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 "semaphore_posix.hpp" 67 #include "services/attachListener.hpp" 68 #include "services/memTracker.hpp" 69 #include "services/runtimeService.hpp" 70 #include "utilities/align.hpp" 71 #include "utilities/decoder.hpp" 72 #include "utilities/defaultStream.hpp" 73 #include "utilities/events.hpp" 74 #include "utilities/elfFile.hpp" 75 #include "utilities/growableArray.hpp" 76 #include "utilities/macros.hpp" 77 #include "utilities/vmError.hpp" 78 79 // put OS-includes here 80 # include <sys/types.h> 81 # include <sys/mman.h> 82 # include <sys/stat.h> 83 # include <sys/select.h> 84 # include <pthread.h> 85 # include <signal.h> 86 # include <errno.h> 87 # include <dlfcn.h> 88 # include <stdio.h> 89 # include <unistd.h> 90 # include <sys/resource.h> 91 # include <pthread.h> 92 # include <sys/stat.h> 93 # include <sys/time.h> 94 # include <sys/times.h> 95 # include <sys/utsname.h> 96 # include <sys/socket.h> 97 # include <sys/wait.h> 98 # include <pwd.h> 99 # include <poll.h> 100 # include <fcntl.h> 101 # include <string.h> 102 # include <syscall.h> 103 # include <sys/sysinfo.h> 104 # include <gnu/libc-version.h> 105 # include <sys/ipc.h> 106 # include <sys/shm.h> 107 # include <link.h> 108 # include <stdint.h> 109 # include <inttypes.h> 110 # include <sys/ioctl.h> 111 112 #ifndef _GNU_SOURCE 113 #define _GNU_SOURCE 114 #include <sched.h> 115 #undef _GNU_SOURCE 116 #else 117 #include <sched.h> 118 #endif 119 120 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 121 // getrusage() is prepared to handle the associated failure. 122 #ifndef RUSAGE_THREAD 123 #define RUSAGE_THREAD (1) /* only the calling thread */ 124 #endif 125 126 #define MAX_PATH (2 * K) 127 128 #define MAX_SECS 100000000 129 130 // for timer info max values which include all bits 131 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 132 133 enum CoredumpFilterBit { 134 FILE_BACKED_PVT_BIT = 1 << 2, 135 FILE_BACKED_SHARED_BIT = 1 << 3, 136 LARGEPAGES_BIT = 1 << 6, 137 DAX_SHARED_BIT = 1 << 8 138 }; 139 140 //////////////////////////////////////////////////////////////////////////////// 141 // global variables 142 julong os::Linux::_physical_memory = 0; 143 144 address os::Linux::_initial_thread_stack_bottom = NULL; 145 uintptr_t os::Linux::_initial_thread_stack_size = 0; 146 147 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 148 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL; 149 Mutex* os::Linux::_createThread_lock = NULL; 150 pthread_t os::Linux::_main_thread; 151 int os::Linux::_page_size = -1; 152 bool os::Linux::_supports_fast_thread_cpu_time = false; 153 uint32_t os::Linux::_os_version = 0; 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 // Return true if user is running as root. 231 232 bool os::have_special_privileges() { 233 static bool init = false; 234 static bool privileges = false; 235 if (!init) { 236 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 237 init = true; 238 } 239 return privileges; 240 } 241 242 243 #ifndef SYS_gettid 244 // i386: 224, ia64: 1105, amd64: 186, sparc 143 245 #ifdef __ia64__ 246 #define SYS_gettid 1105 247 #else 248 #ifdef __i386__ 249 #define SYS_gettid 224 250 #else 251 #ifdef __amd64__ 252 #define SYS_gettid 186 253 #else 254 #ifdef __sparc__ 255 #define SYS_gettid 143 256 #else 257 #error define gettid for the arch 258 #endif 259 #endif 260 #endif 261 #endif 262 #endif 263 264 265 // pid_t gettid() 266 // 267 // Returns the kernel thread id of the currently running thread. Kernel 268 // thread id is used to access /proc. 269 pid_t os::Linux::gettid() { 270 int rslt = syscall(SYS_gettid); 271 assert(rslt != -1, "must be."); // old linuxthreads implementation? 272 return (pid_t)rslt; 273 } 274 275 // Most versions of linux have a bug where the number of processors are 276 // determined by looking at the /proc file system. In a chroot environment, 277 // the system call returns 1. 278 static bool unsafe_chroot_detected = false; 279 static const char *unstable_chroot_error = "/proc file system not found.\n" 280 "Java may be unstable running multithreaded in a chroot " 281 "environment on Linux when /proc filesystem is not mounted."; 282 283 void os::Linux::initialize_system_info() { 284 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 285 if (processor_count() == 1) { 286 pid_t pid = os::Linux::gettid(); 287 char fname[32]; 288 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 289 FILE *fp = fopen(fname, "r"); 290 if (fp == NULL) { 291 unsafe_chroot_detected = true; 292 } else { 293 fclose(fp); 294 } 295 } 296 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 297 assert(processor_count() > 0, "linux error"); 298 } 299 300 void os::init_system_properties_values() { 301 // The next steps are taken in the product version: 302 // 303 // Obtain the JAVA_HOME value from the location of libjvm.so. 304 // This library should be located at: 305 // <JAVA_HOME>/lib/{client|server}/libjvm.so. 306 // 307 // If "/jre/lib/" appears at the right place in the path, then we 308 // assume libjvm.so is installed in a JDK and we use this path. 309 // 310 // Otherwise exit with message: "Could not create the Java virtual machine." 311 // 312 // The following extra steps are taken in the debugging version: 313 // 314 // If "/jre/lib/" does NOT appear at the right place in the path 315 // instead of exit check for $JAVA_HOME environment variable. 316 // 317 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 318 // then we append a fake suffix "hotspot/libjvm.so" to this path so 319 // it looks like libjvm.so is installed there 320 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 321 // 322 // Otherwise exit. 323 // 324 // Important note: if the location of libjvm.so changes this 325 // code needs to be changed accordingly. 326 327 // See ld(1): 328 // The linker uses the following search paths to locate required 329 // shared libraries: 330 // 1: ... 331 // ... 332 // 7: The default directories, normally /lib and /usr/lib. 333 #ifndef OVERRIDE_LIBPATH 334 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390) 335 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 336 #else 337 #define DEFAULT_LIBPATH "/lib:/usr/lib" 338 #endif 339 #else 340 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH 341 #endif 342 343 // Base path of extensions installed on the system. 344 #define SYS_EXT_DIR "/usr/java/packages" 345 #define EXTENSIONS_DIR "/lib/ext" 346 347 // Buffer that fits several sprintfs. 348 // Note that the space for the colon and the trailing null are provided 349 // by the nulls included by the sizeof operator. 350 const size_t bufsize = 351 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 352 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 353 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 354 355 // sysclasspath, java_home, dll_dir 356 { 357 char *pslash; 358 os::jvm_path(buf, bufsize); 359 360 // Found the full path to libjvm.so. 361 // Now cut the path to <java_home>/jre if we can. 362 pslash = strrchr(buf, '/'); 363 if (pslash != NULL) { 364 *pslash = '\0'; // Get rid of /libjvm.so. 365 } 366 pslash = strrchr(buf, '/'); 367 if (pslash != NULL) { 368 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 369 } 370 Arguments::set_dll_dir(buf); 371 372 if (pslash != NULL) { 373 pslash = strrchr(buf, '/'); 374 if (pslash != NULL) { 375 *pslash = '\0'; // Get rid of /lib. 376 } 377 } 378 Arguments::set_java_home(buf); 379 if (!set_boot_path('/', ':')) { 380 vm_exit_during_initialization("Failed setting boot class path.", NULL); 381 } 382 } 383 384 // Where to look for native libraries. 385 // 386 // Note: Due to a legacy implementation, most of the library path 387 // is set in the launcher. This was to accomodate linking restrictions 388 // on legacy Linux implementations (which are no longer supported). 389 // Eventually, all the library path setting will be done here. 390 // 391 // However, to prevent the proliferation of improperly built native 392 // libraries, the new path component /usr/java/packages is added here. 393 // Eventually, all the library path setting will be done here. 394 { 395 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 396 // should always exist (until the legacy problem cited above is 397 // addressed). 398 const char *v = ::getenv("LD_LIBRARY_PATH"); 399 const char *v_colon = ":"; 400 if (v == NULL) { v = ""; v_colon = ""; } 401 // That's +1 for the colon and +1 for the trailing '\0'. 402 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 403 strlen(v) + 1 + 404 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1, 405 mtInternal); 406 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon); 407 Arguments::set_library_path(ld_library_path); 408 FREE_C_HEAP_ARRAY(char, ld_library_path); 409 } 410 411 // Extensions directories. 412 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 413 Arguments::set_ext_dirs(buf); 414 415 FREE_C_HEAP_ARRAY(char, buf); 416 417 #undef DEFAULT_LIBPATH 418 #undef SYS_EXT_DIR 419 #undef EXTENSIONS_DIR 420 } 421 422 //////////////////////////////////////////////////////////////////////////////// 423 // breakpoint support 424 425 void os::breakpoint() { 426 BREAKPOINT; 427 } 428 429 extern "C" void breakpoint() { 430 // use debugger to set breakpoint here 431 } 432 433 //////////////////////////////////////////////////////////////////////////////// 434 // signal support 435 436 debug_only(static bool signal_sets_initialized = false); 437 static sigset_t unblocked_sigs, vm_sigs; 438 439 void os::Linux::signal_sets_init() { 440 // Should also have an assertion stating we are still single-threaded. 441 assert(!signal_sets_initialized, "Already initialized"); 442 // Fill in signals that are necessarily unblocked for all threads in 443 // the VM. Currently, we unblock the following signals: 444 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 445 // by -Xrs (=ReduceSignalUsage)); 446 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 447 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 448 // the dispositions or masks wrt these signals. 449 // Programs embedding the VM that want to use the above signals for their 450 // own purposes must, at this time, use the "-Xrs" option to prevent 451 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 452 // (See bug 4345157, and other related bugs). 453 // In reality, though, unblocking these signals is really a nop, since 454 // these signals are not blocked by default. 455 sigemptyset(&unblocked_sigs); 456 sigaddset(&unblocked_sigs, SIGILL); 457 sigaddset(&unblocked_sigs, SIGSEGV); 458 sigaddset(&unblocked_sigs, SIGBUS); 459 sigaddset(&unblocked_sigs, SIGFPE); 460 #if defined(PPC64) 461 sigaddset(&unblocked_sigs, SIGTRAP); 462 #endif 463 sigaddset(&unblocked_sigs, SR_signum); 464 465 if (!ReduceSignalUsage) { 466 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 467 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 468 } 469 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 470 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 471 } 472 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 473 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 474 } 475 } 476 // Fill in signals that are blocked by all but the VM thread. 477 sigemptyset(&vm_sigs); 478 if (!ReduceSignalUsage) { 479 sigaddset(&vm_sigs, BREAK_SIGNAL); 480 } 481 debug_only(signal_sets_initialized = true); 482 483 } 484 485 // These are signals that are unblocked while a thread is running Java. 486 // (For some reason, they get blocked by default.) 487 sigset_t* os::Linux::unblocked_signals() { 488 assert(signal_sets_initialized, "Not initialized"); 489 return &unblocked_sigs; 490 } 491 492 // These are the signals that are blocked while a (non-VM) thread is 493 // running Java. Only the VM thread handles these signals. 494 sigset_t* os::Linux::vm_signals() { 495 assert(signal_sets_initialized, "Not initialized"); 496 return &vm_sigs; 497 } 498 499 void os::Linux::hotspot_sigmask(Thread* thread) { 500 501 //Save caller's signal mask before setting VM signal mask 502 sigset_t caller_sigmask; 503 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 504 505 OSThread* osthread = thread->osthread(); 506 osthread->set_caller_sigmask(caller_sigmask); 507 508 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 509 510 if (!ReduceSignalUsage) { 511 if (thread->is_VM_thread()) { 512 // Only the VM thread handles BREAK_SIGNAL ... 513 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 514 } else { 515 // ... all other threads block BREAK_SIGNAL 516 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 517 } 518 } 519 } 520 521 ////////////////////////////////////////////////////////////////////////////// 522 // detecting pthread library 523 524 void os::Linux::libpthread_init() { 525 // Save glibc and pthread version strings. 526 #if !defined(_CS_GNU_LIBC_VERSION) || \ 527 !defined(_CS_GNU_LIBPTHREAD_VERSION) 528 #error "glibc too old (< 2.3.2)" 529 #endif 530 531 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 532 assert(n > 0, "cannot retrieve glibc version"); 533 char *str = (char *)malloc(n, mtInternal); 534 confstr(_CS_GNU_LIBC_VERSION, str, n); 535 os::Linux::set_glibc_version(str); 536 537 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 538 assert(n > 0, "cannot retrieve pthread version"); 539 str = (char *)malloc(n, mtInternal); 540 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 541 os::Linux::set_libpthread_version(str); 542 } 543 544 ///////////////////////////////////////////////////////////////////////////// 545 // thread stack expansion 546 547 // os::Linux::manually_expand_stack() takes care of expanding the thread 548 // stack. Note that this is normally not needed: pthread stacks allocate 549 // thread stack using mmap() without MAP_NORESERVE, so the stack is already 550 // committed. Therefore it is not necessary to expand the stack manually. 551 // 552 // Manually expanding the stack was historically needed on LinuxThreads 553 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays 554 // it is kept to deal with very rare corner cases: 555 // 556 // For one, user may run the VM on an own implementation of threads 557 // whose stacks are - like the old LinuxThreads - implemented using 558 // mmap(MAP_GROWSDOWN). 559 // 560 // Also, this coding may be needed if the VM is running on the primordial 561 // thread. Normally we avoid running on the primordial thread; however, 562 // user may still invoke the VM on the primordial thread. 563 // 564 // The following historical comment describes the details about running 565 // on a thread stack allocated with mmap(MAP_GROWSDOWN): 566 567 568 // Force Linux kernel to expand current thread stack. If "bottom" is close 569 // to the stack guard, caller should block all signals. 570 // 571 // MAP_GROWSDOWN: 572 // A special mmap() flag that is used to implement thread stacks. It tells 573 // kernel that the memory region should extend downwards when needed. This 574 // allows early versions of LinuxThreads to only mmap the first few pages 575 // when creating a new thread. Linux kernel will automatically expand thread 576 // stack as needed (on page faults). 577 // 578 // However, because the memory region of a MAP_GROWSDOWN stack can grow on 579 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 580 // region, it's hard to tell if the fault is due to a legitimate stack 581 // access or because of reading/writing non-exist memory (e.g. buffer 582 // overrun). As a rule, if the fault happens below current stack pointer, 583 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the 584 // application (see Linux kernel fault.c). 585 // 586 // This Linux feature can cause SIGSEGV when VM bangs thread stack for 587 // stack overflow detection. 588 // 589 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 590 // not use MAP_GROWSDOWN. 591 // 592 // To get around the problem and allow stack banging on Linux, we need to 593 // manually expand thread stack after receiving the SIGSEGV. 594 // 595 // There are two ways to expand thread stack to address "bottom", we used 596 // both of them in JVM before 1.5: 597 // 1. adjust stack pointer first so that it is below "bottom", and then 598 // touch "bottom" 599 // 2. mmap() the page in question 600 // 601 // Now alternate signal stack is gone, it's harder to use 2. For instance, 602 // if current sp is already near the lower end of page 101, and we need to 603 // call mmap() to map page 100, it is possible that part of the mmap() frame 604 // will be placed in page 100. When page 100 is mapped, it is zero-filled. 605 // That will destroy the mmap() frame and cause VM to crash. 606 // 607 // The following code works by adjusting sp first, then accessing the "bottom" 608 // page to force a page fault. Linux kernel will then automatically expand the 609 // stack mapping. 610 // 611 // _expand_stack_to() assumes its frame size is less than page size, which 612 // should always be true if the function is not inlined. 613 614 static void NOINLINE _expand_stack_to(address bottom) { 615 address sp; 616 size_t size; 617 volatile char *p; 618 619 // Adjust bottom to point to the largest address within the same page, it 620 // gives us a one-page buffer if alloca() allocates slightly more memory. 621 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size()); 622 bottom += os::Linux::page_size() - 1; 623 624 // sp might be slightly above current stack pointer; if that's the case, we 625 // will alloca() a little more space than necessary, which is OK. Don't use 626 // os::current_stack_pointer(), as its result can be slightly below current 627 // stack pointer, causing us to not alloca enough to reach "bottom". 628 sp = (address)&sp; 629 630 if (sp > bottom) { 631 size = sp - bottom; 632 p = (volatile char *)alloca(size); 633 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 634 p[0] = '\0'; 635 } 636 } 637 638 void os::Linux::expand_stack_to(address bottom) { 639 _expand_stack_to(bottom); 640 } 641 642 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 643 assert(t!=NULL, "just checking"); 644 assert(t->osthread()->expanding_stack(), "expand should be set"); 645 assert(t->stack_base() != NULL, "stack_base was not initialized"); 646 647 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) { 648 sigset_t mask_all, old_sigset; 649 sigfillset(&mask_all); 650 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 651 _expand_stack_to(addr); 652 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 653 return true; 654 } 655 return false; 656 } 657 658 ////////////////////////////////////////////////////////////////////////////// 659 // create new thread 660 661 // Thread start routine for all newly created threads 662 static void *thread_native_entry(Thread *thread) { 663 664 thread->record_stack_base_and_size(); 665 666 // Try to randomize the cache line index of hot stack frames. 667 // This helps when threads of the same stack traces evict each other's 668 // cache lines. The threads can be either from the same JVM instance, or 669 // from different JVM instances. The benefit is especially true for 670 // processors with hyperthreading technology. 671 static int counter = 0; 672 int pid = os::current_process_id(); 673 alloca(((pid ^ counter++) & 7) * 128); 674 675 thread->initialize_thread_current(); 676 677 OSThread* osthread = thread->osthread(); 678 Monitor* sync = osthread->startThread_lock(); 679 680 osthread->set_thread_id(os::current_thread_id()); 681 682 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 683 os::current_thread_id(), (uintx) pthread_self()); 684 685 if (UseNUMA) { 686 int lgrp_id = os::numa_get_group_id(); 687 if (lgrp_id != -1) { 688 thread->set_lgrp_id(lgrp_id); 689 } 690 } 691 // initialize signal mask for this thread 692 os::Linux::hotspot_sigmask(thread); 693 694 // initialize floating point control register 695 os::Linux::init_thread_fpu_state(); 696 697 // handshaking with parent thread 698 { 699 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 700 701 // notify parent thread 702 osthread->set_state(INITIALIZED); 703 sync->notify_all(); 704 705 // wait until os::start_thread() 706 while (osthread->get_state() == INITIALIZED) { 707 sync->wait(Mutex::_no_safepoint_check_flag); 708 } 709 } 710 711 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected"); 712 713 // call one more level start routine 714 thread->call_run(); 715 716 // Note: at this point the thread object may already have deleted itself. 717 // Prevent dereferencing it from here on out. 718 thread = NULL; 719 720 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 721 os::current_thread_id(), (uintx) pthread_self()); 722 723 return 0; 724 } 725 726 bool os::create_thread(Thread* thread, ThreadType thr_type, 727 size_t req_stack_size) { 728 assert(thread->osthread() == NULL, "caller responsible"); 729 730 // Allocate the OSThread object 731 OSThread* osthread = new OSThread(NULL, NULL); 732 if (osthread == NULL) { 733 return false; 734 } 735 736 // set the correct thread state 737 osthread->set_thread_type(thr_type); 738 739 // Initial state is ALLOCATED but not INITIALIZED 740 osthread->set_state(ALLOCATED); 741 742 thread->set_osthread(osthread); 743 744 // init thread attributes 745 pthread_attr_t attr; 746 pthread_attr_init(&attr); 747 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 748 749 // Calculate stack size if it's not specified by caller. 750 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size); 751 // In the Linux NPTL pthread implementation the guard size mechanism 752 // is not implemented properly. The posix standard requires adding 753 // the size of the guard pages to the stack size, instead Linux 754 // takes the space out of 'stacksize'. Thus we adapt the requested 755 // stack_size by the size of the guard pages to mimick proper 756 // behaviour. However, be careful not to end up with a size 757 // of zero due to overflow. Don't add the guard page in that case. 758 size_t guard_size = os::Linux::default_guard_size(thr_type); 759 if (stack_size <= SIZE_MAX - guard_size) { 760 stack_size += guard_size; 761 } 762 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned"); 763 764 int status = pthread_attr_setstacksize(&attr, stack_size); 765 assert_status(status == 0, status, "pthread_attr_setstacksize"); 766 767 // Configure glibc guard page. 768 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 769 770 ThreadState state; 771 772 { 773 pthread_t tid; 774 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread); 775 776 char buf[64]; 777 if (ret == 0) { 778 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ", 779 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 780 } else { 781 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.", 782 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 783 } 784 785 pthread_attr_destroy(&attr); 786 787 if (ret != 0) { 788 // Need to clean up stuff we've allocated so far 789 thread->set_osthread(NULL); 790 delete osthread; 791 return false; 792 } 793 794 // Store pthread info into the OSThread 795 osthread->set_pthread_id(tid); 796 797 // Wait until child thread is either initialized or aborted 798 { 799 Monitor* sync_with_child = osthread->startThread_lock(); 800 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 801 while ((state = osthread->get_state()) == ALLOCATED) { 802 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 803 } 804 } 805 } 806 807 // Aborted due to thread limit being reached 808 if (state == ZOMBIE) { 809 thread->set_osthread(NULL); 810 delete osthread; 811 return false; 812 } 813 814 // The thread is returned suspended (in state INITIALIZED), 815 // and is started higher up in the call chain 816 assert(state == INITIALIZED, "race condition"); 817 return true; 818 } 819 820 ///////////////////////////////////////////////////////////////////////////// 821 // attach existing thread 822 823 // bootstrap the main thread 824 bool os::create_main_thread(JavaThread* thread) { 825 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 826 return create_attached_thread(thread); 827 } 828 829 bool os::create_attached_thread(JavaThread* thread) { 830 #ifdef ASSERT 831 thread->verify_not_published(); 832 #endif 833 834 // Allocate the OSThread object 835 OSThread* osthread = new OSThread(NULL, NULL); 836 837 if (osthread == NULL) { 838 return false; 839 } 840 841 // Store pthread info into the OSThread 842 osthread->set_thread_id(os::Linux::gettid()); 843 osthread->set_pthread_id(::pthread_self()); 844 845 // initialize floating point control register 846 os::Linux::init_thread_fpu_state(); 847 848 // Initial thread state is RUNNABLE 849 osthread->set_state(RUNNABLE); 850 851 thread->set_osthread(osthread); 852 853 if (UseNUMA) { 854 int lgrp_id = os::numa_get_group_id(); 855 if (lgrp_id != -1) { 856 thread->set_lgrp_id(lgrp_id); 857 } 858 } 859 860 if (os::is_primordial_thread()) { 861 // If current thread is primordial thread, its stack is mapped on demand, 862 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 863 // the entire stack region to avoid SEGV in stack banging. 864 // It is also useful to get around the heap-stack-gap problem on SuSE 865 // kernel (see 4821821 for details). We first expand stack to the top 866 // of yellow zone, then enable stack yellow zone (order is significant, 867 // enabling yellow zone first will crash JVM on SuSE Linux), so there 868 // is no gap between the last two virtual memory regions. 