1 /* 2 * Copyright (c) 1999, 2020, 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 "jvm.h" 27 #include "classfile/classLoader.hpp" 28 #include "classfile/systemDictionary.hpp" 29 #include "classfile/vmSymbols.hpp" 30 #include "code/icBuffer.hpp" 31 #include "code/vtableStubs.hpp" 32 #include "compiler/compileBroker.hpp" 33 #include "compiler/disassembler.hpp" 34 #include "interpreter/interpreter.hpp" 35 #include "logging/log.hpp" 36 #include "logging/logStream.hpp" 37 #include "memory/allocation.inline.hpp" 38 #include "memory/filemap.hpp" 39 #include "oops/oop.inline.hpp" 40 #include "os_linux.inline.hpp" 41 #include "os_posix.inline.hpp" 42 #include "os_share_linux.hpp" 43 #include "osContainer_linux.hpp" 44 #include "prims/jniFastGetField.hpp" 45 #include "prims/jvm_misc.hpp" 46 #include "runtime/arguments.hpp" 47 #include "runtime/atomic.hpp" 48 #include "runtime/extendedPC.hpp" 49 #include "runtime/globals.hpp" 50 #include "runtime/interfaceSupport.inline.hpp" 51 #include "runtime/init.hpp" 52 #include "runtime/java.hpp" 53 #include "runtime/javaCalls.hpp" 54 #include "runtime/mutexLocker.hpp" 55 #include "runtime/objectMonitor.hpp" 56 #include "runtime/osThread.hpp" 57 #include "runtime/perfMemory.hpp" 58 #include "runtime/sharedRuntime.hpp" 59 #include "runtime/statSampler.hpp" 60 #include "runtime/stubRoutines.hpp" 61 #include "runtime/thread.inline.hpp" 62 #include "runtime/threadCritical.hpp" 63 #include "runtime/threadSMR.hpp" 64 #include "runtime/timer.hpp" 65 #include "runtime/vm_version.hpp" 66 #include "semaphore_posix.hpp" 67 #include "services/attachListener.hpp" 68 #include "services/memTracker.hpp" 69 #include "services/runtimeService.hpp" 70 #include "utilities/align.hpp" 71 #include "utilities/decoder.hpp" 72 #include "utilities/defaultStream.hpp" 73 #include "utilities/events.hpp" 74 #include "utilities/elfFile.hpp" 75 #include "utilities/growableArray.hpp" 76 #include "utilities/macros.hpp" 77 #include "utilities/powerOfTwo.hpp" 78 #include "utilities/vmError.hpp" 79 80 // put OS-includes here 81 # include <sys/types.h> 82 # include <sys/mman.h> 83 # include <sys/stat.h> 84 # include <sys/select.h> 85 # include <pthread.h> 86 # include <signal.h> 87 # include <endian.h> 88 # include <errno.h> 89 # include <dlfcn.h> 90 # include <stdio.h> 91 # include <unistd.h> 92 # include <sys/resource.h> 93 # include <pthread.h> 94 # include <sys/stat.h> 95 # include <sys/time.h> 96 # include <sys/times.h> 97 # include <sys/utsname.h> 98 # include <sys/socket.h> 99 # include <sys/wait.h> 100 # include <pwd.h> 101 # include <poll.h> 102 # include <fcntl.h> 103 # include <string.h> 104 # include <syscall.h> 105 # include <sys/sysinfo.h> 106 # include <gnu/libc-version.h> 107 # include <sys/ipc.h> 108 # include <sys/shm.h> 109 # include <link.h> 110 # include <stdint.h> 111 # include <inttypes.h> 112 # include <sys/ioctl.h> 113 114 #ifndef _GNU_SOURCE 115 #define _GNU_SOURCE 116 #include <sched.h> 117 #undef _GNU_SOURCE 118 #else 119 #include <sched.h> 120 #endif 121 122 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 123 // getrusage() is prepared to handle the associated failure. 124 #ifndef RUSAGE_THREAD 125 #define RUSAGE_THREAD (1) /* only the calling thread */ 126 #endif 127 128 #define MAX_PATH (2 * K) 129 130 #define MAX_SECS 100000000 131 132 // for timer info max values which include all bits 133 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 134 135 enum CoredumpFilterBit { 136 FILE_BACKED_PVT_BIT = 1 << 2, 137 FILE_BACKED_SHARED_BIT = 1 << 3, 138 LARGEPAGES_BIT = 1 << 6, 139 DAX_SHARED_BIT = 1 << 8 140 }; 141 142 //////////////////////////////////////////////////////////////////////////////// 143 // global variables 144 julong os::Linux::_physical_memory = 0; 145 146 address os::Linux::_initial_thread_stack_bottom = NULL; 147 uintptr_t os::Linux::_initial_thread_stack_size = 0; 148 149 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 150 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL; 151 pthread_t os::Linux::_main_thread; 152 int os::Linux::_page_size = -1; 153 bool os::Linux::_supports_fast_thread_cpu_time = false; 154 const char * os::Linux::_glibc_version = NULL; 155 const char * os::Linux::_libpthread_version = NULL; 156 157 static jlong initial_time_count=0; 158 159 static int clock_tics_per_sec = 100; 160 161 // If the VM might have been created on the primordial thread, we need to resolve the 162 // primordial thread stack bounds and check if the current thread might be the 163 // primordial thread in places. If we know that the primordial thread is never used, 164 // such as when the VM was created by one of the standard java launchers, we can 165 // avoid this 166 static bool suppress_primordial_thread_resolution = false; 167 168 // For diagnostics to print a message once. see run_periodic_checks 169 static sigset_t check_signal_done; 170 static bool check_signals = true; 171 172 // Signal number used to suspend/resume a thread 173 174 // do not use any signal number less than SIGSEGV, see 4355769 175 static int SR_signum = SIGUSR2; 176 sigset_t SR_sigset; 177 178 // utility functions 179 180 static int SR_initialize(); 181 182 julong os::available_memory() { 183 return Linux::available_memory(); 184 } 185 186 julong os::Linux::available_memory() { 187 // values in struct sysinfo are "unsigned long" 188 struct sysinfo si; 189 julong avail_mem; 190 191 if (OSContainer::is_containerized()) { 192 jlong mem_limit, mem_usage; 193 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) { 194 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value", 195 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit); 196 } 197 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) { 198 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage); 199 } 200 if (mem_limit > 0 && mem_usage > 0 ) { 201 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0; 202 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem); 203 return avail_mem; 204 } 205 } 206 207 sysinfo(&si); 208 avail_mem = (julong)si.freeram * si.mem_unit; 209 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem); 210 return avail_mem; 211 } 212 213 julong os::physical_memory() { 214 jlong phys_mem = 0; 215 if (OSContainer::is_containerized()) { 216 jlong mem_limit; 217 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) { 218 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit); 219 return mem_limit; 220 } 221 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value", 222 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit); 223 } 224 225 phys_mem = Linux::physical_memory(); 226 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem); 227 return phys_mem; 228 } 229 230 static uint64_t initial_total_ticks = 0; 231 static uint64_t initial_steal_ticks = 0; 232 static bool has_initial_tick_info = false; 233 234 static void next_line(FILE *f) { 235 int c; 236 do { 237 c = fgetc(f); 238 } while (c != '\n' && c != EOF); 239 } 240 241 bool os::Linux::get_tick_information(CPUPerfTicks* pticks, int which_logical_cpu) { 242 FILE* fh; 243 uint64_t userTicks, niceTicks, systemTicks, idleTicks; 244 // since at least kernel 2.6 : iowait: time waiting for I/O to complete 245 // irq: time servicing interrupts; softirq: time servicing softirqs 246 uint64_t iowTicks = 0, irqTicks = 0, sirqTicks= 0; 247 // steal (since kernel 2.6.11): time spent in other OS when running in a virtualized environment 248 uint64_t stealTicks = 0; 249 // guest (since kernel 2.6.24): time spent running a virtual CPU for guest OS under the 250 // control of the Linux kernel 251 uint64_t guestNiceTicks = 0; 252 int logical_cpu = -1; 253 const int required_tickinfo_count = (which_logical_cpu == -1) ? 4 : 5; 254 int n; 255 256 memset(pticks, 0, sizeof(CPUPerfTicks)); 257 258 if ((fh = fopen("/proc/stat", "r")) == NULL) { 259 return false; 260 } 261 262 if (which_logical_cpu == -1) { 263 n = fscanf(fh, "cpu " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " 264 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " 265 UINT64_FORMAT " " UINT64_FORMAT " ", 266 &userTicks, &niceTicks, &systemTicks, &idleTicks, 267 &iowTicks, &irqTicks, &sirqTicks, 268 &stealTicks, &guestNiceTicks); 269 } else { 270 // Move to next line 271 next_line(fh); 272 273 // find the line for requested cpu faster to just iterate linefeeds? 274 for (int i = 0; i < which_logical_cpu; i++) { 275 next_line(fh); 276 } 277 278 n = fscanf(fh, "cpu%u " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " 279 UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " 280 UINT64_FORMAT " " UINT64_FORMAT " ", 281 &logical_cpu, &userTicks, &niceTicks, 282 &systemTicks, &idleTicks, &iowTicks, &irqTicks, &sirqTicks, 283 &stealTicks, &guestNiceTicks); 284 } 285 286 fclose(fh); 287 if (n < required_tickinfo_count || logical_cpu != which_logical_cpu) { 288 return false; 289 } 290 pticks->used = userTicks + niceTicks; 291 pticks->usedKernel = systemTicks + irqTicks + sirqTicks; 292 pticks->total = userTicks + niceTicks + systemTicks + idleTicks + 293 iowTicks + irqTicks + sirqTicks + stealTicks + guestNiceTicks; 294 295 if (n > required_tickinfo_count + 3) { 296 pticks->steal = stealTicks; 297 pticks->has_steal_ticks = true; 298 } else { 299 pticks->steal = 0; 300 pticks->has_steal_ticks = false; 301 } 302 303 return true; 304 } 305 306 // Return true if user is running as root. 307 308 bool os::have_special_privileges() { 309 static bool init = false; 310 static bool privileges = false; 311 if (!init) { 312 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 313 init = true; 314 } 315 return privileges; 316 } 317 318 319 #ifndef SYS_gettid 320 // i386: 224, ia64: 1105, amd64: 186, sparc 143 321 #ifdef __ia64__ 322 #define SYS_gettid 1105 323 #else 324 #ifdef __i386__ 325 #define SYS_gettid 224 326 #else 327 #ifdef __amd64__ 328 #define SYS_gettid 186 329 #else 330 #ifdef __sparc__ 331 #define SYS_gettid 143 332 #else 333 #error define gettid for the arch 334 #endif 335 #endif 336 #endif 337 #endif 338 #endif 339 340 341 // pid_t gettid() 342 // 343 // Returns the kernel thread id of the currently running thread. Kernel 344 // thread id is used to access /proc. 345 pid_t os::Linux::gettid() { 346 int rslt = syscall(SYS_gettid); 347 assert(rslt != -1, "must be."); // old linuxthreads implementation? 348 return (pid_t)rslt; 349 } 350 351 // Most versions of linux have a bug where the number of processors are 352 // determined by looking at the /proc file system. In a chroot environment, 353 // the system call returns 1. 354 static bool unsafe_chroot_detected = false; 355 static const char *unstable_chroot_error = "/proc file system not found.\n" 356 "Java may be unstable running multithreaded in a chroot " 357 "environment on Linux when /proc filesystem is not mounted."; 358 359 void os::Linux::initialize_system_info() { 360 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 361 if (processor_count() == 1) { 362 pid_t pid = os::Linux::gettid(); 363 char fname[32]; 364 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 365 FILE *fp = fopen(fname, "r"); 366 if (fp == NULL) { 367 unsafe_chroot_detected = true; 368 } else { 369 fclose(fp); 370 } 371 } 372 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 373 assert(processor_count() > 0, "linux error"); 374 } 375 376 void os::init_system_properties_values() { 377 // The next steps are taken in the product version: 378 // 379 // Obtain the JAVA_HOME value from the location of libjvm.so. 380 // This library should be located at: 381 // <JAVA_HOME>/lib/{client|server}/libjvm.so. 382 // 383 // If "/jre/lib/" appears at the right place in the path, then we 384 // assume libjvm.so is installed in a JDK and we use this path. 385 // 386 // Otherwise exit with message: "Could not create the Java virtual machine." 387 // 388 // The following extra steps are taken in the debugging version: 389 // 390 // If "/jre/lib/" does NOT appear at the right place in the path 391 // instead of exit check for $JAVA_HOME environment variable. 392 // 393 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 394 // then we append a fake suffix "hotspot/libjvm.so" to this path so 395 // it looks like libjvm.so is installed there 396 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 397 // 398 // Otherwise exit. 399 // 400 // Important note: if the location of libjvm.so changes this 401 // code needs to be changed accordingly. 402 403 // See ld(1): 404 // The linker uses the following search paths to locate required 405 // shared libraries: 406 // 1: ... 407 // ... 408 // 7: The default directories, normally /lib and /usr/lib. 409 #ifndef OVERRIDE_LIBPATH 410 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390) 411 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 412 #else 413 #define DEFAULT_LIBPATH "/lib:/usr/lib" 414 #endif 415 #else 416 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH 417 #endif 418 419 // Base path of extensions installed on the system. 420 #define SYS_EXT_DIR "/usr/java/packages" 421 #define EXTENSIONS_DIR "/lib/ext" 422 423 // Buffer that fits several sprintfs. 424 // Note that the space for the colon and the trailing null are provided 425 // by the nulls included by the sizeof operator. 426 const size_t bufsize = 427 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 428 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 429 char *buf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 430 431 // sysclasspath, java_home, dll_dir 432 { 433 char *pslash; 434 os::jvm_path(buf, bufsize); 435 436 // Found the full path to libjvm.so. 437 // Now cut the path to <java_home>/jre if we can. 438 pslash = strrchr(buf, '/'); 439 if (pslash != NULL) { 440 *pslash = '\0'; // Get rid of /libjvm.so. 441 } 442 pslash = strrchr(buf, '/'); 443 if (pslash != NULL) { 444 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 445 } 446 Arguments::set_dll_dir(buf); 447 448 if (pslash != NULL) { 449 pslash = strrchr(buf, '/'); 450 if (pslash != NULL) { 451 *pslash = '\0'; // Get rid of /lib. 452 } 453 } 454 Arguments::set_java_home(buf); 455 if (!set_boot_path('/', ':')) { 456 vm_exit_during_initialization("Failed setting boot class path.", NULL); 457 } 458 } 459 460 // Where to look for native libraries. 461 // 462 // Note: Due to a legacy implementation, most of the library path 463 // is set in the launcher. This was to accomodate linking restrictions 464 // on legacy Linux implementations (which are no longer supported). 465 // Eventually, all the library path setting will be done here. 466 // 467 // However, to prevent the proliferation of improperly built native 468 // libraries, the new path component /usr/java/packages is added here. 469 // Eventually, all the library path setting will be done here. 470 { 471 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 472 // should always exist (until the legacy problem cited above is 473 // addressed). 474 const char *v = ::getenv("LD_LIBRARY_PATH"); 475 const char *v_colon = ":"; 476 if (v == NULL) { v = ""; v_colon = ""; } 477 // That's +1 for the colon and +1 for the trailing '\0'. 478 char *ld_library_path = NEW_C_HEAP_ARRAY(char, 479 strlen(v) + 1 + 480 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1, 481 mtInternal); 482 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon); 483 Arguments::set_library_path(ld_library_path); 484 FREE_C_HEAP_ARRAY(char, ld_library_path); 485 } 486 487 // Extensions directories. 488 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 489 Arguments::set_ext_dirs(buf); 490 491 FREE_C_HEAP_ARRAY(char, buf); 492 493 #undef DEFAULT_LIBPATH 494 #undef SYS_EXT_DIR 495 #undef EXTENSIONS_DIR 496 } 497 498 //////////////////////////////////////////////////////////////////////////////// 499 // breakpoint support 500 501 void os::breakpoint() { 502 BREAKPOINT; 503 } 504 505 extern "C" void breakpoint() { 506 // use debugger to set breakpoint here 507 } 508 509 //////////////////////////////////////////////////////////////////////////////// 510 // signal support 511 512 debug_only(static bool signal_sets_initialized = false); 513 static sigset_t unblocked_sigs, vm_sigs; 514 515 void os::Linux::signal_sets_init() { 516 // Should also have an assertion stating we are still single-threaded. 517 assert(!signal_sets_initialized, "Already initialized"); 518 // Fill in signals that are necessarily unblocked for all threads in 519 // the VM. Currently, we unblock the following signals: 520 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 521 // by -Xrs (=ReduceSignalUsage)); 522 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 523 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 524 // the dispositions or masks wrt these signals. 525 // Programs embedding the VM that want to use the above signals for their 526 // own purposes must, at this time, use the "-Xrs" option to prevent 527 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 528 // (See bug 4345157, and other related bugs). 529 // In reality, though, unblocking these signals is really a nop, since 530 // these signals are not blocked by default. 531 sigemptyset(&unblocked_sigs); 532 sigaddset(&unblocked_sigs, SIGILL); 533 sigaddset(&unblocked_sigs, SIGSEGV); 534 sigaddset(&unblocked_sigs, SIGBUS); 535 sigaddset(&unblocked_sigs, SIGFPE); 536 #if defined(PPC64) 537 sigaddset(&unblocked_sigs, SIGTRAP); 538 #endif 539 sigaddset(&unblocked_sigs, SR_signum); 540 541 if (!ReduceSignalUsage) { 542 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 543 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 544 } 545 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 546 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 547 } 548 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 549 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 550 } 551 } 552 // Fill in signals that are blocked by all but the VM thread. 553 sigemptyset(&vm_sigs); 554 if (!ReduceSignalUsage) { 555 sigaddset(&vm_sigs, BREAK_SIGNAL); 556 } 557 debug_only(signal_sets_initialized = true); 558 559 } 560 561 // These are signals that are unblocked while a thread is running Java. 562 // (For some reason, they get blocked by default.) 563 sigset_t* os::Linux::unblocked_signals() { 564 assert(signal_sets_initialized, "Not initialized"); 565 return &unblocked_sigs; 566 } 567 568 // These are the signals that are blocked while a (non-VM) thread is 569 // running Java. Only the VM thread handles these signals. 570 sigset_t* os::Linux::vm_signals() { 571 assert(signal_sets_initialized, "Not initialized"); 572 return &vm_sigs; 573 } 574 575 void os::Linux::hotspot_sigmask(Thread* thread) { 576 577 //Save caller's signal mask before setting VM signal mask 578 sigset_t caller_sigmask; 579 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 580 581 OSThread* osthread = thread->osthread(); 582 osthread->set_caller_sigmask(caller_sigmask); 583 584 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 585 586 if (!ReduceSignalUsage) { 587 if (thread->is_VM_thread()) { 588 // Only the VM thread handles BREAK_SIGNAL ... 589 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 590 } else { 591 // ... all other threads block BREAK_SIGNAL 592 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 593 } 594 } 595 } 596 597 ////////////////////////////////////////////////////////////////////////////// 598 // detecting pthread library 599 600 void os::Linux::libpthread_init() { 601 // Save glibc and pthread version strings. 602 #if !defined(_CS_GNU_LIBC_VERSION) || \ 603 !defined(_CS_GNU_LIBPTHREAD_VERSION) 604 #error "glibc too old (< 2.3.2)" 605 #endif 606 607 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 608 assert(n > 0, "cannot retrieve glibc version"); 609 char *str = (char *)malloc(n, mtInternal); 610 confstr(_CS_GNU_LIBC_VERSION, str, n); 611 os::Linux::set_glibc_version(str); 612 613 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 614 assert(n > 0, "cannot retrieve pthread version"); 615 str = (char *)malloc(n, mtInternal); 616 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 617 os::Linux::set_libpthread_version(str); 618 } 619 620 ///////////////////////////////////////////////////////////////////////////// 621 // thread stack expansion 622 623 // os::Linux::manually_expand_stack() takes care of expanding the thread 624 // stack. Note that this is normally not needed: pthread stacks allocate 625 // thread stack using mmap() without MAP_NORESERVE, so the stack is already 626 // committed. Therefore it is not necessary to expand the stack manually. 627 // 628 // Manually expanding the stack was historically needed on LinuxThreads 629 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays 630 // it is kept to deal with very rare corner cases: 631 // 632 // For one, user may run the VM on an own implementation of threads 633 // whose stacks are - like the old LinuxThreads - implemented using 634 // mmap(MAP_GROWSDOWN). 635 // 636 // Also, this coding may be needed if the VM is running on the primordial 637 // thread. Normally we avoid running on the primordial thread; however, 638 // user may still invoke the VM on the primordial thread. 639 // 640 // The following historical comment describes the details about running 641 // on a thread stack allocated with mmap(MAP_GROWSDOWN): 642 643 644 // Force Linux kernel to expand current thread stack. If "bottom" is close 645 // to the stack guard, caller should block all signals. 646 // 647 // MAP_GROWSDOWN: 648 // A special mmap() flag that is used to implement thread stacks. It tells 649 // kernel that the memory region should extend downwards when needed. This 650 // allows early versions of LinuxThreads to only mmap the first few pages 651 // when creating a new thread. Linux kernel will automatically expand thread 652 // stack as needed (on page faults). 653 // 654 // However, because the memory region of a MAP_GROWSDOWN stack can grow on 655 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 656 // region, it's hard to tell if the fault is due to a legitimate stack 657 // access or because of reading/writing non-exist memory (e.g. buffer 658 // overrun). As a rule, if the fault happens below current stack pointer, 659 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the 660 // application (see Linux kernel fault.c). 661 // 662 // This Linux feature can cause SIGSEGV when VM bangs thread stack for 663 // stack overflow detection. 664 // 665 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 666 // not use MAP_GROWSDOWN. 667 // 668 // To get around the problem and allow stack banging on Linux, we need to 669 // manually expand thread stack after receiving the SIGSEGV. 670 // 671 // There are two ways to expand thread stack to address "bottom", we used 672 // both of them in JVM before 1.5: 673 // 1. adjust stack pointer first so that it is below "bottom", and then 674 // touch "bottom" 675 // 2. mmap() the page in question 676 // 677 // Now alternate signal stack is gone, it's harder to use 2. For instance, 678 // if current sp is already near the lower end of page 101, and we need to 679 // call mmap() to map page 100, it is possible that part of the mmap() frame 680 // will be placed in page 100. When page 100 is mapped, it is zero-filled. 681 // That will destroy the mmap() frame and cause VM to crash. 682 // 683 // The following code works by adjusting sp first, then accessing the "bottom" 684 // page to force a page fault. Linux kernel will then automatically expand the 685 // stack mapping. 686 // 687 // _expand_stack_to() assumes its frame size is less than page size, which 688 // should always be true if the function is not inlined. 689 690 static void NOINLINE _expand_stack_to(address bottom) { 691 address sp; 692 size_t size; 693 volatile char *p; 694 695 // Adjust bottom to point to the largest address within the same page, it 696 // gives us a one-page buffer if alloca() allocates slightly more memory. 697 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size()); 698 bottom += os::Linux::page_size() - 1; 699 700 // sp might be slightly above current stack pointer; if that's the case, we 701 // will alloca() a little more space than necessary, which is OK. Don't use 702 // os::current_stack_pointer(), as its result can be slightly below current 703 // stack pointer, causing us to not alloca enough to reach "bottom". 704 sp = (address)&sp; 705 706 if (sp > bottom) { 707 size = sp - bottom; 708 p = (volatile char *)alloca(size); 709 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 710 p[0] = '\0'; 711 } 712 } 713 714 void os::Linux::expand_stack_to(address bottom) { 715 _expand_stack_to(bottom); 716 } 717 718 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 719 assert(t!=NULL, "just checking"); 720 assert(t->osthread()->expanding_stack(), "expand should be set"); 721 assert(t->stack_base() != NULL, "stack_base was not initialized"); 722 723 if (t->is_in_usable_stack(addr)) { 724 sigset_t mask_all, old_sigset; 725 sigfillset(&mask_all); 726 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 727 _expand_stack_to(addr); 728 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 729 return true; 730 } 731 return false; 732 } 733 734 ////////////////////////////////////////////////////////////////////////////// 735 // create new thread 736 737 // Thread start routine for all newly created threads 738 static void *thread_native_entry(Thread *thread) { 739 740 thread->record_stack_base_and_size(); 741 742 // Try to randomize the cache line index of hot stack frames. 743 // This helps when threads of the same stack traces evict each other's 744 // cache lines. The threads can be either from the same JVM instance, or 745 // from different JVM instances. The benefit is especially true for 746 // processors with hyperthreading technology. 747 static int counter = 0; 748 int pid = os::current_process_id(); 749 alloca(((pid ^ counter++) & 7) * 128); 750 751 thread->initialize_thread_current(); 752 753 OSThread* osthread = thread->osthread(); 754 Monitor* sync = osthread->startThread_lock(); 755 756 osthread->set_thread_id(os::current_thread_id()); 757 758 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 759 os::current_thread_id(), (uintx) pthread_self()); 760 761 if (UseNUMA) { 762 int lgrp_id = os::numa_get_group_id(); 763 if (lgrp_id != -1) { 764 thread->set_lgrp_id(lgrp_id); 765 } 766 } 767 // initialize signal mask for this thread 768 os::Linux::hotspot_sigmask(thread); 769 770 // initialize floating point control register 771 os::Linux::init_thread_fpu_state(); 772 773 // handshaking with parent thread 774 { 775 MutexLocker ml(sync, Mutex::_no_safepoint_check_flag); 776 777 // notify parent thread 778 osthread->set_state(INITIALIZED); 779 sync->notify_all(); 780 781 // wait until os::start_thread() 782 while (osthread->get_state() == INITIALIZED) { 783 sync->wait_without_safepoint_check(); 784 } 785 } 786 787 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected"); 788 789 // call one more level start routine 790 thread->call_run(); 791 792 // Note: at this point the thread object may already have deleted itself. 