1 /* 2 * Copyright (c) 1999, 2018, 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 "memory/allocation.inline.hpp" 37 #include "memory/filemap.hpp" 38 #include "oops/oop.inline.hpp" 39 #include "os_linux.inline.hpp" 40 #include "os_share_linux.hpp" 41 #include "osContainer_linux.hpp" 42 #include "prims/jniFastGetField.hpp" 43 #include "prims/jvm_misc.hpp" 44 #include "runtime/arguments.hpp" 45 #include "runtime/atomic.hpp" 46 #include "runtime/extendedPC.hpp" 47 #include "runtime/globals.hpp" 48 #include "runtime/interfaceSupport.hpp" 49 #include "runtime/init.hpp" 50 #include "runtime/java.hpp" 51 #include "runtime/javaCalls.hpp" 52 #include "runtime/mutexLocker.hpp" 53 #include "runtime/objectMonitor.hpp" 54 #include "runtime/orderAccess.inline.hpp" 55 #include "runtime/osThread.hpp" 56 #include "runtime/perfMemory.hpp" 57 #include "runtime/sharedRuntime.hpp" 58 #include "runtime/statSampler.hpp" 59 #include "runtime/stubRoutines.hpp" 60 #include "runtime/thread.inline.hpp" 61 #include "runtime/threadCritical.hpp" 62 #include "runtime/threadSMR.hpp" 63 #include "runtime/timer.hpp" 64 #include "semaphore_posix.hpp" 65 #include "services/attachListener.hpp" 66 #include "services/memTracker.hpp" 67 #include "services/runtimeService.hpp" 68 #include "utilities/align.hpp" 69 #include "utilities/decoder.hpp" 70 #include "utilities/defaultStream.hpp" 71 #include "utilities/events.hpp" 72 #include "utilities/elfFile.hpp" 73 #include "utilities/growableArray.hpp" 74 #include "utilities/macros.hpp" 75 #include "utilities/vmError.hpp" 76 77 // put OS-includes here 78 # include <sys/types.h> 79 # include <sys/mman.h> 80 # include <sys/stat.h> 81 # include <sys/select.h> 82 # include <pthread.h> 83 # include <signal.h> 84 # include <errno.h> 85 # include <dlfcn.h> 86 # include <stdio.h> 87 # include <unistd.h> 88 # include <sys/resource.h> 89 # include <pthread.h> 90 # include <sys/stat.h> 91 # include <sys/time.h> 92 # include <sys/times.h> 93 # include <sys/utsname.h> 94 # include <sys/socket.h> 95 # include <sys/wait.h> 96 # include <pwd.h> 97 # include <poll.h> 98 # include <fcntl.h> 99 # include <string.h> 100 # include <syscall.h> 101 # include <sys/sysinfo.h> 102 # include <gnu/libc-version.h> 103 # include <sys/ipc.h> 104 # include <sys/shm.h> 105 # include <link.h> 106 # include <stdint.h> 107 # include <inttypes.h> 108 # include <sys/ioctl.h> 109 110 #ifndef _GNU_SOURCE 111 #define _GNU_SOURCE 112 #include <sched.h> 113 #undef _GNU_SOURCE 114 #else 115 #include <sched.h> 116 #endif 117 118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 119 // getrusage() is prepared to handle the associated failure. 120 #ifndef RUSAGE_THREAD 121 #define RUSAGE_THREAD (1) /* only the calling thread */ 122 #endif 123 124 #define MAX_PATH (2 * K) 125 126 #define MAX_SECS 100000000 127 128 // for timer info max values which include all bits 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 130 131 #define LARGEPAGES_BIT (1 << 6) 132 #define DAX_SHARED_BIT (1 << 8) 133 //////////////////////////////////////////////////////////////////////////////// 134 // global variables 135 julong os::Linux::_physical_memory = 0; 136 137 address os::Linux::_initial_thread_stack_bottom = NULL; 138 uintptr_t os::Linux::_initial_thread_stack_size = 0; 139 140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 142 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL; 143 Mutex* os::Linux::_createThread_lock = NULL; 144 pthread_t os::Linux::_main_thread; 145 int os::Linux::_page_size = -1; 146 bool os::Linux::_supports_fast_thread_cpu_time = false; 147 uint32_t os::Linux::_os_version = 0; 148 const char * os::Linux::_glibc_version = NULL; 149 const char * os::Linux::_libpthread_version = NULL; 150 151 static jlong initial_time_count=0; 152 153 static int clock_tics_per_sec = 100; 154 155 // For diagnostics to print a message once. see run_periodic_checks 156 static sigset_t check_signal_done; 157 static bool check_signals = true; 158 159 // Signal number used to suspend/resume a thread 160 161 // do not use any signal number less than SIGSEGV, see 4355769 162 static int SR_signum = SIGUSR2; 163 sigset_t SR_sigset; 164 165 // utility functions 166 167 static int SR_initialize(); 168 169 julong os::available_memory() { 170 return Linux::available_memory(); 171 } 172 173 julong os::Linux::available_memory() { 174 // values in struct sysinfo are "unsigned long" 175 struct sysinfo si; 176 julong avail_mem; 177 178 if (OSContainer::is_containerized()) { 179 jlong mem_limit, mem_usage; 180 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) { 181 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value", 182 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit); 183 } 184 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) { 185 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage); 186 } 187 if (mem_limit > 0 && mem_usage > 0 ) { 188 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0; 189 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem); 190 return avail_mem; 191 } 192 } 193 194 sysinfo(&si); 195 avail_mem = (julong)si.freeram * si.mem_unit; 196 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem); 197 return avail_mem; 198 } 199 200 julong os::physical_memory() { 201 jlong phys_mem = 0; 202 if (OSContainer::is_containerized()) { 203 jlong mem_limit; 204 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) { 205 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit); 206 return mem_limit; 207 } 208 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value", 209 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit); 210 } 211 212 phys_mem = Linux::physical_memory(); 213 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem); 214 return phys_mem; 215 } 216 217 // Return true if user is running as root. 218 219 bool os::have_special_privileges() { 220 static bool init = false; 221 static bool privileges = false; 222 if (!init) { 223 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 224 init = true; 225 } 226 return privileges; 227 } 228 229 230 #ifndef SYS_gettid 231 // i386: 224, ia64: 1105, amd64: 186, sparc 143 232 #ifdef __ia64__ 233 #define SYS_gettid 1105 234 #else 235 #ifdef __i386__ 236 #define SYS_gettid 224 237 #else 238 #ifdef __amd64__ 239 #define SYS_gettid 186 240 #else 241 #ifdef __sparc__ 242 #define SYS_gettid 143 243 #else 244 #error define gettid for the arch 245 #endif 246 #endif 247 #endif 248 #endif 249 #endif 250 251 252 // pid_t gettid() 253 // 254 // Returns the kernel thread id of the currently running thread. Kernel 255 // thread id is used to access /proc. 256 pid_t os::Linux::gettid() { 257 int rslt = syscall(SYS_gettid); 258 assert(rslt != -1, "must be."); // old linuxthreads implementation? 259 return (pid_t)rslt; 260 } 261 262 // Most versions of linux have a bug where the number of processors are 263 // determined by looking at the /proc file system. In a chroot environment, 264 // the system call returns 1. This causes the VM to act as if it is 265 // a single processor and elide locking (see is_MP() call). 266 static bool unsafe_chroot_detected = false; 267 static const char *unstable_chroot_error = "/proc file system not found.\n" 268 "Java may be unstable running multithreaded in a chroot " 269 "environment on Linux when /proc filesystem is not mounted."; 270 271 void os::Linux::initialize_system_info() { 272 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 273 if (processor_count() == 1) { 274 pid_t pid = os::Linux::gettid(); 275 char fname[32]; 276 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 277 FILE *fp = fopen(fname, "r"); 278 if (fp == NULL) { 279 unsafe_chroot_detected = true; 280 } else { 281 fclose(fp); 282 } 283 } 284 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 285 assert(processor_count() > 0, "linux error"); 286 } 287 288 void os::init_system_properties_values() { 289 // The next steps are taken in the product version: 290 // 291 // Obtain the JAVA_HOME value from the location of libjvm.so. 292 // This library should be located at: 293 // <JAVA_HOME>/lib/{client|server}/libjvm.so. 294 // 295 // If "/jre/lib/" appears at the right place in the path, then we 296 // assume libjvm.so is installed in a JDK and we use this path. 297 // 298 // Otherwise exit with message: "Could not create the Java virtual machine." 299 // 300 // The following extra steps are taken in the debugging version: 301 // 302 // If "/jre/lib/" does NOT appear at the right place in the path 303 // instead of exit check for $JAVA_HOME environment variable. 304 // 305 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 306 // then we append a fake suffix "hotspot/libjvm.so" to this path so 307 // it looks like libjvm.so is installed there 308 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 309 // 310 // Otherwise exit. 311 // 312 // Important note: if the location of libjvm.so changes this 313 // code needs to be changed accordingly. 314 315 // See ld(1): 316 // The linker uses the following search paths to locate required 317 // shared libraries: 318 // 1: ... 319 // ... 320 // 7: The default directories, normally /lib and /usr/lib. 321 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390) 322 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 323 #else 324 #define DEFAULT_LIBPATH "/lib:/usr/lib" 325 #endif 326 327 // Base path of extensions installed on the system. 328 #define SYS_EXT_DIR "/usr/java/packages" 329 #define EXTENSIONS_DIR "/lib/ext" 330 331 // Buffer that fits several sprintfs. 332 // Note that the space for the colon and the trailing null are provided 333 // by the nulls included by the sizeof operator. 334 const size_t bufsize = 335 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 336 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 337 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 338 339 // sysclasspath, java_home, dll_dir 340 { 341 char *pslash; 342 os::jvm_path(buf, bufsize); 343 344 // Found the full path to libjvm.so. 345 // Now cut the path to <java_home>/jre if we can. 346 pslash = strrchr(buf, '/'); 347 if (pslash != NULL) { 348 *pslash = '\0'; // Get rid of /libjvm.so. 349 } 350 pslash = strrchr(buf, '/'); 351 if (pslash != NULL) { 352 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 353 } 354 Arguments::set_dll_dir(buf); 355 356 if (pslash != NULL) { 357 pslash = strrchr(buf, '/'); 358 if (pslash != NULL) { 359 *pslash = '\0'; // Get rid of /lib. 360 } 361 } 362 Arguments::set_java_home(buf); 363 set_boot_path('/', ':'); 364 } 365 366 // Where to look for native libraries. 367 // 368 // Note: Due to a legacy implementation, most of the library path 369 // is set in the launcher. This was to accomodate linking restrictions 370 // on legacy Linux implementations (which are no longer supported). 371 // Eventually, all the library path setting will be done here. 372 // 373 // However, to prevent the proliferation of improperly built native 374 // libraries, the new path component /usr/java/packages is added here. 375 // Eventually, all the library path setting will be done here. 376 { 377 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 378 // should always exist (until the legacy problem cited above is 379 // addressed). 380 const char *v = ::getenv("LD_LIBRARY_PATH"); 381 const char *v_colon = ":"; 382 if (v == NULL) { v = ""; v_colon = ""; } 383 // That's +1 for the colon and +1 for the trailing '\0'. 384 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 385 strlen(v) + 1 + 386 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1, 387 mtInternal); 388 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon); 389 Arguments::set_library_path(ld_library_path); 390 FREE_C_HEAP_ARRAY(char, ld_library_path); 391 } 392 393 // Extensions directories. 394 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 395 Arguments::set_ext_dirs(buf); 396 397 FREE_C_HEAP_ARRAY(char, buf); 398 399 #undef DEFAULT_LIBPATH 400 #undef SYS_EXT_DIR 401 #undef EXTENSIONS_DIR 402 } 403 404 //////////////////////////////////////////////////////////////////////////////// 405 // breakpoint support 406 407 void os::breakpoint() { 408 BREAKPOINT; 409 } 410 411 extern "C" void breakpoint() { 412 // use debugger to set breakpoint here 413 } 414 415 //////////////////////////////////////////////////////////////////////////////// 416 // signal support 417 418 debug_only(static bool signal_sets_initialized = false); 419 static sigset_t unblocked_sigs, vm_sigs; 420 421 bool os::Linux::is_sig_ignored(int sig) { 422 struct sigaction oact; 423 sigaction(sig, (struct sigaction*)NULL, &oact); 424 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 425 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 426 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) { 427 return true; 428 } else { 429 return false; 430 } 431 } 432 433 void os::Linux::signal_sets_init() { 434 // Should also have an assertion stating we are still single-threaded. 435 assert(!signal_sets_initialized, "Already initialized"); 436 // Fill in signals that are necessarily unblocked for all threads in 437 // the VM. Currently, we unblock the following signals: 438 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 439 // by -Xrs (=ReduceSignalUsage)); 440 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 441 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 442 // the dispositions or masks wrt these signals. 443 // Programs embedding the VM that want to use the above signals for their 444 // own purposes must, at this time, use the "-Xrs" option to prevent 445 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 446 // (See bug 4345157, and other related bugs). 447 // In reality, though, unblocking these signals is really a nop, since 448 // these signals are not blocked by default. 449 sigemptyset(&unblocked_sigs); 450 sigaddset(&unblocked_sigs, SIGILL); 451 sigaddset(&unblocked_sigs, SIGSEGV); 452 sigaddset(&unblocked_sigs, SIGBUS); 453 sigaddset(&unblocked_sigs, SIGFPE); 454 #if defined(PPC64) 455 sigaddset(&unblocked_sigs, SIGTRAP); 456 #endif 457 sigaddset(&unblocked_sigs, SR_signum); 458 459 if (!ReduceSignalUsage) { 460 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 461 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 462 } 463 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 464 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 465 } 466 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 467 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 468 } 469 } 470 // Fill in signals that are blocked by all but the VM thread. 471 sigemptyset(&vm_sigs); 472 if (!ReduceSignalUsage) { 473 sigaddset(&vm_sigs, BREAK_SIGNAL); 474 } 475 debug_only(signal_sets_initialized = true); 476 477 } 478 479 // These are signals that are unblocked while a thread is running Java. 480 // (For some reason, they get blocked by default.) 481 sigset_t* os::Linux::unblocked_signals() { 482 assert(signal_sets_initialized, "Not initialized"); 483 return &unblocked_sigs; 484 } 485 486 // These are the signals that are blocked while a (non-VM) thread is 487 // running Java. Only the VM thread handles these signals. 488 sigset_t* os::Linux::vm_signals() { 489 assert(signal_sets_initialized, "Not initialized"); 490 return &vm_sigs; 491 } 492 493 void os::Linux::hotspot_sigmask(Thread* thread) { 494 495 //Save caller's signal mask before setting VM signal mask 496 sigset_t caller_sigmask; 497 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 498 499 OSThread* osthread = thread->osthread(); 500 osthread->set_caller_sigmask(caller_sigmask); 501 502 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 503 504 if (!ReduceSignalUsage) { 505 if (thread->is_VM_thread()) { 506 // Only the VM thread handles BREAK_SIGNAL ... 507 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 508 } else { 509 // ... all other threads block BREAK_SIGNAL 510 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 511 } 512 } 513 } 514 515 ////////////////////////////////////////////////////////////////////////////// 516 // detecting pthread library 517 518 void os::Linux::libpthread_init() { 519 // Save glibc and pthread version strings. 520 #if !defined(_CS_GNU_LIBC_VERSION) || \ 521 !defined(_CS_GNU_LIBPTHREAD_VERSION) 522 #error "glibc too old (< 2.3.2)" 523 #endif 524 525 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 526 assert(n > 0, "cannot retrieve glibc version"); 527 char *str = (char *)malloc(n, mtInternal); 528 confstr(_CS_GNU_LIBC_VERSION, str, n); 529 os::Linux::set_glibc_version(str); 530 531 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 532 assert(n > 0, "cannot retrieve pthread version"); 533 str = (char *)malloc(n, mtInternal); 534 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 535 os::Linux::set_libpthread_version(str); 536 } 537 538 ///////////////////////////////////////////////////////////////////////////// 539 // thread stack expansion 540 541 // os::Linux::manually_expand_stack() takes care of expanding the thread 542 // stack. Note that this is normally not needed: pthread stacks allocate 543 // thread stack using mmap() without MAP_NORESERVE, so the stack is already 544 // committed. Therefore it is not necessary to expand the stack manually. 545 // 546 // Manually expanding the stack was historically needed on LinuxThreads 547 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays 548 // it is kept to deal with very rare corner cases: 549 // 550 // For one, user may run the VM on an own implementation of threads 551 // whose stacks are - like the old LinuxThreads - implemented using 552 // mmap(MAP_GROWSDOWN). 553 // 554 // Also, this coding may be needed if the VM is running on the primordial 555 // thread. Normally we avoid running on the primordial thread; however, 556 // user may still invoke the VM on the primordial thread. 557 // 558 // The following historical comment describes the details about running 559 // on a thread stack allocated with mmap(MAP_GROWSDOWN): 560 561 562 // Force Linux kernel to expand current thread stack. If "bottom" is close 563 // to the stack guard, caller should block all signals. 564 // 565 // MAP_GROWSDOWN: 566 // A special mmap() flag that is used to implement thread stacks. It tells 567 // kernel that the memory region should extend downwards when needed. This 568 // allows early versions of LinuxThreads to only mmap the first few pages 569 // when creating a new thread. Linux kernel will automatically expand thread 570 // stack as needed (on page faults). 571 // 572 // However, because the memory region of a MAP_GROWSDOWN stack can grow on 573 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 574 // region, it's hard to tell if the fault is due to a legitimate stack 575 // access or because of reading/writing non-exist memory (e.g. buffer 576 // overrun). As a rule, if the fault happens below current stack pointer, 577 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the 578 // application (see Linux kernel fault.c). 579 // 580 // This Linux feature can cause SIGSEGV when VM bangs thread stack for 581 // stack overflow detection. 582 // 583 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 584 // not use MAP_GROWSDOWN. 585 // 586 // To get around the problem and allow stack banging on Linux, we need to 587 // manually expand thread stack after receiving the SIGSEGV. 