869 870 JavaThread *jt = (JavaThread *)thread; 871 address addr = jt->stack_reserved_zone_base(); 872 assert(addr != NULL, "initialization problem?"); 873 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 874 875 osthread->set_expanding_stack(); 876 os::Linux::manually_expand_stack(jt, addr); 877 osthread->clear_expanding_stack(); 878 } 879 880 // initialize signal mask for this thread 881 // and save the caller's signal mask 882 os::Linux::hotspot_sigmask(thread); 883 884 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 885 os::current_thread_id(), (uintx) pthread_self()); 886 887 return true; 888 } 889 890 void os::pd_start_thread(Thread* thread) { 891 OSThread * osthread = thread->osthread(); 892 assert(osthread->get_state() != INITIALIZED, "just checking"); 893 Monitor* sync_with_child = osthread->startThread_lock(); 894 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 895 sync_with_child->notify(); 896 } 897 898 // Free Linux resources related to the OSThread 899 void os::free_thread(OSThread* osthread) { 900 assert(osthread != NULL, "osthread not set"); 901 902 // We are told to free resources of the argument thread, 903 // but we can only really operate on the current thread. 904 assert(Thread::current()->osthread() == osthread, 905 "os::free_thread but not current thread"); 906 907 #ifdef ASSERT 908 sigset_t current; 909 sigemptyset(¤t); 910 pthread_sigmask(SIG_SETMASK, NULL, ¤t); 911 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!"); 912 #endif 913 914 // Restore caller's signal mask 915 sigset_t sigmask = osthread->caller_sigmask(); 916 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 917 918 delete osthread; 919 } 920 921 ////////////////////////////////////////////////////////////////////////////// 922 // primordial thread 923 924 // Check if current thread is the primordial thread, similar to Solaris thr_main. 925 bool os::is_primordial_thread(void) { 926 if (suppress_primordial_thread_resolution) { 927 return false; 928 } 929 char dummy; 930 // If called before init complete, thread stack bottom will be null. 931 // Can be called if fatal error occurs before initialization. 932 if (os::Linux::initial_thread_stack_bottom() == NULL) return false; 933 assert(os::Linux::initial_thread_stack_bottom() != NULL && 934 os::Linux::initial_thread_stack_size() != 0, 935 "os::init did not locate primordial thread's stack region"); 936 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() && 937 (address)&dummy < os::Linux::initial_thread_stack_bottom() + 938 os::Linux::initial_thread_stack_size()) { 939 return true; 940 } else { 941 return false; 942 } 943 } 944 945 // Find the virtual memory area that contains addr 946 static bool find_vma(address addr, address* vma_low, address* vma_high) { 947 FILE *fp = fopen("/proc/self/maps", "r"); 948 if (fp) { 949 address low, high; 950 while (!feof(fp)) { 951 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 952 if (low <= addr && addr < high) { 953 if (vma_low) *vma_low = low; 954 if (vma_high) *vma_high = high; 955 fclose(fp); 956 return true; 957 } 958 } 959 for (;;) { 960 int ch = fgetc(fp); 961 if (ch == EOF || ch == (int)'\n') break; 962 } 963 } 964 fclose(fp); 965 } 966 return false; 967 } 968 969 // Locate primordial thread stack. This special handling of primordial thread stack 970 // is needed because pthread_getattr_np() on most (all?) Linux distros returns 971 // bogus value for the primordial process thread. While the launcher has created 972 // the VM in a new thread since JDK 6, we still have to allow for the use of the 973 // JNI invocation API from a primordial thread. 974 void os::Linux::capture_initial_stack(size_t max_size) { 975 976 // max_size is either 0 (which means accept OS default for thread stacks) or 977 // a user-specified value known to be at least the minimum needed. If we 978 // are actually on the primordial thread we can make it appear that we have a 979 // smaller max_size stack by inserting the guard pages at that location. But we 980 // cannot do anything to emulate a larger stack than what has been provided by 981 // the OS or threading library. In fact if we try to use a stack greater than 982 // what is set by rlimit then we will crash the hosting process. 983 984 // Maximum stack size is the easy part, get it from RLIMIT_STACK. 985 // If this is "unlimited" then it will be a huge value. 986 struct rlimit rlim; 987 getrlimit(RLIMIT_STACK, &rlim); 988 size_t stack_size = rlim.rlim_cur; 989 990 // 6308388: a bug in ld.so will relocate its own .data section to the 991 // lower end of primordial stack; reduce ulimit -s value a little bit 992 // so we won't install guard page on ld.so's data section. 993 // But ensure we don't underflow the stack size - allow 1 page spare 994 if (stack_size >= (size_t)(3 * page_size())) { 995 stack_size -= 2 * page_size(); 996 } 997 998 // Try to figure out where the stack base (top) is. This is harder. 999 // 1000 // When an application is started, glibc saves the initial stack pointer in 1001 // a global variable "__libc_stack_end", which is then used by system 1002 // libraries. __libc_stack_end should be pretty close to stack top. The 1003 // variable is available since the very early days. However, because it is 1004 // a private interface, it could disappear in the future. 1005 // 1006 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1007 // to __libc_stack_end, it is very close to stack top, but isn't the real 1008 // stack top. Note that /proc may not exist if VM is running as a chroot 1009 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1010 // /proc/<pid>/stat could change in the future (though unlikely). 1011 // 1012 // We try __libc_stack_end first. If that doesn't work, look for 1013 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1014 // as a hint, which should work well in most cases. 1015 1016 uintptr_t stack_start; 1017 1018 // try __libc_stack_end first 1019 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1020 if (p && *p) { 1021 stack_start = *p; 1022 } else { 1023 // see if we can get the start_stack field from /proc/self/stat 1024 FILE *fp; 1025 int pid; 1026 char state; 1027 int ppid; 1028 int pgrp; 1029 int session; 1030 int nr; 1031 int tpgrp; 1032 unsigned long flags; 1033 unsigned long minflt; 1034 unsigned long cminflt; 1035 unsigned long majflt; 1036 unsigned long cmajflt; 1037 unsigned long utime; 1038 unsigned long stime; 1039 long cutime; 1040 long cstime; 1041 long prio; 1042 long nice; 1043 long junk; 1044 long it_real; 1045 uintptr_t start; 1046 uintptr_t vsize; 1047 intptr_t rss; 1048 uintptr_t rsslim; 1049 uintptr_t scodes; 1050 uintptr_t ecode; 1051 int i; 1052 1053 // Figure what the primordial thread stack base is. Code is inspired 1054 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1055 // followed by command name surrounded by parentheses, state, etc. 1056 char stat[2048]; 1057 int statlen; 1058 1059 fp = fopen("/proc/self/stat", "r"); 1060 if (fp) { 1061 statlen = fread(stat, 1, 2047, fp); 1062 stat[statlen] = '\0'; 1063 fclose(fp); 1064 1065 // Skip pid and the command string. Note that we could be dealing with 1066 // weird command names, e.g. user could decide to rename java launcher 1067 // to "java 1.4.2 :)", then the stat file would look like 1068 // 1234 (java 1.4.2 :)) R ... ... 1069 // We don't really need to know the command string, just find the last 1070 // occurrence of ")" and then start parsing from there. See bug 4726580. 1071 char * s = strrchr(stat, ')'); 1072 1073 i = 0; 1074 if (s) { 1075 // Skip blank chars 1076 do { s++; } while (s && isspace(*s)); 1077 1078 #define _UFM UINTX_FORMAT 1079 #define _DFM INTX_FORMAT 1080 1081 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1082 // 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 1083 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, 1084 &state, // 3 %c 1085 &ppid, // 4 %d 1086 &pgrp, // 5 %d 1087 &session, // 6 %d 1088 &nr, // 7 %d 1089 &tpgrp, // 8 %d 1090 &flags, // 9 %lu 1091 &minflt, // 10 %lu 1092 &cminflt, // 11 %lu 1093 &majflt, // 12 %lu 1094 &cmajflt, // 13 %lu 1095 &utime, // 14 %lu 1096 &stime, // 15 %lu 1097 &cutime, // 16 %ld 1098 &cstime, // 17 %ld 1099 &prio, // 18 %ld 1100 &nice, // 19 %ld 1101 &junk, // 20 %ld 1102 &it_real, // 21 %ld 1103 &start, // 22 UINTX_FORMAT 1104 &vsize, // 23 UINTX_FORMAT 1105 &rss, // 24 INTX_FORMAT 1106 &rsslim, // 25 UINTX_FORMAT 1107 &scodes, // 26 UINTX_FORMAT 1108 &ecode, // 27 UINTX_FORMAT 1109 &stack_start); // 28 UINTX_FORMAT 1110 } 1111 1112 #undef _UFM 1113 #undef _DFM 1114 1115 if (i != 28 - 2) { 1116 assert(false, "Bad conversion from /proc/self/stat"); 1117 // product mode - assume we are the primordial thread, good luck in the 1118 // embedded case. 1119 warning("Can't detect primordial thread stack location - bad conversion"); 1120 stack_start = (uintptr_t) &rlim; 1121 } 1122 } else { 1123 // For some reason we can't open /proc/self/stat (for example, running on 1124 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1125 // most cases, so don't abort: 1126 warning("Can't detect primordial thread stack location - no /proc/self/stat"); 1127 stack_start = (uintptr_t) &rlim; 1128 } 1129 } 1130 1131 // Now we have a pointer (stack_start) very close to the stack top, the 1132 // next thing to do is to figure out the exact location of stack top. We 1133 // can find out the virtual memory area that contains stack_start by 1134 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1135 // and its upper limit is the real stack top. (again, this would fail if 1136 // running inside chroot, because /proc may not exist.) 1137 1138 uintptr_t stack_top; 1139 address low, high; 1140 if (find_vma((address)stack_start, &low, &high)) { 1141 // success, "high" is the true stack top. (ignore "low", because initial 1142 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1143 stack_top = (uintptr_t)high; 1144 } else { 1145 // failed, likely because /proc/self/maps does not exist 1146 warning("Can't detect primordial thread stack location - find_vma failed"); 1147 // best effort: stack_start is normally within a few pages below the real 1148 // stack top, use it as stack top, and reduce stack size so we won't put 1149 // guard page outside stack. 1150 stack_top = stack_start; 1151 stack_size -= 16 * page_size(); 1152 } 1153 1154 // stack_top could be partially down the page so align it 1155 stack_top = align_up(stack_top, page_size()); 1156 1157 // Allowed stack value is minimum of max_size and what we derived from rlimit 1158 if (max_size > 0) { 1159 _initial_thread_stack_size = MIN2(max_size, stack_size); 1160 } else { 1161 // Accept the rlimit max, but if stack is unlimited then it will be huge, so 1162 // clamp it at 8MB as we do on Solaris 1163 _initial_thread_stack_size = MIN2(stack_size, 8*M); 1164 } 1165 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size()); 1166 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1167 1168 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!"); 1169 1170 if (log_is_enabled(Info, os, thread)) { 1171 // See if we seem to be on primordial process thread 1172 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) && 1173 uintptr_t(&rlim) < stack_top; 1174 1175 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: " 1176 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT, 1177 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K, 1178 stack_top, intptr_t(_initial_thread_stack_bottom)); 1179 } 1180 } 1181 1182 //////////////////////////////////////////////////////////////////////////////// 1183 // time support 1184 1185 #ifndef SUPPORTS_CLOCK_MONOTONIC 1186 #error "Build platform doesn't support clock_gettime and related functionality" 1187 #endif 1188 1189 // Time since start-up in seconds to a fine granularity. 1190 // Used by VMSelfDestructTimer and the MemProfiler. 1191 double os::elapsedTime() { 1192 1193 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1194 } 1195 1196 jlong os::elapsed_counter() { 1197 return javaTimeNanos() - initial_time_count; 1198 } 1199 1200 jlong os::elapsed_frequency() { 1201 return NANOSECS_PER_SEC; // nanosecond resolution 1202 } 1203 1204 bool os::supports_vtime() { return true; } 1205 bool os::enable_vtime() { return false; } 1206 bool os::vtime_enabled() { return false; } 1207 1208 double os::elapsedVTime() { 1209 struct rusage usage; 1210 int retval = getrusage(RUSAGE_THREAD, &usage); 1211 if (retval == 0) { 1212 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); 1213 } else { 1214 // better than nothing, but not much 1215 return elapsedTime(); 1216 } 1217 } 1218 1219 jlong os::javaTimeMillis() { 1220 timeval time; 1221 int status = gettimeofday(&time, NULL); 1222 assert(status != -1, "linux error"); 1223 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1224 } 1225 1226 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) { 1227 timeval time; 1228 int status = gettimeofday(&time, NULL); 1229 assert(status != -1, "linux error"); 1230 seconds = jlong(time.tv_sec); 1231 nanos = jlong(time.tv_usec) * 1000; 1232 } 1233 1234 void os::Linux::fast_thread_clock_init() { 1235 if (!UseLinuxPosixThreadCPUClocks) { 1236 return; 1237 } 1238 clockid_t clockid; 1239 struct timespec tp; 1240 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1241 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1242 1243 // Switch to using fast clocks for thread cpu time if 1244 // the clock_getres() returns 0 error code. 1245 // Note, that some kernels may support the current thread 1246 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1247 // returned by the pthread_getcpuclockid(). 1248 // If the fast Posix clocks are supported then the clock_getres() 1249 // must return at least tp.tv_sec == 0 which means a resolution 1250 // better than 1 sec. This is extra check for reliability. 1251 1252 if (pthread_getcpuclockid_func && 1253 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1254 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1255 _supports_fast_thread_cpu_time = true; 1256 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1257 } 1258 } 1259 1260 jlong os::javaTimeNanos() { 1261 if (os::supports_monotonic_clock()) { 1262 struct timespec tp; 1263 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp); 1264 assert(status == 0, "gettime error"); 1265 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1266 return result; 1267 } else { 1268 timeval time; 1269 int status = gettimeofday(&time, NULL); 1270 assert(status != -1, "linux error"); 1271 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1272 return 1000 * usecs; 1273 } 1274 } 1275 1276 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1277 if (os::supports_monotonic_clock()) { 1278 info_ptr->max_value = ALL_64_BITS; 1279 1280 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1281 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1282 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1283 } else { 1284 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1285 info_ptr->max_value = ALL_64_BITS; 1286 1287 // gettimeofday is a real time clock so it skips 1288 info_ptr->may_skip_backward = true; 1289 info_ptr->may_skip_forward = true; 1290 } 1291 1292 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1293 } 1294 1295 // Return the real, user, and system times in seconds from an 1296 // arbitrary fixed point in the past. 1297 bool os::getTimesSecs(double* process_real_time, 1298 double* process_user_time, 1299 double* process_system_time) { 1300 struct tms ticks; 1301 clock_t real_ticks = times(&ticks); 1302 1303 if (real_ticks == (clock_t) (-1)) { 1304 return false; 1305 } else { 1306 double ticks_per_second = (double) clock_tics_per_sec; 1307 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1308 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1309 *process_real_time = ((double) real_ticks) / ticks_per_second; 1310 1311 return true; 1312 } 1313 } 1314 1315 1316 char * os::local_time_string(char *buf, size_t buflen) { 1317 struct tm t; 1318 time_t long_time; 1319 time(&long_time); 1320 localtime_r(&long_time, &t); 1321 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1322 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1323 t.tm_hour, t.tm_min, t.tm_sec); 1324 return buf; 1325 } 1326 1327 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1328 return localtime_r(clock, res); 1329 } 1330 1331 //////////////////////////////////////////////////////////////////////////////// 1332 // runtime exit support 1333 1334 // Note: os::shutdown() might be called very early during initialization, or 1335 // called from signal handler. Before adding something to os::shutdown(), make 1336 // sure it is async-safe and can handle partially initialized VM. 1337 void os::shutdown() { 1338 1339 // allow PerfMemory to attempt cleanup of any persistent resources 1340 perfMemory_exit(); 1341 1342 // needs to remove object in file system 1343 AttachListener::abort(); 1344 1345 // flush buffered output, finish log files 1346 ostream_abort(); 1347 1348 // Check for abort hook 1349 abort_hook_t abort_hook = Arguments::abort_hook(); 1350 if (abort_hook != NULL) { 1351 abort_hook(); 1352 } 1353 1354 } 1355 1356 // Note: os::abort() might be called very early during initialization, or 1357 // called from signal handler. Before adding something to os::abort(), make 1358 // sure it is async-safe and can handle partially initialized VM. 1359 void os::abort(bool dump_core, void* siginfo, const void* context) { 1360 os::shutdown(); 1361 if (dump_core) { 1362 if (DumpPrivateMappingsInCore) { 1363 ClassLoader::close_jrt_image(); 1364 } 1365 #ifndef PRODUCT 1366 fdStream out(defaultStream::output_fd()); 1367 out.print_raw("Current thread is "); 1368 char buf[16]; 1369 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1370 out.print_raw_cr(buf); 1371 out.print_raw_cr("Dumping core ..."); 1372 #endif 1373 ::abort(); // dump core 1374 } 1375 1376 ::exit(1); 1377 } 1378 1379 // Die immediately, no exit hook, no abort hook, no cleanup. 1380 void os::die() { 1381 ::abort(); 1382 } 1383 1384 // thread_id is kernel thread id (similar to Solaris LWP id) 1385 intx os::current_thread_id() { return os::Linux::gettid(); } 1386 int os::current_process_id() { 1387 return ::getpid(); 1388 } 1389 1390 // DLL functions 1391 1392 const char* os::dll_file_extension() { return ".so"; } 1393 1394 // This must be hard coded because it's the system's temporary 1395 // directory not the java application's temp directory, ala java.io.tmpdir. 1396 const char* os::get_temp_directory() { return "/tmp"; } 1397 1398 static bool file_exists(const char* filename) { 1399 struct stat statbuf; 1400 if (filename == NULL || strlen(filename) == 0) { 1401 return false; 1402 } 1403 return os::stat(filename, &statbuf) == 0; 1404 } 1405 1406 // check if addr is inside libjvm.so 1407 bool os::address_is_in_vm(address addr) { 1408 static address libjvm_base_addr; 1409 Dl_info dlinfo; 1410 1411 if (libjvm_base_addr == NULL) { 1412 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1413 libjvm_base_addr = (address)dlinfo.dli_fbase; 1414 } 1415 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1416 } 1417 1418 if (dladdr((void *)addr, &dlinfo) != 0) { 1419 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1420 } 1421 1422 return false; 1423 } 1424 1425 bool os::dll_address_to_function_name(address addr, char *buf, 1426 int buflen, int *offset, 1427 bool demangle) { 1428 // buf is not optional, but offset is optional 1429 assert(buf != NULL, "sanity check"); 1430 1431 Dl_info dlinfo; 1432 1433 if (dladdr((void*)addr, &dlinfo) != 0) { 1434 // see if we have a matching symbol 1435 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1436 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) { 1437 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1438 } 1439 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1440 return true; 1441 } 1442 // no matching symbol so try for just file info 1443 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1444 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1445 buf, buflen, offset, dlinfo.dli_fname, demangle)) { 1446 return true; 1447 } 1448 } 1449 } 1450 1451 buf[0] = '\0'; 1452 if (offset != NULL) *offset = -1; 1453 return false; 1454 } 1455 1456 struct _address_to_library_name { 1457 address addr; // input : memory address 1458 size_t buflen; // size of fname 1459 char* fname; // output: library name 1460 address base; // library base addr 1461 }; 1462 1463 static int address_to_library_name_callback(struct dl_phdr_info *info, 1464 size_t size, void *data) { 1465 int i; 1466 bool found = false; 1467 address libbase = NULL; 1468 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1469 1470 // iterate through all loadable segments 1471 for (i = 0; i < info->dlpi_phnum; i++) { 1472 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1473 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1474 // base address of a library is the lowest address of its loaded 1475 // segments. 1476 if (libbase == NULL || libbase > segbase) { 1477 libbase = segbase; 1478 } 1479 // see if 'addr' is within current segment 1480 if (segbase <= d->addr && 1481 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1482 found = true; 1483 } 1484 } 1485 } 1486 1487 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1488 // so dll_address_to_library_name() can fall through to use dladdr() which 1489 // can figure out executable name from argv[0]. 1490 if (found && info->dlpi_name && info->dlpi_name[0]) { 1491 d->base = libbase; 1492 if (d->fname) { 1493 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1494 } 1495 return 1; 1496 } 1497 return 0; 1498 } 1499 1500 bool os::dll_address_to_library_name(address addr, char* buf, 1501 int buflen, int* offset) { 1502 // buf is not optional, but offset is optional 1503 assert(buf != NULL, "sanity check"); 1504 1505 Dl_info dlinfo; 1506 struct _address_to_library_name data; 1507 1508 // There is a bug in old glibc dladdr() implementation that it could resolve 1509 // to wrong library name if the .so file has a base address != NULL. Here 1510 // we iterate through the program headers of all loaded libraries to find 1511 // out which library 'addr' really belongs to. This workaround can be 1512 // removed once the minimum requirement for glibc is moved to 2.3.x. 1513 data.addr = addr; 1514 data.fname = buf; 1515 data.buflen = buflen; 1516 data.base = NULL; 1517 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1518 1519 if (rslt) { 1520 // buf already contains library name 1521 if (offset) *offset = addr - data.base; 1522 return true; 1523 } 1524 if (dladdr((void*)addr, &dlinfo) != 0) { 1525 if (dlinfo.dli_fname != NULL) { 1526 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1527 } 1528 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1529 *offset = addr - (address)dlinfo.dli_fbase; 1530 } 1531 return true; 1532 } 1533 1534 buf[0] = '\0'; 1535 if (offset) *offset = -1; 1536 return false; 1537 } 1538 1539 // Loads .dll/.so and 1540 // in case of error it checks if .dll/.so was built for the 1541 // same architecture as Hotspot is running on 1542 1543 1544 // Remember the stack's state. The Linux dynamic linker will change 1545 // the stack to 'executable' at most once, so we must safepoint only once. 1546 bool os::Linux::_stack_is_executable = false; 1547 1548 // VM operation that loads a library. This is necessary if stack protection 1549 // of the Java stacks can be lost during loading the library. If we 1550 // do not stop the Java threads, they can stack overflow before the stacks 1551 // are protected again. 1552 class VM_LinuxDllLoad: public VM_Operation { 1553 private: 1554 const char *_filename; 1555 char *_ebuf; 1556 int _ebuflen; 1557 void *_lib; 1558 public: 1559 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1560 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1561 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1562 void doit() { 1563 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1564 os::Linux::_stack_is_executable = true; 1565 } 1566 void* loaded_library() { return _lib; } 1567 }; 1568 1569 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1570 void * result = NULL; 1571 bool load_attempted = false; 1572 1573 // Check whether the library to load might change execution rights 1574 // of the stack. If they are changed, the protection of the stack 1575 // guard pages will be lost. We need a safepoint to fix this. 1576 // 1577 // See Linux man page execstack(8) for more info. 1578 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1579 if (!ElfFile::specifies_noexecstack(filename)) { 1580 if (!is_init_completed()) { 1581 os::Linux::_stack_is_executable = true; 1582 // This is OK - No Java threads have been created yet, and hence no 1583 // stack guard pages to fix. 1584 // 1585 // Dynamic loader will make all stacks executable after 1586 // this function returns, and will not do that again. 1587 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet."); 1588 } else { 1589 warning("You have loaded library %s which might have disabled stack guard. " 1590 "The VM will try to fix the stack guard now.