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_share_linux.hpp" 42 #include "osContainer_linux.hpp" 43 #include "prims/jniFastGetField.hpp" 44 #include "prims/jvm_misc.hpp" 45 #include "runtime/arguments.hpp" 46 #include "runtime/atomic.hpp" 47 #include "runtime/extendedPC.hpp" 48 #include "runtime/globals.hpp" 49 #include "runtime/interfaceSupport.inline.hpp" 50 #include "runtime/init.hpp" 51 #include "runtime/java.hpp" 52 #include "runtime/javaCalls.hpp" 53 #include "runtime/mutexLocker.hpp" 54 #include "runtime/objectMonitor.hpp" 55 #include "runtime/orderAccess.hpp" 56 #include "runtime/osThread.hpp" 57 #include "runtime/perfMemory.hpp" 58 #include "runtime/sharedRuntime.hpp" 59 #include "runtime/statSampler.hpp" 60 #include "runtime/stubRoutines.hpp" 61 #include "runtime/thread.inline.hpp" 62 #include "runtime/threadCritical.hpp" 63 #include "runtime/threadSMR.hpp" 64 #include "runtime/timer.hpp" 65 #include "semaphore_posix.hpp" 66 #include "services/attachListener.hpp" 67 #include "services/memTracker.hpp" 68 #include "services/runtimeService.hpp" 69 #include "utilities/align.hpp" 70 #include "utilities/decoder.hpp" 71 #include "utilities/defaultStream.hpp" 72 #include "utilities/events.hpp" 73 #include "utilities/elfFile.hpp" 74 #include "utilities/growableArray.hpp" 75 #include "utilities/macros.hpp" 76 #include "utilities/vmError.hpp" 77 78 // put OS-includes here 79 # include <sys/types.h> 80 # include <sys/mman.h> 81 # include <sys/stat.h> 82 # include <sys/select.h> 83 # include <pthread.h> 84 # include <signal.h> 85 # include <errno.h> 86 # include <dlfcn.h> 87 # include <stdio.h> 88 # include <unistd.h> 89 # include <sys/resource.h> 90 # include <pthread.h> 91 # include <sys/stat.h> 92 # include <sys/time.h> 93 # include <sys/times.h> 94 # include <sys/utsname.h> 95 # include <sys/socket.h> 96 # include <sys/wait.h> 97 # include <pwd.h> 98 # include <poll.h> 99 # include <fcntl.h> 100 # include <string.h> 101 # include <syscall.h> 102 # include <sys/sysinfo.h> 103 # include <gnu/libc-version.h> 104 # include <sys/ipc.h> 105 # include <sys/shm.h> 106 # include <link.h> 107 # include <stdint.h> 108 # include <inttypes.h> 109 # include <sys/ioctl.h> 110 111 #ifndef _GNU_SOURCE 112 #define _GNU_SOURCE 113 #include <sched.h> 114 #undef _GNU_SOURCE 115 #else 116 #include <sched.h> 117 #endif 118 119 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 120 // getrusage() is prepared to handle the associated failure. 121 #ifndef RUSAGE_THREAD 122 #define RUSAGE_THREAD (1) /* only the calling thread */ 123 #endif 124 125 #define MAX_PATH (2 * K) 126 127 #define MAX_SECS 100000000 128 129 // for timer info max values which include all bits 130 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 131 132 enum CoredumpFilterBit { 133 FILE_BACKED_PVT_BIT = 1 << 2, 134 FILE_BACKED_SHARED_BIT = 1 << 3, 135 LARGEPAGES_BIT = 1 << 6, 136 DAX_SHARED_BIT = 1 << 8 137 }; 138 139 //////////////////////////////////////////////////////////////////////////////// 140 // global variables 141 julong os::Linux::_physical_memory = 0; 142 143 address os::Linux::_initial_thread_stack_bottom = NULL; 144 uintptr_t os::Linux::_initial_thread_stack_size = 0; 145 146 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 147 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL; 148 Mutex* os::Linux::_createThread_lock = NULL; 149 pthread_t os::Linux::_main_thread; 150 int os::Linux::_page_size = -1; 151 bool os::Linux::_supports_fast_thread_cpu_time = false; 152 uint32_t os::Linux::_os_version = 0; 153 const char * os::Linux::_glibc_version = NULL; 154 const char * os::Linux::_libpthread_version = NULL; 155 156 static jlong initial_time_count=0; 157 158 static int clock_tics_per_sec = 100; 159 160 // If the VM might have been created on the primordial thread, we need to resolve the 161 // primordial thread stack bounds and check if the current thread might be the 162 // primordial thread in places. If we know that the primordial thread is never used, 163 // such as when the VM was created by one of the standard java launchers, we can 164 // avoid this 165 static bool suppress_primordial_thread_resolution = false; 166 167 // For diagnostics to print a message once. see run_periodic_checks 168 static sigset_t check_signal_done; 169 static bool check_signals = true; 170 171 // Signal number used to suspend/resume a thread 172 173 // do not use any signal number less than SIGSEGV, see 4355769 174 static int SR_signum = SIGUSR2; 175 sigset_t SR_sigset; 176 177 // utility functions 178 179 static int SR_initialize(); 180 181 julong os::available_memory() { 182 return Linux::available_memory(); 183 } 184 185 julong os::Linux::available_memory() { 186 // values in struct sysinfo are "unsigned long" 187 struct sysinfo si; 188 julong avail_mem; 189 190 if (OSContainer::is_containerized()) { 191 jlong mem_limit, mem_usage; 192 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) { 193 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value", 194 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit); 195 } 196 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) { 197 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage); 198 } 199 if (mem_limit > 0 && mem_usage > 0 ) { 200 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0; 201 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem); 202 return avail_mem; 203 } 204 } 205 206 sysinfo(&si); 207 avail_mem = (julong)si.freeram * si.mem_unit; 208 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem); 209 return avail_mem; 210 } 211 212 julong os::physical_memory() { 213 jlong phys_mem = 0; 214 if (OSContainer::is_containerized()) { 215 jlong mem_limit; 216 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) { 217 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit); 218 return mem_limit; 219 } 220 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value", 221 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit); 222 } 223 224 phys_mem = Linux::physical_memory(); 225 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem); 226 return phys_mem; 227 } 228 229 // Return true if user is running as root. 230 231 bool os::have_special_privileges() { 232 static bool init = false; 233 static bool privileges = false; 234 if (!init) { 235 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 236 init = true; 237 } 238 return privileges; 239 } 240 241 242 #ifndef SYS_gettid 243 // i386: 224, ia64: 1105, amd64: 186, sparc 143 244 #ifdef __ia64__ 245 #define SYS_gettid 1105 246 #else 247 #ifdef __i386__ 248 #define SYS_gettid 224 249 #else 250 #ifdef __amd64__ 251 #define SYS_gettid 186 252 #else 253 #ifdef __sparc__ 254 #define SYS_gettid 143 255 #else 256 #error define gettid for the arch 257 #endif 258 #endif 259 #endif 260 #endif 261 #endif 262 263 264 // pid_t gettid() 265 // 266 // Returns the kernel thread id of the currently running thread. Kernel 267 // thread id is used to access /proc. 268 pid_t os::Linux::gettid() { 269 int rslt = syscall(SYS_gettid); 270 assert(rslt != -1, "must be."); // old linuxthreads implementation? 271 return (pid_t)rslt; 272 } 273 274 // Most versions of linux have a bug where the number of processors are 275 // determined by looking at the /proc file system. In a chroot environment, 276 // the system call returns 1. 277 static bool unsafe_chroot_detected = false; 278 static const char *unstable_chroot_error = "/proc file system not found.\n" 279 "Java may be unstable running multithreaded in a chroot " 280 "environment on Linux when /proc filesystem is not mounted."; 281 282 void os::Linux::initialize_system_info() { 283 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 284 if (processor_count() == 1) { 285 pid_t pid = os::Linux::gettid(); 286 char fname[32]; 287 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 288 FILE *fp = fopen(fname, "r"); 289 if (fp == NULL) { 290 unsafe_chroot_detected = true; 291 } else { 292 fclose(fp); 293 } 294 } 295 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 296 assert(processor_count() > 0, "linux error"); 297 } 298 299 void os::init_system_properties_values() { 300 // The next steps are taken in the product version: 301 // 302 // Obtain the JAVA_HOME value from the location of libjvm.so. 303 // This library should be located at: 304 // <JAVA_HOME>/lib/{client|server}/libjvm.so. 305 // 306 // If "/jre/lib/" appears at the right place in the path, then we 307 // assume libjvm.so is installed in a JDK and we use this path. 308 // 309 // Otherwise exit with message: "Could not create the Java virtual machine." 310 // 311 // The following extra steps are taken in the debugging version: 312 // 313 // If "/jre/lib/" does NOT appear at the right place in the path 314 // instead of exit check for $JAVA_HOME environment variable. 315 // 316 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 317 // then we append a fake suffix "hotspot/libjvm.so" to this path so 318 // it looks like libjvm.so is installed there 319 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 320 // 321 // Otherwise exit. 322 // 323 // Important note: if the location of libjvm.so changes this 324 // code needs to be changed accordingly. 325 326 // See ld(1): 327 // The linker uses the following search paths to locate required 328 // shared libraries: 329 // 1: ... 330 // ... 331 // 7: The default directories, normally /lib and /usr/lib. 332 #ifndef OVERRIDE_LIBPATH 333 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390) 334 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 335 #else 336 #define DEFAULT_LIBPATH "/lib:/usr/lib" 337 #endif 338 #else 339 #define DEFAULT_LIBPATH OVERRIDE_LIBPATH 340 #endif 341 342 // Base path of extensions installed on the system. 343 #define SYS_EXT_DIR "/usr/java/packages" 344 #define EXTENSIONS_DIR "/lib/ext" 345 346 // Buffer that fits several sprintfs. 347 // Note that the space for the colon and the trailing null are provided 348 // by the nulls included by the sizeof operator. 349 const size_t bufsize = 350 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 351 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 352 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 353 354 // sysclasspath, java_home, dll_dir 355 { 356 char *pslash; 357 os::jvm_path(buf, bufsize); 358 359 // Found the full path to libjvm.so. 360 // Now cut the path to <java_home>/jre if we can. 361 pslash = strrchr(buf, '/'); 362 if (pslash != NULL) { 363 *pslash = '\0'; // Get rid of /libjvm.so. 364 } 365 pslash = strrchr(buf, '/'); 366 if (pslash != NULL) { 367 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 368 } 369 Arguments::set_dll_dir(buf); 370 371 if (pslash != NULL) { 372 pslash = strrchr(buf, '/'); 373 if (pslash != NULL) { 374 *pslash = '\0'; // Get rid of /lib. 375 } 376 } 377 Arguments::set_java_home(buf); 378 if (!set_boot_path('/', ':')) { 379 vm_exit_during_initialization("Failed setting boot class path.", NULL); 380 } 381 } 382 383 // Where to look for native libraries. 384 // 385 // Note: Due to a legacy implementation, most of the library path 386 // is set in the launcher. This was to accomodate linking restrictions 387 // on legacy Linux implementations (which are no longer supported). 388 // Eventually, all the library path setting will be done here. 389 // 390 // However, to prevent the proliferation of improperly built native 391 // libraries, the new path component /usr/java/packages is added here. 392 // Eventually, all the library path setting will be done here. 393 { 394 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 395 // should always exist (until the legacy problem cited above is 396 // addressed). 397 const char *v = ::getenv("LD_LIBRARY_PATH"); 398 const char *v_colon = ":"; 399 if (v == NULL) { v = ""; v_colon = ""; } 400 // That's +1 for the colon and +1 for the trailing '\0'. 401 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 402 strlen(v) + 1 + 403 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1, 404 mtInternal); 405 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon); 406 Arguments::set_library_path(ld_library_path); 407 FREE_C_HEAP_ARRAY(char, ld_library_path); 408 } 409 410 // Extensions directories. 411 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 412 Arguments::set_ext_dirs(buf); 413 414 FREE_C_HEAP_ARRAY(char, buf); 415 416 #undef DEFAULT_LIBPATH 417 #undef SYS_EXT_DIR 418 #undef EXTENSIONS_DIR 419 } 420 421 //////////////////////////////////////////////////////////////////////////////// 422 // breakpoint support 423 424 void os::breakpoint() { 425 BREAKPOINT; 426 } 427 428 extern "C" void breakpoint() { 429 // use debugger to set breakpoint here 430 } 431 432 //////////////////////////////////////////////////////////////////////////////// 433 // signal support 434 435 debug_only(static bool signal_sets_initialized = false); 436 static sigset_t unblocked_sigs, vm_sigs; 437 438 void os::Linux::signal_sets_init() { 439 // Should also have an assertion stating we are still single-threaded. 440 assert(!signal_sets_initialized, "Already initialized"); 441 // Fill in signals that are necessarily unblocked for all threads in 442 // the VM. Currently, we unblock the following signals: 443 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 444 // by -Xrs (=ReduceSignalUsage)); 445 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 446 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 447 // the dispositions or masks wrt these signals. 448 // Programs embedding the VM that want to use the above signals for their 449 // own purposes must, at this time, use the "-Xrs" option to prevent 450 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 451 // (See bug 4345157, and other related bugs). 452 // In reality, though, unblocking these signals is really a nop, since 453 // these signals are not blocked by default. 454 sigemptyset(&unblocked_sigs); 455 sigaddset(&unblocked_sigs, SIGILL); 456 sigaddset(&unblocked_sigs, SIGSEGV); 457 sigaddset(&unblocked_sigs, SIGBUS); 458 sigaddset(&unblocked_sigs, SIGFPE); 459 #if defined(PPC64) 460 sigaddset(&unblocked_sigs, SIGTRAP); 461 #endif 462 sigaddset(&unblocked_sigs, SR_signum); 463 464 if (!ReduceSignalUsage) { 465 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 466 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 467 } 468 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 469 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 470 } 471 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 472 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 473 } 474 } 475 // Fill in signals that are blocked by all but the VM thread. 476 sigemptyset(&vm_sigs); 477 if (!ReduceSignalUsage) { 478 sigaddset(&vm_sigs, BREAK_SIGNAL); 479 } 480 debug_only(signal_sets_initialized = true); 481 482 } 483 484 // These are signals that are unblocked while a thread is running Java. 485 // (For some reason, they get blocked by default.) 486 sigset_t* os::Linux::unblocked_signals() { 487 assert(signal_sets_initialized, "Not initialized"); 488 return &unblocked_sigs; 489 } 490 491 // These are the signals that are blocked while a (non-VM) thread is 492 // running Java. Only the VM thread handles these signals. 493 sigset_t* os::Linux::vm_signals() { 494 assert(signal_sets_initialized, "Not initialized"); 495 return &vm_sigs; 496 } 497 498 void os::Linux::hotspot_sigmask(Thread* thread) { 499 500 //Save caller's signal mask before setting VM signal mask 501 sigset_t caller_sigmask; 502 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 503 504 OSThread* osthread = thread->osthread(); 505 osthread->set_caller_sigmask(caller_sigmask); 506 507 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 508 509 if (!ReduceSignalUsage) { 510 if (thread->is_VM_thread()) { 511 // Only the VM thread handles BREAK_SIGNAL ... 512 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 513 } else { 514 // ... all other threads block BREAK_SIGNAL 515 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 516 } 517 } 518 } 519 520 ////////////////////////////////////////////////////////////////////////////// 521 // detecting pthread library 522 523 void os::Linux::libpthread_init() { 524 // Save glibc and pthread version strings. 525 #if !defined(_CS_GNU_LIBC_VERSION) || \ 526 !defined(_CS_GNU_LIBPTHREAD_VERSION) 527 #error "glibc too old (< 2.3.2)" 528 #endif 529 530 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 531 assert(n > 0, "cannot retrieve glibc version"); 532 char *str = (char *)malloc(n, mtInternal); 533 confstr(_CS_GNU_LIBC_VERSION, str, n); 534 os::Linux::set_glibc_version(str); 535 536 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 537 assert(n > 0, "cannot retrieve pthread version"); 538 str = (char *)malloc(n, mtInternal); 539 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 540 os::Linux::set_libpthread_version(str); 541 } 542 543 ///////////////////////////////////////////////////////////////////////////// 544 // thread stack expansion 545 546 // os::Linux::manually_expand_stack() takes care of expanding the thread 547 // stack. Note that this is normally not needed: pthread stacks allocate 548 // thread stack using mmap() without MAP_NORESERVE, so the stack is already 549 // committed. Therefore it is not necessary to expand the stack manually. 550 // 551 // Manually expanding the stack was historically needed on LinuxThreads 552 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays 553 // it is kept to deal with very rare corner cases: 554 // 555 // For one, user may run the VM on an own implementation of threads 556 // whose stacks are - like the old LinuxThreads - implemented using 557 // mmap(MAP_GROWSDOWN). 558 // 559 // Also, this coding may be needed if the VM is running on the primordial 560 // thread. Normally we avoid running on the primordial thread; however, 561 // user may still invoke the VM on the primordial thread. 562 // 563 // The following historical comment describes the details about running 564 // on a thread stack allocated with mmap(MAP_GROWSDOWN): 565 566 567 // Force Linux kernel to expand current thread stack. If "bottom" is close 568 // to the stack guard, caller should block all signals. 569 // 570 // MAP_GROWSDOWN: 571 // A special mmap() flag that is used to implement thread stacks. It tells 572 // kernel that the memory region should extend downwards when needed. This 573 // allows early versions of LinuxThreads to only mmap the first few pages 574 // when creating a new thread. Linux kernel will automatically expand thread 575 // stack as needed (on page faults). 576 // 577 // However, because the memory region of a MAP_GROWSDOWN stack can grow on 578 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 579 // region, it's hard to tell if the fault is due to a legitimate stack 580 // access or because of reading/writing non-exist memory (e.g. buffer 581 // overrun). As a rule, if the fault happens below current stack pointer, 582 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the 583 // application (see Linux kernel fault.c). 584 // 585 // This Linux feature can cause SIGSEGV when VM bangs thread stack for 586 // stack overflow detection. 587 // 588 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 589 // not use MAP_GROWSDOWN. 590 // 591 // To get around the problem and allow stack banging on Linux, we need to 592 // manually expand thread stack after receiving the SIGSEGV. 593 // 594 // There are two ways to expand thread stack to address "bottom", we used 595 // both of them in JVM before 1.5: 596 // 1. adjust stack pointer first so that it is below "bottom", and then 597 // touch "bottom" 598 // 2. mmap() the page in question 599 // 600 // Now alternate signal stack is gone, it's harder to use 2. For instance, 601 // if current sp is already near the lower end of page 101, and we need to 602 // call mmap() to map page 100, it is possible that part of the mmap() frame 603 // will be placed in page 100. When page 100 is mapped, it is zero-filled. 604 // That will destroy the mmap() frame and cause VM to crash. 605 // 606 // The following code works by adjusting sp first, then accessing the "bottom" 607 // page to force a page fault. Linux kernel will then automatically expand the 608 // stack mapping. 609 // 610 // _expand_stack_to() assumes its frame size is less than page size, which 611 // should always be true if the function is not inlined. 612 613 static void NOINLINE _expand_stack_to(address bottom) { 614 address sp; 615 size_t size; 616 volatile char *p; 617 618 // Adjust bottom to point to the largest address within the same page, it 619 // gives us a one-page buffer if alloca() allocates slightly more memory. 620 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size()); 621 bottom += os::Linux::page_size() - 1; 622 623 // sp might be slightly above current stack pointer; if that's the case, we 624 // will alloca() a little more space than necessary, which is OK. Don't use 625 // os::current_stack_pointer(), as its result can be slightly below current 626 // stack pointer, causing us to not alloca enough to reach "bottom". 627 sp = (address)&sp; 628 629 if (sp > bottom) { 630 size = sp - bottom; 631 p = (volatile char *)alloca(size); 632 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 633 p[0] = '\0'; 634 } 635 } 636 637 void os::Linux::expand_stack_to(address bottom) { 638 _expand_stack_to(bottom); 639 } 640 641 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 642 assert(t!=NULL, "just checking"); 643 assert(t->osthread()->expanding_stack(), "expand should be set"); 644 assert(t->stack_base() != NULL, "stack_base was not initialized"); 645 646 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) { 647 sigset_t mask_all, old_sigset; 648 sigfillset(&mask_all); 649 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 650 _expand_stack_to(addr); 651 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 652 return true; 653 } 654 return false; 655 } 656 657 ////////////////////////////////////////////////////////////////////////////// 658 // create new thread 659 660 // Thread start routine for all newly created threads 661 static void *thread_native_entry(Thread *thread) { 662 663 thread->record_stack_base_and_size(); 664 665 // Try to randomize the cache line index of hot stack frames. 666 // This helps when threads of the same stack traces evict each other's 667 // cache lines. The threads can be either from the same JVM instance, or 668 // from different JVM instances. The benefit is especially true for 669 // processors with hyperthreading technology. 670 static int counter = 0; 671 int pid = os::current_process_id(); 672 alloca(((pid ^ counter++) & 7) * 128); 673 674 thread->initialize_thread_current(); 675 676 OSThread* osthread = thread->osthread(); 677 Monitor* sync = osthread->startThread_lock(); 678 679 osthread->set_thread_id(os::current_thread_id()); 680 681 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 682 os::current_thread_id(), (uintx) pthread_self()); 683 684 if (UseNUMA) { 685 int lgrp_id = os::numa_get_group_id(); 686 if (lgrp_id != -1) { 687 thread->set_lgrp_id(lgrp_id); 688 } 689 } 690 // initialize signal mask for this thread 691 os::Linux::hotspot_sigmask(thread); 692 693 // initialize floating point control register 694 os::Linux::init_thread_fpu_state(); 695 696 // handshaking with parent thread 697 { 698 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 699 700 // notify parent thread 701 osthread->set_state(INITIALIZED); 702 sync->notify_all(); 703 704 // wait until os::start_thread() 705 while (osthread->get_state() == INITIALIZED) { 706 sync->wait(Mutex::_no_safepoint_check_flag); 707 } 708 } 709 710 assert(osthread->pthread_id() != 0, "pthread_id was not set as expected"); 711 712 // call one more level start routine 713 thread->call_run(); 714 715 // Note: at this point the thread object may already have deleted itself. 716 // Prevent dereferencing it from here on out. 717 thread = NULL; 718 719 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 720 os::current_thread_id(), (uintx) pthread_self()); 721 722 return 0; 723 } 724 725 bool os::create_thread(Thread* thread, ThreadType thr_type, 726 size_t req_stack_size) { 727 assert(thread->osthread() == NULL, "caller responsible"); 728 729 // Allocate the OSThread object 730 OSThread* osthread = new OSThread(NULL, NULL); 731 if (osthread == NULL) { 732 return false; 733 } 734 735 // set the correct thread state 736 osthread->set_thread_type(thr_type); 737 738 // Initial state is ALLOCATED but not INITIALIZED 739 osthread->set_state(ALLOCATED); 740 741 thread->set_osthread(osthread); 742 743 // init thread attributes 744 pthread_attr_t attr; 745 pthread_attr_init(&attr); 746 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 747 748 // Calculate stack size if it's not specified by caller. 