1 /* 2 * Copyright (c) 1999, 2019, 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/orderAccess.hpp" 57 #include "runtime/osThread.hpp" 58 #include "runtime/perfMemory.hpp" 59 #include "runtime/sharedRuntime.hpp" 60 #include "runtime/statSampler.hpp" 61 #include "runtime/stubRoutines.hpp" 62 #include "runtime/thread.inline.hpp" 63 #include "runtime/threadCritical.hpp" 64 #include "runtime/threadSMR.hpp" 65 #include "runtime/timer.hpp" 66 #include "runtime/vm_version.hpp" 67 #include "semaphore_posix.hpp" 68 #include "services/attachListener.hpp" 69 #include "services/memTracker.hpp" 70 #include "services/runtimeService.hpp" 71 #include "utilities/align.hpp" 72 #include "utilities/decoder.hpp" 73 #include "utilities/defaultStream.hpp" 74 #include "utilities/events.hpp" 75 #include "utilities/elfFile.hpp" 76 #include "utilities/growableArray.hpp" 77 #include "utilities/macros.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 (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) { 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[PATH_MAX + 1]; 2086 2087 // Parse fields from line 2088 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s", 2089 &base, &top, permissions, &offset, device, &inode, name); 2090 2091 // Filter by device id '00:00' so that we only get file system mapped files. 2092 if (strcmp(device, "00:00") != 0) { 2093 2094 // Call callback with the fields of interest 2095 if(callback(name, (address)base, (address)top, param)) { 2096 // Oops abort, callback aborted 2097 fclose(procmapsFile); 2098 return 1; 2099 } 2100 } 2101 } 2102 fclose(procmapsFile); 2103 } 2104 return 0; 2105 } 2106 2107 void os::print_os_info_brief(outputStream* st) { 2108 os::Linux::print_distro_info(st); 2109 2110 os::Posix::print_uname_info(st); 2111 2112 os::Linux::print_libversion_info(st); 2113 2114 } 2115 2116 void os::print_os_info(outputStream* st) { 2117 st->print("OS:"); 2118 2119 os::Linux::print_distro_info(st); 2120 2121 os::Posix::print_uname_info(st); 2122 2123 // Print warning if unsafe chroot environment detected 2124 if (unsafe_chroot_detected) { 2125 st->print("WARNING!! "); 2126 st->print_cr("%s", unstable_chroot_error); 2127 } 2128 2129 os::Linux::print_libversion_info(st); 2130 2131 os::Posix::print_rlimit_info(st); 2132 2133 os::Posix::print_load_average(st); 2134 2135 os::Linux::print_full_memory_info(st); 2136 2137 os::Linux::print_proc_sys_info(st); 2138 2139 os::Linux::print_ld_preload_file(st); 2140 2141 os::Linux::print_container_info(st); 2142 2143 VM_Version::print_platform_virtualization_info(st); 2144 2145 os::Linux::print_steal_info(st); 2146 } 2147 2148 // Try to identify popular distros. 2149 // Most Linux distributions have a /etc/XXX-release file, which contains 2150 // the OS version string. Newer Linux distributions have a /etc/lsb-release 2151 // file that also contains the OS version string. Some have more than one 2152 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2153 // /etc/redhat-release.), so the order is important. 2154 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2155 // their own specific XXX-release file as well as a redhat-release file. 2156 // Because of this the XXX-release file needs to be searched for before the 2157 // redhat-release file. 2158 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the 2159 // search for redhat-release / SuSE-release needs to be before lsb-release. 2160 // Since the lsb-release file is the new standard it needs to be searched 2161 // before the older style release files. 2162 // Searching system-release (Red Hat) and os-release (other Linuxes) are a 2163 // next to last resort. The os-release file is a new standard that contains 2164 // distribution information and the system-release file seems to be an old 2165 // standard that has been replaced by the lsb-release and os-release files. 2166 // Searching for the debian_version file is the last resort. It contains 2167 // an informative string like "6.0.6" or "wheezy/sid". Because of this 2168 // "Debian " is printed before the contents of the debian_version file. 2169 2170 const char* distro_files[] = { 2171 "/etc/oracle-release", 2172 "/etc/mandriva-release", 2173 "/etc/mandrake-release", 2174 "/etc/sun-release", 2175 "/etc/redhat-release", 2176 "/etc/SuSE-release", 2177 "/etc/lsb-release", 2178 "/etc/turbolinux-release", 2179 "/etc/gentoo-release", 2180 "/etc/ltib-release", 2181 "/etc/angstrom-version", 2182 "/etc/system-release", 2183 "/etc/os-release", 2184 NULL }; 2185 2186 void os::Linux::print_distro_info(outputStream* st) { 2187 for (int i = 0;; i++) { 2188 const char* file = distro_files[i]; 2189 if (file == NULL) { 2190 break; // done 2191 } 2192 // If file prints, we found it. 2193 if (_print_ascii_file(file, st)) { 2194 return; 2195 } 2196 } 2197 2198 if (file_exists("/etc/debian_version")) { 2199 st->print("Debian "); 2200 _print_ascii_file("/etc/debian_version", st); 2201 } else { 2202 st->print("Linux"); 2203 } 2204 st->cr(); 2205 } 2206 2207 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) { 2208 char buf[256]; 2209 while (fgets(buf, sizeof(buf), fp)) { 2210 // Edit out extra stuff in expected format 2211 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) { 2212 char* ptr = strstr(buf, "\""); // the name is in quotes 2213 if (ptr != NULL) { 2214 ptr++; // go beyond first quote 2215 char* nl = strchr(ptr, '\"'); 2216 if (nl != NULL) *nl = '\0'; 2217 strncpy(distro, ptr, length); 2218 } else { 2219 ptr = strstr(buf, "="); 2220 ptr++; // go beyond equals then 2221 char* nl = strchr(ptr, '\n'); 2222 if (nl != NULL) *nl = '\0'; 2223 strncpy(distro, ptr, length); 2224 } 2225 return; 2226 } else if (get_first_line) { 2227 char* nl = strchr(buf, '\n'); 2228 if (nl != NULL) *nl = '\0'; 2229 strncpy(distro, buf, length); 2230 return; 2231 } 2232 } 2233 // print last line and close 2234 char* nl = strchr(buf, '\n'); 2235 if (nl != NULL) *nl = '\0'; 2236 strncpy(distro, buf, length); 2237 } 2238 2239 static void parse_os_info(char* distro, size_t length, const char* file) { 2240 FILE* fp = fopen(file, "r"); 2241 if (fp != NULL) { 2242 // if suse format, print out first line 2243 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0); 2244 parse_os_info_helper(fp, distro, length, get_first_line); 2245 fclose(fp); 2246 } 2247 } 2248 2249 void os::get_summary_os_info(char* buf, size_t buflen) { 2250 for (int i = 0;; i++) { 2251 const char* file = distro_files[i]; 2252 if (file == NULL) { 2253 break; // ran out of distro_files 2254 } 2255 if (file_exists(file)) { 2256 parse_os_info(buf, buflen, file); 2257 return; 2258 } 2259 } 2260 // special case for debian 2261 if (file_exists("/etc/debian_version")) { 2262 strncpy(buf, "Debian ", buflen); 2263 if (buflen > 7) { 2264 parse_os_info(&buf[7], buflen-7, "/etc/debian_version"); 2265 } 2266 } else { 2267 strncpy(buf, "Linux", buflen); 2268 } 2269 } 2270 2271 void os::Linux::print_libversion_info(outputStream* st) { 2272 // libc, pthread 2273 st->print("libc:"); 2274 st->print("%s ", os::Linux::glibc_version()); 2275 st->print("%s ", os::Linux::libpthread_version()); 2276 st->cr(); 2277 } 2278 2279 void os::Linux::print_proc_sys_info(outputStream* st) { 2280 st->cr(); 2281 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):"); 2282 _print_ascii_file("/proc/sys/kernel/threads-max", st); 2283 st->cr(); 2284 st->cr(); 2285 2286 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):"); 2287 _print_ascii_file("/proc/sys/vm/max_map_count", st); 2288 st->cr(); 2289 st->cr(); 2290 2291 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):"); 2292 _print_ascii_file("/proc/sys/kernel/pid_max", st); 2293 st->cr(); 2294 st->cr(); 2295 } 2296 2297 void os::Linux::print_full_memory_info(outputStream* st) { 2298 st->print("\n/proc/meminfo:\n"); 2299 _print_ascii_file("/proc/meminfo", st); 2300 st->cr(); 2301 } 2302 2303 void os::Linux::print_ld_preload_file(outputStream* st) { 2304 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:"); 2305 st->cr(); 2306 } 2307 2308 void os::Linux::print_container_info(outputStream* st) { 2309 if (!OSContainer::is_containerized()) { 2310 return; 2311 } 2312 2313 st->print("container (cgroup) information:\n"); 2314 2315 const char *p_ct = OSContainer::container_type(); 2316 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported"); 2317 2318 char *p = OSContainer::cpu_cpuset_cpus(); 2319 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported"); 2320 free(p); 2321 2322 p = OSContainer::cpu_cpuset_memory_nodes(); 2323 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported"); 2324 free(p); 2325 2326 int i = OSContainer::active_processor_count(); 2327 st->print("active_processor_count: "); 2328 if (i > 0) { 2329 st->print("%d\n", i); 2330 } else { 2331 st->print("not supported\n"); 2332 } 2333 2334 i = OSContainer::cpu_quota(); 2335 st->print("cpu_quota: "); 2336 if (i > 0) { 2337 st->print("%d\n", i); 2338 } else { 2339 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota"); 2340 } 2341 2342 i = OSContainer::cpu_period(); 2343 st->print("cpu_period: "); 2344 if (i > 0) { 2345 st->print("%d\n", i); 2346 } else { 2347 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period"); 2348 } 2349 2350 i = OSContainer::cpu_shares(); 2351 st->print("cpu_shares: "); 2352 if (i > 0) { 2353 st->print("%d\n", i); 2354 } else { 2355 st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares"); 2356 } 2357 2358 jlong j = OSContainer::memory_limit_in_bytes(); 2359 st->print("memory_limit_in_bytes: "); 2360 if (j > 0) { 2361 st->print(JLONG_FORMAT "\n", j); 2362 } else { 2363 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited"); 2364 } 2365 2366 j = OSContainer::memory_and_swap_limit_in_bytes(); 2367 st->print("memory_and_swap_limit_in_bytes: "); 2368 if (j > 0) { 2369 st->print(JLONG_FORMAT "\n", j); 2370 } else { 2371 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited"); 2372 } 2373 2374 j = OSContainer::memory_soft_limit_in_bytes(); 2375 st->print("memory_soft_limit_in_bytes: "); 2376 if (j > 0) { 2377 st->print(JLONG_FORMAT "\n", j); 2378 } else { 2379 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited"); 2380 } 2381 2382 j = OSContainer::OSContainer::memory_usage_in_bytes(); 2383 st->print("memory_usage_in_bytes: "); 2384 if (j > 0) { 2385 st->print(JLONG_FORMAT "\n", j); 2386 } else { 2387 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited"); 2388 } 2389 2390 j = OSContainer::OSContainer::memory_max_usage_in_bytes(); 2391 st->print("memory_max_usage_in_bytes: "); 2392 if (j > 0) { 2393 st->print(JLONG_FORMAT "\n", j); 2394 } else { 2395 st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited"); 2396 } 2397 st->cr(); 2398 } 2399 2400 void os::Linux::print_steal_info(outputStream* st) { 2401 if (has_initial_tick_info) { 2402 CPUPerfTicks pticks; 2403 bool res = os::Linux::get_tick_information(&pticks, -1); 2404 2405 if (res && pticks.has_steal_ticks) { 2406 uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks; 2407 uint64_t total_ticks_difference = pticks.total - initial_total_ticks; 2408 double steal_ticks_perc = 0.0; 2409 if (total_ticks_difference != 0) { 2410 steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference; 2411 } 2412 st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference); 2413 st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc); 2414 } 2415 } 2416 } 2417 2418 void os::print_memory_info(outputStream* st) { 2419 2420 st->print("Memory:"); 2421 st->print(" %dk page", os::vm_page_size()>>10); 2422 2423 // values in struct sysinfo are "unsigned long" 2424 struct sysinfo si; 2425 sysinfo(&si); 2426 2427 st->print(", physical " UINT64_FORMAT "k", 2428 os::physical_memory() >> 10); 2429 st->print("(" UINT64_FORMAT "k free)", 2430 os::available_memory() >> 10); 2431 st->print(", swap " UINT64_FORMAT "k", 2432 ((jlong)si.totalswap * si.mem_unit) >> 10); 2433 st->print("(" UINT64_FORMAT "k free)", 2434 ((jlong)si.freeswap * si.mem_unit) >> 10); 2435 st->cr(); 2436 } 2437 2438 // Print the first "model name" line and the first "flags" line 2439 // that we find and nothing more. We assume "model name" comes 2440 // before "flags" so if we find a second "model name", then the 2441 // "flags" field is considered missing. 2442 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) { 2443 #if defined(IA32) || defined(AMD64) 2444 // Other platforms have less repetitive cpuinfo files 2445 FILE *fp = fopen("/proc/cpuinfo", "r"); 2446 if (fp) { 2447 while (!feof(fp)) { 2448 if (fgets(buf, buflen, fp)) { 2449 // Assume model name comes before flags 2450 bool model_name_printed = false; 2451 if (strstr(buf, "model name") != NULL) { 2452 if (!model_name_printed) { 2453 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n"); 2454 st->print_raw(buf); 2455 model_name_printed = true; 2456 } else { 2457 // model name printed but not flags? Odd, just return 2458 fclose(fp); 2459 return true; 2460 } 2461 } 2462 // print the flags line too 2463 if (strstr(buf, "flags") != NULL) { 2464 st->print_raw(buf); 2465 fclose(fp); 2466 return true; 2467 } 2468 } 2469 } 2470 fclose(fp); 2471 } 2472 #endif // x86 platforms 2473 return false; 2474 } 2475 2476 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) { 2477 // Only print the model name if the platform provides this as a summary 2478 if (!print_model_name_and_flags(st, buf, buflen)) { 2479 st->print("\n/proc/cpuinfo:\n"); 2480 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2481 st->print_cr(" <Not Available>"); 2482 } 2483 } 2484 } 2485 2486 #if defined(AMD64) || defined(IA32) || defined(X32) 2487 const char* search_string = "model name"; 2488 #elif defined(M68K) 2489 const char* search_string = "CPU"; 2490 #elif defined(PPC64) 2491 const char* search_string = "cpu"; 2492 #elif defined(S390) 2493 const char* search_string = "machine ="; 2494 #elif defined(SPARC) 2495 const char* search_string = "cpu"; 2496 #else 2497 const char* search_string = "Processor"; 2498 #endif 2499 2500 // Parses the cpuinfo file for string representing the model name. 2501 void os::get_summary_cpu_info(char* cpuinfo, size_t length) { 2502 FILE* fp = fopen("/proc/cpuinfo", "r"); 2503 if (fp != NULL) { 2504 while (!feof(fp)) { 2505 char buf[256]; 2506 if (fgets(buf, sizeof(buf), fp)) { 2507 char* start = strstr(buf, search_string); 2508 if (start != NULL) { 2509 char *ptr = start + strlen(search_string); 2510 char *end = buf + strlen(buf); 2511 while (ptr != end) { 2512 // skip whitespace and colon for the rest of the name. 2513 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') { 2514 break; 2515 } 2516 ptr++; 2517 } 2518 if (ptr != end) { 2519 // reasonable string, get rid of newline and keep the rest 2520 char* nl = strchr(buf, '\n'); 2521 if (nl != NULL) *nl = '\0'; 2522 strncpy(cpuinfo, ptr, length); 2523 fclose(fp); 2524 return; 2525 } 2526 } 2527 } 2528 } 2529 fclose(fp); 2530 } 2531 // cpuinfo not found or parsing failed, just print generic string. The entire 2532 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86) 2533 #if defined(AARCH64) 2534 strncpy(cpuinfo, "AArch64", length); 2535 #elif defined(AMD64) 2536 strncpy(cpuinfo, "x86_64", length); 2537 #elif defined(ARM) // Order wrt. AARCH64 is relevant! 2538 strncpy(cpuinfo, "ARM", length); 2539 #elif defined(IA32) 2540 strncpy(cpuinfo, "x86_32", length); 2541 #elif defined(IA64) 2542 strncpy(cpuinfo, "IA64", length); 2543 #elif defined(PPC) 2544 strncpy(cpuinfo, "PPC64", length); 2545 #elif defined(S390) 2546 strncpy(cpuinfo, "S390", length); 2547 #elif defined(SPARC) 2548 strncpy(cpuinfo, "sparcv9", length); 2549 #elif defined(ZERO_LIBARCH) 2550 strncpy(cpuinfo, ZERO_LIBARCH, length); 2551 #else 2552 strncpy(cpuinfo, "unknown", length); 2553 #endif 2554 } 2555 2556 static void print_signal_handler(outputStream* st, int sig, 2557 char* buf, size_t buflen); 2558 2559 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2560 st->print_cr("Signal Handlers:"); 2561 print_signal_handler(st, SIGSEGV, buf, buflen); 2562 print_signal_handler(st, SIGBUS , buf, buflen); 2563 print_signal_handler(st, SIGFPE , buf, buflen); 2564 print_signal_handler(st, SIGPIPE, buf, buflen); 2565 print_signal_handler(st, SIGXFSZ, buf, buflen); 2566 print_signal_handler(st, SIGILL , buf, buflen); 2567 print_signal_handler(st, SR_signum, buf, buflen); 2568 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2569 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2570 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2571 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2572 #if defined(PPC64) 2573 print_signal_handler(st, SIGTRAP, buf, buflen); 2574 #endif 2575 } 2576 2577 static char saved_jvm_path[MAXPATHLEN] = {0}; 2578 2579 // Find the full path to the current module, libjvm.so 2580 void os::jvm_path(char *buf, jint buflen) { 2581 // Error checking. 2582 if (buflen < MAXPATHLEN) { 2583 assert(false, "must use a large-enough buffer"); 2584 buf[0] = '\0'; 2585 return; 2586 } 2587 // Lazy resolve the path to current module. 2588 if (saved_jvm_path[0] != 0) { 2589 strcpy(buf, saved_jvm_path); 2590 return; 2591 } 2592 2593 char dli_fname[MAXPATHLEN]; 2594 bool ret = dll_address_to_library_name( 2595 CAST_FROM_FN_PTR(address, os::jvm_path), 2596 dli_fname, sizeof(dli_fname), NULL); 2597 assert(ret, "cannot locate libjvm"); 2598 char *rp = NULL; 2599 if (ret && dli_fname[0] != '\0') { 2600 rp = os::Posix::realpath(dli_fname, buf, buflen); 2601 } 2602 if (rp == NULL) { 2603 return; 2604 } 2605 2606 if (Arguments::sun_java_launcher_is_altjvm()) { 2607 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2608 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so". 2609 // If "/jre/lib/" appears at the right place in the string, then 2610 // assume we are installed in a JDK and we're done. Otherwise, check 2611 // for a JAVA_HOME environment variable and fix up the path so it 2612 // looks like libjvm.so is installed there (append a fake suffix 2613 // hotspot/libjvm.so). 