793 // Prevent dereferencing it from here on out. 794 thread = NULL; 795 796 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 797 os::current_thread_id(), (uintx) pthread_self()); 798 799 return 0; 800 } 801 802 // On Linux, glibc places static TLS blocks (for __thread variables) on 803 // the thread stack. This decreases the stack size actually available 804 // to threads. 805 // 806 // For large static TLS sizes, this may cause threads to malfunction due 807 // to insufficient stack space. This is a well-known issue in glibc: 808 // http://sourceware.org/bugzilla/show_bug.cgi?id=11787. 809 // 810 // As a workaround, we call a private but assumed-stable glibc function, 811 // __pthread_get_minstack() to obtain the minstack size and derive the 812 // static TLS size from it. We then increase the user requested stack 813 // size by this TLS size. 814 // 815 // Due to compatibility concerns, this size adjustment is opt-in and 816 // controlled via AdjustStackSizeForTLS. 817 typedef size_t (*GetMinStack)(const pthread_attr_t *attr); 818 819 GetMinStack _get_minstack_func = NULL; 820 821 static void get_minstack_init() { 822 _get_minstack_func = 823 (GetMinStack)dlsym(RTLD_DEFAULT, "__pthread_get_minstack"); 824 log_info(os, thread)("Lookup of __pthread_get_minstack %s", 825 _get_minstack_func == NULL ? "failed" : "succeeded"); 826 } 827 828 // Returns the size of the static TLS area glibc puts on thread stacks. 829 // The value is cached on first use, which occurs when the first thread 830 // is created during VM initialization. 831 static size_t get_static_tls_area_size(const pthread_attr_t *attr) { 832 size_t tls_size = 0; 833 if (_get_minstack_func != NULL) { 834 // Obtain the pthread minstack size by calling __pthread_get_minstack. 835 size_t minstack_size = _get_minstack_func(attr); 836 837 // Remove non-TLS area size included in minstack size returned 838 // by __pthread_get_minstack() to get the static TLS size. 839 // In glibc before 2.27, minstack size includes guard_size. 840 // In glibc 2.27 and later, guard_size is automatically added 841 // to the stack size by pthread_create and is no longer included 842 // in minstack size. In both cases, the guard_size is taken into 843 // account, so there is no need to adjust the result for that. 844 // 845 // Although __pthread_get_minstack() is a private glibc function, 846 // it is expected to have a stable behavior across future glibc 847 // versions while glibc still allocates the static TLS blocks off 848 // the stack. Following is glibc 2.28 __pthread_get_minstack(): 849 // 850 // size_t 851 // __pthread_get_minstack (const pthread_attr_t *attr) 852 // { 853 // return GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN; 854 // } 855 // 856 // 857 // The following 'minstack_size > os::vm_page_size() + PTHREAD_STACK_MIN' 858 // if check is done for precaution. 859 if (minstack_size > (size_t)os::vm_page_size() + PTHREAD_STACK_MIN) { 860 tls_size = minstack_size - os::vm_page_size() - PTHREAD_STACK_MIN; 861 } 862 } 863 864 log_info(os, thread)("Stack size adjustment for TLS is " SIZE_FORMAT, 865 tls_size); 866 return tls_size; 867 } 868 869 bool os::create_thread(Thread* thread, ThreadType thr_type, 870 size_t req_stack_size) { 871 assert(thread->osthread() == NULL, "caller responsible"); 872 873 // Allocate the OSThread object 874 OSThread* osthread = new OSThread(NULL, NULL); 875 if (osthread == NULL) { 876 return false; 877 } 878 879 // set the correct thread state 880 osthread->set_thread_type(thr_type); 881 882 // Initial state is ALLOCATED but not INITIALIZED 883 osthread->set_state(ALLOCATED); 884 885 thread->set_osthread(osthread); 886 887 // init thread attributes 888 pthread_attr_t attr; 889 pthread_attr_init(&attr); 890 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 891 892 // Calculate stack size if it's not specified by caller. 893 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size); 894 // In glibc versions prior to 2.7 the guard size mechanism 895 // is not implemented properly. The posix standard requires adding 896 // the size of the guard pages to the stack size, instead Linux 897 // takes the space out of 'stacksize'. Thus we adapt the requested 898 // stack_size by the size of the guard pages to mimick proper 899 // behaviour. However, be careful not to end up with a size 900 // of zero due to overflow. Don't add the guard page in that case. 901 size_t guard_size = os::Linux::default_guard_size(thr_type); 902 // Configure glibc guard page. Must happen before calling 903 // get_static_tls_area_size(), which uses the guard_size. 904 pthread_attr_setguardsize(&attr, guard_size); 905 906 size_t stack_adjust_size = 0; 907 if (AdjustStackSizeForTLS) { 908 // Adjust the stack_size for on-stack TLS - see get_static_tls_area_size(). 909 stack_adjust_size += get_static_tls_area_size(&attr); 910 } else { 911 stack_adjust_size += guard_size; 912 } 913 914 stack_adjust_size = align_up(stack_adjust_size, os::vm_page_size()); 915 if (stack_size <= SIZE_MAX - stack_adjust_size) { 916 stack_size += stack_adjust_size; 917 } 918 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned"); 919 920 int status = pthread_attr_setstacksize(&attr, stack_size); 921 assert_status(status == 0, status, "pthread_attr_setstacksize"); 922 923 ThreadState state; 924 925 { 926 pthread_t tid; 927 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread); 928 929 char buf[64]; 930 if (ret == 0) { 931 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ", 932 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 933 } else { 934 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.", 935 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 936 // Log some OS information which might explain why creating the thread failed. 937 log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads()); 938 LogStream st(Log(os, thread)::info()); 939 os::Posix::print_rlimit_info(&st); 940 os::print_memory_info(&st); 941 os::Linux::print_proc_sys_info(&st); 942 os::Linux::print_container_info(&st); 943 } 944 945 pthread_attr_destroy(&attr); 946 947 if (ret != 0) { 948 // Need to clean up stuff we've allocated so far 949 thread->set_osthread(NULL); 950 delete osthread; 951 return false; 952 } 953 954 // Store pthread info into the OSThread 955 osthread->set_pthread_id(tid); 956 957 // Wait until child thread is either initialized or aborted 958 { 959 Monitor* sync_with_child = osthread->startThread_lock(); 960 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag); 961 while ((state = osthread->get_state()) == ALLOCATED) { 962 sync_with_child->wait_without_safepoint_check(); 963 } 964 } 965 } 966 967 // Aborted due to thread limit being reached 968 if (state == ZOMBIE) { 969 thread->set_osthread(NULL); 970 delete osthread; 971 return false; 972 } 973 974 // The thread is returned suspended (in state INITIALIZED), 975 // and is started higher up in the call chain 976 assert(state == INITIALIZED, "race condition"); 977 return true; 978 } 979 980 ///////////////////////////////////////////////////////////////////////////// 981 // attach existing thread 982 983 // bootstrap the main thread 984 bool os::create_main_thread(JavaThread* thread) { 985 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 986 return create_attached_thread(thread); 987 } 988 989 bool os::create_attached_thread(JavaThread* thread) { 990 #ifdef ASSERT 991 thread->verify_not_published(); 992 #endif 993 994 // Allocate the OSThread object 995 OSThread* osthread = new OSThread(NULL, NULL); 996 997 if (osthread == NULL) { 998 return false; 999 } 1000 1001 // Store pthread info into the OSThread 1002 osthread->set_thread_id(os::Linux::gettid()); 1003 osthread->set_pthread_id(::pthread_self()); 1004 1005 // initialize floating point control register 1006 os::Linux::init_thread_fpu_state(); 1007 1008 // Initial thread state is RUNNABLE 1009 osthread->set_state(RUNNABLE); 1010 1011 thread->set_osthread(osthread); 1012 1013 if (UseNUMA) { 1014 int lgrp_id = os::numa_get_group_id(); 1015 if (lgrp_id != -1) { 1016 thread->set_lgrp_id(lgrp_id); 1017 } 1018 } 1019 1020 if (os::is_primordial_thread()) { 1021 // If current thread is primordial thread, its stack is mapped on demand, 1022 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 1023 // the entire stack region to avoid SEGV in stack banging. 1024 // It is also useful to get around the heap-stack-gap problem on SuSE 1025 // kernel (see 4821821 for details). We first expand stack to the top 1026 // of yellow zone, then enable stack yellow zone (order is significant, 1027 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1028 // is no gap between the last two virtual memory regions. 1029 1030 JavaThread *jt = (JavaThread *)thread; 1031 address addr = jt->stack_reserved_zone_base(); 1032 assert(addr != NULL, "initialization problem?"); 1033 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1034 1035 osthread->set_expanding_stack(); 1036 os::Linux::manually_expand_stack(jt, addr); 1037 osthread->clear_expanding_stack(); 1038 } 1039 1040 // initialize signal mask for this thread 1041 // and save the caller's signal mask 1042 os::Linux::hotspot_sigmask(thread); 1043 1044 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 1045 os::current_thread_id(), (uintx) pthread_self()); 1046 1047 return true; 1048 } 1049 1050 void os::pd_start_thread(Thread* thread) { 1051 OSThread * osthread = thread->osthread(); 1052 assert(osthread->get_state() != INITIALIZED, "just checking"); 1053 Monitor* sync_with_child = osthread->startThread_lock(); 1054 MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1055 sync_with_child->notify(); 1056 } 1057 1058 // Free Linux resources related to the OSThread 1059 void os::free_thread(OSThread* osthread) { 1060 assert(osthread != NULL, "osthread not set"); 1061 1062 // We are told to free resources of the argument thread, 1063 // but we can only really operate on the current thread. 1064 assert(Thread::current()->osthread() == osthread, 1065 "os::free_thread but not current thread"); 1066 1067 #ifdef ASSERT 1068 sigset_t current; 1069 sigemptyset(¤t); 1070 pthread_sigmask(SIG_SETMASK, NULL, ¤t); 1071 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!"); 1072 #endif 1073 1074 // Restore caller's signal mask 1075 sigset_t sigmask = osthread->caller_sigmask(); 1076 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1077 1078 delete osthread; 1079 } 1080 1081 ////////////////////////////////////////////////////////////////////////////// 1082 // primordial thread 1083 1084 // Check if current thread is the primordial thread, similar to Solaris thr_main. 1085 bool os::is_primordial_thread(void) { 1086 if (suppress_primordial_thread_resolution) { 1087 return false; 1088 } 1089 char dummy; 1090 // If called before init complete, thread stack bottom will be null. 1091 // Can be called if fatal error occurs before initialization. 1092 if (os::Linux::initial_thread_stack_bottom() == NULL) return false; 1093 assert(os::Linux::initial_thread_stack_bottom() != NULL && 1094 os::Linux::initial_thread_stack_size() != 0, 1095 "os::init did not locate primordial thread's stack region"); 1096 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() && 1097 (address)&dummy < os::Linux::initial_thread_stack_bottom() + 1098 os::Linux::initial_thread_stack_size()) { 1099 return true; 1100 } else { 1101 return false; 1102 } 1103 } 1104 1105 // Find the virtual memory area that contains addr 1106 static bool find_vma(address addr, address* vma_low, address* vma_high) { 1107 FILE *fp = fopen("/proc/self/maps", "r"); 1108 if (fp) { 1109 address low, high; 1110 while (!feof(fp)) { 1111 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1112 if (low <= addr && addr < high) { 1113 if (vma_low) *vma_low = low; 1114 if (vma_high) *vma_high = high; 1115 fclose(fp); 1116 return true; 1117 } 1118 } 1119 for (;;) { 1120 int ch = fgetc(fp); 1121 if (ch == EOF || ch == (int)'\n') break; 1122 } 1123 } 1124 fclose(fp); 1125 } 1126 return false; 1127 } 1128 1129 // Locate primordial thread stack. This special handling of primordial thread stack 1130 // is needed because pthread_getattr_np() on most (all?) Linux distros returns 1131 // bogus value for the primordial process thread. While the launcher has created 1132 // the VM in a new thread since JDK 6, we still have to allow for the use of the 1133 // JNI invocation API from a primordial thread. 1134 void os::Linux::capture_initial_stack(size_t max_size) { 1135 1136 // max_size is either 0 (which means accept OS default for thread stacks) or 1137 // a user-specified value known to be at least the minimum needed. If we 1138 // are actually on the primordial thread we can make it appear that we have a 1139 // smaller max_size stack by inserting the guard pages at that location. But we 1140 // cannot do anything to emulate a larger stack than what has been provided by 1141 // the OS or threading library. In fact if we try to use a stack greater than 1142 // what is set by rlimit then we will crash the hosting process. 1143 1144 // Maximum stack size is the easy part, get it from RLIMIT_STACK. 1145 // If this is "unlimited" then it will be a huge value. 1146 struct rlimit rlim; 1147 getrlimit(RLIMIT_STACK, &rlim); 1148 size_t stack_size = rlim.rlim_cur; 1149 1150 // 6308388: a bug in ld.so will relocate its own .data section to the 1151 // lower end of primordial stack; reduce ulimit -s value a little bit 1152 // so we won't install guard page on ld.so's data section. 1153 // But ensure we don't underflow the stack size - allow 1 page spare 1154 if (stack_size >= (size_t)(3 * page_size())) { 1155 stack_size -= 2 * page_size(); 1156 } 1157 1158 // Try to figure out where the stack base (top) is. This is harder. 1159 // 1160 // When an application is started, glibc saves the initial stack pointer in 1161 // a global variable "__libc_stack_end", which is then used by system 1162 // libraries. __libc_stack_end should be pretty close to stack top. The 1163 // variable is available since the very early days. However, because it is 1164 // a private interface, it could disappear in the future. 1165 // 1166 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1167 // to __libc_stack_end, it is very close to stack top, but isn't the real 1168 // stack top. Note that /proc may not exist if VM is running as a chroot 1169 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1170 // /proc/<pid>/stat could change in the future (though unlikely). 1171 // 1172 // We try __libc_stack_end first. If that doesn't work, look for 1173 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1174 // as a hint, which should work well in most cases. 1175 1176 uintptr_t stack_start; 1177 1178 // try __libc_stack_end first 1179 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1180 if (p && *p) { 1181 stack_start = *p; 1182 } else { 1183 // see if we can get the start_stack field from /proc/self/stat 1184 FILE *fp; 1185 int pid; 1186 char state; 1187 int ppid; 1188 int pgrp; 1189 int session; 1190 int nr; 1191 int tpgrp; 1192 unsigned long flags; 1193 unsigned long minflt; 1194 unsigned long cminflt; 1195 unsigned long majflt; 1196 unsigned long cmajflt; 1197 unsigned long utime; 1198 unsigned long stime; 1199 long cutime; 1200 long cstime; 1201 long prio; 1202 long nice; 1203 long junk; 1204 long it_real; 1205 uintptr_t start; 1206 uintptr_t vsize; 1207 intptr_t rss; 1208 uintptr_t rsslim; 1209 uintptr_t scodes; 1210 uintptr_t ecode; 1211 int i; 1212 1213 // Figure what the primordial thread stack base is. Code is inspired 1214 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1215 // followed by command name surrounded by parentheses, state, etc. 1216 char stat[2048]; 1217 int statlen; 1218 1219 fp = fopen("/proc/self/stat", "r"); 1220 if (fp) { 1221 statlen = fread(stat, 1, 2047, fp); 1222 stat[statlen] = '\0'; 1223 fclose(fp); 1224 1225 // Skip pid and the command string. Note that we could be dealing with 1226 // weird command names, e.g. user could decide to rename java launcher 1227 // to "java 1.4.2 :)", then the stat file would look like 1228 // 1234 (java 1.4.2 :)) R ... ... 1229 // We don't really need to know the command string, just find the last 1230 // occurrence of ")" and then start parsing from there. See bug 4726580. 1231 char * s = strrchr(stat, ')'); 1232 1233 i = 0; 1234 if (s) { 1235 // Skip blank chars 1236 do { s++; } while (s && isspace(*s)); 1237 1238 #define _UFM UINTX_FORMAT 1239 #define _DFM INTX_FORMAT 1240 1241 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1242 // 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 1243 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, 1244 &state, // 3 %c 1245 &ppid, // 4 %d 1246 &pgrp, // 5 %d 1247 &session, // 6 %d 1248 &nr, // 7 %d 1249 &tpgrp, // 8 %d 1250 &flags, // 9 %lu 1251 &minflt, // 10 %lu 1252 &cminflt, // 11 %lu 1253 &majflt, // 12 %lu 1254 &cmajflt, // 13 %lu 1255 &utime, // 14 %lu 1256 &stime, // 15 %lu 1257 &cutime, // 16 %ld 1258 &cstime, // 17 %ld 1259 &prio, // 18 %ld 1260 &nice, // 19 %ld 1261 &junk, // 20 %ld 1262 &it_real, // 21 %ld 1263 &start, // 22 UINTX_FORMAT 1264 &vsize, // 23 UINTX_FORMAT 1265 &rss, // 24 INTX_FORMAT 1266 &rsslim, // 25 UINTX_FORMAT 1267 &scodes, // 26 UINTX_FORMAT 1268 &ecode, // 27 UINTX_FORMAT 1269 &stack_start); // 28 UINTX_FORMAT 1270 } 1271 1272 #undef _UFM 1273 #undef _DFM 1274 1275 if (i != 28 - 2) { 1276 assert(false, "Bad conversion from /proc/self/stat"); 1277 // product mode - assume we are the primordial thread, good luck in the 1278 // embedded case. 1279 warning("Can't detect primordial thread stack location - bad conversion"); 1280 stack_start = (uintptr_t) &rlim; 1281 } 1282 } else { 1283 // For some reason we can't open /proc/self/stat (for example, running on 1284 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1285 // most cases, so don't abort: 1286 warning("Can't detect primordial thread stack location - no /proc/self/stat"); 1287 stack_start = (uintptr_t) &rlim; 1288 } 1289 } 1290 1291 // Now we have a pointer (stack_start) very close to the stack top, the 1292 // next thing to do is to figure out the exact location of stack top. We 1293 // can find out the virtual memory area that contains stack_start by 1294 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1295 // and its upper limit is the real stack top. (again, this would fail if 1296 // running inside chroot, because /proc may not exist.) 1297 1298 uintptr_t stack_top; 1299 address low, high; 1300 if (find_vma((address)stack_start, &low, &high)) { 1301 // success, "high" is the true stack top. (ignore "low", because initial 1302 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1303 stack_top = (uintptr_t)high; 1304 } else { 1305 // failed, likely because /proc/self/maps does not exist 1306 warning("Can't detect primordial thread stack location - find_vma failed"); 1307 // best effort: stack_start is normally within a few pages below the real 1308 // stack top, use it as stack top, and reduce stack size so we won't put 1309 // guard page outside stack. 1310 stack_top = stack_start; 1311 stack_size -= 16 * page_size(); 1312 } 1313 1314 // stack_top could be partially down the page so align it 1315 stack_top = align_up(stack_top, page_size()); 1316 1317 // Allowed stack value is minimum of max_size and what we derived from rlimit 1318 if (max_size > 0) { 1319 _initial_thread_stack_size = MIN2(max_size, stack_size); 1320 } else { 1321 // Accept the rlimit max, but if stack is unlimited then it will be huge, so 1322 // clamp it at 8MB as we do on Solaris 1323 _initial_thread_stack_size = MIN2(stack_size, 8*M); 1324 } 1325 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size()); 1326 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1327 1328 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!"); 1329 1330 if (log_is_enabled(Info, os, thread)) { 1331 // See if we seem to be on primordial process thread 1332 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) && 1333 uintptr_t(&rlim) < stack_top; 1334 1335 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: " 1336 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT, 1337 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K, 1338 stack_top, intptr_t(_initial_thread_stack_bottom)); 1339 } 1340 } 1341 1342 //////////////////////////////////////////////////////////////////////////////// 1343 // time support 1344 1345 #ifndef SUPPORTS_CLOCK_MONOTONIC 1346 #error "Build platform doesn't support clock_gettime and related functionality" 1347 #endif 1348 1349 // Time since start-up in seconds to a fine granularity. 1350 // Used by VMSelfDestructTimer and the MemProfiler. 1351 double os::elapsedTime() { 1352 1353 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1354 } 1355 1356 jlong os::elapsed_counter() { 1357 return javaTimeNanos() - initial_time_count; 1358 } 1359 1360 jlong os::elapsed_frequency() { 1361 return NANOSECS_PER_SEC; // nanosecond resolution 1362 } 1363 1364 bool os::supports_vtime() { return true; } 1365 1366 double os::elapsedVTime() { 1367 struct rusage usage; 1368 int retval = getrusage(RUSAGE_THREAD, &usage); 1369 if (retval == 0) { 1370 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); 1371 } else { 1372 // better than nothing, but not much 1373 return elapsedTime(); 1374 } 1375 } 1376 1377 jlong os::javaTimeMillis() { 1378 timeval time; 1379 int status = gettimeofday(&time, NULL); 1380 assert(status != -1, "linux error"); 1381 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1382 } 1383 1384 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) { 1385 timeval time; 1386 int status = gettimeofday(&time, NULL); 1387 assert(status != -1, "linux error"); 1388 seconds = jlong(time.tv_sec); 1389 nanos = jlong(time.tv_usec) * 1000; 1390 } 1391 1392 void os::Linux::fast_thread_clock_init() { 1393 if (!UseLinuxPosixThreadCPUClocks) { 1394 return; 1395 } 1396 clockid_t clockid; 1397 struct timespec tp; 1398 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1399 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1400 1401 // Switch to using fast clocks for thread cpu time if 1402 // the clock_getres() returns 0 error code. 1403 // Note, that some kernels may support the current thread 1404 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1405 // returned by the pthread_getcpuclockid(). 1406 // If the fast Posix clocks are supported then the clock_getres() 1407 // must return at least tp.tv_sec == 0 which means a resolution 1408 // better than 1 sec. This is extra check for reliability. 1409 1410 if (pthread_getcpuclockid_func && 1411 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1412 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1413 _supports_fast_thread_cpu_time = true; 1414 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1415 } 1416 } 1417 1418 jlong os::javaTimeNanos() { 1419 if (os::supports_monotonic_clock()) { 1420 struct timespec tp; 1421 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp); 1422 assert(status == 0, "gettime error"); 1423 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1424 return result; 1425 } else { 1426 timeval time; 1427 int status = gettimeofday(&time, NULL); 1428 assert(status != -1, "linux error"); 1429 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1430 return 1000 * usecs; 1431 } 1432 } 1433 1434 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1435 if (os::supports_monotonic_clock()) { 1436 info_ptr->max_value = ALL_64_BITS; 1437 1438 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1439 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1440 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1441 } else { 1442 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1443 info_ptr->max_value = ALL_64_BITS; 1444 1445 // gettimeofday is a real time clock so it skips 1446 info_ptr->may_skip_backward = true; 1447 info_ptr->may_skip_forward = true; 1448 } 1449 1450 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1451 } 1452 1453 // Return the real, user, and system times in seconds from an 1454 // arbitrary fixed point in the past. 1455 bool os::getTimesSecs(double* process_real_time, 1456 double* process_user_time, 1457 double* process_system_time) { 1458 struct tms ticks; 1459 clock_t real_ticks = times(&ticks); 1460 1461 if (real_ticks == (clock_t) (-1)) { 1462 return false; 1463 } else { 1464 double ticks_per_second = (double) clock_tics_per_sec; 1465 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1466 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1467 *process_real_time = ((double) real_ticks) / ticks_per_second; 1468 1469 return true; 1470 } 1471 } 1472 1473 1474 char * os::local_time_string(char *buf, size_t buflen) { 1475 struct tm t; 1476 time_t long_time; 1477 time(&long_time); 1478 localtime_r(&long_time, &t); 1479 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1480 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1481 t.tm_hour, t.tm_min, t.tm_sec); 1482 return buf; 1483 } 1484 1485 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1486 return localtime_r(clock, res); 1487 } 1488 1489 //////////////////////////////////////////////////////////////////////////////// 1490 // runtime exit support 1491 1492 // Note: os::shutdown() might be called very early during initialization, or 1493 // called from signal handler. Before adding something to os::shutdown(), make 1494 // sure it is async-safe and can handle partially initialized VM. 1495 void os::shutdown() { 1496 1497 // allow PerfMemory to attempt cleanup of any persistent resources 1498 perfMemory_exit(); 1499 1500 // needs to remove object in file system 1501 AttachListener::abort(); 1502 1503 // flush buffered output, finish log files 1504 ostream_abort(); 1505 1506 // Check for abort hook 1507 abort_hook_t abort_hook = Arguments::abort_hook(); 1508 if (abort_hook != NULL) { 1509 abort_hook(); 1510 } 1511 1512 } 1513 1514 // Note: os::abort() might be called very early during initialization, or 1515 // called from signal handler. Before adding something to os::abort(), make 1516 // sure it is async-safe and can handle partially initialized VM. 1517 void os::abort(bool dump_core, void* siginfo, const void* context) { 1518 os::shutdown(); 1519 if (dump_core) { 1520 if (DumpPrivateMappingsInCore) { 1521 ClassLoader::close_jrt_image(); 1522 } 1523 #ifndef PRODUCT 1524 fdStream out(defaultStream::output_fd()); 1525 out.print_raw("Current thread is "); 1526 char buf[16]; 1527 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1528 out.print_raw_cr(buf); 1529 out.print_raw_cr("Dumping core ..."); 1530 #endif 1531 ::abort(); // dump core 1532 } 1533 1534 ::exit(1); 1535 } 1536 1537 // Die immediately, no exit hook, no abort hook, no cleanup. 