749 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size); 750 // In the Linux NPTL pthread implementation the guard size mechanism 751 // is not implemented properly. The posix standard requires adding 752 // the size of the guard pages to the stack size, instead Linux 753 // takes the space out of 'stacksize'. Thus we adapt the requested 754 // stack_size by the size of the guard pages to mimick proper 755 // behaviour. However, be careful not to end up with a size 756 // of zero due to overflow. Don't add the guard page in that case. 757 size_t guard_size = os::Linux::default_guard_size(thr_type); 758 if (stack_size <= SIZE_MAX - guard_size) { 759 stack_size += guard_size; 760 } 761 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned"); 762 763 int status = pthread_attr_setstacksize(&attr, stack_size); 764 assert_status(status == 0, status, "pthread_attr_setstacksize"); 765 766 // Configure glibc guard page. 767 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 768 769 ThreadState state; 770 771 { 772 pthread_t tid; 773 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread); 774 775 char buf[64]; 776 if (ret == 0) { 777 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ", 778 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 779 } else { 780 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.", 781 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 782 } 783 784 pthread_attr_destroy(&attr); 785 786 if (ret != 0) { 787 // Need to clean up stuff we've allocated so far 788 thread->set_osthread(NULL); 789 delete osthread; 790 return false; 791 } 792 793 // Store pthread info into the OSThread 794 osthread->set_pthread_id(tid); 795 796 // Wait until child thread is either initialized or aborted 797 { 798 Monitor* sync_with_child = osthread->startThread_lock(); 799 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 800 while ((state = osthread->get_state()) == ALLOCATED) { 801 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 802 } 803 } 804 } 805 806 // Aborted due to thread limit being reached 807 if (state == ZOMBIE) { 808 thread->set_osthread(NULL); 809 delete osthread; 810 return false; 811 } 812 813 // The thread is returned suspended (in state INITIALIZED), 814 // and is started higher up in the call chain 815 assert(state == INITIALIZED, "race condition"); 816 return true; 817 } 818 819 ///////////////////////////////////////////////////////////////////////////// 820 // attach existing thread 821 822 // bootstrap the main thread 823 bool os::create_main_thread(JavaThread* thread) { 824 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 825 return create_attached_thread(thread); 826 } 827 828 bool os::create_attached_thread(JavaThread* thread) { 829 #ifdef ASSERT 830 thread->verify_not_published(); 831 #endif 832 833 // Allocate the OSThread object 834 OSThread* osthread = new OSThread(NULL, NULL); 835 836 if (osthread == NULL) { 837 return false; 838 } 839 840 // Store pthread info into the OSThread 841 osthread->set_thread_id(os::Linux::gettid()); 842 osthread->set_pthread_id(::pthread_self()); 843 844 // initialize floating point control register 845 os::Linux::init_thread_fpu_state(); 846 847 // Initial thread state is RUNNABLE 848 osthread->set_state(RUNNABLE); 849 850 thread->set_osthread(osthread); 851 852 if (UseNUMA) { 853 int lgrp_id = os::numa_get_group_id(); 854 if (lgrp_id != -1) { 855 thread->set_lgrp_id(lgrp_id); 856 } 857 } 858 859 if (os::is_primordial_thread()) { 860 // If current thread is primordial thread, its stack is mapped on demand, 861 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 862 // the entire stack region to avoid SEGV in stack banging. 863 // It is also useful to get around the heap-stack-gap problem on SuSE 864 // kernel (see 4821821 for details). We first expand stack to the top 865 // of yellow zone, then enable stack yellow zone (order is significant, 866 // enabling yellow zone first will crash JVM on SuSE Linux), so there 867 // is no gap between the last two virtual memory regions. 868 869 JavaThread *jt = (JavaThread *)thread; 870 address addr = jt->stack_reserved_zone_base(); 871 assert(addr != NULL, "initialization problem?"); 872 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 873 874 osthread->set_expanding_stack(); 875 os::Linux::manually_expand_stack(jt, addr); 876 osthread->clear_expanding_stack(); 877 } 878 879 // initialize signal mask for this thread 880 // and save the caller's signal mask 881 os::Linux::hotspot_sigmask(thread); 882 883 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 884 os::current_thread_id(), (uintx) pthread_self()); 885 886 return true; 887 } 888 889 void os::pd_start_thread(Thread* thread) { 890 OSThread * osthread = thread->osthread(); 891 assert(osthread->get_state() != INITIALIZED, "just checking"); 892 Monitor* sync_with_child = osthread->startThread_lock(); 893 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 894 sync_with_child->notify(); 895 } 896 897 // Free Linux resources related to the OSThread 898 void os::free_thread(OSThread* osthread) { 899 assert(osthread != NULL, "osthread not set"); 900 901 // We are told to free resources of the argument thread, 902 // but we can only really operate on the current thread. 903 assert(Thread::current()->osthread() == osthread, 904 "os::free_thread but not current thread"); 905 906 #ifdef ASSERT 907 sigset_t current; 908 sigemptyset(¤t); 909 pthread_sigmask(SIG_SETMASK, NULL, ¤t); 910 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!"); 911 #endif 912 913 // Restore caller's signal mask 914 sigset_t sigmask = osthread->caller_sigmask(); 915 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 916 917 delete osthread; 918 } 919 920 ////////////////////////////////////////////////////////////////////////////// 921 // primordial thread 922 923 // Check if current thread is the primordial thread, similar to Solaris thr_main. 924 bool os::is_primordial_thread(void) { 925 if (suppress_primordial_thread_resolution) { 926 return false; 927 } 928 char dummy; 929 // If called before init complete, thread stack bottom will be null. 930 // Can be called if fatal error occurs before initialization. 931 if (os::Linux::initial_thread_stack_bottom() == NULL) return false; 932 assert(os::Linux::initial_thread_stack_bottom() != NULL && 933 os::Linux::initial_thread_stack_size() != 0, 934 "os::init did not locate primordial thread's stack region"); 935 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() && 936 (address)&dummy < os::Linux::initial_thread_stack_bottom() + 937 os::Linux::initial_thread_stack_size()) { 938 return true; 939 } else { 940 return false; 941 } 942 } 943 944 // Find the virtual memory area that contains addr 945 static bool find_vma(address addr, address* vma_low, address* vma_high) { 946 FILE *fp = fopen("/proc/self/maps", "r"); 947 if (fp) { 948 address low, high; 949 while (!feof(fp)) { 950 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 951 if (low <= addr && addr < high) { 952 if (vma_low) *vma_low = low; 953 if (vma_high) *vma_high = high; 954 fclose(fp); 955 return true; 956 } 957 } 958 for (;;) { 959 int ch = fgetc(fp); 960 if (ch == EOF || ch == (int)'\n') break; 961 } 962 } 963 fclose(fp); 964 } 965 return false; 966 } 967 968 // Locate primordial thread stack. This special handling of primordial thread stack 969 // is needed because pthread_getattr_np() on most (all?) Linux distros returns 970 // bogus value for the primordial process thread. While the launcher has created 971 // the VM in a new thread since JDK 6, we still have to allow for the use of the 972 // JNI invocation API from a primordial thread. 973 void os::Linux::capture_initial_stack(size_t max_size) { 974 975 // max_size is either 0 (which means accept OS default for thread stacks) or 976 // a user-specified value known to be at least the minimum needed. If we 977 // are actually on the primordial thread we can make it appear that we have a 978 // smaller max_size stack by inserting the guard pages at that location. But we 979 // cannot do anything to emulate a larger stack than what has been provided by 980 // the OS or threading library. In fact if we try to use a stack greater than 981 // what is set by rlimit then we will crash the hosting process. 982 983 // Maximum stack size is the easy part, get it from RLIMIT_STACK. 984 // If this is "unlimited" then it will be a huge value. 985 struct rlimit rlim; 986 getrlimit(RLIMIT_STACK, &rlim); 987 size_t stack_size = rlim.rlim_cur; 988 989 // 6308388: a bug in ld.so will relocate its own .data section to the 990 // lower end of primordial stack; reduce ulimit -s value a little bit 991 // so we won't install guard page on ld.so's data section. 992 // But ensure we don't underflow the stack size - allow 1 page spare 993 if (stack_size >= (size_t)(3 * page_size())) { 994 stack_size -= 2 * page_size(); 995 } 996 997 // Try to figure out where the stack base (top) is. This is harder. 998 // 999 // When an application is started, glibc saves the initial stack pointer in 1000 // a global variable "__libc_stack_end", which is then used by system 1001 // libraries. __libc_stack_end should be pretty close to stack top. The 1002 // variable is available since the very early days. However, because it is 1003 // a private interface, it could disappear in the future. 1004 // 1005 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1006 // to __libc_stack_end, it is very close to stack top, but isn't the real 1007 // stack top. Note that /proc may not exist if VM is running as a chroot 1008 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1009 // /proc/<pid>/stat could change in the future (though unlikely). 1010 // 1011 // We try __libc_stack_end first. If that doesn't work, look for 1012 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1013 // as a hint, which should work well in most cases. 1014 1015 uintptr_t stack_start; 1016 1017 // try __libc_stack_end first 1018 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1019 if (p && *p) { 1020 stack_start = *p; 1021 } else { 1022 // see if we can get the start_stack field from /proc/self/stat 1023 FILE *fp; 1024 int pid; 1025 char state; 1026 int ppid; 1027 int pgrp; 1028 int session; 1029 int nr; 1030 int tpgrp; 1031 unsigned long flags; 1032 unsigned long minflt; 1033 unsigned long cminflt; 1034 unsigned long majflt; 1035 unsigned long cmajflt; 1036 unsigned long utime; 1037 unsigned long stime; 1038 long cutime; 1039 long cstime; 1040 long prio; 1041 long nice; 1042 long junk; 1043 long it_real; 1044 uintptr_t start; 1045 uintptr_t vsize; 1046 intptr_t rss; 1047 uintptr_t rsslim; 1048 uintptr_t scodes; 1049 uintptr_t ecode; 1050 int i; 1051 1052 // Figure what the primordial thread stack base is. Code is inspired 1053 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1054 // followed by command name surrounded by parentheses, state, etc. 1055 char stat[2048]; 1056 int statlen; 1057 1058 fp = fopen("/proc/self/stat", "r"); 1059 if (fp) { 1060 statlen = fread(stat, 1, 2047, fp); 1061 stat[statlen] = '\0'; 1062 fclose(fp); 1063 1064 // Skip pid and the command string. Note that we could be dealing with 1065 // weird command names, e.g. user could decide to rename java launcher 1066 // to "java 1.4.2 :)", then the stat file would look like 1067 // 1234 (java 1.4.2 :)) R ... ... 1068 // We don't really need to know the command string, just find the last 1069 // occurrence of ")" and then start parsing from there. See bug 4726580. 1070 char * s = strrchr(stat, ')'); 1071 1072 i = 0; 1073 if (s) { 1074 // Skip blank chars 1075 do { s++; } while (s && isspace(*s)); 1076 1077 #define _UFM UINTX_FORMAT 1078 #define _DFM INTX_FORMAT 1079 1080 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1081 // 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 1082 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, 1083 &state, // 3 %c 1084 &ppid, // 4 %d 1085 &pgrp, // 5 %d 1086 &session, // 6 %d 1087 &nr, // 7 %d 1088 &tpgrp, // 8 %d 1089 &flags, // 9 %lu 1090 &minflt, // 10 %lu 1091 &cminflt, // 11 %lu 1092 &majflt, // 12 %lu 1093 &cmajflt, // 13 %lu 1094 &utime, // 14 %lu 1095 &stime, // 15 %lu 1096 &cutime, // 16 %ld 1097 &cstime, // 17 %ld 1098 &prio, // 18 %ld 1099 &nice, // 19 %ld 1100 &junk, // 20 %ld 1101 &it_real, // 21 %ld 1102 &start, // 22 UINTX_FORMAT 1103 &vsize, // 23 UINTX_FORMAT 1104 &rss, // 24 INTX_FORMAT 1105 &rsslim, // 25 UINTX_FORMAT 1106 &scodes, // 26 UINTX_FORMAT 1107 &ecode, // 27 UINTX_FORMAT 1108 &stack_start); // 28 UINTX_FORMAT 1109 } 1110 1111 #undef _UFM 1112 #undef _DFM 1113 1114 if (i != 28 - 2) { 1115 assert(false, "Bad conversion from /proc/self/stat"); 1116 // product mode - assume we are the primordial thread, good luck in the 1117 // embedded case. 1118 warning("Can't detect primordial thread stack location - bad conversion"); 1119 stack_start = (uintptr_t) &rlim; 1120 } 1121 } else { 1122 // For some reason we can't open /proc/self/stat (for example, running on 1123 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1124 // most cases, so don't abort: 1125 warning("Can't detect primordial thread stack location - no /proc/self/stat"); 1126 stack_start = (uintptr_t) &rlim; 1127 } 1128 } 1129 1130 // Now we have a pointer (stack_start) very close to the stack top, the 1131 // next thing to do is to figure out the exact location of stack top. We 1132 // can find out the virtual memory area that contains stack_start by 1133 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1134 // and its upper limit is the real stack top. (again, this would fail if 1135 // running inside chroot, because /proc may not exist.) 1136 1137 uintptr_t stack_top; 1138 address low, high; 1139 if (find_vma((address)stack_start, &low, &high)) { 1140 // success, "high" is the true stack top. (ignore "low", because initial 1141 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1142 stack_top = (uintptr_t)high; 1143 } else { 1144 // failed, likely because /proc/self/maps does not exist 1145 warning("Can't detect primordial thread stack location - find_vma failed"); 1146 // best effort: stack_start is normally within a few pages below the real 1147 // stack top, use it as stack top, and reduce stack size so we won't put 1148 // guard page outside stack. 1149 stack_top = stack_start; 1150 stack_size -= 16 * page_size(); 1151 } 1152 1153 // stack_top could be partially down the page so align it 1154 stack_top = align_up(stack_top, page_size()); 1155 1156 // Allowed stack value is minimum of max_size and what we derived from rlimit 1157 if (max_size > 0) { 1158 _initial_thread_stack_size = MIN2(max_size, stack_size); 1159 } else { 1160 // Accept the rlimit max, but if stack is unlimited then it will be huge, so 1161 // clamp it at 8MB as we do on Solaris 1162 _initial_thread_stack_size = MIN2(stack_size, 8*M); 1163 } 1164 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size()); 1165 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1166 1167 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!"); 1168 1169 if (log_is_enabled(Info, os, thread)) { 1170 // See if we seem to be on primordial process thread 1171 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) && 1172 uintptr_t(&rlim) < stack_top; 1173 1174 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: " 1175 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT, 1176 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K, 1177 stack_top, intptr_t(_initial_thread_stack_bottom)); 1178 } 1179 } 1180 1181 //////////////////////////////////////////////////////////////////////////////// 1182 // time support 1183 1184 #ifndef SUPPORTS_CLOCK_MONOTONIC 1185 #error "Build platform doesn't support clock_gettime and related functionality" 1186 #endif 1187 1188 // Time since start-up in seconds to a fine granularity. 1189 // Used by VMSelfDestructTimer and the MemProfiler. 1190 double os::elapsedTime() { 1191 1192 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1193 } 1194 1195 jlong os::elapsed_counter() { 1196 return javaTimeNanos() - initial_time_count; 1197 } 1198 1199 jlong os::elapsed_frequency() { 1200 return NANOSECS_PER_SEC; // nanosecond resolution 1201 } 1202 1203 bool os::supports_vtime() { return true; } 1204 bool os::enable_vtime() { return false; } 1205 bool os::vtime_enabled() { return false; } 1206 1207 double os::elapsedVTime() { 1208 struct rusage usage; 1209 int retval = getrusage(RUSAGE_THREAD, &usage); 1210 if (retval == 0) { 1211 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); 1212 } else { 1213 // better than nothing, but not much 1214 return elapsedTime(); 1215 } 1216 } 1217 1218 jlong os::javaTimeMillis() { 1219 timeval time; 1220 int status = gettimeofday(&time, NULL); 1221 assert(status != -1, "linux error"); 1222 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1223 } 1224 1225 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) { 1226 timeval time; 1227 int status = gettimeofday(&time, NULL); 1228 assert(status != -1, "linux error"); 1229 seconds = jlong(time.tv_sec); 1230 nanos = jlong(time.tv_usec) * 1000; 1231 } 1232 1233 void os::Linux::fast_thread_clock_init() { 1234 if (!UseLinuxPosixThreadCPUClocks) { 1235 return; 1236 } 1237 clockid_t clockid; 1238 struct timespec tp; 1239 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1240 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1241 1242 // Switch to using fast clocks for thread cpu time if 1243 // the clock_getres() returns 0 error code. 1244 // Note, that some kernels may support the current thread 1245 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1246 // returned by the pthread_getcpuclockid(). 1247 // If the fast Posix clocks are supported then the clock_getres() 1248 // must return at least tp.tv_sec == 0 which means a resolution 1249 // better than 1 sec. This is extra check for reliability. 1250 1251 if (pthread_getcpuclockid_func && 1252 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1253 os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1254 _supports_fast_thread_cpu_time = true; 1255 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1256 } 1257 } 1258 1259 jlong os::javaTimeNanos() { 1260 if (os::supports_monotonic_clock()) { 1261 struct timespec tp; 1262 int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp); 1263 assert(status == 0, "gettime error"); 1264 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1265 return result; 1266 } else { 1267 timeval time; 1268 int status = gettimeofday(&time, NULL); 1269 assert(status != -1, "linux error"); 1270 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1271 return 1000 * usecs; 1272 } 1273 } 1274 1275 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1276 if (os::supports_monotonic_clock()) { 1277 info_ptr->max_value = ALL_64_BITS; 1278 1279 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1280 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1281 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1282 } else { 1283 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1284 info_ptr->max_value = ALL_64_BITS; 1285 1286 // gettimeofday is a real time clock so it skips 1287 info_ptr->may_skip_backward = true; 1288 info_ptr->may_skip_forward = true; 1289 } 1290 1291 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1292 } 1293 1294 // Return the real, user, and system times in seconds from an 1295 // arbitrary fixed point in the past. 1296 bool os::getTimesSecs(double* process_real_time, 1297 double* process_user_time, 1298 double* process_system_time) { 1299 struct tms ticks; 1300 clock_t real_ticks = times(&ticks); 1301 1302 if (real_ticks == (clock_t) (-1)) { 1303 return false; 1304 } else { 1305 double ticks_per_second = (double) clock_tics_per_sec; 1306 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1307 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1308 *process_real_time = ((double) real_ticks) / ticks_per_second; 1309 1310 return true; 1311 } 1312 } 1313 1314 1315 char * os::local_time_string(char *buf, size_t buflen) { 1316 struct tm t; 1317 time_t long_time; 1318 time(&long_time); 1319 localtime_r(&long_time, &t); 1320 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1321 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1322 t.tm_hour, t.tm_min, t.tm_sec); 1323 return buf; 1324 } 1325 1326 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1327 return localtime_r(clock, res); 1328 } 1329 1330 //////////////////////////////////////////////////////////////////////////////// 1331 // runtime exit support 1332 1333 // Note: os::shutdown() might be called very early during initialization, or 1334 // called from signal handler. Before adding something to os::shutdown(), make 1335 // sure it is async-safe and can handle partially initialized VM. 1336 void os::shutdown() { 1337 1338 // allow PerfMemory to attempt cleanup of any persistent resources 1339 perfMemory_exit(); 1340 1341 // needs to remove object in file system 1342 AttachListener::abort(); 1343 1344 // flush buffered output, finish log files 1345 ostream_abort(); 1346 1347 // Check for abort hook 1348 abort_hook_t abort_hook = Arguments::abort_hook(); 1349 if (abort_hook != NULL) { 1350 abort_hook(); 1351 } 1352 1353 } 1354 1355 // Note: os::abort() might be called very early during initialization, or 1356 // called from signal handler. Before adding something to os::abort(), make 1357 // sure it is async-safe and can handle partially initialized VM. 1358 void os::abort(bool dump_core, void* siginfo, const void* context) { 1359 os::shutdown(); 1360 if (dump_core) { 1361 #ifndef ZERO 1362 if (DumpPrivateMappingsInCore) { 1363 ClassLoader::close_jrt_image(); 1364 } 1365 #endif 1366 #ifndef PRODUCT 1367 fdStream out(defaultStream::output_fd()); 1368 out.print_raw("Current thread is "); 1369 char buf[16]; 1370 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1371 out.print_raw_cr(buf); 1372 out.print_raw_cr("Dumping core ..."); 1373 #endif 1374 ::abort(); // dump core 1375 } 1376 1377 ::exit(1); 1378 } 1379 1380 // Die immediately, no exit hook, no abort hook, no cleanup. 1381 void os::die() { 1382 ::abort(); 1383 } 1384 1385 // thread_id is kernel thread id (similar to Solaris LWP id) 1386 intx os::current_thread_id() { return os::Linux::gettid(); } 1387 int os::current_process_id() { 1388 return ::getpid(); 1389 } 1390 1391 // DLL functions 1392 1393 const char* os::dll_file_extension() { return ".so"; } 1394 1395 // This must be hard coded because it's the system's temporary 1396 // directory not the java application's temp directory, ala java.io.tmpdir. 1397 const char* os::get_temp_directory() { return "/tmp"; } 1398 1399 static bool file_exists(const char* filename) { 1400 struct stat statbuf; 1401 if (filename == NULL || strlen(filename) == 0) { 1402 return false; 1403 } 1404 return os::stat(filename, &statbuf) == 0; 1405 } 1406 1407 // check if addr is inside libjvm.so 1408 bool os::address_is_in_vm(address addr) { 1409 static address libjvm_base_addr; 1410 Dl_info dlinfo; 1411 1412 if (libjvm_base_addr == NULL) { 1413 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1414 libjvm_base_addr = (address)dlinfo.dli_fbase; 1415 } 1416 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1417 } 1418 1419 if (dladdr((void *)addr, &dlinfo) != 0) { 1420 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1421 } 1422 1423 return false; 1424 } 1425 1426 bool os::dll_address_to_function_name(address addr, char *buf, 1427 int buflen, int *offset, 1428 bool demangle) { 1429 // buf is not optional, but offset is optional 1430 assert(buf != NULL, "sanity check"); 1431 1432 Dl_info dlinfo; 1433 1434 if (dladdr((void*)addr, &dlinfo) != 0) { 1435 // see if we have a matching symbol 1436 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1437 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) { 1438 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1439 } 1440 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1441 return true; 1442 } 1443 // no matching symbol so try for just file info 1444 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1445 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1446 buf, buflen, offset, dlinfo.dli_fname, demangle)) { 1447 return true; 1448 } 1449 } 1450 } 1451 1452 buf[0] = '\0'; 1453 if (offset != NULL) *offset = -1; 1454 return false; 1455 } 1456 1457 struct _address_to_library_name { 1458 address addr; // input : memory address 1459 size_t buflen; // size of fname 1460 char* fname; // output: library name 1461 address base; // library base addr 1462 }; 1463 1464 static int address_to_library_name_callback(struct dl_phdr_info *info, 1465 size_t size, void *data) { 1466 int i; 1467 bool found = false; 1468 address libbase = NULL; 1469 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1470 1471 // iterate through all loadable segments 1472 for (i = 0; i < info->dlpi_phnum; i++) { 1473 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1474 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1475 // base address of a library is the lowest address of its loaded 1476 // segments. 1477 if (libbase == NULL || libbase > segbase) { 1478 libbase = segbase; 1479 } 1480 // see if 'addr' is within current segment 1481 if (segbase <= d->addr && 1482 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1483 found = true; 1484 } 1485 } 1486 } 1487 1488 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1489 // so dll_address_to_library_name() can fall through to use dladdr() which 1490 // can figure out executable name from argv[0]. 1491 if (found && info->dlpi_name && info->dlpi_name[0]) { 1492 d->base = libbase; 1493 if (d->fname) { 1494 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1495 } 1496 return 1; 1497 } 1498 return 0; 1499 } 1500 1501 bool os::dll_address_to_library_name(address addr, char* buf, 1502 int buflen, int* offset) { 1503 // buf is not optional, but offset is optional 1504 assert(buf != NULL, "sanity check"); 1505 1506 Dl_info dlinfo; 1507 struct _address_to_library_name data; 1508 1509 // There is a bug in old glibc dladdr() implementation that it could resolve 1510 // to wrong library name if the .so file has a base address != NULL. Here 1511 // we iterate through the program headers of all loaded libraries to find 1512 // out which library 'addr' really belongs to. This workaround can be 1513 // removed once the minimum requirement for glibc is moved to 2.3.x. 1514 data.addr = addr; 1515 data.fname = buf; 1516 data.buflen = buflen; 1517 data.base = NULL; 1518 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1519 1520 if (rslt) { 1521 // buf already contains library name 1522 if (offset) *offset = addr - data.base; 1523 return true; 1524 } 1525 if (dladdr((void*)addr, &dlinfo) != 0) { 1526 if (dlinfo.dli_fname != NULL) { 1527 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1528 } 1529 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1530 *offset = addr - (address)dlinfo.dli_fbase; 1531 } 1532 return true; 1533 } 1534 1535 buf[0] = '\0'; 1536 if (offset) *offset = -1; 1537 return false; 1538 } 1539 1540 // Loads .dll/.so and 1541 // in case of error it checks if .dll/.so was built for the 1542 // same architecture as Hotspot is running on 1543 1544 1545 // Remember the stack's state. The Linux dynamic linker will change 1546 // the stack to 'executable' at most once, so we must safepoint only once. 1547 bool os::Linux::_stack_is_executable = false; 1548 1549 // VM operation that loads a library. This is necessary if stack protection 1550 // of the Java stacks can be lost during loading the library. If we 1551 // do not stop the Java threads, they can stack overflow before the stacks 1552 // are protected again. 