\n" 1591 "It's highly recommended that you fix the library with " 1592 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1593 filename); 1594 1595 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1596 JavaThread *jt = JavaThread::current(); 1597 if (jt->thread_state() != _thread_in_native) { 1598 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1599 // that requires ExecStack. Cannot enter safe point. Let's give up. 1600 warning("Unable to fix stack guard. Giving up."); 1601 } else { 1602 if (!LoadExecStackDllInVMThread) { 1603 // This is for the case where the DLL has an static 1604 // constructor function that executes JNI code. We cannot 1605 // load such DLLs in the VMThread. 1606 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1607 } 1608 1609 ThreadInVMfromNative tiv(jt); 1610 debug_only(VMNativeEntryWrapper vew;) 1611 1612 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1613 VMThread::execute(&op); 1614 if (LoadExecStackDllInVMThread) { 1615 result = op.loaded_library(); 1616 } 1617 load_attempted = true; 1618 } 1619 } 1620 } 1621 } 1622 1623 if (!load_attempted) { 1624 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1625 } 1626 1627 if (result != NULL) { 1628 // Successful loading 1629 return result; 1630 } 1631 1632 Elf32_Ehdr elf_head; 1633 int diag_msg_max_length=ebuflen-strlen(ebuf); 1634 char* diag_msg_buf=ebuf+strlen(ebuf); 1635 1636 if (diag_msg_max_length==0) { 1637 // No more space in ebuf for additional diagnostics message 1638 return NULL; 1639 } 1640 1641 1642 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1643 1644 if (file_descriptor < 0) { 1645 // Can't open library, report dlerror() message 1646 return NULL; 1647 } 1648 1649 bool failed_to_read_elf_head= 1650 (sizeof(elf_head)!= 1651 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1652 1653 ::close(file_descriptor); 1654 if (failed_to_read_elf_head) { 1655 // file i/o error - report dlerror() msg 1656 return NULL; 1657 } 1658 1659 typedef struct { 1660 Elf32_Half code; // Actual value as defined in elf.h 1661 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1662 unsigned char elf_class; // 32 or 64 bit 1663 unsigned char endianess; // MSB or LSB 1664 char* name; // String representation 1665 } arch_t; 1666 1667 #ifndef EM_486 1668 #define EM_486 6 /* Intel 80486 */ 1669 #endif 1670 #ifndef EM_AARCH64 1671 #define EM_AARCH64 183 /* ARM AARCH64 */ 1672 #endif 1673 1674 static const arch_t arch_array[]={ 1675 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1676 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1677 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1678 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1679 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1680 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1681 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1682 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1683 #if defined(VM_LITTLE_ENDIAN) 1684 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"}, 1685 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"}, 1686 #else 1687 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1688 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"}, 1689 #endif 1690 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1691 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1692 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1693 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1694 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1695 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1696 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1697 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1698 }; 1699 1700 #if (defined IA32) 1701 static Elf32_Half running_arch_code=EM_386; 1702 #elif (defined AMD64) || (defined X32) 1703 static Elf32_Half running_arch_code=EM_X86_64; 1704 #elif (defined IA64) 1705 static Elf32_Half running_arch_code=EM_IA_64; 1706 #elif (defined __sparc) && (defined _LP64) 1707 static Elf32_Half running_arch_code=EM_SPARCV9; 1708 #elif (defined __sparc) && (!defined _LP64) 1709 static Elf32_Half running_arch_code=EM_SPARC; 1710 #elif (defined __powerpc64__) 1711 static Elf32_Half running_arch_code=EM_PPC64; 1712 #elif (defined __powerpc__) 1713 static Elf32_Half running_arch_code=EM_PPC; 1714 #elif (defined AARCH64) 1715 static Elf32_Half running_arch_code=EM_AARCH64; 1716 #elif (defined ARM) 1717 static Elf32_Half running_arch_code=EM_ARM; 1718 #elif (defined S390) 1719 static Elf32_Half running_arch_code=EM_S390; 1720 #elif (defined ALPHA) 1721 static Elf32_Half running_arch_code=EM_ALPHA; 1722 #elif (defined MIPSEL) 1723 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1724 #elif (defined PARISC) 1725 static Elf32_Half running_arch_code=EM_PARISC; 1726 #elif (defined MIPS) 1727 static Elf32_Half running_arch_code=EM_MIPS; 1728 #elif (defined M68K) 1729 static Elf32_Half running_arch_code=EM_68K; 1730 #elif (defined SH) 1731 static Elf32_Half running_arch_code=EM_SH; 1732 #else 1733 #error Method os::dll_load requires that one of following is defined:\ 1734 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc 1735 #endif 1736 1737 // Identify compatability class for VM's architecture and library's architecture 1738 // Obtain string descriptions for architectures 1739 1740 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1741 int running_arch_index=-1; 1742 1743 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1744 if (running_arch_code == arch_array[i].code) { 1745 running_arch_index = i; 1746 } 1747 if (lib_arch.code == arch_array[i].code) { 1748 lib_arch.compat_class = arch_array[i].compat_class; 1749 lib_arch.name = arch_array[i].name; 1750 } 1751 } 1752 1753 assert(running_arch_index != -1, 1754 "Didn't find running architecture code (running_arch_code) in arch_array"); 1755 if (running_arch_index == -1) { 1756 // Even though running architecture detection failed 1757 // we may still continue with reporting dlerror() message 1758 return NULL; 1759 } 1760 1761 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1762 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1763 return NULL; 1764 } 1765 1766 #ifndef S390 1767 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1768 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1769 return NULL; 1770 } 1771 #endif // !S390 1772 1773 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1774 if (lib_arch.name!=NULL) { 1775 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1776 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1777 lib_arch.name, arch_array[running_arch_index].name); 1778 } else { 1779 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1780 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1781 lib_arch.code, 1782 arch_array[running_arch_index].name); 1783 } 1784 } 1785 1786 return NULL; 1787 } 1788 1789 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 1790 int ebuflen) { 1791 void * result = ::dlopen(filename, RTLD_LAZY); 1792 if (result == NULL) { 1793 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 1794 ebuf[ebuflen-1] = '\0'; 1795 } 1796 return result; 1797 } 1798 1799 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 1800 int ebuflen) { 1801 void * result = NULL; 1802 if (LoadExecStackDllInVMThread) { 1803 result = dlopen_helper(filename, ebuf, ebuflen); 1804 } 1805 1806 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 1807 // library that requires an executable stack, or which does not have this 1808 // stack attribute set, dlopen changes the stack attribute to executable. The 1809 // read protection of the guard pages gets lost. 1810 // 1811 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 1812 // may have been queued at the same time. 1813 1814 if (!_stack_is_executable) { 1815 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) { 1816 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 1817 jt->stack_guards_enabled()) { // No pending stack overflow exceptions 1818 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) { 1819 warning("Attempt to reguard stack yellow zone failed."); 1820 } 1821 } 1822 } 1823 } 1824 1825 return result; 1826 } 1827 1828 void* os::dll_lookup(void* handle, const char* name) { 1829 void* res = dlsym(handle, name); 1830 return res; 1831 } 1832 1833 void* os::get_default_process_handle() { 1834 return (void*)::dlopen(NULL, RTLD_LAZY); 1835 } 1836 1837 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) { 1838 int fd = ::open(filename, O_RDONLY); 1839 if (fd == -1) { 1840 return false; 1841 } 1842 1843 if (hdr != NULL) { 1844 st->print_cr("%s", hdr); 1845 } 1846 1847 char buf[33]; 1848 int bytes; 1849 buf[32] = '\0'; 1850 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) { 1851 st->print_raw(buf, bytes); 1852 } 1853 1854 ::close(fd); 1855 1856 return true; 1857 } 1858 1859 #if defined(S390) || defined(PPC64) 1860 // keywords_to_match - NULL terminated array of keywords 1861 static bool print_matching_lines_from_file(const char* filename, outputStream* st, const char* keywords_to_match[]) { 1862 char* line = NULL; 1863 size_t length = 0; 1864 FILE* fp = fopen(filename, "r"); 1865 if (fp == NULL) { 1866 return false; 1867 } 1868 1869 st->print_cr("Virtualization information:"); 1870 while (getline(&line, &length, fp) != -1) { 1871 int i = 0; 1872 while (keywords_to_match[i] != NULL) { 1873 if (strncmp(line, keywords_to_match[i], strlen(keywords_to_match[i])) == 0) { 1874 st->print("%s", line); 1875 break; 1876 } 1877 i++; 1878 } 1879 } 1880 1881 free(line); 1882 fclose(fp); 1883 1884 return true; 1885 } 1886 #endif 1887 1888 void os::print_dll_info(outputStream *st) { 1889 st->print_cr("Dynamic libraries:"); 1890 1891 char fname[32]; 1892 pid_t pid = os::Linux::gettid(); 1893 1894 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1895 1896 if (!_print_ascii_file(fname, st)) { 1897 st->print("Can not get library information for pid = %d\n", pid); 1898 } 1899 } 1900 1901 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 1902 FILE *procmapsFile = NULL; 1903 1904 // Open the procfs maps file for the current process 1905 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 1906 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 1907 char line[PATH_MAX + 100]; 1908 1909 // Read line by line from 'file' 1910 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 1911 u8 base, top, offset, inode; 1912 char permissions[5]; 1913 char device[6]; 1914 char name[PATH_MAX + 1]; 1915 1916 // Parse fields from line 1917 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s", 1918 &base, &top, permissions, &offset, device, &inode, name); 1919 1920 // Filter by device id '00:00' so that we only get file system mapped files. 1921 if (strcmp(device, "00:00") != 0) { 1922 1923 // Call callback with the fields of interest 1924 if(callback(name, (address)base, (address)top, param)) { 1925 // Oops abort, callback aborted 1926 fclose(procmapsFile); 1927 return 1; 1928 } 1929 } 1930 } 1931 fclose(procmapsFile); 1932 } 1933 return 0; 1934 } 1935 1936 void os::print_os_info_brief(outputStream* st) { 1937 os::Linux::print_distro_info(st); 1938 1939 os::Posix::print_uname_info(st); 1940 1941 os::Linux::print_libversion_info(st); 1942 1943 } 1944 1945 void os::print_os_info(outputStream* st) { 1946 st->print("OS:"); 1947 1948 os::Linux::print_distro_info(st); 1949 1950 os::Posix::print_uname_info(st); 1951 1952 // Print warning if unsafe chroot environment detected 1953 if (unsafe_chroot_detected) { 1954 st->print("WARNING!! "); 1955 st->print_cr("%s", unstable_chroot_error); 1956 } 1957 1958 os::Linux::print_libversion_info(st); 1959 1960 os::Posix::print_rlimit_info(st); 1961 1962 os::Posix::print_load_average(st); 1963 1964 os::Linux::print_full_memory_info(st); 1965 1966 os::Linux::print_proc_sys_info(st); 1967 1968 os::Linux::print_ld_preload_file(st); 1969 1970 os::Linux::print_container_info(st); 1971 1972 os::Linux::print_virtualization_info(st); 1973 } 1974 1975 // Try to identify popular distros. 1976 // Most Linux distributions have a /etc/XXX-release file, which contains 1977 // the OS version string. Newer Linux distributions have a /etc/lsb-release 1978 // file that also contains the OS version string. Some have more than one 1979 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 1980 // /etc/redhat-release.), so the order is important. 1981 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 1982 // their own specific XXX-release file as well as a redhat-release file. 1983 // Because of this the XXX-release file needs to be searched for before the 1984 // redhat-release file. 1985 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the 1986 // search for redhat-release / SuSE-release needs to be before lsb-release. 1987 // Since the lsb-release file is the new standard it needs to be searched 1988 // before the older style release files. 1989 // Searching system-release (Red Hat) and os-release (other Linuxes) are a 1990 // next to last resort. The os-release file is a new standard that contains 1991 // distribution information and the system-release file seems to be an old 1992 // standard that has been replaced by the lsb-release and os-release files. 1993 // Searching for the debian_version file is the last resort. It contains 1994 // an informative string like "6.0.6" or "wheezy/sid". Because of this 1995 // "Debian " is printed before the contents of the debian_version file. 1996 1997 const char* distro_files[] = { 1998 "/etc/oracle-release", 1999 "/etc/mandriva-release", 2000 "/etc/mandrake-release", 2001 "/etc/sun-release", 2002 "/etc/redhat-release", 2003 "/etc/SuSE-release", 2004 "/etc/lsb-release", 2005 "/etc/turbolinux-release", 2006 "/etc/gentoo-release", 2007 "/etc/ltib-release", 2008 "/etc/angstrom-version", 2009 "/etc/system-release", 2010 "/etc/os-release", 2011 NULL }; 2012 2013 void os::Linux::print_distro_info(outputStream* st) { 2014 for (int i = 0;; i++) { 2015 const char* file = distro_files[i]; 2016 if (file == NULL) { 2017 break; // done 2018 } 2019 // If file prints, we found it. 2020 if (_print_ascii_file(file, st)) { 2021 return; 2022 } 2023 } 2024 2025 if (file_exists("/etc/debian_version")) { 2026 st->print("Debian "); 2027 _print_ascii_file("/etc/debian_version", st); 2028 } else { 2029 st->print("Linux"); 2030 } 2031 st->cr(); 2032 } 2033 2034 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) { 2035 char buf[256]; 2036 while (fgets(buf, sizeof(buf), fp)) { 2037 // Edit out extra stuff in expected format 2038 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) { 2039 char* ptr = strstr(buf, "\""); // the name is in quotes 2040 if (ptr != NULL) { 2041 ptr++; // go beyond first quote 2042 char* nl = strchr(ptr, '\"'); 2043 if (nl != NULL) *nl = '\0'; 2044 strncpy(distro, ptr, length); 2045 } else { 2046 ptr = strstr(buf, "="); 2047 ptr++; // go beyond equals then 2048 char* nl = strchr(ptr, '\n'); 2049 if (nl != NULL) *nl = '\0'; 2050 strncpy(distro, ptr, length); 2051 } 2052 return; 2053 } else if (get_first_line) { 2054 char* nl = strchr(buf, '\n'); 2055 if (nl != NULL) *nl = '\0'; 2056 strncpy(distro, buf, length); 2057 return; 2058 } 2059 } 2060 // print last line and close 2061 char* nl = strchr(buf, '\n'); 2062 if (nl != NULL) *nl = '\0'; 2063 strncpy(distro, buf, length); 2064 } 2065 2066 static void parse_os_info(char* distro, size_t length, const char* file) { 2067 FILE* fp = fopen(file, "r"); 2068 if (fp != NULL) { 2069 // if suse format, print out first line 2070 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0); 2071 parse_os_info_helper(fp, distro, length, get_first_line); 2072 fclose(fp); 2073 } 2074 } 2075 2076 void os::get_summary_os_info(char* buf, size_t buflen) { 2077 for (int i = 0;; i++) { 2078 const char* file = distro_files[i]; 2079 if (file == NULL) { 2080 break; // ran out of distro_files 2081 } 2082 if (file_exists(file)) { 2083 parse_os_info(buf, buflen, file); 2084 return; 2085 } 2086 } 2087 // special case for debian 2088 if (file_exists("/etc/debian_version")) { 2089 strncpy(buf, "Debian ", buflen); 2090 if (buflen > 7) { 2091 parse_os_info(&buf[7], buflen-7, "/etc/debian_version"); 2092 } 2093 } else { 2094 strncpy(buf, "Linux", buflen); 2095 } 2096 } 2097 2098 void os::Linux::print_libversion_info(outputStream* st) { 2099 // libc, pthread 2100 st->print("libc:"); 2101 st->print("%s ", os::Linux::glibc_version()); 2102 st->print("%s ", os::Linux::libpthread_version()); 2103 st->cr(); 2104 } 2105 2106 void os::Linux::print_proc_sys_info(outputStream* st) { 2107 st->cr(); 2108 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):"); 2109 _print_ascii_file("/proc/sys/kernel/threads-max", st); 2110 st->cr(); 2111 st->cr(); 2112 2113 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):"); 2114 _print_ascii_file("/proc/sys/vm/max_map_count", st); 2115 st->cr(); 2116 st->cr(); 2117 2118 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):"); 2119 _print_ascii_file("/proc/sys/kernel/pid_max", st); 2120 st->cr(); 2121 st->cr(); 2122 } 2123 2124 void os::Linux::print_full_memory_info(outputStream* st) { 2125 st->print("\n/proc/meminfo:\n"); 2126 _print_ascii_file("/proc/meminfo", st); 2127 st->cr(); 2128 } 2129 2130 void os::Linux::print_ld_preload_file(outputStream* st) { 2131 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:"); 2132 st->cr(); 2133 } 2134 2135 void os::Linux::print_container_info(outputStream* st) { 2136 if (!OSContainer::is_containerized()) { 2137 return; 2138 } 2139 2140 st->print("container (cgroup) information:\n"); 2141 2142 const char *p_ct = OSContainer::container_type(); 2143 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed"); 2144 2145 char *p = OSContainer::cpu_cpuset_cpus(); 2146 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed"); 2147 free(p); 2148 2149 p = OSContainer::cpu_cpuset_memory_nodes(); 2150 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed"); 2151 free(p); 2152 2153 int i = OSContainer::active_processor_count(); 2154 if (i > 0) { 2155 st->print("active_processor_count: %d\n", i); 2156 } else { 2157 st->print("active_processor_count: failed\n"); 2158 } 2159 2160 i = OSContainer::cpu_quota(); 2161 st->print("cpu_quota: %d\n", i); 2162 2163 i = OSContainer::cpu_period(); 2164 st->print("cpu_period: %d\n", i); 2165 2166 i = OSContainer::cpu_shares(); 2167 st->print("cpu_shares: %d\n", i); 2168 2169 jlong j = OSContainer::memory_limit_in_bytes(); 2170 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j); 2171 2172 j = OSContainer::memory_and_swap_limit_in_bytes(); 2173 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j); 2174 2175 j = OSContainer::memory_soft_limit_in_bytes(); 2176 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j); 2177 2178 j = OSContainer::OSContainer::memory_usage_in_bytes(); 2179 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j); 2180 2181 j = OSContainer::OSContainer::memory_max_usage_in_bytes(); 2182 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j); 2183 st->cr(); 2184 } 2185 2186 void os::Linux::print_virtualization_info(outputStream* st) { 2187 #if defined(S390) 2188 // /proc/sysinfo contains interesting information about 2189 // - LPAR 2190 // - whole "Box" (CPUs ) 2191 // - z/VM / KVM (VM<nn>); this is not available in an LPAR-only setup 2192 const char* kw[] = { "LPAR", "CPUs", "VM", NULL }; 2193 const char* info_file = "/proc/sysinfo"; 2194 2195 if (!print_matching_lines_from_file(info_file, st, kw)) { 2196 st->print_cr(" <%s Not Available>", info_file); 2197 } 2198 #elif defined(PPC64) 2199 const char* info_file = "/proc/ppc64/lparcfg"; 2200 const char* kw[] = { "system_type=", // qemu indicates PowerKVM 2201 "partition_entitled_capacity=", // entitled processor capacity percentage 2202 "partition_max_entitled_capacity=", 2203 "capacity_weight=", // partition CPU weight 2204 "partition_active_processors=", 2205 "partition_potential_processors=", 2206 "entitled_proc_capacity_available=", 2207 "capped=", // 0 - uncapped, 1 - vcpus capped at entitled processor capacity percentage 2208 "shared_processor_mode=", // (non)dedicated partition 2209 "system_potential_processors=", 2210 "pool=", // CPU-pool number 2211 "pool_capacity=", 2212 "NumLpars=", // on non-KVM machines, NumLpars is not found for full partition mode machines 2213 NULL }; 2214 if (!print_matching_lines_from_file(info_file, st, kw)) { 2215 st->print_cr(" <%s Not Available>", info_file); 2216 } 2217 #endif 2218 } 2219 2220 void os::print_memory_info(outputStream* st) { 2221 2222 st->print("Memory:"); 2223 st->print(" %dk page", os::vm_page_size()>>10); 2224 2225 // values in struct sysinfo are "unsigned long" 2226 struct sysinfo si; 2227 sysinfo(&si); 2228 2229 st->print(", physical " UINT64_FORMAT "k", 2230 os::physical_memory() >> 10); 2231 st->print("(" UINT64_FORMAT "k free)", 2232 os::available_memory() >> 10); 2233 st->print(", swap " UINT64_FORMAT "k", 2234 ((jlong)si.totalswap * si.mem_unit) >> 10); 2235 st->print("(" UINT64_FORMAT "k free)", 2236 ((jlong)si.freeswap * si.mem_unit) >> 10); 2237 st->cr(); 2238 } 2239 2240 // Print the first "model name" line and the first "flags" line 2241 // that we find and nothing more. We assume "model name" comes 2242 // before "flags" so if we find a second "model name", then the 2243 // "flags" field is considered missing. 2244 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) { 2245 #if defined(IA32) || defined(AMD64) 2246 // Other platforms have less repetitive cpuinfo files 2247 FILE *fp = fopen("/proc/cpuinfo", "r"); 2248 if (fp) { 2249 while (!feof(fp)) { 2250 if (fgets(buf, buflen, fp)) { 2251 // Assume model name comes before flags 2252 bool model_name_printed = false; 2253 if (strstr(buf, "model name") != NULL) { 2254 if (!model_name_printed) { 2255 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n"); 2256 st->print_raw(buf); 2257 model_name_printed = true; 2258 } else { 2259 // model name printed but not flags? Odd, just return 2260 fclose(fp); 2261 return true; 2262 } 2263 } 2264 // print the flags line too 2265 if (strstr(buf, "flags") != NULL) { 2266 st->print_raw(buf); 2267 fclose(fp); 2268 return true; 2269 } 2270 } 2271 } 2272 fclose(fp); 2273 } 2274 #endif // x86 platforms 2275 return false; 2276 } 2277 2278 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) { 2279 // Only print the model name if the platform provides this as a summary 2280 if (!print_model_name_and_flags(st, buf, buflen)) { 2281 st->print("\n/proc/cpuinfo:\n"); 2282 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2283 st->print_cr(" <Not Available>"); 2284 } 2285 } 2286 } 2287 2288 #if defined(AMD64) || defined(IA32) || defined(X32) 2289 const char* search_string = "model name"; 2290 #elif defined(M68K) 2291 const char* search_string = "CPU"; 2292 #elif defined(PPC64) 2293 const char* search_string = "cpu"; 2294 #elif defined(S390) 2295 const char* search_string = "processor"; 2296 #elif defined(SPARC) 2297 const char* search_string = "cpu"; 2298 #else 2299 const char* search_string = "Processor"; 2300 #endif 2301 2302 // Parses the cpuinfo file for string representing the model name. 2303 void os::get_summary_cpu_info(char* cpuinfo, size_t length) { 2304 FILE* fp = fopen("/proc/cpuinfo", "r"); 2305 if (fp != NULL) { 2306 while (!feof(fp)) { 2307 char buf[256]; 2308 if (fgets(buf, sizeof(buf), fp)) { 2309 char* start = strstr(buf, search_string); 2310 if (start != NULL) { 2311 char *ptr = start + strlen(search_string); 2312 char *end = buf + strlen(buf); 2313 while (ptr != end) { 2314 // skip whitespace and colon for the rest of the name. 2315 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') { 2316 break; 2317 } 2318 ptr++; 2319 } 2320 if (ptr != end) { 2321 // reasonable string, get rid of newline and keep the rest 2322 char* nl = strchr(buf, '\n'); 2323 if (nl != NULL) *nl = '\0'; 2324 strncpy(cpuinfo, ptr, length); 2325 fclose(fp); 2326 return; 2327 } 2328 } 2329 } 2330 } 2331 fclose(fp); 2332 } 2333 // cpuinfo not found or parsing failed, just print generic string. The entire 2334 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86) 2335 #if defined(AARCH64) 2336 strncpy(cpuinfo, "AArch64", length); 2337 #elif defined(AMD64) 2338 strncpy(cpuinfo, "x86_64", length); 2339 #elif defined(ARM) // Order wrt. AARCH64 is relevant! 2340 strncpy(cpuinfo, "ARM", length); 2341 #elif defined(IA32) 2342 strncpy(cpuinfo, "x86_32", length); 2343 #elif defined(IA64) 2344 strncpy(cpuinfo, "IA64", length); 2345 #elif defined(PPC) 2346 strncpy(cpuinfo, "PPC64", length); 2347 #elif defined(S390) 2348 strncpy(cpuinfo, "S390", length); 2349 #elif defined(SPARC) 2350 strncpy(cpuinfo, "sparcv9", length); 2351 #elif defined(ZERO_LIBARCH) 2352 strncpy(cpuinfo, ZERO_LIBARCH, length); 2353 #else 2354 strncpy(cpuinfo, "unknown", length); 2355 #endif 2356 } 2357 2358 static void print_signal_handler(outputStream* st, int sig, 2359 char* buf, size_t buflen); 2360 2361 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2362 st->print_cr("Signal Handlers:"); 2363 print_signal_handler(st, SIGSEGV, buf, buflen); 2364 print_signal_handler(st, SIGBUS , buf, buflen); 2365 print_signal_handler(st, SIGFPE , buf, buflen); 2366 print_signal_handler(st, SIGPIPE, buf, buflen); 2367 print_signal_handler(st, SIGXFSZ, buf, buflen); 2368 print_signal_handler(st, SIGILL , buf, buflen); 2369 print_signal_handler(st, SR_signum, buf, buflen); 2370 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2371 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2372 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2373 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2374 #if defined(PPC64) 2375 print_signal_handler(st, SIGTRAP, buf, buflen); 2376 #endif 2377 } 2378 2379 static char saved_jvm_path[MAXPATHLEN] = {0}; 2380 2381 // Find the full path to the current module, libjvm.so 2382 void os::jvm_path(char *buf, jint buflen) { 2383 // Error checking. 