1538 // Dump a core file, if possible, for debugging. 1539 void os::die() { 1540 if (TestUnresponsiveErrorHandler && !CreateCoredumpOnCrash) { 1541 // For TimeoutInErrorHandlingTest.java, we just kill the VM 1542 // and don't take the time to generate a core file. 1543 os::signal_raise(SIGKILL); 1544 } else { 1545 ::abort(); 1546 } 1547 } 1548 1549 // thread_id is kernel thread id (similar to Solaris LWP id) 1550 intx os::current_thread_id() { return os::Linux::gettid(); } 1551 int os::current_process_id() { 1552 return ::getpid(); 1553 } 1554 1555 // DLL functions 1556 1557 const char* os::dll_file_extension() { return ".so"; } 1558 1559 // This must be hard coded because it's the system's temporary 1560 // directory not the java application's temp directory, ala java.io.tmpdir. 1561 const char* os::get_temp_directory() { return "/tmp"; } 1562 1563 static bool file_exists(const char* filename) { 1564 struct stat statbuf; 1565 if (filename == NULL || strlen(filename) == 0) { 1566 return false; 1567 } 1568 return os::stat(filename, &statbuf) == 0; 1569 } 1570 1571 // check if addr is inside libjvm.so 1572 bool os::address_is_in_vm(address addr) { 1573 static address libjvm_base_addr; 1574 Dl_info dlinfo; 1575 1576 if (libjvm_base_addr == NULL) { 1577 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1578 libjvm_base_addr = (address)dlinfo.dli_fbase; 1579 } 1580 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1581 } 1582 1583 if (dladdr((void *)addr, &dlinfo) != 0) { 1584 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1585 } 1586 1587 return false; 1588 } 1589 1590 bool os::dll_address_to_function_name(address addr, char *buf, 1591 int buflen, int *offset, 1592 bool demangle) { 1593 // buf is not optional, but offset is optional 1594 assert(buf != NULL, "sanity check"); 1595 1596 Dl_info dlinfo; 1597 1598 if (dladdr((void*)addr, &dlinfo) != 0) { 1599 // see if we have a matching symbol 1600 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1601 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) { 1602 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1603 } 1604 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1605 return true; 1606 } 1607 // no matching symbol so try for just file info 1608 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1609 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1610 buf, buflen, offset, dlinfo.dli_fname, demangle)) { 1611 return true; 1612 } 1613 } 1614 } 1615 1616 buf[0] = '\0'; 1617 if (offset != NULL) *offset = -1; 1618 return false; 1619 } 1620 1621 struct _address_to_library_name { 1622 address addr; // input : memory address 1623 size_t buflen; // size of fname 1624 char* fname; // output: library name 1625 address base; // library base addr 1626 }; 1627 1628 static int address_to_library_name_callback(struct dl_phdr_info *info, 1629 size_t size, void *data) { 1630 int i; 1631 bool found = false; 1632 address libbase = NULL; 1633 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1634 1635 // iterate through all loadable segments 1636 for (i = 0; i < info->dlpi_phnum; i++) { 1637 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1638 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1639 // base address of a library is the lowest address of its loaded 1640 // segments. 1641 if (libbase == NULL || libbase > segbase) { 1642 libbase = segbase; 1643 } 1644 // see if 'addr' is within current segment 1645 if (segbase <= d->addr && 1646 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1647 found = true; 1648 } 1649 } 1650 } 1651 1652 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1653 // so dll_address_to_library_name() can fall through to use dladdr() which 1654 // can figure out executable name from argv[0]. 1655 if (found && info->dlpi_name && info->dlpi_name[0]) { 1656 d->base = libbase; 1657 if (d->fname) { 1658 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1659 } 1660 return 1; 1661 } 1662 return 0; 1663 } 1664 1665 bool os::dll_address_to_library_name(address addr, char* buf, 1666 int buflen, int* offset) { 1667 // buf is not optional, but offset is optional 1668 assert(buf != NULL, "sanity check"); 1669 1670 Dl_info dlinfo; 1671 struct _address_to_library_name data; 1672 1673 // There is a bug in old glibc dladdr() implementation that it could resolve 1674 // to wrong library name if the .so file has a base address != NULL. Here 1675 // we iterate through the program headers of all loaded libraries to find 1676 // out which library 'addr' really belongs to. This workaround can be 1677 // removed once the minimum requirement for glibc is moved to 2.3.x. 1678 data.addr = addr; 1679 data.fname = buf; 1680 data.buflen = buflen; 1681 data.base = NULL; 1682 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1683 1684 if (rslt) { 1685 // buf already contains library name 1686 if (offset) *offset = addr - data.base; 1687 return true; 1688 } 1689 if (dladdr((void*)addr, &dlinfo) != 0) { 1690 if (dlinfo.dli_fname != NULL) { 1691 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1692 } 1693 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1694 *offset = addr - (address)dlinfo.dli_fbase; 1695 } 1696 return true; 1697 } 1698 1699 buf[0] = '\0'; 1700 if (offset) *offset = -1; 1701 return false; 1702 } 1703 1704 // Loads .dll/.so and 1705 // in case of error it checks if .dll/.so was built for the 1706 // same architecture as Hotspot is running on 1707 1708 1709 // Remember the stack's state. The Linux dynamic linker will change 1710 // the stack to 'executable' at most once, so we must safepoint only once. 1711 bool os::Linux::_stack_is_executable = false; 1712 1713 // VM operation that loads a library. This is necessary if stack protection 1714 // of the Java stacks can be lost during loading the library. If we 1715 // do not stop the Java threads, they can stack overflow before the stacks 1716 // are protected again. 1717 class VM_LinuxDllLoad: public VM_Operation { 1718 private: 1719 const char *_filename; 1720 char *_ebuf; 1721 int _ebuflen; 1722 void *_lib; 1723 public: 1724 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1725 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1726 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1727 void doit() { 1728 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1729 os::Linux::_stack_is_executable = true; 1730 } 1731 void* loaded_library() { return _lib; } 1732 }; 1733 1734 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1735 void * result = NULL; 1736 bool load_attempted = false; 1737 1738 log_info(os)("attempting shared library load of %s", filename); 1739 1740 // Check whether the library to load might change execution rights 1741 // of the stack. If they are changed, the protection of the stack 1742 // guard pages will be lost. We need a safepoint to fix this. 1743 // 1744 // See Linux man page execstack(8) for more info. 1745 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1746 if (!ElfFile::specifies_noexecstack(filename)) { 1747 if (!is_init_completed()) { 1748 os::Linux::_stack_is_executable = true; 1749 // This is OK - No Java threads have been created yet, and hence no 1750 // stack guard pages to fix. 1751 // 1752 // Dynamic loader will make all stacks executable after 1753 // this function returns, and will not do that again. 1754 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet."); 1755 } else { 1756 warning("You have loaded library %s which might have disabled stack guard. " 1757 "The VM will try to fix the stack guard now.\n" 1758 "It's highly recommended that you fix the library with " 1759 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1760 filename); 1761 1762 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1763 JavaThread *jt = JavaThread::current(); 1764 if (jt->thread_state() != _thread_in_native) { 1765 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1766 // that requires ExecStack. Cannot enter safe point. Let's give up. 1767 warning("Unable to fix stack guard. Giving up."); 1768 } else { 1769 if (!LoadExecStackDllInVMThread) { 1770 // This is for the case where the DLL has an static 1771 // constructor function that executes JNI code. We cannot 1772 // load such DLLs in the VMThread. 1773 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1774 } 1775 1776 ThreadInVMfromNative tiv(jt); 1777 debug_only(VMNativeEntryWrapper vew;) 1778 1779 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1780 VMThread::execute(&op); 1781 if (LoadExecStackDllInVMThread) { 1782 result = op.loaded_library(); 1783 } 1784 load_attempted = true; 1785 } 1786 } 1787 } 1788 } 1789 1790 if (!load_attempted) { 1791 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1792 } 1793 1794 if (result != NULL) { 1795 // Successful loading 1796 return result; 1797 } 1798 1799 Elf32_Ehdr elf_head; 1800 int diag_msg_max_length=ebuflen-strlen(ebuf); 1801 char* diag_msg_buf=ebuf+strlen(ebuf); 1802 1803 if (diag_msg_max_length==0) { 1804 // No more space in ebuf for additional diagnostics message 1805 return NULL; 1806 } 1807 1808 1809 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1810 1811 if (file_descriptor < 0) { 1812 // Can't open library, report dlerror() message 1813 return NULL; 1814 } 1815 1816 bool failed_to_read_elf_head= 1817 (sizeof(elf_head)!= 1818 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1819 1820 ::close(file_descriptor); 1821 if (failed_to_read_elf_head) { 1822 // file i/o error - report dlerror() msg 1823 return NULL; 1824 } 1825 1826 if (elf_head.e_ident[EI_DATA] != LITTLE_ENDIAN_ONLY(ELFDATA2LSB) BIG_ENDIAN_ONLY(ELFDATA2MSB)) { 1827 // handle invalid/out of range endianness values 1828 if (elf_head.e_ident[EI_DATA] == 0 || elf_head.e_ident[EI_DATA] > 2) { 1829 return NULL; 1830 } 1831 1832 #if defined(VM_LITTLE_ENDIAN) 1833 // VM is LE, shared object BE 1834 elf_head.e_machine = be16toh(elf_head.e_machine); 1835 #else 1836 // VM is BE, shared object LE 1837 elf_head.e_machine = le16toh(elf_head.e_machine); 1838 #endif 1839 } 1840 1841 typedef struct { 1842 Elf32_Half code; // Actual value as defined in elf.h 1843 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1844 unsigned char elf_class; // 32 or 64 bit 1845 unsigned char endianness; // MSB or LSB 1846 char* name; // String representation 1847 } arch_t; 1848 1849 #ifndef EM_486 1850 #define EM_486 6 /* Intel 80486 */ 1851 #endif 1852 #ifndef EM_AARCH64 1853 #define EM_AARCH64 183 /* ARM AARCH64 */ 1854 #endif 1855 1856 static const arch_t arch_array[]={ 1857 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1858 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1859 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1860 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1861 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1862 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1863 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1864 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1865 #if defined(VM_LITTLE_ENDIAN) 1866 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"}, 1867 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"}, 1868 #else 1869 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1870 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"}, 1871 #endif 1872 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1873 // we only support 64 bit z architecture 1874 {EM_S390, EM_S390, ELFCLASS64, ELFDATA2MSB, (char*)"IBM System/390"}, 1875 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1876 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1877 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1878 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1879 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1880 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1881 }; 1882 1883 #if (defined IA32) 1884 static Elf32_Half running_arch_code=EM_386; 1885 #elif (defined AMD64) || (defined X32) 1886 static Elf32_Half running_arch_code=EM_X86_64; 1887 #elif (defined IA64) 1888 static Elf32_Half running_arch_code=EM_IA_64; 1889 #elif (defined __sparc) && (defined _LP64) 1890 static Elf32_Half running_arch_code=EM_SPARCV9; 1891 #elif (defined __sparc) && (!defined _LP64) 1892 static Elf32_Half running_arch_code=EM_SPARC; 1893 #elif (defined __powerpc64__) 1894 static Elf32_Half running_arch_code=EM_PPC64; 1895 #elif (defined __powerpc__) 1896 static Elf32_Half running_arch_code=EM_PPC; 1897 #elif (defined AARCH64) 1898 static Elf32_Half running_arch_code=EM_AARCH64; 1899 #elif (defined ARM) 1900 static Elf32_Half running_arch_code=EM_ARM; 1901 #elif (defined S390) 1902 static Elf32_Half running_arch_code=EM_S390; 1903 #elif (defined ALPHA) 1904 static Elf32_Half running_arch_code=EM_ALPHA; 1905 #elif (defined MIPSEL) 1906 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1907 #elif (defined PARISC) 1908 static Elf32_Half running_arch_code=EM_PARISC; 1909 #elif (defined MIPS) 1910 static Elf32_Half running_arch_code=EM_MIPS; 1911 #elif (defined M68K) 1912 static Elf32_Half running_arch_code=EM_68K; 1913 #elif (defined SH) 1914 static Elf32_Half running_arch_code=EM_SH; 1915 #else 1916 #error Method os::dll_load requires that one of following is defined:\ 1917 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc 1918 #endif 1919 1920 // Identify compatibility class for VM's architecture and library's architecture 1921 // Obtain string descriptions for architectures 1922 1923 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1924 int running_arch_index=-1; 1925 1926 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1927 if (running_arch_code == arch_array[i].code) { 1928 running_arch_index = i; 1929 } 1930 if (lib_arch.code == arch_array[i].code) { 1931 lib_arch.compat_class = arch_array[i].compat_class; 1932 lib_arch.name = arch_array[i].name; 1933 } 1934 } 1935 1936 assert(running_arch_index != -1, 1937 "Didn't find running architecture code (running_arch_code) in arch_array"); 1938 if (running_arch_index == -1) { 1939 // Even though running architecture detection failed 1940 // we may still continue with reporting dlerror() message 1941 return NULL; 1942 } 1943 1944 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1945 if (lib_arch.name != NULL) { 1946 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1947 " (Possible cause: can't load %s .so on a %s platform)", 1948 lib_arch.name, arch_array[running_arch_index].name); 1949 } else { 1950 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1951 " (Possible cause: can't load this .so (machine code=0x%x) on a %s platform)", 1952 lib_arch.code, arch_array[running_arch_index].name); 1953 } 1954 return NULL; 1955 } 1956 1957 if (lib_arch.endianness != arch_array[running_arch_index].endianness) { 1958 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: endianness mismatch)"); 1959 return NULL; 1960 } 1961 1962 // ELF file class/capacity : 0 - invalid, 1 - 32bit, 2 - 64bit 1963 if (lib_arch.elf_class > 2 || lib_arch.elf_class < 1) { 1964 ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: invalid ELF file class)"); 1965 return NULL; 1966 } 1967 1968 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1969 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1970 " (Possible cause: architecture word width mismatch, can't load %d-bit .so on a %d-bit platform)", 1971 (int) lib_arch.elf_class * 32, arch_array[running_arch_index].elf_class * 32); 1972 return NULL; 1973 } 1974 1975 return NULL; 1976 } 1977 1978 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 1979 int ebuflen) { 1980 void * result = ::dlopen(filename, RTLD_LAZY); 1981 if (result == NULL) { 1982 const char* error_report = ::dlerror(); 1983 if (error_report == NULL) { 1984 error_report = "dlerror returned no error description"; 1985 } 1986 if (ebuf != NULL && ebuflen > 0) { 1987 ::strncpy(ebuf, error_report, ebuflen-1); 1988 ebuf[ebuflen-1]='\0'; 1989 } 1990 Events::log(NULL, "Loading shared library %s failed, %s", filename, error_report); 1991 log_info(os)("shared library load of %s failed, %s", filename, error_report); 1992 } else { 1993 Events::log(NULL, "Loaded shared library %s", filename); 1994 log_info(os)("shared library load of %s was successful", filename); 1995 } 1996 return result; 1997 } 1998 1999 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 2000 int ebuflen) { 2001 void * result = NULL; 2002 if (LoadExecStackDllInVMThread) { 2003 result = dlopen_helper(filename, ebuf, ebuflen); 2004 } 2005 2006 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2007 // library that requires an executable stack, or which does not have this 2008 // stack attribute set, dlopen changes the stack attribute to executable. The 2009 // read protection of the guard pages gets lost. 2010 // 2011 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2012 // may have been queued at the same time. 2013 2014 if (!_stack_is_executable) { 2015 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) { 2016 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2017 jt->stack_guards_enabled()) { // No pending stack overflow exceptions 2018 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) { 2019 warning("Attempt to reguard stack yellow zone failed."); 2020 } 2021 } 2022 } 2023 } 2024 2025 return result; 2026 } 2027 2028 void* os::dll_lookup(void* handle, const char* name) { 2029 void* res = dlsym(handle, name); 2030 return res; 2031 } 2032 2033 void* os::get_default_process_handle() { 2034 return (void*)::dlopen(NULL, RTLD_LAZY); 2035 } 2036 2037 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) { 2038 int fd = ::open(filename, O_RDONLY); 2039 if (fd == -1) { 2040 return false; 2041 } 2042 2043 if (hdr != NULL) { 2044 st->print_cr("%s", hdr); 2045 } 2046 2047 char buf[33]; 2048 int bytes; 2049 buf[32] = '\0'; 2050 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) { 2051 st->print_raw(buf, bytes); 2052 } 2053 2054 ::close(fd); 2055 2056 return true; 2057 } 2058 2059 void os::print_dll_info(outputStream *st) { 2060 st->print_cr("Dynamic libraries:"); 2061 2062 char fname[32]; 2063 pid_t pid = os::Linux::gettid(); 2064 2065 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2066 2067 if (!_print_ascii_file(fname, st)) { 2068 st->print("Can not get library information for pid = %d\n", pid); 2069 } 2070 } 2071 2072 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 2073 FILE *procmapsFile = NULL; 2074 2075 // Open the procfs maps file for the current process 2076 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 2077 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 2078 char line[PATH_MAX + 100]; 2079 2080 // Read line by line from 'file' 2081 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 2082 u8 base, top, offset, inode; 2083 char permissions[5]; 2084 char device[6]; 2085 char name[sizeof(line)]; 2086 2087 // Parse fields from line 2088 int matches = sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s", 2089 &base, &top, permissions, &offset, device, &inode, name); 2090 // the last entry 'name' is empty for some entries, so we might have 6 matches instead of 7 for some lines 2091 if (matches < 6) continue; 2092 if (matches == 6) name[0] = '\0'; 2093 2094 // Filter by device id '00:00' so that we only get file system mapped files. 2095 if (strcmp(device, "00:00") != 0) { 2096 2097 // Call callback with the fields of interest 2098 if(callback(name, (address)base, (address)top, param)) { 2099 // Oops abort, callback aborted 2100 fclose(procmapsFile); 2101 return 1; 2102 } 2103 } 2104 } 2105 fclose(procmapsFile); 2106 } 2107 return 0; 2108 } 2109 2110 void os::print_os_info_brief(outputStream* st) { 2111 os::Linux::print_distro_info(st); 2112 2113 os::Posix::print_uname_info(st); 2114 2115 os::Linux::print_libversion_info(st); 2116 2117 } 2118 2119 void os::print_os_info(outputStream* st) { 2120 st->print("OS:"); 2121 2122 os::Linux::print_distro_info(st); 2123 2124 os::Posix::print_uname_info(st); 2125 2126 os::Linux::print_uptime_info(st); 2127 2128 // Print warning if unsafe chroot environment detected 2129 if (unsafe_chroot_detected) { 2130 st->print("WARNING!! "); 2131 st->print_cr("%s", unstable_chroot_error); 2132 } 2133 2134 os::Linux::print_libversion_info(st); 2135 2136 os::Posix::print_rlimit_info(st); 2137 2138 os::Posix::print_load_average(st); 2139 2140 os::Linux::print_full_memory_info(st); 2141 2142 os::Linux::print_proc_sys_info(st); 2143 2144 os::Linux::print_ld_preload_file(st); 2145 2146 os::Linux::print_container_info(st); 2147 2148 VM_Version::print_platform_virtualization_info(st); 2149 2150 os::Linux::print_steal_info(st); 2151 } 2152 2153 // Try to identify popular distros. 2154 // Most Linux distributions have a /etc/XXX-release file, which contains 2155 // the OS version string. Newer Linux distributions have a /etc/lsb-release 2156 // file that also contains the OS version string. Some have more than one 2157 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2158 // /etc/redhat-release.), so the order is important. 2159 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2160 // their own specific XXX-release file as well as a redhat-release file. 2161 // Because of this the XXX-release file needs to be searched for before the 2162 // redhat-release file. 2163 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the 2164 // search for redhat-release / SuSE-release needs to be before lsb-release. 2165 // Since the lsb-release file is the new standard it needs to be searched 2166 // before the older style release files. 2167 // Searching system-release (Red Hat) and os-release (other Linuxes) are a 2168 // next to last resort. The os-release file is a new standard that contains 2169 // distribution information and the system-release file seems to be an old 2170 // standard that has been replaced by the lsb-release and os-release files. 2171 // Searching for the debian_version file is the last resort. It contains 2172 // an informative string like "6.0.6" or "wheezy/sid". Because of this 2173 // "Debian " is printed before the contents of the debian_version file. 2174 2175 const char* distro_files[] = { 2176 "/etc/oracle-release", 2177 "/etc/mandriva-release", 2178 "/etc/mandrake-release", 2179 "/etc/sun-release", 2180 "/etc/redhat-release", 2181 "/etc/SuSE-release", 2182 "/etc/lsb-release", 2183 "/etc/turbolinux-release", 2184 "/etc/gentoo-release", 2185 "/etc/ltib-release", 2186 "/etc/angstrom-version", 2187 "/etc/system-release", 2188 "/etc/os-release", 2189 NULL }; 2190 2191 void os::Linux::print_distro_info(outputStream* st) { 2192 for (int i = 0;; i++) { 2193 const char* file = distro_files[i]; 2194 if (file == NULL) { 2195 break; // done 2196 } 2197 // If file prints, we found it. 2198 if (_print_ascii_file(file, st)) { 2199 return; 2200 } 2201 } 2202 2203 if (file_exists("/etc/debian_version")) { 2204 st->print("Debian "); 2205 _print_ascii_file("/etc/debian_version", st); 2206 } else { 2207 st->print("Linux"); 2208 } 2209 st->cr(); 2210 } 2211 2212 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) { 2213 char buf[256]; 2214 while (fgets(buf, sizeof(buf), fp)) { 2215 // Edit out extra stuff in expected format 2216 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) { 2217 char* ptr = strstr(buf, "\""); // the name is in quotes 2218 if (ptr != NULL) { 2219 ptr++; // go beyond first quote 2220 char* nl = strchr(ptr, '\"'); 2221 if (nl != NULL) *nl = '\0'; 2222 strncpy(distro, ptr, length); 2223 } else { 2224 ptr = strstr(buf, "="); 2225 ptr++; // go beyond equals then 2226 char* nl = strchr(ptr, '\n'); 2227 if (nl != NULL) *nl = '\0'; 2228 strncpy(distro, ptr, length); 2229 } 2230 return; 2231 } else if (get_first_line) { 2232 char* nl = strchr(buf, '\n'); 2233 if (nl != NULL) *nl = '\0'; 2234 strncpy(distro, buf, length); 2235 return; 2236 } 2237 } 2238 // print last line and close 2239 char* nl = strchr(buf, '\n'); 2240 if (nl != NULL) *nl = '\0'; 2241 strncpy(distro, buf, length); 2242 } 2243 2244 static void parse_os_info(char* distro, size_t length, const char* file) { 2245 FILE* fp = fopen(file, "r"); 2246 if (fp != NULL) { 2247 // if suse format, print out first line 2248 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0); 2249 parse_os_info_helper(fp, distro, length, get_first_line); 2250 fclose(fp); 2251 } 2252 } 2253 2254 void os::get_summary_os_info(char* buf, size_t buflen) { 2255 for (int i = 0;; i++) { 2256 const char* file = distro_files[i]; 2257 if (file == NULL) { 2258 break; // ran out of distro_files 2259 } 2260 if (file_exists(file)) { 2261 parse_os_info(buf, buflen, file); 2262 return; 2263 } 2264 } 2265 // special case for debian 2266 if (file_exists("/etc/debian_version")) { 2267 strncpy(buf, "Debian ", buflen); 2268 if (buflen > 7) { 2269 parse_os_info(&buf[7], buflen-7, "/etc/debian_version"); 2270 } 2271 } else { 2272 strncpy(buf, "Linux", buflen); 2273 } 2274 } 2275 2276 void os::Linux::print_libversion_info(outputStream* st) { 2277 // libc, pthread 2278 st->print("libc:"); 2279 st->print("%s ", os::Linux::glibc_version()); 2280 st->print("%s ", os::Linux::libpthread_version()); 2281 st->cr(); 2282 } 2283 2284 void os::Linux::print_proc_sys_info(outputStream* st) { 2285 st->cr(); 2286 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):"); 2287 _print_ascii_file("/proc/sys/kernel/threads-max", st); 2288 st->cr(); 2289 st->cr(); 2290 2291 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):"); 2292 _print_ascii_file("/proc/sys/vm/max_map_count", st); 2293 st->cr(); 2294 st->cr(); 2295 2296 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):"); 2297 _print_ascii_file("/proc/sys/kernel/pid_max", st); 2298 st->cr(); 2299 st->cr(); 2300 } 2301 2302 void os::Linux::print_full_memory_info(outputStream* st) { 2303 st->print("\n/proc/meminfo:\n"); 2304 _print_ascii_file("/proc/meminfo", st); 2305 st->cr(); 2306 } 2307 2308 void os::Linux::print_ld_preload_file(outputStream* st) { 2309 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:"); 2310 st->cr(); 2311 } 2312 2313 void os::Linux::print_uptime_info(outputStream* st) { 2314 struct sysinfo sinfo; 2315 int ret = sysinfo(&sinfo); 2316 if (ret == 0) { 2317 os::print_dhm(st, "OS uptime:", (long) sinfo.uptime); 2318 } 2319 } 2320 2321 2322 void os::Linux::print_container_info(outputStream* st) { 2323 if (!OSContainer::is_containerized()) { 2324 return; 2325 } 2326 2327 st->print("container (cgroup) information:\n"); 2328 2329 const char *p_ct = OSContainer::container_type(); 2330 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported"); 2331 2332 char *p = OSContainer::cpu_cpuset_cpus(); 2333 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported"); 2334 free(p); 2335 2336 p = OSContainer::cpu_cpuset_memory_nodes(); 2337 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported"); 2338 free(p); 2339 2340 int i = OSContainer::active_processor_count(); 2341 st->print("active_processor_count: "); 2342 if (i > 0) { 2343 st->print("%d\n", i); 2344 } else { 2345 st->print("not supported\n"); 2346 } 2347 2348 i = OSContainer::cpu_quota(); 2349 st->print("cpu_quota: "); 2350 if (i > 0) { 2351 st->print("%d\n", i); 2352 } else { 2353 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota"); 2354 } 2355 2356 i = OSContainer::cpu_period(); 2357 st->print("cpu_period: "); 2358 if (i > 0) { 2359 st->print("%d\n", i); 2360 } else { 2361 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period"); 2362 } 2363 2364 i = OSContainer::cpu_shares(); 2365 st->print("cpu_shares: "); 2366 if (i > 0) { 2367 st->print("%d\n", i); 2368 } else { 2369 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares"); 2370 } 2371 2372 jlong j = OSContainer::memory_limit_in_bytes(); 2373 st->print("memory_limit_in_bytes: "); 2374 if (j > 0) { 2375 st->print(JLONG_FORMAT "\n", j); 2376 } else { 2377 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited"); 2378 } 2379 2380 j = OSContainer::memory_and_swap_limit_in_bytes(); 2381 st->print("memory_and_swap_limit_in_bytes: "); 2382 if (j > 0) { 2383 st->print(JLONG_FORMAT "\n", j); 2384 } else { 2385 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited"); 2386 } 2387 2388 j = OSContainer::memory_soft_limit_in_bytes(); 2389 st->print("memory_soft_limit_in_bytes: "); 2390 if (j > 0) { 2391 st->print(JLONG_FORMAT "\n", j); 2392 } else { 2393 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited"); 2394 } 2395 2396 j = OSContainer::OSContainer::memory_usage_in_bytes(); 2397 st->print("memory_usage_in_bytes: "); 2398 if (j > 0) { 2399 st->print(JLONG_FORMAT "\n", j); 2400 } else { 2401 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited"); 2402 } 2403 2404 j = OSContainer::OSContainer::memory_max_usage_in_bytes(); 2405 st->print("memory_max_usage_in_bytes: "); 2406 if (j > 0) { 2407 st->print(JLONG_FORMAT "\n", j); 2408 } else { 2409 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited"); 2410 } 2411 st->cr(); 2412 } 2413 2414 void os::Linux::print_steal_info(outputStream* st) { 2415 if (has_initial_tick_info) { 2416 CPUPerfTicks pticks; 2417 bool res = os::Linux::get_tick_information(&pticks, -1); 2418 2419 if (res && pticks.has_steal_ticks) { 2420 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks; 2421 uint64_t total_ticks_difference = pticks.total - initial_total_ticks; 2422 double steal_ticks_perc = 0.0; 2423 if (total_ticks_difference != 0) { 2424 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference; 2425 } 2426 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference); 2427 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc); 2428 } 2429 } 2430 } 2431 2432 void os::print_memory_info(outputStream* st) { 2433 2434 st->print("Memory:"); 2435 st->print(" %dk page", os::vm_page_size()>>10); 2436 2437 // values in struct sysinfo are "unsigned long" 2438 struct sysinfo si; 2439 sysinfo(&si); 2440 2441 st->print(", physical " UINT64_FORMAT "k", 2442 os::physical_memory() >> 10); 2443 st->print("(" UINT64_FORMAT "k free)", 2444 os::available_memory() >> 10); 2445 st->print(", swap " UINT64_FORMAT "k", 2446 ((jlong)si.totalswap * si.mem_unit) >> 10); 2447 st->print("(" UINT64_FORMAT "k free)", 2448 ((jlong)si.freeswap * si.mem_unit) >> 10); 2449 st->cr(); 2450 } 2451 2452 // Print the first "model name" line and the first "flags" line 2453 // that we find and nothing more. We assume "model name" comes 2454 // before "flags" so if we find a second "model name", then the 2455 // "flags" field is considered missing. 