1553 class VM_LinuxDllLoad: public VM_Operation { 1554 private: 1555 const char *_filename; 1556 char *_ebuf; 1557 int _ebuflen; 1558 void *_lib; 1559 public: 1560 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1561 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1562 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1563 void doit() { 1564 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1565 os::Linux::_stack_is_executable = true; 1566 } 1567 void* loaded_library() { return _lib; } 1568 }; 1569 1570 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1571 void * result = NULL; 1572 bool load_attempted = false; 1573 1574 // Check whether the library to load might change execution rights 1575 // of the stack. If they are changed, the protection of the stack 1576 // guard pages will be lost. We need a safepoint to fix this. 1577 // 1578 // See Linux man page execstack(8) for more info. 1579 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1580 if (!ElfFile::specifies_noexecstack(filename)) { 1581 if (!is_init_completed()) { 1582 os::Linux::_stack_is_executable = true; 1583 // This is OK - No Java threads have been created yet, and hence no 1584 // stack guard pages to fix. 1585 // 1586 // Dynamic loader will make all stacks executable after 1587 // this function returns, and will not do that again. 1588 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet."); 1589 } else { 1590 warning("You have loaded library %s which might have disabled stack guard. " 1591 "The VM will try to fix the stack guard now.\n" 1592 "It's highly recommended that you fix the library with " 1593 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1594 filename); 1595 1596 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1597 JavaThread *jt = JavaThread::current(); 1598 if (jt->thread_state() != _thread_in_native) { 1599 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1600 // that requires ExecStack. Cannot enter safe point. Let's give up. 1601 warning("Unable to fix stack guard. Giving up."); 1602 } else { 1603 if (!LoadExecStackDllInVMThread) { 1604 // This is for the case where the DLL has an static 1605 // constructor function that executes JNI code. We cannot 1606 // load such DLLs in the VMThread. 1607 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1608 } 1609 1610 ThreadInVMfromNative tiv(jt); 1611 debug_only(VMNativeEntryWrapper vew;) 1612 1613 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1614 VMThread::execute(&op); 1615 if (LoadExecStackDllInVMThread) { 1616 result = op.loaded_library(); 1617 } 1618 load_attempted = true; 1619 } 1620 } 1621 } 1622 } 1623 1624 if (!load_attempted) { 1625 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1626 } 1627 1628 if (result != NULL) { 1629 // Successful loading 1630 return result; 1631 } 1632 1633 Elf32_Ehdr elf_head; 1634 int diag_msg_max_length=ebuflen-strlen(ebuf); 1635 char* diag_msg_buf=ebuf+strlen(ebuf); 1636 1637 if (diag_msg_max_length==0) { 1638 // No more space in ebuf for additional diagnostics message 1639 return NULL; 1640 } 1641 1642 1643 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1644 1645 if (file_descriptor < 0) { 1646 // Can't open library, report dlerror() message 1647 return NULL; 1648 } 1649 1650 bool failed_to_read_elf_head= 1651 (sizeof(elf_head)!= 1652 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1653 1654 ::close(file_descriptor); 1655 if (failed_to_read_elf_head) { 1656 // file i/o error - report dlerror() msg 1657 return NULL; 1658 } 1659 1660 typedef struct { 1661 Elf32_Half code; // Actual value as defined in elf.h 1662 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1663 unsigned char elf_class; // 32 or 64 bit 1664 unsigned char endianess; // MSB or LSB 1665 char* name; // String representation 1666 } arch_t; 1667 1668 #ifndef EM_486 1669 #define EM_486 6 /* Intel 80486 */ 1670 #endif 1671 #ifndef EM_AARCH64 1672 #define EM_AARCH64 183 /* ARM AARCH64 */ 1673 #endif 1674 1675 static const arch_t arch_array[]={ 1676 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1677 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1678 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1679 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1680 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1681 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1682 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1683 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1684 #if defined(VM_LITTLE_ENDIAN) 1685 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"}, 1686 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"}, 1687 #else 1688 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1689 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"}, 1690 #endif 1691 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1692 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1693 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1694 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1695 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1696 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1697 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1698 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1699 }; 1700 1701 #if (defined IA32) 1702 static Elf32_Half running_arch_code=EM_386; 1703 #elif (defined AMD64) || (defined X32) 1704 static Elf32_Half running_arch_code=EM_X86_64; 1705 #elif (defined IA64) 1706 static Elf32_Half running_arch_code=EM_IA_64; 1707 #elif (defined __sparc) && (defined _LP64) 1708 static Elf32_Half running_arch_code=EM_SPARCV9; 1709 #elif (defined __sparc) && (!defined _LP64) 1710 static Elf32_Half running_arch_code=EM_SPARC; 1711 #elif (defined __powerpc64__) 1712 static Elf32_Half running_arch_code=EM_PPC64; 1713 #elif (defined __powerpc__) 1714 static Elf32_Half running_arch_code=EM_PPC; 1715 #elif (defined AARCH64) 1716 static Elf32_Half running_arch_code=EM_AARCH64; 1717 #elif (defined ARM) 1718 static Elf32_Half running_arch_code=EM_ARM; 1719 #elif (defined S390) 1720 static Elf32_Half running_arch_code=EM_S390; 1721 #elif (defined ALPHA) 1722 static Elf32_Half running_arch_code=EM_ALPHA; 1723 #elif (defined MIPSEL) 1724 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1725 #elif (defined PARISC) 1726 static Elf32_Half running_arch_code=EM_PARISC; 1727 #elif (defined MIPS) 1728 static Elf32_Half running_arch_code=EM_MIPS; 1729 #elif (defined M68K) 1730 static Elf32_Half running_arch_code=EM_68K; 1731 #elif (defined SH) 1732 static Elf32_Half running_arch_code=EM_SH; 1733 #else 1734 #error Method os::dll_load requires that one of following is defined:\ 1735 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc 1736 #endif 1737 1738 // Identify compatability class for VM's architecture and library's architecture 1739 // Obtain string descriptions for architectures 1740 1741 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1742 int running_arch_index=-1; 1743 1744 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1745 if (running_arch_code == arch_array[i].code) { 1746 running_arch_index = i; 1747 } 1748 if (lib_arch.code == arch_array[i].code) { 1749 lib_arch.compat_class = arch_array[i].compat_class; 1750 lib_arch.name = arch_array[i].name; 1751 } 1752 } 1753 1754 assert(running_arch_index != -1, 1755 "Didn't find running architecture code (running_arch_code) in arch_array"); 1756 if (running_arch_index == -1) { 1757 // Even though running architecture detection failed 1758 // we may still continue with reporting dlerror() message 1759 return NULL; 1760 } 1761 1762 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1763 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1764 return NULL; 1765 } 1766 1767 #ifndef S390 1768 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1769 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1770 return NULL; 1771 } 1772 #endif // !S390 1773 1774 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1775 if (lib_arch.name!=NULL) { 1776 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1777 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1778 lib_arch.name, arch_array[running_arch_index].name); 1779 } else { 1780 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1781 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1782 lib_arch.code, 1783 arch_array[running_arch_index].name); 1784 } 1785 } 1786 1787 return NULL; 1788 } 1789 1790 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 1791 int ebuflen) { 1792 void * result = ::dlopen(filename, RTLD_LAZY); 1793 if (result == NULL) { 1794 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 1795 ebuf[ebuflen-1] = '\0'; 1796 } 1797 return result; 1798 } 1799 1800 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 1801 int ebuflen) { 1802 void * result = NULL; 1803 if (LoadExecStackDllInVMThread) { 1804 result = dlopen_helper(filename, ebuf, ebuflen); 1805 } 1806 1807 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 1808 // library that requires an executable stack, or which does not have this 1809 // stack attribute set, dlopen changes the stack attribute to executable. The 1810 // read protection of the guard pages gets lost. 1811 // 1812 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 1813 // may have been queued at the same time. 1814 1815 if (!_stack_is_executable) { 1816 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) { 1817 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 1818 jt->stack_guards_enabled()) { // No pending stack overflow exceptions 1819 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) { 1820 warning("Attempt to reguard stack yellow zone failed."); 1821 } 1822 } 1823 } 1824 } 1825 1826 return result; 1827 } 1828 1829 void* os::dll_lookup(void* handle, const char* name) { 1830 void* res = dlsym(handle, name); 1831 return res; 1832 } 1833 1834 void* os::get_default_process_handle() { 1835 return (void*)::dlopen(NULL, RTLD_LAZY); 1836 } 1837 1838 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) { 1839 int fd = ::open(filename, O_RDONLY); 1840 if (fd == -1) { 1841 return false; 1842 } 1843 1844 if (hdr != NULL) { 1845 st->print_cr("%s", hdr); 1846 } 1847 1848 char buf[33]; 1849 int bytes; 1850 buf[32] = '\0'; 1851 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) { 1852 st->print_raw(buf, bytes); 1853 } 1854 1855 ::close(fd); 1856 1857 return true; 1858 } 1859 1860 #if defined(S390) 1861 // keywords_to_match - NULL terminated array of keywords 1862 static bool print_matching_lines_from_sysinfo_file(outputStream* st, const char* keywords_to_match[]) { 1863 const char* filename = "/proc/sysinfo"; 1864 char* line = NULL; 1865 size_t length = 0; 1866 FILE* fp = fopen(filename, "r"); 1867 if (fp == NULL) { 1868 return false; 1869 } 1870 1871 st->print_cr("Virtualization information:"); 1872 while (getline(&line, &length, fp) != -1) { 1873 int i = 0; 1874 while (keywords_to_match[i] != NULL) { 1875 if (strncmp(line, keywords_to_match[i], strlen(keywords_to_match[i])) == 0) { 1876 st->print("%s", line); 1877 break; 1878 } 1879 i++; 1880 } 1881 } 1882 1883 free(line); 1884 fclose(fp); 1885 1886 return true; 1887 } 1888 #endif 1889 1890 void os::print_dll_info(outputStream *st) { 1891 st->print_cr("Dynamic libraries:"); 1892 1893 char fname[32]; 1894 pid_t pid = os::Linux::gettid(); 1895 1896 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1897 1898 if (!_print_ascii_file(fname, st)) { 1899 st->print("Can not get library information for pid = %d\n", pid); 1900 } 1901 } 1902 1903 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 1904 FILE *procmapsFile = NULL; 1905 1906 // Open the procfs maps file for the current process 1907 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 1908 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 1909 char line[PATH_MAX + 100]; 1910 1911 // Read line by line from 'file' 1912 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 1913 u8 base, top, offset, inode; 1914 char permissions[5]; 1915 char device[6]; 1916 char name[PATH_MAX + 1]; 1917 1918 // Parse fields from line 1919 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s", 1920 &base, &top, permissions, &offset, device, &inode, name); 1921 1922 // Filter by device id '00:00' so that we only get file system mapped files. 1923 if (strcmp(device, "00:00") != 0) { 1924 1925 // Call callback with the fields of interest 1926 if(callback(name, (address)base, (address)top, param)) { 1927 // Oops abort, callback aborted 1928 fclose(procmapsFile); 1929 return 1; 1930 } 1931 } 1932 } 1933 fclose(procmapsFile); 1934 } 1935 return 0; 1936 } 1937 1938 void os::print_os_info_brief(outputStream* st) { 1939 os::Linux::print_distro_info(st); 1940 1941 os::Posix::print_uname_info(st); 1942 1943 os::Linux::print_libversion_info(st); 1944 1945 } 1946 1947 void os::print_os_info(outputStream* st) { 1948 st->print("OS:"); 1949 1950 os::Linux::print_distro_info(st); 1951 1952 os::Posix::print_uname_info(st); 1953 1954 // Print warning if unsafe chroot environment detected 1955 if (unsafe_chroot_detected) { 1956 st->print("WARNING!! "); 1957 st->print_cr("%s", unstable_chroot_error); 1958 } 1959 1960 os::Linux::print_libversion_info(st); 1961 1962 os::Posix::print_rlimit_info(st); 1963 1964 os::Posix::print_load_average(st); 1965 1966 os::Linux::print_full_memory_info(st); 1967 1968 os::Linux::print_proc_sys_info(st); 1969 1970 os::Linux::print_ld_preload_file(st); 1971 1972 os::Linux::print_container_info(st); 1973 1974 os::Linux::print_virtualization_info(st); 1975 } 1976 1977 // Try to identify popular distros. 1978 // Most Linux distributions have a /etc/XXX-release file, which contains 1979 // the OS version string. Newer Linux distributions have a /etc/lsb-release 1980 // file that also contains the OS version string. Some have more than one 1981 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 1982 // /etc/redhat-release.), so the order is important. 1983 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 1984 // their own specific XXX-release file as well as a redhat-release file. 1985 // Because of this the XXX-release file needs to be searched for before the 1986 // redhat-release file. 1987 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the 1988 // search for redhat-release / SuSE-release needs to be before lsb-release. 1989 // Since the lsb-release file is the new standard it needs to be searched 1990 // before the older style release files. 1991 // Searching system-release (Red Hat) and os-release (other Linuxes) are a 1992 // next to last resort. The os-release file is a new standard that contains 1993 // distribution information and the system-release file seems to be an old 1994 // standard that has been replaced by the lsb-release and os-release files. 1995 // Searching for the debian_version file is the last resort. It contains 1996 // an informative string like "6.0.6" or "wheezy/sid". Because of this 1997 // "Debian " is printed before the contents of the debian_version file. 1998 1999 const char* distro_files[] = { 2000 "/etc/oracle-release", 2001 "/etc/mandriva-release", 2002 "/etc/mandrake-release", 2003 "/etc/sun-release", 2004 "/etc/redhat-release", 2005 "/etc/SuSE-release", 2006 "/etc/lsb-release", 2007 "/etc/turbolinux-release", 2008 "/etc/gentoo-release", 2009 "/etc/ltib-release", 2010 "/etc/angstrom-version", 2011 "/etc/system-release", 2012 "/etc/os-release", 2013 NULL }; 2014 2015 void os::Linux::print_distro_info(outputStream* st) { 2016 for (int i = 0;; i++) { 2017 const char* file = distro_files[i]; 2018 if (file == NULL) { 2019 break; // done 2020 } 2021 // If file prints, we found it. 2022 if (_print_ascii_file(file, st)) { 2023 return; 2024 } 2025 } 2026 2027 if (file_exists("/etc/debian_version")) { 2028 st->print("Debian "); 2029 _print_ascii_file("/etc/debian_version", st); 2030 } else { 2031 st->print("Linux"); 2032 } 2033 st->cr(); 2034 } 2035 2036 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) { 2037 char buf[256]; 2038 while (fgets(buf, sizeof(buf), fp)) { 2039 // Edit out extra stuff in expected format 2040 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) { 2041 char* ptr = strstr(buf, "\""); // the name is in quotes 2042 if (ptr != NULL) { 2043 ptr++; // go beyond first quote 2044 char* nl = strchr(ptr, '\"'); 2045 if (nl != NULL) *nl = '\0'; 2046 strncpy(distro, ptr, length); 2047 } else { 2048 ptr = strstr(buf, "="); 2049 ptr++; // go beyond equals then 2050 char* nl = strchr(ptr, '\n'); 2051 if (nl != NULL) *nl = '\0'; 2052 strncpy(distro, ptr, length); 2053 } 2054 return; 2055 } else if (get_first_line) { 2056 char* nl = strchr(buf, '\n'); 2057 if (nl != NULL) *nl = '\0'; 2058 strncpy(distro, buf, length); 2059 return; 2060 } 2061 } 2062 // print last line and close 2063 char* nl = strchr(buf, '\n'); 2064 if (nl != NULL) *nl = '\0'; 2065 strncpy(distro, buf, length); 2066 } 2067 2068 static void parse_os_info(char* distro, size_t length, const char* file) { 2069 FILE* fp = fopen(file, "r"); 2070 if (fp != NULL) { 2071 // if suse format, print out first line 2072 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0); 2073 parse_os_info_helper(fp, distro, length, get_first_line); 2074 fclose(fp); 2075 } 2076 } 2077 2078 void os::get_summary_os_info(char* buf, size_t buflen) { 2079 for (int i = 0;; i++) { 2080 const char* file = distro_files[i]; 2081 if (file == NULL) { 2082 break; // ran out of distro_files 2083 } 2084 if (file_exists(file)) { 2085 parse_os_info(buf, buflen, file); 2086 return; 2087 } 2088 } 2089 // special case for debian 2090 if (file_exists("/etc/debian_version")) { 2091 strncpy(buf, "Debian ", buflen); 2092 if (buflen > 7) { 2093 parse_os_info(&buf[7], buflen-7, "/etc/debian_version"); 2094 } 2095 } else { 2096 strncpy(buf, "Linux", buflen); 2097 } 2098 } 2099 2100 void os::Linux::print_libversion_info(outputStream* st) { 2101 // libc, pthread 2102 st->print("libc:"); 2103 st->print("%s ", os::Linux::glibc_version()); 2104 st->print("%s ", os::Linux::libpthread_version()); 2105 st->cr(); 2106 } 2107 2108 void os::Linux::print_proc_sys_info(outputStream* st) { 2109 st->cr(); 2110 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):"); 2111 _print_ascii_file("/proc/sys/kernel/threads-max", st); 2112 st->cr(); 2113 st->cr(); 2114 2115 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):"); 2116 _print_ascii_file("/proc/sys/vm/max_map_count", st); 2117 st->cr(); 2118 st->cr(); 2119 2120 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):"); 2121 _print_ascii_file("/proc/sys/kernel/pid_max", st); 2122 st->cr(); 2123 st->cr(); 2124 } 2125 2126 void os::Linux::print_full_memory_info(outputStream* st) { 2127 st->print("\n/proc/meminfo:\n"); 2128 _print_ascii_file("/proc/meminfo", st); 2129 st->cr(); 2130 } 2131 2132 void os::Linux::print_ld_preload_file(outputStream* st) { 2133 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:"); 2134 st->cr(); 2135 } 2136 2137 void os::Linux::print_container_info(outputStream* st) { 2138 if (!OSContainer::is_containerized()) { 2139 return; 2140 } 2141 2142 st->print("container (cgroup) information:\n"); 2143 2144 const char *p_ct = OSContainer::container_type(); 2145 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed"); 2146 2147 char *p = OSContainer::cpu_cpuset_cpus(); 2148 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed"); 2149 free(p); 2150 2151 p = OSContainer::cpu_cpuset_memory_nodes(); 2152 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed"); 2153 free(p); 2154 2155 int i = OSContainer::active_processor_count(); 2156 if (i > 0) { 2157 st->print("active_processor_count: %d\n", i); 2158 } else { 2159 st->print("active_processor_count: failed\n"); 2160 } 2161 2162 i = OSContainer::cpu_quota(); 2163 st->print("cpu_quota: %d\n", i); 2164 2165 i = OSContainer::cpu_period(); 2166 st->print("cpu_period: %d\n", i); 2167 2168 i = OSContainer::cpu_shares(); 2169 st->print("cpu_shares: %d\n", i); 2170 2171 jlong j = OSContainer::memory_limit_in_bytes(); 2172 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j); 2173 2174 j = OSContainer::memory_and_swap_limit_in_bytes(); 2175 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j); 2176 2177 j = OSContainer::memory_soft_limit_in_bytes(); 2178 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j); 2179 2180 j = OSContainer::OSContainer::memory_usage_in_bytes(); 2181 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j); 2182 2183 j = OSContainer::OSContainer::memory_max_usage_in_bytes(); 2184 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j); 2185 st->cr(); 2186 } 2187 2188 void os::Linux::print_virtualization_info(outputStream* st) { 2189 #if defined(S390) 2190 // /proc/sysinfo contains interesting information about 2191 // - LPAR 2192 // - whole "Box" (CPUs ) 2193 // - z/VM / KVM (VM<nn>); this is not available in an LPAR-only setup 2194 const char* kw[] = { "LPAR", "CPUs", "VM", NULL }; 2195 2196 if (! print_matching_lines_from_sysinfo_file(st, kw)) { 2197 st->print_cr(" </proc/sysinfo Not Available>"); 2198 } 2199 #endif 2200 } 2201 2202 void os::print_memory_info(outputStream* st) { 2203 2204 st->print("Memory:"); 2205 st->print(" %dk page", os::vm_page_size()>>10); 2206 2207 // values in struct sysinfo are "unsigned long" 2208 struct sysinfo si; 2209 sysinfo(&si); 2210 2211 st->print(", physical " UINT64_FORMAT "k", 2212 os::physical_memory() >> 10); 2213 st->print("(" UINT64_FORMAT "k free)", 2214 os::available_memory() >> 10); 2215 st->print(", swap " UINT64_FORMAT "k", 2216 ((jlong)si.totalswap * si.mem_unit) >> 10); 2217 st->print("(" UINT64_FORMAT "k free)", 2218 ((jlong)si.freeswap * si.mem_unit) >> 10); 2219 st->cr(); 2220 } 2221 2222 // Print the first "model name" line and the first "flags" line 2223 // that we find and nothing more. We assume "model name" comes 2224 // before "flags" so if we find a second "model name", then the 2225 // "flags" field is considered missing. 2226 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) { 2227 #if defined(IA32) || defined(AMD64) 2228 // Other platforms have less repetitive cpuinfo files 2229 FILE *fp = fopen("/proc/cpuinfo", "r"); 2230 if (fp) { 2231 while (!feof(fp)) { 2232 if (fgets(buf, buflen, fp)) { 2233 // Assume model name comes before flags 2234 bool model_name_printed = false; 2235 if (strstr(buf, "model name") != NULL) { 2236 if (!model_name_printed) { 2237 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n"); 2238 st->print_raw(buf); 2239 model_name_printed = true; 2240 } else { 2241 // model name printed but not flags? Odd, just return 2242 fclose(fp); 2243 return true; 2244 } 2245 } 2246 // print the flags line too 2247 if (strstr(buf, "flags") != NULL) { 2248 st->print_raw(buf); 2249 fclose(fp); 2250 return true; 2251 } 2252 } 2253 } 2254 fclose(fp); 2255 } 2256 #endif // x86 platforms 2257 return false; 2258 } 2259 2260 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) { 2261 // Only print the model name if the platform provides this as a summary 2262 if (!print_model_name_and_flags(st, buf, buflen)) { 2263 st->print("\n/proc/cpuinfo:\n"); 2264 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2265 st->print_cr(" <Not Available>"); 2266 } 2267 } 2268 } 2269 2270 #if defined(AMD64) || defined(IA32) || defined(X32) 2271 const char* search_string = "model name"; 2272 #elif defined(M68K) 2273 const char* search_string = "CPU"; 2274 #elif defined(PPC64) 2275 const char* search_string = "cpu"; 2276 #elif defined(S390) 2277 const char* search_string = "processor"; 2278 #elif defined(SPARC) 2279 const char* search_string = "cpu"; 2280 #else 2281 const char* search_string = "Processor"; 2282 #endif 2283 2284 // Parses the cpuinfo file for string representing the model name. 2285 void os::get_summary_cpu_info(char* cpuinfo, size_t length) { 2286 FILE* fp = fopen("/proc/cpuinfo", "r"); 2287 if (fp != NULL) { 2288 while (!feof(fp)) { 2289 char buf[256]; 2290 if (fgets(buf, sizeof(buf), fp)) { 2291 char* start = strstr(buf, search_string); 2292 if (start != NULL) { 2293 char *ptr = start + strlen(search_string); 2294 char *end = buf + strlen(buf); 2295 while (ptr != end) { 2296 // skip whitespace and colon for the rest of the name. 2297 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') { 2298 break; 2299 } 2300 ptr++; 2301 } 2302 if (ptr != end) { 2303 // reasonable string, get rid of newline and keep the rest 2304 char* nl = strchr(buf, '\n'); 2305 if (nl != NULL) *nl = '\0'; 2306 strncpy(cpuinfo, ptr, length); 2307 fclose(fp); 2308 return; 2309 } 2310 } 2311 } 2312 } 2313 fclose(fp); 2314 } 2315 // cpuinfo not found or parsing failed, just print generic string. The entire 2316 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86) 2317 #if defined(AARCH64) 2318 strncpy(cpuinfo, "AArch64", length); 2319 #elif defined(AMD64) 2320 strncpy(cpuinfo, "x86_64", length); 2321 #elif defined(ARM) // Order wrt. AARCH64 is relevant! 