2614 const char *p = buf + strlen(buf) - 1; 2615 for (int count = 0; p > buf && count < 5; ++count) { 2616 for (--p; p > buf && *p != '/'; --p) 2617 /* empty */ ; 2618 } 2619 2620 if (strncmp(p, "/jre/lib/", 9) != 0) { 2621 // Look for JAVA_HOME in the environment. 2622 char* java_home_var = ::getenv("JAVA_HOME"); 2623 if (java_home_var != NULL && java_home_var[0] != 0) { 2624 char* jrelib_p; 2625 int len; 2626 2627 // Check the current module name "libjvm.so". 2628 p = strrchr(buf, '/'); 2629 if (p == NULL) { 2630 return; 2631 } 2632 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2633 2634 rp = os::Posix::realpath(java_home_var, buf, buflen); 2635 if (rp == NULL) { 2636 return; 2637 } 2638 2639 // determine if this is a legacy image or modules image 2640 // modules image doesn't have "jre" subdirectory 2641 len = strlen(buf); 2642 assert(len < buflen, "Ran out of buffer room"); 2643 jrelib_p = buf + len; 2644 snprintf(jrelib_p, buflen-len, "/jre/lib"); 2645 if (0 != access(buf, F_OK)) { 2646 snprintf(jrelib_p, buflen-len, "/lib"); 2647 } 2648 2649 if (0 == access(buf, F_OK)) { 2650 // Use current module name "libjvm.so" 2651 len = strlen(buf); 2652 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2653 } else { 2654 // Go back to path of .so 2655 rp = os::Posix::realpath(dli_fname, buf, buflen); 2656 if (rp == NULL) { 2657 return; 2658 } 2659 } 2660 } 2661 } 2662 } 2663 2664 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2665 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2666 } 2667 2668 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2669 // no prefix required, not even "_" 2670 } 2671 2672 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2673 // no suffix required 2674 } 2675 2676 //////////////////////////////////////////////////////////////////////////////// 2677 // sun.misc.Signal support 2678 2679 static void UserHandler(int sig, void *siginfo, void *context) { 2680 // Ctrl-C is pressed during error reporting, likely because the error 2681 // handler fails to abort. Let VM die immediately. 2682 if (sig == SIGINT && VMError::is_error_reported()) { 2683 os::die(); 2684 } 2685 2686 os::signal_notify(sig); 2687 } 2688 2689 void* os::user_handler() { 2690 return CAST_FROM_FN_PTR(void*, UserHandler); 2691 } 2692 2693 extern "C" { 2694 typedef void (*sa_handler_t)(int); 2695 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2696 } 2697 2698 void* os::signal(int signal_number, void* handler) { 2699 struct sigaction sigAct, oldSigAct; 2700 2701 sigfillset(&(sigAct.sa_mask)); 2702 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2703 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2704 2705 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2706 // -1 means registration failed 2707 return (void *)-1; 2708 } 2709 2710 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2711 } 2712 2713 void os::signal_raise(int signal_number) { 2714 ::raise(signal_number); 2715 } 2716 2717 // The following code is moved from os.cpp for making this 2718 // code platform specific, which it is by its very nature. 2719 2720 // Will be modified when max signal is changed to be dynamic 2721 int os::sigexitnum_pd() { 2722 return NSIG; 2723 } 2724 2725 // a counter for each possible signal value 2726 static volatile jint pending_signals[NSIG+1] = { 0 }; 2727 2728 // Linux(POSIX) specific hand shaking semaphore. 2729 static Semaphore* sig_sem = NULL; 2730 static PosixSemaphore sr_semaphore; 2731 2732 static void jdk_misc_signal_init() { 2733 // Initialize signal structures 2734 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2735 2736 // Initialize signal semaphore 2737 sig_sem = new Semaphore(); 2738 } 2739 2740 void os::signal_notify(int sig) { 2741 if (sig_sem != NULL) { 2742 Atomic::inc(&pending_signals[sig]); 2743 sig_sem->signal(); 2744 } else { 2745 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init 2746 // initialization isn't called. 2747 assert(ReduceSignalUsage, "signal semaphore should be created"); 2748 } 2749 } 2750 2751 static int check_pending_signals() { 2752 for (;;) { 2753 for (int i = 0; i < NSIG + 1; i++) { 2754 jint n = pending_signals[i]; 2755 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2756 return i; 2757 } 2758 } 2759 JavaThread *thread = JavaThread::current(); 2760 ThreadBlockInVM tbivm(thread); 2761 2762 bool threadIsSuspended; 2763 do { 2764 thread->set_suspend_equivalent(); 2765 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2766 sig_sem->wait(); 2767 2768 // were we externally suspended while we were waiting? 2769 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2770 if (threadIsSuspended) { 2771 // The semaphore has been incremented, but while we were waiting 2772 // another thread suspended us. We don't want to continue running 2773 // while suspended because that would surprise the thread that 2774 // suspended us. 2775 sig_sem->signal(); 2776 2777 thread->java_suspend_self(); 2778 } 2779 } while (threadIsSuspended); 2780 } 2781 } 2782 2783 int os::signal_wait() { 2784 return check_pending_signals(); 2785 } 2786 2787 //////////////////////////////////////////////////////////////////////////////// 2788 // Virtual Memory 2789 2790 int os::vm_page_size() { 2791 // Seems redundant as all get out 2792 assert(os::Linux::page_size() != -1, "must call os::init"); 2793 return os::Linux::page_size(); 2794 } 2795 2796 // Solaris allocates memory by pages. 2797 int os::vm_allocation_granularity() { 2798 assert(os::Linux::page_size() != -1, "must call os::init"); 2799 return os::Linux::page_size(); 2800 } 2801 2802 // Rationale behind this function: 2803 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2804 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2805 // samples for JITted code. Here we create private executable mapping over the code cache 2806 // and then we can use standard (well, almost, as mapping can change) way to provide 2807 // info for the reporting script by storing timestamp and location of symbol 2808 void linux_wrap_code(char* base, size_t size) { 2809 static volatile jint cnt = 0; 2810 2811 if (!UseOprofile) { 2812 return; 2813 } 2814 2815 char buf[PATH_MAX+1]; 2816 int num = Atomic::add(1, &cnt); 2817 2818 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2819 os::get_temp_directory(), os::current_process_id(), num); 2820 unlink(buf); 2821 2822 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2823 2824 if (fd != -1) { 2825 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2826 if (rv != (off_t)-1) { 2827 if (::write(fd, "", 1) == 1) { 2828 mmap(base, size, 2829 PROT_READ|PROT_WRITE|PROT_EXEC, 2830 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2831 } 2832 } 2833 ::close(fd); 2834 unlink(buf); 2835 } 2836 } 2837 2838 static bool recoverable_mmap_error(int err) { 2839 // See if the error is one we can let the caller handle. This 2840 // list of errno values comes from JBS-6843484. I can't find a 2841 // Linux man page that documents this specific set of errno 2842 // values so while this list currently matches Solaris, it may 2843 // change as we gain experience with this failure mode. 2844 switch (err) { 2845 case EBADF: 2846 case EINVAL: 2847 case ENOTSUP: 2848 // let the caller deal with these errors 2849 return true; 2850 2851 default: 2852 // Any remaining errors on this OS can cause our reserved mapping 2853 // to be lost. That can cause confusion where different data 2854 // structures think they have the same memory mapped. The worst 2855 // scenario is if both the VM and a library think they have the 2856 // same memory mapped. 2857 return false; 2858 } 2859 } 2860 2861 static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2862 int err) { 2863 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2864 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec, 2865 os::strerror(err), err); 2866 } 2867 2868 static void warn_fail_commit_memory(char* addr, size_t size, 2869 size_t alignment_hint, bool exec, 2870 int err) { 2871 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2872 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, 2873 alignment_hint, exec, os::strerror(err), err); 2874 } 2875 2876 // NOTE: Linux kernel does not really reserve the pages for us. 2877 // All it does is to check if there are enough free pages 2878 // left at the time of mmap(). This could be a potential 2879 // problem. 2880 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2881 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2882 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2883 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2884 if (res != (uintptr_t) MAP_FAILED) { 2885 if (UseNUMAInterleaving) { 2886 numa_make_global(addr, size); 2887 } 2888 return 0; 2889 } 2890 2891 int err = errno; // save errno from mmap() call above 2892 2893 if (!recoverable_mmap_error(err)) { 2894 warn_fail_commit_memory(addr, size, exec, err); 2895 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2896 } 2897 2898 return err; 2899 } 2900 2901 bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2902 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2903 } 2904 2905 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2906 const char* mesg) { 2907 assert(mesg != NULL, "mesg must be specified"); 2908 int err = os::Linux::commit_memory_impl(addr, size, exec); 2909 if (err != 0) { 2910 // the caller wants all commit errors to exit with the specified mesg: 2911 warn_fail_commit_memory(addr, size, exec, err); 2912 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2913 } 2914 } 2915 2916 // Define MAP_HUGETLB here so we can build HotSpot on old systems. 2917 #ifndef MAP_HUGETLB 2918 #define MAP_HUGETLB 0x40000 2919 #endif 2920 2921 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2922 #ifndef MADV_HUGEPAGE 2923 #define MADV_HUGEPAGE 14 2924 #endif 2925 2926 int os::Linux::commit_memory_impl(char* addr, size_t size, 2927 size_t alignment_hint, bool exec) { 2928 int err = os::Linux::commit_memory_impl(addr, size, exec); 2929 if (err == 0) { 2930 realign_memory(addr, size, alignment_hint); 2931 } 2932 return err; 2933 } 2934 2935 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2936 bool exec) { 2937 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2938 } 2939 2940 void os::pd_commit_memory_or_exit(char* addr, size_t size, 2941 size_t alignment_hint, bool exec, 2942 const char* mesg) { 2943 assert(mesg != NULL, "mesg must be specified"); 2944 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2945 if (err != 0) { 2946 // the caller wants all commit errors to exit with the specified mesg: 2947 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2948 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2949 } 2950 } 2951 2952 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2953 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2954 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2955 // be supported or the memory may already be backed by huge pages. 2956 ::madvise(addr, bytes, MADV_HUGEPAGE); 2957 } 2958 } 2959 2960 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2961 // This method works by doing an mmap over an existing mmaping and effectively discarding 2962 // the existing pages. However it won't work for SHM-based large pages that cannot be 2963 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2964 // small pages on top of the SHM segment. This method always works for small pages, so we 2965 // allow that in any case. 2966 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2967 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2968 } 2969 } 2970 2971 void os::numa_make_global(char *addr, size_t bytes) { 2972 Linux::numa_interleave_memory(addr, bytes); 2973 } 2974 2975 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2976 // bind policy to MPOL_PREFERRED for the current thread. 2977 #define USE_MPOL_PREFERRED 0 2978 2979 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2980 // To make NUMA and large pages more robust when both enabled, we need to ease 2981 // the requirements on where the memory should be allocated. MPOL_BIND is the 2982 // default policy and it will force memory to be allocated on the specified 2983 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2984 // the specified node, but will not force it. Using this policy will prevent 2985 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2986 // free large pages. 2987 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2988 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2989 } 2990 2991 bool os::numa_topology_changed() { return false; } 2992 2993 size_t os::numa_get_groups_num() { 2994 // Return just the number of nodes in which it's possible to allocate memory 2995 // (in numa terminology, configured nodes). 2996 return Linux::numa_num_configured_nodes(); 2997 } 2998 2999 int os::numa_get_group_id() { 3000 int cpu_id = Linux::sched_getcpu(); 3001 if (cpu_id != -1) { 3002 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 3003 if (lgrp_id != -1) { 3004 return lgrp_id; 3005 } 3006 } 3007 return 0; 3008 } 3009 3010 int os::numa_get_group_id_for_address(const void* address) { 3011 void** pages = const_cast<void**>(&address); 3012 int id = -1; 3013 3014 if (os::Linux::numa_move_pages(0, 1, pages, NULL, &id, 0) == -1) { 3015 return -1; 3016 } 3017 if (id < 0) { 3018 return -1; 3019 } 3020 return id; 3021 } 3022 3023 int os::Linux::get_existing_num_nodes() { 3024 int node; 3025 int highest_node_number = Linux::numa_max_node(); 3026 int num_nodes = 0; 3027 3028 // Get the total number of nodes in the system including nodes without memory. 3029 for (node = 0; node <= highest_node_number; node++) { 3030 if (is_node_in_existing_nodes(node)) { 3031 num_nodes++; 3032 } 3033 } 3034 return num_nodes; 3035 } 3036 3037 size_t os::numa_get_leaf_groups(int *ids, size_t size) { 3038 int highest_node_number = Linux::numa_max_node(); 3039 size_t i = 0; 3040 3041 // Map all node ids in which it is possible to allocate memory. Also nodes are 3042 // not always consecutively available, i.e. available from 0 to the highest 3043 // node number. If the nodes have been bound explicitly using numactl membind, 3044 // then allocate memory from those nodes only. 3045 for (int node = 0; node <= highest_node_number; node++) { 3046 if (Linux::is_node_in_bound_nodes((unsigned int)node)) { 3047 ids[i++] = node; 3048 } 3049 } 3050 return i; 3051 } 3052 3053 bool os::get_page_info(char *start, page_info* info) { 3054 return false; 3055 } 3056 3057 char *os::scan_pages(char *start, char* end, page_info* page_expected, 3058 page_info* page_found) { 3059 return end; 3060 } 3061 3062 3063 int os::Linux::sched_getcpu_syscall(void) { 3064 unsigned int cpu = 0; 3065 int retval = -1; 3066 3067 #if defined(IA32) 3068 #ifndef SYS_getcpu 3069 #define SYS_getcpu 318 3070 #endif 3071 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 3072 #elif defined(AMD64) 3073 // Unfortunately we have to bring all these macros here from vsyscall.h 3074 // to be able to compile on old linuxes. 3075 #define __NR_vgetcpu 2 3076 #define VSYSCALL_START (-10UL << 20) 3077 #define VSYSCALL_SIZE 1024 3078 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 3079 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 3080 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 3081 retval = vgetcpu(&cpu, NULL, NULL); 3082 #endif 3083 3084 return (retval == -1) ? retval : cpu; 3085 } 3086 3087 void os::Linux::sched_getcpu_init() { 3088 // sched_getcpu() should be in libc. 3089 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 3090 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 3091 3092 // If it's not, try a direct syscall. 3093 if (sched_getcpu() == -1) { 3094 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 3095 (void*)&sched_getcpu_syscall)); 3096 } 3097 3098 if (sched_getcpu() == -1) { 3099 vm_exit_during_initialization("getcpu(2) system call not supported by kernel"); 3100 } 3101 } 3102 3103 // Something to do with the numa-aware allocator needs these symbols 3104 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 3105 extern "C" JNIEXPORT void numa_error(char *where) { } 3106 3107 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails 3108 // load symbol from base version instead. 3109 void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 3110 void *f = dlvsym(handle, name, "libnuma_1.1"); 3111 if (f == NULL) { 3112 f = dlsym(handle, name); 3113 } 3114 return f; 3115 } 3116 3117 // Handle request to load libnuma symbol version 1.2 (API v2) only. 3118 // Return NULL if the symbol is not defined in this particular version. 