588 // 589 // There are two ways to expand thread stack to address "bottom", we used 590 // both of them in JVM before 1.5: 591 // 1. adjust stack pointer first so that it is below "bottom", and then 592 // touch "bottom" 593 // 2. mmap() the page in question 594 // 595 // Now alternate signal stack is gone, it's harder to use 2. For instance, 596 // if current sp is already near the lower end of page 101, and we need to 597 // call mmap() to map page 100, it is possible that part of the mmap() frame 598 // will be placed in page 100. When page 100 is mapped, it is zero-filled. 599 // That will destroy the mmap() frame and cause VM to crash. 600 // 601 // The following code works by adjusting sp first, then accessing the "bottom" 602 // page to force a page fault. Linux kernel will then automatically expand the 603 // stack mapping. 604 // 605 // _expand_stack_to() assumes its frame size is less than page size, which 606 // should always be true if the function is not inlined. 607 608 static void NOINLINE _expand_stack_to(address bottom) { 609 address sp; 610 size_t size; 611 volatile char *p; 612 613 // Adjust bottom to point to the largest address within the same page, it 614 // gives us a one-page buffer if alloca() allocates slightly more memory. 615 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size()); 616 bottom += os::Linux::page_size() - 1; 617 618 // sp might be slightly above current stack pointer; if that's the case, we 619 // will alloca() a little more space than necessary, which is OK. Don't use 620 // os::current_stack_pointer(), as its result can be slightly below current 621 // stack pointer, causing us to not alloca enough to reach "bottom". 622 sp = (address)&sp; 623 624 if (sp > bottom) { 625 size = sp - bottom; 626 p = (volatile char *)alloca(size); 627 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 628 p[0] = '\0'; 629 } 630 } 631 632 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 633 assert(t!=NULL, "just checking"); 634 assert(t->osthread()->expanding_stack(), "expand should be set"); 635 assert(t->stack_base() != NULL, "stack_base was not initialized"); 636 637 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) { 638 sigset_t mask_all, old_sigset; 639 sigfillset(&mask_all); 640 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 641 _expand_stack_to(addr); 642 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 643 return true; 644 } 645 return false; 646 } 647 648 ////////////////////////////////////////////////////////////////////////////// 649 // create new thread 650 651 // Thread start routine for all newly created threads 652 static void *thread_native_entry(Thread *thread) { 653 // Try to randomize the cache line index of hot stack frames. 654 // This helps when threads of the same stack traces evict each other's 655 // cache lines. The threads can be either from the same JVM instance, or 656 // from different JVM instances. The benefit is especially true for 657 // processors with hyperthreading technology. 658 static int counter = 0; 659 int pid = os::current_process_id(); 660 alloca(((pid ^ counter++) & 7) * 128); 661 662 thread->initialize_thread_current(); 663 664 OSThread* osthread = thread->osthread(); 665 Monitor* sync = osthread->startThread_lock(); 666 667 osthread->set_thread_id(os::current_thread_id()); 668 669 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 670 os::current_thread_id(), (uintx) pthread_self()); 671 672 if (UseNUMA) { 673 int lgrp_id = os::numa_get_group_id(); 674 if (lgrp_id != -1) { 675 thread->set_lgrp_id(lgrp_id); 676 } 677 } 678 // initialize signal mask for this thread 679 os::Linux::hotspot_sigmask(thread); 680 681 // initialize floating point control register 682 os::Linux::init_thread_fpu_state(); 683 684 // handshaking with parent thread 685 { 686 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 687 688 // notify parent thread 689 osthread->set_state(INITIALIZED); 690 sync->notify_all(); 691 692 // wait until os::start_thread() 693 while (osthread->get_state() == INITIALIZED) { 694 sync->wait(Mutex::_no_safepoint_check_flag); 695 } 696 } 697 698 // call one more level start routine 699 thread->run(); 700 701 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 702 os::current_thread_id(), (uintx) pthread_self()); 703 704 // If a thread has not deleted itself ("delete this") as part of its 705 // termination sequence, we have to ensure thread-local-storage is 706 // cleared before we actually terminate. No threads should ever be 707 // deleted asynchronously with respect to their termination. 708 if (Thread::current_or_null_safe() != NULL) { 709 assert(Thread::current_or_null_safe() == thread, "current thread is wrong"); 710 thread->clear_thread_current(); 711 } 712 713 return 0; 714 } 715 716 bool os::create_thread(Thread* thread, ThreadType thr_type, 717 size_t req_stack_size) { 718 assert(thread->osthread() == NULL, "caller responsible"); 719 720 // Allocate the OSThread object 721 OSThread* osthread = new OSThread(NULL, NULL); 722 if (osthread == NULL) { 723 return false; 724 } 725 726 // set the correct thread state 727 osthread->set_thread_type(thr_type); 728 729 // Initial state is ALLOCATED but not INITIALIZED 730 osthread->set_state(ALLOCATED); 731 732 thread->set_osthread(osthread); 733 734 // init thread attributes 735 pthread_attr_t attr; 736 pthread_attr_init(&attr); 737 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 738 739 // Calculate stack size if it's not specified by caller. 740 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size); 741 // In the Linux NPTL pthread implementation the guard size mechanism 742 // is not implemented properly. The posix standard requires adding 743 // the size of the guard pages to the stack size, instead Linux 744 // takes the space out of 'stacksize'. Thus we adapt the requested 745 // stack_size by the size of the guard pages to mimick proper 746 // behaviour. However, be careful not to end up with a size 747 // of zero due to overflow. Don't add the guard page in that case. 748 size_t guard_size = os::Linux::default_guard_size(thr_type); 749 if (stack_size <= SIZE_MAX - guard_size) { 750 stack_size += guard_size; 751 } 752 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned"); 753 754 int status = pthread_attr_setstacksize(&attr, stack_size); 755 assert_status(status == 0, status, "pthread_attr_setstacksize"); 756 757 // Configure glibc guard page. 758 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 759 760 ThreadState state; 761 762 { 763 pthread_t tid; 764 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread); 765 766 char buf[64]; 767 if (ret == 0) { 768 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ", 769 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 770 } else { 771 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.", 772 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 773 } 774 775 pthread_attr_destroy(&attr); 776 777 if (ret != 0) { 778 // Need to clean up stuff we've allocated so far 779 thread->set_osthread(NULL); 780 delete osthread; 781 return false; 782 } 783 784 // Store pthread info into the OSThread 785 osthread->set_pthread_id(tid); 786 787 // Wait until child thread is either initialized or aborted 788 { 789 Monitor* sync_with_child = osthread->startThread_lock(); 790 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 791 while ((state = osthread->get_state()) == ALLOCATED) { 792 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 793 } 794 } 795 } 796 797 // Aborted due to thread limit being reached 798 if (state == ZOMBIE) { 799 thread->set_osthread(NULL); 800 delete osthread; 801 return false; 802 } 803 804 // The thread is returned suspended (in state INITIALIZED), 805 // and is started higher up in the call chain 806 assert(state == INITIALIZED, "race condition"); 807 return true; 808 } 809 810 ///////////////////////////////////////////////////////////////////////////// 811 // attach existing thread 812 813 // bootstrap the main thread 814 bool os::create_main_thread(JavaThread* thread) { 815 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 816 return create_attached_thread(thread); 817 } 818 819 bool os::create_attached_thread(JavaThread* thread) { 820 #ifdef ASSERT 821 thread->verify_not_published(); 822 #endif 823 824 // Allocate the OSThread object 825 OSThread* osthread = new OSThread(NULL, NULL); 826 827 if (osthread == NULL) { 828 return false; 829 } 830 831 // Store pthread info into the OSThread 832 osthread->set_thread_id(os::Linux::gettid()); 833 osthread->set_pthread_id(::pthread_self()); 834 835 // initialize floating point control register 836 os::Linux::init_thread_fpu_state(); 837 838 // Initial thread state is RUNNABLE 839 osthread->set_state(RUNNABLE); 840 841 thread->set_osthread(osthread); 842 843 if (UseNUMA) { 844 int lgrp_id = os::numa_get_group_id(); 845 if (lgrp_id != -1) { 846 thread->set_lgrp_id(lgrp_id); 847 } 848 } 849 850 if (os::is_primordial_thread()) { 851 // If current thread is primordial thread, its stack is mapped on demand, 852 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 853 // the entire stack region to avoid SEGV in stack banging. 854 // It is also useful to get around the heap-stack-gap problem on SuSE 855 // kernel (see 4821821 for details). We first expand stack to the top 856 // of yellow zone, then enable stack yellow zone (order is significant, 857 // enabling yellow zone first will crash JVM on SuSE Linux), so there 858 // is no gap between the last two virtual memory regions. 859 860 JavaThread *jt = (JavaThread *)thread; 861 address addr = jt->stack_reserved_zone_base(); 862 assert(addr != NULL, "initialization problem?"); 863 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 864 865 osthread->set_expanding_stack(); 866 os::Linux::manually_expand_stack(jt, addr); 867 osthread->clear_expanding_stack(); 868 } 869 870 // initialize signal mask for this thread 871 // and save the caller's signal mask 872 os::Linux::hotspot_sigmask(thread); 873 874 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 875 os::current_thread_id(), (uintx) pthread_self()); 876 877 return true; 878 } 879 880 void os::pd_start_thread(Thread* thread) { 881 OSThread * osthread = thread->osthread(); 882 assert(osthread->get_state() != INITIALIZED, "just checking"); 883 Monitor* sync_with_child = osthread->startThread_lock(); 884 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 885 sync_with_child->notify(); 886 } 887 888 // Free Linux resources related to the OSThread 889 void os::free_thread(OSThread* osthread) { 890 assert(osthread != NULL, "osthread not set"); 891 892 // We are told to free resources of the argument thread, 893 // but we can only really operate on the current thread. 894 assert(Thread::current()->osthread() == osthread, 895 "os::free_thread but not current thread"); 896 897 #ifdef ASSERT 898 sigset_t current; 899 sigemptyset(¤t); 900 pthread_sigmask(SIG_SETMASK, NULL, ¤t); 901 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!"); 902 #endif 903 904 // Restore caller's signal mask 905 sigset_t sigmask = osthread->caller_sigmask(); 906 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 907 908 delete osthread; 909 } 910 911 ////////////////////////////////////////////////////////////////////////////// 912 // primordial thread 913 914 // Check if current thread is the primordial thread, similar to Solaris thr_main. 915 bool os::is_primordial_thread(void) { 916 char dummy; 917 // If called before init complete, thread stack bottom will be null. 918 // Can be called if fatal error occurs before initialization. 919 if (os::Linux::initial_thread_stack_bottom() == NULL) return false; 920 assert(os::Linux::initial_thread_stack_bottom() != NULL && 921 os::Linux::initial_thread_stack_size() != 0, 922 "os::init did not locate primordial thread's stack region"); 923 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() && 924 (address)&dummy < os::Linux::initial_thread_stack_bottom() + 925 os::Linux::initial_thread_stack_size()) { 926 return true; 927 } else { 928 return false; 929 } 930 } 931 932 // Find the virtual memory area that contains addr 933 static bool find_vma(address addr, address* vma_low, address* vma_high) { 934 FILE *fp = fopen("/proc/self/maps", "r"); 935 if (fp) { 936 address low, high; 937 while (!feof(fp)) { 938 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 939 if (low <= addr && addr < high) { 940 if (vma_low) *vma_low = low; 941 if (vma_high) *vma_high = high; 942 fclose(fp); 943 return true; 944 } 945 } 946 for (;;) { 947 int ch = fgetc(fp); 948 if (ch == EOF || ch == (int)'\n') break; 949 } 950 } 951 fclose(fp); 952 } 953 return false; 954 } 955 956 // Locate primordial thread stack. This special handling of primordial thread stack 957 // is needed because pthread_getattr_np() on most (all?) Linux distros returns 958 // bogus value for the primordial process thread. While the launcher has created 959 // the VM in a new thread since JDK 6, we still have to allow for the use of the 960 // JNI invocation API from a primordial thread. 961 void os::Linux::capture_initial_stack(size_t max_size) { 962 963 // max_size is either 0 (which means accept OS default for thread stacks) or 964 // a user-specified value known to be at least the minimum needed. If we 965 // are actually on the primordial thread we can make it appear that we have a 966 // smaller max_size stack by inserting the guard pages at that location. But we 967 // cannot do anything to emulate a larger stack than what has been provided by 968 // the OS or threading library. In fact if we try to use a stack greater than 969 // what is set by rlimit then we will crash the hosting process. 970 971 // Maximum stack size is the easy part, get it from RLIMIT_STACK. 972 // If this is "unlimited" then it will be a huge value. 973 struct rlimit rlim; 974 getrlimit(RLIMIT_STACK, &rlim); 975 size_t stack_size = rlim.rlim_cur; 976 977 // 6308388: a bug in ld.so will relocate its own .data section to the 978 // lower end of primordial stack; reduce ulimit -s value a little bit 979 // so we won't install guard page on ld.so's data section. 980 // But ensure we don't underflow the stack size - allow 1 page spare 981 if (stack_size >= (size_t)(3 * page_size())) { 982 stack_size -= 2 * page_size(); 983 } 984 985 // Try to figure out where the stack base (top) is. This is harder. 986 // 987 // When an application is started, glibc saves the initial stack pointer in 988 // a global variable "__libc_stack_end", which is then used by system 989 // libraries. __libc_stack_end should be pretty close to stack top. The 990 // variable is available since the very early days. However, because it is 991 // a private interface, it could disappear in the future. 992 // 993 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 994 // to __libc_stack_end, it is very close to stack top, but isn't the real 995 // stack top. Note that /proc may not exist if VM is running as a chroot 996 // program, so reading /proc/<pid>/stat could fail. Also the contents of 997 // /proc/<pid>/stat could change in the future (though unlikely). 998 // 999 // We try __libc_stack_end first. If that doesn't work, look for 1000 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1001 // as a hint, which should work well in most cases. 1002 1003 uintptr_t stack_start; 1004 1005 // try __libc_stack_end first 1006 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1007 if (p && *p) { 1008 stack_start = *p; 1009 } else { 1010 // see if we can get the start_stack field from /proc/self/stat 1011 FILE *fp; 1012 int pid; 1013 char state; 1014 int ppid; 1015 int pgrp; 1016 int session; 1017 int nr; 1018 int tpgrp; 1019 unsigned long flags; 1020 unsigned long minflt; 1021 unsigned long cminflt; 1022 unsigned long majflt; 1023 unsigned long cmajflt; 1024 unsigned long utime; 1025 unsigned long stime; 1026 long cutime; 1027 long cstime; 1028 long prio; 1029 long nice; 1030 long junk; 1031 long it_real; 1032 uintptr_t start; 1033 uintptr_t vsize; 1034 intptr_t rss; 1035 uintptr_t rsslim; 1036 uintptr_t scodes; 1037 uintptr_t ecode; 1038 int i; 1039 1040 // Figure what the primordial thread stack base is. Code is inspired 1041 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1042 // followed by command name surrounded by parentheses, state, etc. 1043 char stat[2048]; 1044 int statlen; 1045 1046 fp = fopen("/proc/self/stat", "r"); 1047 if (fp) { 1048 statlen = fread(stat, 1, 2047, fp); 1049 stat[statlen] = '\0'; 1050 fclose(fp); 1051 1052 // Skip pid and the command string. Note that we could be dealing with 1053 // weird command names, e.g. user could decide to rename java launcher 1054 // to "java 1.4.2 :)", then the stat file would look like 1055 // 1234 (java 1.4.2 :)) R ... ... 1056 // We don't really need to know the command string, just find the last 1057 // occurrence of ")" and then start parsing from there. See bug 4726580. 1058 char * s = strrchr(stat, ')'); 1059 1060 i = 0; 1061 if (s) { 1062 // Skip blank chars 1063 do { s++; } while (s && isspace(*s)); 1064 1065 #define _UFM UINTX_FORMAT 1066 #define _DFM INTX_FORMAT 1067 1068 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1069 // 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 1070 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, 1071 &state, // 3 %c 1072 &ppid, // 4 %d 1073 &pgrp, // 5 %d 1074 &session, // 6 %d 1075 &nr, // 7 %d 1076 &tpgrp, // 8 %d 1077 &flags, // 9 %lu 1078 &minflt, // 10 %lu 1079 &cminflt, // 11 %lu 1080 &majflt, // 12 %lu 1081 &cmajflt, // 13 %lu 1082 &utime, // 14 %lu 1083 &stime, // 15 %lu 1084 &cutime, // 16 %ld 1085 &cstime, // 17 %ld 1086 &prio, // 18 %ld 1087 &nice, // 19 %ld 1088 &junk, // 20 %ld 1089 &it_real, // 21 %ld 1090 &start, // 22 UINTX_FORMAT 1091 &vsize, // 23 UINTX_FORMAT 1092 &rss, // 24 INTX_FORMAT 1093 &rsslim, // 25 UINTX_FORMAT 1094 &scodes, // 26 UINTX_FORMAT 1095 &ecode, // 27 UINTX_FORMAT 1096 &stack_start); // 28 UINTX_FORMAT 1097 } 1098 1099 #undef _UFM 1100 #undef _DFM 1101 1102 if (i != 28 - 2) { 1103 assert(false, "Bad conversion from /proc/self/stat"); 1104 // product mode - assume we are the primordial thread, good luck in the 1105 // embedded case. 1106 warning("Can't detect primordial thread stack location - bad conversion"); 1107 stack_start = (uintptr_t) &rlim; 1108 } 1109 } else { 1110 // For some reason we can't open /proc/self/stat (for example, running on 1111 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1112 // most cases, so don't abort: 1113 warning("Can't detect primordial thread stack location - no /proc/self/stat"); 1114 stack_start = (uintptr_t) &rlim; 1115 } 1116 } 1117 1118 // Now we have a pointer (stack_start) very close to the stack top, the 1119 // next thing to do is to figure out the exact location of stack top. We 1120 // can find out the virtual memory area that contains stack_start by 1121 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1122 // and its upper limit is the real stack top. (again, this would fail if 1123 // running inside chroot, because /proc may not exist.) 1124 1125 uintptr_t stack_top; 1126 address low, high; 1127 if (find_vma((address)stack_start, &low, &high)) { 1128 // success, "high" is the true stack top. (ignore "low", because initial 1129 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1130 stack_top = (uintptr_t)high; 1131 } else { 1132 // failed, likely because /proc/self/maps does not exist 1133 warning("Can't detect primordial thread stack location - find_vma failed"); 1134 // best effort: stack_start is normally within a few pages below the real 1135 // stack top, use it as stack top, and reduce stack size so we won't put 1136 // guard page outside stack. 