2384 if (buflen < MAXPATHLEN) { 2385 assert(false, "must use a large-enough buffer"); 2386 buf[0] = '\0'; 2387 return; 2388 } 2389 // Lazy resolve the path to current module. 2390 if (saved_jvm_path[0] != 0) { 2391 strcpy(buf, saved_jvm_path); 2392 return; 2393 } 2394 2395 char dli_fname[MAXPATHLEN]; 2396 bool ret = dll_address_to_library_name( 2397 CAST_FROM_FN_PTR(address, os::jvm_path), 2398 dli_fname, sizeof(dli_fname), NULL); 2399 assert(ret, "cannot locate libjvm"); 2400 char *rp = NULL; 2401 if (ret && dli_fname[0] != '\0') { 2402 rp = os::Posix::realpath(dli_fname, buf, buflen); 2403 } 2404 if (rp == NULL) { 2405 return; 2406 } 2407 2408 if (Arguments::sun_java_launcher_is_altjvm()) { 2409 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2410 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so". 2411 // If "/jre/lib/" appears at the right place in the string, then 2412 // assume we are installed in a JDK and we're done. Otherwise, check 2413 // for a JAVA_HOME environment variable and fix up the path so it 2414 // looks like libjvm.so is installed there (append a fake suffix 2415 // hotspot/libjvm.so). 2416 const char *p = buf + strlen(buf) - 1; 2417 for (int count = 0; p > buf && count < 5; ++count) { 2418 for (--p; p > buf && *p != '/'; --p) 2419 /* empty */ ; 2420 } 2421 2422 if (strncmp(p, "/jre/lib/", 9) != 0) { 2423 // Look for JAVA_HOME in the environment. 2424 char* java_home_var = ::getenv("JAVA_HOME"); 2425 if (java_home_var != NULL && java_home_var[0] != 0) { 2426 char* jrelib_p; 2427 int len; 2428 2429 // Check the current module name "libjvm.so". 2430 p = strrchr(buf, '/'); 2431 if (p == NULL) { 2432 return; 2433 } 2434 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2435 2436 rp = os::Posix::realpath(java_home_var, buf, buflen); 2437 if (rp == NULL) { 2438 return; 2439 } 2440 2441 // determine if this is a legacy image or modules image 2442 // modules image doesn't have "jre" subdirectory 2443 len = strlen(buf); 2444 assert(len < buflen, "Ran out of buffer room"); 2445 jrelib_p = buf + len; 2446 snprintf(jrelib_p, buflen-len, "/jre/lib"); 2447 if (0 != access(buf, F_OK)) { 2448 snprintf(jrelib_p, buflen-len, "/lib"); 2449 } 2450 2451 if (0 == access(buf, F_OK)) { 2452 // Use current module name "libjvm.so" 2453 len = strlen(buf); 2454 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2455 } else { 2456 // Go back to path of .so 2457 rp = os::Posix::realpath(dli_fname, buf, buflen); 2458 if (rp == NULL) { 2459 return; 2460 } 2461 } 2462 } 2463 } 2464 } 2465 2466 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2467 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2468 } 2469 2470 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2471 // no prefix required, not even "_" 2472 } 2473 2474 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2475 // no suffix required 2476 } 2477 2478 //////////////////////////////////////////////////////////////////////////////// 2479 // sun.misc.Signal support 2480 2481 static volatile jint sigint_count = 0; 2482 2483 static void UserHandler(int sig, void *siginfo, void *context) { 2484 // 4511530 - sem_post is serialized and handled by the manager thread. When 2485 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2486 // don't want to flood the manager thread with sem_post requests. 2487 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2488 return; 2489 } 2490 2491 // Ctrl-C is pressed during error reporting, likely because the error 2492 // handler fails to abort. Let VM die immediately. 2493 if (sig == SIGINT && VMError::is_error_reported()) { 2494 os::die(); 2495 } 2496 2497 os::signal_notify(sig); 2498 } 2499 2500 void* os::user_handler() { 2501 return CAST_FROM_FN_PTR(void*, UserHandler); 2502 } 2503 2504 extern "C" { 2505 typedef void (*sa_handler_t)(int); 2506 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2507 } 2508 2509 void* os::signal(int signal_number, void* handler) { 2510 struct sigaction sigAct, oldSigAct; 2511 2512 sigfillset(&(sigAct.sa_mask)); 2513 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2514 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2515 2516 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2517 // -1 means registration failed 2518 return (void *)-1; 2519 } 2520 2521 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2522 } 2523 2524 void os::signal_raise(int signal_number) { 2525 ::raise(signal_number); 2526 } 2527 2528 // The following code is moved from os.cpp for making this 2529 // code platform specific, which it is by its very nature. 2530 2531 // Will be modified when max signal is changed to be dynamic 2532 int os::sigexitnum_pd() { 2533 return NSIG; 2534 } 2535 2536 // a counter for each possible signal value 2537 static volatile jint pending_signals[NSIG+1] = { 0 }; 2538 2539 // Linux(POSIX) specific hand shaking semaphore. 2540 static Semaphore* sig_sem = NULL; 2541 static PosixSemaphore sr_semaphore; 2542 2543 static void jdk_misc_signal_init() { 2544 // Initialize signal structures 2545 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2546 2547 // Initialize signal semaphore 2548 sig_sem = new Semaphore(); 2549 } 2550 2551 void os::signal_notify(int sig) { 2552 if (sig_sem != NULL) { 2553 Atomic::inc(&pending_signals[sig]); 2554 sig_sem->signal(); 2555 } else { 2556 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init 2557 // initialization isn't called. 2558 assert(ReduceSignalUsage, "signal semaphore should be created"); 2559 } 2560 } 2561 2562 static int check_pending_signals() { 2563 Atomic::store(0, &sigint_count); 2564 for (;;) { 2565 for (int i = 0; i < NSIG + 1; i++) { 2566 jint n = pending_signals[i]; 2567 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2568 return i; 2569 } 2570 } 2571 JavaThread *thread = JavaThread::current(); 2572 ThreadBlockInVM tbivm(thread); 2573 2574 bool threadIsSuspended; 2575 do { 2576 thread->set_suspend_equivalent(); 2577 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2578 sig_sem->wait(); 2579 2580 // were we externally suspended while we were waiting? 2581 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2582 if (threadIsSuspended) { 2583 // The semaphore has been incremented, but while we were waiting 2584 // another thread suspended us. We don't want to continue running 2585 // while suspended because that would surprise the thread that 2586 // suspended us. 2587 sig_sem->signal(); 2588 2589 thread->java_suspend_self(); 2590 } 2591 } while (threadIsSuspended); 2592 } 2593 } 2594 2595 int os::signal_wait() { 2596 return check_pending_signals(); 2597 } 2598 2599 //////////////////////////////////////////////////////////////////////////////// 2600 // Virtual Memory 2601 2602 int os::vm_page_size() { 2603 // Seems redundant as all get out 2604 assert(os::Linux::page_size() != -1, "must call os::init"); 2605 return os::Linux::page_size(); 2606 } 2607 2608 // Solaris allocates memory by pages. 2609 int os::vm_allocation_granularity() { 2610 assert(os::Linux::page_size() != -1, "must call os::init"); 2611 return os::Linux::page_size(); 2612 } 2613 2614 // Rationale behind this function: 2615 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2616 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2617 // samples for JITted code. Here we create private executable mapping over the code cache 2618 // and then we can use standard (well, almost, as mapping can change) way to provide 2619 // info for the reporting script by storing timestamp and location of symbol 2620 void linux_wrap_code(char* base, size_t size) { 2621 static volatile jint cnt = 0; 2622 2623 if (!UseOprofile) { 2624 return; 2625 } 2626 2627 char buf[PATH_MAX+1]; 2628 int num = Atomic::add(1, &cnt); 2629 2630 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2631 os::get_temp_directory(), os::current_process_id(), num); 2632 unlink(buf); 2633 2634 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2635 2636 if (fd != -1) { 2637 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2638 if (rv != (off_t)-1) { 2639 if (::write(fd, "", 1) == 1) { 2640 mmap(base, size, 2641 PROT_READ|PROT_WRITE|PROT_EXEC, 2642 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2643 } 2644 } 2645 ::close(fd); 2646 unlink(buf); 2647 } 2648 } 2649 2650 static bool recoverable_mmap_error(int err) { 2651 // See if the error is one we can let the caller handle. This 2652 // list of errno values comes from JBS-6843484. I can't find a 2653 // Linux man page that documents this specific set of errno 2654 // values so while this list currently matches Solaris, it may 2655 // change as we gain experience with this failure mode. 2656 switch (err) { 2657 case EBADF: 2658 case EINVAL: 2659 case ENOTSUP: 2660 // let the caller deal with these errors 2661 return true; 2662 2663 default: 2664 // Any remaining errors on this OS can cause our reserved mapping 2665 // to be lost. That can cause confusion where different data 2666 // structures think they have the same memory mapped. The worst 2667 // scenario is if both the VM and a library think they have the 2668 // same memory mapped. 2669 return false; 2670 } 2671 } 2672 2673 static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2674 int err) { 2675 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2676 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec, 2677 os::strerror(err), err); 2678 } 2679 2680 static void warn_fail_commit_memory(char* addr, size_t size, 2681 size_t alignment_hint, bool exec, 2682 int err) { 2683 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2684 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, 2685 alignment_hint, exec, os::strerror(err), err); 2686 } 2687 2688 // NOTE: Linux kernel does not really reserve the pages for us. 2689 // All it does is to check if there are enough free pages 2690 // left at the time of mmap(). This could be a potential 2691 // problem. 2692 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2693 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2694 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2695 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2696 if (res != (uintptr_t) MAP_FAILED) { 2697 if (UseNUMAInterleaving) { 2698 numa_make_global(addr, size); 2699 } 2700 return 0; 2701 } 2702 2703 int err = errno; // save errno from mmap() call above 2704 2705 if (!recoverable_mmap_error(err)) { 2706 warn_fail_commit_memory(addr, size, exec, err); 2707 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2708 } 2709 2710 return err; 2711 } 2712 2713 bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2714 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2715 } 2716 2717 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2718 const char* mesg) { 2719 assert(mesg != NULL, "mesg must be specified"); 2720 int err = os::Linux::commit_memory_impl(addr, size, exec); 2721 if (err != 0) { 2722 // the caller wants all commit errors to exit with the specified mesg: 2723 warn_fail_commit_memory(addr, size, exec, err); 2724 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2725 } 2726 } 2727 2728 // Define MAP_HUGETLB here so we can build HotSpot on old systems. 2729 #ifndef MAP_HUGETLB 2730 #define MAP_HUGETLB 0x40000 2731 #endif 2732 2733 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2734 #ifndef MADV_HUGEPAGE 2735 #define MADV_HUGEPAGE 14 2736 #endif 2737 2738 int os::Linux::commit_memory_impl(char* addr, size_t size, 2739 size_t alignment_hint, bool exec) { 2740 int err = os::Linux::commit_memory_impl(addr, size, exec); 2741 if (err == 0) { 2742 realign_memory(addr, size, alignment_hint); 2743 } 2744 return err; 2745 } 2746 2747 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2748 bool exec) { 2749 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2750 } 2751 2752 void os::pd_commit_memory_or_exit(char* addr, size_t size, 2753 size_t alignment_hint, bool exec, 2754 const char* mesg) { 2755 assert(mesg != NULL, "mesg must be specified"); 2756 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2757 if (err != 0) { 2758 // the caller wants all commit errors to exit with the specified mesg: 2759 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2760 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2761 } 2762 } 2763 2764 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2765 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2766 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2767 // be supported or the memory may already be backed by huge pages. 2768 ::madvise(addr, bytes, MADV_HUGEPAGE); 2769 } 2770 } 2771 2772 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2773 // This method works by doing an mmap over an existing mmaping and effectively discarding 2774 // the existing pages. However it won't work for SHM-based large pages that cannot be 2775 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2776 // small pages on top of the SHM segment. This method always works for small pages, so we 2777 // allow that in any case. 2778 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2779 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2780 } 2781 } 2782 2783 void os::numa_make_global(char *addr, size_t bytes) { 2784 Linux::numa_interleave_memory(addr, bytes); 2785 } 2786 2787 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2788 // bind policy to MPOL_PREFERRED for the current thread. 2789 #define USE_MPOL_PREFERRED 0 2790 2791 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2792 // To make NUMA and large pages more robust when both enabled, we need to ease 2793 // the requirements on where the memory should be allocated. MPOL_BIND is the 2794 // default policy and it will force memory to be allocated on the specified 2795 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2796 // the specified node, but will not force it. Using this policy will prevent 2797 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2798 // free large pages. 2799 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2800 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2801 } 2802 2803 bool os::numa_topology_changed() { return false; } 2804 2805 size_t os::numa_get_groups_num() { 2806 // Return just the number of nodes in which it's possible to allocate memory 2807 // (in numa terminology, configured nodes). 2808 return Linux::numa_num_configured_nodes(); 2809 } 2810 2811 int os::numa_get_group_id() { 2812 int cpu_id = Linux::sched_getcpu(); 2813 if (cpu_id != -1) { 2814 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2815 if (lgrp_id != -1) { 2816 return lgrp_id; 2817 } 2818 } 2819 return 0; 2820 } 2821 2822 int os::Linux::get_existing_num_nodes() { 2823 int node; 2824 int highest_node_number = Linux::numa_max_node(); 2825 int num_nodes = 0; 2826 2827 // Get the total number of nodes in the system including nodes without memory. 2828 for (node = 0; node <= highest_node_number; node++) { 2829 if (is_node_in_existing_nodes(node)) { 2830 num_nodes++; 2831 } 2832 } 2833 return num_nodes; 2834 } 2835 2836 size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2837 int highest_node_number = Linux::numa_max_node(); 2838 size_t i = 0; 2839 2840 // Map all node ids in which it is possible to allocate memory. Also nodes are 2841 // not always consecutively available, i.e. available from 0 to the highest 2842 // node number. If the nodes have been bound explicitly using numactl membind, 2843 // then allocate memory from those nodes only. 2844 for (int node = 0; node <= highest_node_number; node++) { 2845 if (Linux::is_node_in_bound_nodes((unsigned int)node)) { 2846 ids[i++] = node; 2847 } 2848 } 2849 return i; 2850 } 2851 2852 bool os::get_page_info(char *start, page_info* info) { 2853 return false; 2854 } 2855 2856 char *os::scan_pages(char *start, char* end, page_info* page_expected, 2857 page_info* page_found) { 2858 return end; 2859 } 2860 2861 2862 int os::Linux::sched_getcpu_syscall(void) { 2863 unsigned int cpu = 0; 2864 int retval = -1; 2865 2866 #if defined(IA32) 2867 #ifndef SYS_getcpu 2868 #define SYS_getcpu 318 2869 #endif 2870 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2871 #elif defined(AMD64) 2872 // Unfortunately we have to bring all these macros here from vsyscall.h 2873 // to be able to compile on old linuxes. 2874 #define __NR_vgetcpu 2 2875 #define VSYSCALL_START (-10UL << 20) 2876 #define VSYSCALL_SIZE 1024 2877 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2878 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2879 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2880 retval = vgetcpu(&cpu, NULL, NULL); 2881 #endif 2882 2883 return (retval == -1) ? retval : cpu; 2884 } 2885 2886 void os::Linux::sched_getcpu_init() { 2887 // sched_getcpu() should be in libc. 2888 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2889 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2890 2891 // If it's not, try a direct syscall. 2892 if (sched_getcpu() == -1) { 2893 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2894 (void*)&sched_getcpu_syscall)); 2895 } 2896 2897 if (sched_getcpu() == -1) { 2898 vm_exit_during_initialization("getcpu(2) system call not supported by kernel"); 2899 } 2900 } 2901 2902 // Something to do with the numa-aware allocator needs these symbols 2903 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2904 extern "C" JNIEXPORT void numa_error(char *where) { } 2905 2906 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails 2907 // load symbol from base version instead. 2908 void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2909 void *f = dlvsym(handle, name, "libnuma_1.1"); 2910 if (f == NULL) { 2911 f = dlsym(handle, name); 2912 } 2913 return f; 2914 } 2915 2916 // Handle request to load libnuma symbol version 1.2 (API v2) only. 2917 // Return NULL if the symbol is not defined in this particular version. 2918 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) { 2919 return dlvsym(handle, name, "libnuma_1.2"); 2920 } 2921 2922 bool os::Linux::libnuma_init() { 2923 if (sched_getcpu() != -1) { // Requires sched_getcpu() support 2924 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2925 if (handle != NULL) { 2926 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2927 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2928 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2929 libnuma_dlsym(handle, "numa_max_node"))); 2930 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t, 2931 libnuma_dlsym(handle, "numa_num_configured_nodes"))); 2932 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2933 libnuma_dlsym(handle, "numa_available"))); 2934 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2935 libnuma_dlsym(handle, "numa_tonode_memory"))); 2936 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2937 libnuma_dlsym(handle, "numa_interleave_memory"))); 2938 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t, 2939 libnuma_v2_dlsym(handle, "numa_interleave_memory"))); 2940 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2941 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2942 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t, 2943 libnuma_dlsym(handle, "numa_bitmask_isbitset"))); 2944 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t, 2945 libnuma_dlsym(handle, "numa_distance"))); 2946 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t, 2947 libnuma_v2_dlsym(handle, "numa_get_membind"))); 2948 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t, 2949 libnuma_v2_dlsym(handle, "numa_get_interleave_mask"))); 2950 2951 if (numa_available() != -1) { 2952 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2953 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr")); 2954 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr")); 2955 set_numa_interleave_bitmask(_numa_get_interleave_mask()); 2956 set_numa_membind_bitmask(_numa_get_membind()); 2957 // Create an index -> node mapping, since nodes are not always consecutive 2958 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2959 rebuild_nindex_to_node_map(); 2960 // Create a cpu -> node mapping 2961 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2962 rebuild_cpu_to_node_map(); 2963 return true; 2964 } 2965 } 2966 } 2967 return false; 2968 } 2969 2970 size_t os::Linux::default_guard_size(os::ThreadType thr_type) { 2971 // Creating guard page is very expensive. Java thread has HotSpot 2972 // guard pages, only enable glibc guard page for non-Java threads. 2973 // (Remember: compiler thread is a Java thread, too!) 2974 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size()); 2975 } 2976 2977 void os::Linux::rebuild_nindex_to_node_map() { 2978 int highest_node_number = Linux::numa_max_node(); 2979 2980 nindex_to_node()->clear(); 2981 for (int node = 0; node <= highest_node_number; node++) { 2982 if (Linux::is_node_in_existing_nodes(node)) { 2983 nindex_to_node()->append(node); 2984 } 2985 } 2986 } 2987 2988 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2989 // The table is later used in get_node_by_cpu(). 2990 void os::Linux::rebuild_cpu_to_node_map() { 2991 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2992 // in libnuma (possible values are starting from 16, 2993 // and continuing up with every other power of 2, but less 2994 // than the maximum number of CPUs supported by kernel), and 2995 // is a subject to change (in libnuma version 2 the requirements 2996 // are more reasonable) we'll just hardcode the number they use 2997 // in the library. 2998 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2999 3000 size_t cpu_num = processor_count(); 3001 size_t cpu_map_size = NCPUS / BitsPerCLong; 3002 size_t cpu_map_valid_size = 3003 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 3004 3005 cpu_to_node()->clear(); 3006 cpu_to_node()->at_grow(cpu_num - 1); 3007 3008 size_t node_num = get_existing_num_nodes(); 3009 3010 int distance = 0; 3011 int closest_distance = INT_MAX; 3012 int closest_node = 0; 3013 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 3014 for (size_t i = 0; i < node_num; i++) { 3015 // Check if node is configured (not a memory-less node). If it is not, find 3016 // the closest configured node. Check also if node is bound, i.e. it's allowed 3017 // to allocate memory from the node. If it's not allowed, map cpus in that node 3018 // to the closest node from which memory allocation is allowed. 3019 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) || 3020 !is_node_in_bound_nodes(nindex_to_node()->at(i))) { 3021 closest_distance = INT_MAX; 3022 // Check distance from all remaining nodes in the system. Ignore distance 3023 // from itself, from another non-configured node, and from another non-bound 3024 // node. 3025 for (size_t m = 0; m < node_num; m++) { 3026 if (m != i && 3027 is_node_in_configured_nodes(nindex_to_node()->at(m)) && 3028 is_node_in_bound_nodes(nindex_to_node()->at(m))) { 3029 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m)); 3030 // If a closest node is found, update. There is always at least one 3031 // configured and bound node in the system so there is always at least 3032 // one node close. 