pid_pos = strstr(core_pattern, "%p"); 5624 int written; 5625 5626 if (core_pattern[0] == '/') { 5627 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern); 5628 } else { 5629 char cwd[PATH_MAX]; 5630 5631 const char* p = get_current_directory(cwd, PATH_MAX); 5632 if (p == NULL) { 5633 return -1; 5634 } 5635 5636 if (core_pattern[0] == '|') { 5637 written = jio_snprintf(buffer, bufferSize, 5638 "\"%s\" (or dumping to %s/core.%d)", 5639 &core_pattern[1], p, current_process_id()); 5640 } else { 5641 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 5642 } 5643 } 5644 5645 if (written < 0) { 5646 return -1; 5647 } 5648 5649 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 5650 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 5651 5652 if (core_uses_pid_file != -1) { 5653 char core_uses_pid = 0; 5654 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 5655 ::close(core_uses_pid_file); 5656 5657 if (core_uses_pid == '1') { 5658 jio_snprintf(buffer + written, bufferSize - written, 5659 ".%d", current_process_id()); 5660 } 5661 } 5662 } 5663 5664 return strlen(buffer); 5665 } 5666 5667 bool os::start_debugging(char *buf, int buflen) { 5668 int len = (int)strlen(buf); 5669 char *p = &buf[len]; 5670 5671 jio_snprintf(p, buflen-len, 5672 "\n\n" 5673 "Do you want to debug the problem?\n\n" 5674 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n" 5675 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n" 5676 "Otherwise, press RETURN to abort...", 5677 os::current_process_id(), os::current_process_id(), 5678 os::current_thread_id(), os::current_thread_id()); 5679 5680 bool yes = os::message_box("Unexpected Error", buf); 5681 5682 if (yes) { 5683 // yes, user asked VM to launch debugger 5684 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d", 5685 os::current_process_id(), os::current_process_id()); 5686 5687 os::fork_and_exec(buf); 5688 yes = false; 5689 } 5690 return yes; 5691 } 5692 5693 5694 // Java/Compiler thread: 5695 // 5696 // Low memory addresses 5697 // P0 +------------------------+ 5698 // | |\ Java thread created by VM does not have glibc 5699 // | glibc guard page | - guard page, attached Java thread usually has 5700 // | |/ 1 glibc guard page. 5701 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 5702 // | |\ 5703 // | HotSpot Guard Pages | - red, yellow and reserved pages 5704 // | |/ 5705 // +------------------------+ JavaThread::stack_reserved_zone_base() 5706 // | |\ 5707 // | Normal Stack | - 5708 // | |/ 5709 // P2 +------------------------+ Thread::stack_base() 5710 // 5711 // Non-Java thread: 5712 // 5713 // Low memory addresses 5714 // P0 +------------------------+ 5715 // | |\ 5716 // | glibc guard page | - usually 1 page 5717 // | |/ 5718 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 5719 // | |\ 5720 // | Normal Stack | - 5721 // | |/ 5722 // P2 +------------------------+ Thread::stack_base() 5723 // 5724 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size 5725 // returned from pthread_attr_getstack(). 5726 // ** Due to NPTL implementation error, linux takes the glibc guard page out 5727 // of the stack size given in pthread_attr. We work around this for 5728 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.) 5729 // 5730 #ifndef ZERO 5731 static void current_stack_region(address * bottom, size_t * size) { 5732 if (os::Linux::is_initial_thread()) { 5733 // initial thread needs special handling because pthread_getattr_np() 5734 // may return bogus value. 5735 *bottom = os::Linux::initial_thread_stack_bottom(); 5736 *size = os::Linux::initial_thread_stack_size(); 5737 } else { 5738 pthread_attr_t attr; 5739 5740 int rslt = pthread_getattr_np(pthread_self(), &attr); 5741 5742 // JVM needs to know exact stack location, abort if it fails 5743 if (rslt != 0) { 5744 if (rslt == ENOMEM) { 5745 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np"); 5746 } else { 5747 fatal("pthread_getattr_np failed with error = %d", rslt); 5748 } 5749 } 5750 5751 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) { 5752 fatal("Cannot locate current stack attributes!"); 5753 } 5754 5755 // Work around NPTL stack guard error. 