2456 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) { 2457 #if defined(IA32) || defined(AMD64) 2458 // Other platforms have less repetitive cpuinfo files 2459 FILE *fp = fopen("/proc/cpuinfo", "r"); 2460 if (fp) { 2461 while (!feof(fp)) { 2462 if (fgets(buf, buflen, fp)) { 2463 // Assume model name comes before flags 2464 bool model_name_printed = false; 2465 if (strstr(buf, "model name") != NULL) { 2466 if (!model_name_printed) { 2467 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n"); 2468 st->print_raw(buf); 2469 model_name_printed = true; 2470 } else { 2471 // model name printed but not flags? Odd, just return 2472 fclose(fp); 2473 return true; 2474 } 2475 } 2476 // print the flags line too 2477 if (strstr(buf, "flags") != NULL) { 2478 st->print_raw(buf); 2479 fclose(fp); 2480 return true; 2481 } 2482 } 2483 } 2484 fclose(fp); 2485 } 2486 #endif // x86 platforms 2487 return false; 2488 } 2489 2490 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) { 2491 // Only print the model name if the platform provides this as a summary 2492 if (!print_model_name_and_flags(st, buf, buflen)) { 2493 st->print("\n/proc/cpuinfo:\n"); 2494 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2495 st->print_cr(" <Not Available>"); 2496 } 2497 } 2498 } 2499 2500 #if defined(AMD64) || defined(IA32) || defined(X32) 2501 const char* search_string = "model name"; 2502 #elif defined(M68K) 2503 const char* search_string = "CPU"; 2504 #elif defined(PPC64) 2505 const char* search_string = "cpu"; 2506 #elif defined(S390) 2507 const char* search_string = "machine ="; 2508 #elif defined(SPARC) 2509 const char* search_string = "cpu"; 2510 #else 2511 const char* search_string = "Processor"; 2512 #endif 2513 2514 // Parses the cpuinfo file for string representing the model name. 2515 void os::get_summary_cpu_info(char* cpuinfo, size_t length) { 2516 FILE* fp = fopen("/proc/cpuinfo", "r"); 2517 if (fp != NULL) { 2518 while (!feof(fp)) { 2519 char buf[256]; 2520 if (fgets(buf, sizeof(buf), fp)) { 2521 char* start = strstr(buf, search_string); 2522 if (start != NULL) { 2523 char *ptr = start + strlen(search_string); 2524 char *end = buf + strlen(buf); 2525 while (ptr != end) { 2526 // skip whitespace and colon for the rest of the name. 2527 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') { 2528 break; 2529 } 2530 ptr++; 2531 } 2532 if (ptr != end) { 2533 // reasonable string, get rid of newline and keep the rest 2534 char* nl = strchr(buf, '\n'); 2535 if (nl != NULL) *nl = '\0'; 2536 strncpy(cpuinfo, ptr, length); 2537 fclose(fp); 2538 return; 2539 } 2540 } 2541 } 2542 } 2543 fclose(fp); 2544 } 2545 // cpuinfo not found or parsing failed, just print generic string. The entire 2546 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86) 2547 #if defined(AARCH64) 2548 strncpy(cpuinfo, "AArch64", length); 2549 #elif defined(AMD64) 2550 strncpy(cpuinfo, "x86_64", length); 2551 #elif defined(ARM) // Order wrt. AARCH64 is relevant! 2552 strncpy(cpuinfo, "ARM", length); 2553 #elif defined(IA32) 2554 strncpy(cpuinfo, "x86_32", length); 2555 #elif defined(IA64) 2556 strncpy(cpuinfo, "IA64", length); 2557 #elif defined(PPC) 2558 strncpy(cpuinfo, "PPC64", length); 2559 #elif defined(S390) 2560 strncpy(cpuinfo, "S390", length); 2561 #elif defined(SPARC) 2562 strncpy(cpuinfo, "sparcv9", length); 2563 #elif defined(ZERO_LIBARCH) 2564 strncpy(cpuinfo, ZERO_LIBARCH, length); 2565 #else 2566 strncpy(cpuinfo, "unknown", length); 2567 #endif 2568 } 2569 2570 static void print_signal_handler(outputStream* st, int sig, 2571 char* buf, size_t buflen); 2572 2573 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2574 st->print_cr("Signal Handlers:"); 2575 print_signal_handler(st, SIGSEGV, buf, buflen); 2576 print_signal_handler(st, SIGBUS , buf, buflen); 2577 print_signal_handler(st, SIGFPE , buf, buflen); 2578 print_signal_handler(st, SIGPIPE, buf, buflen); 2579 print_signal_handler(st, SIGXFSZ, buf, buflen); 2580 print_signal_handler(st, SIGILL , buf, buflen); 2581 print_signal_handler(st, SR_signum, buf, buflen); 2582 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2583 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2584 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2585 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2586 #if defined(PPC64) 2587 print_signal_handler(st, SIGTRAP, buf, buflen); 2588 #endif 2589 } 2590 2591 static char saved_jvm_path[MAXPATHLEN] = {0}; 2592 2593 // Find the full path to the current module, libjvm.so 2594 void os::jvm_path(char *buf, jint buflen) { 2595 // Error checking. 2596 if (buflen < MAXPATHLEN) { 2597 assert(false, "must use a large-enough buffer"); 2598 buf[0] = '\0'; 2599 return; 2600 } 2601 // Lazy resolve the path to current module. 2602 if (saved_jvm_path[0] != 0) { 2603 strcpy(buf, saved_jvm_path); 2604 return; 2605 } 2606 2607 char dli_fname[MAXPATHLEN]; 2608 bool ret = dll_address_to_library_name( 2609 CAST_FROM_FN_PTR(address, os::jvm_path), 2610 dli_fname, sizeof(dli_fname), NULL); 2611 assert(ret, "cannot locate libjvm"); 2612 char *rp = NULL; 2613 if (ret && dli_fname[0] != '\0') { 2614 rp = os::Posix::realpath(dli_fname, buf, buflen); 2615 } 2616 if (rp == NULL) { 2617 return; 2618 } 2619 2620 if (Arguments::sun_java_launcher_is_altjvm()) { 2621 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2622 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so". 2623 // If "/jre/lib/" appears at the right place in the string, then 2624 // assume we are installed in a JDK and we're done. Otherwise, check 2625 // for a JAVA_HOME environment variable and fix up the path so it 2626 // looks like libjvm.so is installed there (append a fake suffix 2627 // hotspot/libjvm.so). 2628 const char *p = buf + strlen(buf) - 1; 2629 for (int count = 0; p > buf && count < 5; ++count) { 2630 for (--p; p > buf && *p != '/'; --p) 2631 /* empty */ ; 2632 } 2633 2634 if (strncmp(p, "/jre/lib/", 9) != 0) { 2635 // Look for JAVA_HOME in the environment. 2636 char* java_home_var = ::getenv("JAVA_HOME"); 2637 if (java_home_var != NULL && java_home_var[0] != 0) { 2638 char* jrelib_p; 2639 int len; 2640 2641 // Check the current module name "libjvm.so". 2642 p = strrchr(buf, '/'); 2643 if (p == NULL) { 2644 return; 2645 } 2646 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2647 2648 rp = os::Posix::realpath(java_home_var, buf, buflen); 2649 if (rp == NULL) { 2650 return; 2651 } 2652 2653 // determine if this is a legacy image or modules image 2654 // modules image doesn't have "jre" subdirectory 2655 len = strlen(buf); 2656 assert(len < buflen, "Ran out of buffer room"); 2657 jrelib_p = buf + len; 2658 snprintf(jrelib_p, buflen-len, "/jre/lib"); 2659 if (0 != access(buf, F_OK)) { 2660 snprintf(jrelib_p, buflen-len, "/lib"); 2661 } 2662 2663 if (0 == access(buf, F_OK)) { 2664 // Use current module name "libjvm.so" 2665 len = strlen(buf); 2666 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2667 } else { 2668 // Go back to path of .so 2669 rp = os::Posix::realpath(dli_fname, buf, buflen); 2670 if (rp == NULL) { 2671 return; 2672 } 2673 } 2674 } 2675 } 2676 } 2677 2678 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2679 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2680 } 2681 2682 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2683 // no prefix required, not even "_" 2684 } 2685 2686 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2687 // no suffix required 2688 } 2689 2690 //////////////////////////////////////////////////////////////////////////////// 2691 // sun.misc.Signal support 2692 2693 static void UserHandler(int sig, void *siginfo, void *context) { 2694 // Ctrl-C is pressed during error reporting, likely because the error 2695 // handler fails to abort. Let VM die immediately. 2696 if (sig == SIGINT && VMError::is_error_reported()) { 2697 os::die(); 2698 } 2699 2700 os::signal_notify(sig); 2701 } 2702 2703 void* os::user_handler() { 2704 return CAST_FROM_FN_PTR(void*, UserHandler); 2705 } 2706 2707 extern "C" { 2708 typedef void (*sa_handler_t)(int); 2709 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2710 } 2711 2712 void* os::signal(int signal_number, void* handler) { 2713 struct sigaction sigAct, oldSigAct; 2714 2715 sigfillset(&(sigAct.sa_mask)); 2716 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2717 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2718 2719 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2720 // -1 means registration failed 2721 return (void *)-1; 2722 } 2723 2724 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2725 } 2726 2727 void os::signal_raise(int signal_number) { 2728 ::raise(signal_number); 2729 } 2730 2731 // The following code is moved from os.cpp for making this 2732 // code platform specific, which it is by its very nature. 2733 2734 // Will be modified when max signal is changed to be dynamic 2735 int os::sigexitnum_pd() { 2736 return NSIG; 2737 } 2738 2739 // a counter for each possible signal value 2740 static volatile jint pending_signals[NSIG+1] = { 0 }; 2741 2742 // Linux(POSIX) specific hand shaking semaphore. 2743 static Semaphore* sig_sem = NULL; 2744 static PosixSemaphore sr_semaphore; 2745 2746 static void jdk_misc_signal_init() { 2747 // Initialize signal structures 2748 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2749 2750 // Initialize signal semaphore 2751 sig_sem = new Semaphore(); 2752 } 2753 2754 void os::signal_notify(int sig) { 2755 if (sig_sem != NULL) { 2756 Atomic::inc(&pending_signals[sig]); 2757 sig_sem->signal(); 2758 } else { 2759 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init 2760 // initialization isn't called. 2761 assert(ReduceSignalUsage, "signal semaphore should be created"); 2762 } 2763 } 2764 2765 static int check_pending_signals() { 2766 for (;;) { 2767 for (int i = 0; i < NSIG + 1; i++) { 2768 jint n = pending_signals[i]; 2769 if (n > 0 && n == Atomic::cmpxchg(&pending_signals[i], n, n - 1)) { 2770 return i; 2771 } 2772 } 2773 JavaThread *thread = JavaThread::current(); 2774 ThreadBlockInVM tbivm(thread); 2775 2776 bool threadIsSuspended; 2777 do { 2778 thread->set_suspend_equivalent(); 2779 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2780 sig_sem->wait(); 2781 2782 // were we externally suspended while we were waiting? 2783 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2784 if (threadIsSuspended) { 2785 // The semaphore has been incremented, but while we were waiting 2786 // another thread suspended us. We don't want to continue running 2787 // while suspended because that would surprise the thread that 2788 // suspended us. 2789 sig_sem->signal(); 2790 2791 thread->java_suspend_self(); 2792 } 2793 } while (threadIsSuspended); 2794 } 2795 } 2796 2797 int os::signal_wait() { 2798 return check_pending_signals(); 2799 } 2800 2801 //////////////////////////////////////////////////////////////////////////////// 2802 // Virtual Memory 2803 2804 int os::vm_page_size() { 2805 // Seems redundant as all get out 2806 assert(os::Linux::page_size() != -1, "must call os::init"); 2807 return os::Linux::page_size(); 2808 } 2809 2810 // Solaris allocates memory by pages. 2811 int os::vm_allocation_granularity() { 2812 assert(os::Linux::page_size() != -1, "must call os::init"); 2813 return os::Linux::page_size(); 2814 } 2815 2816 // Rationale behind this function: 2817 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2818 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2819 // samples for JITted code. Here we create private executable mapping over the code cache 2820 // and then we can use standard (well, almost, as mapping can change) way to provide 2821 // info for the reporting script by storing timestamp and location of symbol 2822 void linux_wrap_code(char* base, size_t size) { 2823 static volatile jint cnt = 0; 2824 2825 if (!UseOprofile) { 2826 return; 2827 } 2828 2829 char buf[PATH_MAX+1]; 2830 int num = Atomic::add(&cnt, 1); 2831 2832 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2833 os::get_temp_directory(), os::current_process_id(), num); 2834 unlink(buf); 2835 2836 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2837 2838 if (fd != -1) { 2839 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2840 if (rv != (off_t)-1) { 2841 if (::write(fd, "", 1) == 1) { 2842 mmap(base, size, 2843 PROT_READ|PROT_WRITE|PROT_EXEC, 2844 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2845 } 2846 } 2847 ::close(fd); 2848 unlink(buf); 2849 } 2850 } 2851 2852 static bool recoverable_mmap_error(int err) { 2853 // See if the error is one we can let the caller handle. This 2854 // list of errno values comes from JBS-6843484. I can't find a 2855 // Linux man page that documents this specific set of errno 2856 // values so while this list currently matches Solaris, it may 2857 // change as we gain experience with this failure mode. 2858 switch (err) { 2859 case EBADF: 2860 case EINVAL: 2861 case ENOTSUP: 2862 // let the caller deal with these errors 2863 return true; 2864 2865 default: 2866 // Any remaining errors on this OS can cause our reserved mapping 2867 // to be lost. That can cause confusion where different data 2868 // structures think they have the same memory mapped. The worst 2869 // scenario is if both the VM and a library think they have the 2870 // same memory mapped. 2871 return false; 2872 } 2873 } 2874 2875 static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2876 int err) { 2877 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2878 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec, 2879 os::strerror(err), err); 2880 } 2881 2882 static void warn_fail_commit_memory(char* addr, size_t size, 2883 size_t alignment_hint, bool exec, 2884 int err) { 2885 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2886 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, 2887 alignment_hint, exec, os::strerror(err), err); 2888 } 2889 2890 // NOTE: Linux kernel does not really reserve the pages for us. 2891 // All it does is to check if there are enough free pages 2892 // left at the time of mmap(). This could be a potential 2893 // problem. 2894 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2895 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2896 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2897 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2898 if (res != (uintptr_t) MAP_FAILED) { 2899 if (UseNUMAInterleaving) { 2900 numa_make_global(addr, size); 2901 } 2902 return 0; 2903 } 2904 2905 int err = errno; // save errno from mmap() call above 2906 2907 if (!recoverable_mmap_error(err)) { 2908 warn_fail_commit_memory(addr, size, exec, err); 2909 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2910 } 2911 2912 return err; 2913 } 2914 2915 bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2916 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2917 } 2918 2919 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2920 const char* mesg) { 2921 assert(mesg != NULL, "mesg must be specified"); 2922 int err = os::Linux::commit_memory_impl(addr, size, exec); 2923 if (err != 0) { 2924 // the caller wants all commit errors to exit with the specified mesg: 2925 warn_fail_commit_memory(addr, size, exec, err); 2926 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2927 } 2928 } 2929 2930 // Define MAP_HUGETLB here so we can build HotSpot on old systems. 2931 #ifndef MAP_HUGETLB 2932 #define MAP_HUGETLB 0x40000 2933 #endif 2934 2935 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2936 #ifndef MADV_HUGEPAGE 2937 #define MADV_HUGEPAGE 14 2938 #endif 2939 2940 int os::Linux::commit_memory_impl(char* addr, size_t size, 2941 size_t alignment_hint, bool exec) { 2942 int err = os::Linux::commit_memory_impl(addr, size, exec); 2943 if (err == 0) { 2944 realign_memory(addr, size, alignment_hint); 2945 } 2946 return err; 2947 } 2948 2949 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2950 bool exec) { 2951 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2952 } 2953 2954 void os::pd_commit_memory_or_exit(char* addr, size_t size, 2955 size_t alignment_hint, bool exec, 2956 const char* mesg) { 2957 assert(mesg != NULL, "mesg must be specified"); 2958 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2959 if (err != 0) { 2960 // the caller wants all commit errors to exit with the specified mesg: 2961 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2962 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2963 } 2964 } 2965 2966 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2967 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2968 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2969 // be supported or the memory may already be backed by huge pages. 2970 ::madvise(addr, bytes, MADV_HUGEPAGE); 2971 } 2972 } 2973 2974 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2975 // This method works by doing an mmap over an existing mmaping and effectively discarding 2976 // the existing pages. However it won't work for SHM-based large pages that cannot be 2977 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2978 // small pages on top of the SHM segment. This method always works for small pages, so we 2979 // allow that in any case. 2980 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2981 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2982 } 2983 } 2984 2985 void os::numa_make_global(char *addr, size_t bytes) { 2986 Linux::numa_interleave_memory(addr, bytes); 2987 } 2988 2989 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2990 // bind policy to MPOL_PREFERRED for the current thread. 2991 #define USE_MPOL_PREFERRED 0 2992 2993 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2994 // To make NUMA and large pages more robust when both enabled, we need to ease 2995 // the requirements on where the memory should be allocated. MPOL_BIND is the 2996 // default policy and it will force memory to be allocated on the specified 2997 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2998 // the specified node, but will not force it. Using this policy will prevent 2999 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 3000 // free large pages. 3001 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 3002 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 3003 } 3004 3005 bool os::numa_topology_changed() { return false; } 3006 3007 size_t os::numa_get_groups_num() { 3008 // Return just the number of nodes in which it's possible to allocate memory 3009 // (in numa terminology, configured nodes). 3010 return Linux::numa_num_configured_nodes(); 3011 } 3012 3013 int os::numa_get_group_id() { 3014 int cpu_id = Linux::sched_getcpu(); 3015 if (cpu_id != -1) { 3016 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 3017 if (lgrp_id != -1) { 3018 return lgrp_id; 3019 } 3020 } 3021 return 0; 3022 } 3023 3024 int os::numa_get_group_id_for_address(const void* address) { 3025 void** pages = const_cast<void**>(&address); 3026 int id = -1; 3027 3028 if (os::Linux::numa_move_pages(0, 1, pages, NULL, &id, 0) == -1) { 3029 return -1; 3030 } 3031 if (id < 0) { 3032 return -1; 3033 } 3034 return id; 3035 } 3036 3037 int os::Linux::get_existing_num_nodes() { 3038 int node; 3039 int highest_node_number = Linux::numa_max_node(); 3040 int num_nodes = 0; 3041 3042 // Get the total number of nodes in the system including nodes without memory. 3043 for (node = 0; node <= highest_node_number; node++) { 3044 if (is_node_in_existing_nodes(node)) { 3045 num_nodes++; 3046 } 3047 } 3048 return num_nodes; 3049 } 3050 3051 size_t os::numa_get_leaf_groups(int *ids, size_t size) { 3052 int highest_node_number = Linux::numa_max_node(); 3053 size_t i = 0; 3054 3055 // Map all node ids in which it is possible to allocate memory. Also nodes are 3056 // not always consecutively available, i.e. available from 0 to the highest 3057 // node number. If the nodes have been bound explicitly using numactl membind, 3058 // then allocate memory from those nodes only. 3059 for (int node = 0; node <= highest_node_number; node++) { 3060 if (Linux::is_node_in_bound_nodes((unsigned int)node)) { 3061 ids[i++] = node; 3062 } 3063 } 3064 return i; 3065 } 3066 3067 bool os::get_page_info(char *start, page_info* info) { 3068 return false; 3069 } 3070 3071 char *os::scan_pages(char *start, char* end, page_info* page_expected, 3072 page_info* page_found) { 3073 return end; 3074 } 3075 3076 3077 int os::Linux::sched_getcpu_syscall(void) { 3078 unsigned int cpu = 0; 3079 int retval = -1; 3080 3081 #if defined(IA32) 3082 #ifndef SYS_getcpu 3083 #define SYS_getcpu 318 3084 #endif 3085 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 3086 #elif defined(AMD64) 3087 // Unfortunately we have to bring all these macros here from vsyscall.h 3088 // to be able to compile on old linuxes. 3089 #define __NR_vgetcpu 2 3090 #define VSYSCALL_START (-10UL << 20) 3091 #define VSYSCALL_SIZE 1024 3092 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 3093 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 3094 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 3095 retval = vgetcpu(&cpu, NULL, NULL); 3096 #endif 3097 3098 return (retval == -1) ? retval : cpu; 3099 } 3100 3101 void os::Linux::sched_getcpu_init() { 3102 // sched_getcpu() should be in libc. 3103 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 3104 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 3105 3106 // If it's not, try a direct syscall. 3107 if (sched_getcpu() == -1) { 3108 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 3109 (void*)&sched_getcpu_syscall)); 3110 } 3111 3112 if (sched_getcpu() == -1) { 3113 vm_exit_during_initialization("getcpu(2) system call not supported by kernel"); 3114 } 3115 } 3116 3117 // Something to do with the numa-aware allocator needs these symbols 3118 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 3119 extern "C" JNIEXPORT void numa_error(char *where) { } 3120 3121 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails 3122 // load symbol from base version instead. 3123 void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 3124 void *f = dlvsym(handle, name, "libnuma_1.1"); 3125 if (f == NULL) { 3126 f = dlsym(handle, name); 3127 } 3128 return f; 3129 } 3130 3131 // Handle request to load libnuma symbol version 1.2 (API v2) only. 3132 // Return NULL if the symbol is not defined in this particular version. 