3119 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) { 3120 return dlvsym(handle, name, "libnuma_1.2"); 3121 } 3122 3123 bool os::Linux::libnuma_init() { 3124 if (sched_getcpu() != -1) { // Requires sched_getcpu() support 3125 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 3126 if (handle != NULL) { 3127 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 3128 libnuma_dlsym(handle, "numa_node_to_cpus"))); 3129 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 3130 libnuma_dlsym(handle, "numa_max_node"))); 3131 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t, 3132 libnuma_dlsym(handle, "numa_num_configured_nodes"))); 3133 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 3134 libnuma_dlsym(handle, "numa_available"))); 3135 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 3136 libnuma_dlsym(handle, "numa_tonode_memory"))); 3137 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 3138 libnuma_dlsym(handle, "numa_interleave_memory"))); 3139 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t, 3140 libnuma_v2_dlsym(handle, "numa_interleave_memory"))); 3141 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 3142 libnuma_dlsym(handle, "numa_set_bind_policy"))); 3143 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t, 3144 libnuma_dlsym(handle, "numa_bitmask_isbitset"))); 3145 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t, 3146 libnuma_dlsym(handle, "numa_distance"))); 3147 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t, 3148 libnuma_v2_dlsym(handle, "numa_get_membind"))); 3149 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t, 3150 libnuma_v2_dlsym(handle, "numa_get_interleave_mask"))); 3151 set_numa_move_pages(CAST_TO_FN_PTR(numa_move_pages_func_t, 3152 libnuma_dlsym(handle, "numa_move_pages"))); 3153 3154 if (numa_available() != -1) { 3155 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 3156 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr")); 3157 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr")); 3158 set_numa_interleave_bitmask(_numa_get_interleave_mask()); 3159 set_numa_membind_bitmask(_numa_get_membind()); 3160 // Create an index -> node mapping, since nodes are not always consecutive 3161 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 3162 rebuild_nindex_to_node_map(); 3163 // Create a cpu -> node mapping 3164 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 3165 rebuild_cpu_to_node_map(); 3166 return true; 3167 } 3168 } 3169 } 3170 return false; 3171 } 3172 3173 size_t os::Linux::default_guard_size(os::ThreadType thr_type) { 3174 // Creating guard page is very expensive. Java thread has HotSpot 3175 // guard pages, only enable glibc guard page for non-Java threads. 3176 // (Remember: compiler thread is a Java thread, too!) 3177 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size()); 3178 } 3179 3180 void os::Linux::rebuild_nindex_to_node_map() { 3181 int highest_node_number = Linux::numa_max_node(); 3182 3183 nindex_to_node()->clear(); 3184 for (int node = 0; node <= highest_node_number; node++) { 3185 if (Linux::is_node_in_existing_nodes(node)) { 3186 nindex_to_node()->append(node); 3187 } 3188 } 3189 } 3190 3191 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 3192 // The table is later used in get_node_by_cpu(). 3193 void os::Linux::rebuild_cpu_to_node_map() { 3194 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 3195 // in libnuma (possible values are starting from 16, 3196 // and continuing up with every other power of 2, but less 3197 // than the maximum number of CPUs supported by kernel), and 3198 // is a subject to change (in libnuma version 2 the requirements 3199 // are more reasonable) we'll just hardcode the number they use 3200 // in the library. 3201 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 3202 3203 size_t cpu_num = processor_count(); 3204 size_t cpu_map_size = NCPUS / BitsPerCLong; 3205 size_t cpu_map_valid_size = 3206 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 3207 3208 cpu_to_node()->clear(); 3209 cpu_to_node()->at_grow(cpu_num - 1); 3210 3211 size_t node_num = get_existing_num_nodes(); 3212 3213 int distance = 0; 3214 int closest_distance = INT_MAX; 3215 int closest_node = 0; 3216 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 3217 for (size_t i = 0; i < node_num; i++) { 3218 // Check if node is configured (not a memory-less node). If it is not, find 3219 // the closest configured node. Check also if node is bound, i.e. it's allowed 3220 // to allocate memory from the node. If it's not allowed, map cpus in that node 3221 // to the closest node from which memory allocation is allowed. 3222 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) || 3223 !is_node_in_bound_nodes(nindex_to_node()->at(i))) { 3224 closest_distance = INT_MAX; 3225 // Check distance from all remaining nodes in the system. Ignore distance 3226 // from itself, from another non-configured node, and from another non-bound 3227 // node. 3228 for (size_t m = 0; m < node_num; m++) { 3229 if (m != i && 3230 is_node_in_configured_nodes(nindex_to_node()->at(m)) && 3231 is_node_in_bound_nodes(nindex_to_node()->at(m))) { 3232 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m)); 3233 // If a closest node is found, update. There is always at least one 3234 // configured and bound node in the system so there is always at least 3235 // one node close. 3236 if (distance != 0 && distance < closest_distance) { 3237 closest_distance = distance; 3238 closest_node = nindex_to_node()->at(m); 3239 } 3240 } 3241 } 3242 } else { 3243 // Current node is already a configured node. 3244 closest_node = nindex_to_node()->at(i); 3245 } 3246 3247 // Get cpus from the original node and map them to the closest node. If node 3248 // is a configured node (not a memory-less node), then original node and 3249 // closest node are the same. 3250 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 3251 for (size_t j = 0; j < cpu_map_valid_size; j++) { 3252 if (cpu_map[j] != 0) { 3253 for (size_t k = 0; k < BitsPerCLong; k++) { 3254 if (cpu_map[j] & (1UL << k)) { 3255 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node); 3256 } 3257 } 3258 } 3259 } 3260 } 3261 } 3262 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 3263 } 3264 3265 int os::Linux::get_node_by_cpu(int cpu_id) { 3266 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 3267 return cpu_to_node()->at(cpu_id); 3268 } 3269 return -1; 3270 } 3271 3272 GrowableArray<int>* os::Linux::_cpu_to_node; 3273 GrowableArray<int>* os::Linux::_nindex_to_node; 3274 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 3275 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 3276 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 3277 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes; 3278 os::Linux::numa_available_func_t os::Linux::_numa_available; 3279 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 3280 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 3281 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2; 3282 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 3283 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset; 3284 os::Linux::numa_distance_func_t os::Linux::_numa_distance; 3285 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind; 3286 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask; 3287 os::Linux::numa_move_pages_func_t os::Linux::_numa_move_pages; 3288 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy; 3289 unsigned long* os::Linux::_numa_all_nodes; 3290 struct bitmask* os::Linux::_numa_all_nodes_ptr; 3291 struct bitmask* os::Linux::_numa_nodes_ptr; 3292 struct bitmask* os::Linux::_numa_interleave_bitmask; 3293 struct bitmask* os::Linux::_numa_membind_bitmask; 3294 3295 bool os::pd_uncommit_memory(char* addr, size_t size) { 3296 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 3297 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 3298 return res != (uintptr_t) MAP_FAILED; 3299 } 3300 3301 static address get_stack_commited_bottom(address bottom, size_t size) { 3302 address nbot = bottom; 3303 address ntop = bottom + size; 3304 3305 size_t page_sz = os::vm_page_size(); 3306 unsigned pages = size / page_sz; 3307 3308 unsigned char vec[1]; 3309 unsigned imin = 1, imax = pages + 1, imid; 3310 int mincore_return_value = 0; 3311 3312 assert(imin <= imax, "Unexpected page size"); 3313 3314 while (imin < imax) { 3315 imid = (imax + imin) / 2; 3316 nbot = ntop - (imid * page_sz); 3317 3318 // Use a trick with mincore to check whether the page is mapped or not. 3319 // mincore sets vec to 1 if page resides in memory and to 0 if page 3320 // is swapped output but if page we are asking for is unmapped 3321 // it returns -1,ENOMEM 3322 mincore_return_value = mincore(nbot, page_sz, vec); 3323 3324 if (mincore_return_value == -1) { 3325 // Page is not mapped go up 3326 // to find first mapped page 3327 if (errno != EAGAIN) { 3328 assert(errno == ENOMEM, "Unexpected mincore errno"); 3329 imax = imid; 3330 } 3331 } else { 3332 // Page is mapped go down 3333 // to find first not mapped page 3334 imin = imid + 1; 3335 } 3336 } 3337 3338 nbot = nbot + page_sz; 3339 3340 // Adjust stack bottom one page up if last checked page is not mapped 3341 if (mincore_return_value == -1) { 3342 nbot = nbot + page_sz; 3343 } 3344 3345 return nbot; 3346 } 3347 3348 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) { 3349 int mincore_return_value; 3350 const size_t stripe = 1024; // query this many pages each time 3351 unsigned char vec[stripe + 1]; 3352 // set a guard 3353 vec[stripe] = 'X'; 3354 3355 const size_t page_sz = os::vm_page_size(); 3356 size_t pages = size / page_sz; 3357 3358 assert(is_aligned(start, page_sz), "Start address must be page aligned"); 3359 assert(is_aligned(size, page_sz), "Size must be page aligned"); 3360 3361 committed_start = NULL; 3362 3363 int loops = (pages + stripe - 1) / stripe; 3364 int committed_pages = 0; 3365 address loop_base = start; 3366 bool found_range = false; 3367 3368 for (int index = 0; index < loops && !found_range; index ++) { 3369 assert(pages > 0, "Nothing to do"); 3370 int pages_to_query = (pages >= stripe) ? stripe : pages; 3371 pages -= pages_to_query; 3372 3373 // Get stable read 3374 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN); 3375 3376 // During shutdown, some memory goes away without properly notifying NMT, 3377 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object. 3378 // Bailout and return as not committed for now. 3379 if (mincore_return_value == -1 && errno == ENOMEM) { 3380 return false; 3381 } 3382 3383 assert(vec[stripe] == 'X', "overflow guard"); 3384 assert(mincore_return_value == 0, "Range must be valid"); 3385 // Process this stripe 3386 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) { 3387 if ((vec[vecIdx] & 0x01) == 0) { // not committed 3388 // End of current contiguous region 3389 if (committed_start != NULL) { 3390 found_range = true; 3391 break; 3392 } 3393 } else { // committed 3394 // Start of region 3395 if (committed_start == NULL) { 3396 committed_start = loop_base + page_sz * vecIdx; 3397 } 3398 committed_pages ++; 3399 } 3400 } 3401 3402 loop_base += pages_to_query * page_sz; 3403 } 3404 3405 if (committed_start != NULL) { 3406 assert(committed_pages > 0, "Must have committed region"); 3407 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has"); 3408 assert(committed_start >= start && committed_start < start + size, "Out of range"); 3409 committed_size = page_sz * committed_pages; 3410 return true; 3411 } else { 3412 assert(committed_pages == 0, "Should not have committed region"); 3413 return false; 3414 } 3415 } 3416 3417 3418 // Linux uses a growable mapping for the stack, and if the mapping for 3419 // the stack guard pages is not removed when we detach a thread the 3420 // stack cannot grow beyond the pages where the stack guard was 3421 // mapped. If at some point later in the process the stack expands to 3422 // that point, the Linux kernel cannot expand the stack any further 3423 // because the guard pages are in the way, and a segfault occurs. 3424 // 3425 // However, it's essential not to split the stack region by unmapping 3426 // a region (leaving a hole) that's already part of the stack mapping, 3427 // so if the stack mapping has already grown beyond the guard pages at 3428 // the time we create them, we have to truncate the stack mapping. 3429 // So, we need to know the extent of the stack mapping when 3430 // create_stack_guard_pages() is called. 3431 3432 // We only need this for stacks that are growable: at the time of 3433 // writing thread stacks don't use growable mappings (i.e. those 3434 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3435 // only applies to the main thread. 3436 3437 // If the (growable) stack mapping already extends beyond the point 3438 // where we're going to put our guard pages, truncate the mapping at 3439 // that point by munmap()ping it. This ensures that when we later 3440 // munmap() the guard pages we don't leave a hole in the stack 3441 // mapping. This only affects the main/primordial thread 3442 3443 bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3444 if (os::is_primordial_thread()) { 3445 // As we manually grow stack up to bottom inside create_attached_thread(), 3446 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3447 // we don't need to do anything special. 3448 // Check it first, before calling heavy function. 3449 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3450 unsigned char vec[1]; 3451 3452 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3453 // Fallback to slow path on all errors, including EAGAIN 3454 stack_extent = (uintptr_t) get_stack_commited_bottom( 3455 os::Linux::initial_thread_stack_bottom(), 3456 (size_t)addr - stack_extent); 3457 } 3458 3459 if (stack_extent < (uintptr_t)addr) { 3460 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3461 } 3462 } 3463 3464 return os::commit_memory(addr, size, !ExecMem); 3465 } 3466 3467 // If this is a growable mapping, remove the guard pages entirely by 3468 // munmap()ping them. If not, just call uncommit_memory(). This only 3469 // affects the main/primordial thread, but guard against future OS changes. 3470 // It's safe to always unmap guard pages for primordial thread because we 3471 // always place it right after end of the mapped region. 3472 3473 bool os::remove_stack_guard_pages(char* addr, size_t size) { 3474 uintptr_t stack_extent, stack_base; 3475 3476 if (os::is_primordial_thread()) { 3477 return ::munmap(addr, size) == 0; 3478 } 3479 3480 return os::uncommit_memory(addr, size); 3481 } 3482 3483 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3484 // at 'requested_addr'. If there are existing memory mappings at the same 3485 // location, however, they will be overwritten. If 'fixed' is false, 3486 // 'requested_addr' is only treated as a hint, the return value may or 3487 // may not start from the requested address. Unlike Linux mmap(), this 3488 // function returns NULL to indicate failure. 3489 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3490 char * addr; 3491 int flags; 3492 3493 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3494 if (fixed) { 3495 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3496 flags |= MAP_FIXED; 3497 } 3498 3499 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3500 // touch an uncommitted page. Otherwise, the read/write might 3501 // succeed if we have enough swap space to back the physical page. 3502 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3503 flags, -1, 0); 3504 3505 return addr == MAP_FAILED ? NULL : addr; 3506 } 3507 3508 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address 3509 // (req_addr != NULL) or with a given alignment. 3510 // - bytes shall be a multiple of alignment. 3511 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3512 // - alignment sets the alignment at which memory shall be allocated. 3513 // It must be a multiple of allocation granularity. 3514 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3515 // req_addr or NULL. 3516 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) { 3517 3518 size_t extra_size = bytes; 3519 if (req_addr == NULL && alignment > 0) { 3520 extra_size += alignment; 3521 } 3522 3523 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE, 3524 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 3525 -1, 0); 3526 if (start == MAP_FAILED) { 3527 start = NULL; 3528 } else { 3529 if (req_addr != NULL) { 3530 if (start != req_addr) { 3531 ::munmap(start, extra_size); 3532 start = NULL; 3533 } 3534 } else { 3535 char* const start_aligned = align_up(start, alignment); 3536 char* const end_aligned = start_aligned + bytes; 3537 char* const end = start + extra_size; 3538 if (start_aligned > start) { 3539 ::munmap(start, start_aligned - start); 3540 } 3541 if (end_aligned < end) { 3542 ::munmap(end_aligned, end - end_aligned); 3543 } 3544 start = start_aligned; 3545 } 3546 } 3547 return start; 3548 } 3549 3550 static int anon_munmap(char * addr, size_t size) { 3551 return ::munmap(addr, size) == 0; 3552 } 3553 3554 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3555 size_t alignment_hint) { 3556 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3557 } 3558 3559 bool os::pd_release_memory(char* addr, size_t size) { 3560 return anon_munmap(addr, size); 3561 } 3562 3563 static bool linux_mprotect(char* addr, size_t size, int prot) { 3564 // Linux wants the mprotect address argument to be page aligned. 3565 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size()); 3566 3567 // According to SUSv3, mprotect() should only be used with mappings 3568 // established by mmap(), and mmap() always maps whole pages. Unaligned 3569 // 'addr' likely indicates problem in the VM (e.g. trying to change 3570 // protection of malloc'ed or statically allocated memory). Check the 3571 // caller if you hit this assert. 3572 assert(addr == bottom, "sanity check"); 3573 3574 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3575 Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot); 3576 return ::mprotect(bottom, size, prot) == 0; 3577 } 3578 3579 // Set protections specified 3580 bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3581 bool is_committed) { 3582 unsigned int p = 0; 3583 switch (prot) { 3584 case MEM_PROT_NONE: p = PROT_NONE; break; 3585 case MEM_PROT_READ: p = PROT_READ; break; 3586 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3587 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3588 default: 3589 ShouldNotReachHere(); 3590 } 3591 // is_committed is unused. 