5756 size_t guard_size = 0; 5757 rslt = pthread_attr_getguardsize(&attr, &guard_size); 5758 if (rslt != 0) { 5759 fatal("pthread_attr_getguardsize failed with error = %d", rslt); 5760 } 5761 *bottom += guard_size; 5762 *size -= guard_size; 5763 5764 pthread_attr_destroy(&attr); 5765 5766 } 5767 assert(os::current_stack_pointer() >= *bottom && 5768 os::current_stack_pointer() < *bottom + *size, "just checking"); 5769 } 5770 5771 address os::current_stack_base() { 5772 address bottom; 5773 size_t size; 5774 current_stack_region(&bottom, &size); 5775 return (bottom + size); 5776 } 5777 5778 size_t os::current_stack_size() { 5779 // This stack size includes the usable stack and HotSpot guard pages 5780 // (for the threads that have Hotspot guard pages). 5781 address bottom; 5782 size_t size; 5783 current_stack_region(&bottom, &size); 5784 return size; 5785 } 5786 #endif 5787 5788 static inline struct timespec get_mtime(const char* filename) { 5789 struct stat st; 5790 int ret = os::stat(filename, &st); 5791 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno)); 5792 return st.st_mtim; 5793 } 5794 5795 int os::compare_file_modified_times(const char* file1, const char* file2) { 5796 struct timespec filetime1 = get_mtime(file1); 5797 struct timespec filetime2 = get_mtime(file2); 5798 int diff = filetime1.tv_sec - filetime2.tv_sec; 5799 if (diff == 0) { 5800 return filetime1.tv_nsec - filetime2.tv_nsec; 5801 } 5802 return diff; 5803 } 5804 5805 /////////////// Unit tests /////////////// 5806 5807 #ifndef PRODUCT 5808 5809 #define test_log(...) \ 5810 do { \ 5811 if (VerboseInternalVMTests) { \ 5812 tty->print_cr(__VA_ARGS__); \ 5813 tty->flush(); \ 5814 } \ 5815 } while (false) 5816 5817 class TestReserveMemorySpecial : AllStatic { 5818 public: 5819 static void small_page_write(void* addr, size_t size) { 5820 size_t page_size = os::vm_page_size(); 5821 5822 char* end = (char*)addr + size; 5823 for (char* p = (char*)addr; p < end; p += page_size) { 5824 *p = 1; 5825 } 5826 } 5827 5828 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 5829 if (!UseHugeTLBFS) { 5830 return; 5831 } 5832 5833 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 5834 5835 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 5836 5837 if (addr != NULL) { 5838 small_page_write(addr, size); 5839 5840 os::Linux::release_memory_special_huge_tlbfs(addr, size); 5841 } 5842 } 5843 5844 static void test_reserve_memory_special_huge_tlbfs_only() { 5845 if (!UseHugeTLBFS) { 5846 return; 5847 } 5848 5849 size_t lp = os::large_page_size(); 5850 5851 for (size_t size = lp; size <= lp * 10; size += lp) { 5852 test_reserve_memory_special_huge_tlbfs_only(size); 5853 } 5854 } 5855 5856 static void test_reserve_memory_special_huge_tlbfs_mixed() { 5857 size_t lp = os::large_page_size(); 5858 size_t ag = os::vm_allocation_granularity(); 5859 5860 // sizes to test 5861 const size_t sizes[] = { 5862 lp, lp + ag, lp + lp / 2, lp * 2, 5863 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2, 5864 lp * 10, lp * 10 + lp / 2 5865 }; 5866 const int num_sizes = sizeof(sizes) / sizeof(size_t); 5867 5868 // For each size/alignment combination, we test three scenarios: 5869 // 1) with req_addr == NULL 5870 // 2) with a non-null req_addr at which we expect to successfully allocate 5871 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we 5872 // expect the allocation to either fail or to ignore req_addr 5873 5874 // Pre-allocate two areas; they shall be as large as the largest allocation 5875 // and aligned to the largest alignment we will be testing. 