1137 stack_top = stack_start; 1138 stack_size -= 16 * page_size(); 1139 } 1140 1141 // stack_top could be partially down the page so align it 1142 stack_top = align_up(stack_top, page_size()); 1143 1144 // Allowed stack value is minimum of max_size and what we derived from rlimit 1145 if (max_size > 0) { 1146 _initial_thread_stack_size = MIN2(max_size, stack_size); 1147 } else { 1148 // Accept the rlimit max, but if stack is unlimited then it will be huge, so 1149 // clamp it at 8MB as we do on Solaris 1150 _initial_thread_stack_size = MIN2(stack_size, 8*M); 1151 } 1152 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size()); 1153 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1154 1155 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!"); 1156 1157 if (log_is_enabled(Info, os, thread)) { 1158 // See if we seem to be on primordial process thread 1159 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) && 1160 uintptr_t(&rlim) < stack_top; 1161 1162 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: " 1163 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT, 1164 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K, 1165 stack_top, intptr_t(_initial_thread_stack_bottom)); 1166 } 1167 } 1168 1169 //////////////////////////////////////////////////////////////////////////////// 1170 // time support 1171 1172 // Time since start-up in seconds to a fine granularity. 1173 // Used by VMSelfDestructTimer and the MemProfiler. 1174 double os::elapsedTime() { 1175 1176 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1177 } 1178 1179 jlong os::elapsed_counter() { 1180 return javaTimeNanos() - initial_time_count; 1181 } 1182 1183 jlong os::elapsed_frequency() { 1184 return NANOSECS_PER_SEC; // nanosecond resolution 1185 } 1186 1187 bool os::supports_vtime() { return true; } 1188 bool os::enable_vtime() { return false; } 1189 bool os::vtime_enabled() { return false; } 1190 1191 double os::elapsedVTime() { 1192 struct rusage usage; 1193 int retval = getrusage(RUSAGE_THREAD, &usage); 1194 if (retval == 0) { 1195 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); 1196 } else { 1197 // better than nothing, but not much 1198 return elapsedTime(); 1199 } 1200 } 1201 1202 jlong os::javaTimeMillis() { 1203 timeval time; 1204 int status = gettimeofday(&time, NULL); 1205 assert(status != -1, "linux error"); 1206 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1207 } 1208 1209 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) { 1210 timeval time; 1211 int status = gettimeofday(&time, NULL); 1212 assert(status != -1, "linux error"); 1213 seconds = jlong(time.tv_sec); 1214 nanos = jlong(time.tv_usec) * 1000; 1215 } 1216 1217 1218 #ifndef CLOCK_MONOTONIC 1219 #define CLOCK_MONOTONIC (1) 1220 #endif 1221 1222 void os::Linux::clock_init() { 1223 // we do dlopen's in this particular order due to bug in linux 1224 // dynamical loader (see 6348968) leading to crash on exit 1225 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1226 if (handle == NULL) { 1227 handle = dlopen("librt.so", RTLD_LAZY); 1228 } 1229 1230 if (handle) { 1231 int (*clock_getres_func)(clockid_t, struct timespec*) = 1232 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1233 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1234 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1235 if (clock_getres_func && clock_gettime_func) { 1236 // See if monotonic clock is supported by the kernel. Note that some 1237 // early implementations simply return kernel jiffies (updated every 1238 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1239 // for nano time (though the monotonic property is still nice to have). 1240 // It's fixed in newer kernels, however clock_getres() still returns 1241 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1242 // resolution for now. Hopefully as people move to new kernels, this 1243 // won't be a problem. 1244 struct timespec res; 1245 struct timespec tp; 1246 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1247 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1248 // yes, monotonic clock is supported 1249 _clock_gettime = clock_gettime_func; 1250 return; 1251 } else { 1252 // close librt if there is no monotonic clock 1253 dlclose(handle); 1254 } 1255 } 1256 } 1257 warning("No monotonic clock was available - timed services may " \ 1258 "be adversely affected if the time-of-day clock changes"); 1259 } 1260 1261 #ifndef SYS_clock_getres 1262 #if defined(X86) || defined(PPC64) || defined(S390) 1263 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261) 1264 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1265 #else 1266 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1267 #define sys_clock_getres(x,y) -1 1268 #endif 1269 #else 1270 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1271 #endif 1272 1273 void os::Linux::fast_thread_clock_init() { 1274 if (!UseLinuxPosixThreadCPUClocks) { 1275 return; 1276 } 1277 clockid_t clockid; 1278 struct timespec tp; 1279 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1280 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1281 1282 // Switch to using fast clocks for thread cpu time if 1283 // the sys_clock_getres() returns 0 error code. 1284 // Note, that some kernels may support the current thread 1285 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1286 // returned by the pthread_getcpuclockid(). 1287 // If the fast Posix clocks are supported then the sys_clock_getres() 1288 // must return at least tp.tv_sec == 0 which means a resolution 1289 // better than 1 sec. This is extra check for reliability. 1290 1291 if (pthread_getcpuclockid_func && 1292 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1293 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1294 _supports_fast_thread_cpu_time = true; 1295 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1296 } 1297 } 1298 1299 jlong os::javaTimeNanos() { 1300 if (os::supports_monotonic_clock()) { 1301 struct timespec tp; 1302 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1303 assert(status == 0, "gettime error"); 1304 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1305 return result; 1306 } else { 1307 timeval time; 1308 int status = gettimeofday(&time, NULL); 1309 assert(status != -1, "linux error"); 1310 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1311 return 1000 * usecs; 1312 } 1313 } 1314 1315 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1316 if (os::supports_monotonic_clock()) { 1317 info_ptr->max_value = ALL_64_BITS; 1318 1319 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1320 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1321 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1322 } else { 1323 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1324 info_ptr->max_value = ALL_64_BITS; 1325 1326 // gettimeofday is a real time clock so it skips 1327 info_ptr->may_skip_backward = true; 1328 info_ptr->may_skip_forward = true; 1329 } 1330 1331 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1332 } 1333 1334 // Return the real, user, and system times in seconds from an 1335 // arbitrary fixed point in the past. 1336 bool os::getTimesSecs(double* process_real_time, 1337 double* process_user_time, 1338 double* process_system_time) { 1339 struct tms ticks; 1340 clock_t real_ticks = times(&ticks); 1341 1342 if (real_ticks == (clock_t) (-1)) { 1343 return false; 1344 } else { 1345 double ticks_per_second = (double) clock_tics_per_sec; 1346 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1347 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1348 *process_real_time = ((double) real_ticks) / ticks_per_second; 1349 1350 return true; 1351 } 1352 } 1353 1354 1355 char * os::local_time_string(char *buf, size_t buflen) { 1356 struct tm t; 1357 time_t long_time; 1358 time(&long_time); 1359 localtime_r(&long_time, &t); 1360 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1361 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1362 t.tm_hour, t.tm_min, t.tm_sec); 1363 return buf; 1364 } 1365 1366 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1367 return localtime_r(clock, res); 1368 } 1369 1370 //////////////////////////////////////////////////////////////////////////////// 1371 // runtime exit support 1372 1373 // Note: os::shutdown() might be called very early during initialization, or 1374 // called from signal handler. Before adding something to os::shutdown(), make 1375 // sure it is async-safe and can handle partially initialized VM. 1376 void os::shutdown() { 1377 1378 // allow PerfMemory to attempt cleanup of any persistent resources 1379 perfMemory_exit(); 1380 1381 // needs to remove object in file system 1382 AttachListener::abort(); 1383 1384 // flush buffered output, finish log files 1385 ostream_abort(); 1386 1387 // Check for abort hook 1388 abort_hook_t abort_hook = Arguments::abort_hook(); 1389 if (abort_hook != NULL) { 1390 abort_hook(); 1391 } 1392 1393 } 1394 1395 // Note: os::abort() might be called very early during initialization, or 1396 // called from signal handler. Before adding something to os::abort(), make 1397 // sure it is async-safe and can handle partially initialized VM. 1398 void os::abort(bool dump_core, void* siginfo, const void* context) { 1399 os::shutdown(); 1400 if (dump_core) { 1401 #ifndef PRODUCT 1402 fdStream out(defaultStream::output_fd()); 1403 out.print_raw("Current thread is "); 1404 char buf[16]; 1405 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1406 out.print_raw_cr(buf); 1407 out.print_raw_cr("Dumping core ..."); 1408 #endif 1409 ::abort(); // dump core 1410 } 1411 1412 ::exit(1); 1413 } 1414 1415 // Die immediately, no exit hook, no abort hook, no cleanup. 1416 void os::die() { 1417 ::abort(); 1418 } 1419 1420 1421 // This method is a copy of JDK's sysGetLastErrorString 1422 // from src/solaris/hpi/src/system_md.c 1423 1424 size_t os::lasterror(char *buf, size_t len) { 1425 if (errno == 0) return 0; 1426 1427 const char *s = os::strerror(errno); 1428 size_t n = ::strlen(s); 1429 if (n >= len) { 1430 n = len - 1; 1431 } 1432 ::strncpy(buf, s, n); 1433 buf[n] = '\0'; 1434 return n; 1435 } 1436 1437 // thread_id is kernel thread id (similar to Solaris LWP id) 1438 intx os::current_thread_id() { return os::Linux::gettid(); } 1439 int os::current_process_id() { 1440 return ::getpid(); 1441 } 1442 1443 // DLL functions 1444 1445 const char* os::dll_file_extension() { return ".so"; } 1446 1447 // This must be hard coded because it's the system's temporary 1448 // directory not the java application's temp directory, ala java.io.tmpdir. 1449 const char* os::get_temp_directory() { return "/tmp"; } 1450 1451 static bool file_exists(const char* filename) { 1452 struct stat statbuf; 1453 if (filename == NULL || strlen(filename) == 0) { 1454 return false; 1455 } 1456 return os::stat(filename, &statbuf) == 0; 1457 } 1458 1459 // check if addr is inside libjvm.so 1460 bool os::address_is_in_vm(address addr) { 1461 static address libjvm_base_addr; 1462 Dl_info dlinfo; 1463 1464 if (libjvm_base_addr == NULL) { 1465 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1466 libjvm_base_addr = (address)dlinfo.dli_fbase; 1467 } 1468 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1469 } 1470 1471 if (dladdr((void *)addr, &dlinfo) != 0) { 1472 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1473 } 1474 1475 return false; 1476 } 1477 1478 bool os::dll_address_to_function_name(address addr, char *buf, 1479 int buflen, int *offset, 1480 bool demangle) { 1481 // buf is not optional, but offset is optional 1482 assert(buf != NULL, "sanity check"); 1483 1484 Dl_info dlinfo; 1485 1486 if (dladdr((void*)addr, &dlinfo) != 0) { 1487 // see if we have a matching symbol 1488 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1489 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) { 1490 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1491 } 1492 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1493 return true; 1494 } 1495 // no matching symbol so try for just file info 1496 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1497 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1498 buf, buflen, offset, dlinfo.dli_fname, demangle)) { 1499 return true; 1500 } 1501 } 1502 } 1503 1504 buf[0] = '\0'; 1505 if (offset != NULL) *offset = -1; 1506 return false; 1507 } 1508 1509 struct _address_to_library_name { 1510 address addr; // input : memory address 1511 size_t buflen; // size of fname 1512 char* fname; // output: library name 1513 address base; // library base addr 1514 }; 1515 1516 static int address_to_library_name_callback(struct dl_phdr_info *info, 1517 size_t size, void *data) { 1518 int i; 1519 bool found = false; 1520 address libbase = NULL; 1521 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1522 1523 // iterate through all loadable segments 1524 for (i = 0; i < info->dlpi_phnum; i++) { 1525 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1526 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1527 // base address of a library is the lowest address of its loaded 1528 // segments. 1529 if (libbase == NULL || libbase > segbase) { 1530 libbase = segbase; 1531 } 1532 // see if 'addr' is within current segment 1533 if (segbase <= d->addr && 1534 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1535 found = true; 1536 } 1537 } 1538 } 1539 1540 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1541 // so dll_address_to_library_name() can fall through to use dladdr() which 1542 // can figure out executable name from argv[0]. 1543 if (found && info->dlpi_name && info->dlpi_name[0]) { 1544 d->base = libbase; 1545 if (d->fname) { 1546 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1547 } 1548 return 1; 1549 } 1550 return 0; 1551 } 1552 1553 bool os::dll_address_to_library_name(address addr, char* buf, 1554 int buflen, int* offset) { 1555 // buf is not optional, but offset is optional 1556 assert(buf != NULL, "sanity check"); 1557 1558 Dl_info dlinfo; 1559 struct _address_to_library_name data; 1560 1561 // There is a bug in old glibc dladdr() implementation that it could resolve 1562 // to wrong library name if the .so file has a base address != NULL. Here 1563 // we iterate through the program headers of all loaded libraries to find 1564 // out which library 'addr' really belongs to. This workaround can be 1565 // removed once the minimum requirement for glibc is moved to 2.3.x. 1566 data.addr = addr; 1567 data.fname = buf; 1568 data.buflen = buflen; 1569 data.base = NULL; 1570 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1571 1572 if (rslt) { 1573 // buf already contains library name 1574 if (offset) *offset = addr - data.base; 1575 return true; 1576 } 1577 if (dladdr((void*)addr, &dlinfo) != 0) { 1578 if (dlinfo.dli_fname != NULL) { 1579 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1580 } 1581 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1582 *offset = addr - (address)dlinfo.dli_fbase; 1583 } 1584 return true; 1585 } 1586 1587 buf[0] = '\0'; 1588 if (offset) *offset = -1; 1589 return false; 1590 } 1591 1592 // Loads .dll/.so and 1593 // in case of error it checks if .dll/.so was built for the 1594 // same architecture as Hotspot is running on 1595 1596 1597 // Remember the stack's state. The Linux dynamic linker will change 1598 // the stack to 'executable' at most once, so we must safepoint only once. 1599 bool os::Linux::_stack_is_executable = false; 1600 1601 // VM operation that loads a library. This is necessary if stack protection 1602 // of the Java stacks can be lost during loading the library. If we 1603 // do not stop the Java threads, they can stack overflow before the stacks 1604 // are protected again. 1605 class VM_LinuxDllLoad: public VM_Operation { 1606 private: 1607 const char *_filename; 1608 char *_ebuf; 1609 int _ebuflen; 1610 void *_lib; 1611 public: 1612 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1613 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1614 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1615 void doit() { 1616 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1617 os::Linux::_stack_is_executable = true; 1618 } 1619 void* loaded_library() { return _lib; } 1620 }; 1621 1622 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1623 void * result = NULL; 1624 bool load_attempted = false; 1625 1626 // Check whether the library to load might change execution rights 1627 // of the stack. If they are changed, the protection of the stack 1628 // guard pages will be lost. We need a safepoint to fix this. 1629 // 1630 // See Linux man page execstack(8) for more info. 1631 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1632 if (!ElfFile::specifies_noexecstack(filename)) { 1633 if (!is_init_completed()) { 1634 os::Linux::_stack_is_executable = true; 1635 // This is OK - No Java threads have been created yet, and hence no 1636 // stack guard pages to fix. 1637 // 1638 // This should happen only when you are building JDK7 using a very 1639 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1640 // 1641 // Dynamic loader will make all stacks executable after 1642 // this function returns, and will not do that again. 1643 #ifdef ASSERT 1644 ThreadsListHandle tlh; 1645 assert(tlh.length() == 0, "no Java threads should exist yet."); 1646 #endif 1647 } else { 1648 warning("You have loaded library %s which might have disabled stack guard. " 1649 "The VM will try to fix the stack guard now.\n" 1650 "It's highly recommended that you fix the library with " 1651 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1652 filename); 1653 1654 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1655 JavaThread *jt = JavaThread::current(); 1656 if (jt->thread_state() != _thread_in_native) { 1657 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1658 // that requires ExecStack. Cannot enter safe point. Let's give up. 1659 warning("Unable to fix stack guard. Giving up."); 1660 } else { 1661 if (!LoadExecStackDllInVMThread) { 1662 // This is for the case where the DLL has an static 1663 // constructor function that executes JNI code. We cannot 1664 // load such DLLs in the VMThread. 1665 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1666 } 1667 1668 ThreadInVMfromNative tiv(jt); 1669 debug_only(VMNativeEntryWrapper vew;) 1670 1671 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1672 VMThread::execute(&op); 1673 if (LoadExecStackDllInVMThread) { 1674 result = op.loaded_library(); 1675 } 1676 load_attempted = true; 1677 } 1678 } 1679 } 1680 } 1681 1682 if (!load_attempted) { 1683 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1684 } 1685 1686 if (result != NULL) { 1687 // Successful loading 1688 return result; 1689 } 1690 1691 Elf32_Ehdr elf_head; 1692 int diag_msg_max_length=ebuflen-strlen(ebuf); 1693 char* diag_msg_buf=ebuf+strlen(ebuf); 1694 1695 if (diag_msg_max_length==0) { 1696 // No more space in ebuf for additional diagnostics message 1697 return NULL; 1698 } 1699 1700 1701 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1702 1703 if (file_descriptor < 0) { 1704 // Can't open library, report dlerror() message 1705 return NULL; 1706 } 1707 1708 bool failed_to_read_elf_head= 1709 (sizeof(elf_head)!