3592 return linux_mprotect(addr, bytes, p); 3593 } 3594 3595 bool os::guard_memory(char* addr, size_t size) { 3596 return linux_mprotect(addr, size, PROT_NONE); 3597 } 3598 3599 bool os::unguard_memory(char* addr, size_t size) { 3600 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3601 } 3602 3603 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3604 size_t page_size) { 3605 bool result = false; 3606 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3607 MAP_ANONYMOUS|MAP_PRIVATE, 3608 -1, 0); 3609 if (p != MAP_FAILED) { 3610 void *aligned_p = align_up(p, page_size); 3611 3612 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3613 3614 munmap(p, page_size * 2); 3615 } 3616 3617 if (warn && !result) { 3618 warning("TransparentHugePages is not supported by the operating system."); 3619 } 3620 3621 return result; 3622 } 3623 3624 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3625 bool result = false; 3626 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3627 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3628 -1, 0); 3629 3630 if (p != MAP_FAILED) { 3631 // We don't know if this really is a huge page or not. 3632 FILE *fp = fopen("/proc/self/maps", "r"); 3633 if (fp) { 3634 while (!feof(fp)) { 3635 char chars[257]; 3636 long x = 0; 3637 if (fgets(chars, sizeof(chars), fp)) { 3638 if (sscanf(chars, "%lx-%*x", &x) == 1 3639 && x == (long)p) { 3640 if (strstr (chars, "hugepage")) { 3641 result = true; 3642 break; 3643 } 3644 } 3645 } 3646 } 3647 fclose(fp); 3648 } 3649 munmap(p, page_size); 3650 } 3651 3652 if (warn && !result) { 3653 warning("HugeTLBFS is not supported by the operating system."); 3654 } 3655 3656 return result; 3657 } 3658 3659 // From the coredump_filter documentation: 3660 // 3661 // - (bit 0) anonymous private memory 3662 // - (bit 1) anonymous shared memory 3663 // - (bit 2) file-backed private memory 3664 // - (bit 3) file-backed shared memory 3665 // - (bit 4) ELF header pages in file-backed private memory areas (it is 3666 // effective only if the bit 2 is cleared) 3667 // - (bit 5) hugetlb private memory 3668 // - (bit 6) hugetlb shared memory 3669 // - (bit 7) dax private memory 3670 // - (bit 8) dax shared memory 3671 // 3672 static void set_coredump_filter(CoredumpFilterBit bit) { 3673 FILE *f; 3674 long cdm; 3675 3676 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3677 return; 3678 } 3679 3680 if (fscanf(f, "%lx", &cdm) != 1) { 3681 fclose(f); 3682 return; 3683 } 3684 3685 long saved_cdm = cdm; 3686 rewind(f); 3687 cdm |= bit; 3688 3689 if (cdm != saved_cdm) { 3690 fprintf(f, "%#lx", cdm); 3691 } 3692 3693 fclose(f); 3694 } 3695 3696 // Large page support 3697 3698 static size_t _large_page_size = 0; 3699 3700 size_t os::Linux::find_large_page_size() { 3701 size_t large_page_size = 0; 3702 3703 // large_page_size on Linux is used to round up heap size. x86 uses either 3704 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3705 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3706 // page as large as 256M. 3707 // 3708 // Here we try to figure out page size by parsing /proc/meminfo and looking 3709 // for a line with the following format: 3710 // Hugepagesize: 2048 kB 3711 // 3712 // If we can't determine the value (e.g. /proc is not mounted, or the text 3713 // format has been changed), we'll use the largest page size supported by 3714 // the processor. 3715 3716 #ifndef ZERO 3717 large_page_size = 3718 AARCH64_ONLY(2 * M) 3719 AMD64_ONLY(2 * M) 3720 ARM32_ONLY(2 * M) 3721 IA32_ONLY(4 * M) 3722 IA64_ONLY(256 * M) 3723 PPC_ONLY(4 * M) 3724 S390_ONLY(1 * M) 3725 SPARC_ONLY(4 * M); 3726 #endif // ZERO 3727 3728 FILE *fp = fopen("/proc/meminfo", "r"); 3729 if (fp) { 3730 while (!feof(fp)) { 3731 int x = 0; 3732 char buf[16]; 3733 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3734 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3735 large_page_size = x * K; 3736 break; 3737 } 3738 } else { 3739 // skip to next line 3740 for (;;) { 3741 int ch = fgetc(fp); 3742 if (ch == EOF || ch == (int)'\n') break; 3743 } 3744 } 3745 } 3746 fclose(fp); 3747 } 3748 3749 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3750 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3751 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3752 proper_unit_for_byte_size(large_page_size)); 3753 } 3754 3755 return large_page_size; 3756 } 3757 3758 size_t os::Linux::setup_large_page_size() { 3759 _large_page_size = Linux::find_large_page_size(); 3760 const size_t default_page_size = (size_t)Linux::page_size(); 3761 if (_large_page_size > default_page_size) { 3762 _page_sizes[0] = _large_page_size; 3763 _page_sizes[1] = default_page_size; 3764 _page_sizes[2] = 0; 3765 } 3766 3767 return _large_page_size; 3768 } 3769 3770 bool os::Linux::setup_large_page_type(size_t page_size) { 3771 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3772 FLAG_IS_DEFAULT(UseSHM) && 3773 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3774 3775 // The type of large pages has not been specified by the user. 3776 3777 // Try UseHugeTLBFS and then UseSHM. 3778 UseHugeTLBFS = UseSHM = true; 3779 3780 // Don't try UseTransparentHugePages since there are known 3781 // performance issues with it turned on. This might change in the future. 3782 UseTransparentHugePages = false; 3783 } 3784 3785 if (UseTransparentHugePages) { 3786 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3787 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3788 UseHugeTLBFS = false; 3789 UseSHM = false; 3790 return true; 3791 } 3792 UseTransparentHugePages = false; 3793 } 3794 3795 if (UseHugeTLBFS) { 3796 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3797 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3798 UseSHM = false; 3799 return true; 3800 } 3801 UseHugeTLBFS = false; 3802 } 3803 3804 return UseSHM; 3805 } 3806 3807 void os::large_page_init() { 3808 if (!UseLargePages && 3809 !UseTransparentHugePages && 3810 !UseHugeTLBFS && 3811 !UseSHM) { 3812 // Not using large pages. 3813 return; 3814 } 3815 3816 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3817 // The user explicitly turned off large pages. 3818 // Ignore the rest of the large pages flags. 3819 UseTransparentHugePages = false; 3820 UseHugeTLBFS = false; 3821 UseSHM = false; 3822 return; 3823 } 3824 3825 size_t large_page_size = Linux::setup_large_page_size(); 3826 UseLargePages = Linux::setup_large_page_type(large_page_size); 3827 3828 set_coredump_filter(LARGEPAGES_BIT); 3829 } 3830 3831 #ifndef SHM_HUGETLB 3832 #define SHM_HUGETLB 04000 3833 #endif 3834 3835 #define shm_warning_format(format, ...) \ 3836 do { \ 3837 if (UseLargePages && \ 3838 (!FLAG_IS_DEFAULT(UseLargePages) || \ 3839 !FLAG_IS_DEFAULT(UseSHM) || \ 3840 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \ 3841 warning(format, __VA_ARGS__); \ 3842 } \ 3843 } while (0) 3844 3845 #define shm_warning(str) shm_warning_format("%s", str) 3846 3847 #define shm_warning_with_errno(str) \ 3848 do { \ 3849 int err = errno; \ 3850 shm_warning_format(str " (error = %d)", err); \ 3851 } while (0) 3852 3853 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) { 3854 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment"); 3855 3856 if (!is_aligned(alignment, SHMLBA)) { 3857 assert(false, "Code below assumes that alignment is at least SHMLBA aligned"); 3858 return NULL; 3859 } 3860 3861 // To ensure that we get 'alignment' aligned memory from shmat, 3862 // we pre-reserve aligned virtual memory and then attach to that. 3863 3864 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL); 3865 if (pre_reserved_addr == NULL) { 3866 // Couldn't pre-reserve aligned memory. 3867 shm_warning("Failed to pre-reserve aligned memory for shmat."); 3868 return NULL; 3869 } 3870 3871 // SHM_REMAP is needed to allow shmat to map over an existing mapping. 3872 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP); 3873 3874 if ((intptr_t)addr == -1) { 3875 int err = errno; 3876 shm_warning_with_errno("Failed to attach shared memory."); 3877 3878 assert(err != EACCES, "Unexpected error"); 3879 assert(err != EIDRM, "Unexpected error"); 3880 assert(err != EINVAL, "Unexpected error"); 3881 3882 // Since we don't know if the kernel unmapped the pre-reserved memory area 3883 // we can't unmap it, since that would potentially unmap memory that was 3884 // mapped from other threads. 3885 return NULL; 3886 } 3887 3888 return addr; 3889 } 3890 3891 static char* shmat_at_address(int shmid, char* req_addr) { 3892 if (!is_aligned(req_addr, SHMLBA)) { 3893 assert(false, "Requested address needs to be SHMLBA aligned"); 3894 return NULL; 3895 } 3896 3897 char* addr = (char*)shmat(shmid, req_addr, 0); 3898 3899 if ((intptr_t)addr == -1) { 3900 shm_warning_with_errno("Failed to attach shared memory."); 3901 return NULL; 3902 } 3903 3904 return addr; 3905 } 3906 3907 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) { 3908 // If a req_addr has been provided, we assume that the caller has already aligned the address. 3909 if (req_addr != NULL) { 3910 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size"); 3911 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment"); 3912 return shmat_at_address(shmid, req_addr); 3913 } 3914 3915 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically 3916 // return large page size aligned memory addresses when req_addr == NULL. 3917 // However, if the alignment is larger than the large page size, we have 3918 // to manually ensure that the memory returned is 'alignment' aligned. 3919 if (alignment > os::large_page_size()) { 3920 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size"); 3921 return shmat_with_alignment(shmid, bytes, alignment); 3922 } else { 3923 return shmat_at_address(shmid, NULL); 3924 } 3925 } 3926 3927 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3928 char* req_addr, bool exec) { 3929 // "exec" is passed in but not used. Creating the shared image for 3930 // the code cache doesn't have an SHM_X executable permission to check. 3931 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3932 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3933 assert(is_aligned(req_addr, alignment), "Unaligned address"); 3934 3935 if (!is_aligned(bytes, os::large_page_size())) { 3936 return NULL; // Fallback to small pages. 3937 } 3938 3939 // Create a large shared memory region to attach to based on size. 3940 // Currently, size is the total size of the heap. 3941 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3942 if (shmid == -1) { 3943 // Possible reasons for shmget failure: 3944 // 1. shmmax is too small for Java heap. 3945 // > check shmmax value: cat /proc/sys/kernel/shmmax 3946 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3947 // 2. not enough large page memory. 3948 // > check available large pages: cat /proc/meminfo 3949 // > increase amount of large pages: 3950 // echo new_value > /proc/sys/vm/nr_hugepages 3951 // Note 1: different Linux may use different name for this property, 3952 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3953 // Note 2: it's possible there's enough physical memory available but 3954 // they are so fragmented after a long run that they can't 3955 // coalesce into large pages. Try to reserve large pages when 3956 // the system is still "fresh". 3957 shm_warning_with_errno("Failed to reserve shared memory."); 3958 return NULL; 3959 } 3960 3961 // Attach to the region. 3962 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr); 3963 3964 // Remove shmid. If shmat() is successful, the actual shared memory segment 3965 // will be deleted when it's detached by shmdt() or when the process 3966 // terminates. If shmat() is not successful this will remove the shared 3967 // segment immediately. 3968 shmctl(shmid, IPC_RMID, NULL); 3969 3970 return addr; 3971 } 3972 3973 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3974 int error) { 3975 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3976 3977 bool warn_on_failure = UseLargePages && 3978 (!FLAG_IS_DEFAULT(UseLargePages) || 3979 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3980 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3981 3982 if (warn_on_failure) { 3983 char msg[128]; 3984 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3985 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3986 warning("%s", msg); 3987 } 3988 } 3989 3990 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3991 char* req_addr, 3992 bool exec) { 3993 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3994 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size"); 3995 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3996 3997 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3998 char* addr = (char*)::mmap(req_addr, bytes, prot, 3999 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 4000 -1, 0); 4001 4002 if (addr == MAP_FAILED) { 4003 warn_on_large_pages_failure(req_addr, bytes, errno); 4004 return NULL; 4005 } 4006 4007 assert(is_aligned(addr, os::large_page_size()), "Must be"); 4008 4009 return addr; 4010 } 4011 4012 // Reserve memory using mmap(MAP_HUGETLB). 4013 // - bytes shall be a multiple of alignment. 4014 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 4015 // - alignment sets the alignment at which memory shall be allocated. 4016 // It must be a multiple of allocation granularity. 4017 // Returns address of memory or NULL. If req_addr was not NULL, will only return 4018 // req_addr or NULL. 4019 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 4020 size_t alignment, 4021 char* req_addr, 4022 bool exec) { 4023 size_t large_page_size = os::large_page_size(); 4024 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 4025 4026 assert(is_aligned(req_addr, alignment), "Must be"); 4027 assert(is_aligned(bytes, alignment), "Must be"); 4028 4029 // First reserve - but not commit - the address range in small pages. 4030 char* const start = anon_mmap_aligned(bytes, alignment, req_addr); 4031 4032 if (start == NULL) { 4033 return NULL; 4034 } 4035 4036 assert(is_aligned(start, alignment), "Must be"); 4037 4038 char* end = start + bytes; 4039 4040 // Find the regions of the allocated chunk that can be promoted to large pages. 4041 char* lp_start = align_up(start, large_page_size); 4042 char* lp_end = align_down(end, large_page_size); 4043 4044 size_t lp_bytes = lp_end - lp_start; 4045 4046 assert(is_aligned(lp_bytes, large_page_size), "Must be"); 4047 4048 if (lp_bytes == 0) { 4049 // The mapped region doesn't even span the start and the end of a large page. 4050 // Fall back to allocate a non-special area. 4051 ::munmap(start, end - start); 4052 return NULL; 4053 } 4054 4055 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 4056 4057 void* result; 4058 4059 // Commit small-paged leading area. 4060 if (start != lp_start) { 4061 result = ::mmap(start, lp_start - start, prot, 4062 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 4063 -1, 0); 4064 if (result == MAP_FAILED) { 4065 ::munmap(lp_start, end - lp_start); 4066 return NULL; 4067 } 4068 } 4069 4070 // Commit large-paged area. 4071 result = ::mmap(lp_start, lp_bytes, prot, 4072 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 4073 -1, 0); 4074 if (result == MAP_FAILED) { 4075 warn_on_large_pages_failure(lp_start, lp_bytes, errno); 4076 // If the mmap above fails, the large pages region will be unmapped and we 4077 // have regions before and after with small pages. Release these regions. 4078 // 4079 // | mapped | unmapped | mapped | 4080 // ^ ^ ^ ^ 4081 // start lp_start lp_end end 4082 // 4083 ::munmap(start, lp_start - start); 4084 ::munmap(lp_end, end - lp_end); 4085 return NULL; 4086 } 4087 4088 // Commit small-paged trailing area. 4089 if (lp_end != end) { 4090 result = ::mmap(lp_end, end - lp_end, prot, 4091 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 4092 -1, 0); 4093 if (result == MAP_FAILED) { 4094 ::munmap(start, lp_end - start); 4095 return NULL; 4096 } 4097 } 4098 4099 return start; 4100 } 4101 4102 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 4103 size_t alignment, 4104 char* req_addr, 4105 bool exec) { 4106 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 4107 assert(is_aligned(req_addr, alignment), "Must be"); 4108 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be"); 4109 assert(is_power_of_2(os::large_page_size()), "Must be"); 4110 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 4111 4112 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 4113 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 4114 } else { 4115 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 4116 } 4117 } 4118 4119 char* os::reserve_memory_special(size_t bytes, size_t alignment, 4120 char* req_addr, bool exec) { 4121 assert(UseLargePages, "only for large pages"); 4122 4123 char* addr; 4124 if (UseSHM) { 4125 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 4126 } else { 4127 assert(UseHugeTLBFS, "must be"); 4128 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 4129 } 4130 4131 if (addr != NULL) { 4132 if (UseNUMAInterleaving) { 4133 numa_make_global(addr, bytes); 4134 } 4135 4136 // The memory is committed 4137 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 4138 } 4139 4140 return addr; 4141 } 4142 4143 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 4144 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 4145 return shmdt(base) == 0; 4146 } 4147 4148 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 4149 return pd_release_memory(base, bytes); 4150 } 4151 4152 bool os::release_memory_special(char* base, size_t bytes) { 4153 bool res; 4154 if (MemTracker::tracking_level() > NMT_minimal) { 4155 Tracker tkr(Tracker::release); 4156 res = os::Linux::release_memory_special_impl(base, bytes); 4157 if (res) { 4158 tkr.record((address)base, bytes); 4159 } 4160 4161 } else { 4162 res = os::Linux::release_memory_special_impl(base, bytes); 4163 } 4164 return res; 4165 } 4166 4167 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 4168 assert(UseLargePages, "only for large pages"); 4169 bool res; 4170 4171 if (UseSHM) { 4172 res = os::Linux::release_memory_special_shm(base, bytes); 4173 } else { 4174 assert(UseHugeTLBFS, "must be"); 4175 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 4176 } 4177 return res; 4178 } 4179 4180 size_t os::large_page_size() { 4181 return _large_page_size; 4182 } 4183 4184 // With SysV SHM the entire memory region must be allocated as shared 4185 // memory. 