3133 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) { 3134 return dlvsym(handle, name, "libnuma_1.2"); 3135 } 3136 3137 bool os::Linux::libnuma_init() { 3138 if (sched_getcpu() != -1) { // Requires sched_getcpu() support 3139 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 3140 if (handle != NULL) { 3141 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 3142 libnuma_dlsym(handle, "numa_node_to_cpus"))); 3143 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 3144 libnuma_dlsym(handle, "numa_max_node"))); 3145 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t, 3146 libnuma_dlsym(handle, "numa_num_configured_nodes"))); 3147 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 3148 libnuma_dlsym(handle, "numa_available"))); 3149 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 3150 libnuma_dlsym(handle, "numa_tonode_memory"))); 3151 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 3152 libnuma_dlsym(handle, "numa_interleave_memory"))); 3153 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t, 3154 libnuma_v2_dlsym(handle, "numa_interleave_memory"))); 3155 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 3156 libnuma_dlsym(handle, "numa_set_bind_policy"))); 3157 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t, 3158 libnuma_dlsym(handle, "numa_bitmask_isbitset"))); 3159 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t, 3160 libnuma_dlsym(handle, "numa_distance"))); 3161 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t, 3162 libnuma_v2_dlsym(handle, "numa_get_membind"))); 3163 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t, 3164 libnuma_v2_dlsym(handle, "numa_get_interleave_mask"))); 3165 set_numa_move_pages(CAST_TO_FN_PTR(numa_move_pages_func_t, 3166 libnuma_dlsym(handle, "numa_move_pages"))); 3167 set_numa_set_preferred(CAST_TO_FN_PTR(numa_set_preferred_func_t, 3168 libnuma_dlsym(handle, "numa_set_preferred"))); 3169 3170 if (numa_available() != -1) { 3171 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 3172 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr")); 3173 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr")); 3174 set_numa_interleave_bitmask(_numa_get_interleave_mask()); 3175 set_numa_membind_bitmask(_numa_get_membind()); 3176 // Create an index -> node mapping, since nodes are not always consecutive 3177 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 3178 rebuild_nindex_to_node_map(); 3179 // Create a cpu -> node mapping 3180 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 3181 rebuild_cpu_to_node_map(); 3182 return true; 3183 } 3184 } 3185 } 3186 return false; 3187 } 3188 3189 size_t os::Linux::default_guard_size(os::ThreadType thr_type) { 3190 // Creating guard page is very expensive. Java thread has HotSpot 3191 // guard pages, only enable glibc guard page for non-Java threads. 3192 // (Remember: compiler thread is a Java thread, too!) 3193 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size()); 3194 } 3195 3196 void os::Linux::rebuild_nindex_to_node_map() { 3197 int highest_node_number = Linux::numa_max_node(); 3198 3199 nindex_to_node()->clear(); 3200 for (int node = 0; node <= highest_node_number; node++) { 3201 if (Linux::is_node_in_existing_nodes(node)) { 3202 nindex_to_node()->append(node); 3203 } 3204 } 3205 } 3206 3207 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 3208 // The table is later used in get_node_by_cpu(). 3209 void os::Linux::rebuild_cpu_to_node_map() { 3210 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 3211 // in libnuma (possible values are starting from 16, 3212 // and continuing up with every other power of 2, but less 3213 // than the maximum number of CPUs supported by kernel), and 3214 // is a subject to change (in libnuma version 2 the requirements 3215 // are more reasonable) we'll just hardcode the number they use 3216 // in the library. 3217 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 3218 3219 size_t cpu_num = processor_count(); 3220 size_t cpu_map_size = NCPUS / BitsPerCLong; 3221 size_t cpu_map_valid_size = 3222 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 3223 3224 cpu_to_node()->clear(); 3225 cpu_to_node()->at_grow(cpu_num - 1); 3226 3227 size_t node_num = get_existing_num_nodes(); 3228 3229 int distance = 0; 3230 int closest_distance = INT_MAX; 3231 int closest_node = 0; 3232 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 3233 for (size_t i = 0; i < node_num; i++) { 3234 // Check if node is configured (not a memory-less node). If it is not, find 3235 // the closest configured node. Check also if node is bound, i.e. it's allowed 3236 // to allocate memory from the node. If it's not allowed, map cpus in that node 3237 // to the closest node from which memory allocation is allowed. 3238 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) || 3239 !is_node_in_bound_nodes(nindex_to_node()->at(i))) { 3240 closest_distance = INT_MAX; 3241 // Check distance from all remaining nodes in the system. Ignore distance 3242 // from itself, from another non-configured node, and from another non-bound 3243 // node. 3244 for (size_t m = 0; m < node_num; m++) { 3245 if (m != i && 3246 is_node_in_configured_nodes(nindex_to_node()->at(m)) && 3247 is_node_in_bound_nodes(nindex_to_node()->at(m))) { 3248 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m)); 3249 // If a closest node is found, update. There is always at least one 3250 // configured and bound node in the system so there is always at least 3251 // one node close. 3252 if (distance != 0 && distance < closest_distance) { 3253 closest_distance = distance; 3254 closest_node = nindex_to_node()->at(m); 3255 } 3256 } 3257 } 3258 } else { 3259 // Current node is already a configured node. 3260 closest_node = nindex_to_node()->at(i); 3261 } 3262 3263 // Get cpus from the original node and map them to the closest node. If node 3264 // is a configured node (not a memory-less node), then original node and 3265 // closest node are the same. 3266 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 3267 for (size_t j = 0; j < cpu_map_valid_size; j++) { 3268 if (cpu_map[j] != 0) { 3269 for (size_t k = 0; k < BitsPerCLong; k++) { 3270 if (cpu_map[j] & (1UL << k)) { 3271 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node); 3272 } 3273 } 3274 } 3275 } 3276 } 3277 } 3278 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 3279 } 3280 3281 int os::Linux::get_node_by_cpu(int cpu_id) { 3282 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 3283 return cpu_to_node()->at(cpu_id); 3284 } 3285 return -1; 3286 } 3287 3288 GrowableArray<int>* os::Linux::_cpu_to_node; 3289 GrowableArray<int>* os::Linux::_nindex_to_node; 3290 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 3291 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 3292 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 3293 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes; 3294 os::Linux::numa_available_func_t os::Linux::_numa_available; 3295 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 3296 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 3297 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2; 3298 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 3299 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset; 3300 os::Linux::numa_distance_func_t os::Linux::_numa_distance; 3301 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind; 3302 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask; 3303 os::Linux::numa_move_pages_func_t os::Linux::_numa_move_pages; 3304 os::Linux::numa_set_preferred_func_t os::Linux::_numa_set_preferred; 3305 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy; 3306 unsigned long* os::Linux::_numa_all_nodes; 3307 struct bitmask* os::Linux::_numa_all_nodes_ptr; 3308 struct bitmask* os::Linux::_numa_nodes_ptr; 3309 struct bitmask* os::Linux::_numa_interleave_bitmask; 3310 struct bitmask* os::Linux::_numa_membind_bitmask; 3311 3312 bool os::pd_uncommit_memory(char* addr, size_t size) { 3313 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 3314 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 3315 return res != (uintptr_t) MAP_FAILED; 3316 } 3317 3318 static address get_stack_commited_bottom(address bottom, size_t size) { 3319 address nbot = bottom; 3320 address ntop = bottom + size; 3321 3322 size_t page_sz = os::vm_page_size(); 3323 unsigned pages = size / page_sz; 3324 3325 unsigned char vec[1]; 3326 unsigned imin = 1, imax = pages + 1, imid; 3327 int mincore_return_value = 0; 3328 3329 assert(imin <= imax, "Unexpected page size"); 3330 3331 while (imin < imax) { 3332 imid = (imax + imin) / 2; 3333 nbot = ntop - (imid * page_sz); 3334 3335 // Use a trick with mincore to check whether the page is mapped or not. 3336 // mincore sets vec to 1 if page resides in memory and to 0 if page 3337 // is swapped output but if page we are asking for is unmapped 3338 // it returns -1,ENOMEM 3339 mincore_return_value = mincore(nbot, page_sz, vec); 3340 3341 if (mincore_return_value == -1) { 3342 // Page is not mapped go up 3343 // to find first mapped page 3344 if (errno != EAGAIN) { 3345 assert(errno == ENOMEM, "Unexpected mincore errno"); 3346 imax = imid; 3347 } 3348 } else { 3349 // Page is mapped go down 3350 // to find first not mapped page 3351 imin = imid + 1; 3352 } 3353 } 3354 3355 nbot = nbot + page_sz; 3356 3357 // Adjust stack bottom one page up if last checked page is not mapped 3358 if (mincore_return_value == -1) { 3359 nbot = nbot + page_sz; 3360 } 3361 3362 return nbot; 3363 } 3364 3365 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) { 3366 int mincore_return_value; 3367 const size_t stripe = 1024; // query this many pages each time 3368 unsigned char vec[stripe + 1]; 3369 // set a guard 3370 vec[stripe] = 'X'; 3371 3372 const size_t page_sz = os::vm_page_size(); 3373 size_t pages = size / page_sz; 3374 3375 assert(is_aligned(start, page_sz), "Start address must be page aligned"); 3376 assert(is_aligned(size, page_sz), "Size must be page aligned"); 3377 3378 committed_start = NULL; 3379 3380 int loops = (pages + stripe - 1) / stripe; 3381 int committed_pages = 0; 3382 address loop_base = start; 3383 bool found_range = false; 3384 3385 for (int index = 0; index < loops && !found_range; index ++) { 3386 assert(pages > 0, "Nothing to do"); 3387 int pages_to_query = (pages >= stripe) ? stripe : pages; 3388 pages -= pages_to_query; 3389 3390 // Get stable read 3391 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN); 3392 3393 // During shutdown, some memory goes away without properly notifying NMT, 3394 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object. 3395 // Bailout and return as not committed for now. 3396 if (mincore_return_value == -1 && errno == ENOMEM) { 3397 return false; 3398 } 3399 3400 assert(vec[stripe] == 'X', "overflow guard"); 3401 assert(mincore_return_value == 0, "Range must be valid"); 3402 // Process this stripe 3403 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) { 3404 if ((vec[vecIdx] & 0x01) == 0) { // not committed 3405 // End of current contiguous region 3406 if (committed_start != NULL) { 3407 found_range = true; 3408 break; 3409 } 3410 } else { // committed 3411 // Start of region 3412 if (committed_start == NULL) { 3413 committed_start = loop_base + page_sz * vecIdx; 3414 } 3415 committed_pages ++; 3416 } 3417 } 3418 3419 loop_base += pages_to_query * page_sz; 3420 } 3421 3422 if (committed_start != NULL) { 3423 assert(committed_pages > 0, "Must have committed region"); 3424 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has"); 3425 assert(committed_start >= start && committed_start < start + size, "Out of range"); 3426 committed_size = page_sz * committed_pages; 3427 return true; 3428 } else { 3429 assert(committed_pages == 0, "Should not have committed region"); 3430 return false; 3431 } 3432 } 3433 3434 3435 // Linux uses a growable mapping for the stack, and if the mapping for 3436 // the stack guard pages is not removed when we detach a thread the 3437 // stack cannot grow beyond the pages where the stack guard was 3438 // mapped. If at some point later in the process the stack expands to 3439 // that point, the Linux kernel cannot expand the stack any further 3440 // because the guard pages are in the way, and a segfault occurs. 3441 // 3442 // However, it's essential not to split the stack region by unmapping 3443 // a region (leaving a hole) that's already part of the stack mapping, 3444 // so if the stack mapping has already grown beyond the guard pages at 3445 // the time we create them, we have to truncate the stack mapping. 3446 // So, we need to know the extent of the stack mapping when 3447 // create_stack_guard_pages() is called. 3448 3449 // We only need this for stacks that are growable: at the time of 3450 // writing thread stacks don't use growable mappings (i.e. those 3451 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3452 // only applies to the main thread. 3453 3454 // If the (growable) stack mapping already extends beyond the point 3455 // where we're going to put our guard pages, truncate the mapping at 3456 // that point by munmap()ping it. This ensures that when we later 3457 // munmap() the guard pages we don't leave a hole in the stack 3458 // mapping. This only affects the main/primordial thread 3459 3460 bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3461 if (os::is_primordial_thread()) { 3462 // As we manually grow stack up to bottom inside create_attached_thread(), 3463 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3464 // we don't need to do anything special. 3465 // Check it first, before calling heavy function. 3466 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3467 unsigned char vec[1]; 3468 3469 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3470 // Fallback to slow path on all errors, including EAGAIN 3471 stack_extent = (uintptr_t) get_stack_commited_bottom( 3472 os::Linux::initial_thread_stack_bottom(), 3473 (size_t)addr - stack_extent); 3474 } 3475 3476 if (stack_extent < (uintptr_t)addr) { 3477 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3478 } 3479 } 3480 3481 return os::commit_memory(addr, size, !ExecMem); 3482 } 3483 3484 // If this is a growable mapping, remove the guard pages entirely by 3485 // munmap()ping them. If not, just call uncommit_memory(). This only 3486 // affects the main/primordial thread, but guard against future OS changes. 3487 // It's safe to always unmap guard pages for primordial thread because we 3488 // always place it right after end of the mapped region. 3489 3490 bool os::remove_stack_guard_pages(char* addr, size_t size) { 3491 uintptr_t stack_extent, stack_base; 3492 3493 if (os::is_primordial_thread()) { 3494 return ::munmap(addr, size) == 0; 3495 } 3496 3497 return os::uncommit_memory(addr, size); 3498 } 3499 3500 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3501 // at 'requested_addr'. If there are existing memory mappings at the same 3502 // location, however, they will be overwritten. If 'fixed' is false, 3503 // 'requested_addr' is only treated as a hint, the return value may or 3504 // may not start from the requested address. Unlike Linux mmap(), this 3505 // function returns NULL to indicate failure. 3506 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3507 char * addr; 3508 int flags; 3509 3510 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3511 if (fixed) { 3512 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3513 flags |= MAP_FIXED; 3514 } 3515 3516 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3517 // touch an uncommitted page. Otherwise, the read/write might 3518 // succeed if we have enough swap space to back the physical page. 3519 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3520 flags, -1, 0); 3521 3522 return addr == MAP_FAILED ? NULL : addr; 3523 } 3524 3525 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address 3526 // (req_addr != NULL) or with a given alignment. 3527 // - bytes shall be a multiple of alignment. 3528 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3529 // - alignment sets the alignment at which memory shall be allocated. 3530 // It must be a multiple of allocation granularity. 3531 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3532 // req_addr or NULL. 3533 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) { 3534 3535 size_t extra_size = bytes; 3536 if (req_addr == NULL && alignment > 0) { 3537 extra_size += alignment; 3538 } 3539 3540 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE, 3541 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 3542 -1, 0); 3543 if (start == MAP_FAILED) { 3544 start = NULL; 3545 } else { 3546 if (req_addr != NULL) { 3547 if (start != req_addr) { 3548 ::munmap(start, extra_size); 3549 start = NULL; 3550 } 3551 } else { 3552 char* const start_aligned = align_up(start, alignment); 3553 char* const end_aligned = start_aligned + bytes; 3554 char* const end = start + extra_size; 3555 if (start_aligned > start) { 3556 ::munmap(start, start_aligned - start); 3557 } 3558 if (end_aligned < end) { 3559 ::munmap(end_aligned, end - end_aligned); 3560 } 3561 start = start_aligned; 3562 } 3563 } 3564 return start; 3565 } 3566 3567 static int anon_munmap(char * addr, size_t size) { 3568 return ::munmap(addr, size) == 0; 3569 } 3570 3571 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3572 size_t alignment_hint) { 3573 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3574 } 3575 3576 bool os::pd_release_memory(char* addr, size_t size) { 3577 return anon_munmap(addr, size); 3578 } 3579 3580 static bool linux_mprotect(char* addr, size_t size, int prot) { 3581 // Linux wants the mprotect address argument to be page aligned. 3582 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size()); 3583 3584 // According to SUSv3, mprotect() should only be used with mappings 3585 // established by mmap(), and mmap() always maps whole pages. Unaligned 3586 // 'addr' likely indicates problem in the VM (e.g. trying to change 3587 // protection of malloc'ed or statically allocated memory). Check the 3588 // caller if you hit this assert. 3589 assert(addr == bottom, "sanity check"); 3590 3591 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3592 Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot); 3593 return ::mprotect(bottom, size, prot) == 0; 3594 } 3595 3596 // Set protections specified 3597 bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3598 bool is_committed) { 3599 unsigned int p = 0; 3600 switch (prot) { 3601 case MEM_PROT_NONE: p = PROT_NONE; break; 3602 case MEM_PROT_READ: p = PROT_READ; break; 3603 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3604 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3605 default: 3606 ShouldNotReachHere(); 3607 } 3608 // is_committed is unused. 3609 return linux_mprotect(addr, bytes, p); 3610 } 3611 3612 bool os::guard_memory(char* addr, size_t size) { 3613 return linux_mprotect(addr, size, PROT_NONE); 3614 } 3615 3616 bool os::unguard_memory(char* addr, size_t size) { 3617 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3618 } 3619 3620 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3621 size_t page_size) { 3622 bool result = false; 3623 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3624 MAP_ANONYMOUS|MAP_PRIVATE, 3625 -1, 0); 3626 if (p != MAP_FAILED) { 3627 void *aligned_p = align_up(p, page_size); 3628 3629 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3630 3631 munmap(p, page_size * 2); 3632 } 3633 3634 if (warn && !result) { 3635 warning("TransparentHugePages is not supported by the operating system."); 3636 } 3637 3638 return result; 3639 } 3640 3641 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3642 bool result = false; 3643 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3644 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3645 -1, 0); 3646 3647 if (p != MAP_FAILED) { 3648 // We don't know if this really is a huge page or not. 3649 FILE *fp = fopen("/proc/self/maps", "r"); 3650 if (fp) { 3651 while (!feof(fp)) { 3652 char chars[257]; 3653 long x = 0; 3654 if (fgets(chars, sizeof(chars), fp)) { 3655 if (sscanf(chars, "%lx-%*x", &x) == 1 3656 && x == (long)p) { 3657 if (strstr (chars, "hugepage")) { 3658 result = true; 3659 break; 3660 } 3661 } 3662 } 3663 } 3664 fclose(fp); 3665 } 3666 munmap(p, page_size); 3667 } 3668 3669 if (warn && !result) { 3670 warning("HugeTLBFS is not supported by the operating system."); 3671 } 3672 3673 return result; 3674 } 3675 3676 // From the coredump_filter documentation: 3677 // 3678 // - (bit 0) anonymous private memory 3679 // - (bit 1) anonymous shared memory 3680 // - (bit 2) file-backed private memory 3681 // - (bit 3) file-backed shared memory 3682 // - (bit 4) ELF header pages in file-backed private memory areas (it is 3683 // effective only if the bit 2 is cleared) 3684 // - (bit 5) hugetlb private memory 3685 // - (bit 6) hugetlb shared memory 3686 // - (bit 7) dax private memory 3687 // - (bit 8) dax shared memory 3688 // 3689 static void set_coredump_filter(CoredumpFilterBit bit) { 3690 FILE *f; 3691 long cdm; 3692 3693 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3694 return; 3695 } 3696 3697 if (fscanf(f, "%lx", &cdm) != 1) { 3698 fclose(f); 3699 return; 3700 } 3701 3702 long saved_cdm = cdm; 3703 rewind(f); 3704 cdm |= bit; 3705 3706 if (cdm != saved_cdm) { 3707 fprintf(f, "%#lx", cdm); 3708 } 3709 3710 fclose(f); 3711 } 3712 3713 // Large page support 3714 3715 static size_t _large_page_size = 0; 3716 3717 size_t os::Linux::find_large_page_size() { 3718 size_t large_page_size = 0; 3719 3720 // large_page_size on Linux is used to round up heap size. x86 uses either 3721 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3722 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3723 // page as large as 256M. 3724 // 3725 // Here we try to figure out page size by parsing /proc/meminfo and looking 3726 // for a line with the following format: 3727 // Hugepagesize: 2048 kB 3728 // 3729 // If we can't determine the value (e.g. /proc is not mounted, or the text 3730 // format has been changed), we'll use the largest page size supported by 3731 // the processor. 3732 3733 #ifndef ZERO 3734 large_page_size = 3735 AARCH64_ONLY(2 * M) 3736 AMD64_ONLY(2 * M) 3737 ARM32_ONLY(2 * M) 3738 IA32_ONLY(4 * M) 3739 IA64_ONLY(256 * M) 3740 PPC_ONLY(4 * M) 3741 S390_ONLY(1 * M) 3742 SPARC_ONLY(4 * M); 3743 #endif // ZERO 3744 3745 FILE *fp = fopen("/proc/meminfo", "r"); 3746 if (fp) { 3747 while (!feof(fp)) { 3748 int x = 0; 3749 char buf[16]; 3750 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3751 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3752 large_page_size = x * K; 3753 break; 3754 } 3755 } else { 3756 // skip to next line 3757 for (;;) { 3758 int ch = fgetc(fp); 3759 if (ch == EOF || ch == (int)'\n') break; 3760 } 3761 } 3762 } 3763 fclose(fp); 3764 } 3765 3766 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3767 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3768 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3769 proper_unit_for_byte_size(large_page_size)); 3770 } 3771 3772 return large_page_size; 3773 } 3774 3775 size_t os::Linux::setup_large_page_size() { 3776 _large_page_size = Linux::find_large_page_size(); 3777 const size_t default_page_size = (size_t)Linux::page_size(); 3778 if (_large_page_size > default_page_size) { 3779 _page_sizes[0] = _large_page_size; 3780 _page_sizes[1] = default_page_size; 3781 _page_sizes[2] = 0; 3782 } 3783 3784 return _large_page_size; 3785 } 3786 3787 bool os::Linux::setup_large_page_type(size_t page_size) { 3788 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3789 FLAG_IS_DEFAULT(UseSHM) && 3790 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3791 3792 // The type of large pages has not been specified by the user. 3793 3794 // Try UseHugeTLBFS and then UseSHM. 3795 UseHugeTLBFS = UseSHM = true; 3796 3797 // Don't try UseTransparentHugePages since there are known 3798 // performance issues with it turned on. This might change in the future. 3799 UseTransparentHugePages = false; 3800 } 3801 3802 if (UseTransparentHugePages) { 3803 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3804 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3805 UseHugeTLBFS = false; 3806 UseSHM = false; 3807 return true; 3808 } 3809 UseTransparentHugePages = false; 3810 } 3811 3812 if (UseHugeTLBFS) { 3813 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3814 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3815 UseSHM = false; 3816 return true; 3817 } 3818 UseHugeTLBFS = false; 3819 } 3820 3821 return UseSHM; 3822 } 3823 3824 void os::large_page_init() { 3825 if (!UseLargePages && 3826 !UseTransparentHugePages && 3827 !UseHugeTLBFS && 3828 !UseSHM) { 3829 // Not using large pages. 3830 return; 3831 } 3832 3833 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3834 // The user explicitly turned off large pages. 3835 // Ignore the rest of the large pages flags. 3836 UseTransparentHugePages = false; 3837 UseHugeTLBFS = false; 3838 UseSHM = false; 3839 return; 3840 } 3841 3842 size_t large_page_size = Linux::setup_large_page_size(); 3843 UseLargePages = Linux::setup_large_page_type(large_page_size); 3844 3845 set_coredump_filter(LARGEPAGES_BIT); 3846 } 3847 3848 #ifndef SHM_HUGETLB 3849 #define SHM_HUGETLB 04000 3850 #endif 3851 3852 #define shm_warning_format(format, ...) \ 3853 do { \ 3854 if (UseLargePages && \ 3855 (!FLAG_IS_DEFAULT(UseLargePages) || \ 3856 !FLAG_IS_DEFAULT(UseSHM) || \ 3857 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \ 3858 warning(format, __VA_ARGS__); \ 3859 } \ 3860 } while (0) 3861 3862 #define shm_warning(str) shm_warning_format("%s", str) 3863 3864 #define shm_warning_with_errno(str) \ 3865 do { \ 3866 int err = errno; \ 3867 shm_warning_format(str " (error = %d)", err); \ 3868 } while (0) 3869 3870 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) { 3871 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment"); 3872 3873 if (!is_aligned(alignment, SHMLBA)) { 3874 assert(false, "Code below assumes that alignment is at least SHMLBA aligned"); 3875 return NULL; 3876 } 3877 3878 // To ensure that we get 'alignment' aligned memory from shmat, 3879 // we pre-reserve aligned virtual memory and then attach to that. 3880 3881 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL); 3882 if (pre_reserved_addr == NULL) { 3883 // Couldn't pre-reserve aligned memory. 3884 shm_warning("Failed to pre-reserve aligned memory for shmat."); 3885 return NULL; 3886 } 3887 3888 // SHM_REMAP is needed to allow shmat to map over an existing mapping. 3889 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP); 3890 3891 if ((intptr_t)addr == -1) { 3892 int err = errno; 3893 shm_warning_with_errno("Failed to attach shared memory."); 3894 3895 assert(err != EACCES, "Unexpected error"); 3896 assert(err != EIDRM, "Unexpected error"); 3897 assert(err != EINVAL, "Unexpected error"); 3898 3899 // Since we don't know if the kernel unmapped the pre-reserved memory area 3900 // we can't unmap it, since that would potentially unmap memory that was 3901 // mapped from other threads. 3902 return NULL; 3903 } 3904 3905 return addr; 3906 } 3907 3908 static char* shmat_at_address(int shmid, char* req_addr) { 3909 if (!is_aligned(req_addr, SHMLBA)) { 3910 assert(false, "Requested address needs to be SHMLBA aligned"); 3911 return NULL; 3912 } 3913 3914 char* addr = (char*)shmat(shmid, req_addr, 0); 3915 3916 if ((intptr_t)addr == -1) { 3917 shm_warning_with_errno("Failed to attach shared memory."); 3918 return NULL; 3919 } 3920 3921 return addr; 3922 } 3923 3924 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) { 3925 // If a req_addr has been provided, we assume that the caller has already aligned the address. 