= 1710 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1711 1712 ::close(file_descriptor); 1713 if (failed_to_read_elf_head) { 1714 // file i/o error - report dlerror() msg 1715 return NULL; 1716 } 1717 1718 typedef struct { 1719 Elf32_Half code; // Actual value as defined in elf.h 1720 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1721 unsigned char elf_class; // 32 or 64 bit 1722 unsigned char endianess; // MSB or LSB 1723 char* name; // String representation 1724 } arch_t; 1725 1726 #ifndef EM_486 1727 #define EM_486 6 /* Intel 80486 */ 1728 #endif 1729 #ifndef EM_AARCH64 1730 #define EM_AARCH64 183 /* ARM AARCH64 */ 1731 #endif 1732 1733 static const arch_t arch_array[]={ 1734 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1735 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1736 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1737 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1738 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1739 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1740 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1741 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1742 #if defined(VM_LITTLE_ENDIAN) 1743 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"}, 1744 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"}, 1745 #else 1746 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1747 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"}, 1748 #endif 1749 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1750 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1751 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1752 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1753 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1754 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1755 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1756 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1757 }; 1758 1759 #if (defined IA32) 1760 static Elf32_Half running_arch_code=EM_386; 1761 #elif (defined AMD64) 1762 static Elf32_Half running_arch_code=EM_X86_64; 1763 #elif (defined IA64) 1764 static Elf32_Half running_arch_code=EM_IA_64; 1765 #elif (defined __sparc) && (defined _LP64) 1766 static Elf32_Half running_arch_code=EM_SPARCV9; 1767 #elif (defined __sparc) && (!defined _LP64) 1768 static Elf32_Half running_arch_code=EM_SPARC; 1769 #elif (defined __powerpc64__) 1770 static Elf32_Half running_arch_code=EM_PPC64; 1771 #elif (defined __powerpc__) 1772 static Elf32_Half running_arch_code=EM_PPC; 1773 #elif (defined AARCH64) 1774 static Elf32_Half running_arch_code=EM_AARCH64; 1775 #elif (defined ARM) 1776 static Elf32_Half running_arch_code=EM_ARM; 1777 #elif (defined S390) 1778 static Elf32_Half running_arch_code=EM_S390; 1779 #elif (defined ALPHA) 1780 static Elf32_Half running_arch_code=EM_ALPHA; 1781 #elif (defined MIPSEL) 1782 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1783 #elif (defined PARISC) 1784 static Elf32_Half running_arch_code=EM_PARISC; 1785 #elif (defined MIPS) 1786 static Elf32_Half running_arch_code=EM_MIPS; 1787 #elif (defined M68K) 1788 static Elf32_Half running_arch_code=EM_68K; 1789 #elif (defined SH) 1790 static Elf32_Half running_arch_code=EM_SH; 1791 #else 1792 #error Method os::dll_load requires that one of following is defined:\ 1793 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc 1794 #endif 1795 1796 // Identify compatability class for VM's architecture and library's architecture 1797 // Obtain string descriptions for architectures 1798 1799 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1800 int running_arch_index=-1; 1801 1802 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1803 if (running_arch_code == arch_array[i].code) { 1804 running_arch_index = i; 1805 } 1806 if (lib_arch.code == arch_array[i].code) { 1807 lib_arch.compat_class = arch_array[i].compat_class; 1808 lib_arch.name = arch_array[i].name; 1809 } 1810 } 1811 1812 assert(running_arch_index != -1, 1813 "Didn't find running architecture code (running_arch_code) in arch_array"); 1814 if (running_arch_index == -1) { 1815 // Even though running architecture detection failed 1816 // we may still continue with reporting dlerror() message 1817 return NULL; 1818 } 1819 1820 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1821 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1822 return NULL; 1823 } 1824 1825 #ifndef S390 1826 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1827 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1828 return NULL; 1829 } 1830 #endif // !S390 1831 1832 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1833 if (lib_arch.name!=NULL) { 1834 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1835 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1836 lib_arch.name, arch_array[running_arch_index].name); 1837 } else { 1838 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1839 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1840 lib_arch.code, 1841 arch_array[running_arch_index].name); 1842 } 1843 } 1844 1845 return NULL; 1846 } 1847 1848 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 1849 int ebuflen) { 1850 void * result = ::dlopen(filename, RTLD_LAZY); 1851 if (result == NULL) { 1852 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 1853 ebuf[ebuflen-1] = '\0'; 1854 } 1855 return result; 1856 } 1857 1858 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 1859 int ebuflen) { 1860 void * result = NULL; 1861 if (LoadExecStackDllInVMThread) { 1862 result = dlopen_helper(filename, ebuf, ebuflen); 1863 } 1864 1865 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 1866 // library that requires an executable stack, or which does not have this 1867 // stack attribute set, dlopen changes the stack attribute to executable. The 1868 // read protection of the guard pages gets lost. 1869 // 1870 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 1871 // may have been queued at the same time. 1872 1873 if (!_stack_is_executable) { 1874 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) { 1875 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 1876 jt->stack_guards_enabled()) { // No pending stack overflow exceptions 1877 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) { 1878 warning("Attempt to reguard stack yellow zone failed."); 1879 } 1880 } 1881 } 1882 } 1883 1884 return result; 1885 } 1886 1887 void* os::dll_lookup(void* handle, const char* name) { 1888 void* res = dlsym(handle, name); 1889 return res; 1890 } 1891 1892 void* os::get_default_process_handle() { 1893 return (void*)::dlopen(NULL, RTLD_LAZY); 1894 } 1895 1896 static bool _print_ascii_file(const char* filename, outputStream* st) { 1897 int fd = ::open(filename, O_RDONLY); 1898 if (fd == -1) { 1899 return false; 1900 } 1901 1902 char buf[33]; 1903 int bytes; 1904 buf[32] = '\0'; 1905 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) { 1906 st->print_raw(buf, bytes); 1907 } 1908 1909 ::close(fd); 1910 1911 return true; 1912 } 1913 1914 void os::print_dll_info(outputStream *st) { 1915 st->print_cr("Dynamic libraries:"); 1916 1917 char fname[32]; 1918 pid_t pid = os::Linux::gettid(); 1919 1920 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1921 1922 if (!_print_ascii_file(fname, st)) { 1923 st->print("Can not get library information for pid = %d\n", pid); 1924 } 1925 } 1926 1927 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 1928 FILE *procmapsFile = NULL; 1929 1930 // Open the procfs maps file for the current process 1931 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 1932 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 1933 char line[PATH_MAX + 100]; 1934 1935 // Read line by line from 'file' 1936 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 1937 u8 base, top, offset, inode; 1938 char permissions[5]; 1939 char device[6]; 1940 char name[PATH_MAX + 1]; 1941 1942 // Parse fields from line 1943 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s", 1944 &base, &top, permissions, &offset, device, &inode, name); 1945 1946 // Filter by device id '00:00' so that we only get file system mapped files. 1947 if (strcmp(device, "00:00") != 0) { 1948 1949 // Call callback with the fields of interest 1950 if(callback(name, (address)base, (address)top, param)) { 1951 // Oops abort, callback aborted 1952 fclose(procmapsFile); 1953 return 1; 1954 } 1955 } 1956 } 1957 fclose(procmapsFile); 1958 } 1959 return 0; 1960 } 1961 1962 void os::print_os_info_brief(outputStream* st) { 1963 os::Linux::print_distro_info(st); 1964 1965 os::Posix::print_uname_info(st); 1966 1967 os::Linux::print_libversion_info(st); 1968 1969 } 1970 1971 void os::print_os_info(outputStream* st) { 1972 st->print("OS:"); 1973 1974 os::Linux::print_distro_info(st); 1975 1976 os::Posix::print_uname_info(st); 1977 1978 // Print warning if unsafe chroot environment detected 1979 if (unsafe_chroot_detected) { 1980 st->print("WARNING!! "); 1981 st->print_cr("%s", unstable_chroot_error); 1982 } 1983 1984 os::Linux::print_libversion_info(st); 1985 1986 os::Posix::print_rlimit_info(st); 1987 1988 os::Posix::print_load_average(st); 1989 1990 os::Linux::print_full_memory_info(st); 1991 1992 os::Linux::print_container_info(st); 1993 } 1994 1995 // Try to identify popular distros. 1996 // Most Linux distributions have a /etc/XXX-release file, which contains 1997 // the OS version string. Newer Linux distributions have a /etc/lsb-release 1998 // file that also contains the OS version string. Some have more than one 1999 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2000 // /etc/redhat-release.), so the order is important. 2001 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2002 // their own specific XXX-release file as well as a redhat-release file. 2003 // Because of this the XXX-release file needs to be searched for before the 2004 // redhat-release file. 2005 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the 2006 // search for redhat-release / SuSE-release needs to be before lsb-release. 2007 // Since the lsb-release file is the new standard it needs to be searched 2008 // before the older style release files. 2009 // Searching system-release (Red Hat) and os-release (other Linuxes) are a 2010 // next to last resort. The os-release file is a new standard that contains 2011 // distribution information and the system-release file seems to be an old 2012 // standard that has been replaced by the lsb-release and os-release files. 2013 // Searching for the debian_version file is the last resort. It contains 2014 // an informative string like "6.0.6" or "wheezy/sid". Because of this 2015 // "Debian " is printed before the contents of the debian_version file. 2016 2017 const char* distro_files[] = { 2018 "/etc/oracle-release", 2019 "/etc/mandriva-release", 2020 "/etc/mandrake-release", 2021 "/etc/sun-release", 2022 "/etc/redhat-release", 2023 "/etc/SuSE-release", 2024 "/etc/lsb-release", 2025 "/etc/turbolinux-release", 2026 "/etc/gentoo-release", 2027 "/etc/ltib-release", 2028 "/etc/angstrom-version", 2029 "/etc/system-release", 2030 "/etc/os-release", 2031 NULL }; 2032 2033 void os::Linux::print_distro_info(outputStream* st) { 2034 for (int i = 0;; i++) { 2035 const char* file = distro_files[i]; 2036 if (file == NULL) { 2037 break; // done 2038 } 2039 // If file prints, we found it. 2040 if (_print_ascii_file(file, st)) { 2041 return; 2042 } 2043 } 2044 2045 if (file_exists("/etc/debian_version")) { 2046 st->print("Debian "); 2047 _print_ascii_file("/etc/debian_version", st); 2048 } else { 2049 st->print("Linux"); 2050 } 2051 st->cr(); 2052 } 2053 2054 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) { 2055 char buf[256]; 2056 while (fgets(buf, sizeof(buf), fp)) { 2057 // Edit out extra stuff in expected format 2058 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) { 2059 char* ptr = strstr(buf, "\""); // the name is in quotes 2060 if (ptr != NULL) { 2061 ptr++; // go beyond first quote 2062 char* nl = strchr(ptr, '\"'); 2063 if (nl != NULL) *nl = '\0'; 2064 strncpy(distro, ptr, length); 2065 } else { 2066 ptr = strstr(buf, "="); 2067 ptr++; // go beyond equals then 2068 char* nl = strchr(ptr, '\n'); 2069 if (nl != NULL) *nl = '\0'; 2070 strncpy(distro, ptr, length); 2071 } 2072 return; 2073 } else if (get_first_line) { 2074 char* nl = strchr(buf, '\n'); 2075 if (nl != NULL) *nl = '\0'; 2076 strncpy(distro, buf, length); 2077 return; 2078 } 2079 } 2080 // print last line and close 2081 char* nl = strchr(buf, '\n'); 2082 if (nl != NULL) *nl = '\0'; 2083 strncpy(distro, buf, length); 2084 } 2085 2086 static void parse_os_info(char* distro, size_t length, const char* file) { 2087 FILE* fp = fopen(file, "r"); 2088 if (fp != NULL) { 2089 // if suse format, print out first line 2090 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0); 2091 parse_os_info_helper(fp, distro, length, get_first_line); 2092 fclose(fp); 2093 } 2094 } 2095 2096 void os::get_summary_os_info(char* buf, size_t buflen) { 2097 for (int i = 0;; i++) { 2098 const char* file = distro_files[i]; 2099 if (file == NULL) { 2100 break; // ran out of distro_files 2101 } 2102 if (file_exists(file)) { 2103 parse_os_info(buf, buflen, file); 2104 return; 2105 } 2106 } 2107 // special case for debian 2108 if (file_exists("/etc/debian_version")) { 2109 strncpy(buf, "Debian ", buflen); 2110 parse_os_info(&buf[7], buflen-7, "/etc/debian_version"); 2111 } else { 2112 strncpy(buf, "Linux", buflen); 2113 } 2114 } 2115 2116 void os::Linux::print_libversion_info(outputStream* st) { 2117 // libc, pthread 2118 st->print("libc:"); 2119 st->print("%s ", os::Linux::glibc_version()); 2120 st->print("%s ", os::Linux::libpthread_version()); 2121 st->cr(); 2122 } 2123 2124 void os::Linux::print_full_memory_info(outputStream* st) { 2125 st->print("\n/proc/meminfo:\n"); 2126 _print_ascii_file("/proc/meminfo", st); 2127 st->cr(); 2128 } 2129 2130 void os::Linux::print_container_info(outputStream* st) { 2131 if (!OSContainer::is_containerized()) { 2132 return; 2133 } 2134 2135 st->print("container (cgroup) information:\n"); 2136 2137 const char *p_ct = OSContainer::container_type(); 2138 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed"); 2139 2140 char *p = OSContainer::cpu_cpuset_cpus(); 2141 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed"); 2142 free(p); 2143 2144 p = OSContainer::cpu_cpuset_memory_nodes(); 2145 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed"); 2146 free(p); 2147 2148 int i = OSContainer::active_processor_count(); 2149 if (i > 0) { 2150 st->print("active_processor_count: %d\n", i); 2151 } else { 2152 st->print("active_processor_count: failed\n"); 2153 } 2154 2155 i = OSContainer::cpu_quota(); 2156 st->print("cpu_quota: %d\n", i); 2157 2158 i = OSContainer::cpu_period(); 2159 st->print("cpu_period: %d\n", i); 2160 2161 i = OSContainer::cpu_shares(); 2162 st->print("cpu_shares: %d\n", i); 2163 2164 jlong j = OSContainer::memory_limit_in_bytes(); 2165 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j); 2166 2167 j = OSContainer::memory_and_swap_limit_in_bytes(); 2168 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j); 2169 2170 j = OSContainer::memory_soft_limit_in_bytes(); 2171 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j); 2172 2173 j = OSContainer::OSContainer::memory_usage_in_bytes(); 2174 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j); 2175 2176 j = OSContainer::OSContainer::memory_max_usage_in_bytes(); 2177 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j); 2178 st->cr(); 2179 } 2180 2181 void os::print_memory_info(outputStream* st) { 2182 2183 st->print("Memory:"); 2184 st->print(" %dk page", os::vm_page_size()>>10); 2185 2186 // values in struct sysinfo are "unsigned long" 2187 struct sysinfo si; 2188 sysinfo(&si); 2189 2190 st->print(", physical " UINT64_FORMAT "k", 2191 os::physical_memory() >> 10); 2192 st->print("(" UINT64_FORMAT "k free)", 2193 os::available_memory() >> 10); 2194 st->print(", swap " UINT64_FORMAT "k", 2195 ((jlong)si.totalswap * si.mem_unit) >> 10); 2196 st->print("(" UINT64_FORMAT "k free)", 2197 ((jlong)si.freeswap * si.mem_unit) >> 10); 2198 st->cr(); 2199 } 2200 2201 // Print the first "model name" line and the first "flags" line 2202 // that we find and nothing more. We assume "model name" comes 2203 // before "flags" so if we find a second "model name", then the 2204 // "flags" field is considered missing. 2205 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) { 2206 #if defined(IA32) || defined(AMD64) 2207 // Other platforms have less repetitive cpuinfo files 2208 FILE *fp = fopen("/proc/cpuinfo", "r"); 2209 if (fp) { 2210 while (!feof(fp)) { 2211 if (fgets(buf, buflen, fp)) { 2212 // Assume model name comes before flags 2213 bool model_name_printed = false; 2214 if (strstr(buf, "model name") != NULL) { 2215 if (!model_name_printed) { 2216 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n"); 2217 st->print_raw(buf); 2218 model_name_printed = true; 2219 } else { 2220 // model name printed but not flags? Odd, just return 2221 fclose(fp); 2222 return true; 2223 } 2224 } 2225 // print the flags line too 2226 if (strstr(buf, "flags") != NULL) { 2227 st->print_raw(buf); 2228 fclose(fp); 2229 return true; 2230 } 2231 } 2232 } 2233 fclose(fp); 2234 } 2235 #endif // x86 platforms 2236 return false; 2237 } 2238 2239 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) { 2240 // Only print the model name if the platform provides this as a summary 2241 if (!print_model_name_and_flags(st, buf, buflen)) { 2242 st->print("\n/proc/cpuinfo:\n"); 2243 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2244 st->print_cr(" <Not Available>"); 2245 } 2246 } 2247 } 2248 2249 #if defined(AMD64) || defined(IA32) || defined(X32) 2250 const char* search_string = "model name"; 2251 #elif defined(M68K) 2252 const char* search_string = "CPU"; 2253 #elif defined(PPC64) 2254 const char* search_string = "cpu"; 2255 #elif defined(S390) 2256 const char* search_string = "processor"; 2257 #elif defined(SPARC) 2258 const char* search_string = "cpu"; 2259 #else 2260 const char* search_string = "Processor"; 2261 #endif 2262 2263 // Parses the cpuinfo file for string representing the model name. 2264 void os::get_summary_cpu_info(char* cpuinfo, size_t length) { 2265 FILE* fp = fopen("/proc/cpuinfo", "r"); 2266 if (fp != NULL) { 2267 while (!feof(fp)) { 2268 char buf[256]; 2269 if (fgets(buf, sizeof(buf), fp)) { 2270 char* start = strstr(buf, search_string); 2271 if (start != NULL) { 2272 char *ptr = start + strlen(search_string); 2273 char *end = buf + strlen(buf); 2274 while (ptr != end) { 2275 // skip whitespace and colon for the rest of the name. 2276 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') { 2277 break; 2278 } 2279 ptr++; 2280 } 2281 if (ptr != end) { 2282 // reasonable string, get rid of newline and keep the rest 2283 char* nl = strchr(buf, '\n'); 2284 if (nl != NULL) *nl = '\0'; 2285 strncpy(cpuinfo, ptr, length); 2286 fclose(fp); 2287 return; 2288 } 2289 } 2290 } 2291 } 2292 fclose(fp); 2293 } 2294 // cpuinfo not found or parsing failed, just print generic string. The entire 2295 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86) 2296 #if defined(AARCH64) 2297 strncpy(cpuinfo, "AArch64", length); 2298 #elif defined(AMD64) 2299 strncpy(cpuinfo, "x86_64", length); 2300 #elif defined(ARM) // Order wrt. AARCH64 is relevant! 