4186 // HugeTLBFS allows application to commit large page memory on demand. 4187 // However, when committing memory with HugeTLBFS fails, the region 4188 // that was supposed to be committed will lose the old reservation 4189 // and allow other threads to steal that memory region. Because of this 4190 // behavior we can't commit HugeTLBFS memory. 4191 bool os::can_commit_large_page_memory() { 4192 return UseTransparentHugePages; 4193 } 4194 4195 bool os::can_execute_large_page_memory() { 4196 return UseTransparentHugePages || UseHugeTLBFS; 4197 } 4198 4199 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) { 4200 assert(file_desc >= 0, "file_desc is not valid"); 4201 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr); 4202 if (result != NULL) { 4203 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) { 4204 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory")); 4205 } 4206 } 4207 return result; 4208 } 4209 4210 // Reserve memory at an arbitrary address, only if that area is 4211 // available (and not reserved for something else). 4212 4213 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 4214 // Assert only that the size is a multiple of the page size, since 4215 // that's all that mmap requires, and since that's all we really know 4216 // about at this low abstraction level. If we need higher alignment, 4217 // we can either pass an alignment to this method or verify alignment 4218 // in one of the methods further up the call chain. See bug 5044738. 4219 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 4220 4221 // Repeatedly allocate blocks until the block is allocated at the 4222 // right spot. 4223 4224 // Linux mmap allows caller to pass an address as hint; give it a try first, 4225 // if kernel honors the hint then we can return immediately. 4226 char * addr = anon_mmap(requested_addr, bytes, false); 4227 if (addr == requested_addr) { 4228 return requested_addr; 4229 } 4230 4231 if (addr != NULL) { 4232 // mmap() is successful but it fails to reserve at the requested address 4233 anon_munmap(addr, bytes); 4234 } 4235 4236 return NULL; 4237 } 4238 4239 // Sleep forever; naked call to OS-specific sleep; use with CAUTION 4240 void os::infinite_sleep() { 4241 while (true) { // sleep forever ... 4242 ::sleep(100); // ... 100 seconds at a time 4243 } 4244 } 4245 4246 // Used to convert frequent JVM_Yield() to nops 4247 bool os::dont_yield() { 4248 return DontYieldALot; 4249 } 4250 4251 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will 4252 // actually give up the CPU. Since skip buddy (v2.6.28): 4253 // 4254 // * Sets the yielding task as skip buddy for current CPU's run queue. 4255 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task). 4256 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy). 4257 // 4258 // An alternative is calling os::naked_short_nanosleep with a small number to avoid 4259 // getting re-scheduled immediately. 4260 // 4261 void os::naked_yield() { 4262 sched_yield(); 4263 } 4264 4265 //////////////////////////////////////////////////////////////////////////////// 4266 // thread priority support 4267 4268 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 4269 // only supports dynamic priority, static priority must be zero. For real-time 4270 // applications, Linux supports SCHED_RR which allows static priority (1-99). 4271 // However, for large multi-threaded applications, SCHED_RR is not only slower 4272 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 4273 // of 5 runs - Sep 2005). 4274 // 4275 // The following code actually changes the niceness of kernel-thread/LWP. It 4276 // has an assumption that setpriority() only modifies one kernel-thread/LWP, 4277 // not the entire user process, and user level threads are 1:1 mapped to kernel 4278 // threads. It has always been the case, but could change in the future. For 4279 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). 4280 // It is only used when ThreadPriorityPolicy=1 and may require system level permission 4281 // (e.g., root privilege or CAP_SYS_NICE capability). 4282 4283 int os::java_to_os_priority[CriticalPriority + 1] = { 4284 19, // 0 Entry should never be used 4285 4286 4, // 1 MinPriority 4287 3, // 2 4288 2, // 3 4289 4290 1, // 4 4291 0, // 5 NormPriority 4292 -1, // 6 4293 4294 -2, // 7 4295 -3, // 8 4296 -4, // 9 NearMaxPriority 4297 4298 -5, // 10 MaxPriority 4299 4300 -5 // 11 CriticalPriority 4301 }; 4302 4303 static int prio_init() { 4304 if (ThreadPriorityPolicy == 1) { 4305 if (geteuid() != 0) { 4306 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 4307 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \ 4308 "e.g., being the root user. If the necessary permission is not " \ 4309 "possessed, changes to priority will be silently ignored."); 4310 } 4311 } 4312 } 4313 if (UseCriticalJavaThreadPriority) { 4314 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 4315 } 4316 return 0; 4317 } 4318 4319 OSReturn os::set_native_priority(Thread* thread, int newpri) { 4320 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 4321 4322 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 4323 return (ret == 0) ? OS_OK : OS_ERR; 4324 } 4325 4326 OSReturn os::get_native_priority(const Thread* const thread, 4327 int *priority_ptr) { 4328 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 4329 *priority_ptr = java_to_os_priority[NormPriority]; 4330 return OS_OK; 4331 } 4332 4333 errno = 0; 4334 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 4335 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 4336 } 4337 4338 //////////////////////////////////////////////////////////////////////////////// 4339 // suspend/resume support 4340 4341 // The low-level signal-based suspend/resume support is a remnant from the 4342 // old VM-suspension that used to be for java-suspension, safepoints etc, 4343 // within hotspot. Currently used by JFR's OSThreadSampler 4344 // 4345 // The remaining code is greatly simplified from the more general suspension 4346 // code that used to be used. 4347 // 4348 // The protocol is quite simple: 4349 // - suspend: 4350 // - sends a signal to the target thread 4351 // - polls the suspend state of the osthread using a yield loop 4352 // - target thread signal handler (SR_handler) sets suspend state 4353 // and blocks in sigsuspend until continued 4354 // - resume: 4355 // - sets target osthread state to continue 4356 // - sends signal to end the sigsuspend loop in the SR_handler 4357 // 4358 // Note that the SR_lock plays no role in this suspend/resume protocol, 4359 // but is checked for NULL in SR_handler as a thread termination indicator. 4360 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs. 4361 // 4362 // Note that resume_clear_context() and suspend_save_context() are needed 4363 // by SR_handler(), so that fetch_frame_from_ucontext() works, 4364 // which in part is used by: 4365 // - Forte Analyzer: AsyncGetCallTrace() 4366 // - StackBanging: get_frame_at_stack_banging_point() 4367 4368 static void resume_clear_context(OSThread *osthread) { 4369 osthread->set_ucontext(NULL); 4370 osthread->set_siginfo(NULL); 4371 } 4372 4373 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 4374 ucontext_t* context) { 4375 osthread->set_ucontext(context); 4376 osthread->set_siginfo(siginfo); 4377 } 4378 4379 // Handler function invoked when a thread's execution is suspended or 4380 // resumed. We have to be careful that only async-safe functions are 4381 // called here (Note: most pthread functions are not async safe and 4382 // should be avoided.) 4383 // 4384 // Note: sigwait() is a more natural fit than sigsuspend() from an 4385 // interface point of view, but sigwait() prevents the signal hander 4386 // from being run. libpthread would get very confused by not having 4387 // its signal handlers run and prevents sigwait()'s use with the 4388 // mutex granting granting signal. 4389 // 4390 // Currently only ever called on the VMThread and JavaThreads (PC sampling) 4391 // 4392 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 4393 // Save and restore errno to avoid confusing native code with EINTR 4394 // after sigsuspend. 4395 int old_errno = errno; 4396 4397 Thread* thread = Thread::current_or_null_safe(); 4398 assert(thread != NULL, "Missing current thread in SR_handler"); 4399 4400 // On some systems we have seen signal delivery get "stuck" until the signal 4401 // mask is changed as part of thread termination. Check that the current thread 4402 // has not already terminated (via SR_lock()) - else the following assertion 4403 // will fail because the thread is no longer a JavaThread as the ~JavaThread 4404 // destructor has completed. 4405 4406 if (thread->SR_lock() == NULL) { 4407 return; 4408 } 4409 4410 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 4411 4412 OSThread* osthread = thread->osthread(); 4413 4414 os::SuspendResume::State current = osthread->sr.state(); 4415 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 4416 suspend_save_context(osthread, siginfo, context); 4417 4418 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4419 os::SuspendResume::State state = osthread->sr.suspended(); 4420 if (state == os::SuspendResume::SR_SUSPENDED) { 4421 sigset_t suspend_set; // signals for sigsuspend() 4422 sigemptyset(&suspend_set); 4423 // get current set of blocked signals and unblock resume signal 4424 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4425 sigdelset(&suspend_set, SR_signum); 4426 4427 sr_semaphore.signal(); 4428 // wait here until we are resumed 4429 while (1) { 4430 sigsuspend(&suspend_set); 4431 4432 os::SuspendResume::State result = osthread->sr.running(); 4433 if (result == os::SuspendResume::SR_RUNNING) { 4434 sr_semaphore.signal(); 4435 break; 4436 } 4437 } 4438 4439 } else if (state == os::SuspendResume::SR_RUNNING) { 4440 // request was cancelled, continue 4441 } else { 4442 ShouldNotReachHere(); 4443 } 4444 4445 resume_clear_context(osthread); 4446 } else if (current == os::SuspendResume::SR_RUNNING) { 4447 // request was cancelled, continue 4448 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4449 // ignore 4450 } else { 4451 // ignore 4452 } 4453 4454 errno = old_errno; 4455 } 4456 4457 static int SR_initialize() { 4458 struct sigaction act; 4459 char *s; 4460 4461 // Get signal number to use for suspend/resume 4462 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4463 int sig = ::strtol(s, 0, 10); 4464 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769. 4465 sig < NSIG) { // Must be legal signal and fit into sigflags[]. 4466 SR_signum = sig; 4467 } else { 4468 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.", 4469 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum); 4470 } 4471 } 4472 4473 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4474 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4475 4476 sigemptyset(&SR_sigset); 4477 sigaddset(&SR_sigset, SR_signum); 4478 4479 // Set up signal handler for suspend/resume 4480 act.sa_flags = SA_RESTART|SA_SIGINFO; 4481 act.sa_handler = (void (*)(int)) SR_handler; 4482 4483 // SR_signum is blocked by default. 4484 // 4528190 - We also need to block pthread restart signal (32 on all 4485 // supported Linux platforms). Note that LinuxThreads need to block 4486 // this signal for all threads to work properly. So we don't have 4487 // to use hard-coded signal number when setting up the mask. 4488 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4489 4490 if (sigaction(SR_signum, &act, 0) == -1) { 4491 return -1; 4492 } 4493 4494 // Save signal flag 4495 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4496 return 0; 4497 } 4498 4499 static int sr_notify(OSThread* osthread) { 4500 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4501 assert_status(status == 0, status, "pthread_kill"); 4502 return status; 4503 } 4504 4505 // "Randomly" selected value for how long we want to spin 4506 // before bailing out on suspending a thread, also how often 4507 // we send a signal to a thread we want to resume 4508 static const int RANDOMLY_LARGE_INTEGER = 1000000; 4509 static const int RANDOMLY_LARGE_INTEGER2 = 100; 4510 4511 // returns true on success and false on error - really an error is fatal 4512 // but this seems the normal response to library errors 4513 static bool do_suspend(OSThread* osthread) { 4514 assert(osthread->sr.is_running(), "thread should be running"); 4515 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4516 4517 // mark as suspended and send signal 4518 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4519 // failed to switch, state wasn't running? 4520 ShouldNotReachHere(); 4521 return false; 4522 } 4523 4524 if (sr_notify(osthread) != 0) { 4525 ShouldNotReachHere(); 4526 } 4527 4528 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4529 while (true) { 4530 if (sr_semaphore.timedwait(2)) { 4531 break; 4532 } else { 4533 // timeout 4534 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4535 if (cancelled == os::SuspendResume::SR_RUNNING) { 4536 return false; 4537 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4538 // make sure that we consume the signal on the semaphore as well 4539 sr_semaphore.wait(); 4540 break; 4541 } else { 4542 ShouldNotReachHere(); 4543 return false; 4544 } 4545 } 4546 } 4547 4548 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4549 return true; 4550 } 4551 4552 static void do_resume(OSThread* osthread) { 4553 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4554 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4555 4556 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4557 // failed to switch to WAKEUP_REQUEST 4558 ShouldNotReachHere(); 4559 return; 4560 } 4561 4562 while (true) { 4563 if (sr_notify(osthread) == 0) { 4564 if (sr_semaphore.timedwait(2)) { 4565 if (osthread->sr.is_running()) { 4566 return; 4567 } 4568 } 4569 } else { 4570 ShouldNotReachHere(); 4571 } 4572 } 4573 4574 guarantee(osthread->sr.is_running(), "Must be running!"); 4575 } 4576 4577 /////////////////////////////////////////////////////////////////////////////////// 4578 // signal handling (except suspend/resume) 4579 4580 // This routine may be used by user applications as a "hook" to catch signals. 4581 // The user-defined signal handler must pass unrecognized signals to this 4582 // routine, and if it returns true (non-zero), then the signal handler must 4583 // return immediately. If the flag "abort_if_unrecognized" is true, then this 4584 // routine will never retun false (zero), but instead will execute a VM panic 4585 // routine kill the process. 4586 // 4587 // If this routine returns false, it is OK to call it again. This allows 4588 // the user-defined signal handler to perform checks either before or after 4589 // the VM performs its own checks. Naturally, the user code would be making 4590 // a serious error if it tried to handle an exception (such as a null check 4591 // or breakpoint) that the VM was generating for its own correct operation. 4592 // 4593 // This routine may recognize any of the following kinds of signals: 4594 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4595 // It should be consulted by handlers for any of those signals. 4596 // 4597 // The caller of this routine must pass in the three arguments supplied 4598 // to the function referred to in the "sa_sigaction" (not the "sa_handler") 4599 // field of the structure passed to sigaction(). This routine assumes that 4600 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4601 // 4602 // Note that the VM will print warnings if it detects conflicting signal 4603 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4604 // 4605 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4606 siginfo_t* siginfo, 4607 void* ucontext, 4608 int abort_if_unrecognized); 4609 4610 static void signalHandler(int sig, siginfo_t* info, void* uc) { 4611 assert(info != NULL && uc != NULL, "it must be old kernel"); 4612 int orig_errno = errno; // Preserve errno value over signal handler. 4613 JVM_handle_linux_signal(sig, info, uc, true); 4614 errno = orig_errno; 4615 } 4616 4617 4618 // This boolean allows users to forward their own non-matching signals 4619 // to JVM_handle_linux_signal, harmlessly. 4620 bool os::Linux::signal_handlers_are_installed = false; 4621 4622 // For signal-chaining 4623 bool os::Linux::libjsig_is_loaded = false; 4624 typedef struct sigaction *(*get_signal_t)(int); 4625 get_signal_t os::Linux::get_signal_action = NULL; 4626 4627 struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4628 struct sigaction *actp = NULL; 4629 4630 if (libjsig_is_loaded) { 4631 // Retrieve the old signal handler from libjsig 4632 actp = (*get_signal_action)(sig); 4633 } 4634 if (actp == NULL) { 4635 // Retrieve the preinstalled signal handler from jvm 4636 actp = os::Posix::get_preinstalled_handler(sig); 4637 } 4638 4639 return actp; 4640 } 4641 4642 static bool call_chained_handler(struct sigaction *actp, int sig, 4643 siginfo_t *siginfo, void *context) { 4644 // Call the old signal handler 4645 if (actp->sa_handler == SIG_DFL) { 4646 // It's more reasonable to let jvm treat it as an unexpected exception 4647 // instead of taking the default action. 