3033 if (distance != 0 && distance < closest_distance) { 3034 closest_distance = distance; 3035 closest_node = nindex_to_node()->at(m); 3036 } 3037 } 3038 } 3039 } else { 3040 // Current node is already a configured node. 3041 closest_node = nindex_to_node()->at(i); 3042 } 3043 3044 // Get cpus from the original node and map them to the closest node. If node 3045 // is a configured node (not a memory-less node), then original node and 3046 // closest node are the same. 3047 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 3048 for (size_t j = 0; j < cpu_map_valid_size; j++) { 3049 if (cpu_map[j] != 0) { 3050 for (size_t k = 0; k < BitsPerCLong; k++) { 3051 if (cpu_map[j] & (1UL << k)) { 3052 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node); 3053 } 3054 } 3055 } 3056 } 3057 } 3058 } 3059 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 3060 } 3061 3062 int os::Linux::get_node_by_cpu(int cpu_id) { 3063 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 3064 return cpu_to_node()->at(cpu_id); 3065 } 3066 return -1; 3067 } 3068 3069 GrowableArray<int>* os::Linux::_cpu_to_node; 3070 GrowableArray<int>* os::Linux::_nindex_to_node; 3071 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 3072 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 3073 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 3074 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes; 3075 os::Linux::numa_available_func_t os::Linux::_numa_available; 3076 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 3077 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 3078 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2; 3079 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 3080 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset; 3081 os::Linux::numa_distance_func_t os::Linux::_numa_distance; 3082 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind; 3083 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask; 3084 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy; 3085 unsigned long* os::Linux::_numa_all_nodes; 3086 struct bitmask* os::Linux::_numa_all_nodes_ptr; 3087 struct bitmask* os::Linux::_numa_nodes_ptr; 3088 struct bitmask* os::Linux::_numa_interleave_bitmask; 3089 struct bitmask* os::Linux::_numa_membind_bitmask; 3090 3091 bool os::pd_uncommit_memory(char* addr, size_t size) { 3092 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 3093 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 3094 return res != (uintptr_t) MAP_FAILED; 3095 } 3096 3097 static address get_stack_commited_bottom(address bottom, size_t size) { 3098 address nbot = bottom; 3099 address ntop = bottom + size; 3100 3101 size_t page_sz = os::vm_page_size(); 3102 unsigned pages = size / page_sz; 3103 3104 unsigned char vec[1]; 3105 unsigned imin = 1, imax = pages + 1, imid; 3106 int mincore_return_value = 0; 3107 3108 assert(imin <= imax, "Unexpected page size"); 3109 3110 while (imin < imax) { 3111 imid = (imax + imin) / 2; 3112 nbot = ntop - (imid * page_sz); 3113 3114 // Use a trick with mincore to check whether the page is mapped or not. 3115 // mincore sets vec to 1 if page resides in memory and to 0 if page 3116 // is swapped output but if page we are asking for is unmapped 3117 // it returns -1,ENOMEM 3118 mincore_return_value = mincore(nbot, page_sz, vec); 3119 3120 if (mincore_return_value == -1) { 3121 // Page is not mapped go up 3122 // to find first mapped page 3123 if (errno != EAGAIN) { 3124 assert(errno == ENOMEM, "Unexpected mincore errno"); 3125 imax = imid; 3126 } 3127 } else { 3128 // Page is mapped go down 3129 // to find first not mapped page 3130 imin = imid + 1; 3131 } 3132 } 3133 3134 nbot = nbot + page_sz; 3135 3136 // Adjust stack bottom one page up if last checked page is not mapped 3137 if (mincore_return_value == -1) { 3138 nbot = nbot + page_sz; 3139 } 3140 3141 return nbot; 3142 } 3143 3144 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) { 3145 int mincore_return_value; 3146 const size_t stripe = 1024; // query this many pages each time 3147 unsigned char vec[stripe + 1]; 3148 // set a guard 3149 vec[stripe] = 'X'; 3150 3151 const size_t page_sz = os::vm_page_size(); 3152 size_t pages = size / page_sz; 3153 3154 assert(is_aligned(start, page_sz), "Start address must be page aligned"); 3155 assert(is_aligned(size, page_sz), "Size must be page aligned"); 3156 3157 committed_start = NULL; 3158 3159 int loops = (pages + stripe - 1) / stripe; 3160 int committed_pages = 0; 3161 address loop_base = start; 3162 bool found_range = false; 3163 3164 for (int index = 0; index < loops && !found_range; index ++) { 3165 assert(pages > 0, "Nothing to do"); 3166 int pages_to_query = (pages >= stripe) ? stripe : pages; 3167 pages -= pages_to_query; 3168 3169 // Get stable read 3170 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN); 3171 3172 // During shutdown, some memory goes away without properly notifying NMT, 3173 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object. 3174 // Bailout and return as not committed for now. 3175 if (mincore_return_value == -1 && errno == ENOMEM) { 3176 return false; 3177 } 3178 3179 assert(vec[stripe] == 'X', "overflow guard"); 3180 assert(mincore_return_value == 0, "Range must be valid"); 3181 // Process this stripe 3182 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) { 3183 if ((vec[vecIdx] & 0x01) == 0) { // not committed 3184 // End of current contiguous region 3185 if (committed_start != NULL) { 3186 found_range = true; 3187 break; 3188 } 3189 } else { // committed 3190 // Start of region 3191 if (committed_start == NULL) { 3192 committed_start = loop_base + page_sz * vecIdx; 3193 } 3194 committed_pages ++; 3195 } 3196 } 3197 3198 loop_base += pages_to_query * page_sz; 3199 } 3200 3201 if (committed_start != NULL) { 3202 assert(committed_pages > 0, "Must have committed region"); 3203 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has"); 3204 assert(committed_start >= start && committed_start < start + size, "Out of range"); 3205 committed_size = page_sz * committed_pages; 3206 return true; 3207 } else { 3208 assert(committed_pages == 0, "Should not have committed region"); 3209 return false; 3210 } 3211 } 3212 3213 3214 // Linux uses a growable mapping for the stack, and if the mapping for 3215 // the stack guard pages is not removed when we detach a thread the 3216 // stack cannot grow beyond the pages where the stack guard was 3217 // mapped. If at some point later in the process the stack expands to 3218 // that point, the Linux kernel cannot expand the stack any further 3219 // because the guard pages are in the way, and a segfault occurs. 3220 // 3221 // However, it's essential not to split the stack region by unmapping 3222 // a region (leaving a hole) that's already part of the stack mapping, 3223 // so if the stack mapping has already grown beyond the guard pages at 3224 // the time we create them, we have to truncate the stack mapping. 3225 // So, we need to know the extent of the stack mapping when 3226 // create_stack_guard_pages() is called. 3227 3228 // We only need this for stacks that are growable: at the time of 3229 // writing thread stacks don't use growable mappings (i.e. those 3230 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3231 // only applies to the main thread. 3232 3233 // If the (growable) stack mapping already extends beyond the point 3234 // where we're going to put our guard pages, truncate the mapping at 3235 // that point by munmap()ping it. This ensures that when we later 3236 // munmap() the guard pages we don't leave a hole in the stack 3237 // mapping. This only affects the main/primordial thread 3238 3239 bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3240 if (os::is_primordial_thread()) { 3241 // As we manually grow stack up to bottom inside create_attached_thread(), 3242 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3243 // we don't need to do anything special. 3244 // Check it first, before calling heavy function. 3245 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3246 unsigned char vec[1]; 3247 3248 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3249 // Fallback to slow path on all errors, including EAGAIN 3250 stack_extent = (uintptr_t) get_stack_commited_bottom( 3251 os::Linux::initial_thread_stack_bottom(), 3252 (size_t)addr - stack_extent); 3253 } 3254 3255 if (stack_extent < (uintptr_t)addr) { 3256 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3257 } 3258 } 3259 3260 return os::commit_memory(addr, size, !ExecMem); 3261 } 3262 3263 // If this is a growable mapping, remove the guard pages entirely by 3264 // munmap()ping them. If not, just call uncommit_memory(). This only 3265 // affects the main/primordial thread, but guard against future OS changes. 3266 // It's safe to always unmap guard pages for primordial thread because we 3267 // always place it right after end of the mapped region. 3268 3269 bool os::remove_stack_guard_pages(char* addr, size_t size) { 3270 uintptr_t stack_extent, stack_base; 3271 3272 if (os::is_primordial_thread()) { 3273 return ::munmap(addr, size) == 0; 3274 } 3275 3276 return os::uncommit_memory(addr, size); 3277 } 3278 3279 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3280 // at 'requested_addr'. If there are existing memory mappings at the same 3281 // location, however, they will be overwritten. If 'fixed' is false, 3282 // 'requested_addr' is only treated as a hint, the return value may or 3283 // may not start from the requested address. Unlike Linux mmap(), this 3284 // function returns NULL to indicate failure. 3285 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3286 char * addr; 3287 int flags; 3288 3289 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3290 if (fixed) { 3291 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3292 flags |= MAP_FIXED; 3293 } 3294 3295 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3296 // touch an uncommitted page. Otherwise, the read/write might 3297 // succeed if we have enough swap space to back the physical page. 3298 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3299 flags, -1, 0); 3300 3301 return addr == MAP_FAILED ? NULL : addr; 3302 } 3303 3304 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address 3305 // (req_addr != NULL) or with a given alignment. 3306 // - bytes shall be a multiple of alignment. 3307 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3308 // - alignment sets the alignment at which memory shall be allocated. 3309 // It must be a multiple of allocation granularity. 3310 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3311 // req_addr or NULL. 3312 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) { 3313 3314 size_t extra_size = bytes; 3315 if (req_addr == NULL && alignment > 0) { 3316 extra_size += alignment; 3317 } 3318 3319 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE, 3320 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 3321 -1, 0); 3322 if (start == MAP_FAILED) { 3323 start = NULL; 3324 } else { 3325 if (req_addr != NULL) { 3326 if (start != req_addr) { 3327 ::munmap(start, extra_size); 3328 start = NULL; 3329 } 3330 } else { 3331 char* const start_aligned = align_up(start, alignment); 3332 char* const end_aligned = start_aligned + bytes; 3333 char* const end = start + extra_size; 3334 if (start_aligned > start) { 3335 ::munmap(start, start_aligned - start); 3336 } 3337 if (end_aligned < end) { 3338 ::munmap(end_aligned, end - end_aligned); 3339 } 3340 start = start_aligned; 3341 } 3342 } 3343 return start; 3344 } 3345 3346 static int anon_munmap(char * addr, size_t size) { 3347 return ::munmap(addr, size) == 0; 3348 } 3349 3350 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3351 size_t alignment_hint) { 3352 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3353 } 3354 3355 bool os::pd_release_memory(char* addr, size_t size) { 3356 return anon_munmap(addr, size); 3357 } 3358 3359 static bool linux_mprotect(char* addr, size_t size, int prot) { 3360 // Linux wants the mprotect address argument to be page aligned. 3361 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size()); 3362 3363 // According to SUSv3, mprotect() should only be used with mappings 3364 // established by mmap(), and mmap() always maps whole pages. Unaligned 3365 // 'addr' likely indicates problem in the VM (e.g. trying to change 3366 // protection of malloc'ed or statically allocated memory). Check the 3367 // caller if you hit this assert. 3368 assert(addr == bottom, "sanity check"); 3369 3370 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3371 return ::mprotect(bottom, size, prot) == 0; 3372 } 3373 3374 // Set protections specified 3375 bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3376 bool is_committed) { 3377 unsigned int p = 0; 3378 switch (prot) { 3379 case MEM_PROT_NONE: p = PROT_NONE; break; 3380 case MEM_PROT_READ: p = PROT_READ; break; 3381 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3382 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3383 default: 3384 ShouldNotReachHere(); 3385 } 3386 // is_committed is unused. 3387 return linux_mprotect(addr, bytes, p); 3388 } 3389 3390 bool os::guard_memory(char* addr, size_t size) { 3391 return linux_mprotect(addr, size, PROT_NONE); 3392 } 3393 3394 bool os::unguard_memory(char* addr, size_t size) { 3395 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3396 } 3397 3398 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3399 size_t page_size) { 3400 bool result = false; 3401 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3402 MAP_ANONYMOUS|MAP_PRIVATE, 3403 -1, 0); 3404 if (p != MAP_FAILED) { 3405 void *aligned_p = align_up(p, page_size); 3406 3407 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3408 3409 munmap(p, page_size * 2); 3410 } 3411 3412 if (warn && !result) { 3413 warning("TransparentHugePages is not supported by the operating system."); 3414 } 3415 3416 return result; 3417 } 3418 3419 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3420 bool result = false; 3421 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3422 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3423 -1, 0); 3424 3425 if (p != MAP_FAILED) { 3426 // We don't know if this really is a huge page or not. 3427 FILE *fp = fopen("/proc/self/maps", "r"); 3428 if (fp) { 3429 while (!feof(fp)) { 3430 char chars[257]; 3431 long x = 0; 3432 if (fgets(chars, sizeof(chars), fp)) { 3433 if (sscanf(chars, "%lx-%*x", &x) == 1 3434 && x == (long)p) { 3435 if (strstr (chars, "hugepage")) { 3436 result = true; 3437 break; 3438 } 3439 } 3440 } 3441 } 3442 fclose(fp); 3443 } 3444 munmap(p, page_size); 3445 } 3446 3447 if (warn && !result) { 3448 warning("HugeTLBFS is not supported by the operating system."); 3449 } 3450 3451 return result; 3452 } 3453 3454 // From the coredump_filter documentation: 3455 // 3456 // - (bit 0) anonymous private memory 3457 // - (bit 1) anonymous shared memory 3458 // - (bit 2) file-backed private memory 3459 // - (bit 3) file-backed shared memory 3460 // - (bit 4) ELF header pages in file-backed private memory areas (it is 3461 // effective only if the bit 2 is cleared) 3462 // - (bit 5) hugetlb private memory 3463 // - (bit 6) hugetlb shared memory 3464 // - (bit 7) dax private memory 3465 // - (bit 8) dax shared memory 3466 // 3467 static void set_coredump_filter(CoredumpFilterBit bit) { 3468 FILE *f; 3469 long cdm; 3470 3471 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3472 return; 3473 } 3474 3475 if (fscanf(f, "%lx", &cdm) != 1) { 3476 fclose(f); 3477 return; 3478 } 3479 3480 long saved_cdm = cdm; 3481 rewind(f); 3482 cdm |= bit; 3483 3484 if (cdm != saved_cdm) { 3485 fprintf(f, "%#lx", cdm); 3486 } 3487 3488 fclose(f); 3489 } 3490 3491 // Large page support 3492 3493 static size_t _large_page_size = 0; 3494 3495 size_t os::Linux::find_large_page_size() { 3496 size_t large_page_size = 0; 3497 3498 // large_page_size on Linux is used to round up heap size. x86 uses either 3499 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3500 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3501 // page as large as 256M. 3502 // 3503 // Here we try to figure out page size by parsing /proc/meminfo and looking 3504 // for a line with the following format: 3505 // Hugepagesize: 2048 kB 3506 // 3507 // If we can't determine the value (e.g. /proc is not mounted, or the text 3508 // format has been changed), we'll use the largest page size supported by 3509 // the processor. 3510 3511 #ifndef ZERO 3512 large_page_size = 3513 AARCH64_ONLY(2 * M) 3514 AMD64_ONLY(2 * M) 3515 ARM32_ONLY(2 * M) 3516 IA32_ONLY(4 * M) 3517 IA64_ONLY(256 * M) 3518 PPC_ONLY(4 * M) 3519 S390_ONLY(1 * M) 3520 SPARC_ONLY(4 * M); 3521 #endif // ZERO 3522 3523 FILE *fp = fopen("/proc/meminfo", "r"); 3524 if (fp) { 3525 while (!feof(fp)) { 3526 int x = 0; 3527 char buf[16]; 3528 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3529 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3530 large_page_size = x * K; 3531 break; 3532 } 3533 } else { 3534 // skip to next line 3535 for (;;) { 3536 int ch = fgetc(fp); 3537 if (ch == EOF || ch == (int)'\n') break; 3538 } 3539 } 3540 } 3541 fclose(fp); 3542 } 3543 3544 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3545 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3546 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3547 proper_unit_for_byte_size(large_page_size)); 3548 } 3549 3550 return large_page_size; 3551 } 3552 3553 size_t os::Linux::setup_large_page_size() { 3554 _large_page_size = Linux::find_large_page_size(); 3555 const size_t default_page_size = (size_t)Linux::page_size(); 3556 if (_large_page_size > default_page_size) { 3557 _page_sizes[0] = _large_page_size; 3558 _page_sizes[1] = default_page_size; 3559 _page_sizes[2] = 0; 3560 } 3561 3562 return _large_page_size; 3563 } 3564 3565 bool os::Linux::setup_large_page_type(size_t page_size) { 3566 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3567 FLAG_IS_DEFAULT(UseSHM) && 3568 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3569 3570 // The type of large pages has not been specified by the user. 3571 3572 // Try UseHugeTLBFS and then UseSHM. 3573 UseHugeTLBFS = UseSHM = true; 3574 3575 // Don't try UseTransparentHugePages since there are known 3576 // performance issues with it turned on. This might change in the future. 3577 UseTransparentHugePages = false; 3578 } 3579 3580 if (UseTransparentHugePages) { 3581 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3582 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3583 UseHugeTLBFS = false; 3584 UseSHM = false; 3585 return true; 3586 } 3587 UseTransparentHugePages = false; 3588 } 3589 3590 if (UseHugeTLBFS) { 3591 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3592 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3593 UseSHM = false; 3594 return true; 3595 } 3596 UseHugeTLBFS = false; 3597 } 3598 3599 return UseSHM; 3600 } 3601 3602 void os::large_page_init() { 3603 if (!UseLargePages && 3604 !UseTransparentHugePages && 3605 !UseHugeTLBFS && 3606 !UseSHM) { 3607 // Not using large pages. 3608 return; 3609 } 3610 3611 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3612 // The user explicitly turned off large pages. 3613 // Ignore the rest of the large pages flags. 3614 UseTransparentHugePages = false; 3615 UseHugeTLBFS = false; 3616 UseSHM = false; 3617 return; 3618 } 3619 3620 size_t large_page_size = Linux::setup_large_page_size(); 3621 UseLargePages = Linux::setup_large_page_type(large_page_size); 3622 3623 set_coredump_filter(LARGEPAGES_BIT); 3624 } 3625 3626 #ifndef SHM_HUGETLB 3627 #define SHM_HUGETLB 04000 3628 #endif 3629 3630 #define shm_warning_format(format, ...) \ 3631 do { \ 3632 if (UseLargePages && \ 3633 (!FLAG_IS_DEFAULT(UseLargePages) || \ 3634 !FLAG_IS_DEFAULT(UseSHM) || \ 3635 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \ 3636 warning(format, __VA_ARGS__); \ 3637 } \ 3638 } while (0) 3639 3640 #define shm_warning(str) shm_warning_format("%s", str) 3641 3642 #define shm_warning_with_errno(str) \ 3643 do { \ 3644 int err = errno; \ 3645 shm_warning_format(str " (error = %d)", err); \ 3646 } while (0) 3647 3648 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) { 3649 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment"); 3650 3651 if (!is_aligned(alignment, SHMLBA)) { 3652 assert(false, "Code below assumes that alignment is at least SHMLBA aligned"); 3653 return NULL; 3654 } 3655 3656 // To ensure that we get 'alignment' aligned memory from shmat, 3657 // we pre-reserve aligned virtual memory and then attach to that. 3658 3659 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL); 3660 if (pre_reserved_addr == NULL) { 3661 // Couldn't pre-reserve aligned memory. 3662 shm_warning("Failed to pre-reserve aligned memory for shmat."); 3663 return NULL; 3664 } 3665 3666 // SHM_REMAP is needed to allow shmat to map over an existing mapping. 3667 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP); 3668 3669 if ((intptr_t)addr == -1) { 3670 int err = errno; 3671 shm_warning_with_errno("Failed to attach shared memory."); 3672 3673 assert(err != EACCES, "Unexpected error"); 3674 assert(err != EIDRM, "Unexpected error"); 3675 assert(err != EINVAL, "Unexpected error"); 3676 3677 // Since we don't know if the kernel unmapped the pre-reserved memory area 3678 // we can't unmap it, since that would potentially unmap memory that was 3679 // mapped from other threads. 3680 return NULL; 3681 } 3682 3683 return addr; 3684 } 3685 3686 static char* shmat_at_address(int shmid, char* req_addr) { 3687 if (!is_aligned(req_addr, SHMLBA)) { 3688 assert(false, "Requested address needs to be SHMLBA aligned"); 3689 return NULL; 3690 } 3691 3692 char* addr = (char*)shmat(shmid, req_addr, 0); 3693 3694 if ((intptr_t)addr == -1) { 3695 shm_warning_with_errno("Failed to attach shared memory."); 3696 return NULL; 3697 } 3698 3699 return addr; 3700 } 3701 3702 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) { 3703 // If a req_addr has been provided, we assume that the caller has already aligned the address. 3704 if (req_addr != NULL) { 3705 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size"); 3706 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment"); 3707 return shmat_at_address(shmid, req_addr); 3708 } 3709 3710 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically 3711 // return large page size aligned memory addresses when req_addr == NULL. 3712 // However, if the alignment is larger than the large page size, we have 3713 // to manually ensure that the memory returned is 'alignment' aligned. 3714 if (alignment > os::large_page_size()) { 3715 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size"); 3716 return shmat_with_alignment(shmid, bytes, alignment); 3717 } else { 3718 return shmat_at_address(shmid, NULL); 3719 } 3720 } 3721 3722 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3723 char* req_addr, bool exec) { 3724 // "exec" is passed in but not used. Creating the shared image for 3725 // the code cache doesn't have an SHM_X executable permission to check. 3726 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3727 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3728 assert(is_aligned(req_addr, alignment), "Unaligned address"); 3729 3730 if (!is_aligned(bytes, os::large_page_size())) { 3731 return NULL; // Fallback to small pages. 3732 } 3733 3734 // Create a large shared memory region to attach to based on size. 3735 // Currently, size is the total size of the heap. 3736 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3737 if (shmid == -1) { 3738 // Possible reasons for shmget failure: 3739 // 1. shmmax is too small for Java heap. 3740 // > check shmmax value: cat /proc/sys/kernel/shmmax 3741 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3742 // 2. not enough large page memory. 3743 // > check available large pages: cat /proc/meminfo 3744 // > increase amount of large pages: 3745 // echo new_value > /proc/sys/vm/nr_hugepages 3746 // Note 1: different Linux may use different name for this property, 3747 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3748 // Note 2: it's possible there's enough physical memory available but 3749 // they are so fragmented after a long run that they can't 3750 // coalesce into large pages. Try to reserve large pages when 3751 // the system is still "fresh". 3752 shm_warning_with_errno("Failed to reserve shared memory."); 3753 return NULL; 3754 } 3755 3756 // Attach to the region. 3757 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr); 3758 3759 // Remove shmid. If shmat() is successful, the actual shared memory segment 3760 // will be deleted when it's detached by shmdt() or when the process 3761 // terminates. If shmat() is not successful this will remove the shared 3762 // segment immediately. 3763 shmctl(shmid, IPC_RMID, NULL); 3764 3765 return addr; 3766 } 3767 3768 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3769 int error) { 3770 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3771 3772 bool warn_on_failure = UseLargePages && 3773 (!FLAG_IS_DEFAULT(UseLargePages) || 3774 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3775 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3776 3777 if (warn_on_failure) { 3778 char msg[128]; 3779 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3780 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3781 warning("%s", msg); 3782 } 3783 } 3784 3785 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3786 char* req_addr, 3787 bool exec) { 3788 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3789 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size"); 3790 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3791 3792 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3793 char* addr = (char*)::mmap(req_addr, bytes, prot, 3794 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3795 -1, 0); 3796 3797 if (addr == MAP_FAILED) { 3798 warn_on_large_pages_failure(req_addr, bytes, errno); 3799 return NULL; 3800 } 3801 3802 assert(is_aligned(addr, os::large_page_size()), "Must be"); 3803 3804 return addr; 3805 } 3806 3807 // Reserve memory using mmap(MAP_HUGETLB). 3808 // - bytes shall be a multiple of alignment. 3809 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3810 // - alignment sets the alignment at which memory shall be allocated. 3811 // It must be a multiple of allocation granularity. 3812 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3813 // req_addr or NULL. 3814 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3815 size_t alignment, 3816 char* req_addr, 3817 bool exec) { 3818 size_t large_page_size = os::large_page_size(); 3819 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3820 3821 assert(is_aligned(req_addr, alignment), "Must be"); 3822 assert(is_aligned(bytes, alignment), "Must be"); 3823 3824 // First reserve - but not commit - the address range in small pages. 3825 char* const start = anon_mmap_aligned(bytes, alignment, req_addr); 3826 3827 if (start == NULL) { 3828 return NULL; 3829 } 3830 3831 assert(is_aligned(start, alignment), "Must be"); 3832 3833 char* end = start + bytes; 3834 3835 // Find the regions of the allocated chunk that can be promoted to large pages. 3836 char* lp_start = align_up(start, large_page_size); 3837 char* lp_end = align_down(end, large_page_size); 3838 3839 size_t lp_bytes = lp_end - lp_start; 3840 3841 assert(is_aligned(lp_bytes, large_page_size), "Must be"); 3842 3843 if (lp_bytes == 0) { 3844 // The mapped region doesn't even span the start and the end of a large page. 3845 // Fall back to allocate a non-special area. 3846 ::munmap(start, end - start); 3847 return NULL; 3848 } 3849 3850 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3851 3852 void* result; 3853 3854 // Commit small-paged leading area. 3855 if (start != lp_start) { 3856 result = ::mmap(start, lp_start - start, prot, 3857 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3858 -1, 0); 3859 if (result == MAP_FAILED) { 3860 ::munmap(lp_start, end - lp_start); 3861 return NULL; 3862 } 3863 } 3864 3865 // Commit large-paged area. 3866 result = ::mmap(lp_start, lp_bytes, prot, 3867 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3868 -1, 0); 3869 if (result == MAP_FAILED) { 3870 warn_on_large_pages_failure(lp_start, lp_bytes, errno); 3871 // If the mmap above fails, the large pages region will be unmapped and we 3872 // have regions before and after with small pages. Release these regions. 