2301 strncpy(cpuinfo, "ARM", length); 2302 #elif defined(IA32) 2303 strncpy(cpuinfo, "x86_32", length); 2304 #elif defined(IA64) 2305 strncpy(cpuinfo, "IA64", length); 2306 #elif defined(PPC) 2307 strncpy(cpuinfo, "PPC64", length); 2308 #elif defined(S390) 2309 strncpy(cpuinfo, "S390", length); 2310 #elif defined(SPARC) 2311 strncpy(cpuinfo, "sparcv9", length); 2312 #elif defined(ZERO_LIBARCH) 2313 strncpy(cpuinfo, ZERO_LIBARCH, length); 2314 #else 2315 strncpy(cpuinfo, "unknown", length); 2316 #endif 2317 } 2318 2319 static void print_signal_handler(outputStream* st, int sig, 2320 char* buf, size_t buflen); 2321 2322 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2323 st->print_cr("Signal Handlers:"); 2324 print_signal_handler(st, SIGSEGV, buf, buflen); 2325 print_signal_handler(st, SIGBUS , buf, buflen); 2326 print_signal_handler(st, SIGFPE , buf, buflen); 2327 print_signal_handler(st, SIGPIPE, buf, buflen); 2328 print_signal_handler(st, SIGXFSZ, buf, buflen); 2329 print_signal_handler(st, SIGILL , buf, buflen); 2330 print_signal_handler(st, SR_signum, buf, buflen); 2331 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2332 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2333 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2334 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2335 #if defined(PPC64) 2336 print_signal_handler(st, SIGTRAP, buf, buflen); 2337 #endif 2338 } 2339 2340 static char saved_jvm_path[MAXPATHLEN] = {0}; 2341 2342 // Find the full path to the current module, libjvm.so 2343 void os::jvm_path(char *buf, jint buflen) { 2344 // Error checking. 2345 if (buflen < MAXPATHLEN) { 2346 assert(false, "must use a large-enough buffer"); 2347 buf[0] = '\0'; 2348 return; 2349 } 2350 // Lazy resolve the path to current module. 2351 if (saved_jvm_path[0] != 0) { 2352 strcpy(buf, saved_jvm_path); 2353 return; 2354 } 2355 2356 char dli_fname[MAXPATHLEN]; 2357 bool ret = dll_address_to_library_name( 2358 CAST_FROM_FN_PTR(address, os::jvm_path), 2359 dli_fname, sizeof(dli_fname), NULL); 2360 assert(ret, "cannot locate libjvm"); 2361 char *rp = NULL; 2362 if (ret && dli_fname[0] != '\0') { 2363 rp = os::Posix::realpath(dli_fname, buf, buflen); 2364 } 2365 if (rp == NULL) { 2366 return; 2367 } 2368 2369 if (Arguments::sun_java_launcher_is_altjvm()) { 2370 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2371 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so". 2372 // If "/jre/lib/" appears at the right place in the string, then 2373 // assume we are installed in a JDK and we're done. Otherwise, check 2374 // for a JAVA_HOME environment variable and fix up the path so it 2375 // looks like libjvm.so is installed there (append a fake suffix 2376 // hotspot/libjvm.so). 2377 const char *p = buf + strlen(buf) - 1; 2378 for (int count = 0; p > buf && count < 5; ++count) { 2379 for (--p; p > buf && *p != '/'; --p) 2380 /* empty */ ; 2381 } 2382 2383 if (strncmp(p, "/jre/lib/", 9) != 0) { 2384 // Look for JAVA_HOME in the environment. 2385 char* java_home_var = ::getenv("JAVA_HOME"); 2386 if (java_home_var != NULL && java_home_var[0] != 0) { 2387 char* jrelib_p; 2388 int len; 2389 2390 // Check the current module name "libjvm.so". 2391 p = strrchr(buf, '/'); 2392 if (p == NULL) { 2393 return; 2394 } 2395 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2396 2397 rp = os::Posix::realpath(java_home_var, buf, buflen); 2398 if (rp == NULL) { 2399 return; 2400 } 2401 2402 // determine if this is a legacy image or modules image 2403 // modules image doesn't have "jre" subdirectory 2404 len = strlen(buf); 2405 assert(len < buflen, "Ran out of buffer room"); 2406 jrelib_p = buf + len; 2407 snprintf(jrelib_p, buflen-len, "/jre/lib"); 2408 if (0 != access(buf, F_OK)) { 2409 snprintf(jrelib_p, buflen-len, "/lib"); 2410 } 2411 2412 if (0 == access(buf, F_OK)) { 2413 // Use current module name "libjvm.so" 2414 len = strlen(buf); 2415 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2416 } else { 2417 // Go back to path of .so 2418 rp = os::Posix::realpath(dli_fname, buf, buflen); 2419 if (rp == NULL) { 2420 return; 2421 } 2422 } 2423 } 2424 } 2425 } 2426 2427 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2428 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2429 } 2430 2431 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2432 // no prefix required, not even "_" 2433 } 2434 2435 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2436 // no suffix required 2437 } 2438 2439 //////////////////////////////////////////////////////////////////////////////// 2440 // sun.misc.Signal support 2441 2442 static volatile jint sigint_count = 0; 2443 2444 static void UserHandler(int sig, void *siginfo, void *context) { 2445 // 4511530 - sem_post is serialized and handled by the manager thread. When 2446 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2447 // don't want to flood the manager thread with sem_post requests. 2448 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2449 return; 2450 } 2451 2452 // Ctrl-C is pressed during error reporting, likely because the error 2453 // handler fails to abort. Let VM die immediately. 2454 if (sig == SIGINT && VMError::is_error_reported()) { 2455 os::die(); 2456 } 2457 2458 os::signal_notify(sig); 2459 } 2460 2461 void* os::user_handler() { 2462 return CAST_FROM_FN_PTR(void*, UserHandler); 2463 } 2464 2465 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) { 2466 struct timespec ts; 2467 // Semaphore's are always associated with CLOCK_REALTIME 2468 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2469 // see os_posix.cpp for discussion on overflow checking 2470 if (sec >= MAX_SECS) { 2471 ts.tv_sec += MAX_SECS; 2472 ts.tv_nsec = 0; 2473 } else { 2474 ts.tv_sec += sec; 2475 ts.tv_nsec += nsec; 2476 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2477 ts.tv_nsec -= NANOSECS_PER_SEC; 2478 ++ts.tv_sec; // note: this must be <= max_secs 2479 } 2480 } 2481 2482 return ts; 2483 } 2484 2485 extern "C" { 2486 typedef void (*sa_handler_t)(int); 2487 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2488 } 2489 2490 void* os::signal(int signal_number, void* handler) { 2491 struct sigaction sigAct, oldSigAct; 2492 2493 sigfillset(&(sigAct.sa_mask)); 2494 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2495 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2496 2497 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2498 // -1 means registration failed 2499 return (void *)-1; 2500 } 2501 2502 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2503 } 2504 2505 void os::signal_raise(int signal_number) { 2506 ::raise(signal_number); 2507 } 2508 2509 // The following code is moved from os.cpp for making this 2510 // code platform specific, which it is by its very nature. 2511 2512 // Will be modified when max signal is changed to be dynamic 2513 int os::sigexitnum_pd() { 2514 return NSIG; 2515 } 2516 2517 // a counter for each possible signal value 2518 static volatile jint pending_signals[NSIG+1] = { 0 }; 2519 2520 // Linux(POSIX) specific hand shaking semaphore. 2521 static Semaphore* sig_sem = NULL; 2522 static PosixSemaphore sr_semaphore; 2523 2524 void os::signal_init_pd() { 2525 // Initialize signal structures 2526 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2527 2528 // Initialize signal semaphore 2529 sig_sem = new Semaphore(); 2530 } 2531 2532 void os::signal_notify(int sig) { 2533 if (sig_sem != NULL) { 2534 Atomic::inc(&pending_signals[sig]); 2535 sig_sem->signal(); 2536 } else { 2537 // Signal thread is not created with ReduceSignalUsage and signal_init_pd 2538 // initialization isn't called. 2539 assert(ReduceSignalUsage, "signal semaphore should be created"); 2540 } 2541 } 2542 2543 static int check_pending_signals() { 2544 Atomic::store(0, &sigint_count); 2545 for (;;) { 2546 for (int i = 0; i < NSIG + 1; i++) { 2547 jint n = pending_signals[i]; 2548 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2549 return i; 2550 } 2551 } 2552 JavaThread *thread = JavaThread::current(); 2553 ThreadBlockInVM tbivm(thread); 2554 2555 bool threadIsSuspended; 2556 do { 2557 thread->set_suspend_equivalent(); 2558 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2559 sig_sem->wait(); 2560 2561 // were we externally suspended while we were waiting? 2562 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2563 if (threadIsSuspended) { 2564 // The semaphore has been incremented, but while we were waiting 2565 // another thread suspended us. We don't want to continue running 2566 // while suspended because that would surprise the thread that 2567 // suspended us. 2568 sig_sem->signal(); 2569 2570 thread->java_suspend_self(); 2571 } 2572 } while (threadIsSuspended); 2573 } 2574 } 2575 2576 int os::signal_wait() { 2577 return check_pending_signals(); 2578 } 2579 2580 //////////////////////////////////////////////////////////////////////////////// 2581 // Virtual Memory 2582 2583 int os::vm_page_size() { 2584 // Seems redundant as all get out 2585 assert(os::Linux::page_size() != -1, "must call os::init"); 2586 return os::Linux::page_size(); 2587 } 2588 2589 // Solaris allocates memory by pages. 2590 int os::vm_allocation_granularity() { 2591 assert(os::Linux::page_size() != -1, "must call os::init"); 2592 return os::Linux::page_size(); 2593 } 2594 2595 // Rationale behind this function: 2596 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2597 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2598 // samples for JITted code. Here we create private executable mapping over the code cache 2599 // and then we can use standard (well, almost, as mapping can change) way to provide 2600 // info for the reporting script by storing timestamp and location of symbol 2601 void linux_wrap_code(char* base, size_t size) { 2602 static volatile jint cnt = 0; 2603 2604 if (!UseOprofile) { 2605 return; 2606 } 2607 2608 char buf[PATH_MAX+1]; 2609 int num = Atomic::add(1, &cnt); 2610 2611 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2612 os::get_temp_directory(), os::current_process_id(), num); 2613 unlink(buf); 2614 2615 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2616 2617 if (fd != -1) { 2618 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2619 if (rv != (off_t)-1) { 2620 if (::write(fd, "", 1) == 1) { 2621 mmap(base, size, 2622 PROT_READ|PROT_WRITE|PROT_EXEC, 2623 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2624 } 2625 } 2626 ::close(fd); 2627 unlink(buf); 2628 } 2629 } 2630 2631 static bool recoverable_mmap_error(int err) { 2632 // See if the error is one we can let the caller handle. This 2633 // list of errno values comes from JBS-6843484. I can't find a 2634 // Linux man page that documents this specific set of errno 2635 // values so while this list currently matches Solaris, it may 2636 // change as we gain experience with this failure mode. 2637 switch (err) { 2638 case EBADF: 2639 case EINVAL: 2640 case ENOTSUP: 2641 // let the caller deal with these errors 2642 return true; 2643 2644 default: 2645 // Any remaining errors on this OS can cause our reserved mapping 2646 // to be lost. That can cause confusion where different data 2647 // structures think they have the same memory mapped. The worst 2648 // scenario is if both the VM and a library think they have the 2649 // same memory mapped. 2650 return false; 2651 } 2652 } 2653 2654 static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2655 int err) { 2656 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2657 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec, 2658 os::strerror(err), err); 2659 } 2660 2661 static void warn_fail_commit_memory(char* addr, size_t size, 2662 size_t alignment_hint, bool exec, 2663 int err) { 2664 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2665 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, 2666 alignment_hint, exec, os::strerror(err), err); 2667 } 2668 2669 // NOTE: Linux kernel does not really reserve the pages for us. 2670 // All it does is to check if there are enough free pages 2671 // left at the time of mmap(). This could be a potential 2672 // problem. 2673 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2674 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2675 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2676 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2677 if (res != (uintptr_t) MAP_FAILED) { 2678 if (UseNUMAInterleaving) { 2679 numa_make_global(addr, size); 2680 } 2681 return 0; 2682 } 2683 2684 int err = errno; // save errno from mmap() call above 2685 2686 if (!recoverable_mmap_error(err)) { 2687 warn_fail_commit_memory(addr, size, exec, err); 2688 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2689 } 2690 2691 return err; 2692 } 2693 2694 bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2695 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2696 } 2697 2698 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2699 const char* mesg) { 2700 assert(mesg != NULL, "mesg must be specified"); 2701 int err = os::Linux::commit_memory_impl(addr, size, exec); 2702 if (err != 0) { 2703 // the caller wants all commit errors to exit with the specified mesg: 2704 warn_fail_commit_memory(addr, size, exec, err); 2705 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2706 } 2707 } 2708 2709 // Define MAP_HUGETLB here so we can build HotSpot on old systems. 2710 #ifndef MAP_HUGETLB 2711 #define MAP_HUGETLB 0x40000 2712 #endif 2713 2714 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2715 #ifndef MADV_HUGEPAGE 2716 #define MADV_HUGEPAGE 14 2717 #endif 2718 2719 int os::Linux::commit_memory_impl(char* addr, size_t size, 2720 size_t alignment_hint, bool exec) { 2721 int err = os::Linux::commit_memory_impl(addr, size, exec); 2722 if (err == 0) { 2723 realign_memory(addr, size, alignment_hint); 2724 } 2725 return err; 2726 } 2727 2728 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2729 bool exec) { 2730 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2731 } 2732 2733 void os::pd_commit_memory_or_exit(char* addr, size_t size, 2734 size_t alignment_hint, bool exec, 2735 const char* mesg) { 2736 assert(mesg != NULL, "mesg must be specified"); 2737 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2738 if (err != 0) { 2739 // the caller wants all commit errors to exit with the specified mesg: 2740 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2741 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2742 } 2743 } 2744 2745 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2746 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2747 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2748 // be supported or the memory may already be backed by huge pages. 2749 ::madvise(addr, bytes, MADV_HUGEPAGE); 2750 } 2751 } 2752 2753 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2754 // This method works by doing an mmap over an existing mmaping and effectively discarding 2755 // the existing pages. However it won't work for SHM-based large pages that cannot be 2756 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2757 // small pages on top of the SHM segment. This method always works for small pages, so we 2758 // allow that in any case. 2759 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2760 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2761 } 2762 } 2763 2764 void os::numa_make_global(char *addr, size_t bytes) { 2765 Linux::numa_interleave_memory(addr, bytes); 2766 } 2767 2768 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2769 // bind policy to MPOL_PREFERRED for the current thread. 2770 #define USE_MPOL_PREFERRED 0 2771 2772 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2773 // To make NUMA and large pages more robust when both enabled, we need to ease 2774 // the requirements on where the memory should be allocated. MPOL_BIND is the 2775 // default policy and it will force memory to be allocated on the specified 2776 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2777 // the specified node, but will not force it. Using this policy will prevent 2778 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2779 // free large pages. 2780 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2781 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2782 } 2783 2784 bool os::numa_topology_changed() { return false; } 2785 2786 size_t os::numa_get_groups_num() { 2787 // Return just the number of nodes in which it's possible to allocate memory 2788 // (in numa terminology, configured nodes). 2789 return Linux::numa_num_configured_nodes(); 2790 } 2791 2792 int os::numa_get_group_id() { 2793 int cpu_id = Linux::sched_getcpu(); 2794 if (cpu_id != -1) { 2795 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2796 if (lgrp_id != -1) { 2797 return lgrp_id; 2798 } 2799 } 2800 return 0; 2801 } 2802 2803 int os::Linux::get_existing_num_nodes() { 2804 size_t node; 2805 size_t highest_node_number = Linux::numa_max_node(); 2806 int num_nodes = 0; 2807 2808 // Get the total number of nodes in the system including nodes without memory. 2809 for (node = 0; node <= highest_node_number; node++) { 2810 if (isnode_in_existing_nodes(node)) { 2811 num_nodes++; 2812 } 2813 } 2814 return num_nodes; 2815 } 2816 2817 size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2818 size_t highest_node_number = Linux::numa_max_node(); 2819 size_t i = 0; 2820 2821 // Map all node ids in which is possible to allocate memory. Also nodes are 2822 // not always consecutively available, i.e. available from 0 to the highest 2823 // node number. 2824 for (size_t node = 0; node <= highest_node_number; node++) { 2825 if (Linux::isnode_in_configured_nodes(node)) { 2826 ids[i++] = node; 2827 } 2828 } 2829 return i; 2830 } 2831 2832 bool os::get_page_info(char *start, page_info* info) { 2833 return false; 2834 } 2835 2836 char *os::scan_pages(char *start, char* end, page_info* page_expected, 2837 page_info* page_found) { 2838 return end; 2839 } 2840 2841 2842 int os::Linux::sched_getcpu_syscall(void) { 2843 unsigned int cpu = 0; 2844 int retval = -1; 2845 2846 #if defined(IA32) 2847 #ifndef SYS_getcpu 2848 #define SYS_getcpu 318 2849 #endif 2850 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2851 #elif defined(AMD64) 2852 // Unfortunately we have to bring all these macros here from vsyscall.h 2853 // to be able to compile on old linuxes. 2854 #define __NR_vgetcpu 2 2855 #define VSYSCALL_START (-10UL << 20) 2856 #define VSYSCALL_SIZE 1024 2857 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2858 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2859 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2860 retval = vgetcpu(&cpu, NULL, NULL); 2861 #endif 2862 2863 return (retval == -1) ? retval : cpu; 2864 } 2865 2866 void os::Linux::sched_getcpu_init() { 2867 // sched_getcpu() should be in libc. 2868 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2869 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2870 2871 // If it's not, try a direct syscall. 2872 if (sched_getcpu() == -1) { 2873 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2874 (void*)&sched_getcpu_syscall)); 2875 } 2876 } 2877 2878 // Something to do with the numa-aware allocator needs these symbols 2879 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2880 extern "C" JNIEXPORT void numa_error(char *where) { } 2881 2882 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails 2883 // load symbol from base version instead. 