4648 return false; 4649 } else if (actp->sa_handler != SIG_IGN) { 4650 if ((actp->sa_flags & SA_NODEFER) == 0) { 4651 // automaticlly block the signal 4652 sigaddset(&(actp->sa_mask), sig); 4653 } 4654 4655 sa_handler_t hand = NULL; 4656 sa_sigaction_t sa = NULL; 4657 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4658 // retrieve the chained handler 4659 if (siginfo_flag_set) { 4660 sa = actp->sa_sigaction; 4661 } else { 4662 hand = actp->sa_handler; 4663 } 4664 4665 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4666 actp->sa_handler = SIG_DFL; 4667 } 4668 4669 // try to honor the signal mask 4670 sigset_t oset; 4671 sigemptyset(&oset); 4672 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4673 4674 // call into the chained handler 4675 if (siginfo_flag_set) { 4676 (*sa)(sig, siginfo, context); 4677 } else { 4678 (*hand)(sig); 4679 } 4680 4681 // restore the signal mask 4682 pthread_sigmask(SIG_SETMASK, &oset, NULL); 4683 } 4684 // Tell jvm's signal handler the signal is taken care of. 4685 return true; 4686 } 4687 4688 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4689 bool chained = false; 4690 // signal-chaining 4691 if (UseSignalChaining) { 4692 struct sigaction *actp = get_chained_signal_action(sig); 4693 if (actp != NULL) { 4694 chained = call_chained_handler(actp, sig, siginfo, context); 4695 } 4696 } 4697 return chained; 4698 } 4699 4700 // for diagnostic 4701 int sigflags[NSIG]; 4702 4703 int os::Linux::get_our_sigflags(int sig) { 4704 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4705 return sigflags[sig]; 4706 } 4707 4708 void os::Linux::set_our_sigflags(int sig, int flags) { 4709 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4710 if (sig > 0 && sig < NSIG) { 4711 sigflags[sig] = flags; 4712 } 4713 } 4714 4715 void os::Linux::set_signal_handler(int sig, bool set_installed) { 4716 // Check for overwrite. 4717 struct sigaction oldAct; 4718 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4719 4720 void* oldhand = oldAct.sa_sigaction 4721 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4722 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4723 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4724 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4725 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4726 if (AllowUserSignalHandlers || !set_installed) { 4727 // Do not overwrite; user takes responsibility to forward to us. 4728 return; 4729 } else if (UseSignalChaining) { 4730 // save the old handler in jvm 4731 os::Posix::save_preinstalled_handler(sig, oldAct); 4732 // libjsig also interposes the sigaction() call below and saves the 4733 // old sigaction on it own. 4734 } else { 4735 fatal("Encountered unexpected pre-existing sigaction handler " 4736 "%#lx for signal %d.", (long)oldhand, sig); 4737 } 4738 } 4739 4740 struct sigaction sigAct; 4741 sigfillset(&(sigAct.sa_mask)); 4742 sigAct.sa_handler = SIG_DFL; 4743 if (!set_installed) { 4744 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4745 } else { 4746 sigAct.sa_sigaction = signalHandler; 4747 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4748 } 4749 // Save flags, which are set by ours 4750 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4751 sigflags[sig] = sigAct.sa_flags; 4752 4753 int ret = sigaction(sig, &sigAct, &oldAct); 4754 assert(ret == 0, "check"); 4755 4756 void* oldhand2 = oldAct.sa_sigaction 4757 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4758 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4759 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4760 } 4761 4762 // install signal handlers for signals that HotSpot needs to 4763 // handle in order to support Java-level exception handling. 4764 4765 void os::Linux::install_signal_handlers() { 4766 if (!signal_handlers_are_installed) { 4767 signal_handlers_are_installed = true; 4768 4769 // signal-chaining 4770 typedef void (*signal_setting_t)(); 4771 signal_setting_t begin_signal_setting = NULL; 4772 signal_setting_t end_signal_setting = NULL; 4773 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4774 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4775 if (begin_signal_setting != NULL) { 4776 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4777 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4778 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4779 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4780 libjsig_is_loaded = true; 4781 assert(UseSignalChaining, "should enable signal-chaining"); 4782 } 4783 if (libjsig_is_loaded) { 4784 // Tell libjsig jvm is setting signal handlers 4785 (*begin_signal_setting)(); 4786 } 4787 4788 set_signal_handler(SIGSEGV, true); 4789 set_signal_handler(SIGPIPE, true); 4790 set_signal_handler(SIGBUS, true); 4791 set_signal_handler(SIGILL, true); 4792 set_signal_handler(SIGFPE, true); 4793 #if defined(PPC64) 4794 set_signal_handler(SIGTRAP, true); 4795 #endif 4796 set_signal_handler(SIGXFSZ, true); 4797 4798 if (libjsig_is_loaded) { 4799 // Tell libjsig jvm finishes setting signal handlers 4800 (*end_signal_setting)(); 4801 } 4802 4803 // We don't activate signal checker if libjsig is in place, we trust ourselves 4804 // and if UserSignalHandler is installed all bets are off. 4805 // Log that signal checking is off only if -verbose:jni is specified. 4806 if (CheckJNICalls) { 4807 if (libjsig_is_loaded) { 4808 log_debug(jni, resolve)("Info: libjsig is activated, all active signal checking is disabled"); 4809 check_signals = false; 4810 } 4811 if (AllowUserSignalHandlers) { 4812 log_debug(jni, resolve)("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4813 check_signals = false; 4814 } 4815 } 4816 } 4817 } 4818 4819 // This is the fastest way to get thread cpu time on Linux. 4820 // Returns cpu time (user+sys) for any thread, not only for current. 4821 // POSIX compliant clocks are implemented in the kernels 2.6.16+. 4822 // It might work on 2.6.10+ with a special kernel/glibc patch. 4823 // For reference, please, see IEEE Std 1003.1-2004: 4824 // http://www.unix.org/single_unix_specification 4825 4826 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4827 struct timespec tp; 4828 int rc = os::Posix::clock_gettime(clockid, &tp); 4829 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4830 4831 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4832 } 4833 4834 ///// 4835 // glibc on Linux platform uses non-documented flag 4836 // to indicate, that some special sort of signal 4837 // trampoline is used. 4838 // We will never set this flag, and we should 4839 // ignore this flag in our diagnostic 4840 #ifdef SIGNIFICANT_SIGNAL_MASK 4841 #undef SIGNIFICANT_SIGNAL_MASK 4842 #endif 4843 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4844 4845 static const char* get_signal_handler_name(address handler, 4846 char* buf, int buflen) { 4847 int offset = 0; 4848 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4849 if (found) { 4850 // skip directory names 4851 const char *p1, *p2; 4852 p1 = buf; 4853 size_t len = strlen(os::file_separator()); 4854 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4855 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4856 } else { 4857 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4858 } 4859 return buf; 4860 } 4861 4862 static void print_signal_handler(outputStream* st, int sig, 4863 char* buf, size_t buflen) { 4864 struct sigaction sa; 4865 4866 sigaction(sig, NULL, &sa); 4867 4868 // See comment for SIGNIFICANT_SIGNAL_MASK define 4869 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4870 4871 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4872 4873 address handler = (sa.sa_flags & SA_SIGINFO) 4874 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4875 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4876 4877 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4878 st->print("SIG_DFL"); 4879 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4880 st->print("SIG_IGN"); 4881 } else { 4882 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4883 } 4884 4885 st->print(", sa_mask[0]="); 4886 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4887 4888 address rh = VMError::get_resetted_sighandler(sig); 4889 // May be, handler was resetted by VMError? 4890 if (rh != NULL) { 4891 handler = rh; 4892 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4893 } 4894 4895 st->print(", sa_flags="); 4896 os::Posix::print_sa_flags(st, sa.sa_flags); 4897 4898 // Check: is it our handler? 4899 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4900 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4901 // It is our signal handler 4902 // check for flags, reset system-used one! 4903 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4904 st->print( 4905 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4906 os::Linux::get_our_sigflags(sig)); 4907 } 4908 } 4909 st->cr(); 4910 } 4911 4912 4913 #define DO_SIGNAL_CHECK(sig) \ 4914 do { \ 4915 if (!sigismember(&check_signal_done, sig)) { \ 4916 os::Linux::check_signal_handler(sig); \ 4917 } \ 4918 } while (0) 4919 4920 // This method is a periodic task to check for misbehaving JNI applications 4921 // under CheckJNI, we can add any periodic checks here 4922 4923 void os::run_periodic_checks() { 4924 if (check_signals == false) return; 4925 4926 // SEGV and BUS if overridden could potentially prevent 4927 // generation of hs*.log in the event of a crash, debugging 4928 // such a case can be very challenging, so we absolutely 4929 // check the following for a good measure: 4930 DO_SIGNAL_CHECK(SIGSEGV); 4931 DO_SIGNAL_CHECK(SIGILL); 4932 DO_SIGNAL_CHECK(SIGFPE); 4933 DO_SIGNAL_CHECK(SIGBUS); 4934 DO_SIGNAL_CHECK(SIGPIPE); 4935 DO_SIGNAL_CHECK(SIGXFSZ); 4936 #if defined(PPC64) 4937 DO_SIGNAL_CHECK(SIGTRAP); 4938 #endif 4939 4940 // ReduceSignalUsage allows the user to override these handlers 4941 // see comments at the very top and jvm_md.h 4942 if (!ReduceSignalUsage) { 4943 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4944 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4945 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4946 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4947 } 4948 4949 DO_SIGNAL_CHECK(SR_signum); 4950 } 4951 4952 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4953 4954 static os_sigaction_t os_sigaction = NULL; 4955 4956 void os::Linux::check_signal_handler(int sig) { 4957 char buf[O_BUFLEN]; 4958 address jvmHandler = NULL; 4959 4960 4961 struct sigaction act; 4962 if (os_sigaction == NULL) { 4963 // only trust the default sigaction, in case it has been interposed 4964 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4965 if (os_sigaction == NULL) return; 4966 } 4967 4968 os_sigaction(sig, (struct sigaction*)NULL, &act); 4969 4970 4971 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4972 4973 address thisHandler = (act.sa_flags & SA_SIGINFO) 4974 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4975 : CAST_FROM_FN_PTR(address, act.sa_handler); 4976 4977 4978 switch (sig) { 4979 case SIGSEGV: 4980 case SIGBUS: 4981 case SIGFPE: 4982 case SIGPIPE: 4983 case SIGILL: 4984 case SIGXFSZ: 4985 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4986 break; 4987 4988 case SHUTDOWN1_SIGNAL: 4989 case SHUTDOWN2_SIGNAL: 4990 case SHUTDOWN3_SIGNAL: 4991 case BREAK_SIGNAL: 4992 jvmHandler = (address)user_handler(); 4993 break; 4994 4995 default: 4996 if (sig == SR_signum) { 4997 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4998 } else { 4999 return; 5000 } 5001 break; 5002 } 5003 5004 if (thisHandler != jvmHandler) { 5005 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 5006 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 5007 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 5008 // No need to check this sig any longer 5009 sigaddset(&check_signal_done, sig); 5010 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 5011 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 5012 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 5013 exception_name(sig, buf, O_BUFLEN)); 5014 } 5015 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 5016 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 5017 tty->print("expected:"); 5018 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig)); 5019 tty->cr(); 5020 tty->print(" found:"); 5021 os::Posix::print_sa_flags(tty, act.sa_flags); 5022 tty->cr(); 5023 // No need to check this sig any longer 5024 sigaddset(&check_signal_done, sig); 5025 } 5026 5027 // Dump all the signal 5028 if (sigismember(&check_signal_done, sig)) { 5029 print_signal_handlers(tty, buf, O_BUFLEN); 5030 } 5031 } 5032 5033 extern void report_error(char* file_name, int line_no, char* title, 5034 char* format, ...); 5035 5036 // this is called _before_ most of the global arguments have been parsed 5037 void os::init(void) { 5038 char dummy; // used to get a guess on initial stack address 5039 5040 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 5041 5042 init_random(1234567); 5043 5044 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 5045 if (Linux::page_size() == -1) { 5046 fatal("os_linux.cpp: os::init: sysconf failed (%s)", 5047 os::strerror(errno)); 5048 } 5049 init_page_sizes((size_t) Linux::page_size()); 5050 5051 Linux::initialize_system_info(); 5052 5053 os::Linux::CPUPerfTicks pticks; 5054 bool res = os::Linux::get_tick_information(&pticks, -1); 5055 5056 if (res && pticks.has_steal_ticks) { 5057 has_initial_tick_info = true; 5058 initial_total_ticks = pticks.total; 5059 initial_steal_ticks = pticks.steal; 5060 } 5061 5062 // _main_thread points to the thread that created/loaded the JVM. 5063 Linux::_main_thread = pthread_self(); 5064 5065 // retrieve entry point for pthread_setname_np 5066 Linux::_pthread_setname_np = 5067 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 5068 5069 os::Posix::init(); 5070 5071 initial_time_count = javaTimeNanos(); 5072 5073 // Always warn if no monotonic clock available 5074 if (!os::Posix::supports_monotonic_clock()) { 5075 warning("No monotonic clock was available - timed services may " \ 5076 "be adversely affected if the time-of-day clock changes"); 5077 } 5078 } 5079 5080 // To install functions for atexit system call 5081 extern "C" { 5082 static void perfMemory_exit_helper() { 5083 perfMemory_exit(); 5084 } 5085 } 5086 5087 void os::pd_init_container_support() { 5088 OSContainer::init(); 5089 } 5090 5091 void os::Linux::numa_init() { 5092 5093 // Java can be invoked as 5094 // 1. Without numactl and heap will be allocated/configured on all nodes as 5095 // per the system policy. 5096 // 2. With numactl --interleave: 5097 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same 5098 // API for membind case bitmask is reset. 5099 // Interleave is only hint and Kernel can fallback to other nodes if 5100 // no memory is available on the target nodes. 5101 // 3. With numactl --membind: 5102 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for 5103 // interleave case returns bitmask of all nodes. 5104 // numa_all_nodes_ptr holds bitmask of all nodes. 5105 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct 5106 // bitmask when externally configured to run on all or fewer nodes. 5107 5108 if (!Linux::libnuma_init()) { 5109 UseNUMA = false; 5110 } else { 5111 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) { 5112 // If there's only one node (they start from 0) or if the process 5113 // is bound explicitly to a single node using membind, disable NUMA. 5114 UseNUMA = false; 5115 } else { 5116 5117 LogTarget(Info,os) log; 5118 LogStream ls(log); 5119 5120 Linux::set_configured_numa_policy(Linux::identify_numa_policy()); 5121 5122 struct bitmask* bmp = Linux::_numa_membind_bitmask; 5123 const char* numa_mode = "membind"; 5124 5125 if (Linux::is_running_in_interleave_mode()) { 5126 bmp = Linux::_numa_interleave_bitmask; 5127 numa_mode = "interleave"; 5128 } 5129 5130 ls.print("UseNUMA is enabled and invoked in '%s' mode." 5131 " Heap will be configured using NUMA memory nodes:", numa_mode); 5132 5133 for (int node = 0; node <= Linux::numa_max_node(); node++) { 5134 if (Linux::_numa_bitmask_isbitset(bmp, node)) { 5135 ls.print(" %d", node); 5136 } 5137 } 5138 } 5139 } 5140 5141 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 5142 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 5143 // we can make the adaptive lgrp chunk resizing work. If the user specified both 5144 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn 5145 // and disable adaptive resizing. 5146 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) { 5147 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, " 5148 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)"); 5149 UseAdaptiveSizePolicy = false; 5150 UseAdaptiveNUMAChunkSizing = false; 5151 } 5152 } 5153 5154 if (!UseNUMA && ForceNUMA) { 5155 UseNUMA = true; 5156 } 5157 } 5158 5159 // this is called _after_ the global arguments have been parsed 5160 jint os::init_2(void) { 5161 5162 // This could be set after os::Posix::init() but all platforms 5163 // have to set it the same so we have to mirror Solaris. 5164 DEBUG_ONLY(os::set_mutex_init_done();) 5165 5166 os::Posix::init_2(); 5167 5168 Linux::fast_thread_clock_init(); 5169 5170 // initialize suspend/resume support - must do this before signal_sets_init() 5171 if (SR_initialize() != 0) { 5172 perror("SR_initialize failed"); 5173 return JNI_ERR; 5174 } 5175 5176 Linux::signal_sets_init(); 5177 Linux::install_signal_handlers(); 5178 // Initialize data for jdk.internal.misc.Signal 5179 if (!ReduceSignalUsage) { 5180 jdk_misc_signal_init(); 5181 } 5182 5183 if (AdjustStackSizeForTLS) { 5184 get_minstack_init(); 5185 } 5186 5187 // Check and sets minimum stack sizes against command line options 5188 if (Posix::set_minimum_stack_sizes() == JNI_ERR) { 5189 return JNI_ERR; 5190 } 5191 5192 #if defined(IA32) 5193 // Need to ensure we've determined the process's initial stack to 5194 // perform the workaround 5195 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 5196 workaround_expand_exec_shield_cs_limit(); 5197 #else 5198 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher(); 5199 if (!suppress_primordial_thread_resolution) { 5200 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 5201 } 5202 #endif 5203 5204 Linux::libpthread_init(); 5205 Linux::sched_getcpu_init(); 5206 log_info(os)("HotSpot is running with %s, %s", 5207 Linux::glibc_version(), Linux::libpthread_version()); 5208 5209 if (UseNUMA) { 5210 Linux::numa_init(); 5211 } 5212 5213 if (MaxFDLimit) { 5214 // set the number of file descriptors to max. print out error 5215 // if getrlimit/setrlimit fails but continue regardless. 