3873 // 3874 // | mapped | unmapped | mapped | 3875 // ^ ^ ^ ^ 3876 // start lp_start lp_end end 3877 // 3878 ::munmap(start, lp_start - start); 3879 ::munmap(lp_end, end - lp_end); 3880 return NULL; 3881 } 3882 3883 // Commit small-paged trailing area. 3884 if (lp_end != end) { 3885 result = ::mmap(lp_end, end - lp_end, prot, 3886 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3887 -1, 0); 3888 if (result == MAP_FAILED) { 3889 ::munmap(start, lp_end - start); 3890 return NULL; 3891 } 3892 } 3893 3894 return start; 3895 } 3896 3897 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3898 size_t alignment, 3899 char* req_addr, 3900 bool exec) { 3901 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3902 assert(is_aligned(req_addr, alignment), "Must be"); 3903 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be"); 3904 assert(is_power_of_2(os::large_page_size()), "Must be"); 3905 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3906 3907 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3908 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3909 } else { 3910 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3911 } 3912 } 3913 3914 char* os::reserve_memory_special(size_t bytes, size_t alignment, 3915 char* req_addr, bool exec) { 3916 assert(UseLargePages, "only for large pages"); 3917 3918 char* addr; 3919 if (UseSHM) { 3920 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3921 } else { 3922 assert(UseHugeTLBFS, "must be"); 3923 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3924 } 3925 3926 if (addr != NULL) { 3927 if (UseNUMAInterleaving) { 3928 numa_make_global(addr, bytes); 3929 } 3930 3931 // The memory is committed 3932 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3933 } 3934 3935 return addr; 3936 } 3937 3938 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3939 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3940 return shmdt(base) == 0; 3941 } 3942 3943 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3944 return pd_release_memory(base, bytes); 3945 } 3946 3947 bool os::release_memory_special(char* base, size_t bytes) { 3948 bool res; 3949 if (MemTracker::tracking_level() > NMT_minimal) { 3950 Tracker tkr(Tracker::release); 3951 res = os::Linux::release_memory_special_impl(base, bytes); 3952 if (res) { 3953 tkr.record((address)base, bytes); 3954 } 3955 3956 } else { 3957 res = os::Linux::release_memory_special_impl(base, bytes); 3958 } 3959 return res; 3960 } 3961 3962 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3963 assert(UseLargePages, "only for large pages"); 3964 bool res; 3965 3966 if (UseSHM) { 3967 res = os::Linux::release_memory_special_shm(base, bytes); 3968 } else { 3969 assert(UseHugeTLBFS, "must be"); 3970 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3971 } 3972 return res; 3973 } 3974 3975 size_t os::large_page_size() { 3976 return _large_page_size; 3977 } 3978 3979 // With SysV SHM the entire memory region must be allocated as shared 3980 // memory. 3981 // HugeTLBFS allows application to commit large page memory on demand. 3982 // However, when committing memory with HugeTLBFS fails, the region 3983 // that was supposed to be committed will lose the old reservation 3984 // and allow other threads to steal that memory region. Because of this 3985 // behavior we can't commit HugeTLBFS memory. 3986 bool os::can_commit_large_page_memory() { 3987 return UseTransparentHugePages; 3988 } 3989 3990 bool os::can_execute_large_page_memory() { 3991 return UseTransparentHugePages || UseHugeTLBFS; 3992 } 3993 3994 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) { 3995 assert(file_desc >= 0, "file_desc is not valid"); 3996 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr); 3997 if (result != NULL) { 3998 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) { 3999 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory")); 4000 } 4001 } 4002 return result; 4003 } 4004 4005 // Reserve memory at an arbitrary address, only if that area is 4006 // available (and not reserved for something else). 4007 4008 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 4009 const int max_tries = 10; 4010 char* base[max_tries]; 4011 size_t size[max_tries]; 4012 const size_t gap = 0x000000; 4013 4014 // Assert only that the size is a multiple of the page size, since 4015 // that's all that mmap requires, and since that's all we really know 4016 // about at this low abstraction level. If we need higher alignment, 4017 // we can either pass an alignment to this method or verify alignment 4018 // in one of the methods further up the call chain. See bug 5044738. 4019 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 4020 4021 // Repeatedly allocate blocks until the block is allocated at the 4022 // right spot. 4023 4024 // Linux mmap allows caller to pass an address as hint; give it a try first, 4025 // if kernel honors the hint then we can return immediately. 4026 char * addr = anon_mmap(requested_addr, bytes, false); 4027 if (addr == requested_addr) { 4028 return requested_addr; 4029 } 4030 4031 if (addr != NULL) { 4032 // mmap() is successful but it fails to reserve at the requested address 4033 anon_munmap(addr, bytes); 4034 } 4035 4036 int i; 4037 for (i = 0; i < max_tries; ++i) { 4038 base[i] = reserve_memory(bytes); 4039 4040 if (base[i] != NULL) { 4041 // Is this the block we wanted? 4042 if (base[i] == requested_addr) { 4043 size[i] = bytes; 4044 break; 4045 } 4046 4047 // Does this overlap the block we wanted? Give back the overlapped 4048 // parts and try again. 4049 4050 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i]; 4051 if (top_overlap >= 0 && (size_t)top_overlap < bytes) { 4052 unmap_memory(base[i], top_overlap); 4053 base[i] += top_overlap; 4054 size[i] = bytes - top_overlap; 4055 } else { 4056 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr; 4057 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) { 4058 unmap_memory(requested_addr, bottom_overlap); 4059 size[i] = bytes - bottom_overlap; 4060 } else { 4061 size[i] = bytes; 4062 } 4063 } 4064 } 4065 } 4066 4067 // Give back the unused reserved pieces. 4068 4069 for (int j = 0; j < i; ++j) { 4070 if (base[j] != NULL) { 4071 unmap_memory(base[j], size[j]); 4072 } 4073 } 4074 4075 if (i < max_tries) { 4076 return requested_addr; 4077 } else { 4078 return NULL; 4079 } 4080 } 4081 4082 // Sleep forever; naked call to OS-specific sleep; use with CAUTION 4083 void os::infinite_sleep() { 4084 while (true) { // sleep forever ... 4085 ::sleep(100); // ... 100 seconds at a time 4086 } 4087 } 4088 4089 // Used to convert frequent JVM_Yield() to nops 4090 bool os::dont_yield() { 4091 return DontYieldALot; 4092 } 4093 4094 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will 4095 // actually give up the CPU. Since skip buddy (v2.6.28): 4096 // 4097 // * Sets the yielding task as skip buddy for current CPU's run queue. 4098 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task). 4099 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy). 4100 // 4101 // An alternative is calling os::naked_short_nanosleep with a small number to avoid 4102 // getting re-scheduled immediately. 4103 // 4104 void os::naked_yield() { 4105 sched_yield(); 4106 } 4107 4108 //////////////////////////////////////////////////////////////////////////////// 4109 // thread priority support 4110 4111 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 4112 // only supports dynamic priority, static priority must be zero. For real-time 4113 // applications, Linux supports SCHED_RR which allows static priority (1-99). 4114 // However, for large multi-threaded applications, SCHED_RR is not only slower 4115 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 4116 // of 5 runs - Sep 2005). 4117 // 4118 // The following code actually changes the niceness of kernel-thread/LWP. It 4119 // has an assumption that setpriority() only modifies one kernel-thread/LWP, 4120 // not the entire user process, and user level threads are 1:1 mapped to kernel 4121 // threads. It has always been the case, but could change in the future. For 4122 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). 4123 // It is only used when ThreadPriorityPolicy=1 and may require system level permission 4124 // (e.g., root privilege or CAP_SYS_NICE capability). 4125 4126 int os::java_to_os_priority[CriticalPriority + 1] = { 4127 19, // 0 Entry should never be used 4128 4129 4, // 1 MinPriority 4130 3, // 2 4131 2, // 3 4132 4133 1, // 4 4134 0, // 5 NormPriority 4135 -1, // 6 4136 4137 -2, // 7 4138 -3, // 8 4139 -4, // 9 NearMaxPriority 4140 4141 -5, // 10 MaxPriority 4142 4143 -5 // 11 CriticalPriority 4144 }; 4145 4146 static int prio_init() { 4147 if (ThreadPriorityPolicy == 1) { 4148 if (geteuid() != 0) { 4149 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 4150 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \ 4151 "e.g., being the root user. If the necessary permission is not " \ 4152 "possessed, changes to priority will be silently ignored."); 4153 } 4154 } 4155 } 4156 if (UseCriticalJavaThreadPriority) { 4157 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 4158 } 4159 return 0; 4160 } 4161 4162 OSReturn os::set_native_priority(Thread* thread, int newpri) { 4163 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 4164 4165 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 4166 return (ret == 0) ? OS_OK : OS_ERR; 4167 } 4168 4169 OSReturn os::get_native_priority(const Thread* const thread, 4170 int *priority_ptr) { 4171 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 4172 *priority_ptr = java_to_os_priority[NormPriority]; 4173 return OS_OK; 4174 } 4175 4176 errno = 0; 4177 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 4178 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 4179 } 4180 4181 //////////////////////////////////////////////////////////////////////////////// 4182 // suspend/resume support 4183 4184 // The low-level signal-based suspend/resume support is a remnant from the 4185 // old VM-suspension that used to be for java-suspension, safepoints etc, 4186 // within hotspot. Currently used by JFR's OSThreadSampler 4187 // 4188 // The remaining code is greatly simplified from the more general suspension 4189 // code that used to be used. 4190 // 4191 // The protocol is quite simple: 4192 // - suspend: 4193 // - sends a signal to the target thread 4194 // - polls the suspend state of the osthread using a yield loop 4195 // - target thread signal handler (SR_handler) sets suspend state 4196 // and blocks in sigsuspend until continued 4197 // - resume: 4198 // - sets target osthread state to continue 4199 // - sends signal to end the sigsuspend loop in the SR_handler 4200 // 4201 // Note that the SR_lock plays no role in this suspend/resume protocol, 4202 // but is checked for NULL in SR_handler as a thread termination indicator. 4203 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs. 4204 // 4205 // Note that resume_clear_context() and suspend_save_context() are needed 4206 // by SR_handler(), so that fetch_frame_from_ucontext() works, 4207 // which in part is used by: 4208 // - Forte Analyzer: AsyncGetCallTrace() 4209 // - StackBanging: get_frame_at_stack_banging_point() 4210 4211 static void resume_clear_context(OSThread *osthread) { 4212 osthread->set_ucontext(NULL); 4213 osthread->set_siginfo(NULL); 4214 } 4215 4216 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 4217 ucontext_t* context) { 4218 osthread->set_ucontext(context); 4219 osthread->set_siginfo(siginfo); 4220 } 4221 4222 // Handler function invoked when a thread's execution is suspended or 4223 // resumed. We have to be careful that only async-safe functions are 4224 // called here (Note: most pthread functions are not async safe and 4225 // should be avoided.) 4226 // 4227 // Note: sigwait() is a more natural fit than sigsuspend() from an 4228 // interface point of view, but sigwait() prevents the signal hander 4229 // from being run. libpthread would get very confused by not having 4230 // its signal handlers run and prevents sigwait()'s use with the 4231 // mutex granting granting signal. 4232 // 4233 // Currently only ever called on the VMThread and JavaThreads (PC sampling) 4234 // 4235 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 4236 // Save and restore errno to avoid confusing native code with EINTR 4237 // after sigsuspend. 4238 int old_errno = errno; 4239 4240 Thread* thread = Thread::current_or_null_safe(); 4241 assert(thread != NULL, "Missing current thread in SR_handler"); 4242 4243 // On some systems we have seen signal delivery get "stuck" until the signal 4244 // mask is changed as part of thread termination. Check that the current thread 4245 // has not already terminated (via SR_lock()) - else the following assertion 4246 // will fail because the thread is no longer a JavaThread as the ~JavaThread 4247 // destructor has completed. 4248 4249 if (thread->SR_lock() == NULL) { 4250 return; 4251 } 4252 4253 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 4254 4255 OSThread* osthread = thread->osthread(); 4256 4257 os::SuspendResume::State current = osthread->sr.state(); 4258 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 4259 suspend_save_context(osthread, siginfo, context); 4260 4261 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4262 os::SuspendResume::State state = osthread->sr.suspended(); 4263 if (state == os::SuspendResume::SR_SUSPENDED) { 4264 sigset_t suspend_set; // signals for sigsuspend() 4265 sigemptyset(&suspend_set); 4266 // get current set of blocked signals and unblock resume signal 4267 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4268 sigdelset(&suspend_set, SR_signum); 4269 4270 sr_semaphore.signal(); 4271 // wait here until we are resumed 4272 while (1) { 4273 sigsuspend(&suspend_set); 4274 4275 os::SuspendResume::State result = osthread->sr.running(); 4276 if (result == os::SuspendResume::SR_RUNNING) { 4277 sr_semaphore.signal(); 4278 break; 4279 } 4280 } 4281 4282 } else if (state == os::SuspendResume::SR_RUNNING) { 4283 // request was cancelled, continue 4284 } else { 4285 ShouldNotReachHere(); 4286 } 4287 4288 resume_clear_context(osthread); 4289 } else if (current == os::SuspendResume::SR_RUNNING) { 4290 // request was cancelled, continue 4291 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4292 // ignore 4293 } else { 4294 // ignore 4295 } 4296 4297 errno = old_errno; 4298 } 4299 4300 static int SR_initialize() { 4301 struct sigaction act; 4302 char *s; 4303 4304 // Get signal number to use for suspend/resume 4305 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4306 int sig = ::strtol(s, 0, 10); 4307 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769. 4308 sig < NSIG) { // Must be legal signal and fit into sigflags[]. 4309 SR_signum = sig; 4310 } else { 4311 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.", 4312 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum); 4313 } 4314 } 4315 4316 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4317 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4318 4319 sigemptyset(&SR_sigset); 4320 sigaddset(&SR_sigset, SR_signum); 4321 4322 // Set up signal handler for suspend/resume 4323 act.sa_flags = SA_RESTART|SA_SIGINFO; 4324 act.sa_handler = (void (*)(int)) SR_handler; 4325 4326 // SR_signum is blocked by default. 4327 // 4528190 - We also need to block pthread restart signal (32 on all 4328 // supported Linux platforms). Note that LinuxThreads need to block 4329 // this signal for all threads to work properly. So we don't have 4330 // to use hard-coded signal number when setting up the mask. 4331 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4332 4333 if (sigaction(SR_signum, &act, 0) == -1) { 4334 return -1; 4335 } 4336 4337 // Save signal flag 4338 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4339 return 0; 4340 } 4341 4342 static int sr_notify(OSThread* osthread) { 4343 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4344 assert_status(status == 0, status, "pthread_kill"); 4345 return status; 4346 } 4347 4348 // "Randomly" selected value for how long we want to spin 4349 // before bailing out on suspending a thread, also how often 4350 // we send a signal to a thread we want to resume 4351 static const int RANDOMLY_LARGE_INTEGER = 1000000; 4352 static const int RANDOMLY_LARGE_INTEGER2 = 100; 4353 4354 // returns true on success and false on error - really an error is fatal 4355 // but this seems the normal response to library errors 4356 static bool do_suspend(OSThread* osthread) { 4357 assert(osthread->sr.is_running(), "thread should be running"); 4358 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4359 4360 // mark as suspended and send signal 4361 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4362 // failed to switch, state wasn't running? 4363 ShouldNotReachHere(); 4364 return false; 4365 } 4366 4367 if (sr_notify(osthread) != 0) { 4368 ShouldNotReachHere(); 4369 } 4370 4371 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4372 while (true) { 4373 if (sr_semaphore.timedwait(2)) { 4374 break; 4375 } else { 4376 // timeout 4377 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4378 if (cancelled == os::SuspendResume::SR_RUNNING) { 4379 return false; 4380 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4381 // make sure that we consume the signal on the semaphore as well 4382 sr_semaphore.wait(); 4383 break; 4384 } else { 4385 ShouldNotReachHere(); 4386 return false; 4387 } 4388 } 4389 } 4390 4391 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4392 return true; 4393 } 4394 4395 static void do_resume(OSThread* osthread) { 4396 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4397 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4398 4399 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4400 // failed to switch to WAKEUP_REQUEST 4401 ShouldNotReachHere(); 4402 return; 4403 } 4404 4405 while (true) { 4406 if (sr_notify(osthread) == 0) { 4407 if (sr_semaphore.timedwait(2)) { 4408 if (osthread->sr.is_running()) { 4409 return; 4410 } 4411 } 4412 } else { 4413 ShouldNotReachHere(); 4414 } 4415 } 4416 4417 guarantee(osthread->sr.is_running(), "Must be running!"); 4418 } 4419 4420 /////////////////////////////////////////////////////////////////////////////////// 4421 // signal handling (except suspend/resume) 4422 4423 // This routine may be used by user applications as a "hook" to catch signals. 4424 // The user-defined signal handler must pass unrecognized signals to this 4425 // routine, and if it returns true (non-zero), then the signal handler must 4426 // return immediately. If the flag "abort_if_unrecognized" is true, then this 4427 // routine will never retun false (zero), but instead will execute a VM panic 4428 // routine kill the process. 4429 // 4430 // If this routine returns false, it is OK to call it again. This allows 4431 // the user-defined signal handler to perform checks either before or after 4432 // the VM performs its own checks. Naturally, the user code would be making 4433 // a serious error if it tried to handle an exception (such as a null check 4434 // or breakpoint) that the VM was generating for its own correct operation. 4435 // 4436 // This routine may recognize any of the following kinds of signals: 4437 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4438 // It should be consulted by handlers for any of those signals. 4439 // 4440 // The caller of this routine must pass in the three arguments supplied 4441 // to the function referred to in the "sa_sigaction" (not the "sa_handler") 4442 // field of the structure passed to sigaction(). This routine assumes that 4443 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4444 // 4445 // Note that the VM will print warnings if it detects conflicting signal 4446 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4447 // 4448 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4449 siginfo_t* siginfo, 4450 void* ucontext, 4451 int abort_if_unrecognized); 4452 4453 static void signalHandler(int sig, siginfo_t* info, void* uc) { 4454 assert(info != NULL && uc != NULL, "it must be old kernel"); 4455 int orig_errno = errno; // Preserve errno value over signal handler. 4456 JVM_handle_linux_signal(sig, info, uc, true); 4457 errno = orig_errno; 4458 } 4459 4460 4461 // This boolean allows users to forward their own non-matching signals 4462 // to JVM_handle_linux_signal, harmlessly. 4463 bool os::Linux::signal_handlers_are_installed = false; 4464 4465 // For signal-chaining 4466 bool os::Linux::libjsig_is_loaded = false; 4467 typedef struct sigaction *(*get_signal_t)(int); 4468 get_signal_t os::Linux::get_signal_action = NULL; 4469 4470 struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4471 struct sigaction *actp = NULL; 4472 4473 if (libjsig_is_loaded) { 4474 // Retrieve the old signal handler from libjsig 4475 actp = (*get_signal_action)(sig); 4476 } 4477 if (actp == NULL) { 4478 // Retrieve the preinstalled signal handler from jvm 4479 actp = os::Posix::get_preinstalled_handler(sig); 4480 } 4481 4482 return actp; 4483 } 4484 4485 static bool call_chained_handler(struct sigaction *actp, int sig, 4486 siginfo_t *siginfo, void *context) { 4487 // Call the old signal handler 4488 if (actp->sa_handler == SIG_DFL) { 4489 // It's more reasonable to let jvm treat it as an unexpected exception 4490 // instead of taking the default action. 4491 return false; 4492 } else if (actp->sa_handler != SIG_IGN) { 4493 if ((actp->sa_flags & SA_NODEFER) == 0) { 4494 // automaticlly block the signal 4495 sigaddset(&(actp->sa_mask), sig); 4496 } 4497 4498 sa_handler_t hand = NULL; 4499 sa_sigaction_t sa = NULL; 4500 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4501 // retrieve the chained handler 4502 if (siginfo_flag_set) { 4503 sa = actp->sa_sigaction; 4504 } else { 4505 hand = actp->sa_handler; 4506 } 4507 4508 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4509 actp->sa_handler = SIG_DFL; 4510 } 4511 4512 // try to honor the signal mask 4513 sigset_t oset; 4514 sigemptyset(&oset); 4515 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4516 4517 // call into the chained handler 4518 if (siginfo_flag_set) { 4519 (*sa)(sig, siginfo, context); 4520 } else { 4521 (*hand)(sig); 4522 } 4523 4524 // restore the signal mask 4525 pthread_sigmask(SIG_SETMASK, &oset, NULL); 4526 } 4527 // Tell jvm's signal handler the signal is taken care of. 4528 return true; 4529 } 4530 4531 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4532 bool chained = false; 4533 // signal-chaining 4534 if (UseSignalChaining) { 4535 struct sigaction *actp = get_chained_signal_action(sig); 4536 if (actp != NULL) { 4537 chained = call_chained_handler(actp, sig, siginfo, context); 4538 } 4539 } 4540 return chained; 4541 } 4542 4543 // for diagnostic 4544 int sigflags[NSIG]; 4545 4546 int os::Linux::get_our_sigflags(int sig) { 4547 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4548 return sigflags[sig]; 4549 } 4550 4551 void os::Linux::set_our_sigflags(int sig, int flags) { 4552 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4553 if (sig > 0 && sig < NSIG) { 4554 sigflags[sig] = flags; 4555 } 4556 } 4557 4558 void os::Linux::set_signal_handler(int sig, bool set_installed) { 4559 // Check for overwrite. 4560 struct sigaction oldAct; 4561 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4562 4563 void* oldhand = oldAct.sa_sigaction 4564 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4565 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4566 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4567 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4568 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4569 if (AllowUserSignalHandlers || !set_installed) { 4570 // Do not overwrite; user takes responsibility to forward to us. 4571 return; 4572 } else if (UseSignalChaining) { 4573 // save the old handler in jvm 4574 os::Posix::save_preinstalled_handler(sig, oldAct); 4575 // libjsig also interposes the sigaction() call below and saves the 4576 // old sigaction on it own. 4577 } else { 4578 fatal("Encountered unexpected pre-existing sigaction handler " 4579 "%#lx for signal %d.", (long)oldhand, sig); 4580 } 4581 } 4582 4583 struct sigaction sigAct; 4584 sigfillset(&(sigAct.sa_mask)); 4585 sigAct.sa_handler = SIG_DFL; 4586 if (!set_installed) { 4587 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4588 } else { 4589 sigAct.sa_sigaction = signalHandler; 4590 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4591 } 4592 // Save flags, which are set by ours 4593 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4594 sigflags[sig] = sigAct.sa_flags; 4595 4596 int ret = sigaction(sig, &sigAct, &oldAct); 4597 assert(ret == 0, "check"); 4598 4599 void* oldhand2 = oldAct.sa_sigaction 4600 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4601 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4602 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4603 } 4604 4605 // install signal handlers for signals that HotSpot needs to 4606 // handle in order to support Java-level exception handling. 