2884 void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2885 void *f = dlvsym(handle, name, "libnuma_1.1"); 2886 if (f == NULL) { 2887 f = dlsym(handle, name); 2888 } 2889 return f; 2890 } 2891 2892 // Handle request to load libnuma symbol version 1.2 (API v2) only. 2893 // Return NULL if the symbol is not defined in this particular version. 2894 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) { 2895 return dlvsym(handle, name, "libnuma_1.2"); 2896 } 2897 2898 bool os::Linux::libnuma_init() { 2899 if (sched_getcpu() != -1) { // Requires sched_getcpu() support 2900 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2901 if (handle != NULL) { 2902 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2903 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2904 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2905 libnuma_dlsym(handle, "numa_max_node"))); 2906 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t, 2907 libnuma_dlsym(handle, "numa_num_configured_nodes"))); 2908 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2909 libnuma_dlsym(handle, "numa_available"))); 2910 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2911 libnuma_dlsym(handle, "numa_tonode_memory"))); 2912 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2913 libnuma_dlsym(handle, "numa_interleave_memory"))); 2914 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t, 2915 libnuma_v2_dlsym(handle, "numa_interleave_memory"))); 2916 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2917 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2918 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t, 2919 libnuma_dlsym(handle, "numa_bitmask_isbitset"))); 2920 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t, 2921 libnuma_dlsym(handle, "numa_distance"))); 2922 2923 if (numa_available() != -1) { 2924 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2925 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr")); 2926 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr")); 2927 // Create an index -> node mapping, since nodes are not always consecutive 2928 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2929 rebuild_nindex_to_node_map(); 2930 // Create a cpu -> node mapping 2931 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2932 rebuild_cpu_to_node_map(); 2933 return true; 2934 } 2935 } 2936 } 2937 return false; 2938 } 2939 2940 size_t os::Linux::default_guard_size(os::ThreadType thr_type) { 2941 // Creating guard page is very expensive. Java thread has HotSpot 2942 // guard pages, only enable glibc guard page for non-Java threads. 2943 // (Remember: compiler thread is a Java thread, too!) 2944 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size()); 2945 } 2946 2947 void os::Linux::rebuild_nindex_to_node_map() { 2948 int highest_node_number = Linux::numa_max_node(); 2949 2950 nindex_to_node()->clear(); 2951 for (int node = 0; node <= highest_node_number; node++) { 2952 if (Linux::isnode_in_existing_nodes(node)) { 2953 nindex_to_node()->append(node); 2954 } 2955 } 2956 } 2957 2958 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2959 // The table is later used in get_node_by_cpu(). 2960 void os::Linux::rebuild_cpu_to_node_map() { 2961 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2962 // in libnuma (possible values are starting from 16, 2963 // and continuing up with every other power of 2, but less 2964 // than the maximum number of CPUs supported by kernel), and 2965 // is a subject to change (in libnuma version 2 the requirements 2966 // are more reasonable) we'll just hardcode the number they use 2967 // in the library. 2968 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2969 2970 size_t cpu_num = processor_count(); 2971 size_t cpu_map_size = NCPUS / BitsPerCLong; 2972 size_t cpu_map_valid_size = 2973 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2974 2975 cpu_to_node()->clear(); 2976 cpu_to_node()->at_grow(cpu_num - 1); 2977 2978 size_t node_num = get_existing_num_nodes(); 2979 2980 int distance = 0; 2981 int closest_distance = INT_MAX; 2982 int closest_node = 0; 2983 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2984 for (size_t i = 0; i < node_num; i++) { 2985 // Check if node is configured (not a memory-less node). If it is not, find 2986 // the closest configured node. 2987 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) { 2988 closest_distance = INT_MAX; 2989 // Check distance from all remaining nodes in the system. Ignore distance 2990 // from itself and from another non-configured node. 2991 for (size_t m = 0; m < node_num; m++) { 2992 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) { 2993 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m)); 2994 // If a closest node is found, update. There is always at least one 2995 // configured node in the system so there is always at least one node 2996 // close. 2997 if (distance != 0 && distance < closest_distance) { 2998 closest_distance = distance; 2999 closest_node = nindex_to_node()->at(m); 3000 } 3001 } 3002 } 3003 } else { 3004 // Current node is already a configured node. 3005 closest_node = nindex_to_node()->at(i); 3006 } 3007 3008 // Get cpus from the original node and map them to the closest node. If node 3009 // is a configured node (not a memory-less node), then original node and 3010 // closest node are the same. 3011 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 3012 for (size_t j = 0; j < cpu_map_valid_size; j++) { 3013 if (cpu_map[j] != 0) { 3014 for (size_t k = 0; k < BitsPerCLong; k++) { 3015 if (cpu_map[j] & (1UL << k)) { 3016 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node); 3017 } 3018 } 3019 } 3020 } 3021 } 3022 } 3023 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 3024 } 3025 3026 int os::Linux::get_node_by_cpu(int cpu_id) { 3027 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 3028 return cpu_to_node()->at(cpu_id); 3029 } 3030 return -1; 3031 } 3032 3033 GrowableArray<int>* os::Linux::_cpu_to_node; 3034 GrowableArray<int>* os::Linux::_nindex_to_node; 3035 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 3036 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 3037 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 3038 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes; 3039 os::Linux::numa_available_func_t os::Linux::_numa_available; 3040 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 3041 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 3042 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2; 3043 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 3044 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset; 3045 os::Linux::numa_distance_func_t os::Linux::_numa_distance; 3046 unsigned long* os::Linux::_numa_all_nodes; 3047 struct bitmask* os::Linux::_numa_all_nodes_ptr; 3048 struct bitmask* os::Linux::_numa_nodes_ptr; 3049 3050 bool os::pd_uncommit_memory(char* addr, size_t size) { 3051 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 3052 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 3053 return res != (uintptr_t) MAP_FAILED; 3054 } 3055 3056 static address get_stack_commited_bottom(address bottom, size_t size) { 3057 address nbot = bottom; 3058 address ntop = bottom + size; 3059 3060 size_t page_sz = os::vm_page_size(); 3061 unsigned pages = size / page_sz; 3062 3063 unsigned char vec[1]; 3064 unsigned imin = 1, imax = pages + 1, imid; 3065 int mincore_return_value = 0; 3066 3067 assert(imin <= imax, "Unexpected page size"); 3068 3069 while (imin < imax) { 3070 imid = (imax + imin) / 2; 3071 nbot = ntop - (imid * page_sz); 3072 3073 // Use a trick with mincore to check whether the page is mapped or not. 3074 // mincore sets vec to 1 if page resides in memory and to 0 if page 3075 // is swapped output but if page we are asking for is unmapped 3076 // it returns -1,ENOMEM 3077 mincore_return_value = mincore(nbot, page_sz, vec); 3078 3079 if (mincore_return_value == -1) { 3080 // Page is not mapped go up 3081 // to find first mapped page 3082 if (errno != EAGAIN) { 3083 assert(errno == ENOMEM, "Unexpected mincore errno"); 3084 imax = imid; 3085 } 3086 } else { 3087 // Page is mapped go down 3088 // to find first not mapped page 3089 imin = imid + 1; 3090 } 3091 } 3092 3093 nbot = nbot + page_sz; 3094 3095 // Adjust stack bottom one page up if last checked page is not mapped 3096 if (mincore_return_value == -1) { 3097 nbot = nbot + page_sz; 3098 } 3099 3100 return nbot; 3101 } 3102 3103 3104 // Linux uses a growable mapping for the stack, and if the mapping for 3105 // the stack guard pages is not removed when we detach a thread the 3106 // stack cannot grow beyond the pages where the stack guard was 3107 // mapped. If at some point later in the process the stack expands to 3108 // that point, the Linux kernel cannot expand the stack any further 3109 // because the guard pages are in the way, and a segfault occurs. 3110 // 3111 // However, it's essential not to split the stack region by unmapping 3112 // a region (leaving a hole) that's already part of the stack mapping, 3113 // so if the stack mapping has already grown beyond the guard pages at 3114 // the time we create them, we have to truncate the stack mapping. 3115 // So, we need to know the extent of the stack mapping when 3116 // create_stack_guard_pages() is called. 3117 3118 // We only need this for stacks that are growable: at the time of 3119 // writing thread stacks don't use growable mappings (i.e. those 3120 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3121 // only applies to the main thread. 3122 3123 // If the (growable) stack mapping already extends beyond the point 3124 // where we're going to put our guard pages, truncate the mapping at 3125 // that point by munmap()ping it. This ensures that when we later 3126 // munmap() the guard pages we don't leave a hole in the stack 3127 // mapping. This only affects the main/primordial thread 3128 3129 bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3130 if (os::is_primordial_thread()) { 3131 // As we manually grow stack up to bottom inside create_attached_thread(), 3132 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3133 // we don't need to do anything special. 3134 // Check it first, before calling heavy function. 3135 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3136 unsigned char vec[1]; 3137 3138 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3139 // Fallback to slow path on all errors, including EAGAIN 3140 stack_extent = (uintptr_t) get_stack_commited_bottom( 3141 os::Linux::initial_thread_stack_bottom(), 3142 (size_t)addr - stack_extent); 3143 } 3144 3145 if (stack_extent < (uintptr_t)addr) { 3146 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3147 } 3148 } 3149 3150 return os::commit_memory(addr, size, !ExecMem); 3151 } 3152 3153 // If this is a growable mapping, remove the guard pages entirely by 3154 // munmap()ping them. If not, just call uncommit_memory(). This only 3155 // affects the main/primordial thread, but guard against future OS changes. 3156 // It's safe to always unmap guard pages for primordial thread because we 3157 // always place it right after end of the mapped region. 3158 3159 bool os::remove_stack_guard_pages(char* addr, size_t size) { 3160 uintptr_t stack_extent, stack_base; 3161 3162 if (os::is_primordial_thread()) { 3163 return ::munmap(addr, size) == 0; 3164 } 3165 3166 return os::uncommit_memory(addr, size); 3167 } 3168 3169 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3170 // at 'requested_addr'. If there are existing memory mappings at the same 3171 // location, however, they will be overwritten. If 'fixed' is false, 3172 // 'requested_addr' is only treated as a hint, the return value may or 3173 // may not start from the requested address. Unlike Linux mmap(), this 3174 // function returns NULL to indicate failure. 3175 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3176 char * addr; 3177 int flags; 3178 3179 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3180 if (fixed) { 3181 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3182 flags |= MAP_FIXED; 3183 } 3184 3185 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3186 // touch an uncommitted page. Otherwise, the read/write might 3187 // succeed if we have enough swap space to back the physical page. 3188 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3189 flags, -1, 0); 3190 3191 return addr == MAP_FAILED ? NULL : addr; 3192 } 3193 3194 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address 3195 // (req_addr != NULL) or with a given alignment. 3196 // - bytes shall be a multiple of alignment. 3197 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3198 // - alignment sets the alignment at which memory shall be allocated. 3199 // It must be a multiple of allocation granularity. 3200 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3201 // req_addr or NULL. 3202 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) { 3203 3204 size_t extra_size = bytes; 3205 if (req_addr == NULL && alignment > 0) { 3206 extra_size += alignment; 3207 } 3208 3209 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE, 3210 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 3211 -1, 0); 3212 if (start == MAP_FAILED) { 3213 start = NULL; 3214 } else { 3215 if (req_addr != NULL) { 3216 if (start != req_addr) { 3217 ::munmap(start, extra_size); 3218 start = NULL; 3219 } 3220 } else { 3221 char* const start_aligned = align_up(start, alignment); 3222 char* const end_aligned = start_aligned + bytes; 3223 char* const end = start + extra_size; 3224 if (start_aligned > start) { 3225 ::munmap(start, start_aligned - start); 3226 } 3227 if (end_aligned < end) { 3228 ::munmap(end_aligned, end - end_aligned); 3229 } 3230 start = start_aligned; 3231 } 3232 } 3233 return start; 3234 } 3235 3236 static int anon_munmap(char * addr, size_t size) { 3237 return ::munmap(addr, size) == 0; 3238 } 3239 3240 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3241 size_t alignment_hint) { 3242 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3243 } 3244 3245 bool os::pd_release_memory(char* addr, size_t size) { 3246 return anon_munmap(addr, size); 3247 } 3248 3249 static bool linux_mprotect(char* addr, size_t size, int prot) { 3250 // Linux wants the mprotect address argument to be page aligned. 3251 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size()); 3252 3253 // According to SUSv3, mprotect() should only be used with mappings 3254 // established by mmap(), and mmap() always maps whole pages. Unaligned 3255 // 'addr' likely indicates problem in the VM (e.g. trying to change 3256 // protection of malloc'ed or statically allocated memory). Check the 3257 // caller if you hit this assert. 3258 assert(addr == bottom, "sanity check"); 3259 3260 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3261 return ::mprotect(bottom, size, prot) == 0; 3262 } 3263 3264 // Set protections specified 3265 bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3266 bool is_committed) { 3267 unsigned int p = 0; 3268 switch (prot) { 3269 case MEM_PROT_NONE: p = PROT_NONE; break; 3270 case MEM_PROT_READ: p = PROT_READ; break; 3271 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3272 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3273 default: 3274 ShouldNotReachHere(); 3275 } 3276 // is_committed is unused. 3277 return linux_mprotect(addr, bytes, p); 3278 } 3279 3280 bool os::guard_memory(char* addr, size_t size) { 3281 return linux_mprotect(addr, size, PROT_NONE); 3282 } 3283 3284 bool os::unguard_memory(char* addr, size_t size) { 3285 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3286 } 3287 3288 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3289 size_t page_size) { 3290 bool result = false; 3291 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3292 MAP_ANONYMOUS|MAP_PRIVATE, 3293 -1, 0); 3294 if (p != MAP_FAILED) { 3295 void *aligned_p = align_up(p, page_size); 3296 3297 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3298 3299 munmap(p, page_size * 2); 3300 } 3301 3302 if (warn && !result) { 3303 warning("TransparentHugePages is not supported by the operating system."); 3304 } 3305 3306 return result; 3307 } 3308 3309 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3310 bool result = false; 3311 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3312 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3313 -1, 0); 3314 3315 if (p != MAP_FAILED) { 3316 // We don't know if this really is a huge page or not. 3317 FILE *fp = fopen("/proc/self/maps", "r"); 3318 if (fp) { 3319 while (!feof(fp)) { 3320 char chars[257]; 3321 long x = 0; 3322 if (fgets(chars, sizeof(chars), fp)) { 3323 if (sscanf(chars, "%lx-%*x", &x) == 1 3324 && x == (long)p) { 3325 if (strstr (chars, "hugepage")) { 3326 result = true; 3327 break; 3328 } 3329 } 3330 } 3331 } 3332 fclose(fp); 3333 } 3334 munmap(p, page_size); 3335 } 3336 3337 if (warn && !result) { 3338 warning("HugeTLBFS is not supported by the operating system."); 3339 } 3340 3341 return result; 3342 } 3343 3344 // Set the coredump_filter bits to include largepages in core dump (bit 6) 3345 // 3346 // From the coredump_filter documentation: 3347 // 3348 // - (bit 0) anonymous private memory 3349 // - (bit 1) anonymous shared memory 3350 // - (bit 2) file-backed private memory 3351 // - (bit 3) file-backed shared memory 3352 // - (bit 4) ELF header pages in file-backed private memory areas (it is 3353 // effective only if the bit 2 is cleared) 3354 // - (bit 5) hugetlb private memory 3355 // - (bit 6) hugetlb shared memory 3356 // - (bit 7) dax private memory 3357 // - (bit 8) dax shared memory 3358 // 3359 static void set_coredump_filter(bool largepages, bool dax_shared) { 3360 FILE *f; 3361 long cdm; 3362 bool filter_changed = false; 3363 3364 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3365 return; 3366 } 3367 3368 if (fscanf(f, "%lx", &cdm) != 1) { 3369 fclose(f); 3370 return; 3371 } 3372 3373 rewind(f); 3374 3375 if (largepages && (cdm & LARGEPAGES_BIT) == 0) { 3376 cdm |= LARGEPAGES_BIT; 3377 filter_changed = true; 3378 } 3379 if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) { 3380 cdm |= DAX_SHARED_BIT; 3381 filter_changed = true; 3382 } 3383 if (filter_changed) { 3384 fprintf(f, "%#lx", cdm); 3385 } 3386 3387 fclose(f); 3388 } 3389 3390 // Large page support 3391 3392 static size_t _large_page_size = 0; 3393 3394 size_t os::Linux::find_large_page_size() { 3395 size_t large_page_size = 0; 3396 3397 // large_page_size on Linux is used to round up heap size. x86 uses either 3398 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3399 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3400 // page as large as 256M. 3401 // 3402 // Here we try to figure out page size by parsing /proc/meminfo and looking 3403 // for a line with the following format: 3404 // Hugepagesize: 2048 kB 3405 // 3406 // If we can't determine the value (e.g. /proc is not mounted, or the text 3407 // format has been changed), we'll use the largest page size supported by 3408 // the processor. 