2322 strncpy(cpuinfo, "ARM", length); 2323 #elif defined(IA32) 2324 strncpy(cpuinfo, "x86_32", length); 2325 #elif defined(IA64) 2326 strncpy(cpuinfo, "IA64", length); 2327 #elif defined(PPC) 2328 strncpy(cpuinfo, "PPC64", length); 2329 #elif defined(S390) 2330 strncpy(cpuinfo, "S390", length); 2331 #elif defined(SPARC) 2332 strncpy(cpuinfo, "sparcv9", length); 2333 #elif defined(ZERO_LIBARCH) 2334 strncpy(cpuinfo, ZERO_LIBARCH, length); 2335 #else 2336 strncpy(cpuinfo, "unknown", length); 2337 #endif 2338 } 2339 2340 static void print_signal_handler(outputStream* st, int sig, 2341 char* buf, size_t buflen); 2342 2343 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2344 st->print_cr("Signal Handlers:"); 2345 print_signal_handler(st, SIGSEGV, buf, buflen); 2346 print_signal_handler(st, SIGBUS , buf, buflen); 2347 print_signal_handler(st, SIGFPE , buf, buflen); 2348 print_signal_handler(st, SIGPIPE, buf, buflen); 2349 print_signal_handler(st, SIGXFSZ, buf, buflen); 2350 print_signal_handler(st, SIGILL , buf, buflen); 2351 print_signal_handler(st, SR_signum, buf, buflen); 2352 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2353 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2354 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2355 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2356 #if defined(PPC64) 2357 print_signal_handler(st, SIGTRAP, buf, buflen); 2358 #endif 2359 } 2360 2361 static char saved_jvm_path[MAXPATHLEN] = {0}; 2362 2363 // Find the full path to the current module, libjvm.so 2364 void os::jvm_path(char *buf, jint buflen) { 2365 // Error checking. 2366 if (buflen < MAXPATHLEN) { 2367 assert(false, "must use a large-enough buffer"); 2368 buf[0] = '\0'; 2369 return; 2370 } 2371 // Lazy resolve the path to current module. 2372 if (saved_jvm_path[0] != 0) { 2373 strcpy(buf, saved_jvm_path); 2374 return; 2375 } 2376 2377 char dli_fname[MAXPATHLEN]; 2378 bool ret = dll_address_to_library_name( 2379 CAST_FROM_FN_PTR(address, os::jvm_path), 2380 dli_fname, sizeof(dli_fname), NULL); 2381 assert(ret, "cannot locate libjvm"); 2382 char *rp = NULL; 2383 if (ret && dli_fname[0] != '\0') { 2384 rp = os::Posix::realpath(dli_fname, buf, buflen); 2385 } 2386 if (rp == NULL) { 2387 return; 2388 } 2389 2390 if (Arguments::sun_java_launcher_is_altjvm()) { 2391 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2392 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so". 2393 // If "/jre/lib/" appears at the right place in the string, then 2394 // assume we are installed in a JDK and we're done. Otherwise, check 2395 // for a JAVA_HOME environment variable and fix up the path so it 2396 // looks like libjvm.so is installed there (append a fake suffix 2397 // hotspot/libjvm.so). 2398 const char *p = buf + strlen(buf) - 1; 2399 for (int count = 0; p > buf && count < 5; ++count) { 2400 for (--p; p > buf && *p != '/'; --p) 2401 /* empty */ ; 2402 } 2403 2404 if (strncmp(p, "/jre/lib/", 9) != 0) { 2405 // Look for JAVA_HOME in the environment. 2406 char* java_home_var = ::getenv("JAVA_HOME"); 2407 if (java_home_var != NULL && java_home_var[0] != 0) { 2408 char* jrelib_p; 2409 int len; 2410 2411 // Check the current module name "libjvm.so". 2412 p = strrchr(buf, '/'); 2413 if (p == NULL) { 2414 return; 2415 } 2416 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2417 2418 rp = os::Posix::realpath(java_home_var, buf, buflen); 2419 if (rp == NULL) { 2420 return; 2421 } 2422 2423 // determine if this is a legacy image or modules image 2424 // modules image doesn't have "jre" subdirectory 2425 len = strlen(buf); 2426 assert(len < buflen, "Ran out of buffer room"); 2427 jrelib_p = buf + len; 2428 snprintf(jrelib_p, buflen-len, "/jre/lib"); 2429 if (0 != access(buf, F_OK)) { 2430 snprintf(jrelib_p, buflen-len, "/lib"); 2431 } 2432 2433 if (0 == access(buf, F_OK)) { 2434 // Use current module name "libjvm.so" 2435 len = strlen(buf); 2436 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2437 } else { 2438 // Go back to path of .so 2439 rp = os::Posix::realpath(dli_fname, buf, buflen); 2440 if (rp == NULL) { 2441 return; 2442 } 2443 } 2444 } 2445 } 2446 } 2447 2448 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2449 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2450 } 2451 2452 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2453 // no prefix required, not even "_" 2454 } 2455 2456 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2457 // no suffix required 2458 } 2459 2460 //////////////////////////////////////////////////////////////////////////////// 2461 // sun.misc.Signal support 2462 2463 static volatile jint sigint_count = 0; 2464 2465 static void UserHandler(int sig, void *siginfo, void *context) { 2466 // 4511530 - sem_post is serialized and handled by the manager thread. When 2467 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2468 // don't want to flood the manager thread with sem_post requests. 2469 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2470 return; 2471 } 2472 2473 // Ctrl-C is pressed during error reporting, likely because the error 2474 // handler fails to abort. Let VM die immediately. 2475 if (sig == SIGINT && VMError::is_error_reported()) { 2476 os::die(); 2477 } 2478 2479 os::signal_notify(sig); 2480 } 2481 2482 void* os::user_handler() { 2483 return CAST_FROM_FN_PTR(void*, UserHandler); 2484 } 2485 2486 extern "C" { 2487 typedef void (*sa_handler_t)(int); 2488 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2489 } 2490 2491 void* os::signal(int signal_number, void* handler) { 2492 struct sigaction sigAct, oldSigAct; 2493 2494 sigfillset(&(sigAct.sa_mask)); 2495 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2496 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2497 2498 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2499 // -1 means registration failed 2500 return (void *)-1; 2501 } 2502 2503 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2504 } 2505 2506 void os::signal_raise(int signal_number) { 2507 ::raise(signal_number); 2508 } 2509 2510 // The following code is moved from os.cpp for making this 2511 // code platform specific, which it is by its very nature. 2512 2513 // Will be modified when max signal is changed to be dynamic 2514 int os::sigexitnum_pd() { 2515 return NSIG; 2516 } 2517 2518 // a counter for each possible signal value 2519 static volatile jint pending_signals[NSIG+1] = { 0 }; 2520 2521 // Linux(POSIX) specific hand shaking semaphore. 2522 static Semaphore* sig_sem = NULL; 2523 static PosixSemaphore sr_semaphore; 2524 2525 static void jdk_misc_signal_init() { 2526 // Initialize signal structures 2527 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2528 2529 // Initialize signal semaphore 2530 sig_sem = new Semaphore(); 2531 } 2532 2533 void os::signal_notify(int sig) { 2534 if (sig_sem != NULL) { 2535 Atomic::inc(&pending_signals[sig]); 2536 sig_sem->signal(); 2537 } else { 2538 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init 2539 // initialization isn't called. 2540 assert(ReduceSignalUsage, "signal semaphore should be created"); 2541 } 2542 } 2543 2544 static int check_pending_signals() { 2545 Atomic::store(0, &sigint_count); 2546 for (;;) { 2547 for (int i = 0; i < NSIG + 1; i++) { 2548 jint n = pending_signals[i]; 2549 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2550 return i; 2551 } 2552 } 2553 JavaThread *thread = JavaThread::current(); 2554 ThreadBlockInVM tbivm(thread); 2555 2556 bool threadIsSuspended; 2557 do { 2558 thread->set_suspend_equivalent(); 2559 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2560 sig_sem->wait(); 2561 2562 // were we externally suspended while we were waiting? 2563 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2564 if (threadIsSuspended) { 2565 // The semaphore has been incremented, but while we were waiting 2566 // another thread suspended us. We don't want to continue running 2567 // while suspended because that would surprise the thread that 2568 // suspended us. 2569 sig_sem->signal(); 2570 2571 thread->java_suspend_self(); 2572 } 2573 } while (threadIsSuspended); 2574 } 2575 } 2576 2577 int os::signal_wait() { 2578 return check_pending_signals(); 2579 } 2580 2581 //////////////////////////////////////////////////////////////////////////////// 2582 // Virtual Memory 2583 2584 int os::vm_page_size() { 2585 // Seems redundant as all get out 2586 assert(os::Linux::page_size() != -1, "must call os::init"); 2587 return os::Linux::page_size(); 2588 } 2589 2590 // Solaris allocates memory by pages. 2591 int os::vm_allocation_granularity() { 2592 assert(os::Linux::page_size() != -1, "must call os::init"); 2593 return os::Linux::page_size(); 2594 } 2595 2596 // Rationale behind this function: 2597 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2598 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2599 // samples for JITted code. Here we create private executable mapping over the code cache 2600 // and then we can use standard (well, almost, as mapping can change) way to provide 2601 // info for the reporting script by storing timestamp and location of symbol 2602 void linux_wrap_code(char* base, size_t size) { 2603 static volatile jint cnt = 0; 2604 2605 if (!UseOprofile) { 2606 return; 2607 } 2608 2609 char buf[PATH_MAX+1]; 2610 int num = Atomic::add(1, &cnt); 2611 2612 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2613 os::get_temp_directory(), os::current_process_id(), num); 2614 unlink(buf); 2615 2616 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2617 2618 if (fd != -1) { 2619 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2620 if (rv != (off_t)-1) { 2621 if (::write(fd, "", 1) == 1) { 2622 mmap(base, size, 2623 PROT_READ|PROT_WRITE|PROT_EXEC, 2624 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2625 } 2626 } 2627 ::close(fd); 2628 unlink(buf); 2629 } 2630 } 2631 2632 static bool recoverable_mmap_error(int err) { 2633 // See if the error is one we can let the caller handle. This 2634 // list of errno values comes from JBS-6843484. I can't find a 2635 // Linux man page that documents this specific set of errno 2636 // values so while this list currently matches Solaris, it may 2637 // change as we gain experience with this failure mode. 2638 switch (err) { 2639 case EBADF: 2640 case EINVAL: 2641 case ENOTSUP: 2642 // let the caller deal with these errors 2643 return true; 2644 2645 default: 2646 // Any remaining errors on this OS can cause our reserved mapping 2647 // to be lost. That can cause confusion where different data 2648 // structures think they have the same memory mapped. The worst 2649 // scenario is if both the VM and a library think they have the 2650 // same memory mapped. 2651 return false; 2652 } 2653 } 2654 2655 static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2656 int err) { 2657 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2658 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec, 2659 os::strerror(err), err); 2660 } 2661 2662 static void warn_fail_commit_memory(char* addr, size_t size, 2663 size_t alignment_hint, bool exec, 2664 int err) { 2665 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2666 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, 2667 alignment_hint, exec, os::strerror(err), err); 2668 } 2669 2670 // NOTE: Linux kernel does not really reserve the pages for us. 2671 // All it does is to check if there are enough free pages 2672 // left at the time of mmap(). This could be a potential 2673 // problem. 2674 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2675 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2676 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2677 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2678 if (res != (uintptr_t) MAP_FAILED) { 2679 if (UseNUMAInterleaving) { 2680 numa_make_global(addr, size); 2681 } 2682 return 0; 2683 } 2684 2685 int err = errno; // save errno from mmap() call above 2686 2687 if (!recoverable_mmap_error(err)) { 2688 warn_fail_commit_memory(addr, size, exec, err); 2689 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2690 } 2691 2692 return err; 2693 } 2694 2695 bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2696 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2697 } 2698 2699 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2700 const char* mesg) { 2701 assert(mesg != NULL, "mesg must be specified"); 2702 int err = os::Linux::commit_memory_impl(addr, size, exec); 2703 if (err != 0) { 2704 // the caller wants all commit errors to exit with the specified mesg: 2705 warn_fail_commit_memory(addr, size, exec, err); 2706 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2707 } 2708 } 2709 2710 // Define MAP_HUGETLB here so we can build HotSpot on old systems. 2711 #ifndef MAP_HUGETLB 2712 #define MAP_HUGETLB 0x40000 2713 #endif 2714 2715 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2716 #ifndef MADV_HUGEPAGE 2717 #define MADV_HUGEPAGE 14 2718 #endif 2719 2720 int os::Linux::commit_memory_impl(char* addr, size_t size, 2721 size_t alignment_hint, bool exec) { 2722 int err = os::Linux::commit_memory_impl(addr, size, exec); 2723 if (err == 0) { 2724 realign_memory(addr, size, alignment_hint); 2725 } 2726 return err; 2727 } 2728 2729 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2730 bool exec) { 2731 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2732 } 2733 2734 void os::pd_commit_memory_or_exit(char* addr, size_t size, 2735 size_t alignment_hint, bool exec, 2736 const char* mesg) { 2737 assert(mesg != NULL, "mesg must be specified"); 2738 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2739 if (err != 0) { 2740 // the caller wants all commit errors to exit with the specified mesg: 2741 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2742 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2743 } 2744 } 2745 2746 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2747 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2748 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2749 // be supported or the memory may already be backed by huge pages. 2750 ::madvise(addr, bytes, MADV_HUGEPAGE); 2751 } 2752 } 2753 2754 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2755 // This method works by doing an mmap over an existing mmaping and effectively discarding 2756 // the existing pages. However it won't work for SHM-based large pages that cannot be 2757 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2758 // small pages on top of the SHM segment. This method always works for small pages, so we 2759 // allow that in any case. 2760 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2761 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2762 } 2763 } 2764 2765 void os::numa_make_global(char *addr, size_t bytes) { 2766 Linux::numa_interleave_memory(addr, bytes); 2767 } 2768 2769 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2770 // bind policy to MPOL_PREFERRED for the current thread. 2771 #define USE_MPOL_PREFERRED 0 2772 2773 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2774 // To make NUMA and large pages more robust when both enabled, we need to ease 2775 // the requirements on where the memory should be allocated. MPOL_BIND is the 2776 // default policy and it will force memory to be allocated on the specified 2777 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2778 // the specified node, but will not force it. Using this policy will prevent 2779 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2780 // free large pages. 2781 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2782 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2783 } 2784 2785 bool os::numa_topology_changed() { return false; } 2786 2787 size_t os::numa_get_groups_num() { 2788 // Return just the number of nodes in which it's possible to allocate memory 2789 // (in numa terminology, configured nodes). 2790 return Linux::numa_num_configured_nodes(); 2791 } 2792 2793 int os::numa_get_group_id() { 2794 int cpu_id = Linux::sched_getcpu(); 2795 if (cpu_id != -1) { 2796 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2797 if (lgrp_id != -1) { 2798 return lgrp_id; 2799 } 2800 } 2801 return 0; 2802 } 2803 2804 int os::Linux::get_existing_num_nodes() { 2805 int node; 2806 int highest_node_number = Linux::numa_max_node(); 2807 int num_nodes = 0; 2808 2809 // Get the total number of nodes in the system including nodes without memory. 2810 for (node = 0; node <= highest_node_number; node++) { 2811 if (is_node_in_existing_nodes(node)) { 2812 num_nodes++; 2813 } 2814 } 2815 return num_nodes; 2816 } 2817 2818 size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2819 int highest_node_number = Linux::numa_max_node(); 2820 size_t i = 0; 2821 2822 // Map all node ids in which it is possible to allocate memory. Also nodes are 2823 // not always consecutively available, i.e. available from 0 to the highest 2824 // node number. If the nodes have been bound explicitly using numactl membind, 2825 // then allocate memory from those nodes only. 2826 for (int node = 0; node <= highest_node_number; node++) { 2827 if (Linux::is_node_in_bound_nodes((unsigned int)node)) { 2828 ids[i++] = node; 2829 } 2830 } 2831 return i; 2832 } 2833 2834 bool os::get_page_info(char *start, page_info* info) { 2835 return false; 2836 } 2837 2838 char *os::scan_pages(char *start, char* end, page_info* page_expected, 2839 page_info* page_found) { 2840 return end; 2841 } 2842 2843 2844 int os::Linux::sched_getcpu_syscall(void) { 2845 unsigned int cpu = 0; 2846 int retval = -1; 2847 2848 #if defined(IA32) 2849 #ifndef SYS_getcpu 2850 #define SYS_getcpu 318 2851 #endif 2852 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2853 #elif defined(AMD64) 2854 // Unfortunately we have to bring all these macros here from vsyscall.h 2855 // to be able to compile on old linuxes. 2856 #define __NR_vgetcpu 2 2857 #define VSYSCALL_START (-10UL << 20) 2858 #define VSYSCALL_SIZE 1024 2859 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2860 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2861 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2862 retval = vgetcpu(&cpu, NULL, NULL); 2863 #endif 2864 2865 return (retval == -1) ? retval : cpu; 2866 } 2867 2868 void os::Linux::sched_getcpu_init() { 2869 // sched_getcpu() should be in libc. 2870 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2871 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2872 2873 // If it's not, try a direct syscall. 2874 if (sched_getcpu() == -1) { 2875 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2876 (void*)&sched_getcpu_syscall)); 2877 } 2878 2879 if (sched_getcpu() == -1) { 2880 vm_exit_during_initialization("getcpu(2) system call not supported by kernel"); 2881 } 2882 } 2883 2884 // Something to do with the numa-aware allocator needs these symbols 2885 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2886 extern "C" JNIEXPORT void numa_error(char *where) { } 2887 2888 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails 2889 // load symbol from base version instead. 2890 void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2891 void *f = dlvsym(handle, name, "libnuma_1.1"); 2892 if (f == NULL) { 2893 f = dlsym(handle, name); 2894 } 2895 return f; 2896 } 2897 2898 // Handle request to load libnuma symbol version 1.2 (API v2) only. 2899 // Return NULL if the symbol is not defined in this particular version. 2900 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) { 2901 return dlvsym(handle, name, "libnuma_1.2"); 2902 } 2903 2904 bool os::Linux::libnuma_init() { 2905 if (sched_getcpu() != -1) { // Requires sched_getcpu() support 2906 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2907 if (handle != NULL) { 2908 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2909 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2910 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2911 libnuma_dlsym(handle, "numa_max_node"))); 2912 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t, 2913 libnuma_dlsym(handle, "numa_num_configured_nodes"))); 2914 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2915 libnuma_dlsym(handle, "numa_available"))); 2916 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2917 libnuma_dlsym(handle, "numa_tonode_memory"))); 2918 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2919 libnuma_dlsym(handle, "numa_interleave_memory"))); 2920 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t, 2921 libnuma_v2_dlsym(handle, "numa_interleave_memory"))); 2922 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2923 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2924 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t, 2925 libnuma_dlsym(handle, "numa_bitmask_isbitset"))); 2926 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t, 2927 libnuma_dlsym(handle, "numa_distance"))); 2928 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t, 2929 libnuma_v2_dlsym(handle, "numa_get_membind"))); 2930 set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t, 2931 libnuma_v2_dlsym(handle, "numa_get_interleave_mask"))); 2932 2933 if (numa_available() != -1) { 2934 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2935 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr")); 2936 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr")); 2937 set_numa_interleave_bitmask(_numa_get_interleave_mask()); 2938 set_numa_membind_bitmask(_numa_get_membind()); 2939 // Create an index -> node mapping, since nodes are not always consecutive 2940 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2941 rebuild_nindex_to_node_map(); 2942 // Create a cpu -> node mapping 2943 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2944 rebuild_cpu_to_node_map(); 2945 return true; 2946 } 2947 } 2948 } 2949 return false; 2950 } 2951 2952 size_t os::Linux::default_guard_size(os::ThreadType thr_type) { 2953 // Creating guard page is very expensive. Java thread has HotSpot 2954 // guard pages, only enable glibc guard page for non-Java threads. 2955 // (Remember: compiler thread is a Java thread, too!) 2956 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size()); 2957 } 2958 2959 void os::Linux::rebuild_nindex_to_node_map() { 2960 int highest_node_number = Linux::numa_max_node(); 2961 2962 nindex_to_node()->clear(); 2963 for (int node = 0; node <= highest_node_number; node++) { 2964 if (Linux::is_node_in_existing_nodes(node)) { 2965 nindex_to_node()->append(node); 2966 } 2967 } 2968 } 2969 2970 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2971 // The table is later used in get_node_by_cpu(). 2972 void os::Linux::rebuild_cpu_to_node_map() { 2973 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2974 // in libnuma (possible values are starting from 16, 2975 // and continuing up with every other power of 2, but less 2976 // than the maximum number of CPUs supported by kernel), and 2977 // is a subject to change (in libnuma version 2 the requirements 2978 // are more reasonable) we'll just hardcode the number they use 2979 // in the library. 2980 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2981 2982 size_t cpu_num = processor_count(); 2983 size_t cpu_map_size = NCPUS / BitsPerCLong; 2984 size_t cpu_map_valid_size = 2985 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2986 2987 cpu_to_node()->clear(); 2988 cpu_to_node()->at_grow(cpu_num - 1); 2989 2990 size_t node_num = get_existing_num_nodes(); 2991 2992 int distance = 0; 2993 int closest_distance = INT_MAX; 2994 int closest_node = 0; 2995 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2996 for (size_t i = 0; i < node_num; i++) { 2997 // Check if node is configured (not a memory-less node). If it is not, find 2998 // the closest configured node. Check also if node is bound, i.e. it's allowed 2999 // to allocate memory from the node. If it's not allowed, map cpus in that node 3000 // to the closest node from which memory allocation is allowed. 3001 if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) || 3002 !is_node_in_bound_nodes(nindex_to_node()->at(i))) { 3003 closest_distance = INT_MAX; 3004 // Check distance from all remaining nodes in the system. Ignore distance 3005 // from itself, from another non-configured node, and from another non-bound 3006 // node. 3007 for (size_t m = 0; m < node_num; m++) { 3008 if (m != i && 3009 is_node_in_configured_nodes(nindex_to_node()->at(m)) && 3010 is_node_in_bound_nodes(nindex_to_node()->at(m))) { 3011 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m)); 3012 // If a closest node is found, update. There is always at least one 3013 // configured and bound node in the system so there is always at least 3014 // one node close. 