5216 struct rlimit nbr_files; 5217 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 5218 if (status != 0) { 5219 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno)); 5220 } else { 5221 nbr_files.rlim_cur = nbr_files.rlim_max; 5222 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 5223 if (status != 0) { 5224 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno)); 5225 } 5226 } 5227 } 5228 5229 // at-exit methods are called in the reverse order of their registration. 5230 // atexit functions are called on return from main or as a result of a 5231 // call to exit(3C). There can be only 32 of these functions registered 5232 // and atexit() does not set errno. 5233 5234 if (PerfAllowAtExitRegistration) { 5235 // only register atexit functions if PerfAllowAtExitRegistration is set. 5236 // atexit functions can be delayed until process exit time, which 5237 // can be problematic for embedded VM situations. Embedded VMs should 5238 // call DestroyJavaVM() to assure that VM resources are released. 5239 5240 // note: perfMemory_exit_helper atexit function may be removed in 5241 // the future if the appropriate cleanup code can be added to the 5242 // VM_Exit VMOperation's doit method. 5243 if (atexit(perfMemory_exit_helper) != 0) { 5244 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 5245 } 5246 } 5247 5248 // initialize thread priority policy 5249 prio_init(); 5250 5251 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) { 5252 set_coredump_filter(DAX_SHARED_BIT); 5253 } 5254 5255 if (DumpPrivateMappingsInCore) { 5256 set_coredump_filter(FILE_BACKED_PVT_BIT); 5257 } 5258 5259 if (DumpSharedMappingsInCore) { 5260 set_coredump_filter(FILE_BACKED_SHARED_BIT); 5261 } 5262 5263 return JNI_OK; 5264 } 5265 5266 // Mark the polling page as unreadable 5267 void os::make_polling_page_unreadable(void) { 5268 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 5269 fatal("Could not disable polling page"); 5270 } 5271 } 5272 5273 // Mark the polling page as readable 5274 void os::make_polling_page_readable(void) { 5275 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 5276 fatal("Could not enable polling page"); 5277 } 5278 } 5279 5280 // older glibc versions don't have this macro (which expands to 5281 // an optimized bit-counting function) so we have to roll our own 5282 #ifndef CPU_COUNT 5283 5284 static int _cpu_count(const cpu_set_t* cpus) { 5285 int count = 0; 5286 // only look up to the number of configured processors 5287 for (int i = 0; i < os::processor_count(); i++) { 5288 if (CPU_ISSET(i, cpus)) { 5289 count++; 5290 } 5291 } 5292 return count; 5293 } 5294 5295 #define CPU_COUNT(cpus) _cpu_count(cpus) 5296 5297 #endif // CPU_COUNT 5298 5299 // Get the current number of available processors for this process. 5300 // This value can change at any time during a process's lifetime. 5301 // sched_getaffinity gives an accurate answer as it accounts for cpusets. 5302 // If it appears there may be more than 1024 processors then we do a 5303 // dynamic check - see 6515172 for details. 5304 // If anything goes wrong we fallback to returning the number of online 5305 // processors - which can be greater than the number available to the process. 5306 int os::Linux::active_processor_count() { 5307 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors 5308 cpu_set_t* cpus_p = &cpus; 5309 int cpus_size = sizeof(cpu_set_t); 5310 5311 int configured_cpus = os::processor_count(); // upper bound on available cpus 5312 int cpu_count = 0; 5313 5314 // old build platforms may not support dynamic cpu sets 5315 #ifdef CPU_ALLOC 5316 5317 // To enable easy testing of the dynamic path on different platforms we 5318 // introduce a diagnostic flag: UseCpuAllocPath 5319 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) { 5320 // kernel may use a mask bigger than cpu_set_t 5321 log_trace(os)("active_processor_count: using dynamic path %s" 5322 "- configured processors: %d", 5323 UseCpuAllocPath ? "(forced) " : "", 5324 configured_cpus); 5325 cpus_p = CPU_ALLOC(configured_cpus); 5326 if (cpus_p != NULL) { 5327 cpus_size = CPU_ALLOC_SIZE(configured_cpus); 5328 // zero it just to be safe 5329 CPU_ZERO_S(cpus_size, cpus_p); 5330 } 5331 else { 5332 // failed to allocate so fallback to online cpus 5333 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 5334 log_trace(os)("active_processor_count: " 5335 "CPU_ALLOC failed (%s) - using " 5336 "online processor count: %d", 5337 os::strerror(errno), online_cpus); 5338 return online_cpus; 5339 } 5340 } 5341 else { 5342 log_trace(os)("active_processor_count: using static path - configured processors: %d", 5343 configured_cpus); 5344 } 5345 #else // CPU_ALLOC 5346 // these stubs won't be executed 5347 #define CPU_COUNT_S(size, cpus) -1 5348 #define CPU_FREE(cpus) 5349 5350 log_trace(os)("active_processor_count: only static path available - configured processors: %d", 5351 configured_cpus); 5352 #endif // CPU_ALLOC 5353 5354 // pid 0 means the current thread - which we have to assume represents the process 5355 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) { 5356 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5357 cpu_count = CPU_COUNT_S(cpus_size, cpus_p); 5358 } 5359 else { 5360 cpu_count = CPU_COUNT(cpus_p); 5361 } 5362 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count); 5363 } 5364 else { 5365 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN); 5366 warning("sched_getaffinity failed (%s)- using online processor count (%d) " 5367 "which may exceed available processors", os::strerror(errno), cpu_count); 5368 } 5369 5370 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5371 CPU_FREE(cpus_p); 5372 } 5373 5374 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check"); 5375 return cpu_count; 5376 } 5377 5378 // Determine the active processor count from one of 5379 // three different sources: 5380 // 5381 // 1. User option -XX:ActiveProcessorCount 5382 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN) 5383 // 3. extracted from cgroup cpu subsystem (shares and quotas) 5384 // 5385 // Option 1, if specified, will always override. 5386 // If the cgroup subsystem is active and configured, we 5387 // will return the min of the cgroup and option 2 results. 5388 // This is required since tools, such as numactl, that 5389 // alter cpu affinity do not update cgroup subsystem 5390 // cpuset configuration files. 5391 int os::active_processor_count() { 5392 // User has overridden the number of active processors 5393 if (ActiveProcessorCount > 0) { 5394 log_trace(os)("active_processor_count: " 5395 "active processor count set by user : %d", 5396 ActiveProcessorCount); 5397 return ActiveProcessorCount; 5398 } 5399 5400 int active_cpus; 5401 if (OSContainer::is_containerized()) { 5402 active_cpus = OSContainer::active_processor_count(); 5403 log_trace(os)("active_processor_count: determined by OSContainer: %d", 5404 active_cpus); 5405 } else { 5406 active_cpus = os::Linux::active_processor_count(); 5407 } 5408 5409 return active_cpus; 5410 } 5411 5412 uint os::processor_id() { 5413 const int id = Linux::sched_getcpu(); 5414 assert(id >= 0 && id < _processor_count, "Invalid processor id"); 5415 return (uint)id; 5416 } 5417 5418 void os::set_native_thread_name(const char *name) { 5419 if (Linux::_pthread_setname_np) { 5420 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 5421 snprintf(buf, sizeof(buf), "%s", name); 5422 buf[sizeof(buf) - 1] = '\0'; 5423 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 5424 // ERANGE should not happen; all other errors should just be ignored. 5425 assert(rc != ERANGE, "pthread_setname_np failed"); 5426 } 5427 } 5428 5429 bool os::bind_to_processor(uint processor_id) { 5430 // Not yet implemented. 5431 return false; 5432 } 5433 5434 /// 5435 5436 void os::SuspendedThreadTask::internal_do_task() { 5437 if (do_suspend(_thread->osthread())) { 5438 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 5439 do_task(context); 5440 do_resume(_thread->osthread()); 5441 } 5442 } 5443 5444 //////////////////////////////////////////////////////////////////////////////// 5445 // debug support 5446 5447 bool os::find(address addr, outputStream* st) { 5448 Dl_info dlinfo; 5449 memset(&dlinfo, 0, sizeof(dlinfo)); 5450 if (dladdr(addr, &dlinfo) != 0) { 5451 st->print(PTR_FORMAT ": ", p2i(addr)); 5452 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5453 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname, 5454 p2i(addr) - p2i(dlinfo.dli_saddr)); 5455 } else if (dlinfo.dli_fbase != NULL) { 5456 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase)); 5457 } else { 5458 st->print("<absolute address>"); 5459 } 5460 if (dlinfo.dli_fname != NULL) { 5461 st->print(" in %s", dlinfo.dli_fname); 5462 } 5463 if (dlinfo.dli_fbase != NULL) { 5464 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase)); 5465 } 5466 st->cr(); 5467 5468 if (Verbose) { 5469 // decode some bytes around the PC 5470 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5471 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5472 address lowest = (address) dlinfo.dli_sname; 5473 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5474 if (begin < lowest) begin = lowest; 5475 Dl_info dlinfo2; 5476 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5477 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 5478 end = (address) dlinfo2.dli_saddr; 5479 } 5480 Disassembler::decode(begin, end, st); 5481 } 5482 return true; 5483 } 5484 return false; 5485 } 5486 5487 //////////////////////////////////////////////////////////////////////////////// 5488 // misc 5489 5490 // This does not do anything on Linux. This is basically a hook for being 5491 // able to use structured exception handling (thread-local exception filters) 5492 // on, e.g., Win32. 5493 void 5494 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method, 5495 JavaCallArguments* args, Thread* thread) { 5496 f(value, method, args, thread); 5497 } 5498 5499 void os::print_statistics() { 5500 } 5501 5502 bool os::message_box(const char* title, const char* message) { 5503 int i; 5504 fdStream err(defaultStream::error_fd()); 5505 for (i = 0; i < 78; i++) err.print_raw("="); 5506 err.cr(); 5507 err.print_raw_cr(title); 5508 for (i = 0; i < 78; i++) err.print_raw("-"); 5509 err.cr(); 5510 err.print_raw_cr(message); 5511 for (i = 0; i < 78; i++) err.print_raw("="); 5512 err.cr(); 5513 5514 char buf[16]; 5515 // Prevent process from exiting upon "read error" without consuming all CPU 5516 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5517 5518 return buf[0] == 'y' || buf[0] == 'Y'; 5519 } 5520 5521 // Is a (classpath) directory empty? 5522 bool os::dir_is_empty(const char* path) { 5523 DIR *dir = NULL; 5524 struct dirent *ptr; 5525 5526 dir = opendir(path); 5527 if (dir == NULL) return true; 5528 5529 // Scan the directory 5530 bool result = true; 5531 while (result && (ptr = readdir(dir)) != NULL) { 5532 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5533 result = false; 5534 } 5535 } 5536 closedir(dir); 5537 return result; 5538 } 5539 5540 // This code originates from JDK's sysOpen and open64_w 5541 // from src/solaris/hpi/src/system_md.c 5542 5543 int os::open(const char *path, int oflag, int mode) { 5544 if (strlen(path) > MAX_PATH - 1) { 5545 errno = ENAMETOOLONG; 5546 return -1; 5547 } 5548 5549 // All file descriptors that are opened in the Java process and not 5550 // specifically destined for a subprocess should have the close-on-exec 5551 // flag set. If we don't set it, then careless 3rd party native code 5552 // might fork and exec without closing all appropriate file descriptors 5553 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5554 // turn might: 5555 // 5556 // - cause end-of-file to fail to be detected on some file 5557 // descriptors, resulting in mysterious hangs, or 5558 // 5559 // - might cause an fopen in the subprocess to fail on a system 5560 // suffering from bug 1085341. 5561 // 5562 // (Yes, the default setting of the close-on-exec flag is a Unix 5563 // design flaw) 5564 // 5565 // See: 5566 // 1085341: 32-bit stdio routines should support file descriptors >255 5567 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5568 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5569 // 5570 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5571 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5572 // because it saves a system call and removes a small window where the flag 5573 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5574 // and we fall back to using FD_CLOEXEC (see below). 5575 #ifdef O_CLOEXEC 5576 oflag |= O_CLOEXEC; 5577 #endif 5578 5579 int fd = ::open64(path, oflag, mode); 5580 if (fd == -1) return -1; 5581 5582 //If the open succeeded, the file might still be a directory 5583 { 5584 struct stat64 buf64; 5585 int ret = ::fstat64(fd, &buf64); 5586 int st_mode = buf64.st_mode; 5587 5588 if (ret != -1) { 5589 if ((st_mode & S_IFMT) == S_IFDIR) { 5590 errno = EISDIR; 5591 ::close(fd); 5592 return -1; 5593 } 5594 } else { 5595 ::close(fd); 5596 return -1; 5597 } 5598 } 5599 5600 #ifdef FD_CLOEXEC 5601 // Validate that the use of the O_CLOEXEC flag on open above worked. 5602 // With recent kernels, we will perform this check exactly once. 5603 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5604 if (!O_CLOEXEC_is_known_to_work) { 5605 int flags = ::fcntl(fd, F_GETFD); 5606 if (flags != -1) { 5607 if ((flags & FD_CLOEXEC) != 0) 5608 O_CLOEXEC_is_known_to_work = 1; 5609 else 5610 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5611 } 5612 } 5613 #endif 5614 5615 return fd; 5616 } 5617 5618 5619 // create binary file, rewriting existing file if required 5620 int os::create_binary_file(const char* path, bool rewrite_existing) { 5621 int oflags = O_WRONLY | O_CREAT; 5622 if (!rewrite_existing) { 5623 oflags |= O_EXCL; 5624 } 5625 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5626 } 5627 5628 // return current position of file pointer 5629 jlong os::current_file_offset(int fd) { 5630 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5631 } 5632 5633 // move file pointer to the specified offset 5634 jlong os::seek_to_file_offset(int fd, jlong offset) { 5635 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5636 } 5637 5638 // This code originates from JDK's sysAvailable 5639 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5640 5641 int os::available(int fd, jlong *bytes) { 5642 jlong cur, end; 5643 int mode; 5644 struct stat64 buf64; 5645 5646 if (::fstat64(fd, &buf64) >= 0) { 5647 mode = buf64.st_mode; 5648 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5649 int n; 5650 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5651 *bytes = n; 5652 return 1; 5653 } 5654 } 5655 } 5656 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5657 return 0; 5658 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5659 return 0; 5660 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5661 return 0; 5662 } 5663 *bytes = end - cur; 5664 return 1; 5665 } 5666 5667 // Map a block of memory. 5668 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5669 char *addr, size_t bytes, bool read_only, 5670 bool allow_exec) { 5671 int prot; 5672 int flags = MAP_PRIVATE; 5673 5674 if (read_only) { 5675 prot = PROT_READ; 5676 } else { 5677 prot = PROT_READ | PROT_WRITE; 5678 } 5679 5680 if (allow_exec) { 5681 prot |= PROT_EXEC; 5682 } 5683 5684 if (addr != NULL) { 5685 flags |= MAP_FIXED; 5686 } 5687 5688 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5689 fd, file_offset); 5690 if (mapped_address == MAP_FAILED) { 5691 return NULL; 5692 } 5693 return mapped_address; 5694 } 5695 5696 5697 // Remap a block of memory. 5698 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5699 char *addr, size_t bytes, bool read_only, 5700 bool allow_exec) { 5701 // same as map_memory() on this OS 5702 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5703 allow_exec); 5704 } 5705 5706 5707 // Unmap a block of memory. 5708 bool os::pd_unmap_memory(char* addr, size_t bytes) { 5709 return munmap(addr, bytes) == 0; 5710 } 5711 5712 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5713 5714 static jlong fast_cpu_time(Thread *thread) { 5715 clockid_t clockid; 5716 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(), 5717 &clockid); 5718 if (rc == 0) { 5719 return os::Linux::fast_thread_cpu_time(clockid); 5720 } else { 5721 // It's possible to encounter a terminated native thread that failed 5722 // to detach itself from the VM - which should result in ESRCH. 5723 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed"); 5724 return -1; 5725 } 5726 } 5727 5728 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5729 // are used by JVM M&M and JVMTI to get user+sys or user CPU time 5730 // of a thread. 5731 // 5732 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5733 // the fast estimate available on the platform. 