3409 3410 #ifndef ZERO 3411 large_page_size = 3412 AARCH64_ONLY(2 * M) 3413 AMD64_ONLY(2 * M) 3414 ARM32_ONLY(2 * M) 3415 IA32_ONLY(4 * M) 3416 IA64_ONLY(256 * M) 3417 PPC_ONLY(4 * M) 3418 S390_ONLY(1 * M) 3419 SPARC_ONLY(4 * M); 3420 #endif // ZERO 3421 3422 FILE *fp = fopen("/proc/meminfo", "r"); 3423 if (fp) { 3424 while (!feof(fp)) { 3425 int x = 0; 3426 char buf[16]; 3427 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3428 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3429 large_page_size = x * K; 3430 break; 3431 } 3432 } else { 3433 // skip to next line 3434 for (;;) { 3435 int ch = fgetc(fp); 3436 if (ch == EOF || ch == (int)'\n') break; 3437 } 3438 } 3439 } 3440 fclose(fp); 3441 } 3442 3443 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3444 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3445 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3446 proper_unit_for_byte_size(large_page_size)); 3447 } 3448 3449 return large_page_size; 3450 } 3451 3452 size_t os::Linux::setup_large_page_size() { 3453 _large_page_size = Linux::find_large_page_size(); 3454 const size_t default_page_size = (size_t)Linux::page_size(); 3455 if (_large_page_size > default_page_size) { 3456 _page_sizes[0] = _large_page_size; 3457 _page_sizes[1] = default_page_size; 3458 _page_sizes[2] = 0; 3459 } 3460 3461 return _large_page_size; 3462 } 3463 3464 bool os::Linux::setup_large_page_type(size_t page_size) { 3465 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3466 FLAG_IS_DEFAULT(UseSHM) && 3467 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3468 3469 // The type of large pages has not been specified by the user. 3470 3471 // Try UseHugeTLBFS and then UseSHM. 3472 UseHugeTLBFS = UseSHM = true; 3473 3474 // Don't try UseTransparentHugePages since there are known 3475 // performance issues with it turned on. This might change in the future. 3476 UseTransparentHugePages = false; 3477 } 3478 3479 if (UseTransparentHugePages) { 3480 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3481 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3482 UseHugeTLBFS = false; 3483 UseSHM = false; 3484 return true; 3485 } 3486 UseTransparentHugePages = false; 3487 } 3488 3489 if (UseHugeTLBFS) { 3490 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3491 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3492 UseSHM = false; 3493 return true; 3494 } 3495 UseHugeTLBFS = false; 3496 } 3497 3498 return UseSHM; 3499 } 3500 3501 void os::large_page_init() { 3502 if (!UseLargePages && 3503 !UseTransparentHugePages && 3504 !UseHugeTLBFS && 3505 !UseSHM) { 3506 // Not using large pages. 3507 return; 3508 } 3509 3510 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3511 // The user explicitly turned off large pages. 3512 // Ignore the rest of the large pages flags. 3513 UseTransparentHugePages = false; 3514 UseHugeTLBFS = false; 3515 UseSHM = false; 3516 return; 3517 } 3518 3519 size_t large_page_size = Linux::setup_large_page_size(); 3520 UseLargePages = Linux::setup_large_page_type(large_page_size); 3521 3522 set_coredump_filter(true /*largepages*/, false /*dax_shared*/); 3523 } 3524 3525 #ifndef SHM_HUGETLB 3526 #define SHM_HUGETLB 04000 3527 #endif 3528 3529 #define shm_warning_format(format, ...) \ 3530 do { \ 3531 if (UseLargePages && \ 3532 (!FLAG_IS_DEFAULT(UseLargePages) || \ 3533 !FLAG_IS_DEFAULT(UseSHM) || \ 3534 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \ 3535 warning(format, __VA_ARGS__); \ 3536 } \ 3537 } while (0) 3538 3539 #define shm_warning(str) shm_warning_format("%s", str) 3540 3541 #define shm_warning_with_errno(str) \ 3542 do { \ 3543 int err = errno; \ 3544 shm_warning_format(str " (error = %d)", err); \ 3545 } while (0) 3546 3547 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) { 3548 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment"); 3549 3550 if (!is_aligned(alignment, SHMLBA)) { 3551 assert(false, "Code below assumes that alignment is at least SHMLBA aligned"); 3552 return NULL; 3553 } 3554 3555 // To ensure that we get 'alignment' aligned memory from shmat, 3556 // we pre-reserve aligned virtual memory and then attach to that. 3557 3558 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL); 3559 if (pre_reserved_addr == NULL) { 3560 // Couldn't pre-reserve aligned memory. 3561 shm_warning("Failed to pre-reserve aligned memory for shmat."); 3562 return NULL; 3563 } 3564 3565 // SHM_REMAP is needed to allow shmat to map over an existing mapping. 3566 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP); 3567 3568 if ((intptr_t)addr == -1) { 3569 int err = errno; 3570 shm_warning_with_errno("Failed to attach shared memory."); 3571 3572 assert(err != EACCES, "Unexpected error"); 3573 assert(err != EIDRM, "Unexpected error"); 3574 assert(err != EINVAL, "Unexpected error"); 3575 3576 // Since we don't know if the kernel unmapped the pre-reserved memory area 3577 // we can't unmap it, since that would potentially unmap memory that was 3578 // mapped from other threads. 3579 return NULL; 3580 } 3581 3582 return addr; 3583 } 3584 3585 static char* shmat_at_address(int shmid, char* req_addr) { 3586 if (!is_aligned(req_addr, SHMLBA)) { 3587 assert(false, "Requested address needs to be SHMLBA aligned"); 3588 return NULL; 3589 } 3590 3591 char* addr = (char*)shmat(shmid, req_addr, 0); 3592 3593 if ((intptr_t)addr == -1) { 3594 shm_warning_with_errno("Failed to attach shared memory."); 3595 return NULL; 3596 } 3597 3598 return addr; 3599 } 3600 3601 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) { 3602 // If a req_addr has been provided, we assume that the caller has already aligned the address. 3603 if (req_addr != NULL) { 3604 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size"); 3605 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment"); 3606 return shmat_at_address(shmid, req_addr); 3607 } 3608 3609 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically 3610 // return large page size aligned memory addresses when req_addr == NULL. 3611 // However, if the alignment is larger than the large page size, we have 3612 // to manually ensure that the memory returned is 'alignment' aligned. 3613 if (alignment > os::large_page_size()) { 3614 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size"); 3615 return shmat_with_alignment(shmid, bytes, alignment); 3616 } else { 3617 return shmat_at_address(shmid, NULL); 3618 } 3619 } 3620 3621 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3622 char* req_addr, bool exec) { 3623 // "exec" is passed in but not used. Creating the shared image for 3624 // the code cache doesn't have an SHM_X executable permission to check. 3625 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3626 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3627 assert(is_aligned(req_addr, alignment), "Unaligned address"); 3628 3629 if (!is_aligned(bytes, os::large_page_size())) { 3630 return NULL; // Fallback to small pages. 3631 } 3632 3633 // Create a large shared memory region to attach to based on size. 3634 // Currently, size is the total size of the heap. 3635 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3636 if (shmid == -1) { 3637 // Possible reasons for shmget failure: 3638 // 1. shmmax is too small for Java heap. 3639 // > check shmmax value: cat /proc/sys/kernel/shmmax 3640 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3641 // 2. not enough large page memory. 3642 // > check available large pages: cat /proc/meminfo 3643 // > increase amount of large pages: 3644 // echo new_value > /proc/sys/vm/nr_hugepages 3645 // Note 1: different Linux may use different name for this property, 3646 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3647 // Note 2: it's possible there's enough physical memory available but 3648 // they are so fragmented after a long run that they can't 3649 // coalesce into large pages. Try to reserve large pages when 3650 // the system is still "fresh". 3651 shm_warning_with_errno("Failed to reserve shared memory."); 3652 return NULL; 3653 } 3654 3655 // Attach to the region. 3656 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr); 3657 3658 // Remove shmid. If shmat() is successful, the actual shared memory segment 3659 // will be deleted when it's detached by shmdt() or when the process 3660 // terminates. If shmat() is not successful this will remove the shared 3661 // segment immediately. 3662 shmctl(shmid, IPC_RMID, NULL); 3663 3664 return addr; 3665 } 3666 3667 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3668 int error) { 3669 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3670 3671 bool warn_on_failure = UseLargePages && 3672 (!FLAG_IS_DEFAULT(UseLargePages) || 3673 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3674 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3675 3676 if (warn_on_failure) { 3677 char msg[128]; 3678 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3679 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3680 warning("%s", msg); 3681 } 3682 } 3683 3684 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3685 char* req_addr, 3686 bool exec) { 3687 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3688 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size"); 3689 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3690 3691 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3692 char* addr = (char*)::mmap(req_addr, bytes, prot, 3693 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3694 -1, 0); 3695 3696 if (addr == MAP_FAILED) { 3697 warn_on_large_pages_failure(req_addr, bytes, errno); 3698 return NULL; 3699 } 3700 3701 assert(is_aligned(addr, os::large_page_size()), "Must be"); 3702 3703 return addr; 3704 } 3705 3706 // Reserve memory using mmap(MAP_HUGETLB). 3707 // - bytes shall be a multiple of alignment. 3708 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3709 // - alignment sets the alignment at which memory shall be allocated. 3710 // It must be a multiple of allocation granularity. 3711 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3712 // req_addr or NULL. 3713 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3714 size_t alignment, 3715 char* req_addr, 3716 bool exec) { 3717 size_t large_page_size = os::large_page_size(); 3718 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3719 3720 assert(is_aligned(req_addr, alignment), "Must be"); 3721 assert(is_aligned(bytes, alignment), "Must be"); 3722 3723 // First reserve - but not commit - the address range in small pages. 3724 char* const start = anon_mmap_aligned(bytes, alignment, req_addr); 3725 3726 if (start == NULL) { 3727 return NULL; 3728 } 3729 3730 assert(is_aligned(start, alignment), "Must be"); 3731 3732 char* end = start + bytes; 3733 3734 // Find the regions of the allocated chunk that can be promoted to large pages. 3735 char* lp_start = align_up(start, large_page_size); 3736 char* lp_end = align_down(end, large_page_size); 3737 3738 size_t lp_bytes = lp_end - lp_start; 3739 3740 assert(is_aligned(lp_bytes, large_page_size), "Must be"); 3741 3742 if (lp_bytes == 0) { 3743 // The mapped region doesn't even span the start and the end of a large page. 3744 // Fall back to allocate a non-special area. 3745 ::munmap(start, end - start); 3746 return NULL; 3747 } 3748 3749 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3750 3751 void* result; 3752 3753 // Commit small-paged leading area. 3754 if (start != lp_start) { 3755 result = ::mmap(start, lp_start - start, prot, 3756 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3757 -1, 0); 3758 if (result == MAP_FAILED) { 3759 ::munmap(lp_start, end - lp_start); 3760 return NULL; 3761 } 3762 } 3763 3764 // Commit large-paged area. 3765 result = ::mmap(lp_start, lp_bytes, prot, 3766 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3767 -1, 0); 3768 if (result == MAP_FAILED) { 3769 warn_on_large_pages_failure(lp_start, lp_bytes, errno); 3770 // If the mmap above fails, the large pages region will be unmapped and we 3771 // have regions before and after with small pages. Release these regions. 3772 // 3773 // | mapped | unmapped | mapped | 3774 // ^ ^ ^ ^ 3775 // start lp_start lp_end end 3776 // 3777 ::munmap(start, lp_start - start); 3778 ::munmap(lp_end, end - lp_end); 3779 return NULL; 3780 } 3781 3782 // Commit small-paged trailing area. 3783 if (lp_end != end) { 3784 result = ::mmap(lp_end, end - lp_end, prot, 3785 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3786 -1, 0); 3787 if (result == MAP_FAILED) { 3788 ::munmap(start, lp_end - start); 3789 return NULL; 3790 } 3791 } 3792 3793 return start; 3794 } 3795 3796 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3797 size_t alignment, 3798 char* req_addr, 3799 bool exec) { 3800 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3801 assert(is_aligned(req_addr, alignment), "Must be"); 3802 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be"); 3803 assert(is_power_of_2(os::large_page_size()), "Must be"); 3804 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3805 3806 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3807 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3808 } else { 3809 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3810 } 3811 } 3812 3813 char* os::reserve_memory_special(size_t bytes, size_t alignment, 3814 char* req_addr, bool exec) { 3815 assert(UseLargePages, "only for large pages"); 3816 3817 char* addr; 3818 if (UseSHM) { 3819 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3820 } else { 3821 assert(UseHugeTLBFS, "must be"); 3822 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3823 } 3824 3825 if (addr != NULL) { 3826 if (UseNUMAInterleaving) { 3827 numa_make_global(addr, bytes); 3828 } 3829 3830 // The memory is committed 3831 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3832 } 3833 3834 return addr; 3835 } 3836 3837 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3838 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3839 return shmdt(base) == 0; 3840 } 3841 3842 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3843 return pd_release_memory(base, bytes); 3844 } 3845 3846 bool os::release_memory_special(char* base, size_t bytes) { 3847 bool res; 3848 if (MemTracker::tracking_level() > NMT_minimal) { 3849 Tracker tkr(Tracker::release); 3850 res = os::Linux::release_memory_special_impl(base, bytes); 3851 if (res) { 3852 tkr.record((address)base, bytes); 3853 } 3854 3855 } else { 3856 res = os::Linux::release_memory_special_impl(base, bytes); 3857 } 3858 return res; 3859 } 3860 3861 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3862 assert(UseLargePages, "only for large pages"); 3863 bool res; 3864 3865 if (UseSHM) { 3866 res = os::Linux::release_memory_special_shm(base, bytes); 3867 } else { 3868 assert(UseHugeTLBFS, "must be"); 3869 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3870 } 3871 return res; 3872 } 3873 3874 size_t os::large_page_size() { 3875 return _large_page_size; 3876 } 3877 3878 // With SysV SHM the entire memory region must be allocated as shared 3879 // memory. 3880 // HugeTLBFS allows application to commit large page memory on demand. 3881 // However, when committing memory with HugeTLBFS fails, the region 3882 // that was supposed to be committed will lose the old reservation 3883 // and allow other threads to steal that memory region. Because of this 3884 // behavior we can't commit HugeTLBFS memory. 3885 bool os::can_commit_large_page_memory() { 3886 return UseTransparentHugePages; 3887 } 3888 3889 bool os::can_execute_large_page_memory() { 3890 return UseTransparentHugePages || UseHugeTLBFS; 3891 } 3892 3893 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) { 3894 assert(file_desc >= 0, "file_desc is not valid"); 3895 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr); 3896 if (result != NULL) { 3897 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) { 3898 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory")); 3899 } 3900 } 3901 return result; 3902 } 3903 3904 // Reserve memory at an arbitrary address, only if that area is 3905 // available (and not reserved for something else). 3906 3907 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3908 const int max_tries = 10; 3909 char* base[max_tries]; 3910 size_t size[max_tries]; 3911 const size_t gap = 0x000000; 3912 3913 // Assert only that the size is a multiple of the page size, since 3914 // that's all that mmap requires, and since that's all we really know 3915 // about at this low abstraction level. If we need higher alignment, 3916 // we can either pass an alignment to this method or verify alignment 3917 // in one of the methods further up the call chain. See bug 5044738. 3918 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3919 3920 // Repeatedly allocate blocks until the block is allocated at the 3921 // right spot. 3922 3923 // Linux mmap allows caller to pass an address as hint; give it a try first, 3924 // if kernel honors the hint then we can return immediately. 3925 char * addr = anon_mmap(requested_addr, bytes, false); 3926 if (addr == requested_addr) { 3927 return requested_addr; 3928 } 3929 3930 if (addr != NULL) { 3931 // mmap() is successful but it fails to reserve at the requested address 3932 anon_munmap(addr, bytes); 3933 } 3934 3935 int i; 3936 for (i = 0; i < max_tries; ++i) { 3937 base[i] = reserve_memory(bytes); 3938 3939 if (base[i] != NULL) { 3940 // Is this the block we wanted? 3941 if (base[i] == requested_addr) { 3942 size[i] = bytes; 3943 break; 3944 } 3945 3946 // Does this overlap the block we wanted? Give back the overlapped 3947 // parts and try again. 3948 3949 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3950 if (top_overlap >= 0 && (size_t)top_overlap < bytes) { 3951 unmap_memory(base[i], top_overlap); 3952 base[i] += top_overlap; 3953 size[i] = bytes - top_overlap; 3954 } else { 3955 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr; 3956 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) { 3957 unmap_memory(requested_addr, bottom_overlap); 3958 size[i] = bytes - bottom_overlap; 3959 } else { 3960 size[i] = bytes; 3961 } 3962 } 3963 } 3964 } 3965 3966 // Give back the unused reserved pieces. 3967 3968 for (int j = 0; j < i; ++j) { 3969 if (base[j] != NULL) { 3970 unmap_memory(base[j], size[j]); 3971 } 3972 } 3973 3974 if (i < max_tries) { 3975 return requested_addr; 3976 } else { 3977 return NULL; 3978 } 3979 } 3980 3981 size_t os::read(int fd, void *buf, unsigned int nBytes) { 3982 return ::read(fd, buf, nBytes); 3983 } 3984 3985 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) { 3986 return ::pread(fd, buf, nBytes, offset); 3987 } 3988 3989 // Short sleep, direct OS call. 3990 // 3991 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3992 // sched_yield(2) will actually give up the CPU: 3993 // 3994 // * Alone on this pariticular CPU, keeps running. 3995 // * Before the introduction of "skip_buddy" with "compat_yield" disabled 3996 // (pre 2.6.39). 3997 // 3998 // So calling this with 0 is an alternative. 3999 // 4000 void os::naked_short_sleep(jlong ms) { 4001 struct timespec req; 4002 4003 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 4004 req.tv_sec = 0; 4005 if (ms > 0) { 4006 req.tv_nsec = (ms % 1000) * 1000000; 4007 } else { 4008 req.tv_nsec = 1; 4009 } 4010 4011 nanosleep(&req, NULL); 4012 4013 return; 4014 } 4015 4016 // Sleep forever; naked call to OS-specific sleep; use with CAUTION 4017 void os::infinite_sleep() { 4018 while (true) { // sleep forever ... 4019 ::sleep(100); // ... 100 seconds at a time 4020 } 4021 } 4022 4023 // Used to convert frequent JVM_Yield() to nops 4024 bool os::dont_yield() { 4025 return DontYieldALot; 4026 } 4027 4028 void os::naked_yield() { 4029 sched_yield(); 4030 } 4031 4032 //////////////////////////////////////////////////////////////////////////////// 4033 // thread priority support 4034 4035 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 4036 // only supports dynamic priority, static priority must be zero. For real-time 4037 // applications, Linux supports SCHED_RR which allows static priority (1-99). 