3015 if (distance != 0 && distance < closest_distance) { 3016 closest_distance = distance; 3017 closest_node = nindex_to_node()->at(m); 3018 } 3019 } 3020 } 3021 } else { 3022 // Current node is already a configured node. 3023 closest_node = nindex_to_node()->at(i); 3024 } 3025 3026 // Get cpus from the original node and map them to the closest node. If node 3027 // is a configured node (not a memory-less node), then original node and 3028 // closest node are the same. 3029 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 3030 for (size_t j = 0; j < cpu_map_valid_size; j++) { 3031 if (cpu_map[j] != 0) { 3032 for (size_t k = 0; k < BitsPerCLong; k++) { 3033 if (cpu_map[j] & (1UL << k)) { 3034 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node); 3035 } 3036 } 3037 } 3038 } 3039 } 3040 } 3041 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 3042 } 3043 3044 int os::Linux::get_node_by_cpu(int cpu_id) { 3045 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 3046 return cpu_to_node()->at(cpu_id); 3047 } 3048 return -1; 3049 } 3050 3051 GrowableArray<int>* os::Linux::_cpu_to_node; 3052 GrowableArray<int>* os::Linux::_nindex_to_node; 3053 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 3054 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 3055 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 3056 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes; 3057 os::Linux::numa_available_func_t os::Linux::_numa_available; 3058 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 3059 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 3060 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2; 3061 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 3062 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset; 3063 os::Linux::numa_distance_func_t os::Linux::_numa_distance; 3064 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind; 3065 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask; 3066 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy; 3067 unsigned long* os::Linux::_numa_all_nodes; 3068 struct bitmask* os::Linux::_numa_all_nodes_ptr; 3069 struct bitmask* os::Linux::_numa_nodes_ptr; 3070 struct bitmask* os::Linux::_numa_interleave_bitmask; 3071 struct bitmask* os::Linux::_numa_membind_bitmask; 3072 3073 bool os::pd_uncommit_memory(char* addr, size_t size) { 3074 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 3075 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 3076 return res != (uintptr_t) MAP_FAILED; 3077 } 3078 3079 static address get_stack_commited_bottom(address bottom, size_t size) { 3080 address nbot = bottom; 3081 address ntop = bottom + size; 3082 3083 size_t page_sz = os::vm_page_size(); 3084 unsigned pages = size / page_sz; 3085 3086 unsigned char vec[1]; 3087 unsigned imin = 1, imax = pages + 1, imid; 3088 int mincore_return_value = 0; 3089 3090 assert(imin <= imax, "Unexpected page size"); 3091 3092 while (imin < imax) { 3093 imid = (imax + imin) / 2; 3094 nbot = ntop - (imid * page_sz); 3095 3096 // Use a trick with mincore to check whether the page is mapped or not. 3097 // mincore sets vec to 1 if page resides in memory and to 0 if page 3098 // is swapped output but if page we are asking for is unmapped 3099 // it returns -1,ENOMEM 3100 mincore_return_value = mincore(nbot, page_sz, vec); 3101 3102 if (mincore_return_value == -1) { 3103 // Page is not mapped go up 3104 // to find first mapped page 3105 if (errno != EAGAIN) { 3106 assert(errno == ENOMEM, "Unexpected mincore errno"); 3107 imax = imid; 3108 } 3109 } else { 3110 // Page is mapped go down 3111 // to find first not mapped page 3112 imin = imid + 1; 3113 } 3114 } 3115 3116 nbot = nbot + page_sz; 3117 3118 // Adjust stack bottom one page up if last checked page is not mapped 3119 if (mincore_return_value == -1) { 3120 nbot = nbot + page_sz; 3121 } 3122 3123 return nbot; 3124 } 3125 3126 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) { 3127 int mincore_return_value; 3128 const size_t stripe = 1024; // query this many pages each time 3129 unsigned char vec[stripe + 1]; 3130 // set a guard 3131 vec[stripe] = 'X'; 3132 3133 const size_t page_sz = os::vm_page_size(); 3134 size_t pages = size / page_sz; 3135 3136 assert(is_aligned(start, page_sz), "Start address must be page aligned"); 3137 assert(is_aligned(size, page_sz), "Size must be page aligned"); 3138 3139 committed_start = NULL; 3140 3141 int loops = (pages + stripe - 1) / stripe; 3142 int committed_pages = 0; 3143 address loop_base = start; 3144 bool found_range = false; 3145 3146 for (int index = 0; index < loops && !found_range; index ++) { 3147 assert(pages > 0, "Nothing to do"); 3148 int pages_to_query = (pages >= stripe) ? stripe : pages; 3149 pages -= pages_to_query; 3150 3151 // Get stable read 3152 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN); 3153 3154 // During shutdown, some memory goes away without properly notifying NMT, 3155 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object. 3156 // Bailout and return as not committed for now. 3157 if (mincore_return_value == -1 && errno == ENOMEM) { 3158 return false; 3159 } 3160 3161 assert(vec[stripe] == 'X', "overflow guard"); 3162 assert(mincore_return_value == 0, "Range must be valid"); 3163 // Process this stripe 3164 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) { 3165 if ((vec[vecIdx] & 0x01) == 0) { // not committed 3166 // End of current contiguous region 3167 if (committed_start != NULL) { 3168 found_range = true; 3169 break; 3170 } 3171 } else { // committed 3172 // Start of region 3173 if (committed_start == NULL) { 3174 committed_start = loop_base + page_sz * vecIdx; 3175 } 3176 committed_pages ++; 3177 } 3178 } 3179 3180 loop_base += pages_to_query * page_sz; 3181 } 3182 3183 if (committed_start != NULL) { 3184 assert(committed_pages > 0, "Must have committed region"); 3185 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has"); 3186 assert(committed_start >= start && committed_start < start + size, "Out of range"); 3187 committed_size = page_sz * committed_pages; 3188 return true; 3189 } else { 3190 assert(committed_pages == 0, "Should not have committed region"); 3191 return false; 3192 } 3193 } 3194 3195 3196 // Linux uses a growable mapping for the stack, and if the mapping for 3197 // the stack guard pages is not removed when we detach a thread the 3198 // stack cannot grow beyond the pages where the stack guard was 3199 // mapped. If at some point later in the process the stack expands to 3200 // that point, the Linux kernel cannot expand the stack any further 3201 // because the guard pages are in the way, and a segfault occurs. 3202 // 3203 // However, it's essential not to split the stack region by unmapping 3204 // a region (leaving a hole) that's already part of the stack mapping, 3205 // so if the stack mapping has already grown beyond the guard pages at 3206 // the time we create them, we have to truncate the stack mapping. 3207 // So, we need to know the extent of the stack mapping when 3208 // create_stack_guard_pages() is called. 3209 3210 // We only need this for stacks that are growable: at the time of 3211 // writing thread stacks don't use growable mappings (i.e. those 3212 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3213 // only applies to the main thread. 3214 3215 // If the (growable) stack mapping already extends beyond the point 3216 // where we're going to put our guard pages, truncate the mapping at 3217 // that point by munmap()ping it. This ensures that when we later 3218 // munmap() the guard pages we don't leave a hole in the stack 3219 // mapping. This only affects the main/primordial thread 3220 3221 bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3222 if (os::is_primordial_thread()) { 3223 // As we manually grow stack up to bottom inside create_attached_thread(), 3224 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3225 // we don't need to do anything special. 3226 // Check it first, before calling heavy function. 3227 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3228 unsigned char vec[1]; 3229 3230 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3231 // Fallback to slow path on all errors, including EAGAIN 3232 stack_extent = (uintptr_t) get_stack_commited_bottom( 3233 os::Linux::initial_thread_stack_bottom(), 3234 (size_t)addr - stack_extent); 3235 } 3236 3237 if (stack_extent < (uintptr_t)addr) { 3238 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3239 } 3240 } 3241 3242 return os::commit_memory(addr, size, !ExecMem); 3243 } 3244 3245 // If this is a growable mapping, remove the guard pages entirely by 3246 // munmap()ping them. If not, just call uncommit_memory(). This only 3247 // affects the main/primordial thread, but guard against future OS changes. 3248 // It's safe to always unmap guard pages for primordial thread because we 3249 // always place it right after end of the mapped region. 3250 3251 bool os::remove_stack_guard_pages(char* addr, size_t size) { 3252 uintptr_t stack_extent, stack_base; 3253 3254 if (os::is_primordial_thread()) { 3255 return ::munmap(addr, size) == 0; 3256 } 3257 3258 return os::uncommit_memory(addr, size); 3259 } 3260 3261 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3262 // at 'requested_addr'. If there are existing memory mappings at the same 3263 // location, however, they will be overwritten. If 'fixed' is false, 3264 // 'requested_addr' is only treated as a hint, the return value may or 3265 // may not start from the requested address. Unlike Linux mmap(), this 3266 // function returns NULL to indicate failure. 3267 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3268 char * addr; 3269 int flags; 3270 3271 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3272 if (fixed) { 3273 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3274 flags |= MAP_FIXED; 3275 } 3276 3277 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3278 // touch an uncommitted page. Otherwise, the read/write might 3279 // succeed if we have enough swap space to back the physical page. 3280 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3281 flags, -1, 0); 3282 3283 return addr == MAP_FAILED ? NULL : addr; 3284 } 3285 3286 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address 3287 // (req_addr != NULL) or with a given alignment. 3288 // - bytes shall be a multiple of alignment. 3289 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3290 // - alignment sets the alignment at which memory shall be allocated. 3291 // It must be a multiple of allocation granularity. 3292 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3293 // req_addr or NULL. 3294 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) { 3295 3296 size_t extra_size = bytes; 3297 if (req_addr == NULL && alignment > 0) { 3298 extra_size += alignment; 3299 } 3300 3301 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE, 3302 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 3303 -1, 0); 3304 if (start == MAP_FAILED) { 3305 start = NULL; 3306 } else { 3307 if (req_addr != NULL) { 3308 if (start != req_addr) { 3309 ::munmap(start, extra_size); 3310 start = NULL; 3311 } 3312 } else { 3313 char* const start_aligned = align_up(start, alignment); 3314 char* const end_aligned = start_aligned + bytes; 3315 char* const end = start + extra_size; 3316 if (start_aligned > start) { 3317 ::munmap(start, start_aligned - start); 3318 } 3319 if (end_aligned < end) { 3320 ::munmap(end_aligned, end - end_aligned); 3321 } 3322 start = start_aligned; 3323 } 3324 } 3325 return start; 3326 } 3327 3328 static int anon_munmap(char * addr, size_t size) { 3329 return ::munmap(addr, size) == 0; 3330 } 3331 3332 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3333 size_t alignment_hint) { 3334 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3335 } 3336 3337 bool os::pd_release_memory(char* addr, size_t size) { 3338 return anon_munmap(addr, size); 3339 } 3340 3341 static bool linux_mprotect(char* addr, size_t size, int prot) { 3342 // Linux wants the mprotect address argument to be page aligned. 3343 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size()); 3344 3345 // According to SUSv3, mprotect() should only be used with mappings 3346 // established by mmap(), and mmap() always maps whole pages. Unaligned 3347 // 'addr' likely indicates problem in the VM (e.g. trying to change 3348 // protection of malloc'ed or statically allocated memory). Check the 3349 // caller if you hit this assert. 3350 assert(addr == bottom, "sanity check"); 3351 3352 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3353 return ::mprotect(bottom, size, prot) == 0; 3354 } 3355 3356 // Set protections specified 3357 bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3358 bool is_committed) { 3359 unsigned int p = 0; 3360 switch (prot) { 3361 case MEM_PROT_NONE: p = PROT_NONE; break; 3362 case MEM_PROT_READ: p = PROT_READ; break; 3363 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3364 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3365 default: 3366 ShouldNotReachHere(); 3367 } 3368 // is_committed is unused. 3369 return linux_mprotect(addr, bytes, p); 3370 } 3371 3372 bool os::guard_memory(char* addr, size_t size) { 3373 return linux_mprotect(addr, size, PROT_NONE); 3374 } 3375 3376 bool os::unguard_memory(char* addr, size_t size) { 3377 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3378 } 3379 3380 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3381 size_t page_size) { 3382 bool result = false; 3383 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3384 MAP_ANONYMOUS|MAP_PRIVATE, 3385 -1, 0); 3386 if (p != MAP_FAILED) { 3387 void *aligned_p = align_up(p, page_size); 3388 3389 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3390 3391 munmap(p, page_size * 2); 3392 } 3393 3394 if (warn && !result) { 3395 warning("TransparentHugePages is not supported by the operating system."); 3396 } 3397 3398 return result; 3399 } 3400 3401 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3402 bool result = false; 3403 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3404 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3405 -1, 0); 3406 3407 if (p != MAP_FAILED) { 3408 // We don't know if this really is a huge page or not. 3409 FILE *fp = fopen("/proc/self/maps", "r"); 3410 if (fp) { 3411 while (!feof(fp)) { 3412 char chars[257]; 3413 long x = 0; 3414 if (fgets(chars, sizeof(chars), fp)) { 3415 if (sscanf(chars, "%lx-%*x", &x) == 1 3416 && x == (long)p) { 3417 if (strstr (chars, "hugepage")) { 3418 result = true; 3419 break; 3420 } 3421 } 3422 } 3423 } 3424 fclose(fp); 3425 } 3426 munmap(p, page_size); 3427 } 3428 3429 if (warn && !result) { 3430 warning("HugeTLBFS is not supported by the operating system."); 3431 } 3432 3433 return result; 3434 } 3435 3436 // From the coredump_filter documentation: 3437 // 3438 // - (bit 0) anonymous private memory 3439 // - (bit 1) anonymous shared memory 3440 // - (bit 2) file-backed private memory 3441 // - (bit 3) file-backed shared memory 3442 // - (bit 4) ELF header pages in file-backed private memory areas (it is 3443 // effective only if the bit 2 is cleared) 3444 // - (bit 5) hugetlb private memory 3445 // - (bit 6) hugetlb shared memory 3446 // - (bit 7) dax private memory 3447 // - (bit 8) dax shared memory 3448 // 3449 static void set_coredump_filter(CoredumpFilterBit bit) { 3450 FILE *f; 3451 long cdm; 3452 3453 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3454 return; 3455 } 3456 3457 if (fscanf(f, "%lx", &cdm) != 1) { 3458 fclose(f); 3459 return; 3460 } 3461 3462 long saved_cdm = cdm; 3463 rewind(f); 3464 cdm |= bit; 3465 3466 if (cdm != saved_cdm) { 3467 fprintf(f, "%#lx", cdm); 3468 } 3469 3470 fclose(f); 3471 } 3472 3473 // Large page support 3474 3475 static size_t _large_page_size = 0; 3476 3477 size_t os::Linux::find_large_page_size() { 3478 size_t large_page_size = 0; 3479 3480 // large_page_size on Linux is used to round up heap size. x86 uses either 3481 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3482 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3483 // page as large as 256M. 3484 // 3485 // Here we try to figure out page size by parsing /proc/meminfo and looking 3486 // for a line with the following format: 3487 // Hugepagesize: 2048 kB 3488 // 3489 // If we can't determine the value (e.g. /proc is not mounted, or the text 3490 // format has been changed), we'll use the largest page size supported by 3491 // the processor. 3492 3493 #ifndef ZERO 3494 large_page_size = 3495 AARCH64_ONLY(2 * M) 3496 AMD64_ONLY(2 * M) 3497 ARM32_ONLY(2 * M) 3498 IA32_ONLY(4 * M) 3499 IA64_ONLY(256 * M) 3500 PPC_ONLY(4 * M) 3501 S390_ONLY(1 * M) 3502 SPARC_ONLY(4 * M); 3503 #endif // ZERO 3504 3505 FILE *fp = fopen("/proc/meminfo", "r"); 3506 if (fp) { 3507 while (!feof(fp)) { 3508 int x = 0; 3509 char buf[16]; 3510 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3511 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3512 large_page_size = x * K; 3513 break; 3514 } 3515 } else { 3516 // skip to next line 3517 for (;;) { 3518 int ch = fgetc(fp); 3519 if (ch == EOF || ch == (int)'\n') break; 3520 } 3521 } 3522 } 3523 fclose(fp); 3524 } 3525 3526 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3527 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3528 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3529 proper_unit_for_byte_size(large_page_size)); 3530 } 3531 3532 return large_page_size; 3533 } 3534 3535 size_t os::Linux::setup_large_page_size() { 3536 _large_page_size = Linux::find_large_page_size(); 3537 const size_t default_page_size = (size_t)Linux::page_size(); 3538 if (_large_page_size > default_page_size) { 3539 _page_sizes[0] = _large_page_size; 3540 _page_sizes[1] = default_page_size; 3541 _page_sizes[2] = 0; 3542 } 3543 3544 return _large_page_size; 3545 } 3546 3547 bool os::Linux::setup_large_page_type(size_t page_size) { 3548 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3549 FLAG_IS_DEFAULT(UseSHM) && 3550 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3551 3552 // The type of large pages has not been specified by the user. 3553 3554 // Try UseHugeTLBFS and then UseSHM. 3555 UseHugeTLBFS = UseSHM = true; 3556 3557 // Don't try UseTransparentHugePages since there are known 3558 // performance issues with it turned on. This might change in the future. 3559 UseTransparentHugePages = false; 3560 } 3561 3562 if (UseTransparentHugePages) { 3563 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3564 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3565 UseHugeTLBFS = false; 3566 UseSHM = false; 3567 return true; 3568 } 3569 UseTransparentHugePages = false; 3570 } 3571 3572 if (UseHugeTLBFS) { 3573 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3574 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3575 UseSHM = false; 3576 return true; 3577 } 3578 UseHugeTLBFS = false; 3579 } 3580 3581 return UseSHM; 3582 } 3583 3584 void os::large_page_init() { 3585 if (!UseLargePages && 3586 !UseTransparentHugePages && 3587 !UseHugeTLBFS && 3588 !UseSHM) { 3589 // Not using large pages. 3590 return; 3591 } 3592 3593 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3594 // The user explicitly turned off large pages. 3595 // Ignore the rest of the large pages flags. 3596 UseTransparentHugePages = false; 3597 UseHugeTLBFS = false; 3598 UseSHM = false; 3599 return; 3600 } 3601 3602 size_t large_page_size = Linux::setup_large_page_size(); 3603 UseLargePages = Linux::setup_large_page_type(large_page_size); 3604 3605 set_coredump_filter(LARGEPAGES_BIT); 3606 } 3607 3608 #ifndef SHM_HUGETLB 3609 #define SHM_HUGETLB 04000 3610 #endif 3611 3612 #define shm_warning_format(format, ...) \ 3613 do { \ 3614 if (UseLargePages && \ 3615 (!FLAG_IS_DEFAULT(UseLargePages) || \ 3616 !FLAG_IS_DEFAULT(UseSHM) || \ 3617 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \ 3618 warning(format, __VA_ARGS__); \ 3619 } \ 3620 } while (0) 3621 3622 #define shm_warning(str) shm_warning_format("%s", str) 3623 3624 #define shm_warning_with_errno(str) \ 3625 do { \ 3626 int err = errno; \ 3627 shm_warning_format(str " (error = %d)", err); \ 3628 } while (0) 3629 3630 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) { 3631 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment"); 3632 3633 if (!is_aligned(alignment, SHMLBA)) { 3634 assert(false, "Code below assumes that alignment is at least SHMLBA aligned"); 3635 return NULL; 3636 } 3637 3638 // To ensure that we get 'alignment' aligned memory from shmat, 3639 // we pre-reserve aligned virtual memory and then attach to that. 3640 3641 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL); 3642 if (pre_reserved_addr == NULL) { 3643 // Couldn't pre-reserve aligned memory. 3644 shm_warning("Failed to pre-reserve aligned memory for shmat."); 3645 return NULL; 3646 } 3647 3648 // SHM_REMAP is needed to allow shmat to map over an existing mapping. 3649 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP); 3650 3651 if ((intptr_t)addr == -1) { 3652 int err = errno; 3653 shm_warning_with_errno("Failed to attach shared memory."); 3654 3655 assert(err != EACCES, "Unexpected error"); 3656 assert(err != EIDRM, "Unexpected error"); 3657 assert(err != EINVAL, "Unexpected error"); 3658 3659 // Since we don't know if the kernel unmapped the pre-reserved memory area 3660 // we can't unmap it, since that would potentially unmap memory that was 3661 // mapped from other threads. 3662 return NULL; 3663 } 3664 3665 return addr; 3666 } 3667 3668 static char* shmat_at_address(int shmid, char* req_addr) { 3669 if (!is_aligned(req_addr, SHMLBA)) { 3670 assert(false, "Requested address needs to be SHMLBA aligned"); 3671 return NULL; 3672 } 3673 3674 char* addr = (char*)shmat(shmid, req_addr, 0); 3675 3676 if ((intptr_t)addr == -1) { 3677 shm_warning_with_errno("Failed to attach shared memory."); 3678 return NULL; 3679 } 3680 3681 return addr; 3682 } 3683 3684 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) { 3685 // If a req_addr has been provided, we assume that the caller has already aligned the address. 3686 if (req_addr != NULL) { 3687 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size"); 3688 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment"); 3689 return shmat_at_address(shmid, req_addr); 3690 } 3691 3692 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically 3693 // return large page size aligned memory addresses when req_addr == NULL. 3694 // However, if the alignment is larger than the large page size, we have 3695 // to manually ensure that the memory returned is 'alignment' aligned. 3696 if (alignment > os::large_page_size()) { 3697 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size"); 3698 return shmat_with_alignment(shmid, bytes, alignment); 3699 } else { 3700 return shmat_at_address(shmid, NULL); 3701 } 3702 } 3703 3704 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3705 char* req_addr, bool exec) { 3706 // "exec" is passed in but not used. Creating the shared image for 3707 // the code cache doesn't have an SHM_X executable permission to check. 3708 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3709 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3710 assert(is_aligned(req_addr, alignment), "Unaligned address"); 3711 3712 if (!is_aligned(bytes, os::large_page_size())) { 3713 return NULL; // Fallback to small pages. 3714 } 3715 3716 // Create a large shared memory region to attach to based on size. 3717 // Currently, size is the total size of the heap. 3718 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3719 if (shmid == -1) { 3720 // Possible reasons for shmget failure: 3721 // 1. shmmax is too small for Java heap. 3722 // > check shmmax value: cat /proc/sys/kernel/shmmax 3723 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3724 // 2. not enough large page memory. 3725 // > check available large pages: cat /proc/meminfo 3726 // > increase amount of large pages: 3727 // echo new_value > /proc/sys/vm/nr_hugepages 3728 // Note 1: different Linux may use different name for this property, 3729 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3730 // Note 2: it's possible there's enough physical memory available but 3731 // they are so fragmented after a long run that they can't 3732 // coalesce into large pages. Try to reserve large pages when 3733 // the system is still "fresh". 3734 shm_warning_with_errno("Failed to reserve shared memory."); 3735 return NULL; 3736 } 3737 3738 // Attach to the region. 3739 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr); 3740 3741 // Remove shmid. If shmat() is successful, the actual shared memory segment 3742 // will be deleted when it's detached by shmdt() or when the process 3743 // terminates. If shmat() is not successful this will remove the shared 3744 // segment immediately. 3745 shmctl(shmid, IPC_RMID, NULL); 3746 3747 return addr; 3748 } 3749 3750 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3751 int error) { 3752 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3753 3754 bool warn_on_failure = UseLargePages && 3755 (!FLAG_IS_DEFAULT(UseLargePages) || 3756 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3757 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3758 3759 if (warn_on_failure) { 3760 char msg[128]; 3761 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3762 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3763 warning("%s", msg); 3764 } 3765 } 3766 3767 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3768 char* req_addr, 3769 bool exec) { 3770 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3771 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size"); 3772 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3773 3774 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3775 char* addr = (char*)::mmap(req_addr, bytes, prot, 3776 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3777 -1, 0); 3778 3779 if (addr == MAP_FAILED) { 3780 warn_on_large_pages_failure(req_addr, bytes, errno); 3781 return NULL; 3782 } 3783 3784 assert(is_aligned(addr, os::large_page_size()), "Must be"); 3785 3786 return addr; 3787 } 3788 3789 // Reserve memory using mmap(MAP_HUGETLB). 3790 // - bytes shall be a multiple of alignment. 3791 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3792 // - alignment sets the alignment at which memory shall be allocated. 3793 // It must be a multiple of allocation granularity. 3794 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3795 // req_addr or NULL. 3796 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3797 size_t alignment, 3798 char* req_addr, 3799 bool exec) { 3800 size_t large_page_size = os::large_page_size(); 3801 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3802 3803 assert(is_aligned(req_addr, alignment), "Must be"); 3804 assert(is_aligned(bytes, alignment), "Must be"); 3805 3806 // First reserve - but not commit - the address range in small pages. 3807 char* const start = anon_mmap_aligned(bytes, alignment, req_addr); 3808 3809 if (start == NULL) { 3810 return NULL; 3811 } 3812 3813 assert(is_aligned(start, alignment), "Must be"); 3814 3815 char* end = start + bytes; 3816 3817 // Find the regions of the allocated chunk that can be promoted to large pages. 