5876 const size_t mapping_size = sizes[num_sizes - 1] * 2; 5877 char* const mapping1 = (char*) ::mmap(NULL, mapping_size, 5878 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 5879 -1, 0); 5880 assert(mapping1 != MAP_FAILED, "should work"); 5881 5882 char* const mapping2 = (char*) ::mmap(NULL, mapping_size, 5883 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 5884 -1, 0); 5885 assert(mapping2 != MAP_FAILED, "should work"); 5886 5887 // Unmap the first mapping, but leave the second mapping intact: the first 5888 // mapping will serve as a value for a "good" req_addr (case 2). The second 5889 // mapping, still intact, as "bad" req_addr (case 3). 5890 ::munmap(mapping1, mapping_size); 5891 5892 // Case 1 5893 test_log("%s, req_addr NULL:", __FUNCTION__); 5894 test_log("size align result"); 5895 5896 for (int i = 0; i < num_sizes; i++) { 5897 const size_t size = sizes[i]; 5898 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 5899 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 5900 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s", 5901 size, alignment, p2i(p), (p != NULL ? "" : "(failed)")); 5902 if (p != NULL) { 5903 assert(is_aligned(p, alignment), "must be"); 5904 small_page_write(p, size); 5905 os::Linux::release_memory_special_huge_tlbfs(p, size); 5906 } 5907 } 5908 } 5909 5910 // Case 2 5911 test_log("%s, req_addr non-NULL:", __FUNCTION__); 5912 test_log("size align req_addr result"); 5913 5914 for (int i = 0; i < num_sizes; i++) { 5915 const size_t size = sizes[i]; 5916 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 5917 char* const req_addr = align_up(mapping1, alignment); 5918 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 5919 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 5920 size, alignment, p2i(req_addr), p2i(p), 5921 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)"))); 5922 if (p != NULL) { 5923 assert(p == req_addr, "must be"); 5924 small_page_write(p, size); 5925 os::Linux::release_memory_special_huge_tlbfs(p, size); 5926 } 5927 } 5928 } 5929 5930 // Case 3 5931 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__); 5932 test_log("size align req_addr result"); 5933 5934 for (int i = 0; i < num_sizes; i++) { 5935 const size_t size = sizes[i]; 5936 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 5937 char* const req_addr = align_up(mapping2, alignment); 5938 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 5939 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 5940 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)"))); 5941 // as the area around req_addr contains already existing mappings, the API should always 5942 // return NULL (as per contract, it cannot return another address) 5943 assert(p == NULL, "must be"); 5944 } 5945 } 5946 5947 ::munmap(mapping2, mapping_size); 5948 5949 } 5950 5951 static void test_reserve_memory_special_huge_tlbfs() { 5952 if (!UseHugeTLBFS) { 5953 return; 5954 } 5955 5956 test_reserve_memory_special_huge_tlbfs_only(); 5957 test_reserve_memory_special_huge_tlbfs_mixed(); 5958 } 5959 5960 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 5961 if (!UseSHM) { 5962 return; 5963 } 5964 5965 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 5966 5967 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 5968 5969 if (addr != NULL) { 5970 assert(is_aligned(addr, alignment), "Check"); 5971 assert(is_aligned(addr, os::large_page_size()), "Check"); 5972 5973 small_page_write(addr, size); 5974 5975 os::Linux::release_memory_special_shm(addr, size); 5976 } 5977 } 5978 5979 static void test_reserve_memory_special_shm() { 5980 size_t lp = os::large_page_size(); 5981 size_t ag = os::vm_allocation_granularity(); 5982 5983 for (size_t size = ag; size < lp * 3; size += ag) { 5984 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 5985 test_reserve_memory_special_shm(size, alignment); 5986 } 5987 } 5988 } 5989 5990 static void test() { 5991 test_reserve_memory_special_huge_tlbfs(); 5992 test_reserve_memory_special_shm(); 5993 } 5994 }; 5995 5996 void TestReserveMemorySpecial_test() { 5997 TestReserveMemorySpecial::test(); 5998 } 5999 6000 #endif