3926 if (req_addr != NULL) { 3927 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size"); 3928 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment"); 3929 return shmat_at_address(shmid, req_addr); 3930 } 3931 3932 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically 3933 // return large page size aligned memory addresses when req_addr == NULL. 3934 // However, if the alignment is larger than the large page size, we have 3935 // to manually ensure that the memory returned is 'alignment' aligned. 3936 if (alignment > os::large_page_size()) { 3937 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size"); 3938 return shmat_with_alignment(shmid, bytes, alignment); 3939 } else { 3940 return shmat_at_address(shmid, NULL); 3941 } 3942 } 3943 3944 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3945 char* req_addr, bool exec) { 3946 // "exec" is passed in but not used. Creating the shared image for 3947 // the code cache doesn't have an SHM_X executable permission to check. 3948 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3949 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3950 assert(is_aligned(req_addr, alignment), "Unaligned address"); 3951 3952 if (!is_aligned(bytes, os::large_page_size())) { 3953 return NULL; // Fallback to small pages. 3954 } 3955 3956 // Create a large shared memory region to attach to based on size. 3957 // Currently, size is the total size of the heap. 3958 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3959 if (shmid == -1) { 3960 // Possible reasons for shmget failure: 3961 // 1. shmmax is too small for Java heap. 3962 // > check shmmax value: cat /proc/sys/kernel/shmmax 3963 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3964 // 2. not enough large page memory. 3965 // > check available large pages: cat /proc/meminfo 3966 // > increase amount of large pages: 3967 // echo new_value > /proc/sys/vm/nr_hugepages 3968 // Note 1: different Linux may use different name for this property, 3969 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3970 // Note 2: it's possible there's enough physical memory available but 3971 // they are so fragmented after a long run that they can't 3972 // coalesce into large pages. Try to reserve large pages when 3973 // the system is still "fresh". 3974 shm_warning_with_errno("Failed to reserve shared memory."); 3975 return NULL; 3976 } 3977 3978 // Attach to the region. 3979 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr); 3980 3981 // Remove shmid. If shmat() is successful, the actual shared memory segment 3982 // will be deleted when it's detached by shmdt() or when the process 3983 // terminates. If shmat() is not successful this will remove the shared 3984 // segment immediately. 3985 shmctl(shmid, IPC_RMID, NULL); 3986 3987 return addr; 3988 } 3989 3990 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3991 int error) { 3992 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3993 3994 bool warn_on_failure = UseLargePages && 3995 (!FLAG_IS_DEFAULT(UseLargePages) || 3996 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3997 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3998 3999 if (warn_on_failure) { 4000 char msg[128]; 4001 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 4002 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 4003 warning("%s", msg); 4004 } 4005 } 4006 4007 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 4008 char* req_addr, 4009 bool exec) { 4010 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 4011 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size"); 4012 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 4013 4014 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 4015 char* addr = (char*)::mmap(req_addr, bytes, prot, 4016 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 4017 -1, 0); 4018 4019 if (addr == MAP_FAILED) { 4020 warn_on_large_pages_failure(req_addr, bytes, errno); 4021 return NULL; 4022 } 4023 4024 assert(is_aligned(addr, os::large_page_size()), "Must be"); 4025 4026 return addr; 4027 } 4028 4029 // Reserve memory using mmap(MAP_HUGETLB). 4030 // - bytes shall be a multiple of alignment. 4031 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 4032 // - alignment sets the alignment at which memory shall be allocated. 4033 // It must be a multiple of allocation granularity. 4034 // Returns address of memory or NULL. If req_addr was not NULL, will only return 4035 // req_addr or NULL. 4036 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 4037 size_t alignment, 4038 char* req_addr, 4039 bool exec) { 4040 size_t large_page_size = os::large_page_size(); 4041 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 4042 4043 assert(is_aligned(req_addr, alignment), "Must be"); 4044 assert(is_aligned(bytes, alignment), "Must be"); 4045 4046 // First reserve - but not commit - the address range in small pages. 4047 char* const start = anon_mmap_aligned(bytes, alignment, req_addr); 4048 4049 if (start == NULL) { 4050 return NULL; 4051 } 4052 4053 assert(is_aligned(start, alignment), "Must be"); 4054 4055 char* end = start + bytes; 4056 4057 // Find the regions of the allocated chunk that can be promoted to large pages. 4058 char* lp_start = align_up(start, large_page_size); 4059 char* lp_end = align_down(end, large_page_size); 4060 4061 size_t lp_bytes = lp_end - lp_start; 4062 4063 assert(is_aligned(lp_bytes, large_page_size), "Must be"); 4064 4065 if (lp_bytes == 0) { 4066 // The mapped region doesn't even span the start and the end of a large page. 4067 // Fall back to allocate a non-special area. 4068 ::munmap(start, end - start); 4069 return NULL; 4070 } 4071 4072 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 4073 4074 void* result; 4075 4076 // Commit small-paged leading area. 4077 if (start != lp_start) { 4078 result = ::mmap(start, lp_start - start, prot, 4079 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 4080 -1, 0); 4081 if (result == MAP_FAILED) { 4082 ::munmap(lp_start, end - lp_start); 4083 return NULL; 4084 } 4085 } 4086 4087 // Commit large-paged area. 4088 result = ::mmap(lp_start, lp_bytes, prot, 4089 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 4090 -1, 0); 4091 if (result == MAP_FAILED) { 4092 warn_on_large_pages_failure(lp_start, lp_bytes, errno); 4093 // If the mmap above fails, the large pages region will be unmapped and we 4094 // have regions before and after with small pages. Release these regions. 4095 // 4096 // | mapped | unmapped | mapped | 4097 // ^ ^ ^ ^ 4098 // start lp_start lp_end end 4099 // 4100 ::munmap(start, lp_start - start); 4101 ::munmap(lp_end, end - lp_end); 4102 return NULL; 4103 } 4104 4105 // Commit small-paged trailing area. 4106 if (lp_end != end) { 4107 result = ::mmap(lp_end, end - lp_end, prot, 4108 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 4109 -1, 0); 4110 if (result == MAP_FAILED) { 4111 ::munmap(start, lp_end - start); 4112 return NULL; 4113 } 4114 } 4115 4116 return start; 4117 } 4118 4119 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 4120 size_t alignment, 4121 char* req_addr, 4122 bool exec) { 4123 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 4124 assert(is_aligned(req_addr, alignment), "Must be"); 4125 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be"); 4126 assert(is_power_of_2(os::large_page_size()), "Must be"); 4127 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 4128 4129 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 4130 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 4131 } else { 4132 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 4133 } 4134 } 4135 4136 char* os::reserve_memory_special(size_t bytes, size_t alignment, 4137 char* req_addr, bool exec) { 4138 assert(UseLargePages, "only for large pages"); 4139 4140 char* addr; 4141 if (UseSHM) { 4142 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 4143 } else { 4144 assert(UseHugeTLBFS, "must be"); 4145 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 4146 } 4147 4148 if (addr != NULL) { 4149 if (UseNUMAInterleaving) { 4150 numa_make_global(addr, bytes); 4151 } 4152 4153 // The memory is committed 4154 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 4155 } 4156 4157 return addr; 4158 } 4159 4160 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 4161 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 4162 return shmdt(base) == 0; 4163 } 4164 4165 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 4166 return pd_release_memory(base, bytes); 4167 } 4168 4169 bool os::release_memory_special(char* base, size_t bytes) { 4170 bool res; 4171 if (MemTracker::tracking_level() > NMT_minimal) { 4172 Tracker tkr(Tracker::release); 4173 res = os::Linux::release_memory_special_impl(base, bytes); 4174 if (res) { 4175 tkr.record((address)base, bytes); 4176 } 4177 4178 } else { 4179 res = os::Linux::release_memory_special_impl(base, bytes); 4180 } 4181 return res; 4182 } 4183 4184 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 4185 assert(UseLargePages, "only for large pages"); 4186 bool res; 4187 4188 if (UseSHM) { 4189 res = os::Linux::release_memory_special_shm(base, bytes); 4190 } else { 4191 assert(UseHugeTLBFS, "must be"); 4192 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 4193 } 4194 return res; 4195 } 4196 4197 size_t os::large_page_size() { 4198 return _large_page_size; 4199 } 4200 4201 // With SysV SHM the entire memory region must be allocated as shared 4202 // memory. 4203 // HugeTLBFS allows application to commit large page memory on demand. 4204 // However, when committing memory with HugeTLBFS fails, the region 4205 // that was supposed to be committed will lose the old reservation 4206 // and allow other threads to steal that memory region. Because of this 4207 // behavior we can't commit HugeTLBFS memory. 4208 bool os::can_commit_large_page_memory() { 4209 return UseTransparentHugePages; 4210 } 4211 4212 bool os::can_execute_large_page_memory() { 4213 return UseTransparentHugePages || UseHugeTLBFS; 4214 } 4215 4216 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) { 4217 assert(file_desc >= 0, "file_desc is not valid"); 4218 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr); 4219 if (result != NULL) { 4220 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) { 4221 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory")); 4222 } 4223 } 4224 return result; 4225 } 4226 4227 // Reserve memory at an arbitrary address, only if that area is 4228 // available (and not reserved for something else). 4229 4230 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 4231 // Assert only that the size is a multiple of the page size, since 4232 // that's all that mmap requires, and since that's all we really know 4233 // about at this low abstraction level. If we need higher alignment, 4234 // we can either pass an alignment to this method or verify alignment 4235 // in one of the methods further up the call chain. See bug 5044738. 4236 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 4237 4238 // Repeatedly allocate blocks until the block is allocated at the 4239 // right spot. 4240 4241 // Linux mmap allows caller to pass an address as hint; give it a try first, 4242 // if kernel honors the hint then we can return immediately. 4243 char * addr = anon_mmap(requested_addr, bytes, false); 4244 if (addr == requested_addr) { 4245 return requested_addr; 4246 } 4247 4248 if (addr != NULL) { 4249 // mmap() is successful but it fails to reserve at the requested address 4250 anon_munmap(addr, bytes); 4251 } 4252 4253 return NULL; 4254 } 4255 4256 // Sleep forever; naked call to OS-specific sleep; use with CAUTION 4257 void os::infinite_sleep() { 4258 while (true) { // sleep forever ... 4259 ::sleep(100); // ... 100 seconds at a time 4260 } 4261 } 4262 4263 // Used to convert frequent JVM_Yield() to nops 4264 bool os::dont_yield() { 4265 return DontYieldALot; 4266 } 4267 4268 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will 4269 // actually give up the CPU. Since skip buddy (v2.6.28): 4270 // 4271 // * Sets the yielding task as skip buddy for current CPU's run queue. 4272 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task). 4273 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy). 4274 // 4275 // An alternative is calling os::naked_short_nanosleep with a small number to avoid 4276 // getting re-scheduled immediately. 4277 // 4278 void os::naked_yield() { 4279 sched_yield(); 4280 } 4281 4282 //////////////////////////////////////////////////////////////////////////////// 4283 // thread priority support 4284 4285 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 4286 // only supports dynamic priority, static priority must be zero. For real-time 4287 // applications, Linux supports SCHED_RR which allows static priority (1-99). 4288 // However, for large multi-threaded applications, SCHED_RR is not only slower 4289 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 4290 // of 5 runs - Sep 2005). 4291 // 4292 // The following code actually changes the niceness of kernel-thread/LWP. It 4293 // has an assumption that setpriority() only modifies one kernel-thread/LWP, 4294 // not the entire user process, and user level threads are 1:1 mapped to kernel 4295 // threads. It has always been the case, but could change in the future. For 4296 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). 4297 // It is only used when ThreadPriorityPolicy=1 and may require system level permission 4298 // (e.g., root privilege or CAP_SYS_NICE capability). 4299 4300 int os::java_to_os_priority[CriticalPriority + 1] = { 4301 19, // 0 Entry should never be used 4302 4303 4, // 1 MinPriority 4304 3, // 2 4305 2, // 3 4306 4307 1, // 4 4308 0, // 5 NormPriority 4309 -1, // 6 4310 4311 -2, // 7 4312 -3, // 8 4313 -4, // 9 NearMaxPriority 4314 4315 -5, // 10 MaxPriority 4316 4317 -5 // 11 CriticalPriority 4318 }; 4319 4320 static int prio_init() { 4321 if (ThreadPriorityPolicy == 1) { 4322 if (geteuid() != 0) { 4323 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 4324 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \ 4325 "e.g., being the root user. If the necessary permission is not " \ 4326 "possessed, changes to priority will be silently ignored."); 4327 } 4328 } 4329 } 4330 if (UseCriticalJavaThreadPriority) { 4331 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 4332 } 4333 return 0; 4334 } 4335 4336 OSReturn os::set_native_priority(Thread* thread, int newpri) { 4337 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 4338 4339 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 4340 return (ret == 0) ? OS_OK : OS_ERR; 4341 } 4342 4343 OSReturn os::get_native_priority(const Thread* const thread, 4344 int *priority_ptr) { 4345 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 4346 *priority_ptr = java_to_os_priority[NormPriority]; 4347 return OS_OK; 4348 } 4349 4350 errno = 0; 4351 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 4352 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 4353 } 4354 4355 //////////////////////////////////////////////////////////////////////////////// 4356 // suspend/resume support 4357 4358 // The low-level signal-based suspend/resume support is a remnant from the 4359 // old VM-suspension that used to be for java-suspension, safepoints etc, 4360 // within hotspot. Currently used by JFR's OSThreadSampler 4361 // 4362 // The remaining code is greatly simplified from the more general suspension 4363 // code that used to be used. 4364 // 4365 // The protocol is quite simple: 4366 // - suspend: 4367 // - sends a signal to the target thread 4368 // - polls the suspend state of the osthread using a yield loop 4369 // - target thread signal handler (SR_handler) sets suspend state 4370 // and blocks in sigsuspend until continued 4371 // - resume: 4372 // - sets target osthread state to continue 4373 // - sends signal to end the sigsuspend loop in the SR_handler 4374 // 4375 // Note that the SR_lock plays no role in this suspend/resume protocol, 4376 // but is checked for NULL in SR_handler as a thread termination indicator. 4377 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs. 4378 // 4379 // Note that resume_clear_context() and suspend_save_context() are needed 4380 // by SR_handler(), so that fetch_frame_from_ucontext() works, 4381 // which in part is used by: 4382 // - Forte Analyzer: AsyncGetCallTrace() 4383 // - StackBanging: get_frame_at_stack_banging_point() 4384 4385 static void resume_clear_context(OSThread *osthread) { 4386 osthread->set_ucontext(NULL); 4387 osthread->set_siginfo(NULL); 4388 } 4389 4390 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 4391 ucontext_t* context) { 4392 osthread->set_ucontext(context); 4393 osthread->set_siginfo(siginfo); 4394 } 4395 4396 // Handler function invoked when a thread's execution is suspended or 4397 // resumed. We have to be careful that only async-safe functions are 4398 // called here (Note: most pthread functions are not async safe and 4399 // should be avoided.) 4400 // 4401 // Note: sigwait() is a more natural fit than sigsuspend() from an 4402 // interface point of view, but sigwait() prevents the signal hander 4403 // from being run. libpthread would get very confused by not having 4404 // its signal handlers run and prevents sigwait()'s use with the 4405 // mutex granting granting signal. 4406 // 4407 // Currently only ever called on the VMThread and JavaThreads (PC sampling) 4408 // 4409 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 4410 // Save and restore errno to avoid confusing native code with EINTR 4411 // after sigsuspend. 4412 int old_errno = errno; 4413 4414 Thread* thread = Thread::current_or_null_safe(); 4415 assert(thread != NULL, "Missing current thread in SR_handler"); 4416 4417 // On some systems we have seen signal delivery get "stuck" until the signal 4418 // mask is changed as part of thread termination. Check that the current thread 4419 // has not already terminated (via SR_lock()) - else the following assertion 4420 // will fail because the thread is no longer a JavaThread as the ~JavaThread 4421 // destructor has completed. 4422 4423 if (thread->SR_lock() == NULL) { 4424 return; 4425 } 4426 4427 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 4428 4429 OSThread* osthread = thread->osthread(); 4430 4431 os::SuspendResume::State current = osthread->sr.state(); 4432 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 4433 suspend_save_context(osthread, siginfo, context); 4434 4435 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4436 os::SuspendResume::State state = osthread->sr.suspended(); 4437 if (state == os::SuspendResume::SR_SUSPENDED) { 4438 sigset_t suspend_set; // signals for sigsuspend() 4439 sigemptyset(&suspend_set); 4440 // get current set of blocked signals and unblock resume signal 4441 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4442 sigdelset(&suspend_set, SR_signum); 4443 4444 sr_semaphore.signal(); 4445 // wait here until we are resumed 4446 while (1) { 4447 sigsuspend(&suspend_set); 4448 4449 os::SuspendResume::State result = osthread->sr.running(); 4450 if (result == os::SuspendResume::SR_RUNNING) { 4451 sr_semaphore.signal(); 4452 break; 4453 } 4454 } 4455 4456 } else if (state == os::SuspendResume::SR_RUNNING) { 4457 // request was cancelled, continue 4458 } else { 4459 ShouldNotReachHere(); 4460 } 4461 4462 resume_clear_context(osthread); 4463 } else if (current == os::SuspendResume::SR_RUNNING) { 4464 // request was cancelled, continue 4465 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4466 // ignore 4467 } else { 4468 // ignore 4469 } 4470 4471 errno = old_errno; 4472 } 4473 4474 static int SR_initialize() { 4475 struct sigaction act; 4476 char *s; 4477 4478 // Get signal number to use for suspend/resume 4479 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4480 int sig = ::strtol(s, 0, 10); 4481 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769. 4482 sig < NSIG) { // Must be legal signal and fit into sigflags[]. 4483 SR_signum = sig; 4484 } else { 4485 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.", 4486 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum); 4487 } 4488 } 4489 4490 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4491 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4492 4493 sigemptyset(&SR_sigset); 4494 sigaddset(&SR_sigset, SR_signum); 4495 4496 // Set up signal handler for suspend/resume 4497 act.sa_flags = SA_RESTART|SA_SIGINFO; 4498 act.sa_handler = (void (*)(int)) SR_handler; 4499 4500 // SR_signum is blocked by default. 4501 // 4528190 - We also need to block pthread restart signal (32 on all 4502 // supported Linux platforms). Note that LinuxThreads need to block 4503 // this signal for all threads to work properly. So we don't have 4504 // to use hard-coded signal number when setting up the mask. 4505 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4506 4507 if (sigaction(SR_signum, &act, 0) == -1) { 4508 return -1; 4509 } 4510 4511 // Save signal flag 4512 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4513 return 0; 4514 } 4515 4516 static int sr_notify(OSThread* osthread) { 4517 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4518 assert_status(status == 0, status, "pthread_kill"); 4519 return status; 4520 } 4521 4522 // "Randomly" selected value for how long we want to spin 4523 // before bailing out on suspending a thread, also how often 4524 // we send a signal to a thread we want to resume 4525 static const int RANDOMLY_LARGE_INTEGER = 1000000; 4526 static const int RANDOMLY_LARGE_INTEGER2 = 100; 4527 4528 // returns true on success and false on error - really an error is fatal 4529 // but this seems the normal response to library errors 4530 static bool do_suspend(OSThread* osthread) { 4531 assert(osthread->sr.is_running(), "thread should be running"); 4532 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4533 4534 // mark as suspended and send signal 4535 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4536 // failed to switch, state wasn't running? 4537 ShouldNotReachHere(); 4538 return false; 4539 } 4540 4541 if (sr_notify(osthread) != 0) { 4542 ShouldNotReachHere(); 4543 } 4544 4545 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4546 while (true) { 4547 if (sr_semaphore.timedwait(2)) { 4548 break; 4549 } else { 4550 // timeout 4551 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4552 if (cancelled == os::SuspendResume::SR_RUNNING) { 4553 return false; 4554 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4555 // make sure that we consume the signal on the semaphore as well 4556 sr_semaphore.wait(); 4557 break; 4558 } else { 4559 ShouldNotReachHere(); 4560 return false; 4561 } 4562 } 4563 } 4564 4565 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4566 return true; 4567 } 4568 4569 static void do_resume(OSThread* osthread) { 4570 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4571 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4572 4573 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4574 // failed to switch to WAKEUP_REQUEST 4575 ShouldNotReachHere(); 4576 return; 4577 } 4578 4579 while (true) { 4580 if (sr_notify(osthread) == 0) { 4581 if (sr_semaphore.timedwait(2)) { 4582 if (osthread->sr.is_running()) { 4583 return; 4584 } 4585 } 4586 } else { 4587 ShouldNotReachHere(); 4588 } 4589 } 4590 4591 guarantee(osthread->sr.is_running(), "Must be running!"); 4592 } 4593 4594 /////////////////////////////////////////////////////////////////////////////////// 4595 // signal handling (except suspend/resume) 4596 4597 // This routine may be used by user applications as a "hook" to catch signals. 4598 // The user-defined signal handler must pass unrecognized signals to this 4599 // routine, and if it returns true (non-zero), then the signal handler must 4600 // return immediately. If the flag "abort_if_unrecognized" is true, then this 4601 // routine will never retun false (zero), but instead will execute a VM panic 4602 // routine kill the process. 4603 // 4604 // If this routine returns false, it is OK to call it again. This allows 4605 // the user-defined signal handler to perform checks either before or after 4606 // the VM performs its own checks. Naturally, the user code would be making 4607 // a serious error if it tried to handle an exception (such as a null check 4608 // or breakpoint) that the VM was generating for its own correct operation. 4609 // 4610 // This routine may recognize any of the following kinds of signals: 4611 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4612 // It should be consulted by handlers for any of those signals. 4613 // 4614 // The caller of this routine must pass in the three arguments supplied 4615 // to the function referred to in the "sa_sigaction" (not the "sa_handler") 4616 // field of the structure passed to sigaction(). This routine assumes that 4617 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4618 // 4619 // Note that the VM will print warnings if it detects conflicting signal 4620 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4621 // 4622 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4623 siginfo_t* siginfo, 4624 void* ucontext, 4625 int abort_if_unrecognized); 4626 4627 static void signalHandler(int sig, siginfo_t* info, void* uc) { 4628 assert(info != NULL && uc != NULL, "it must be old kernel"); 4629 int orig_errno = errno; // Preserve errno value over signal handler. 4630 JVM_handle_linux_signal(sig, info, uc, true); 4631 errno = orig_errno; 4632 } 4633 4634 4635 // This boolean allows users to forward their own non-matching signals 4636 // to JVM_handle_linux_signal, harmlessly. 4637 bool os::Linux::signal_handlers_are_installed = false; 4638 4639 // For signal-chaining 4640 bool os::Linux::libjsig_is_loaded = false; 4641 typedef struct sigaction *(*get_signal_t)(int); 4642 get_signal_t os::Linux::get_signal_action = NULL; 4643 4644 struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4645 struct sigaction *actp = NULL; 4646 4647 if (libjsig_is_loaded) { 4648 // Retrieve the old signal handler from libjsig 4649 actp = (*get_signal_action)(sig); 4650 } 4651 if (actp == NULL) { 4652 // Retrieve the preinstalled signal handler from jvm 4653 actp = os::Posix::get_preinstalled_handler(sig); 4654 } 4655 4656 return actp; 4657 } 4658 4659 static bool call_chained_handler(struct sigaction *actp, int sig, 4660 siginfo_t *siginfo, void *context) { 4661 // Call the old signal handler 4662 if (actp->sa_handler == SIG_DFL) { 4663 // It's more reasonable to let jvm treat it as an unexpected exception 4664 // instead of taking the default action. 