3818 char* lp_start = align_up(start, large_page_size); 3819 char* lp_end = align_down(end, large_page_size); 3820 3821 size_t lp_bytes = lp_end - lp_start; 3822 3823 assert(is_aligned(lp_bytes, large_page_size), "Must be"); 3824 3825 if (lp_bytes == 0) { 3826 // The mapped region doesn't even span the start and the end of a large page. 3827 // Fall back to allocate a non-special area. 3828 ::munmap(start, end - start); 3829 return NULL; 3830 } 3831 3832 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3833 3834 void* result; 3835 3836 // Commit small-paged leading area. 3837 if (start != lp_start) { 3838 result = ::mmap(start, lp_start - start, prot, 3839 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3840 -1, 0); 3841 if (result == MAP_FAILED) { 3842 ::munmap(lp_start, end - lp_start); 3843 return NULL; 3844 } 3845 } 3846 3847 // Commit large-paged area. 3848 result = ::mmap(lp_start, lp_bytes, prot, 3849 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3850 -1, 0); 3851 if (result == MAP_FAILED) { 3852 warn_on_large_pages_failure(lp_start, lp_bytes, errno); 3853 // If the mmap above fails, the large pages region will be unmapped and we 3854 // have regions before and after with small pages. Release these regions. 3855 // 3856 // | mapped | unmapped | mapped | 3857 // ^ ^ ^ ^ 3858 // start lp_start lp_end end 3859 // 3860 ::munmap(start, lp_start - start); 3861 ::munmap(lp_end, end - lp_end); 3862 return NULL; 3863 } 3864 3865 // Commit small-paged trailing area. 3866 if (lp_end != end) { 3867 result = ::mmap(lp_end, end - lp_end, prot, 3868 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3869 -1, 0); 3870 if (result == MAP_FAILED) { 3871 ::munmap(start, lp_end - start); 3872 return NULL; 3873 } 3874 } 3875 3876 return start; 3877 } 3878 3879 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3880 size_t alignment, 3881 char* req_addr, 3882 bool exec) { 3883 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3884 assert(is_aligned(req_addr, alignment), "Must be"); 3885 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be"); 3886 assert(is_power_of_2(os::large_page_size()), "Must be"); 3887 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3888 3889 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3890 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3891 } else { 3892 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3893 } 3894 } 3895 3896 char* os::reserve_memory_special(size_t bytes, size_t alignment, 3897 char* req_addr, bool exec) { 3898 assert(UseLargePages, "only for large pages"); 3899 3900 char* addr; 3901 if (UseSHM) { 3902 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3903 } else { 3904 assert(UseHugeTLBFS, "must be"); 3905 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3906 } 3907 3908 if (addr != NULL) { 3909 if (UseNUMAInterleaving) { 3910 numa_make_global(addr, bytes); 3911 } 3912 3913 // The memory is committed 3914 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3915 } 3916 3917 return addr; 3918 } 3919 3920 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3921 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3922 return shmdt(base) == 0; 3923 } 3924 3925 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3926 return pd_release_memory(base, bytes); 3927 } 3928 3929 bool os::release_memory_special(char* base, size_t bytes) { 3930 bool res; 3931 if (MemTracker::tracking_level() > NMT_minimal) { 3932 Tracker tkr(Tracker::release); 3933 res = os::Linux::release_memory_special_impl(base, bytes); 3934 if (res) { 3935 tkr.record((address)base, bytes); 3936 } 3937 3938 } else { 3939 res = os::Linux::release_memory_special_impl(base, bytes); 3940 } 3941 return res; 3942 } 3943 3944 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3945 assert(UseLargePages, "only for large pages"); 3946 bool res; 3947 3948 if (UseSHM) { 3949 res = os::Linux::release_memory_special_shm(base, bytes); 3950 } else { 3951 assert(UseHugeTLBFS, "must be"); 3952 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3953 } 3954 return res; 3955 } 3956 3957 size_t os::large_page_size() { 3958 return _large_page_size; 3959 } 3960 3961 // With SysV SHM the entire memory region must be allocated as shared 3962 // memory. 3963 // HugeTLBFS allows application to commit large page memory on demand. 3964 // However, when committing memory with HugeTLBFS fails, the region 3965 // that was supposed to be committed will lose the old reservation 3966 // and allow other threads to steal that memory region. Because of this 3967 // behavior we can't commit HugeTLBFS memory. 3968 bool os::can_commit_large_page_memory() { 3969 return UseTransparentHugePages; 3970 } 3971 3972 bool os::can_execute_large_page_memory() { 3973 return UseTransparentHugePages || UseHugeTLBFS; 3974 } 3975 3976 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) { 3977 assert(file_desc >= 0, "file_desc is not valid"); 3978 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr); 3979 if (result != NULL) { 3980 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) { 3981 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory")); 3982 } 3983 } 3984 return result; 3985 } 3986 3987 // Reserve memory at an arbitrary address, only if that area is 3988 // available (and not reserved for something else). 3989 3990 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3991 const int max_tries = 10; 3992 char* base[max_tries]; 3993 size_t size[max_tries]; 3994 const size_t gap = 0x000000; 3995 3996 // Assert only that the size is a multiple of the page size, since 3997 // that's all that mmap requires, and since that's all we really know 3998 // about at this low abstraction level. If we need higher alignment, 3999 // we can either pass an alignment to this method or verify alignment 4000 // in one of the methods further up the call chain. See bug 5044738. 4001 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 4002 4003 // Repeatedly allocate blocks until the block is allocated at the 4004 // right spot. 4005 4006 // Linux mmap allows caller to pass an address as hint; give it a try first, 4007 // if kernel honors the hint then we can return immediately. 4008 char * addr = anon_mmap(requested_addr, bytes, false); 4009 if (addr == requested_addr) { 4010 return requested_addr; 4011 } 4012 4013 if (addr != NULL) { 4014 // mmap() is successful but it fails to reserve at the requested address 4015 anon_munmap(addr, bytes); 4016 } 4017 4018 int i; 4019 for (i = 0; i < max_tries; ++i) { 4020 base[i] = reserve_memory(bytes); 4021 4022 if (base[i] != NULL) { 4023 // Is this the block we wanted? 4024 if (base[i] == requested_addr) { 4025 size[i] = bytes; 4026 break; 4027 } 4028 4029 // Does this overlap the block we wanted? Give back the overlapped 4030 // parts and try again. 4031 4032 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i]; 4033 if (top_overlap >= 0 && (size_t)top_overlap < bytes) { 4034 unmap_memory(base[i], top_overlap); 4035 base[i] += top_overlap; 4036 size[i] = bytes - top_overlap; 4037 } else { 4038 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr; 4039 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) { 4040 unmap_memory(requested_addr, bottom_overlap); 4041 size[i] = bytes - bottom_overlap; 4042 } else { 4043 size[i] = bytes; 4044 } 4045 } 4046 } 4047 } 4048 4049 // Give back the unused reserved pieces. 4050 4051 for (int j = 0; j < i; ++j) { 4052 if (base[j] != NULL) { 4053 unmap_memory(base[j], size[j]); 4054 } 4055 } 4056 4057 if (i < max_tries) { 4058 return requested_addr; 4059 } else { 4060 return NULL; 4061 } 4062 } 4063 4064 // Sleep forever; naked call to OS-specific sleep; use with CAUTION 4065 void os::infinite_sleep() { 4066 while (true) { // sleep forever ... 4067 ::sleep(100); // ... 100 seconds at a time 4068 } 4069 } 4070 4071 // Used to convert frequent JVM_Yield() to nops 4072 bool os::dont_yield() { 4073 return DontYieldALot; 4074 } 4075 4076 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will 4077 // actually give up the CPU. Since skip buddy (v2.6.28): 4078 // 4079 // * Sets the yielding task as skip buddy for current CPU's run queue. 4080 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task). 4081 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy). 4082 // 4083 // An alternative is calling os::naked_short_nanosleep with a small number to avoid 4084 // getting re-scheduled immediately. 4085 // 4086 void os::naked_yield() { 4087 sched_yield(); 4088 } 4089 4090 //////////////////////////////////////////////////////////////////////////////// 4091 // thread priority support 4092 4093 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 4094 // only supports dynamic priority, static priority must be zero. For real-time 4095 // applications, Linux supports SCHED_RR which allows static priority (1-99). 4096 // However, for large multi-threaded applications, SCHED_RR is not only slower 4097 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 4098 // of 5 runs - Sep 2005). 4099 // 4100 // The following code actually changes the niceness of kernel-thread/LWP. It 4101 // has an assumption that setpriority() only modifies one kernel-thread/LWP, 4102 // not the entire user process, and user level threads are 1:1 mapped to kernel 4103 // threads. It has always been the case, but could change in the future. For 4104 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). 4105 // It is only used when ThreadPriorityPolicy=1 and may require system level permission 4106 // (e.g., root privilege or CAP_SYS_NICE capability). 4107 4108 int os::java_to_os_priority[CriticalPriority + 1] = { 4109 19, // 0 Entry should never be used 4110 4111 4, // 1 MinPriority 4112 3, // 2 4113 2, // 3 4114 4115 1, // 4 4116 0, // 5 NormPriority 4117 -1, // 6 4118 4119 -2, // 7 4120 -3, // 8 4121 -4, // 9 NearMaxPriority 4122 4123 -5, // 10 MaxPriority 4124 4125 -5 // 11 CriticalPriority 4126 }; 4127 4128 static int prio_init() { 4129 if (ThreadPriorityPolicy == 1) { 4130 if (geteuid() != 0) { 4131 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 4132 warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \ 4133 "e.g., being the root user. If the necessary permission is not " \ 4134 "possessed, changes to priority will be silently ignored."); 4135 } 4136 } 4137 } 4138 if (UseCriticalJavaThreadPriority) { 4139 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 4140 } 4141 return 0; 4142 } 4143 4144 OSReturn os::set_native_priority(Thread* thread, int newpri) { 4145 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 4146 4147 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 4148 return (ret == 0) ? OS_OK : OS_ERR; 4149 } 4150 4151 OSReturn os::get_native_priority(const Thread* const thread, 4152 int *priority_ptr) { 4153 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 4154 *priority_ptr = java_to_os_priority[NormPriority]; 4155 return OS_OK; 4156 } 4157 4158 errno = 0; 4159 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 4160 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 4161 } 4162 4163 //////////////////////////////////////////////////////////////////////////////// 4164 // suspend/resume support 4165 4166 // The low-level signal-based suspend/resume support is a remnant from the 4167 // old VM-suspension that used to be for java-suspension, safepoints etc, 4168 // within hotspot. Currently used by JFR's OSThreadSampler 4169 // 4170 // The remaining code is greatly simplified from the more general suspension 4171 // code that used to be used. 4172 // 4173 // The protocol is quite simple: 4174 // - suspend: 4175 // - sends a signal to the target thread 4176 // - polls the suspend state of the osthread using a yield loop 4177 // - target thread signal handler (SR_handler) sets suspend state 4178 // and blocks in sigsuspend until continued 4179 // - resume: 4180 // - sets target osthread state to continue 4181 // - sends signal to end the sigsuspend loop in the SR_handler 4182 // 4183 // Note that the SR_lock plays no role in this suspend/resume protocol, 4184 // but is checked for NULL in SR_handler as a thread termination indicator. 4185 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs. 4186 // 4187 // Note that resume_clear_context() and suspend_save_context() are needed 4188 // by SR_handler(), so that fetch_frame_from_ucontext() works, 4189 // which in part is used by: 4190 // - Forte Analyzer: AsyncGetCallTrace() 4191 // - StackBanging: get_frame_at_stack_banging_point() 4192 4193 static void resume_clear_context(OSThread *osthread) { 4194 osthread->set_ucontext(NULL); 4195 osthread->set_siginfo(NULL); 4196 } 4197 4198 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 4199 ucontext_t* context) { 4200 osthread->set_ucontext(context); 4201 osthread->set_siginfo(siginfo); 4202 } 4203 4204 // Handler function invoked when a thread's execution is suspended or 4205 // resumed. We have to be careful that only async-safe functions are 4206 // called here (Note: most pthread functions are not async safe and 4207 // should be avoided.) 4208 // 4209 // Note: sigwait() is a more natural fit than sigsuspend() from an 4210 // interface point of view, but sigwait() prevents the signal hander 4211 // from being run. libpthread would get very confused by not having 4212 // its signal handlers run and prevents sigwait()'s use with the 4213 // mutex granting granting signal. 4214 // 4215 // Currently only ever called on the VMThread and JavaThreads (PC sampling) 4216 // 4217 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 4218 // Save and restore errno to avoid confusing native code with EINTR 4219 // after sigsuspend. 4220 int old_errno = errno; 4221 4222 Thread* thread = Thread::current_or_null_safe(); 4223 assert(thread != NULL, "Missing current thread in SR_handler"); 4224 4225 // On some systems we have seen signal delivery get "stuck" until the signal 4226 // mask is changed as part of thread termination. Check that the current thread 4227 // has not already terminated (via SR_lock()) - else the following assertion 4228 // will fail because the thread is no longer a JavaThread as the ~JavaThread 4229 // destructor has completed. 4230 4231 if (thread->SR_lock() == NULL) { 4232 return; 4233 } 4234 4235 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 4236 4237 OSThread* osthread = thread->osthread(); 4238 4239 os::SuspendResume::State current = osthread->sr.state(); 4240 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 4241 suspend_save_context(osthread, siginfo, context); 4242 4243 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4244 os::SuspendResume::State state = osthread->sr.suspended(); 4245 if (state == os::SuspendResume::SR_SUSPENDED) { 4246 sigset_t suspend_set; // signals for sigsuspend() 4247 sigemptyset(&suspend_set); 4248 // get current set of blocked signals and unblock resume signal 4249 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4250 sigdelset(&suspend_set, SR_signum); 4251 4252 sr_semaphore.signal(); 4253 // wait here until we are resumed 4254 while (1) { 4255 sigsuspend(&suspend_set); 4256 4257 os::SuspendResume::State result = osthread->sr.running(); 4258 if (result == os::SuspendResume::SR_RUNNING) { 4259 sr_semaphore.signal(); 4260 break; 4261 } 4262 } 4263 4264 } else if (state == os::SuspendResume::SR_RUNNING) { 4265 // request was cancelled, continue 4266 } else { 4267 ShouldNotReachHere(); 4268 } 4269 4270 resume_clear_context(osthread); 4271 } else if (current == os::SuspendResume::SR_RUNNING) { 4272 // request was cancelled, continue 4273 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4274 // ignore 4275 } else { 4276 // ignore 4277 } 4278 4279 errno = old_errno; 4280 } 4281 4282 static int SR_initialize() { 4283 struct sigaction act; 4284 char *s; 4285 4286 // Get signal number to use for suspend/resume 4287 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4288 int sig = ::strtol(s, 0, 10); 4289 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769. 4290 sig < NSIG) { // Must be legal signal and fit into sigflags[]. 4291 SR_signum = sig; 4292 } else { 4293 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.", 4294 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum); 4295 } 4296 } 4297 4298 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4299 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4300 4301 sigemptyset(&SR_sigset); 4302 sigaddset(&SR_sigset, SR_signum); 4303 4304 // Set up signal handler for suspend/resume 4305 act.sa_flags = SA_RESTART|SA_SIGINFO; 4306 act.sa_handler = (void (*)(int)) SR_handler; 4307 4308 // SR_signum is blocked by default. 4309 // 4528190 - We also need to block pthread restart signal (32 on all 4310 // supported Linux platforms). Note that LinuxThreads need to block 4311 // this signal for all threads to work properly. So we don't have 4312 // to use hard-coded signal number when setting up the mask. 4313 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4314 4315 if (sigaction(SR_signum, &act, 0) == -1) { 4316 return -1; 4317 } 4318 4319 // Save signal flag 4320 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4321 return 0; 4322 } 4323 4324 static int sr_notify(OSThread* osthread) { 4325 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4326 assert_status(status == 0, status, "pthread_kill"); 4327 return status; 4328 } 4329 4330 // "Randomly" selected value for how long we want to spin 4331 // before bailing out on suspending a thread, also how often 4332 // we send a signal to a thread we want to resume 4333 static const int RANDOMLY_LARGE_INTEGER = 1000000; 4334 static const int RANDOMLY_LARGE_INTEGER2 = 100; 4335 4336 // returns true on success and false on error - really an error is fatal 4337 // but this seems the normal response to library errors 4338 static bool do_suspend(OSThread* osthread) { 4339 assert(osthread->sr.is_running(), "thread should be running"); 4340 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4341 4342 // mark as suspended and send signal 4343 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4344 // failed to switch, state wasn't running? 4345 ShouldNotReachHere(); 4346 return false; 4347 } 4348 4349 if (sr_notify(osthread) != 0) { 4350 ShouldNotReachHere(); 4351 } 4352 4353 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4354 while (true) { 4355 if (sr_semaphore.timedwait(2)) { 4356 break; 4357 } else { 4358 // timeout 4359 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4360 if (cancelled == os::SuspendResume::SR_RUNNING) { 4361 return false; 4362 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4363 // make sure that we consume the signal on the semaphore as well 4364 sr_semaphore.wait(); 4365 break; 4366 } else { 4367 ShouldNotReachHere(); 4368 return false; 4369 } 4370 } 4371 } 4372 4373 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4374 return true; 4375 } 4376 4377 static void do_resume(OSThread* osthread) { 4378 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4379 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4380 4381 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4382 // failed to switch to WAKEUP_REQUEST 4383 ShouldNotReachHere(); 4384 return; 4385 } 4386 4387 while (true) { 4388 if (sr_notify(osthread) == 0) { 4389 if (sr_semaphore.timedwait(2)) { 4390 if (osthread->sr.is_running()) { 4391 return; 4392 } 4393 } 4394 } else { 4395 ShouldNotReachHere(); 4396 } 4397 } 4398 4399 guarantee(osthread->sr.is_running(), "Must be running!"); 4400 } 4401 4402 /////////////////////////////////////////////////////////////////////////////////// 4403 // signal handling (except suspend/resume) 4404 4405 // This routine may be used by user applications as a "hook" to catch signals. 4406 // The user-defined signal handler must pass unrecognized signals to this 4407 // routine, and if it returns true (non-zero), then the signal handler must 4408 // return immediately. If the flag "abort_if_unrecognized" is true, then this 4409 // routine will never retun false (zero), but instead will execute a VM panic 4410 // routine kill the process. 4411 // 4412 // If this routine returns false, it is OK to call it again. This allows 4413 // the user-defined signal handler to perform checks either before or after 4414 // the VM performs its own checks. Naturally, the user code would be making 4415 // a serious error if it tried to handle an exception (such as a null check 4416 // or breakpoint) that the VM was generating for its own correct operation. 4417 // 4418 // This routine may recognize any of the following kinds of signals: 4419 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4420 // It should be consulted by handlers for any of those signals. 4421 // 4422 // The caller of this routine must pass in the three arguments supplied 4423 // to the function referred to in the "sa_sigaction" (not the "sa_handler") 4424 // field of the structure passed to sigaction(). This routine assumes that 4425 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4426 // 4427 // Note that the VM will print warnings if it detects conflicting signal 4428 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4429 // 4430 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4431 siginfo_t* siginfo, 4432 void* ucontext, 4433 int abort_if_unrecognized); 4434 4435 static void signalHandler(int sig, siginfo_t* info, void* uc) { 4436 assert(info != NULL && uc != NULL, "it must be old kernel"); 4437 int orig_errno = errno; // Preserve errno value over signal handler. 4438 JVM_handle_linux_signal(sig, info, uc, true); 4439 errno = orig_errno; 4440 } 4441 4442 4443 // This boolean allows users to forward their own non-matching signals 4444 // to JVM_handle_linux_signal, harmlessly. 4445 bool os::Linux::signal_handlers_are_installed = false; 4446 4447 // For signal-chaining 4448 struct sigaction sigact[NSIG]; 4449 sigset_t sigs; 4450 bool os::Linux::libjsig_is_loaded = false; 4451 typedef struct sigaction *(*get_signal_t)(int); 4452 get_signal_t os::Linux::get_signal_action = NULL; 4453 4454 struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4455 struct sigaction *actp = NULL; 4456 4457 if (libjsig_is_loaded) { 4458 // Retrieve the old signal handler from libjsig 4459 actp = (*get_signal_action)(sig); 4460 } 4461 if (actp == NULL) { 4462 // Retrieve the preinstalled signal handler from jvm 4463 actp = get_preinstalled_handler(sig); 4464 } 4465 4466 return actp; 4467 } 4468 4469 static bool call_chained_handler(struct sigaction *actp, int sig, 4470 siginfo_t *siginfo, void *context) { 4471 // Call the old signal handler 4472 if (actp->sa_handler == SIG_DFL) { 4473 // It's more reasonable to let jvm treat it as an unexpected exception 4474 // instead of taking the default action. 4475 return false; 4476 } else if (actp->sa_handler != SIG_IGN) { 4477 if ((actp->sa_flags & SA_NODEFER) == 0) { 4478 // automaticlly block the signal 4479 sigaddset(&(actp->sa_mask), sig); 4480 } 4481 4482 sa_handler_t hand = NULL; 4483 sa_sigaction_t sa = NULL; 4484 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4485 // retrieve the chained handler 4486 if (siginfo_flag_set) { 4487 sa = actp->sa_sigaction; 4488 } else { 4489 hand = actp->sa_handler; 4490 } 4491 4492 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4493 actp->sa_handler = SIG_DFL; 4494 } 4495 4496 // try to honor the signal mask 4497 sigset_t oset; 4498 sigemptyset(&oset); 4499 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4500 4501 // call into the chained handler 4502 if (siginfo_flag_set) { 4503 (*sa)(sig, siginfo, context); 4504 } else { 4505 (*hand)(sig); 4506 } 4507 4508 // restore the signal mask 4509 pthread_sigmask(SIG_SETMASK, &oset, NULL); 4510 } 4511 // Tell jvm's signal handler the signal is taken care of. 4512 return true; 4513 } 4514 4515 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4516 bool chained = false; 4517 // signal-chaining 4518 if (UseSignalChaining) { 4519 struct sigaction *actp = get_chained_signal_action(sig); 4520 if (actp != NULL) { 4521 chained = call_chained_handler(actp, sig, siginfo, context); 4522 } 4523 } 4524 return chained; 4525 } 4526 4527 struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4528 if (sigismember(&sigs, sig)) { 4529 return &sigact[sig]; 4530 } 4531 return NULL; 4532 } 4533 4534 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4535 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4536 sigact[sig] = oldAct; 4537 sigaddset(&sigs, sig); 4538 } 4539 4540 // for diagnostic 4541 int sigflags[NSIG]; 4542 4543 int os::Linux::get_our_sigflags(int sig) { 4544 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4545 return sigflags[sig]; 4546 } 4547 4548 void os::Linux::set_our_sigflags(int sig, int flags) { 4549 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4550 if (sig > 0 && sig < NSIG) { 4551 sigflags[sig] = flags; 4552 } 4553 } 4554 4555 void os::Linux::set_signal_handler(int sig, bool set_installed) { 4556 // Check for overwrite. 4557 struct sigaction oldAct; 4558 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4559 4560 void* oldhand = oldAct.sa_sigaction 4561 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4562 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4563 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4564 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4565 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4566 if (AllowUserSignalHandlers || !set_installed) { 4567 // Do not overwrite; user takes responsibility to forward to us. 4568 return; 4569 } else if (UseSignalChaining) { 4570 // save the old handler in jvm 4571 save_preinstalled_handler(sig, oldAct); 4572 // libjsig also interposes the sigaction() call below and saves the 4573 // old sigaction on it own. 4574 } else { 4575 fatal("Encountered unexpected pre-existing sigaction handler " 4576 "%#lx for signal %d.", (long)oldhand, sig); 4577 } 4578 } 4579 4580 struct sigaction sigAct; 4581 sigfillset(&(sigAct.sa_mask)); 4582 sigAct.sa_handler = SIG_DFL; 4583 if (!set_installed) { 4584 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4585 } else { 4586 sigAct.sa_sigaction = signalHandler; 4587 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4588 } 4589 // Save flags, which are set by ours 4590 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4591 sigflags[sig] = sigAct.sa_flags; 4592 4593 int ret = sigaction(sig, &sigAct, &oldAct); 4594 assert(ret == 0, "check"); 4595 4596 void* oldhand2 = oldAct.sa_sigaction 4597 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4598 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4599 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4600 } 4601 4602 // install signal handlers for signals that HotSpot needs to 4603 // handle in order to support Java-level exception handling. 