4607 4608 void os::Linux::install_signal_handlers() { 4609 if (!signal_handlers_are_installed) { 4610 signal_handlers_are_installed = true; 4611 4612 // signal-chaining 4613 typedef void (*signal_setting_t)(); 4614 signal_setting_t begin_signal_setting = NULL; 4615 signal_setting_t end_signal_setting = NULL; 4616 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4617 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4618 if (begin_signal_setting != NULL) { 4619 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4620 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4621 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4622 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4623 libjsig_is_loaded = true; 4624 assert(UseSignalChaining, "should enable signal-chaining"); 4625 } 4626 if (libjsig_is_loaded) { 4627 // Tell libjsig jvm is setting signal handlers 4628 (*begin_signal_setting)(); 4629 } 4630 4631 set_signal_handler(SIGSEGV, true); 4632 set_signal_handler(SIGPIPE, true); 4633 set_signal_handler(SIGBUS, true); 4634 set_signal_handler(SIGILL, true); 4635 set_signal_handler(SIGFPE, true); 4636 #if defined(PPC64) 4637 set_signal_handler(SIGTRAP, true); 4638 #endif 4639 set_signal_handler(SIGXFSZ, true); 4640 4641 if (libjsig_is_loaded) { 4642 // Tell libjsig jvm finishes setting signal handlers 4643 (*end_signal_setting)(); 4644 } 4645 4646 // We don't activate signal checker if libjsig is in place, we trust ourselves 4647 // and if UserSignalHandler is installed all bets are off. 4648 // Log that signal checking is off only if -verbose:jni is specified. 4649 if (CheckJNICalls) { 4650 if (libjsig_is_loaded) { 4651 if (PrintJNIResolving) { 4652 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4653 } 4654 check_signals = false; 4655 } 4656 if (AllowUserSignalHandlers) { 4657 if (PrintJNIResolving) { 4658 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4659 } 4660 check_signals = false; 4661 } 4662 } 4663 } 4664 } 4665 4666 // This is the fastest way to get thread cpu time on Linux. 4667 // Returns cpu time (user+sys) for any thread, not only for current. 4668 // POSIX compliant clocks are implemented in the kernels 2.6.16+. 4669 // It might work on 2.6.10+ with a special kernel/glibc patch. 4670 // For reference, please, see IEEE Std 1003.1-2004: 4671 // http://www.unix.org/single_unix_specification 4672 4673 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4674 struct timespec tp; 4675 int rc = os::Posix::clock_gettime(clockid, &tp); 4676 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4677 4678 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4679 } 4680 4681 void os::Linux::initialize_os_info() { 4682 assert(_os_version == 0, "OS info already initialized"); 4683 4684 struct utsname _uname; 4685 4686 uint32_t major; 4687 uint32_t minor; 4688 uint32_t fix; 4689 4690 int rc; 4691 4692 // Kernel version is unknown if 4693 // verification below fails. 4694 _os_version = 0x01000000; 4695 4696 rc = uname(&_uname); 4697 if (rc != -1) { 4698 4699 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix); 4700 if (rc == 3) { 4701 4702 if (major < 256 && minor < 256 && fix < 256) { 4703 // Kernel version format is as expected, 4704 // set it overriding unknown state. 4705 _os_version = (major << 16) | 4706 (minor << 8 ) | 4707 (fix << 0 ) ; 4708 } 4709 } 4710 } 4711 } 4712 4713 uint32_t os::Linux::os_version() { 4714 assert(_os_version != 0, "not initialized"); 4715 return _os_version & 0x00FFFFFF; 4716 } 4717 4718 bool os::Linux::os_version_is_known() { 4719 assert(_os_version != 0, "not initialized"); 4720 return _os_version & 0x01000000 ? false : true; 4721 } 4722 4723 ///// 4724 // glibc on Linux platform uses non-documented flag 4725 // to indicate, that some special sort of signal 4726 // trampoline is used. 4727 // We will never set this flag, and we should 4728 // ignore this flag in our diagnostic 4729 #ifdef SIGNIFICANT_SIGNAL_MASK 4730 #undef SIGNIFICANT_SIGNAL_MASK 4731 #endif 4732 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4733 4734 static const char* get_signal_handler_name(address handler, 4735 char* buf, int buflen) { 4736 int offset = 0; 4737 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4738 if (found) { 4739 // skip directory names 4740 const char *p1, *p2; 4741 p1 = buf; 4742 size_t len = strlen(os::file_separator()); 4743 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4744 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4745 } else { 4746 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4747 } 4748 return buf; 4749 } 4750 4751 static void print_signal_handler(outputStream* st, int sig, 4752 char* buf, size_t buflen) { 4753 struct sigaction sa; 4754 4755 sigaction(sig, NULL, &sa); 4756 4757 // See comment for SIGNIFICANT_SIGNAL_MASK define 4758 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4759 4760 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4761 4762 address handler = (sa.sa_flags & SA_SIGINFO) 4763 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4764 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4765 4766 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4767 st->print("SIG_DFL"); 4768 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4769 st->print("SIG_IGN"); 4770 } else { 4771 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4772 } 4773 4774 st->print(", sa_mask[0]="); 4775 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4776 4777 address rh = VMError::get_resetted_sighandler(sig); 4778 // May be, handler was resetted by VMError? 4779 if (rh != NULL) { 4780 handler = rh; 4781 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4782 } 4783 4784 st->print(", sa_flags="); 4785 os::Posix::print_sa_flags(st, sa.sa_flags); 4786 4787 // Check: is it our handler? 4788 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4789 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4790 // It is our signal handler 4791 // check for flags, reset system-used one! 4792 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4793 st->print( 4794 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4795 os::Linux::get_our_sigflags(sig)); 4796 } 4797 } 4798 st->cr(); 4799 } 4800 4801 4802 #define DO_SIGNAL_CHECK(sig) \ 4803 do { \ 4804 if (!sigismember(&check_signal_done, sig)) { \ 4805 os::Linux::check_signal_handler(sig); \ 4806 } \ 4807 } while (0) 4808 4809 // This method is a periodic task to check for misbehaving JNI applications 4810 // under CheckJNI, we can add any periodic checks here 4811 4812 void os::run_periodic_checks() { 4813 if (check_signals == false) return; 4814 4815 // SEGV and BUS if overridden could potentially prevent 4816 // generation of hs*.log in the event of a crash, debugging 4817 // such a case can be very challenging, so we absolutely 4818 // check the following for a good measure: 4819 DO_SIGNAL_CHECK(SIGSEGV); 4820 DO_SIGNAL_CHECK(SIGILL); 4821 DO_SIGNAL_CHECK(SIGFPE); 4822 DO_SIGNAL_CHECK(SIGBUS); 4823 DO_SIGNAL_CHECK(SIGPIPE); 4824 DO_SIGNAL_CHECK(SIGXFSZ); 4825 #if defined(PPC64) 4826 DO_SIGNAL_CHECK(SIGTRAP); 4827 #endif 4828 4829 // ReduceSignalUsage allows the user to override these handlers 4830 // see comments at the very top and jvm_md.h 4831 if (!ReduceSignalUsage) { 4832 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4833 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4834 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4835 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4836 } 4837 4838 DO_SIGNAL_CHECK(SR_signum); 4839 } 4840 4841 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4842 4843 static os_sigaction_t os_sigaction = NULL; 4844 4845 void os::Linux::check_signal_handler(int sig) { 4846 char buf[O_BUFLEN]; 4847 address jvmHandler = NULL; 4848 4849 4850 struct sigaction act; 4851 if (os_sigaction == NULL) { 4852 // only trust the default sigaction, in case it has been interposed 4853 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4854 if (os_sigaction == NULL) return; 4855 } 4856 4857 os_sigaction(sig, (struct sigaction*)NULL, &act); 4858 4859 4860 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4861 4862 address thisHandler = (act.sa_flags & SA_SIGINFO) 4863 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4864 : CAST_FROM_FN_PTR(address, act.sa_handler); 4865 4866 4867 switch (sig) { 4868 case SIGSEGV: 4869 case SIGBUS: 4870 case SIGFPE: 4871 case SIGPIPE: 4872 case SIGILL: 4873 case SIGXFSZ: 4874 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4875 break; 4876 4877 case SHUTDOWN1_SIGNAL: 4878 case SHUTDOWN2_SIGNAL: 4879 case SHUTDOWN3_SIGNAL: 4880 case BREAK_SIGNAL: 4881 jvmHandler = (address)user_handler(); 4882 break; 4883 4884 default: 4885 if (sig == SR_signum) { 4886 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4887 } else { 4888 return; 4889 } 4890 break; 4891 } 4892 4893 if (thisHandler != jvmHandler) { 4894 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4895 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4896 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4897 // No need to check this sig any longer 4898 sigaddset(&check_signal_done, sig); 4899 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4900 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4901 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4902 exception_name(sig, buf, O_BUFLEN)); 4903 } 4904 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4905 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4906 tty->print("expected:"); 4907 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig)); 4908 tty->cr(); 4909 tty->print(" found:"); 4910 os::Posix::print_sa_flags(tty, act.sa_flags); 4911 tty->cr(); 4912 // No need to check this sig any longer 4913 sigaddset(&check_signal_done, sig); 4914 } 4915 4916 // Dump all the signal 4917 if (sigismember(&check_signal_done, sig)) { 4918 print_signal_handlers(tty, buf, O_BUFLEN); 4919 } 4920 } 4921 4922 extern void report_error(char* file_name, int line_no, char* title, 4923 char* format, ...); 4924 4925 // this is called _before_ most of the global arguments have been parsed 4926 void os::init(void) { 4927 char dummy; // used to get a guess on initial stack address 4928 4929 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4930 4931 init_random(1234567); 4932 4933 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4934 if (Linux::page_size() == -1) { 4935 fatal("os_linux.cpp: os::init: sysconf failed (%s)", 4936 os::strerror(errno)); 4937 } 4938 init_page_sizes((size_t) Linux::page_size()); 4939 4940 Linux::initialize_system_info(); 4941 4942 Linux::initialize_os_info(); 4943 4944 // _main_thread points to the thread that created/loaded the JVM. 4945 Linux::_main_thread = pthread_self(); 4946 4947 // retrieve entry point for pthread_setname_np 4948 Linux::_pthread_setname_np = 4949 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4950 4951 os::Posix::init(); 4952 4953 initial_time_count = javaTimeNanos(); 4954 4955 // Always warn if no monotonic clock available 4956 if (!os::Posix::supports_monotonic_clock()) { 4957 warning("No monotonic clock was available - timed services may " \ 4958 "be adversely affected if the time-of-day clock changes"); 4959 } 4960 } 4961 4962 // To install functions for atexit system call 4963 extern "C" { 4964 static void perfMemory_exit_helper() { 4965 perfMemory_exit(); 4966 } 4967 } 4968 4969 void os::pd_init_container_support() { 4970 OSContainer::init(); 4971 } 4972 4973 void os::Linux::numa_init() { 4974 4975 // Java can be invoked as 4976 // 1. Without numactl and heap will be allocated/configured on all nodes as 4977 // per the system policy. 4978 // 2. With numactl --interleave: 4979 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same 4980 // API for membind case bitmask is reset. 4981 // Interleave is only hint and Kernel can fallback to other nodes if 4982 // no memory is available on the target nodes. 4983 // 3. With numactl --membind: 4984 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for 4985 // interleave case returns bitmask of all nodes. 4986 // numa_all_nodes_ptr holds bitmask of all nodes. 4987 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct 4988 // bitmask when externally configured to run on all or fewer nodes. 4989 4990 if (!Linux::libnuma_init()) { 4991 UseNUMA = false; 4992 } else { 4993 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) { 4994 // If there's only one node (they start from 0) or if the process 4995 // is bound explicitly to a single node using membind, disable NUMA. 4996 UseNUMA = false; 4997 } else { 4998 4999 LogTarget(Info,os) log; 5000 LogStream ls(log); 5001 5002 Linux::set_configured_numa_policy(Linux::identify_numa_policy()); 5003 5004 struct bitmask* bmp = Linux::_numa_membind_bitmask; 5005 const char* numa_mode = "membind"; 5006 5007 if (Linux::is_running_in_interleave_mode()) { 5008 bmp = Linux::_numa_interleave_bitmask; 5009 numa_mode = "interleave"; 5010 } 5011 5012 ls.print("UseNUMA is enabled and invoked in '%s' mode." 5013 " Heap will be configured using NUMA memory nodes:", numa_mode); 5014 5015 for (int node = 0; node <= Linux::numa_max_node(); node++) { 5016 if (Linux::_numa_bitmask_isbitset(bmp, node)) { 5017 ls.print(" %d", node); 5018 } 5019 } 5020 } 5021 } 5022 5023 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 5024 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 5025 // we can make the adaptive lgrp chunk resizing work. If the user specified both 5026 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn 5027 // and disable adaptive resizing. 5028 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) { 5029 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, " 5030 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)"); 5031 UseAdaptiveSizePolicy = false; 5032 UseAdaptiveNUMAChunkSizing = false; 5033 } 5034 } 5035 5036 if (!UseNUMA && ForceNUMA) { 5037 UseNUMA = true; 5038 } 5039 } 5040 5041 // this is called _after_ the global arguments have been parsed 5042 jint os::init_2(void) { 5043 5044 // This could be set after os::Posix::init() but all platforms 5045 // have to set it the same so we have to mirror Solaris. 5046 DEBUG_ONLY(os::set_mutex_init_done();) 5047 5048 os::Posix::init_2(); 5049 5050 Linux::fast_thread_clock_init(); 5051 5052 // initialize suspend/resume support - must do this before signal_sets_init() 5053 if (SR_initialize() != 0) { 5054 perror("SR_initialize failed"); 5055 return JNI_ERR; 5056 } 5057 5058 Linux::signal_sets_init(); 5059 Linux::install_signal_handlers(); 5060 // Initialize data for jdk.internal.misc.Signal 5061 if (!ReduceSignalUsage) { 5062 jdk_misc_signal_init(); 5063 } 5064 5065 // Check and sets minimum stack sizes against command line options 5066 if (Posix::set_minimum_stack_sizes() == JNI_ERR) { 5067 return JNI_ERR; 5068 } 5069 5070 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher(); 5071 if (!suppress_primordial_thread_resolution) { 5072 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 5073 } 5074 5075 #if defined(IA32) 5076 workaround_expand_exec_shield_cs_limit(); 5077 #endif 5078 5079 Linux::libpthread_init(); 5080 Linux::sched_getcpu_init(); 5081 log_info(os)("HotSpot is running with %s, %s", 5082 Linux::glibc_version(), Linux::libpthread_version()); 5083 5084 if (UseNUMA) { 5085 Linux::numa_init(); 5086 } 5087 5088 if (MaxFDLimit) { 5089 // set the number of file descriptors to max. print out error 5090 // if getrlimit/setrlimit fails but continue regardless. 5091 struct rlimit nbr_files; 5092 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 5093 if (status != 0) { 5094 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno)); 5095 } else { 5096 nbr_files.rlim_cur = nbr_files.rlim_max; 5097 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 5098 if (status != 0) { 5099 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno)); 5100 } 5101 } 5102 } 5103 5104 // Initialize lock used to serialize thread creation (see os::create_thread) 5105 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 5106 5107 // at-exit methods are called in the reverse order of their registration. 5108 // atexit functions are called on return from main or as a result of a 5109 // call to exit(3C). There can be only 32 of these functions registered 5110 // and atexit() does not set errno. 5111 5112 if (PerfAllowAtExitRegistration) { 5113 // only register atexit functions if PerfAllowAtExitRegistration is set. 5114 // atexit functions can be delayed until process exit time, which 5115 // can be problematic for embedded VM situations. Embedded VMs should 5116 // call DestroyJavaVM() to assure that VM resources are released. 5117 5118 // note: perfMemory_exit_helper atexit function may be removed in 5119 // the future if the appropriate cleanup code can be added to the 5120 // VM_Exit VMOperation's doit method. 5121 if (atexit(perfMemory_exit_helper) != 0) { 5122 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 5123 } 5124 } 5125 5126 // initialize thread priority policy 5127 prio_init(); 5128 5129 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) { 5130 set_coredump_filter(DAX_SHARED_BIT); 5131 } 5132 5133 if (DumpPrivateMappingsInCore) { 5134 set_coredump_filter(FILE_BACKED_PVT_BIT); 5135 } 5136 5137 if (DumpSharedMappingsInCore) { 5138 set_coredump_filter(FILE_BACKED_SHARED_BIT); 5139 } 5140 5141 return JNI_OK; 5142 } 5143 5144 // Mark the polling page as unreadable 5145 void os::make_polling_page_unreadable(void) { 5146 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 5147 fatal("Could not disable polling page"); 5148 } 5149 } 5150 5151 // Mark the polling page as readable 5152 void os::make_polling_page_readable(void) { 5153 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 5154 fatal("Could not enable polling page"); 5155 } 5156 } 5157 5158 // older glibc versions don't have this macro (which expands to 5159 // an optimized bit-counting function) so we have to roll our own 5160 #ifndef CPU_COUNT 5161 5162 static int _cpu_count(const cpu_set_t* cpus) { 5163 int count = 0; 5164 // only look up to the number of configured processors 5165 for (int i = 0; i < os::processor_count(); i++) { 5166 if (CPU_ISSET(i, cpus)) { 5167 count++; 5168 } 5169 } 5170 return count; 5171 } 5172 5173 #define CPU_COUNT(cpus) _cpu_count(cpus) 5174 5175 #endif // CPU_COUNT 5176 5177 // Get the current number of available processors for this process. 5178 // This value can change at any time during a process's lifetime. 5179 // sched_getaffinity gives an accurate answer as it accounts for cpusets. 5180 // If it appears there may be more than 1024 processors then we do a 5181 // dynamic check - see 6515172 for details. 5182 // If anything goes wrong we fallback to returning the number of online 5183 // processors - which can be greater than the number available to the process. 5184 int os::Linux::active_processor_count() { 5185 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors 5186 cpu_set_t* cpus_p = &cpus; 5187 int cpus_size = sizeof(cpu_set_t); 5188 5189 int configured_cpus = os::processor_count(); // upper bound on available cpus 5190 int cpu_count = 0; 5191 5192 // old build platforms may not support dynamic cpu sets 5193 #ifdef CPU_ALLOC 5194 5195 // To enable easy testing of the dynamic path on different platforms we 5196 // introduce a diagnostic flag: UseCpuAllocPath 5197 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) { 5198 // kernel may use a mask bigger than cpu_set_t 5199 log_trace(os)("active_processor_count: using dynamic path %s" 5200 "- configured processors: %d", 5201 UseCpuAllocPath ? "(forced) " : "", 5202 configured_cpus); 5203 cpus_p = CPU_ALLOC(configured_cpus); 5204 if (cpus_p != NULL) { 5205 cpus_size = CPU_ALLOC_SIZE(configured_cpus); 5206 // zero it just to be safe 5207 CPU_ZERO_S(cpus_size, cpus_p); 5208 } 5209 else { 5210 // failed to allocate so fallback to online cpus 5211 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 5212 log_trace(os)("active_processor_count: " 5213 "CPU_ALLOC failed (%s) - using " 5214 "online processor count: %d", 5215 os::strerror(errno), online_cpus); 5216 return online_cpus; 5217 } 5218 } 5219 else { 5220 log_trace(os)("active_processor_count: using static path - configured processors: %d", 5221 configured_cpus); 5222 } 5223 #else // CPU_ALLOC 5224 // these stubs won't be executed 5225 #define CPU_COUNT_S(size, cpus) -1 5226 #define CPU_FREE(cpus) 5227 5228 log_trace(os)("active_processor_count: only static path available - configured processors: %d", 5229 configured_cpus); 5230 #endif // CPU_ALLOC 5231 5232 // pid 0 means the current thread - which we have to assume represents the process 5233 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) { 5234 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5235 cpu_count = CPU_COUNT_S(cpus_size, cpus_p); 5236 } 5237 else { 5238 cpu_count = CPU_COUNT(cpus_p); 5239 } 5240 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count); 5241 } 5242 else { 5243 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN); 5244 warning("sched_getaffinity failed (%s)- using online processor count (%d) " 5245 "which may exceed available processors", os::strerror(errno), cpu_count); 5246 } 5247 5248 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5249 CPU_FREE(cpus_p); 5250 } 5251 5252 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check"); 5253 return cpu_count; 5254 } 5255 5256 // Determine the active processor count from one of 5257 // three different sources: 5258 // 5259 // 1. User option -XX:ActiveProcessorCount 5260 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN) 5261 // 3. extracted from cgroup cpu subsystem (shares and quotas) 5262 // 5263 // Option 1, if specified, will always override. 5264 // If the cgroup subsystem is active and configured, we 5265 // will return the min of the cgroup and option 2 results. 5266 // This is required since tools, such as numactl, that 5267 // alter cpu affinity do not update cgroup subsystem 5268 // cpuset configuration files. 5269 int os::active_processor_count() { 5270 // User has overridden the number of active processors 5271 if (ActiveProcessorCount > 0) { 5272 log_trace(os)("active_processor_count: " 5273 "active processor count set by user : %d", 5274 ActiveProcessorCount); 5275 return ActiveProcessorCount; 5276 } 5277 5278 int active_cpus; 5279 if (OSContainer::is_containerized()) { 5280 active_cpus = OSContainer::active_processor_count(); 5281 log_trace(os)("active_processor_count: determined by OSContainer: %d", 5282 active_cpus); 5283 } else { 5284 active_cpus = os::Linux::active_processor_count(); 5285 } 5286 5287 return active_cpus; 5288 } 5289 5290 uint os::processor_id() { 5291 const int id = Linux::sched_getcpu(); 5292 assert(id >= 0 && id < _processor_count, "Invalid processor id"); 5293 return (uint)id; 5294 } 5295 5296 void os::set_native_thread_name(const char *name) { 5297 if (Linux::_pthread_setname_np) { 5298 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 5299 snprintf(buf, sizeof(buf), "%s", name); 5300 buf[sizeof(buf) - 1] = '\0'; 5301 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 5302 // ERANGE should not happen; all other errors should just be ignored. 5303 assert(rc != ERANGE, "pthread_setname_np failed"); 5304 } 5305 } 5306 5307 bool os::distribute_processes(uint length, uint* distribution) { 5308 // Not yet implemented. 5309 return false; 5310 } 5311 5312 bool os::bind_to_processor(uint processor_id) { 5313 // Not yet implemented. 5314 return false; 5315 } 5316 5317 /// 5318 5319 void os::SuspendedThreadTask::internal_do_task() { 5320 if (do_suspend(_thread->osthread())) { 5321 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 5322 do_task(context); 5323 do_resume(_thread->osthread()); 5324 } 5325 } 5326 5327 //////////////////////////////////////////////////////////////////////////////// 5328 // debug support 5329 5330 bool os::find(address addr, outputStream* st) { 5331 Dl_info dlinfo; 5332 memset(&dlinfo, 0, sizeof(dlinfo)); 5333 if (dladdr(addr, &dlinfo) != 0) { 5334 st->print(PTR_FORMAT ": ", p2i(addr)); 5335 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5336 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname, 5337 p2i(addr) - p2i(dlinfo.dli_saddr)); 5338 } else if (dlinfo.dli_fbase != NULL) { 5339 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase)); 5340 } else { 5341 st->print("<absolute address>"); 5342 } 5343 if (dlinfo.dli_fname != NULL) { 5344 st->print(" in %s", dlinfo.dli_fname); 5345 } 5346 if (dlinfo.dli_fbase != NULL) { 5347 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase)); 5348 } 5349 st->cr(); 5350 5351 if (Verbose) { 5352 // decode some bytes around the PC 5353 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5354 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5355 address lowest = (address) dlinfo.dli_sname; 5356 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5357 if (begin < lowest) begin = lowest; 5358 Dl_info dlinfo2; 5359 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5360 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 5361 end = (address) dlinfo2.dli_saddr; 5362 } 5363 Disassembler::decode(begin, end, st); 5364 } 5365 return true; 5366 } 5367 return false; 5368 } 5369 5370 //////////////////////////////////////////////////////////////////////////////// 5371 // misc 5372 5373 // This does not do anything on Linux. This is basically a hook for being 5374 // able to use structured exception handling (thread-local exception filters) 5375 // on, e.g., Win32. 