4665 return false; 4666 } else if (actp->sa_handler != SIG_IGN) { 4667 if ((actp->sa_flags & SA_NODEFER) == 0) { 4668 // automaticlly block the signal 4669 sigaddset(&(actp->sa_mask), sig); 4670 } 4671 4672 sa_handler_t hand = NULL; 4673 sa_sigaction_t sa = NULL; 4674 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4675 // retrieve the chained handler 4676 if (siginfo_flag_set) { 4677 sa = actp->sa_sigaction; 4678 } else { 4679 hand = actp->sa_handler; 4680 } 4681 4682 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4683 actp->sa_handler = SIG_DFL; 4684 } 4685 4686 // try to honor the signal mask 4687 sigset_t oset; 4688 sigemptyset(&oset); 4689 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4690 4691 // call into the chained handler 4692 if (siginfo_flag_set) { 4693 (*sa)(sig, siginfo, context); 4694 } else { 4695 (*hand)(sig); 4696 } 4697 4698 // restore the signal mask 4699 pthread_sigmask(SIG_SETMASK, &oset, NULL); 4700 } 4701 // Tell jvm's signal handler the signal is taken care of. 4702 return true; 4703 } 4704 4705 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4706 bool chained = false; 4707 // signal-chaining 4708 if (UseSignalChaining) { 4709 struct sigaction *actp = get_chained_signal_action(sig); 4710 if (actp != NULL) { 4711 chained = call_chained_handler(actp, sig, siginfo, context); 4712 } 4713 } 4714 return chained; 4715 } 4716 4717 // for diagnostic 4718 int sigflags[NSIG]; 4719 4720 int os::Linux::get_our_sigflags(int sig) { 4721 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4722 return sigflags[sig]; 4723 } 4724 4725 void os::Linux::set_our_sigflags(int sig, int flags) { 4726 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4727 if (sig > 0 && sig < NSIG) { 4728 sigflags[sig] = flags; 4729 } 4730 } 4731 4732 void os::Linux::set_signal_handler(int sig, bool set_installed) { 4733 // Check for overwrite. 4734 struct sigaction oldAct; 4735 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4736 4737 void* oldhand = oldAct.sa_sigaction 4738 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4739 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4740 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4741 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4742 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4743 if (AllowUserSignalHandlers || !set_installed) { 4744 // Do not overwrite; user takes responsibility to forward to us. 4745 return; 4746 } else if (UseSignalChaining) { 4747 // save the old handler in jvm 4748 os::Posix::save_preinstalled_handler(sig, oldAct); 4749 // libjsig also interposes the sigaction() call below and saves the 4750 // old sigaction on it own. 4751 } else { 4752 fatal("Encountered unexpected pre-existing sigaction handler " 4753 "%#lx for signal %d.", (long)oldhand, sig); 4754 } 4755 } 4756 4757 struct sigaction sigAct; 4758 sigfillset(&(sigAct.sa_mask)); 4759 sigAct.sa_handler = SIG_DFL; 4760 if (!set_installed) { 4761 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4762 } else { 4763 sigAct.sa_sigaction = signalHandler; 4764 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4765 } 4766 // Save flags, which are set by ours 4767 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4768 sigflags[sig] = sigAct.sa_flags; 4769 4770 int ret = sigaction(sig, &sigAct, &oldAct); 4771 assert(ret == 0, "check"); 4772 4773 void* oldhand2 = oldAct.sa_sigaction 4774 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4775 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4776 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4777 } 4778 4779 // install signal handlers for signals that HotSpot needs to 4780 // handle in order to support Java-level exception handling. 4781 4782 void os::Linux::install_signal_handlers() { 4783 if (!signal_handlers_are_installed) { 4784 signal_handlers_are_installed = true; 4785 4786 // signal-chaining 4787 typedef void (*signal_setting_t)(); 4788 signal_setting_t begin_signal_setting = NULL; 4789 signal_setting_t end_signal_setting = NULL; 4790 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4791 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4792 if (begin_signal_setting != NULL) { 4793 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4794 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4795 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4796 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4797 libjsig_is_loaded = true; 4798 assert(UseSignalChaining, "should enable signal-chaining"); 4799 } 4800 if (libjsig_is_loaded) { 4801 // Tell libjsig jvm is setting signal handlers 4802 (*begin_signal_setting)(); 4803 } 4804 4805 set_signal_handler(SIGSEGV, true); 4806 set_signal_handler(SIGPIPE, true); 4807 set_signal_handler(SIGBUS, true); 4808 set_signal_handler(SIGILL, true); 4809 set_signal_handler(SIGFPE, true); 4810 #if defined(PPC64) 4811 set_signal_handler(SIGTRAP, true); 4812 #endif 4813 set_signal_handler(SIGXFSZ, true); 4814 4815 if (libjsig_is_loaded) { 4816 // Tell libjsig jvm finishes setting signal handlers 4817 (*end_signal_setting)(); 4818 } 4819 4820 // We don't activate signal checker if libjsig is in place, we trust ourselves 4821 // and if UserSignalHandler is installed all bets are off. 4822 // Log that signal checking is off only if -verbose:jni is specified. 4823 if (CheckJNICalls) { 4824 if (libjsig_is_loaded) { 4825 log_debug(jni, resolve)("Info: libjsig is activated, all active signal checking is disabled"); 4826 check_signals = false; 4827 } 4828 if (AllowUserSignalHandlers) { 4829 log_debug(jni, resolve)("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4830 check_signals = false; 4831 } 4832 } 4833 } 4834 } 4835 4836 // This is the fastest way to get thread cpu time on Linux. 4837 // Returns cpu time (user+sys) for any thread, not only for current. 4838 // POSIX compliant clocks are implemented in the kernels 2.6.16+. 4839 // It might work on 2.6.10+ with a special kernel/glibc patch. 4840 // For reference, please, see IEEE Std 1003.1-2004: 4841 // http://www.unix.org/single_unix_specification 4842 4843 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4844 struct timespec tp; 4845 int rc = os::Posix::clock_gettime(clockid, &tp); 4846 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4847 4848 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4849 } 4850 4851 ///// 4852 // glibc on Linux platform uses non-documented flag 4853 // to indicate, that some special sort of signal 4854 // trampoline is used. 4855 // We will never set this flag, and we should 4856 // ignore this flag in our diagnostic 4857 #ifdef SIGNIFICANT_SIGNAL_MASK 4858 #undef SIGNIFICANT_SIGNAL_MASK 4859 #endif 4860 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4861 4862 static const char* get_signal_handler_name(address handler, 4863 char* buf, int buflen) { 4864 int offset = 0; 4865 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4866 if (found) { 4867 // skip directory names 4868 const char *p1, *p2; 4869 p1 = buf; 4870 size_t len = strlen(os::file_separator()); 4871 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4872 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4873 } else { 4874 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4875 } 4876 return buf; 4877 } 4878 4879 static void print_signal_handler(outputStream* st, int sig, 4880 char* buf, size_t buflen) { 4881 struct sigaction sa; 4882 4883 sigaction(sig, NULL, &sa); 4884 4885 // See comment for SIGNIFICANT_SIGNAL_MASK define 4886 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4887 4888 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4889 4890 address handler = (sa.sa_flags & SA_SIGINFO) 4891 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4892 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4893 4894 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4895 st->print("SIG_DFL"); 4896 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4897 st->print("SIG_IGN"); 4898 } else { 4899 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4900 } 4901 4902 st->print(", sa_mask[0]="); 4903 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4904 4905 address rh = VMError::get_resetted_sighandler(sig); 4906 // May be, handler was resetted by VMError? 4907 if (rh != NULL) { 4908 handler = rh; 4909 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4910 } 4911 4912 st->print(", sa_flags="); 4913 os::Posix::print_sa_flags(st, sa.sa_flags); 4914 4915 // Check: is it our handler? 4916 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4917 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4918 // It is our signal handler 4919 // check for flags, reset system-used one! 4920 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4921 st->print( 4922 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4923 os::Linux::get_our_sigflags(sig)); 4924 } 4925 } 4926 st->cr(); 4927 } 4928 4929 4930 #define DO_SIGNAL_CHECK(sig) \ 4931 do { \ 4932 if (!sigismember(&check_signal_done, sig)) { \ 4933 os::Linux::check_signal_handler(sig); \ 4934 } \ 4935 } while (0) 4936 4937 // This method is a periodic task to check for misbehaving JNI applications 4938 // under CheckJNI, we can add any periodic checks here 4939 4940 void os::run_periodic_checks() { 4941 if (check_signals == false) return; 4942 4943 // SEGV and BUS if overridden could potentially prevent 4944 // generation of hs*.log in the event of a crash, debugging 4945 // such a case can be very challenging, so we absolutely 4946 // check the following for a good measure: 4947 DO_SIGNAL_CHECK(SIGSEGV); 4948 DO_SIGNAL_CHECK(SIGILL); 4949 DO_SIGNAL_CHECK(SIGFPE); 4950 DO_SIGNAL_CHECK(SIGBUS); 4951 DO_SIGNAL_CHECK(SIGPIPE); 4952 DO_SIGNAL_CHECK(SIGXFSZ); 4953 #if defined(PPC64) 4954 DO_SIGNAL_CHECK(SIGTRAP); 4955 #endif 4956 4957 // ReduceSignalUsage allows the user to override these handlers 4958 // see comments at the very top and jvm_md.h 4959 if (!ReduceSignalUsage) { 4960 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4961 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4962 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4963 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4964 } 4965 4966 DO_SIGNAL_CHECK(SR_signum); 4967 } 4968 4969 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4970 4971 static os_sigaction_t os_sigaction = NULL; 4972 4973 void os::Linux::check_signal_handler(int sig) { 4974 char buf[O_BUFLEN]; 4975 address jvmHandler = NULL; 4976 4977 4978 struct sigaction act; 4979 if (os_sigaction == NULL) { 4980 // only trust the default sigaction, in case it has been interposed 4981 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4982 if (os_sigaction == NULL) return; 4983 } 4984 4985 os_sigaction(sig, (struct sigaction*)NULL, &act); 4986 4987 4988 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4989 4990 address thisHandler = (act.sa_flags & SA_SIGINFO) 4991 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4992 : CAST_FROM_FN_PTR(address, act.sa_handler); 4993 4994 4995 switch (sig) { 4996 case SIGSEGV: 4997 case SIGBUS: 4998 case SIGFPE: 4999 case SIGPIPE: 5000 case SIGILL: 5001 case SIGXFSZ: 5002 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 5003 break; 5004 5005 case SHUTDOWN1_SIGNAL: 5006 case SHUTDOWN2_SIGNAL: 5007 case SHUTDOWN3_SIGNAL: 5008 case BREAK_SIGNAL: 5009 jvmHandler = (address)user_handler(); 5010 break; 5011 5012 default: 5013 if (sig == SR_signum) { 5014 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 5015 } else { 5016 return; 5017 } 5018 break; 5019 } 5020 5021 if (thisHandler != jvmHandler) { 5022 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 5023 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 5024 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 5025 // No need to check this sig any longer 5026 sigaddset(&check_signal_done, sig); 5027 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 5028 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 5029 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 5030 exception_name(sig, buf, O_BUFLEN)); 5031 } 5032 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 5033 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 5034 tty->print("expected:"); 5035 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig)); 5036 tty->cr(); 5037 tty->print(" found:"); 5038 os::Posix::print_sa_flags(tty, act.sa_flags); 5039 tty->cr(); 5040 // No need to check this sig any longer 5041 sigaddset(&check_signal_done, sig); 5042 } 5043 5044 // Dump all the signal 5045 if (sigismember(&check_signal_done, sig)) { 5046 print_signal_handlers(tty, buf, O_BUFLEN); 5047 } 5048 } 5049 5050 extern void report_error(char* file_name, int line_no, char* title, 5051 char* format, ...); 5052 5053 // this is called _before_ most of the global arguments have been parsed 5054 void os::init(void) { 5055 char dummy; // used to get a guess on initial stack address 5056 5057 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 5058 5059 init_random(1234567); 5060 5061 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 5062 if (Linux::page_size() == -1) { 5063 fatal("os_linux.cpp: os::init: sysconf failed (%s)", 5064 os::strerror(errno)); 5065 } 5066 init_page_sizes((size_t) Linux::page_size()); 5067 5068 Linux::initialize_system_info(); 5069 5070 os::Linux::CPUPerfTicks pticks; 5071 bool res = os::Linux::get_tick_information(&pticks, -1); 5072 5073 if (res && pticks.has_steal_ticks) { 5074 has_initial_tick_info = true; 5075 initial_total_ticks = pticks.total; 5076 initial_steal_ticks = pticks.steal; 5077 } 5078 5079 // _main_thread points to the thread that created/loaded the JVM. 5080 Linux::_main_thread = pthread_self(); 5081 5082 // retrieve entry point for pthread_setname_np 5083 Linux::_pthread_setname_np = 5084 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 5085 5086 os::Posix::init(); 5087 5088 initial_time_count = javaTimeNanos(); 5089 5090 // Always warn if no monotonic clock available 5091 if (!os::Posix::supports_monotonic_clock()) { 5092 warning("No monotonic clock was available - timed services may " \ 5093 "be adversely affected if the time-of-day clock changes"); 5094 } 5095 } 5096 5097 // To install functions for atexit system call 5098 extern "C" { 5099 static void perfMemory_exit_helper() { 5100 perfMemory_exit(); 5101 } 5102 } 5103 5104 void os::pd_init_container_support() { 5105 OSContainer::init(); 5106 } 5107 5108 void os::Linux::numa_init() { 5109 5110 // Java can be invoked as 5111 // 1. Without numactl and heap will be allocated/configured on all nodes as 5112 // per the system policy. 5113 // 2. With numactl --interleave: 5114 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same 5115 // API for membind case bitmask is reset. 5116 // Interleave is only hint and Kernel can fallback to other nodes if 5117 // no memory is available on the target nodes. 5118 // 3. With numactl --membind: 5119 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for 5120 // interleave case returns bitmask of all nodes. 5121 // numa_all_nodes_ptr holds bitmask of all nodes. 5122 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct 5123 // bitmask when externally configured to run on all or fewer nodes. 5124 5125 if (!Linux::libnuma_init()) { 5126 UseNUMA = false; 5127 } else { 5128 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) { 5129 // If there's only one node (they start from 0) or if the process 5130 // is bound explicitly to a single node using membind, disable NUMA. 5131 UseNUMA = false; 5132 } else { 5133 5134 LogTarget(Info,os) log; 5135 LogStream ls(log); 5136 5137 Linux::set_configured_numa_policy(Linux::identify_numa_policy()); 5138 5139 struct bitmask* bmp = Linux::_numa_membind_bitmask; 5140 const char* numa_mode = "membind"; 5141 5142 if (Linux::is_running_in_interleave_mode()) { 5143 bmp = Linux::_numa_interleave_bitmask; 5144 numa_mode = "interleave"; 5145 } 5146 5147 ls.print("UseNUMA is enabled and invoked in '%s' mode." 5148 " Heap will be configured using NUMA memory nodes:", numa_mode); 5149 5150 for (int node = 0; node <= Linux::numa_max_node(); node++) { 5151 if (Linux::_numa_bitmask_isbitset(bmp, node)) { 5152 ls.print(" %d", node); 5153 } 5154 } 5155 } 5156 } 5157 5158 if (!UseNUMA && ForceNUMA) { 5159 UseNUMA = true; 5160 } 5161 5162 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 5163 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 5164 // we can make the adaptive lgrp chunk resizing work. If the user specified both 5165 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn 5166 // and disable adaptive resizing. 5167 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) { 5168 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, " 5169 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)"); 5170 UseAdaptiveSizePolicy = false; 5171 UseAdaptiveNUMAChunkSizing = false; 5172 } 5173 } 5174 } 5175 5176 // this is called _after_ the global arguments have been parsed 5177 jint os::init_2(void) { 5178 5179 // This could be set after os::Posix::init() but all platforms 5180 // have to set it the same so we have to mirror Solaris. 5181 DEBUG_ONLY(os::set_mutex_init_done();) 5182 5183 os::Posix::init_2(); 5184 5185 Linux::fast_thread_clock_init(); 5186 5187 // initialize suspend/resume support - must do this before signal_sets_init() 5188 if (SR_initialize() != 0) { 5189 perror("SR_initialize failed"); 5190 return JNI_ERR; 5191 } 5192 5193 Linux::signal_sets_init(); 5194 Linux::install_signal_handlers(); 5195 // Initialize data for jdk.internal.misc.Signal 5196 if (!ReduceSignalUsage) { 5197 jdk_misc_signal_init(); 5198 } 5199 5200 if (AdjustStackSizeForTLS) { 5201 get_minstack_init(); 5202 } 5203 5204 // Check and sets minimum stack sizes against command line options 5205 if (Posix::set_minimum_stack_sizes() == JNI_ERR) { 5206 return JNI_ERR; 5207 } 5208 5209 #if defined(IA32) 5210 // Need to ensure we've determined the process's initial stack to 5211 // perform the workaround 5212 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 5213 workaround_expand_exec_shield_cs_limit(); 5214 #else 5215 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher(); 5216 if (!suppress_primordial_thread_resolution) { 5217 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 5218 } 5219 #endif 5220 5221 Linux::libpthread_init(); 5222 Linux::sched_getcpu_init(); 5223 log_info(os)("HotSpot is running with %s, %s", 5224 Linux::glibc_version(), Linux::libpthread_version()); 5225 5226 if (UseNUMA) { 5227 Linux::numa_init(); 5228 } 5229 5230 if (MaxFDLimit) { 5231 // set the number of file descriptors to max. print out error 5232 // if getrlimit/setrlimit fails but continue regardless. 5233 struct rlimit nbr_files; 5234 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 5235 if (status != 0) { 5236 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno)); 5237 } else { 5238 nbr_files.rlim_cur = nbr_files.rlim_max; 5239 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 5240 if (status != 0) { 5241 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno)); 5242 } 5243 } 5244 } 5245 5246 // at-exit methods are called in the reverse order of their registration. 5247 // atexit functions are called on return from main or as a result of a 5248 // call to exit(3C). There can be only 32 of these functions registered 5249 // and atexit() does not set errno. 5250 5251 if (PerfAllowAtExitRegistration) { 5252 // only register atexit functions if PerfAllowAtExitRegistration is set. 5253 // atexit functions can be delayed until process exit time, which 5254 // can be problematic for embedded VM situations. Embedded VMs should 5255 // call DestroyJavaVM() to assure that VM resources are released. 5256 5257 // note: perfMemory_exit_helper atexit function may be removed in 5258 // the future if the appropriate cleanup code can be added to the 5259 // VM_Exit VMOperation's doit method. 5260 if (atexit(perfMemory_exit_helper) != 0) { 5261 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 5262 } 5263 } 5264 5265 // initialize thread priority policy 5266 prio_init(); 5267 5268 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) { 5269 set_coredump_filter(DAX_SHARED_BIT); 5270 } 5271 5272 if (DumpPrivateMappingsInCore) { 5273 set_coredump_filter(FILE_BACKED_PVT_BIT); 5274 } 5275 5276 if (DumpSharedMappingsInCore) { 5277 set_coredump_filter(FILE_BACKED_SHARED_BIT); 5278 } 5279 5280 return JNI_OK; 5281 } 5282 5283 // Mark the polling page as unreadable 5284 void os::make_polling_page_unreadable(void) { 5285 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 5286 fatal("Could not disable polling page"); 5287 } 5288 } 5289 5290 // Mark the polling page as readable 5291 void os::make_polling_page_readable(void) { 5292 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 5293 fatal("Could not enable polling page"); 5294 } 5295 } 5296 5297 // older glibc versions don't have this macro (which expands to 5298 // an optimized bit-counting function) so we have to roll our own 5299 #ifndef CPU_COUNT 5300 5301 static int _cpu_count(const cpu_set_t* cpus) { 5302 int count = 0; 5303 // only look up to the number of configured processors 5304 for (int i = 0; i < os::processor_count(); i++) { 5305 if (CPU_ISSET(i, cpus)) { 5306 count++; 5307 } 5308 } 5309 return count; 5310 } 5311 5312 #define CPU_COUNT(cpus) _cpu_count(cpus) 5313 5314 #endif // CPU_COUNT 5315 5316 // Get the current number of available processors for this process. 5317 // This value can change at any time during a process's lifetime. 5318 // sched_getaffinity gives an accurate answer as it accounts for cpusets. 5319 // If it appears there may be more than 1024 processors then we do a 5320 // dynamic check - see 6515172 for details. 5321 // If anything goes wrong we fallback to returning the number of online 5322 // processors - which can be greater than the number available to the process. 5323 int os::Linux::active_processor_count() { 5324 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors 5325 cpu_set_t* cpus_p = &cpus; 5326 int cpus_size = sizeof(cpu_set_t); 5327 5328 int configured_cpus = os::processor_count(); // upper bound on available cpus 5329 int cpu_count = 0; 5330 5331 // old build platforms may not support dynamic cpu sets 5332 #ifdef CPU_ALLOC 5333 5334 // To enable easy testing of the dynamic path on different platforms we 5335 // introduce a diagnostic flag: UseCpuAllocPath 5336 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) { 5337 // kernel may use a mask bigger than cpu_set_t 5338 log_trace(os)("active_processor_count: using dynamic path %s" 5339 "- configured processors: %d", 5340 UseCpuAllocPath ? "(forced) " : "", 5341 configured_cpus); 5342 cpus_p = CPU_ALLOC(configured_cpus); 5343 if (cpus_p != NULL) { 5344 cpus_size = CPU_ALLOC_SIZE(configured_cpus); 5345 // zero it just to be safe 5346 CPU_ZERO_S(cpus_size, cpus_p); 5347 } 5348 else { 5349 // failed to allocate so fallback to online cpus 5350 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 5351 log_trace(os)("active_processor_count: " 5352 "CPU_ALLOC failed (%s) - using " 5353 "online processor count: %d", 5354 os::strerror(errno), online_cpus); 5355 return online_cpus; 5356 } 5357 } 5358 else { 5359 log_trace(os)("active_processor_count: using static path - configured processors: %d", 5360 configured_cpus); 5361 } 5362 #else // CPU_ALLOC 5363 // these stubs won't be executed 5364 #define CPU_COUNT_S(size, cpus) -1 5365 #define CPU_FREE(cpus) 5366 5367 log_trace(os)("active_processor_count: only static path available - configured processors: %d", 5368 configured_cpus); 5369 #endif // CPU_ALLOC 5370 5371 // pid 0 means the current thread - which we have to assume represents the process 5372 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) { 5373 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5374 cpu_count = CPU_COUNT_S(cpus_size, cpus_p); 5375 } 5376 else { 5377 cpu_count = CPU_COUNT(cpus_p); 5378 } 5379 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count); 5380 } 5381 else { 5382 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN); 5383 warning("sched_getaffinity failed (%s)- using online processor count (%d) " 5384 "which may exceed available processors", os::strerror(errno), cpu_count); 5385 } 5386 5387 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5388 CPU_FREE(cpus_p); 5389 } 5390 5391 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check"); 5392 return cpu_count; 5393 } 5394 5395 // Determine the active processor count from one of 5396 // three different sources: 5397 // 5398 // 1. User option -XX:ActiveProcessorCount 5399 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN) 5400 // 3. extracted from cgroup cpu subsystem (shares and quotas) 5401 // 5402 // Option 1, if specified, will always override. 5403 // If the cgroup subsystem is active and configured, we 5404 // will return the min of the cgroup and option 2 results. 5405 // This is required since tools, such as numactl, that 5406 // alter cpu affinity do not update cgroup subsystem 5407 // cpuset configuration files. 5408 int os::active_processor_count() { 5409 // User has overridden the number of active processors 5410 if (ActiveProcessorCount > 0) { 5411 log_trace(os)("active_processor_count: " 5412 "active processor count set by user : %d", 5413 ActiveProcessorCount); 5414 return ActiveProcessorCount; 5415 } 5416 5417 int active_cpus; 5418 if (OSContainer::is_containerized()) { 5419 active_cpus = OSContainer::active_processor_count(); 5420 log_trace(os)("active_processor_count: determined by OSContainer: %d", 5421 active_cpus); 5422 } else { 5423 active_cpus = os::Linux::active_processor_count(); 5424 } 5425 5426 return active_cpus; 5427 } 5428 5429 uint os::processor_id() { 5430 const int id = Linux::sched_getcpu(); 5431 assert(id >= 0 && id < _processor_count, "Invalid processor id"); 5432 return (uint)id; 5433 } 5434 5435 void os::set_native_thread_name(const char *name) { 5436 if (Linux::_pthread_setname_np) { 5437 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 5438 snprintf(buf, sizeof(buf), "%s", name); 5439 buf[sizeof(buf) - 1] = '\0'; 5440 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 5441 // ERANGE should not happen; all other errors should just be ignored. 5442 assert(rc != ERANGE, "pthread_setname_np failed"); 5443 } 5444 } 5445 5446 bool os::bind_to_processor(uint processor_id) { 5447 // Not yet implemented. 