4604 4605 void os::Linux::install_signal_handlers() { 4606 if (!signal_handlers_are_installed) { 4607 signal_handlers_are_installed = true; 4608 4609 // signal-chaining 4610 typedef void (*signal_setting_t)(); 4611 signal_setting_t begin_signal_setting = NULL; 4612 signal_setting_t end_signal_setting = NULL; 4613 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4614 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4615 if (begin_signal_setting != NULL) { 4616 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4617 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4618 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4619 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4620 libjsig_is_loaded = true; 4621 assert(UseSignalChaining, "should enable signal-chaining"); 4622 } 4623 if (libjsig_is_loaded) { 4624 // Tell libjsig jvm is setting signal handlers 4625 (*begin_signal_setting)(); 4626 } 4627 4628 ::sigemptyset(&sigs); 4629 set_signal_handler(SIGSEGV, true); 4630 set_signal_handler(SIGPIPE, true); 4631 set_signal_handler(SIGBUS, true); 4632 set_signal_handler(SIGILL, true); 4633 set_signal_handler(SIGFPE, true); 4634 #if defined(PPC64) 4635 set_signal_handler(SIGTRAP, true); 4636 #endif 4637 set_signal_handler(SIGXFSZ, true); 4638 4639 if (libjsig_is_loaded) { 4640 // Tell libjsig jvm finishes setting signal handlers 4641 (*end_signal_setting)(); 4642 } 4643 4644 // We don't activate signal checker if libjsig is in place, we trust ourselves 4645 // and if UserSignalHandler is installed all bets are off. 4646 // Log that signal checking is off only if -verbose:jni is specified. 4647 if (CheckJNICalls) { 4648 if (libjsig_is_loaded) { 4649 if (PrintJNIResolving) { 4650 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4651 } 4652 check_signals = false; 4653 } 4654 if (AllowUserSignalHandlers) { 4655 if (PrintJNIResolving) { 4656 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4657 } 4658 check_signals = false; 4659 } 4660 } 4661 } 4662 } 4663 4664 // This is the fastest way to get thread cpu time on Linux. 4665 // Returns cpu time (user+sys) for any thread, not only for current. 4666 // POSIX compliant clocks are implemented in the kernels 2.6.16+. 4667 // It might work on 2.6.10+ with a special kernel/glibc patch. 4668 // For reference, please, see IEEE Std 1003.1-2004: 4669 // http://www.unix.org/single_unix_specification 4670 4671 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4672 struct timespec tp; 4673 int rc = os::Posix::clock_gettime(clockid, &tp); 4674 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4675 4676 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4677 } 4678 4679 void os::Linux::initialize_os_info() { 4680 assert(_os_version == 0, "OS info already initialized"); 4681 4682 struct utsname _uname; 4683 4684 uint32_t major; 4685 uint32_t minor; 4686 uint32_t fix; 4687 4688 int rc; 4689 4690 // Kernel version is unknown if 4691 // verification below fails. 4692 _os_version = 0x01000000; 4693 4694 rc = uname(&_uname); 4695 if (rc != -1) { 4696 4697 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix); 4698 if (rc == 3) { 4699 4700 if (major < 256 && minor < 256 && fix < 256) { 4701 // Kernel version format is as expected, 4702 // set it overriding unknown state. 4703 _os_version = (major << 16) | 4704 (minor << 8 ) | 4705 (fix << 0 ) ; 4706 } 4707 } 4708 } 4709 } 4710 4711 uint32_t os::Linux::os_version() { 4712 assert(_os_version != 0, "not initialized"); 4713 return _os_version & 0x00FFFFFF; 4714 } 4715 4716 bool os::Linux::os_version_is_known() { 4717 assert(_os_version != 0, "not initialized"); 4718 return _os_version & 0x01000000 ? false : true; 4719 } 4720 4721 ///// 4722 // glibc on Linux platform uses non-documented flag 4723 // to indicate, that some special sort of signal 4724 // trampoline is used. 4725 // We will never set this flag, and we should 4726 // ignore this flag in our diagnostic 4727 #ifdef SIGNIFICANT_SIGNAL_MASK 4728 #undef SIGNIFICANT_SIGNAL_MASK 4729 #endif 4730 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4731 4732 static const char* get_signal_handler_name(address handler, 4733 char* buf, int buflen) { 4734 int offset = 0; 4735 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4736 if (found) { 4737 // skip directory names 4738 const char *p1, *p2; 4739 p1 = buf; 4740 size_t len = strlen(os::file_separator()); 4741 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4742 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4743 } else { 4744 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4745 } 4746 return buf; 4747 } 4748 4749 static void print_signal_handler(outputStream* st, int sig, 4750 char* buf, size_t buflen) { 4751 struct sigaction sa; 4752 4753 sigaction(sig, NULL, &sa); 4754 4755 // See comment for SIGNIFICANT_SIGNAL_MASK define 4756 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4757 4758 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4759 4760 address handler = (sa.sa_flags & SA_SIGINFO) 4761 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4762 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4763 4764 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4765 st->print("SIG_DFL"); 4766 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4767 st->print("SIG_IGN"); 4768 } else { 4769 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4770 } 4771 4772 st->print(", sa_mask[0]="); 4773 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4774 4775 address rh = VMError::get_resetted_sighandler(sig); 4776 // May be, handler was resetted by VMError? 4777 if (rh != NULL) { 4778 handler = rh; 4779 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4780 } 4781 4782 st->print(", sa_flags="); 4783 os::Posix::print_sa_flags(st, sa.sa_flags); 4784 4785 // Check: is it our handler? 4786 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4787 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4788 // It is our signal handler 4789 // check for flags, reset system-used one! 4790 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4791 st->print( 4792 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4793 os::Linux::get_our_sigflags(sig)); 4794 } 4795 } 4796 st->cr(); 4797 } 4798 4799 4800 #define DO_SIGNAL_CHECK(sig) \ 4801 do { \ 4802 if (!sigismember(&check_signal_done, sig)) { \ 4803 os::Linux::check_signal_handler(sig); \ 4804 } \ 4805 } while (0) 4806 4807 // This method is a periodic task to check for misbehaving JNI applications 4808 // under CheckJNI, we can add any periodic checks here 4809 4810 void os::run_periodic_checks() { 4811 if (check_signals == false) return; 4812 4813 // SEGV and BUS if overridden could potentially prevent 4814 // generation of hs*.log in the event of a crash, debugging 4815 // such a case can be very challenging, so we absolutely 4816 // check the following for a good measure: 4817 DO_SIGNAL_CHECK(SIGSEGV); 4818 DO_SIGNAL_CHECK(SIGILL); 4819 DO_SIGNAL_CHECK(SIGFPE); 4820 DO_SIGNAL_CHECK(SIGBUS); 4821 DO_SIGNAL_CHECK(SIGPIPE); 4822 DO_SIGNAL_CHECK(SIGXFSZ); 4823 #if defined(PPC64) 4824 DO_SIGNAL_CHECK(SIGTRAP); 4825 #endif 4826 4827 // ReduceSignalUsage allows the user to override these handlers 4828 // see comments at the very top and jvm_md.h 4829 if (!ReduceSignalUsage) { 4830 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4831 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4832 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4833 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4834 } 4835 4836 DO_SIGNAL_CHECK(SR_signum); 4837 } 4838 4839 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4840 4841 static os_sigaction_t os_sigaction = NULL; 4842 4843 void os::Linux::check_signal_handler(int sig) { 4844 char buf[O_BUFLEN]; 4845 address jvmHandler = NULL; 4846 4847 4848 struct sigaction act; 4849 if (os_sigaction == NULL) { 4850 // only trust the default sigaction, in case it has been interposed 4851 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4852 if (os_sigaction == NULL) return; 4853 } 4854 4855 os_sigaction(sig, (struct sigaction*)NULL, &act); 4856 4857 4858 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4859 4860 address thisHandler = (act.sa_flags & SA_SIGINFO) 4861 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4862 : CAST_FROM_FN_PTR(address, act.sa_handler); 4863 4864 4865 switch (sig) { 4866 case SIGSEGV: 4867 case SIGBUS: 4868 case SIGFPE: 4869 case SIGPIPE: 4870 case SIGILL: 4871 case SIGXFSZ: 4872 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4873 break; 4874 4875 case SHUTDOWN1_SIGNAL: 4876 case SHUTDOWN2_SIGNAL: 4877 case SHUTDOWN3_SIGNAL: 4878 case BREAK_SIGNAL: 4879 jvmHandler = (address)user_handler(); 4880 break; 4881 4882 default: 4883 if (sig == SR_signum) { 4884 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4885 } else { 4886 return; 4887 } 4888 break; 4889 } 4890 4891 if (thisHandler != jvmHandler) { 4892 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4893 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4894 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4895 // No need to check this sig any longer 4896 sigaddset(&check_signal_done, sig); 4897 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4898 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4899 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4900 exception_name(sig, buf, O_BUFLEN)); 4901 } 4902 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4903 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4904 tty->print("expected:"); 4905 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig)); 4906 tty->cr(); 4907 tty->print(" found:"); 4908 os::Posix::print_sa_flags(tty, act.sa_flags); 4909 tty->cr(); 4910 // No need to check this sig any longer 4911 sigaddset(&check_signal_done, sig); 4912 } 4913 4914 // Dump all the signal 4915 if (sigismember(&check_signal_done, sig)) { 4916 print_signal_handlers(tty, buf, O_BUFLEN); 4917 } 4918 } 4919 4920 extern void report_error(char* file_name, int line_no, char* title, 4921 char* format, ...); 4922 4923 // this is called _before_ most of the global arguments have been parsed 4924 void os::init(void) { 4925 char dummy; // used to get a guess on initial stack address 4926 4927 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4928 4929 init_random(1234567); 4930 4931 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4932 if (Linux::page_size() == -1) { 4933 fatal("os_linux.cpp: os::init: sysconf failed (%s)", 4934 os::strerror(errno)); 4935 } 4936 init_page_sizes((size_t) Linux::page_size()); 4937 4938 Linux::initialize_system_info(); 4939 4940 Linux::initialize_os_info(); 4941 4942 // _main_thread points to the thread that created/loaded the JVM. 4943 Linux::_main_thread = pthread_self(); 4944 4945 // retrieve entry point for pthread_setname_np 4946 Linux::_pthread_setname_np = 4947 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4948 4949 os::Posix::init(); 4950 4951 initial_time_count = javaTimeNanos(); 4952 4953 // Always warn if no monotonic clock available 4954 if (!os::Posix::supports_monotonic_clock()) { 4955 warning("No monotonic clock was available - timed services may " \ 4956 "be adversely affected if the time-of-day clock changes"); 4957 } 4958 } 4959 4960 // To install functions for atexit system call 4961 extern "C" { 4962 static void perfMemory_exit_helper() { 4963 perfMemory_exit(); 4964 } 4965 } 4966 4967 void os::pd_init_container_support() { 4968 OSContainer::init(); 4969 } 4970 4971 void os::Linux::numa_init() { 4972 4973 // Java can be invoked as 4974 // 1. Without numactl and heap will be allocated/configured on all nodes as 4975 // per the system policy. 4976 // 2. With numactl --interleave: 4977 // Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same 4978 // API for membind case bitmask is reset. 4979 // Interleave is only hint and Kernel can fallback to other nodes if 4980 // no memory is available on the target nodes. 4981 // 3. With numactl --membind: 4982 // Use numa_get_membind(v2) API to get nodes bitmask. The same API for 4983 // interleave case returns bitmask of all nodes. 4984 // numa_all_nodes_ptr holds bitmask of all nodes. 4985 // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct 4986 // bitmask when externally configured to run on all or fewer nodes. 4987 4988 if (!Linux::libnuma_init()) { 4989 UseNUMA = false; 4990 } else { 4991 if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) { 4992 // If there's only one node (they start from 0) or if the process 4993 // is bound explicitly to a single node using membind, disable NUMA. 4994 UseNUMA = false; 4995 } else { 4996 4997 LogTarget(Info,os) log; 4998 LogStream ls(log); 4999 5000 Linux::set_configured_numa_policy(Linux::identify_numa_policy()); 5001 5002 struct bitmask* bmp = Linux::_numa_membind_bitmask; 5003 const char* numa_mode = "membind"; 5004 5005 if (Linux::is_running_in_interleave_mode()) { 5006 bmp = Linux::_numa_interleave_bitmask; 5007 numa_mode = "interleave"; 5008 } 5009 5010 ls.print("UseNUMA is enabled and invoked in '%s' mode." 5011 " Heap will be configured using NUMA memory nodes:", numa_mode); 5012 5013 for (int node = 0; node <= Linux::numa_max_node(); node++) { 5014 if (Linux::_numa_bitmask_isbitset(bmp, node)) { 5015 ls.print(" %d", node); 5016 } 5017 } 5018 } 5019 } 5020 5021 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 5022 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 5023 // we can make the adaptive lgrp chunk resizing work. If the user specified both 5024 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn 5025 // and disable adaptive resizing. 5026 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) { 5027 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, " 5028 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)"); 5029 UseAdaptiveSizePolicy = false; 5030 UseAdaptiveNUMAChunkSizing = false; 5031 } 5032 } 5033 5034 if (!UseNUMA && ForceNUMA) { 5035 UseNUMA = true; 5036 } 5037 } 5038 5039 // this is called _after_ the global arguments have been parsed 5040 jint os::init_2(void) { 5041 5042 // This could be set after os::Posix::init() but all platforms 5043 // have to set it the same so we have to mirror Solaris. 5044 DEBUG_ONLY(os::set_mutex_init_done();) 5045 5046 os::Posix::init_2(); 5047 5048 Linux::fast_thread_clock_init(); 5049 5050 // initialize suspend/resume support - must do this before signal_sets_init() 5051 if (SR_initialize() != 0) { 5052 perror("SR_initialize failed"); 5053 return JNI_ERR; 5054 } 5055 5056 Linux::signal_sets_init(); 5057 Linux::install_signal_handlers(); 5058 // Initialize data for jdk.internal.misc.Signal 5059 if (!ReduceSignalUsage) { 5060 jdk_misc_signal_init(); 5061 } 5062 5063 // Check and sets minimum stack sizes against command line options 5064 if (Posix::set_minimum_stack_sizes() == JNI_ERR) { 5065 return JNI_ERR; 5066 } 5067 5068 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher(); 5069 if (!suppress_primordial_thread_resolution) { 5070 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 5071 } 5072 5073 #if defined(IA32) 5074 workaround_expand_exec_shield_cs_limit(); 5075 #endif 5076 5077 Linux::libpthread_init(); 5078 Linux::sched_getcpu_init(); 5079 log_info(os)("HotSpot is running with %s, %s", 5080 Linux::glibc_version(), Linux::libpthread_version()); 5081 5082 if (UseNUMA) { 5083 Linux::numa_init(); 5084 } 5085 5086 if (MaxFDLimit) { 5087 // set the number of file descriptors to max. print out error 5088 // if getrlimit/setrlimit fails but continue regardless. 5089 struct rlimit nbr_files; 5090 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 5091 if (status != 0) { 5092 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno)); 5093 } else { 5094 nbr_files.rlim_cur = nbr_files.rlim_max; 5095 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 5096 if (status != 0) { 5097 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno)); 5098 } 5099 } 5100 } 5101 5102 // Initialize lock used to serialize thread creation (see os::create_thread) 5103 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 5104 5105 // at-exit methods are called in the reverse order of their registration. 5106 // atexit functions are called on return from main or as a result of a 5107 // call to exit(3C). There can be only 32 of these functions registered 5108 // and atexit() does not set errno. 5109 5110 if (PerfAllowAtExitRegistration) { 5111 // only register atexit functions if PerfAllowAtExitRegistration is set. 5112 // atexit functions can be delayed until process exit time, which 5113 // can be problematic for embedded VM situations. Embedded VMs should 5114 // call DestroyJavaVM() to assure that VM resources are released. 5115 5116 // note: perfMemory_exit_helper atexit function may be removed in 5117 // the future if the appropriate cleanup code can be added to the 5118 // VM_Exit VMOperation's doit method. 5119 if (atexit(perfMemory_exit_helper) != 0) { 5120 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 5121 } 5122 } 5123 5124 // initialize thread priority policy 5125 prio_init(); 5126 5127 if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) { 5128 set_coredump_filter(DAX_SHARED_BIT); 5129 } 5130 5131 if (DumpPrivateMappingsInCore) { 5132 set_coredump_filter(FILE_BACKED_PVT_BIT); 5133 } 5134 5135 if (DumpSharedMappingsInCore) { 5136 set_coredump_filter(FILE_BACKED_SHARED_BIT); 5137 } 5138 5139 return JNI_OK; 5140 } 5141 5142 // Mark the polling page as unreadable 5143 void os::make_polling_page_unreadable(void) { 5144 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 5145 fatal("Could not disable polling page"); 5146 } 5147 } 5148 5149 // Mark the polling page as readable 5150 void os::make_polling_page_readable(void) { 5151 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 5152 fatal("Could not enable polling page"); 5153 } 5154 } 5155 5156 // older glibc versions don't have this macro (which expands to 5157 // an optimized bit-counting function) so we have to roll our own 5158 #ifndef CPU_COUNT 5159 5160 static int _cpu_count(const cpu_set_t* cpus) { 5161 int count = 0; 5162 // only look up to the number of configured processors 5163 for (int i = 0; i < os::processor_count(); i++) { 5164 if (CPU_ISSET(i, cpus)) { 5165 count++; 5166 } 5167 } 5168 return count; 5169 } 5170 5171 #define CPU_COUNT(cpus) _cpu_count(cpus) 5172 5173 #endif // CPU_COUNT 5174 5175 // Get the current number of available processors for this process. 5176 // This value can change at any time during a process's lifetime. 5177 // sched_getaffinity gives an accurate answer as it accounts for cpusets. 5178 // If it appears there may be more than 1024 processors then we do a 5179 // dynamic check - see 6515172 for details. 5180 // If anything goes wrong we fallback to returning the number of online 5181 // processors - which can be greater than the number available to the process. 5182 int os::Linux::active_processor_count() { 5183 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors 5184 cpu_set_t* cpus_p = &cpus; 5185 int cpus_size = sizeof(cpu_set_t); 5186 5187 int configured_cpus = os::processor_count(); // upper bound on available cpus 5188 int cpu_count = 0; 5189 5190 // old build platforms may not support dynamic cpu sets 5191 #ifdef CPU_ALLOC 5192 5193 // To enable easy testing of the dynamic path on different platforms we 5194 // introduce a diagnostic flag: UseCpuAllocPath 5195 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) { 5196 // kernel may use a mask bigger than cpu_set_t 5197 log_trace(os)("active_processor_count: using dynamic path %s" 5198 "- configured processors: %d", 5199 UseCpuAllocPath ? "(forced) " : "", 5200 configured_cpus); 5201 cpus_p = CPU_ALLOC(configured_cpus); 5202 if (cpus_p != NULL) { 5203 cpus_size = CPU_ALLOC_SIZE(configured_cpus); 5204 // zero it just to be safe 5205 CPU_ZERO_S(cpus_size, cpus_p); 5206 } 5207 else { 5208 // failed to allocate so fallback to online cpus 5209 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 5210 log_trace(os)("active_processor_count: " 5211 "CPU_ALLOC failed (%s) - using " 5212 "online processor count: %d", 5213 os::strerror(errno), online_cpus); 5214 return online_cpus; 5215 } 5216 } 5217 else { 5218 log_trace(os)("active_processor_count: using static path - configured processors: %d", 5219 configured_cpus); 5220 } 5221 #else // CPU_ALLOC 5222 // these stubs won't be executed 5223 #define CPU_COUNT_S(size, cpus) -1 5224 #define CPU_FREE(cpus) 5225 5226 log_trace(os)("active_processor_count: only static path available - configured processors: %d", 5227 configured_cpus); 5228 #endif // CPU_ALLOC 5229 5230 // pid 0 means the current thread - which we have to assume represents the process 5231 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) { 5232 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5233 cpu_count = CPU_COUNT_S(cpus_size, cpus_p); 5234 } 5235 else { 5236 cpu_count = CPU_COUNT(cpus_p); 5237 } 5238 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count); 5239 } 5240 else { 5241 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN); 5242 warning("sched_getaffinity failed (%s)- using online processor count (%d) " 5243 "which may exceed available processors", os::strerror(errno), cpu_count); 5244 } 5245 5246 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5247 CPU_FREE(cpus_p); 5248 } 5249 5250 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check"); 5251 return cpu_count; 5252 } 5253 5254 // Determine the active processor count from one of 5255 // three different sources: 5256 // 5257 // 1. User option -XX:ActiveProcessorCount 5258 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN) 5259 // 3. extracted from cgroup cpu subsystem (shares and quotas) 5260 // 5261 // Option 1, if specified, will always override. 5262 // If the cgroup subsystem is active and configured, we 5263 // will return the min of the cgroup and option 2 results. 5264 // This is required since tools, such as numactl, that 5265 // alter cpu affinity do not update cgroup subsystem 5266 // cpuset configuration files. 5267 int os::active_processor_count() { 5268 // User has overridden the number of active processors 5269 if (ActiveProcessorCount > 0) { 5270 log_trace(os)("active_processor_count: " 5271 "active processor count set by user : %d", 5272 ActiveProcessorCount); 5273 return ActiveProcessorCount; 5274 } 5275 5276 int active_cpus; 5277 if (OSContainer::is_containerized()) { 5278 active_cpus = OSContainer::active_processor_count(); 5279 log_trace(os)("active_processor_count: determined by OSContainer: %d", 5280 active_cpus); 5281 } else { 5282 active_cpus = os::Linux::active_processor_count(); 5283 } 5284 5285 return active_cpus; 5286 } 5287 5288 uint os::processor_id() { 5289 const int id = Linux::sched_getcpu(); 5290 assert(id >= 0 && id < _processor_count, "Invalid processor id"); 5291 return (uint)id; 5292 } 5293 5294 void os::set_native_thread_name(const char *name) { 5295 if (Linux::_pthread_setname_np) { 5296 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 5297 snprintf(buf, sizeof(buf), "%s", name); 5298 buf[sizeof(buf) - 1] = '\0'; 5299 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 5300 // ERANGE should not happen; all other errors should just be ignored. 5301 assert(rc != ERANGE, "pthread_setname_np failed"); 5302 } 5303 } 5304 5305 bool os::distribute_processes(uint length, uint* distribution) { 5306 // Not yet implemented. 5307 return false; 5308 } 5309 5310 bool os::bind_to_processor(uint processor_id) { 5311 // Not yet implemented. 5312 return false; 5313 } 5314 5315 /// 5316 5317 void os::SuspendedThreadTask::internal_do_task() { 5318 if (do_suspend(_thread->osthread())) { 5319 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 5320 do_task(context); 5321 do_resume(_thread->osthread()); 5322 } 5323 } 5324 5325 //////////////////////////////////////////////////////////////////////////////// 5326 // debug support 5327 5328 bool os::find(address addr, outputStream* st) { 5329 Dl_info dlinfo; 5330 memset(&dlinfo, 0, sizeof(dlinfo)); 5331 if (dladdr(addr, &dlinfo) != 0) { 5332 st->print(PTR_FORMAT ": ", p2i(addr)); 5333 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5334 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname, 5335 p2i(addr) - p2i(dlinfo.dli_saddr)); 5336 } else if (dlinfo.dli_fbase != NULL) { 5337 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase)); 5338 } else { 5339 st->print("<absolute address>"); 5340 } 5341 if (dlinfo.dli_fname != NULL) { 5342 st->print(" in %s", dlinfo.dli_fname); 5343 } 5344 if (dlinfo.dli_fbase != NULL) { 5345 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase)); 5346 } 5347 st->cr(); 5348 5349 if (Verbose) { 5350 // decode some bytes around the PC 5351 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5352 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5353 address lowest = (address) dlinfo.dli_sname; 5354 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5355 if (begin < lowest) begin = lowest; 5356 Dl_info dlinfo2; 5357 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5358 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 5359 end = (address) dlinfo2.dli_saddr; 5360 } 5361 Disassembler::decode(begin, end, st); 5362 } 5363 return true; 5364 } 5365 return false; 5366 } 5367 5368 //////////////////////////////////////////////////////////////////////////////// 5369 // misc 5370 5371 // This does not do anything on Linux. This is basically a hook for being 5372 // able to use structured exception handling (thread-local exception filters) 5373 // on, e.g., Win32. 