5376 void 5377 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method, 5378 JavaCallArguments* args, Thread* thread) { 5379 f(value, method, args, thread); 5380 } 5381 5382 void os::print_statistics() { 5383 } 5384 5385 bool os::message_box(const char* title, const char* message) { 5386 int i; 5387 fdStream err(defaultStream::error_fd()); 5388 for (i = 0; i < 78; i++) err.print_raw("="); 5389 err.cr(); 5390 err.print_raw_cr(title); 5391 for (i = 0; i < 78; i++) err.print_raw("-"); 5392 err.cr(); 5393 err.print_raw_cr(message); 5394 for (i = 0; i < 78; i++) err.print_raw("="); 5395 err.cr(); 5396 5397 char buf[16]; 5398 // Prevent process from exiting upon "read error" without consuming all CPU 5399 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5400 5401 return buf[0] == 'y' || buf[0] == 'Y'; 5402 } 5403 5404 // Is a (classpath) directory empty? 5405 bool os::dir_is_empty(const char* path) { 5406 DIR *dir = NULL; 5407 struct dirent *ptr; 5408 5409 dir = opendir(path); 5410 if (dir == NULL) return true; 5411 5412 // Scan the directory 5413 bool result = true; 5414 while (result && (ptr = readdir(dir)) != NULL) { 5415 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5416 result = false; 5417 } 5418 } 5419 closedir(dir); 5420 return result; 5421 } 5422 5423 // This code originates from JDK's sysOpen and open64_w 5424 // from src/solaris/hpi/src/system_md.c 5425 5426 int os::open(const char *path, int oflag, int mode) { 5427 if (strlen(path) > MAX_PATH - 1) { 5428 errno = ENAMETOOLONG; 5429 return -1; 5430 } 5431 5432 // All file descriptors that are opened in the Java process and not 5433 // specifically destined for a subprocess should have the close-on-exec 5434 // flag set. If we don't set it, then careless 3rd party native code 5435 // might fork and exec without closing all appropriate file descriptors 5436 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5437 // turn might: 5438 // 5439 // - cause end-of-file to fail to be detected on some file 5440 // descriptors, resulting in mysterious hangs, or 5441 // 5442 // - might cause an fopen in the subprocess to fail on a system 5443 // suffering from bug 1085341. 5444 // 5445 // (Yes, the default setting of the close-on-exec flag is a Unix 5446 // design flaw) 5447 // 5448 // See: 5449 // 1085341: 32-bit stdio routines should support file descriptors >255 5450 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5451 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5452 // 5453 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5454 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5455 // because it saves a system call and removes a small window where the flag 5456 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5457 // and we fall back to using FD_CLOEXEC (see below). 5458 #ifdef O_CLOEXEC 5459 oflag |= O_CLOEXEC; 5460 #endif 5461 5462 int fd = ::open64(path, oflag, mode); 5463 if (fd == -1) return -1; 5464 5465 //If the open succeeded, the file might still be a directory 5466 { 5467 struct stat64 buf64; 5468 int ret = ::fstat64(fd, &buf64); 5469 int st_mode = buf64.st_mode; 5470 5471 if (ret != -1) { 5472 if ((st_mode & S_IFMT) == S_IFDIR) { 5473 errno = EISDIR; 5474 ::close(fd); 5475 return -1; 5476 } 5477 } else { 5478 ::close(fd); 5479 return -1; 5480 } 5481 } 5482 5483 #ifdef FD_CLOEXEC 5484 // Validate that the use of the O_CLOEXEC flag on open above worked. 5485 // With recent kernels, we will perform this check exactly once. 5486 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5487 if (!O_CLOEXEC_is_known_to_work) { 5488 int flags = ::fcntl(fd, F_GETFD); 5489 if (flags != -1) { 5490 if ((flags & FD_CLOEXEC) != 0) 5491 O_CLOEXEC_is_known_to_work = 1; 5492 else 5493 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5494 } 5495 } 5496 #endif 5497 5498 return fd; 5499 } 5500 5501 5502 // create binary file, rewriting existing file if required 5503 int os::create_binary_file(const char* path, bool rewrite_existing) { 5504 int oflags = O_WRONLY | O_CREAT; 5505 if (!rewrite_existing) { 5506 oflags |= O_EXCL; 5507 } 5508 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5509 } 5510 5511 // return current position of file pointer 5512 jlong os::current_file_offset(int fd) { 5513 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5514 } 5515 5516 // move file pointer to the specified offset 5517 jlong os::seek_to_file_offset(int fd, jlong offset) { 5518 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5519 } 5520 5521 // This code originates from JDK's sysAvailable 5522 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5523 5524 int os::available(int fd, jlong *bytes) { 5525 jlong cur, end; 5526 int mode; 5527 struct stat64 buf64; 5528 5529 if (::fstat64(fd, &buf64) >= 0) { 5530 mode = buf64.st_mode; 5531 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5532 int n; 5533 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5534 *bytes = n; 5535 return 1; 5536 } 5537 } 5538 } 5539 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5540 return 0; 5541 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5542 return 0; 5543 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5544 return 0; 5545 } 5546 *bytes = end - cur; 5547 return 1; 5548 } 5549 5550 // Map a block of memory. 5551 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5552 char *addr, size_t bytes, bool read_only, 5553 bool allow_exec) { 5554 int prot; 5555 int flags = MAP_PRIVATE; 5556 5557 if (read_only) { 5558 prot = PROT_READ; 5559 } else { 5560 prot = PROT_READ | PROT_WRITE; 5561 } 5562 5563 if (allow_exec) { 5564 prot |= PROT_EXEC; 5565 } 5566 5567 if (addr != NULL) { 5568 flags |= MAP_FIXED; 5569 } 5570 5571 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5572 fd, file_offset); 5573 if (mapped_address == MAP_FAILED) { 5574 return NULL; 5575 } 5576 return mapped_address; 5577 } 5578 5579 5580 // Remap a block of memory. 5581 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5582 char *addr, size_t bytes, bool read_only, 5583 bool allow_exec) { 5584 // same as map_memory() on this OS 5585 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5586 allow_exec); 5587 } 5588 5589 5590 // Unmap a block of memory. 5591 bool os::pd_unmap_memory(char* addr, size_t bytes) { 5592 return munmap(addr, bytes) == 0; 5593 } 5594 5595 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5596 5597 static jlong fast_cpu_time(Thread *thread) { 5598 clockid_t clockid; 5599 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(), 5600 &clockid); 5601 if (rc == 0) { 5602 return os::Linux::fast_thread_cpu_time(clockid); 5603 } else { 5604 // It's possible to encounter a terminated native thread that failed 5605 // to detach itself from the VM - which should result in ESRCH. 5606 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed"); 5607 return -1; 5608 } 5609 } 5610 5611 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5612 // are used by JVM M&M and JVMTI to get user+sys or user CPU time 5613 // of a thread. 5614 // 5615 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5616 // the fast estimate available on the platform. 5617 5618 jlong os::current_thread_cpu_time() { 5619 if (os::Linux::supports_fast_thread_cpu_time()) { 5620 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5621 } else { 5622 // return user + sys since the cost is the same 5623 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5624 } 5625 } 5626 5627 jlong os::thread_cpu_time(Thread* thread) { 5628 // consistent with what current_thread_cpu_time() returns 5629 if (os::Linux::supports_fast_thread_cpu_time()) { 5630 return fast_cpu_time(thread); 5631 } else { 5632 return slow_thread_cpu_time(thread, true /* user + sys */); 5633 } 5634 } 5635 5636 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5637 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5638 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5639 } else { 5640 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5641 } 5642 } 5643 5644 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5645 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5646 return fast_cpu_time(thread); 5647 } else { 5648 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5649 } 5650 } 5651 5652 // -1 on error. 5653 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5654 pid_t tid = thread->osthread()->thread_id(); 5655 char *s; 5656 char stat[2048]; 5657 int statlen; 5658 char proc_name[64]; 5659 int count; 5660 long sys_time, user_time; 5661 char cdummy; 5662 int idummy; 5663 long ldummy; 5664 FILE *fp; 5665 5666 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5667 fp = fopen(proc_name, "r"); 5668 if (fp == NULL) return -1; 5669 statlen = fread(stat, 1, 2047, fp); 5670 stat[statlen] = '\0'; 5671 fclose(fp); 5672 5673 // Skip pid and the command string. Note that we could be dealing with 5674 // weird command names, e.g. user could decide to rename java launcher 5675 // to "java 1.4.2 :)", then the stat file would look like 5676 // 1234 (java 1.4.2 :)) R ... ... 5677 // We don't really need to know the command string, just find the last 5678 // occurrence of ")" and then start parsing from there. See bug 4726580. 5679 s = strrchr(stat, ')'); 5680 if (s == NULL) return -1; 5681 5682 // Skip blank chars 5683 do { s++; } while (s && isspace(*s)); 5684 5685 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5686 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5687 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5688 &user_time, &sys_time); 5689 if (count != 13) return -1; 5690 if (user_sys_cpu_time) { 5691 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5692 } else { 5693 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5694 } 5695 } 5696 5697 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5698 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5699 info_ptr->may_skip_backward = false; // elapsed time not wall time 5700 info_ptr->may_skip_forward = false; // elapsed time not wall time 5701 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5702 } 5703 5704 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5705 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5706 info_ptr->may_skip_backward = false; // elapsed time not wall time 5707 info_ptr->may_skip_forward = false; // elapsed time not wall time 5708 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5709 } 5710 5711 bool os::is_thread_cpu_time_supported() { 5712 return true; 5713 } 5714 5715 // System loadavg support. Returns -1 if load average cannot be obtained. 5716 // Linux doesn't yet have a (official) notion of processor sets, 5717 // so just return the system wide load average. 5718 int os::loadavg(double loadavg[], int nelem) { 5719 return ::getloadavg(loadavg, nelem); 5720 } 5721 5722 void os::pause() { 5723 char filename[MAX_PATH]; 5724 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5725 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile); 5726 } else { 5727 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5728 } 5729 5730 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5731 if (fd != -1) { 5732 struct stat buf; 5733 ::close(fd); 5734 while (::stat(filename, &buf) == 0) { 5735 (void)::poll(NULL, 0, 100); 5736 } 5737 } else { 5738 jio_fprintf(stderr, 5739 "Could not open pause file '%s', continuing immediately.\n", filename); 5740 } 5741 } 5742 5743 extern char** environ; 5744 5745 // Run the specified command in a separate process. Return its exit value, 5746 // or -1 on failure (e.g. can't fork a new process). 5747 // Unlike system(), this function can be called from signal handler. It 5748 // doesn't block SIGINT et al. 5749 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) { 5750 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5751 5752 pid_t pid ; 5753 5754 if (use_vfork_if_available) { 5755 pid = vfork(); 5756 } else { 5757 pid = fork(); 5758 } 5759 5760 if (pid < 0) { 5761 // fork failed 5762 return -1; 5763 5764 } else if (pid == 0) { 5765 // child process 5766 5767 execve("/bin/sh", (char* const*)argv, environ); 5768 5769 // execve failed 5770 _exit(-1); 5771 5772 } else { 5773 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5774 // care about the actual exit code, for now. 5775 5776 int status; 5777 5778 // Wait for the child process to exit. This returns immediately if 5779 // the child has already exited. */ 5780 while (waitpid(pid, &status, 0) < 0) { 5781 switch (errno) { 5782 case ECHILD: return 0; 5783 case EINTR: break; 5784 default: return -1; 5785 } 5786 } 5787 5788 if (WIFEXITED(status)) { 5789 // The child exited normally; get its exit code. 5790 return WEXITSTATUS(status); 5791 } else if (WIFSIGNALED(status)) { 5792 // The child exited because of a signal 5793 // The best value to return is 0x80 + signal number, 5794 // because that is what all Unix shells do, and because 5795 // it allows callers to distinguish between process exit and 5796 // process death by signal. 5797 return 0x80 + WTERMSIG(status); 5798 } else { 5799 // Unknown exit code; pass it through 5800 return status; 5801 } 5802 } 5803 } 5804 5805 // Get the default path to the core file 5806 // Returns the length of the string 5807 int os::get_core_path(char* buffer, size_t bufferSize) { 5808 /* 5809 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 5810 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 5811 */ 5812 const int core_pattern_len = 129; 5813 char core_pattern[core_pattern_len] = {0}; 5814 5815 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 5816 if (core_pattern_file == -1) { 5817 return -1; 5818 } 5819 5820 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 5821 ::close(core_pattern_file); 5822 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') { 5823 return -1; 5824 } 5825 if (core_pattern[ret-1] == '\n') { 5826 core_pattern[ret-1] = '\0'; 5827 } else { 5828 core_pattern[ret] = '\0'; 5829 } 5830 5831 // Replace the %p in the core pattern with the process id. NOTE: we do this 5832 // only if the pattern doesn't start with "|", and we support only one %p in 5833 // the pattern. 5834 char *pid_pos = strstr(core_pattern, "%p"); 5835 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p" 5836 int written; 5837 5838 if (core_pattern[0] == '/') { 5839 if (pid_pos != NULL) { 5840 *pid_pos = '\0'; 5841 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern, 5842 current_process_id(), tail); 5843 } else { 5844 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern); 5845 } 5846 } else { 5847 char cwd[PATH_MAX]; 5848 5849 const char* p = get_current_directory(cwd, PATH_MAX); 5850 if (p == NULL) { 5851 return -1; 5852 } 5853 5854 if (core_pattern[0] == '|') { 5855 written = jio_snprintf(buffer, bufferSize, 5856 "\"%s\" (or dumping to %s/core.%d)", 5857 &core_pattern[1], p, current_process_id()); 5858 } else if (pid_pos != NULL) { 5859 *pid_pos = '\0'; 5860 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern, 5861 current_process_id(), tail); 5862 } else { 5863 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 5864 } 5865 } 5866 5867 if (written < 0) { 5868 return -1; 5869 } 5870 5871 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 5872 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 5873 5874 if (core_uses_pid_file != -1) { 5875 char core_uses_pid = 0; 5876 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 5877 ::close(core_uses_pid_file); 5878 5879 if (core_uses_pid == '1') { 5880 jio_snprintf(buffer + written, bufferSize - written, 5881 ".%d", current_process_id()); 5882 } 5883 } 5884 } 5885 5886 return strlen(buffer); 5887 } 5888 5889 bool os::start_debugging(char *buf, int buflen) { 5890 int len = (int)strlen(buf); 5891 char *p = &buf[len]; 5892 5893 jio_snprintf(p, buflen-len, 5894 "\n\n" 5895 "Do you want to debug the problem?\n\n" 5896 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n" 5897 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n" 5898 "Otherwise, press RETURN to abort...", 5899 os::current_process_id(), os::current_process_id(), 5900 os::current_thread_id(), os::current_thread_id()); 5901 5902 bool yes = os::message_box("Unexpected Error", buf); 5903 5904 if (yes) { 5905 // yes, user asked VM to launch debugger 5906 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d", 5907 os::current_process_id(), os::current_process_id()); 5908 5909 os::fork_and_exec(buf); 5910 yes = false; 5911 } 5912 return yes; 5913 } 5914 5915 5916 // Java/Compiler thread: 5917 // 5918 // Low memory addresses 5919 // P0 +------------------------+ 5920 // | |\ Java thread created by VM does not have glibc 5921 // | glibc guard page | - guard page, attached Java thread usually has 5922 // | |/ 1 glibc guard page. 5923 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 5924 // | |\ 5925 // | HotSpot Guard Pages | - red, yellow and reserved pages 5926 // | |/ 5927 // +------------------------+ JavaThread::stack_reserved_zone_base() 5928 // | |\ 5929 // | Normal Stack | - 5930 // | |/ 5931 // P2 +------------------------+ Thread::stack_base() 5932 // 5933 // Non-Java thread: 5934 // 5935 // Low memory addresses 5936 // P0 +------------------------+ 5937 // | |\ 5938 // | glibc guard page | - usually 1 page 5939 // | |/ 5940 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 5941 // | |\ 5942 // | Normal Stack | - 5943 // | |/ 5944 // P2 +------------------------+ Thread::stack_base() 5945 // 5946 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size 5947 // returned from pthread_attr_getstack(). 5948 // ** Due to NPTL implementation error, linux takes the glibc guard page out 5949 // of the stack size given in pthread_attr. We work around this for 5950 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.) 5951 // 5952 #ifndef ZERO 5953 static void current_stack_region(address * bottom, size_t * size) { 5954 if (os::is_primordial_thread()) { 5955 // primordial thread needs special handling because pthread_getattr_np() 5956 // may return bogus value. 5957 *bottom = os::Linux::initial_thread_stack_bottom(); 5958 *size = os::Linux::initial_thread_stack_size(); 5959 } else { 5960 pthread_attr_t attr; 5961 5962 int rslt = pthread_getattr_np(pthread_self(), &attr); 5963 5964 // JVM needs to know exact stack location, abort if it fails 5965 if (rslt != 0) { 5966 if (rslt == ENOMEM) { 5967 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np"); 5968 } else { 5969 fatal("pthread_getattr_np failed with error = %d", rslt); 5970 } 5971 } 5972 5973 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) { 5974 fatal("Cannot locate current stack attributes!"); 5975 } 5976 5977 // Work around NPTL stack guard error. 5978 size_t guard_size = 0; 5979 rslt = pthread_attr_getguardsize(&attr, &guard_size); 5980 if (rslt != 0) { 5981 fatal("pthread_attr_getguardsize failed with error = %d", rslt); 5982 } 5983 *bottom += guard_size; 5984 *size -= guard_size; 5985 5986 pthread_attr_destroy(&attr); 5987 5988 } 5989 assert(os::current_stack_pointer() >= *bottom && 5990 os::current_stack_pointer() < *bottom + *size, "just checking"); 5991 } 5992 5993 address os::current_stack_base() { 5994 address bottom; 5995 size_t size; 5996 current_stack_region(&bottom, &size); 5997 return (bottom + size); 5998 } 5999 6000 size_t os::current_stack_size() { 6001 // This stack size includes the usable stack and HotSpot guard pages 6002 // (for the threads that have Hotspot guard pages). 6003 address bottom; 6004 size_t size; 6005 current_stack_region(&bottom, &size); 6006 return size; 6007 } 6008 #endif 6009 6010 static inline struct timespec get_mtime(const char* filename) { 6011 struct stat st; 6012 int ret = os::stat(filename, &st); 6013 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno)); 6014 return st.st_mtim; 6015 } 6016 6017 int os::compare_file_modified_times(const char* file1, const char* file2) { 6018 struct timespec filetime1 = get_mtime(file1); 6019 struct timespec filetime2 = get_mtime(file2); 6020 int diff = filetime1.tv_sec - filetime2.tv_sec; 6021 if (diff == 0) { 6022 return filetime1.tv_nsec - filetime2.tv_nsec; 6023 } 6024 return diff; 6025 } 6026 6027 /////////////// Unit tests /////////////// 6028 6029 #ifndef PRODUCT 6030 6031 class TestReserveMemorySpecial : AllStatic { 6032 public: 6033 static void small_page_write(void* addr, size_t size) { 6034 size_t page_size = os::vm_page_size(); 6035 6036 char* end = (char*)addr + size; 6037 for (char* p = (char*)addr; p < end; p += page_size) { 6038 *p = 1; 6039 } 6040 } 6041 6042 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6043 if (!UseHugeTLBFS) { 6044 return; 6045 } 6046 6047 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6048 6049 if (addr != NULL) { 6050 small_page_write(addr, size); 6051 6052 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6053 } 6054 } 6055 6056 static void test_reserve_memory_special_huge_tlbfs_only() { 6057 if (!UseHugeTLBFS) { 6058 return; 6059 } 6060 6061 size_t lp = os::large_page_size(); 6062 6063 for (size_t size = lp; size <= lp * 10; size += lp) { 6064 test_reserve_memory_special_huge_tlbfs_only(size); 6065 } 6066 } 6067 6068 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6069 size_t lp = os::large_page_size(); 6070 size_t ag = os::vm_allocation_granularity(); 6071 6072 // sizes to test 6073 const size_t sizes[] = { 6074 lp, lp + ag, lp + lp / 2, lp * 2, 6075 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2, 6076 lp * 10, lp * 10 + lp / 2 6077 }; 6078 const int num_sizes = sizeof(sizes) / sizeof(size_t); 6079 6080 // For each size/alignment combination, we test three scenarios: 6081 // 1) with req_addr == NULL 6082 // 2) with a non-null req_addr at which we expect to successfully allocate 6083 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we 6084 // expect the allocation to either fail or to ignore req_addr 6085 6086 // Pre-allocate two areas; they shall be as large as the largest allocation 6087 // and aligned to the largest alignment we will be testing. 6088 const size_t mapping_size = sizes[num_sizes - 1] * 2; 6089 char* const mapping1 = (char*) ::mmap(NULL, mapping_size, 6090 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6091 -1, 0); 6092 assert(mapping1 != MAP_FAILED, "should work"); 6093 6094 char* const mapping2 = (char*) ::mmap(NULL, mapping_size, 6095 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6096 -1, 0); 6097 assert(mapping2 != MAP_FAILED, "should work"); 6098 6099 // Unmap the first mapping, but leave the second mapping intact: the first 6100 // mapping will serve as a value for a "good" req_addr (case 2). The second 6101 // mapping, still intact, as "bad" req_addr (case 3). 6102 ::munmap(mapping1, mapping_size); 6103 6104 // Case 1 6105 for (int i = 0; i < num_sizes; i++) { 6106 const size_t size = sizes[i]; 6107 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6108 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6109 if (p != NULL) { 6110 assert(is_aligned(p, alignment), "must be"); 6111 small_page_write(p, size); 6112 os::Linux::release_memory_special_huge_tlbfs(p, size); 6113 } 6114 } 6115 } 6116 6117 // Case 2 6118 for (int i = 0; i < num_sizes; i++) { 6119 const size_t size = sizes[i]; 6120 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6121 char* const req_addr = align_up(mapping1, alignment); 6122 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6123 if (p != NULL) { 6124 assert(p == req_addr, "must be"); 6125 small_page_write(p, size); 6126 os::Linux::release_memory_special_huge_tlbfs(p, size); 6127 } 6128 } 6129 } 6130 6131 // Case 3 6132 for (int i = 0; i < num_sizes; i++) { 6133 const size_t size = sizes[i]; 6134 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6135 char* const req_addr = align_up(mapping2, alignment); 6136 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6137 // as the area around req_addr contains already existing mappings, the API should always 6138 // return NULL (as per contract, it cannot return another address) 6139 assert(p == NULL, "must be"); 6140 } 6141 } 6142 6143 ::munmap(mapping2, mapping_size); 6144 6145 } 6146 6147 static void test_reserve_memory_special_huge_tlbfs() { 6148 if (!UseHugeTLBFS) { 6149 return; 6150 } 6151 6152 test_reserve_memory_special_huge_tlbfs_only(); 6153 test_reserve_memory_special_huge_tlbfs_mixed(); 6154 } 6155 6156 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6157 if (!UseSHM) { 6158 return; 6159 } 6160 6161 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6162 6163 if (addr != NULL) { 6164 assert(is_aligned(addr, alignment), "Check"); 6165 assert(is_aligned(addr, os::large_page_size()), "Check"); 6166 6167 small_page_write(addr, size); 6168 6169 os::Linux::release_memory_special_shm(addr, size); 6170 } 6171 } 6172 6173 static void test_reserve_memory_special_shm() { 6174 size_t lp = os::large_page_size(); 6175 size_t ag = os::vm_allocation_granularity(); 6176 6177 for (size_t size = ag; size < lp * 3; size += ag) { 6178 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6179 test_reserve_memory_special_shm(size, alignment); 6180 } 6181 } 6182 } 6183 6184 static void test() { 6185 test_reserve_memory_special_huge_tlbfs(); 6186 test_reserve_memory_special_shm(); 6187 } 6188 }; 6189 6190 void TestReserveMemorySpecial_test() { 6191 TestReserveMemorySpecial::test(); 6192 } 6193 6194 #endif