5448 return false; 5449 } 5450 5451 /// 5452 5453 void os::SuspendedThreadTask::internal_do_task() { 5454 if (do_suspend(_thread->osthread())) { 5455 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 5456 do_task(context); 5457 do_resume(_thread->osthread()); 5458 } 5459 } 5460 5461 //////////////////////////////////////////////////////////////////////////////// 5462 // debug support 5463 5464 bool os::find(address addr, outputStream* st) { 5465 Dl_info dlinfo; 5466 memset(&dlinfo, 0, sizeof(dlinfo)); 5467 if (dladdr(addr, &dlinfo) != 0) { 5468 st->print(PTR_FORMAT ": ", p2i(addr)); 5469 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5470 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname, 5471 p2i(addr) - p2i(dlinfo.dli_saddr)); 5472 } else if (dlinfo.dli_fbase != NULL) { 5473 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase)); 5474 } else { 5475 st->print("<absolute address>"); 5476 } 5477 if (dlinfo.dli_fname != NULL) { 5478 st->print(" in %s", dlinfo.dli_fname); 5479 } 5480 if (dlinfo.dli_fbase != NULL) { 5481 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase)); 5482 } 5483 st->cr(); 5484 5485 if (Verbose) { 5486 // decode some bytes around the PC 5487 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5488 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5489 address lowest = (address) dlinfo.dli_sname; 5490 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5491 if (begin < lowest) begin = lowest; 5492 Dl_info dlinfo2; 5493 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5494 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 5495 end = (address) dlinfo2.dli_saddr; 5496 } 5497 Disassembler::decode(begin, end, st); 5498 } 5499 return true; 5500 } 5501 return false; 5502 } 5503 5504 //////////////////////////////////////////////////////////////////////////////// 5505 // misc 5506 5507 // This does not do anything on Linux. This is basically a hook for being 5508 // able to use structured exception handling (thread-local exception filters) 5509 // on, e.g., Win32. 5510 void 5511 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method, 5512 JavaCallArguments* args, Thread* thread) { 5513 f(value, method, args, thread); 5514 } 5515 5516 void os::print_statistics() { 5517 } 5518 5519 bool os::message_box(const char* title, const char* message) { 5520 int i; 5521 fdStream err(defaultStream::error_fd()); 5522 for (i = 0; i < 78; i++) err.print_raw("="); 5523 err.cr(); 5524 err.print_raw_cr(title); 5525 for (i = 0; i < 78; i++) err.print_raw("-"); 5526 err.cr(); 5527 err.print_raw_cr(message); 5528 for (i = 0; i < 78; i++) err.print_raw("="); 5529 err.cr(); 5530 5531 char buf[16]; 5532 // Prevent process from exiting upon "read error" without consuming all CPU 5533 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5534 5535 return buf[0] == 'y' || buf[0] == 'Y'; 5536 } 5537 5538 // Is a (classpath) directory empty? 5539 bool os::dir_is_empty(const char* path) { 5540 DIR *dir = NULL; 5541 struct dirent *ptr; 5542 5543 dir = opendir(path); 5544 if (dir == NULL) return true; 5545 5546 // Scan the directory 5547 bool result = true; 5548 while (result && (ptr = readdir(dir)) != NULL) { 5549 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5550 result = false; 5551 } 5552 } 5553 closedir(dir); 5554 return result; 5555 } 5556 5557 // This code originates from JDK's sysOpen and open64_w 5558 // from src/solaris/hpi/src/system_md.c 5559 5560 int os::open(const char *path, int oflag, int mode) { 5561 if (strlen(path) > MAX_PATH - 1) { 5562 errno = ENAMETOOLONG; 5563 return -1; 5564 } 5565 5566 // All file descriptors that are opened in the Java process and not 5567 // specifically destined for a subprocess should have the close-on-exec 5568 // flag set. If we don't set it, then careless 3rd party native code 5569 // might fork and exec without closing all appropriate file descriptors 5570 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5571 // turn might: 5572 // 5573 // - cause end-of-file to fail to be detected on some file 5574 // descriptors, resulting in mysterious hangs, or 5575 // 5576 // - might cause an fopen in the subprocess to fail on a system 5577 // suffering from bug 1085341. 5578 // 5579 // (Yes, the default setting of the close-on-exec flag is a Unix 5580 // design flaw) 5581 // 5582 // See: 5583 // 1085341: 32-bit stdio routines should support file descriptors >255 5584 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5585 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5586 // 5587 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5588 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5589 // because it saves a system call and removes a small window where the flag 5590 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5591 // and we fall back to using FD_CLOEXEC (see below). 5592 #ifdef O_CLOEXEC 5593 oflag |= O_CLOEXEC; 5594 #endif 5595 5596 int fd = ::open64(path, oflag, mode); 5597 if (fd == -1) return -1; 5598 5599 //If the open succeeded, the file might still be a directory 5600 { 5601 struct stat64 buf64; 5602 int ret = ::fstat64(fd, &buf64); 5603 int st_mode = buf64.st_mode; 5604 5605 if (ret != -1) { 5606 if ((st_mode & S_IFMT) == S_IFDIR) { 5607 errno = EISDIR; 5608 ::close(fd); 5609 return -1; 5610 } 5611 } else { 5612 ::close(fd); 5613 return -1; 5614 } 5615 } 5616 5617 #ifdef FD_CLOEXEC 5618 // Validate that the use of the O_CLOEXEC flag on open above worked. 5619 // With recent kernels, we will perform this check exactly once. 5620 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5621 if (!O_CLOEXEC_is_known_to_work) { 5622 int flags = ::fcntl(fd, F_GETFD); 5623 if (flags != -1) { 5624 if ((flags & FD_CLOEXEC) != 0) 5625 O_CLOEXEC_is_known_to_work = 1; 5626 else 5627 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5628 } 5629 } 5630 #endif 5631 5632 return fd; 5633 } 5634 5635 5636 // create binary file, rewriting existing file if required 5637 int os::create_binary_file(const char* path, bool rewrite_existing) { 5638 int oflags = O_WRONLY | O_CREAT; 5639 if (!rewrite_existing) { 5640 oflags |= O_EXCL; 5641 } 5642 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5643 } 5644 5645 // return current position of file pointer 5646 jlong os::current_file_offset(int fd) { 5647 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5648 } 5649 5650 // move file pointer to the specified offset 5651 jlong os::seek_to_file_offset(int fd, jlong offset) { 5652 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5653 } 5654 5655 // This code originates from JDK's sysAvailable 5656 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5657 5658 int os::available(int fd, jlong *bytes) { 5659 jlong cur, end; 5660 int mode; 5661 struct stat64 buf64; 5662 5663 if (::fstat64(fd, &buf64) >= 0) { 5664 mode = buf64.st_mode; 5665 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5666 int n; 5667 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5668 *bytes = n; 5669 return 1; 5670 } 5671 } 5672 } 5673 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5674 return 0; 5675 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5676 return 0; 5677 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5678 return 0; 5679 } 5680 *bytes = end - cur; 5681 return 1; 5682 } 5683 5684 // Map a block of memory. 5685 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5686 char *addr, size_t bytes, bool read_only, 5687 bool allow_exec) { 5688 int prot; 5689 int flags = MAP_PRIVATE; 5690 5691 if (read_only) { 5692 prot = PROT_READ; 5693 } else { 5694 prot = PROT_READ | PROT_WRITE; 5695 } 5696 5697 if (allow_exec) { 5698 prot |= PROT_EXEC; 5699 } 5700 5701 if (addr != NULL) { 5702 flags |= MAP_FIXED; 5703 } 5704 5705 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5706 fd, file_offset); 5707 if (mapped_address == MAP_FAILED) { 5708 return NULL; 5709 } 5710 return mapped_address; 5711 } 5712 5713 5714 // Remap a block of memory. 5715 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5716 char *addr, size_t bytes, bool read_only, 5717 bool allow_exec) { 5718 // same as map_memory() on this OS 5719 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5720 allow_exec); 5721 } 5722 5723 5724 // Unmap a block of memory. 5725 bool os::pd_unmap_memory(char* addr, size_t bytes) { 5726 return munmap(addr, bytes) == 0; 5727 } 5728 5729 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5730 5731 static jlong fast_cpu_time(Thread *thread) { 5732 clockid_t clockid; 5733 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(), 5734 &clockid); 5735 if (rc == 0) { 5736 return os::Linux::fast_thread_cpu_time(clockid); 5737 } else { 5738 // It's possible to encounter a terminated native thread that failed 5739 // to detach itself from the VM - which should result in ESRCH. 5740 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed"); 5741 return -1; 5742 } 5743 } 5744 5745 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5746 // are used by JVM M&M and JVMTI to get user+sys or user CPU time 5747 // of a thread. 5748 // 5749 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5750 // the fast estimate available on the platform. 5751 5752 jlong os::current_thread_cpu_time() { 5753 if (os::Linux::supports_fast_thread_cpu_time()) { 5754 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5755 } else { 5756 // return user + sys since the cost is the same 5757 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5758 } 5759 } 5760 5761 jlong os::thread_cpu_time(Thread* thread) { 5762 // consistent with what current_thread_cpu_time() returns 5763 if (os::Linux::supports_fast_thread_cpu_time()) { 5764 return fast_cpu_time(thread); 5765 } else { 5766 return slow_thread_cpu_time(thread, true /* user + sys */); 5767 } 5768 } 5769 5770 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5771 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5772 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5773 } else { 5774 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5775 } 5776 } 5777 5778 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5779 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5780 return fast_cpu_time(thread); 5781 } else { 5782 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5783 } 5784 } 5785 5786 // -1 on error. 5787 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5788 pid_t tid = thread->osthread()->thread_id(); 5789 char *s; 5790 char stat[2048]; 5791 int statlen; 5792 char proc_name[64]; 5793 int count; 5794 long sys_time, user_time; 5795 char cdummy; 5796 int idummy; 5797 long ldummy; 5798 FILE *fp; 5799 5800 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5801 fp = fopen(proc_name, "r"); 5802 if (fp == NULL) return -1; 5803 statlen = fread(stat, 1, 2047, fp); 5804 stat[statlen] = '\0'; 5805 fclose(fp); 5806 5807 // Skip pid and the command string. Note that we could be dealing with 5808 // weird command names, e.g. user could decide to rename java launcher 5809 // to "java 1.4.2 :)", then the stat file would look like 5810 // 1234 (java 1.4.2 :)) R ... ... 5811 // We don't really need to know the command string, just find the last 5812 // occurrence of ")" and then start parsing from there. See bug 4726580. 5813 s = strrchr(stat, ')'); 5814 if (s == NULL) return -1; 5815 5816 // Skip blank chars 5817 do { s++; } while (s && isspace(*s)); 5818 5819 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5820 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5821 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5822 &user_time, &sys_time); 5823 if (count != 13) return -1; 5824 if (user_sys_cpu_time) { 5825 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5826 } else { 5827 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5828 } 5829 } 5830 5831 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5832 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5833 info_ptr->may_skip_backward = false; // elapsed time not wall time 5834 info_ptr->may_skip_forward = false; // elapsed time not wall time 5835 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5836 } 5837 5838 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5839 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5840 info_ptr->may_skip_backward = false; // elapsed time not wall time 5841 info_ptr->may_skip_forward = false; // elapsed time not wall time 5842 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5843 } 5844 5845 bool os::is_thread_cpu_time_supported() { 5846 return true; 5847 } 5848 5849 // System loadavg support. Returns -1 if load average cannot be obtained. 5850 // Linux doesn't yet have a (official) notion of processor sets, 5851 // so just return the system wide load average. 5852 int os::loadavg(double loadavg[], int nelem) { 5853 return ::getloadavg(loadavg, nelem); 5854 } 5855 5856 void os::pause() { 5857 char filename[MAX_PATH]; 5858 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5859 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile); 5860 } else { 5861 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5862 } 5863 5864 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5865 if (fd != -1) { 5866 struct stat buf; 5867 ::close(fd); 5868 while (::stat(filename, &buf) == 0) { 5869 (void)::poll(NULL, 0, 100); 5870 } 5871 } else { 5872 jio_fprintf(stderr, 5873 "Could not open pause file '%s', continuing immediately.\n", filename); 5874 } 5875 } 5876 5877 extern char** environ; 5878 5879 // Run the specified command in a separate process. Return its exit value, 5880 // or -1 on failure (e.g. can't fork a new process). 5881 // Unlike system(), this function can be called from signal handler. It 5882 // doesn't block SIGINT et al. 5883 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) { 5884 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5885 5886 pid_t pid ; 5887 5888 if (use_vfork_if_available) { 5889 pid = vfork(); 5890 } else { 5891 pid = fork(); 5892 } 5893 5894 if (pid < 0) { 5895 // fork failed 5896 return -1; 5897 5898 } else if (pid == 0) { 5899 // child process 5900 5901 execve("/bin/sh", (char* const*)argv, environ); 5902 5903 // execve failed 5904 _exit(-1); 5905 5906 } else { 5907 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5908 // care about the actual exit code, for now. 5909 5910 int status; 5911 5912 // Wait for the child process to exit. This returns immediately if 5913 // the child has already exited. */ 5914 while (waitpid(pid, &status, 0) < 0) { 5915 switch (errno) { 5916 case ECHILD: return 0; 5917 case EINTR: break; 5918 default: return -1; 5919 } 5920 } 5921 5922 if (WIFEXITED(status)) { 5923 // The child exited normally; get its exit code. 5924 return WEXITSTATUS(status); 5925 } else if (WIFSIGNALED(status)) { 5926 // The child exited because of a signal 5927 // The best value to return is 0x80 + signal number, 5928 // because that is what all Unix shells do, and because 5929 // it allows callers to distinguish between process exit and 5930 // process death by signal. 5931 return 0x80 + WTERMSIG(status); 5932 } else { 5933 // Unknown exit code; pass it through 5934 return status; 5935 } 5936 } 5937 } 5938 5939 // Get the default path to the core file 5940 // Returns the length of the string 5941 int os::get_core_path(char* buffer, size_t bufferSize) { 5942 /* 5943 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 5944 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 5945 */ 5946 const int core_pattern_len = 129; 5947 char core_pattern[core_pattern_len] = {0}; 5948 5949 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 5950 if (core_pattern_file == -1) { 5951 return -1; 5952 } 5953 5954 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 5955 ::close(core_pattern_file); 5956 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') { 5957 return -1; 5958 } 5959 if (core_pattern[ret-1] == '\n') { 5960 core_pattern[ret-1] = '\0'; 5961 } else { 5962 core_pattern[ret] = '\0'; 5963 } 5964 5965 // Replace the %p in the core pattern with the process id. NOTE: we do this 5966 // only if the pattern doesn't start with "|", and we support only one %p in 5967 // the pattern. 5968 char *pid_pos = strstr(core_pattern, "%p"); 5969 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p" 5970 int written; 5971 5972 if (core_pattern[0] == '/') { 5973 if (pid_pos != NULL) { 5974 *pid_pos = '\0'; 5975 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern, 5976 current_process_id(), tail); 5977 } else { 5978 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern); 5979 } 5980 } else { 5981 char cwd[PATH_MAX]; 5982 5983 const char* p = get_current_directory(cwd, PATH_MAX); 5984 if (p == NULL) { 5985 return -1; 5986 } 5987 5988 if (core_pattern[0] == '|') { 5989 written = jio_snprintf(buffer, bufferSize, 5990 "\"%s\" (or dumping to %s/core.%d)", 5991 &core_pattern[1], p, current_process_id()); 5992 } else if (pid_pos != NULL) { 5993 *pid_pos = '\0'; 5994 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern, 5995 current_process_id(), tail); 5996 } else { 5997 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 5998 } 5999 } 6000 6001 if (written < 0) { 6002 return -1; 6003 } 6004 6005 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 6006 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 6007 6008 if (core_uses_pid_file != -1) { 6009 char core_uses_pid = 0; 6010 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 6011 ::close(core_uses_pid_file); 6012 6013 if (core_uses_pid == '1') { 6014 jio_snprintf(buffer + written, bufferSize - written, 6015 ".%d", current_process_id()); 6016 } 6017 } 6018 } 6019 6020 return strlen(buffer); 6021 } 6022 6023 bool os::start_debugging(char *buf, int buflen) { 6024 int len = (int)strlen(buf); 6025 char *p = &buf[len]; 6026 6027 jio_snprintf(p, buflen-len, 6028 "\n\n" 6029 "Do you want to debug the problem?\n\n" 6030 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n" 6031 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n" 6032 "Otherwise, press RETURN to abort...", 6033 os::current_process_id(), os::current_process_id(), 6034 os::current_thread_id(), os::current_thread_id()); 6035 6036 bool yes = os::message_box("Unexpected Error", buf); 6037 6038 if (yes) { 6039 // yes, user asked VM to launch debugger 6040 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d", 6041 os::current_process_id(), os::current_process_id()); 6042 6043 os::fork_and_exec(buf); 6044 yes = false; 6045 } 6046 return yes; 6047 } 6048 6049 6050 // Java/Compiler thread: 6051 // 6052 // Low memory addresses 6053 // P0 +------------------------+ 6054 // | |\ Java thread created by VM does not have glibc 6055 // | glibc guard page | - guard page, attached Java thread usually has 6056 // | |/ 1 glibc guard page. 6057 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 6058 // | |\ 6059 // | HotSpot Guard Pages | - red, yellow and reserved pages 6060 // | |/ 6061 // +------------------------+ JavaThread::stack_reserved_zone_base() 6062 // | |\ 6063 // | Normal Stack | - 6064 // | |/ 6065 // P2 +------------------------+ Thread::stack_base() 6066 // 6067 // Non-Java thread: 6068 // 6069 // Low memory addresses 6070 // P0 +------------------------+ 6071 // | |\ 6072 // | glibc guard page | - usually 1 page 6073 // | |/ 6074 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 6075 // | |\ 6076 // | Normal Stack | - 6077 // | |/ 6078 // P2 +------------------------+ Thread::stack_base() 6079 // 6080 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size 6081 // returned from pthread_attr_getstack(). 6082 // ** Due to NPTL implementation error, linux takes the glibc guard page out 6083 // of the stack size given in pthread_attr. We work around this for 6084 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.) 6085 // 6086 #ifndef ZERO 6087 static void current_stack_region(address * bottom, size_t * size) { 6088 if (os::is_primordial_thread()) { 6089 // primordial thread needs special handling because pthread_getattr_np() 6090 // may return bogus value. 6091 *bottom = os::Linux::initial_thread_stack_bottom(); 6092 *size = os::Linux::initial_thread_stack_size(); 6093 } else { 6094 pthread_attr_t attr; 6095 6096 int rslt = pthread_getattr_np(pthread_self(), &attr); 6097 6098 // JVM needs to know exact stack location, abort if it fails 6099 if (rslt != 0) { 6100 if (rslt == ENOMEM) { 6101 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np"); 6102 } else { 6103 fatal("pthread_getattr_np failed with error = %d", rslt); 6104 } 6105 } 6106 6107 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) { 6108 fatal("Cannot locate current stack attributes!"); 6109 } 6110 6111 // Work around NPTL stack guard error. 6112 size_t guard_size = 0; 6113 rslt = pthread_attr_getguardsize(&attr, &guard_size); 6114 if (rslt != 0) { 6115 fatal("pthread_attr_getguardsize failed with error = %d", rslt); 6116 } 6117 *bottom += guard_size; 6118 *size -= guard_size; 6119 6120 pthread_attr_destroy(&attr); 6121 6122 } 6123 assert(os::current_stack_pointer() >= *bottom && 6124 os::current_stack_pointer() < *bottom + *size, "just checking"); 6125 } 6126 6127 address os::current_stack_base() { 6128 address bottom; 6129 size_t size; 6130 current_stack_region(&bottom, &size); 6131 return (bottom + size); 6132 } 6133 6134 size_t os::current_stack_size() { 6135 // This stack size includes the usable stack and HotSpot guard pages 6136 // (for the threads that have Hotspot guard pages). 6137 address bottom; 6138 size_t size; 6139 current_stack_region(&bottom, &size); 6140 return size; 6141 } 6142 #endif 6143 6144 static inline struct timespec get_mtime(const char* filename) { 6145 struct stat st; 6146 int ret = os::stat(filename, &st); 6147 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno)); 6148 return st.st_mtim; 6149 } 6150 6151 int os::compare_file_modified_times(const char* file1, const char* file2) { 6152 struct timespec filetime1 = get_mtime(file1); 6153 struct timespec filetime2 = get_mtime(file2); 6154 int diff = filetime1.tv_sec - filetime2.tv_sec; 6155 if (diff == 0) { 6156 return filetime1.tv_nsec - filetime2.tv_nsec; 6157 } 6158 return diff; 6159 } 6160 6161 bool os::supports_map_sync() { 6162 return true; 6163 } 6164 6165 /////////////// Unit tests /////////////// 6166 6167 #ifndef PRODUCT 6168 6169 class TestReserveMemorySpecial : AllStatic { 6170 public: 6171 static void small_page_write(void* addr, size_t size) { 6172 size_t page_size = os::vm_page_size(); 6173 6174 char* end = (char*)addr + size; 6175 for (char* p = (char*)addr; p < end; p += page_size) { 6176 *p = 1; 6177 } 6178 } 6179 6180 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6181 if (!UseHugeTLBFS) { 6182 return; 6183 } 6184 6185 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6186 6187 if (addr != NULL) { 6188 small_page_write(addr, size); 6189 6190 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6191 } 6192 } 6193 6194 static void test_reserve_memory_special_huge_tlbfs_only() { 6195 if (!UseHugeTLBFS) { 6196 return; 6197 } 6198 6199 size_t lp = os::large_page_size(); 6200 6201 for (size_t size = lp; size <= lp * 10; size += lp) { 6202 test_reserve_memory_special_huge_tlbfs_only(size); 6203 } 6204 } 6205 6206 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6207 size_t lp = os::large_page_size(); 6208 size_t ag = os::vm_allocation_granularity(); 6209 6210 // sizes to test 6211 const size_t sizes[] = { 6212 lp, lp + ag, lp + lp / 2, lp * 2, 6213 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2, 6214 lp * 10, lp * 10 + lp / 2 6215 }; 6216 const int num_sizes = sizeof(sizes) / sizeof(size_t); 6217 6218 // For each size/alignment combination, we test three scenarios: 6219 // 1) with req_addr == NULL 6220 // 2) with a non-null req_addr at which we expect to successfully allocate 6221 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we 6222 // expect the allocation to either fail or to ignore req_addr 6223 6224 // Pre-allocate two areas; they shall be as large as the largest allocation 6225 // and aligned to the largest alignment we will be testing. 6226 const size_t mapping_size = sizes[num_sizes - 1] * 2; 6227 char* const mapping1 = (char*) ::mmap(NULL, mapping_size, 6228 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6229 -1, 0); 6230 assert(mapping1 != MAP_FAILED, "should work"); 6231 6232 char* const mapping2 = (char*) ::mmap(NULL, mapping_size, 6233 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6234 -1, 0); 6235 assert(mapping2 != MAP_FAILED, "should work"); 6236 6237 // Unmap the first mapping, but leave the second mapping intact: the first 6238 // mapping will serve as a value for a "good" req_addr (case 2). The second 6239 // mapping, still intact, as "bad" req_addr (case 3). 6240 ::munmap(mapping1, mapping_size); 6241 6242 // Case 1 6243 for (int i = 0; i < num_sizes; i++) { 6244 const size_t size = sizes[i]; 6245 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6246 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6247 if (p != NULL) { 6248 assert(is_aligned(p, alignment), "must be"); 6249 small_page_write(p, size); 6250 os::Linux::release_memory_special_huge_tlbfs(p, size); 6251 } 6252 } 6253 } 6254 6255 // Case 2 6256 for (int i = 0; i < num_sizes; i++) { 6257 const size_t size = sizes[i]; 6258 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6259 char* const req_addr = align_up(mapping1, alignment); 6260 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6261 if (p != NULL) { 6262 assert(p == req_addr, "must be"); 6263 small_page_write(p, size); 6264 os::Linux::release_memory_special_huge_tlbfs(p, size); 6265 } 6266 } 6267 } 6268 6269 // Case 3 6270 for (int i = 0; i < num_sizes; i++) { 6271 const size_t size = sizes[i]; 6272 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6273 char* const req_addr = align_up(mapping2, alignment); 6274 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6275 // as the area around req_addr contains already existing mappings, the API should always 6276 // return NULL (as per contract, it cannot return another address) 6277 assert(p == NULL, "must be"); 6278 } 6279 } 6280 6281 ::munmap(mapping2, mapping_size); 6282 6283 } 6284 6285 static void test_reserve_memory_special_huge_tlbfs() { 6286 if (!UseHugeTLBFS) { 6287 return; 6288 } 6289 6290 test_reserve_memory_special_huge_tlbfs_only(); 6291 test_reserve_memory_special_huge_tlbfs_mixed(); 6292 } 6293 6294 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6295 if (!UseSHM) { 6296 return; 6297 } 6298 6299 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6300 6301 if (addr != NULL) { 6302 assert(is_aligned(addr, alignment), "Check"); 6303 assert(is_aligned(addr, os::large_page_size()), "Check"); 6304 6305 small_page_write(addr, size); 6306 6307 os::Linux::release_memory_special_shm(addr, size); 6308 } 6309 } 6310 6311 static void test_reserve_memory_special_shm() { 6312 size_t lp = os::large_page_size(); 6313 size_t ag = os::vm_allocation_granularity(); 6314 6315 for (size_t size = ag; size < lp * 3; size += ag) { 6316 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6317 test_reserve_memory_special_shm(size, alignment); 6318 } 6319 } 6320 } 6321 6322 static void test() { 6323 test_reserve_memory_special_huge_tlbfs(); 6324 test_reserve_memory_special_shm(); 6325 } 6326 }; 6327 6328 void TestReserveMemorySpecial_test() { 6329 TestReserveMemorySpecial::test(); 6330 } 6331 6332 #endif