5374 void 5375 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method, 5376 JavaCallArguments* args, Thread* thread) { 5377 f(value, method, args, thread); 5378 } 5379 5380 void os::print_statistics() { 5381 } 5382 5383 bool os::message_box(const char* title, const char* message) { 5384 int i; 5385 fdStream err(defaultStream::error_fd()); 5386 for (i = 0; i < 78; i++) err.print_raw("="); 5387 err.cr(); 5388 err.print_raw_cr(title); 5389 for (i = 0; i < 78; i++) err.print_raw("-"); 5390 err.cr(); 5391 err.print_raw_cr(message); 5392 for (i = 0; i < 78; i++) err.print_raw("="); 5393 err.cr(); 5394 5395 char buf[16]; 5396 // Prevent process from exiting upon "read error" without consuming all CPU 5397 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5398 5399 return buf[0] == 'y' || buf[0] == 'Y'; 5400 } 5401 5402 // Is a (classpath) directory empty? 5403 bool os::dir_is_empty(const char* path) { 5404 DIR *dir = NULL; 5405 struct dirent *ptr; 5406 5407 dir = opendir(path); 5408 if (dir == NULL) return true; 5409 5410 // Scan the directory 5411 bool result = true; 5412 while (result && (ptr = readdir(dir)) != NULL) { 5413 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5414 result = false; 5415 } 5416 } 5417 closedir(dir); 5418 return result; 5419 } 5420 5421 // This code originates from JDK's sysOpen and open64_w 5422 // from src/solaris/hpi/src/system_md.c 5423 5424 int os::open(const char *path, int oflag, int mode) { 5425 if (strlen(path) > MAX_PATH - 1) { 5426 errno = ENAMETOOLONG; 5427 return -1; 5428 } 5429 5430 // All file descriptors that are opened in the Java process and not 5431 // specifically destined for a subprocess should have the close-on-exec 5432 // flag set. If we don't set it, then careless 3rd party native code 5433 // might fork and exec without closing all appropriate file descriptors 5434 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5435 // turn might: 5436 // 5437 // - cause end-of-file to fail to be detected on some file 5438 // descriptors, resulting in mysterious hangs, or 5439 // 5440 // - might cause an fopen in the subprocess to fail on a system 5441 // suffering from bug 1085341. 5442 // 5443 // (Yes, the default setting of the close-on-exec flag is a Unix 5444 // design flaw) 5445 // 5446 // See: 5447 // 1085341: 32-bit stdio routines should support file descriptors >255 5448 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5449 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5450 // 5451 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5452 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5453 // because it saves a system call and removes a small window where the flag 5454 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5455 // and we fall back to using FD_CLOEXEC (see below). 5456 #ifdef O_CLOEXEC 5457 oflag |= O_CLOEXEC; 5458 #endif 5459 5460 int fd = ::open64(path, oflag, mode); 5461 if (fd == -1) return -1; 5462 5463 //If the open succeeded, the file might still be a directory 5464 { 5465 struct stat64 buf64; 5466 int ret = ::fstat64(fd, &buf64); 5467 int st_mode = buf64.st_mode; 5468 5469 if (ret != -1) { 5470 if ((st_mode & S_IFMT) == S_IFDIR) { 5471 errno = EISDIR; 5472 ::close(fd); 5473 return -1; 5474 } 5475 } else { 5476 ::close(fd); 5477 return -1; 5478 } 5479 } 5480 5481 #ifdef FD_CLOEXEC 5482 // Validate that the use of the O_CLOEXEC flag on open above worked. 5483 // With recent kernels, we will perform this check exactly once. 5484 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5485 if (!O_CLOEXEC_is_known_to_work) { 5486 int flags = ::fcntl(fd, F_GETFD); 5487 if (flags != -1) { 5488 if ((flags & FD_CLOEXEC) != 0) 5489 O_CLOEXEC_is_known_to_work = 1; 5490 else 5491 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5492 } 5493 } 5494 #endif 5495 5496 return fd; 5497 } 5498 5499 5500 // create binary file, rewriting existing file if required 5501 int os::create_binary_file(const char* path, bool rewrite_existing) { 5502 int oflags = O_WRONLY | O_CREAT; 5503 if (!rewrite_existing) { 5504 oflags |= O_EXCL; 5505 } 5506 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5507 } 5508 5509 // return current position of file pointer 5510 jlong os::current_file_offset(int fd) { 5511 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5512 } 5513 5514 // move file pointer to the specified offset 5515 jlong os::seek_to_file_offset(int fd, jlong offset) { 5516 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5517 } 5518 5519 // This code originates from JDK's sysAvailable 5520 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5521 5522 int os::available(int fd, jlong *bytes) { 5523 jlong cur, end; 5524 int mode; 5525 struct stat64 buf64; 5526 5527 if (::fstat64(fd, &buf64) >= 0) { 5528 mode = buf64.st_mode; 5529 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5530 int n; 5531 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5532 *bytes = n; 5533 return 1; 5534 } 5535 } 5536 } 5537 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5538 return 0; 5539 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5540 return 0; 5541 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5542 return 0; 5543 } 5544 *bytes = end - cur; 5545 return 1; 5546 } 5547 5548 // Map a block of memory. 5549 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5550 char *addr, size_t bytes, bool read_only, 5551 bool allow_exec) { 5552 int prot; 5553 int flags = MAP_PRIVATE; 5554 5555 if (read_only) { 5556 prot = PROT_READ; 5557 } else { 5558 prot = PROT_READ | PROT_WRITE; 5559 } 5560 5561 if (allow_exec) { 5562 prot |= PROT_EXEC; 5563 } 5564 5565 if (addr != NULL) { 5566 flags |= MAP_FIXED; 5567 } 5568 5569 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5570 fd, file_offset); 5571 if (mapped_address == MAP_FAILED) { 5572 return NULL; 5573 } 5574 return mapped_address; 5575 } 5576 5577 5578 // Remap a block of memory. 5579 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5580 char *addr, size_t bytes, bool read_only, 5581 bool allow_exec) { 5582 // same as map_memory() on this OS 5583 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5584 allow_exec); 5585 } 5586 5587 5588 // Unmap a block of memory. 5589 bool os::pd_unmap_memory(char* addr, size_t bytes) { 5590 return munmap(addr, bytes) == 0; 5591 } 5592 5593 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5594 5595 static jlong fast_cpu_time(Thread *thread) { 5596 clockid_t clockid; 5597 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(), 5598 &clockid); 5599 if (rc == 0) { 5600 return os::Linux::fast_thread_cpu_time(clockid); 5601 } else { 5602 // It's possible to encounter a terminated native thread that failed 5603 // to detach itself from the VM - which should result in ESRCH. 5604 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed"); 5605 return -1; 5606 } 5607 } 5608 5609 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5610 // are used by JVM M&M and JVMTI to get user+sys or user CPU time 5611 // of a thread. 5612 // 5613 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5614 // the fast estimate available on the platform. 5615 5616 jlong os::current_thread_cpu_time() { 5617 if (os::Linux::supports_fast_thread_cpu_time()) { 5618 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5619 } else { 5620 // return user + sys since the cost is the same 5621 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5622 } 5623 } 5624 5625 jlong os::thread_cpu_time(Thread* thread) { 5626 // consistent with what current_thread_cpu_time() returns 5627 if (os::Linux::supports_fast_thread_cpu_time()) { 5628 return fast_cpu_time(thread); 5629 } else { 5630 return slow_thread_cpu_time(thread, true /* user + sys */); 5631 } 5632 } 5633 5634 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5635 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5636 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5637 } else { 5638 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5639 } 5640 } 5641 5642 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5643 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5644 return fast_cpu_time(thread); 5645 } else { 5646 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5647 } 5648 } 5649 5650 // -1 on error. 5651 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5652 pid_t tid = thread->osthread()->thread_id(); 5653 char *s; 5654 char stat[2048]; 5655 int statlen; 5656 char proc_name[64]; 5657 int count; 5658 long sys_time, user_time; 5659 char cdummy; 5660 int idummy; 5661 long ldummy; 5662 FILE *fp; 5663 5664 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5665 fp = fopen(proc_name, "r"); 5666 if (fp == NULL) return -1; 5667 statlen = fread(stat, 1, 2047, fp); 5668 stat[statlen] = '\0'; 5669 fclose(fp); 5670 5671 // Skip pid and the command string. Note that we could be dealing with 5672 // weird command names, e.g. user could decide to rename java launcher 5673 // to "java 1.4.2 :)", then the stat file would look like 5674 // 1234 (java 1.4.2 :)) R ... ... 5675 // We don't really need to know the command string, just find the last 5676 // occurrence of ")" and then start parsing from there. See bug 4726580. 5677 s = strrchr(stat, ')'); 5678 if (s == NULL) return -1; 5679 5680 // Skip blank chars 5681 do { s++; } while (s && isspace(*s)); 5682 5683 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5684 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5685 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5686 &user_time, &sys_time); 5687 if (count != 13) return -1; 5688 if (user_sys_cpu_time) { 5689 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5690 } else { 5691 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5692 } 5693 } 5694 5695 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5696 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5697 info_ptr->may_skip_backward = false; // elapsed time not wall time 5698 info_ptr->may_skip_forward = false; // elapsed time not wall time 5699 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5700 } 5701 5702 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5703 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5704 info_ptr->may_skip_backward = false; // elapsed time not wall time 5705 info_ptr->may_skip_forward = false; // elapsed time not wall time 5706 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5707 } 5708 5709 bool os::is_thread_cpu_time_supported() { 5710 return true; 5711 } 5712 5713 // System loadavg support. Returns -1 if load average cannot be obtained. 5714 // Linux doesn't yet have a (official) notion of processor sets, 5715 // so just return the system wide load average. 5716 int os::loadavg(double loadavg[], int nelem) { 5717 return ::getloadavg(loadavg, nelem); 5718 } 5719 5720 void os::pause() { 5721 char filename[MAX_PATH]; 5722 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5723 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile); 5724 } else { 5725 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5726 } 5727 5728 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5729 if (fd != -1) { 5730 struct stat buf; 5731 ::close(fd); 5732 while (::stat(filename, &buf) == 0) { 5733 (void)::poll(NULL, 0, 100); 5734 } 5735 } else { 5736 jio_fprintf(stderr, 5737 "Could not open pause file '%s', continuing immediately.\n", filename); 5738 } 5739 } 5740 5741 extern char** environ; 5742 5743 // Run the specified command in a separate process. Return its exit value, 5744 // or -1 on failure (e.g. can't fork a new process). 5745 // Unlike system(), this function can be called from signal handler. It 5746 // doesn't block SIGINT et al. 5747 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) { 5748 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5749 5750 pid_t pid ; 5751 5752 if (use_vfork_if_available) { 5753 pid = vfork(); 5754 } else { 5755 pid = fork(); 5756 } 5757 5758 if (pid < 0) { 5759 // fork failed 5760 return -1; 5761 5762 } else if (pid == 0) { 5763 // child process 5764 5765 execve("/bin/sh", (char* const*)argv, environ); 5766 5767 // execve failed 5768 _exit(-1); 5769 5770 } else { 5771 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5772 // care about the actual exit code, for now. 5773 5774 int status; 5775 5776 // Wait for the child process to exit. This returns immediately if 5777 // the child has already exited. */ 5778 while (waitpid(pid, &status, 0) < 0) { 5779 switch (errno) { 5780 case ECHILD: return 0; 5781 case EINTR: break; 5782 default: return -1; 5783 } 5784 } 5785 5786 if (WIFEXITED(status)) { 5787 // The child exited normally; get its exit code. 5788 return WEXITSTATUS(status); 5789 } else if (WIFSIGNALED(status)) { 5790 // The child exited because of a signal 5791 // The best value to return is 0x80 + signal number, 5792 // because that is what all Unix shells do, and because 5793 // it allows callers to distinguish between process exit and 5794 // process death by signal. 5795 return 0x80 + WTERMSIG(status); 5796 } else { 5797 // Unknown exit code; pass it through 5798 return status; 5799 } 5800 } 5801 } 5802 5803 // Get the default path to the core file 5804 // Returns the length of the string 5805 int os::get_core_path(char* buffer, size_t bufferSize) { 5806 /* 5807 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 5808 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 5809 */ 5810 const int core_pattern_len = 129; 5811 char core_pattern[core_pattern_len] = {0}; 5812 5813 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 5814 if (core_pattern_file == -1) { 5815 return -1; 5816 } 5817 5818 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 5819 ::close(core_pattern_file); 5820 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') { 5821 return -1; 5822 } 5823 if (core_pattern[ret-1] == '\n') { 5824 core_pattern[ret-1] = '\0'; 5825 } else { 5826 core_pattern[ret] = '\0'; 5827 } 5828 5829 // Replace the %p in the core pattern with the process id. NOTE: we do this 5830 // only if the pattern doesn't start with "|", and we support only one %p in 5831 // the pattern. 5832 char *pid_pos = strstr(core_pattern, "%p"); 5833 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p" 5834 int written; 5835 5836 if (core_pattern[0] == '/') { 5837 if (pid_pos != NULL) { 5838 *pid_pos = '\0'; 5839 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern, 5840 current_process_id(), tail); 5841 } else { 5842 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern); 5843 } 5844 } else { 5845 char cwd[PATH_MAX]; 5846 5847 const char* p = get_current_directory(cwd, PATH_MAX); 5848 if (p == NULL) { 5849 return -1; 5850 } 5851 5852 if (core_pattern[0] == '|') { 5853 written = jio_snprintf(buffer, bufferSize, 5854 "\"%s\" (or dumping to %s/core.%d)", 5855 &core_pattern[1], p, current_process_id()); 5856 } else if (pid_pos != NULL) { 5857 *pid_pos = '\0'; 5858 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern, 5859 current_process_id(), tail); 5860 } else { 5861 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 5862 } 5863 } 5864 5865 if (written < 0) { 5866 return -1; 5867 } 5868 5869 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 5870 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 5871 5872 if (core_uses_pid_file != -1) { 5873 char core_uses_pid = 0; 5874 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 5875 ::close(core_uses_pid_file); 5876 5877 if (core_uses_pid == '1') { 5878 jio_snprintf(buffer + written, bufferSize - written, 5879 ".%d", current_process_id()); 5880 } 5881 } 5882 } 5883 5884 return strlen(buffer); 5885 } 5886 5887 bool os::start_debugging(char *buf, int buflen) { 5888 int len = (int)strlen(buf); 5889 char *p = &buf[len]; 5890 5891 jio_snprintf(p, buflen-len, 5892 "\n\n" 5893 "Do you want to debug the problem?\n\n" 5894 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n" 5895 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n" 5896 "Otherwise, press RETURN to abort...", 5897 os::current_process_id(), os::current_process_id(), 5898 os::current_thread_id(), os::current_thread_id()); 5899 5900 bool yes = os::message_box("Unexpected Error", buf); 5901 5902 if (yes) { 5903 // yes, user asked VM to launch debugger 5904 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d", 5905 os::current_process_id(), os::current_process_id()); 5906 5907 os::fork_and_exec(buf); 5908 yes = false; 5909 } 5910 return yes; 5911 } 5912 5913 5914 // Java/Compiler thread: 5915 // 5916 // Low memory addresses 5917 // P0 +------------------------+ 5918 // | |\ Java thread created by VM does not have glibc 5919 // | glibc guard page | - guard page, attached Java thread usually has 5920 // | |/ 1 glibc guard page. 5921 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 5922 // | |\ 5923 // | HotSpot Guard Pages | - red, yellow and reserved pages 5924 // | |/ 5925 // +------------------------+ JavaThread::stack_reserved_zone_base() 5926 // | |\ 5927 // | Normal Stack | - 5928 // | |/ 5929 // P2 +------------------------+ Thread::stack_base() 5930 // 5931 // Non-Java thread: 5932 // 5933 // Low memory addresses 5934 // P0 +------------------------+ 5935 // | |\ 5936 // | glibc guard page | - usually 1 page 5937 // | |/ 5938 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 5939 // | |\ 5940 // | Normal Stack | - 5941 // | |/ 5942 // P2 +------------------------+ Thread::stack_base() 5943 // 5944 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size 5945 // returned from pthread_attr_getstack(). 5946 // ** Due to NPTL implementation error, linux takes the glibc guard page out 5947 // of the stack size given in pthread_attr. We work around this for 5948 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.) 5949 // 5950 #ifndef ZERO 5951 static void current_stack_region(address * bottom, size_t * size) { 5952 if (os::is_primordial_thread()) { 5953 // primordial thread needs special handling because pthread_getattr_np() 5954 // may return bogus value. 5955 *bottom = os::Linux::initial_thread_stack_bottom(); 5956 *size = os::Linux::initial_thread_stack_size(); 5957 } else { 5958 pthread_attr_t attr; 5959 5960 int rslt = pthread_getattr_np(pthread_self(), &attr); 5961 5962 // JVM needs to know exact stack location, abort if it fails 5963 if (rslt != 0) { 5964 if (rslt == ENOMEM) { 5965 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np"); 5966 } else { 5967 fatal("pthread_getattr_np failed with error = %d", rslt); 5968 } 5969 } 5970 5971 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) { 5972 fatal("Cannot locate current stack attributes!"); 5973 } 5974 5975 // Work around NPTL stack guard error. 5976 size_t guard_size = 0; 5977 rslt = pthread_attr_getguardsize(&attr, &guard_size); 5978 if (rslt != 0) { 5979 fatal("pthread_attr_getguardsize failed with error = %d", rslt); 5980 } 5981 *bottom += guard_size; 5982 *size -= guard_size; 5983 5984 pthread_attr_destroy(&attr); 5985 5986 } 5987 assert(os::current_stack_pointer() >= *bottom && 5988 os::current_stack_pointer() < *bottom + *size, "just checking"); 5989 } 5990 5991 address os::current_stack_base() { 5992 address bottom; 5993 size_t size; 5994 current_stack_region(&bottom, &size); 5995 return (bottom + size); 5996 } 5997 5998 size_t os::current_stack_size() { 5999 // This stack size includes the usable stack and HotSpot guard pages 6000 // (for the threads that have Hotspot guard pages). 6001 address bottom; 6002 size_t size; 6003 current_stack_region(&bottom, &size); 6004 return size; 6005 } 6006 #endif 6007 6008 static inline struct timespec get_mtime(const char* filename) { 6009 struct stat st; 6010 int ret = os::stat(filename, &st); 6011 assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno)); 6012 return st.st_mtim; 6013 } 6014 6015 int os::compare_file_modified_times(const char* file1, const char* file2) { 6016 struct timespec filetime1 = get_mtime(file1); 6017 struct timespec filetime2 = get_mtime(file2); 6018 int diff = filetime1.tv_sec - filetime2.tv_sec; 6019 if (diff == 0) { 6020 return filetime1.tv_nsec - filetime2.tv_nsec; 6021 } 6022 return diff; 6023 } 6024 6025 /////////////// Unit tests /////////////// 6026 6027 #ifndef PRODUCT 6028 6029 class TestReserveMemorySpecial : AllStatic { 6030 public: 6031 static void small_page_write(void* addr, size_t size) { 6032 size_t page_size = os::vm_page_size(); 6033 6034 char* end = (char*)addr + size; 6035 for (char* p = (char*)addr; p < end; p += page_size) { 6036 *p = 1; 6037 } 6038 } 6039 6040 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6041 if (!UseHugeTLBFS) { 6042 return; 6043 } 6044 6045 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6046 6047 if (addr != NULL) { 6048 small_page_write(addr, size); 6049 6050 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6051 } 6052 } 6053 6054 static void test_reserve_memory_special_huge_tlbfs_only() { 6055 if (!UseHugeTLBFS) { 6056 return; 6057 } 6058 6059 size_t lp = os::large_page_size(); 6060 6061 for (size_t size = lp; size <= lp * 10; size += lp) { 6062 test_reserve_memory_special_huge_tlbfs_only(size); 6063 } 6064 } 6065 6066 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6067 size_t lp = os::large_page_size(); 6068 size_t ag = os::vm_allocation_granularity(); 6069 6070 // sizes to test 6071 const size_t sizes[] = { 6072 lp, lp + ag, lp + lp / 2, lp * 2, 6073 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2, 6074 lp * 10, lp * 10 + lp / 2 6075 }; 6076 const int num_sizes = sizeof(sizes) / sizeof(size_t); 6077 6078 // For each size/alignment combination, we test three scenarios: 6079 // 1) with req_addr == NULL 6080 // 2) with a non-null req_addr at which we expect to successfully allocate 6081 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we 6082 // expect the allocation to either fail or to ignore req_addr 6083 6084 // Pre-allocate two areas; they shall be as large as the largest allocation 6085 // and aligned to the largest alignment we will be testing. 6086 const size_t mapping_size = sizes[num_sizes - 1] * 2; 6087 char* const mapping1 = (char*) ::mmap(NULL, mapping_size, 6088 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6089 -1, 0); 6090 assert(mapping1 != MAP_FAILED, "should work"); 6091 6092 char* const mapping2 = (char*) ::mmap(NULL, mapping_size, 6093 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6094 -1, 0); 6095 assert(mapping2 != MAP_FAILED, "should work"); 6096 6097 // Unmap the first mapping, but leave the second mapping intact: the first 6098 // mapping will serve as a value for a "good" req_addr (case 2). The second 6099 // mapping, still intact, as "bad" req_addr (case 3). 6100 ::munmap(mapping1, mapping_size); 6101 6102 // Case 1 6103 for (int i = 0; i < num_sizes; i++) { 6104 const size_t size = sizes[i]; 6105 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6106 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6107 if (p != NULL) { 6108 assert(is_aligned(p, alignment), "must be"); 6109 small_page_write(p, size); 6110 os::Linux::release_memory_special_huge_tlbfs(p, size); 6111 } 6112 } 6113 } 6114 6115 // Case 2 6116 for (int i = 0; i < num_sizes; i++) { 6117 const size_t size = sizes[i]; 6118 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6119 char* const req_addr = align_up(mapping1, alignment); 6120 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6121 if (p != NULL) { 6122 assert(p == req_addr, "must be"); 6123 small_page_write(p, size); 6124 os::Linux::release_memory_special_huge_tlbfs(p, size); 6125 } 6126 } 6127 } 6128 6129 // Case 3 6130 for (int i = 0; i < num_sizes; i++) { 6131 const size_t size = sizes[i]; 6132 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6133 char* const req_addr = align_up(mapping2, alignment); 6134 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6135 // as the area around req_addr contains already existing mappings, the API should always 6136 // return NULL (as per contract, it cannot return another address) 6137 assert(p == NULL, "must be"); 6138 } 6139 } 6140 6141 ::munmap(mapping2, mapping_size); 6142 6143 } 6144 6145 static void test_reserve_memory_special_huge_tlbfs() { 6146 if (!UseHugeTLBFS) { 6147 return; 6148 } 6149 6150 test_reserve_memory_special_huge_tlbfs_only(); 6151 test_reserve_memory_special_huge_tlbfs_mixed(); 6152 } 6153 6154 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6155 if (!UseSHM) { 6156 return; 6157 } 6158 6159 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6160 6161 if (addr != NULL) { 6162 assert(is_aligned(addr, alignment), "Check"); 6163 assert(is_aligned(addr, os::large_page_size()), "Check"); 6164 6165 small_page_write(addr, size); 6166 6167 os::Linux::release_memory_special_shm(addr, size); 6168 } 6169 } 6170 6171 static void test_reserve_memory_special_shm() { 6172 size_t lp = os::large_page_size(); 6173 size_t ag = os::vm_allocation_granularity(); 6174 6175 for (size_t size = ag; size < lp * 3; size += ag) { 6176 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6177 test_reserve_memory_special_shm(size, alignment); 6178 } 6179 } 6180 } 6181 6182 static void test() { 6183 test_reserve_memory_special_huge_tlbfs(); 6184 test_reserve_memory_special_shm(); 6185 } 6186 }; 6187 6188 void TestReserveMemorySpecial_test() { 6189 TestReserveMemorySpecial::test(); 6190 } 6191 6192 #endif