5734 5735 jlong os::current_thread_cpu_time() { 5736 if (os::Linux::supports_fast_thread_cpu_time()) { 5737 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5738 } else { 5739 // return user + sys since the cost is the same 5740 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5741 } 5742 } 5743 5744 jlong os::thread_cpu_time(Thread* thread) { 5745 // consistent with what current_thread_cpu_time() returns 5746 if (os::Linux::supports_fast_thread_cpu_time()) { 5747 return fast_cpu_time(thread); 5748 } else { 5749 return slow_thread_cpu_time(thread, true /* user + sys */); 5750 } 5751 } 5752 5753 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5754 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5755 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5756 } else { 5757 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5758 } 5759 } 5760 5761 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5762 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5763 return fast_cpu_time(thread); 5764 } else { 5765 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5766 } 5767 } 5768 5769 // -1 on error. 5770 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5771 pid_t tid = thread->osthread()->thread_id(); 5772 char *s; 5773 char stat[2048]; 5774 int statlen; 5775 char proc_name[64]; 5776 int count; 5777 long sys_time, user_time; 5778 char cdummy; 5779 int idummy; 5780 long ldummy; 5781 FILE *fp; 5782 5783 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5784 fp = fopen(proc_name, "r"); 5785 if (fp == NULL) return -1; 5786 statlen = fread(stat, 1, 2047, fp); 5787 stat[statlen] = '\0'; 5788 fclose(fp); 5789 5790 // Skip pid and the command string. Note that we could be dealing with 5791 // weird command names, e.g. user could decide to rename java launcher 5792 // to "java 1.4.2 :)", then the stat file would look like 5793 // 1234 (java 1.4.2 :)) R ... ... 5794 // We don't really need to know the command string, just find the last 5795 // occurrence of ")" and then start parsing from there. See bug 4726580. 5796 s = strrchr(stat, ')'); 5797 if (s == NULL) return -1; 5798 5799 // Skip blank chars 5800 do { s++; } while (s && isspace(*s)); 5801 5802 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5803 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5804 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5805 &user_time, &sys_time); 5806 if (count != 13) return -1; 5807 if (user_sys_cpu_time) { 5808 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5809 } else { 5810 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5811 } 5812 } 5813 5814 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5815 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5816 info_ptr->may_skip_backward = false; // elapsed time not wall time 5817 info_ptr->may_skip_forward = false; // elapsed time not wall time 5818 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5819 } 5820 5821 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5822 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5823 info_ptr->may_skip_backward = false; // elapsed time not wall time 5824 info_ptr->may_skip_forward = false; // elapsed time not wall time 5825 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5826 } 5827 5828 bool os::is_thread_cpu_time_supported() { 5829 return true; 5830 } 5831 5832 // System loadavg support. Returns -1 if load average cannot be obtained. 5833 // Linux doesn't yet have a (official) notion of processor sets, 5834 // so just return the system wide load average. 5835 int os::loadavg(double loadavg[], int nelem) { 5836 return ::getloadavg(loadavg, nelem); 5837 } 5838 5839 void os::pause() { 5840 char filename[MAX_PATH]; 5841 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5842 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile); 5843 } else { 5844 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5845 } 5846 5847 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5848 if (fd != -1) { 5849 struct stat buf; 5850 ::close(fd); 5851 while (::stat(filename, &buf) == 0) { 5852 (void)::poll(NULL, 0, 100); 5853 } 5854 } else { 5855 jio_fprintf(stderr, 5856 "Could not open pause file '%s', continuing immediately.\n", filename); 5857 } 5858 } 5859 5860 extern char** environ; 5861 5862 // Run the specified command in a separate process. Return its exit value, 5863 // or -1 on failure (e.g. can't fork a new process). 5864 // Unlike system(), this function can be called from signal handler. It 5865 // doesn't block SIGINT et al. 5866 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) { 5867 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5868 5869 pid_t pid ; 5870 5871 if (use_vfork_if_available) { 5872 pid = vfork(); 5873 } else { 5874 pid = fork(); 5875 } 5876 5877 if (pid < 0) { 5878 // fork failed 5879 return -1; 5880 5881 } else if (pid == 0) { 5882 // child process 5883 5884 execve("/bin/sh", (char* const*)argv, environ); 5885 5886 // execve failed 5887 _exit(-1); 5888 5889 } else { 5890 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5891 // care about the actual exit code, for now. 5892 5893 int status; 5894 5895 // Wait for the child process to exit. This returns immediately if 5896 // the child has already exited. */ 5897 while (waitpid(pid, &status, 0) < 0) { 5898 switch (errno) { 5899 case ECHILD: return 0; 5900 case EINTR: break; 5901 default: return -1; 5902 } 5903 } 5904 5905 if (WIFEXITED(status)) { 5906 // The child exited normally; get its exit code. 5907 return WEXITSTATUS(status); 5908 } else if (WIFSIGNALED(status)) { 5909 // The child exited because of a signal 5910 // The best value to return is 0x80 + signal number, 5911 // because that is what all Unix shells do, and because 5912 // it allows callers to distinguish between process exit and 5913 // process death by signal. 5914 return 0x80 + WTERMSIG(status); 5915 } else { 5916 // Unknown exit code; pass it through 5917 return status; 5918 } 5919 } 5920 } 5921 5922 // Get the default path to the core file 5923 // Returns the length of the string 5924 int os::get_core_path(char* buffer, size_t bufferSize) { 5925 /* 5926 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 5927 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 5928 */ 5929 const int core_pattern_len = 129; 5930 char core_pattern[core_pattern_len] = {0}; 5931 5932 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 5933 if (core_pattern_file == -1) { 5934 return -1; 5935 } 5936 5937 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 5938 ::close(core_pattern_file); 5939 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') { 5940 return -1; 5941 } 5942 if (core_pattern[ret-1] == '\n') { 5943 core_pattern[ret-1] = '\0'; 5944 } else { 5945 core_pattern[ret] = '\0'; 5946 } 5947 5948 // Replace the %p in the core pattern with the process id. NOTE: we do this 5949 // only if the pattern doesn't start with "|", and we support only one %p in 5950 // the pattern. 5951 char *pid_pos = strstr(core_pattern, "%p"); 5952 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p" 5953 int written; 5954 5955 if (core_pattern[0] == '/') { 5956 if (pid_pos != NULL) { 5957 *pid_pos = '\0'; 5958 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern, 5959 current_process_id(), tail); 5960 } else { 5961 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern); 5962 } 5963 } else { 5964 char cwd[PATH_MAX]; 5965 5966 const char* p = get_current_directory(cwd, PATH_MAX); 5967 if (p == NULL) { 5968 return -1; 5969 } 5970 5971 if (core_pattern[0] == '|') { 5972 written = jio_snprintf(buffer, bufferSize, 5973 "\"%s\" (or dumping to %s/core.%d)", 5974 &core_pattern[1], p, current_process_id()); 5975 } else if (pid_pos != NULL) { 5976 *pid_pos = '\0'; 5977 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern, 5978 current_process_id(), tail); 5979 } else { 5980 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 5981 } 5982 } 5983 5984 if (written < 0) { 5985 return -1; 5986 } 5987 5988 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 5989 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 5990 5991 if (core_uses_pid_file != -1) { 5992 char core_uses_pid = 0; 5993 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 5994 ::close(core_uses_pid_file); 5995 5996 if (core_uses_pid == '1') { 5997 jio_snprintf(buffer + written, bufferSize - written, 5998 ".%d", current_process_id()); 5999 } 6000 } 6001 } 6002 6003 return strlen(buffer); 6004 } 6005 6006 bool os::start_debugging(char *buf, int buflen) { 6007 int len = (int)strlen(buf); 6008 char *p = &buf[len]; 6009 6010 jio_snprintf(p, buflen-len, 6011 "\n\n" 6012 "Do you want to debug the problem?\n\n" 6013 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n" 6014 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n" 6015 "Otherwise, press RETURN to abort...", 6016 os::current_process_id(), os::current_process_id(), 6017 os::current_thread_id(), os::current_thread_id()); 6018 6019 bool yes = os::message_box("Unexpected Error", buf); 6020 6021 if (yes) { 6022 // yes, user asked VM to launch debugger 6023 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d", 6024 os::current_process_id(), os::current_process_id()); 6025 6026 os::fork_and_exec(buf); 6027 yes = false; 6028 } 6029 return yes; 6030 } 6031 6032 6033 // Java/Compiler thread: 6034 // 6035 // Low memory addresses 6036 // P0 +------------------------+ 6037 // | |\ Java thread created by VM does not have glibc 6038 // | glibc guard page | - guard page, attached Java thread usually has 6039 // | |/ 1 glibc guard page. 6040 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 6041 // | |\ 6042 // | HotSpot Guard Pages | - red, yellow and reserved pages 6043 // | |/ 6044 // +------------------------+ JavaThread::stack_reserved_zone_base() 6045 // | |\ 6046 // | Normal Stack | - 6047 // | |/ 6048 // P2 +------------------------+ Thread::stack_base() 6049 // 6050 // Non-Java thread: 6051 // 6052 // Low memory addresses 6053 // P0 +------------------------+ 6054 // | |\ 6055 // | glibc guard page | - usually 1 page 6056 // | |/ 6057 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 6058 // | |\ 6059 // | Normal Stack | - 6060 // | |/ 6061 // P2 +------------------------+ Thread::stack_base() 6062 // 6063 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size 6064 // returned from pthread_attr_getstack(). 6065 // ** Due to NPTL implementation error, linux takes the glibc guard page out 6066 // of the stack size given in pthread_attr. We work around this for 6067 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.) 6068 // 6069 #ifndef ZERO 6070 static void current_stack_region(address * bottom, size_t * size) { 6071 if (os::is_primordial_thread()) { 6072 // primordial thread needs special handling because pthread_getattr_np() 6073 // may return bogus value. 6074 *bottom = os::Linux::initial_thread_stack_bottom(); 6075 *size = os::Linux::initial_thread_stack_size(); 6076 } else { 6077 pthread_attr_t attr; 6078 6079 int rslt = pthread_getattr_np(pthread_self(), &attr); 6080 6081 // JVM needs to know exact stack location, abort if it fails 6082 if (rslt != 0) { 6083 if (rslt == ENOMEM) { 6084 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np"); 6085 } else { 6086 fatal("pthread_getattr_np failed with error = %d", rslt); 6087 } 6088 } 6089 6090 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) { 6091 fatal("Cannot locate current stack attributes!"); 6092 } 6093 6094 // Work around NPTL stack guard error. 6095 size_t guard_size = 0; 6096 rslt = pthread_attr_getguardsize(&attr, &guard_size); 6097 if (rslt != 0) { 6098 fatal("pthread_attr_getguardsize failed with error = %d", rslt); 6099 } 6100 *bottom += guard_size; 6101 *size -= guard_size; 6102 6103 pthread_attr_destroy(&attr); 6104 6105 } 6106 assert(os::current_stack_pointer() >= *bottom && 6107 os::current_stack_pointer() < *bottom + *size, "just checking"); 6108 } 6109 6110 address os::current_stack_base() { 6111 address bottom; 6112 size_t size; 6113 current_stack_region(&bottom, &size); 6114 return (bottom + size); 6115 } 6116 6117 size_t os::current_stack_size() { 6118 // This stack size includes the usable stack and HotSpot guard pages 6119 // (for the threads that have Hotspot guard pages). 6120 address bottom; 6121 size_t size; 6122 current_stack_region(&bottom, &size); 6123 return size; 6124 } 6125 #endif 6126 6127 static inline struct timespec get_mtime(const char* filename) { 6128 struct stat st; 6129 int ret = os::stat(filename, &st); 6130 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno)); 6131 return st.st_mtim; 6132 } 6133 6134 int os::compare_file_modified_times(const char* file1, const char* file2) { 6135 struct timespec filetime1 = get_mtime(file1); 6136 struct timespec filetime2 = get_mtime(file2); 6137 int diff = filetime1.tv_sec - filetime2.tv_sec; 6138 if (diff == 0) { 6139 return filetime1.tv_nsec - filetime2.tv_nsec; 6140 } 6141 return diff; 6142 } 6143 6144 bool os::supports_map_sync() { 6145 return true; 6146 } 6147 6148 /////////////// Unit tests /////////////// 6149 6150 #ifndef PRODUCT 6151 6152 class TestReserveMemorySpecial : AllStatic { 6153 public: 6154 static void small_page_write(void* addr, size_t size) { 6155 size_t page_size = os::vm_page_size(); 6156 6157 char* end = (char*)addr + size; 6158 for (char* p = (char*)addr; p < end; p += page_size) { 6159 *p = 1; 6160 } 6161 } 6162 6163 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6164 if (!UseHugeTLBFS) { 6165 return; 6166 } 6167 6168 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6169 6170 if (addr != NULL) { 6171 small_page_write(addr, size); 6172 6173 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6174 } 6175 } 6176 6177 static void test_reserve_memory_special_huge_tlbfs_only() { 6178 if (!UseHugeTLBFS) { 6179 return; 6180 } 6181 6182 size_t lp = os::large_page_size(); 6183 6184 for (size_t size = lp; size <= lp * 10; size += lp) { 6185 test_reserve_memory_special_huge_tlbfs_only(size); 6186 } 6187 } 6188 6189 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6190 size_t lp = os::large_page_size(); 6191 size_t ag = os::vm_allocation_granularity(); 6192 6193 // sizes to test 6194 const size_t sizes[] = { 6195 lp, lp + ag, lp + lp / 2, lp * 2, 6196 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2, 6197 lp * 10, lp * 10 + lp / 2 6198 }; 6199 const int num_sizes = sizeof(sizes) / sizeof(size_t); 6200 6201 // For each size/alignment combination, we test three scenarios: 6202 // 1) with req_addr == NULL 6203 // 2) with a non-null req_addr at which we expect to successfully allocate 6204 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we 6205 // expect the allocation to either fail or to ignore req_addr 6206 6207 // Pre-allocate two areas; they shall be as large as the largest allocation 6208 // and aligned to the largest alignment we will be testing. 6209 const size_t mapping_size = sizes[num_sizes - 1] * 2; 6210 char* const mapping1 = (char*) ::mmap(NULL, mapping_size, 6211 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6212 -1, 0); 6213 assert(mapping1 != MAP_FAILED, "should work"); 6214 6215 char* const mapping2 = (char*) ::mmap(NULL, mapping_size, 6216 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6217 -1, 0); 6218 assert(mapping2 != MAP_FAILED, "should work"); 6219 6220 // Unmap the first mapping, but leave the second mapping intact: the first 6221 // mapping will serve as a value for a "good" req_addr (case 2). The second 6222 // mapping, still intact, as "bad" req_addr (case 3). 6223 ::munmap(mapping1, mapping_size); 6224 6225 // Case 1 6226 for (int i = 0; i < num_sizes; i++) { 6227 const size_t size = sizes[i]; 6228 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6229 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6230 if (p != NULL) { 6231 assert(is_aligned(p, alignment), "must be"); 6232 small_page_write(p, size); 6233 os::Linux::release_memory_special_huge_tlbfs(p, size); 6234 } 6235 } 6236 } 6237 6238 // Case 2 6239 for (int i = 0; i < num_sizes; i++) { 6240 const size_t size = sizes[i]; 6241 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6242 char* const req_addr = align_up(mapping1, alignment); 6243 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6244 if (p != NULL) { 6245 assert(p == req_addr, "must be"); 6246 small_page_write(p, size); 6247 os::Linux::release_memory_special_huge_tlbfs(p, size); 6248 } 6249 } 6250 } 6251 6252 // Case 3 6253 for (int i = 0; i < num_sizes; i++) { 6254 const size_t size = sizes[i]; 6255 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6256 char* const req_addr = align_up(mapping2, alignment); 6257 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6258 // as the area around req_addr contains already existing mappings, the API should always 6259 // return NULL (as per contract, it cannot return another address) 6260 assert(p == NULL, "must be"); 6261 } 6262 } 6263 6264 ::munmap(mapping2, mapping_size); 6265 6266 } 6267 6268 static void test_reserve_memory_special_huge_tlbfs() { 6269 if (!UseHugeTLBFS) { 6270 return; 6271 } 6272 6273 test_reserve_memory_special_huge_tlbfs_only(); 6274 test_reserve_memory_special_huge_tlbfs_mixed(); 6275 } 6276 6277 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6278 if (!UseSHM) { 6279 return; 6280 } 6281 6282 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6283 6284 if (addr != NULL) { 6285 assert(is_aligned(addr, alignment), "Check"); 6286 assert(is_aligned(addr, os::large_page_size()), "Check"); 6287 6288 small_page_write(addr, size); 6289 6290 os::Linux::release_memory_special_shm(addr, size); 6291 } 6292 } 6293 6294 static void test_reserve_memory_special_shm() { 6295 size_t lp = os::large_page_size(); 6296 size_t ag = os::vm_allocation_granularity(); 6297 6298 for (size_t size = ag; size < lp * 3; size += ag) { 6299 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6300 test_reserve_memory_special_shm(size, alignment); 6301 } 6302 } 6303 } 6304 6305 static void test() { 6306 test_reserve_memory_special_huge_tlbfs(); 6307 test_reserve_memory_special_shm(); 6308 } 6309 }; 6310 6311 void TestReserveMemorySpecial_test() { 6312 TestReserveMemorySpecial::test(); 6313 } 6314 6315 #endif