4038 // However, for large multi-threaded applications, SCHED_RR is not only slower 4039 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 4040 // of 5 runs - Sep 2005). 4041 // 4042 // The following code actually changes the niceness of kernel-thread/LWP. It 4043 // has an assumption that setpriority() only modifies one kernel-thread/LWP, 4044 // not the entire user process, and user level threads are 1:1 mapped to kernel 4045 // threads. It has always been the case, but could change in the future. For 4046 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). 4047 // It is only used when ThreadPriorityPolicy=1 and requires root privilege. 4048 4049 int os::java_to_os_priority[CriticalPriority + 1] = { 4050 19, // 0 Entry should never be used 4051 4052 4, // 1 MinPriority 4053 3, // 2 4054 2, // 3 4055 4056 1, // 4 4057 0, // 5 NormPriority 4058 -1, // 6 4059 4060 -2, // 7 4061 -3, // 8 4062 -4, // 9 NearMaxPriority 4063 4064 -5, // 10 MaxPriority 4065 4066 -5 // 11 CriticalPriority 4067 }; 4068 4069 static int prio_init() { 4070 if (ThreadPriorityPolicy == 1) { 4071 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 4072 // if effective uid is not root. Perhaps, a more elegant way of doing 4073 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 4074 if (geteuid() != 0) { 4075 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 4076 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 4077 } 4078 ThreadPriorityPolicy = 0; 4079 } 4080 } 4081 if (UseCriticalJavaThreadPriority) { 4082 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 4083 } 4084 return 0; 4085 } 4086 4087 OSReturn os::set_native_priority(Thread* thread, int newpri) { 4088 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 4089 4090 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 4091 return (ret == 0) ? OS_OK : OS_ERR; 4092 } 4093 4094 OSReturn os::get_native_priority(const Thread* const thread, 4095 int *priority_ptr) { 4096 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 4097 *priority_ptr = java_to_os_priority[NormPriority]; 4098 return OS_OK; 4099 } 4100 4101 errno = 0; 4102 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 4103 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 4104 } 4105 4106 // Hint to the underlying OS that a task switch would not be good. 4107 // Void return because it's a hint and can fail. 4108 void os::hint_no_preempt() {} 4109 4110 //////////////////////////////////////////////////////////////////////////////// 4111 // suspend/resume support 4112 4113 // The low-level signal-based suspend/resume support is a remnant from the 4114 // old VM-suspension that used to be for java-suspension, safepoints etc, 4115 // within hotspot. Currently used by JFR's OSThreadSampler 4116 // 4117 // The remaining code is greatly simplified from the more general suspension 4118 // code that used to be used. 4119 // 4120 // The protocol is quite simple: 4121 // - suspend: 4122 // - sends a signal to the target thread 4123 // - polls the suspend state of the osthread using a yield loop 4124 // - target thread signal handler (SR_handler) sets suspend state 4125 // and blocks in sigsuspend until continued 4126 // - resume: 4127 // - sets target osthread state to continue 4128 // - sends signal to end the sigsuspend loop in the SR_handler 4129 // 4130 // Note that the SR_lock plays no role in this suspend/resume protocol, 4131 // but is checked for NULL in SR_handler as a thread termination indicator. 4132 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs. 4133 // 4134 // Note that resume_clear_context() and suspend_save_context() are needed 4135 // by SR_handler(), so that fetch_frame_from_ucontext() works, 4136 // which in part is used by: 4137 // - Forte Analyzer: AsyncGetCallTrace() 4138 // - StackBanging: get_frame_at_stack_banging_point() 4139 4140 static void resume_clear_context(OSThread *osthread) { 4141 osthread->set_ucontext(NULL); 4142 osthread->set_siginfo(NULL); 4143 } 4144 4145 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 4146 ucontext_t* context) { 4147 osthread->set_ucontext(context); 4148 osthread->set_siginfo(siginfo); 4149 } 4150 4151 // Handler function invoked when a thread's execution is suspended or 4152 // resumed. We have to be careful that only async-safe functions are 4153 // called here (Note: most pthread functions are not async safe and 4154 // should be avoided.) 4155 // 4156 // Note: sigwait() is a more natural fit than sigsuspend() from an 4157 // interface point of view, but sigwait() prevents the signal hander 4158 // from being run. libpthread would get very confused by not having 4159 // its signal handlers run and prevents sigwait()'s use with the 4160 // mutex granting granting signal. 4161 // 4162 // Currently only ever called on the VMThread and JavaThreads (PC sampling) 4163 // 4164 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 4165 // Save and restore errno to avoid confusing native code with EINTR 4166 // after sigsuspend. 4167 int old_errno = errno; 4168 4169 Thread* thread = Thread::current_or_null_safe(); 4170 assert(thread != NULL, "Missing current thread in SR_handler"); 4171 4172 // On some systems we have seen signal delivery get "stuck" until the signal 4173 // mask is changed as part of thread termination. Check that the current thread 4174 // has not already terminated (via SR_lock()) - else the following assertion 4175 // will fail because the thread is no longer a JavaThread as the ~JavaThread 4176 // destructor has completed. 4177 4178 if (thread->SR_lock() == NULL) { 4179 return; 4180 } 4181 4182 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 4183 4184 OSThread* osthread = thread->osthread(); 4185 4186 os::SuspendResume::State current = osthread->sr.state(); 4187 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 4188 suspend_save_context(osthread, siginfo, context); 4189 4190 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4191 os::SuspendResume::State state = osthread->sr.suspended(); 4192 if (state == os::SuspendResume::SR_SUSPENDED) { 4193 sigset_t suspend_set; // signals for sigsuspend() 4194 sigemptyset(&suspend_set); 4195 // get current set of blocked signals and unblock resume signal 4196 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4197 sigdelset(&suspend_set, SR_signum); 4198 4199 sr_semaphore.signal(); 4200 // wait here until we are resumed 4201 while (1) { 4202 sigsuspend(&suspend_set); 4203 4204 os::SuspendResume::State result = osthread->sr.running(); 4205 if (result == os::SuspendResume::SR_RUNNING) { 4206 sr_semaphore.signal(); 4207 break; 4208 } 4209 } 4210 4211 } else if (state == os::SuspendResume::SR_RUNNING) { 4212 // request was cancelled, continue 4213 } else { 4214 ShouldNotReachHere(); 4215 } 4216 4217 resume_clear_context(osthread); 4218 } else if (current == os::SuspendResume::SR_RUNNING) { 4219 // request was cancelled, continue 4220 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4221 // ignore 4222 } else { 4223 // ignore 4224 } 4225 4226 errno = old_errno; 4227 } 4228 4229 static int SR_initialize() { 4230 struct sigaction act; 4231 char *s; 4232 4233 // Get signal number to use for suspend/resume 4234 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4235 int sig = ::strtol(s, 0, 10); 4236 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769. 4237 sig < NSIG) { // Must be legal signal and fit into sigflags[]. 4238 SR_signum = sig; 4239 } else { 4240 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.", 4241 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum); 4242 } 4243 } 4244 4245 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4246 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4247 4248 sigemptyset(&SR_sigset); 4249 sigaddset(&SR_sigset, SR_signum); 4250 4251 // Set up signal handler for suspend/resume 4252 act.sa_flags = SA_RESTART|SA_SIGINFO; 4253 act.sa_handler = (void (*)(int)) SR_handler; 4254 4255 // SR_signum is blocked by default. 4256 // 4528190 - We also need to block pthread restart signal (32 on all 4257 // supported Linux platforms). Note that LinuxThreads need to block 4258 // this signal for all threads to work properly. So we don't have 4259 // to use hard-coded signal number when setting up the mask. 4260 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4261 4262 if (sigaction(SR_signum, &act, 0) == -1) { 4263 return -1; 4264 } 4265 4266 // Save signal flag 4267 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4268 return 0; 4269 } 4270 4271 static int sr_notify(OSThread* osthread) { 4272 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4273 assert_status(status == 0, status, "pthread_kill"); 4274 return status; 4275 } 4276 4277 // "Randomly" selected value for how long we want to spin 4278 // before bailing out on suspending a thread, also how often 4279 // we send a signal to a thread we want to resume 4280 static const int RANDOMLY_LARGE_INTEGER = 1000000; 4281 static const int RANDOMLY_LARGE_INTEGER2 = 100; 4282 4283 // returns true on success and false on error - really an error is fatal 4284 // but this seems the normal response to library errors 4285 static bool do_suspend(OSThread* osthread) { 4286 assert(osthread->sr.is_running(), "thread should be running"); 4287 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4288 4289 // mark as suspended and send signal 4290 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4291 // failed to switch, state wasn't running? 4292 ShouldNotReachHere(); 4293 return false; 4294 } 4295 4296 if (sr_notify(osthread) != 0) { 4297 ShouldNotReachHere(); 4298 } 4299 4300 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4301 while (true) { 4302 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) { 4303 break; 4304 } else { 4305 // timeout 4306 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4307 if (cancelled == os::SuspendResume::SR_RUNNING) { 4308 return false; 4309 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4310 // make sure that we consume the signal on the semaphore as well 4311 sr_semaphore.wait(); 4312 break; 4313 } else { 4314 ShouldNotReachHere(); 4315 return false; 4316 } 4317 } 4318 } 4319 4320 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4321 return true; 4322 } 4323 4324 static void do_resume(OSThread* osthread) { 4325 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4326 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4327 4328 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4329 // failed to switch to WAKEUP_REQUEST 4330 ShouldNotReachHere(); 4331 return; 4332 } 4333 4334 while (true) { 4335 if (sr_notify(osthread) == 0) { 4336 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) { 4337 if (osthread->sr.is_running()) { 4338 return; 4339 } 4340 } 4341 } else { 4342 ShouldNotReachHere(); 4343 } 4344 } 4345 4346 guarantee(osthread->sr.is_running(), "Must be running!"); 4347 } 4348 4349 /////////////////////////////////////////////////////////////////////////////////// 4350 // signal handling (except suspend/resume) 4351 4352 // This routine may be used by user applications as a "hook" to catch signals. 4353 // The user-defined signal handler must pass unrecognized signals to this 4354 // routine, and if it returns true (non-zero), then the signal handler must 4355 // return immediately. If the flag "abort_if_unrecognized" is true, then this 4356 // routine will never retun false (zero), but instead will execute a VM panic 4357 // routine kill the process. 4358 // 4359 // If this routine returns false, it is OK to call it again. This allows 4360 // the user-defined signal handler to perform checks either before or after 4361 // the VM performs its own checks. Naturally, the user code would be making 4362 // a serious error if it tried to handle an exception (such as a null check 4363 // or breakpoint) that the VM was generating for its own correct operation. 4364 // 4365 // This routine may recognize any of the following kinds of signals: 4366 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4367 // It should be consulted by handlers for any of those signals. 4368 // 4369 // The caller of this routine must pass in the three arguments supplied 4370 // to the function referred to in the "sa_sigaction" (not the "sa_handler") 4371 // field of the structure passed to sigaction(). This routine assumes that 4372 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4373 // 4374 // Note that the VM will print warnings if it detects conflicting signal 4375 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4376 // 4377 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4378 siginfo_t* siginfo, 4379 void* ucontext, 4380 int abort_if_unrecognized); 4381 4382 void signalHandler(int sig, siginfo_t* info, void* uc) { 4383 assert(info != NULL && uc != NULL, "it must be old kernel"); 4384 int orig_errno = errno; // Preserve errno value over signal handler. 4385 JVM_handle_linux_signal(sig, info, uc, true); 4386 errno = orig_errno; 4387 } 4388 4389 4390 // This boolean allows users to forward their own non-matching signals 4391 // to JVM_handle_linux_signal, harmlessly. 4392 bool os::Linux::signal_handlers_are_installed = false; 4393 4394 // For signal-chaining 4395 struct sigaction sigact[NSIG]; 4396 uint64_t sigs = 0; 4397 #if (64 < NSIG-1) 4398 #error "Not all signals can be encoded in sigs. Adapt its type!" 4399 #endif 4400 bool os::Linux::libjsig_is_loaded = false; 4401 typedef struct sigaction *(*get_signal_t)(int); 4402 get_signal_t os::Linux::get_signal_action = NULL; 4403 4404 struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4405 struct sigaction *actp = NULL; 4406 4407 if (libjsig_is_loaded) { 4408 // Retrieve the old signal handler from libjsig 4409 actp = (*get_signal_action)(sig); 4410 } 4411 if (actp == NULL) { 4412 // Retrieve the preinstalled signal handler from jvm 4413 actp = get_preinstalled_handler(sig); 4414 } 4415 4416 return actp; 4417 } 4418 4419 static bool call_chained_handler(struct sigaction *actp, int sig, 4420 siginfo_t *siginfo, void *context) { 4421 // Call the old signal handler 4422 if (actp->sa_handler == SIG_DFL) { 4423 // It's more reasonable to let jvm treat it as an unexpected exception 4424 // instead of taking the default action. 4425 return false; 4426 } else if (actp->sa_handler != SIG_IGN) { 4427 if ((actp->sa_flags & SA_NODEFER) == 0) { 4428 // automaticlly block the signal 4429 sigaddset(&(actp->sa_mask), sig); 4430 } 4431 4432 sa_handler_t hand = NULL; 4433 sa_sigaction_t sa = NULL; 4434 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4435 // retrieve the chained handler 4436 if (siginfo_flag_set) { 4437 sa = actp->sa_sigaction; 4438 } else { 4439 hand = actp->sa_handler; 4440 } 4441 4442 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4443 actp->sa_handler = SIG_DFL; 4444 } 4445 4446 // try to honor the signal mask 4447 sigset_t oset; 4448 sigemptyset(&oset); 4449 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4450 4451 // call into the chained handler 4452 if (siginfo_flag_set) { 4453 (*sa)(sig, siginfo, context); 4454 } else { 4455 (*hand)(sig); 4456 } 4457 4458 // restore the signal mask 4459 pthread_sigmask(SIG_SETMASK, &oset, NULL); 4460 } 4461 // Tell jvm's signal handler the signal is taken care of. 4462 return true; 4463 } 4464 4465 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4466 bool chained = false; 4467 // signal-chaining 4468 if (UseSignalChaining) { 4469 struct sigaction *actp = get_chained_signal_action(sig); 4470 if (actp != NULL) { 4471 chained = call_chained_handler(actp, sig, siginfo, context); 4472 } 4473 } 4474 return chained; 4475 } 4476 4477 struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4478 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) { 4479 return &sigact[sig]; 4480 } 4481 return NULL; 4482 } 4483 4484 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4485 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4486 sigact[sig] = oldAct; 4487 sigs |= (uint64_t)1 << (sig-1); 4488 } 4489 4490 // for diagnostic 4491 int sigflags[NSIG]; 4492 4493 int os::Linux::get_our_sigflags(int sig) { 4494 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4495 return sigflags[sig]; 4496 } 4497 4498 void os::Linux::set_our_sigflags(int sig, int flags) { 4499 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4500 if (sig > 0 && sig < NSIG) { 4501 sigflags[sig] = flags; 4502 } 4503 } 4504 4505 void os::Linux::set_signal_handler(int sig, bool set_installed) { 4506 // Check for overwrite. 4507 struct sigaction oldAct; 4508 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4509 4510 void* oldhand = oldAct.sa_sigaction 4511 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4512 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4513 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4514 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4515 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4516 if (AllowUserSignalHandlers || !set_installed) { 4517 // Do not overwrite; user takes responsibility to forward to us. 4518 return; 4519 } else if (UseSignalChaining) { 4520 // save the old handler in jvm 4521 save_preinstalled_handler(sig, oldAct); 4522 // libjsig also interposes the sigaction() call below and saves the 4523 // old sigaction on it own. 4524 } else { 4525 fatal("Encountered unexpected pre-existing sigaction handler " 4526 "%#lx for signal %d.", (long)oldhand, sig); 4527 } 4528 } 4529 4530 struct sigaction sigAct; 4531 sigfillset(&(sigAct.sa_mask)); 4532 sigAct.sa_handler = SIG_DFL; 4533 if (!set_installed) { 4534 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4535 } else { 4536 sigAct.sa_sigaction = signalHandler; 4537 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4538 } 4539 // Save flags, which are set by ours 4540 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4541 sigflags[sig] = sigAct.sa_flags; 4542 4543 int ret = sigaction(sig, &sigAct, &oldAct); 4544 assert(ret == 0, "check"); 4545 4546 void* oldhand2 = oldAct.sa_sigaction 4547 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4548 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4549 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4550 } 4551 4552 // install signal handlers for signals that HotSpot needs to 4553 // handle in order to support Java-level exception handling. 4554 4555 void os::Linux::install_signal_handlers() { 4556 if (!signal_handlers_are_installed) { 4557 signal_handlers_are_installed = true; 4558 4559 // signal-chaining 4560 typedef void (*signal_setting_t)(); 4561 signal_setting_t begin_signal_setting = NULL; 4562 signal_setting_t end_signal_setting = NULL; 4563 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4564 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4565 if (begin_signal_setting != NULL) { 4566 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4567 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4568 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4569 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4570 libjsig_is_loaded = true; 4571 assert(UseSignalChaining, "should enable signal-chaining"); 4572 } 4573 if (libjsig_is_loaded) { 4574 // Tell libjsig jvm is setting signal handlers 4575 (*begin_signal_setting)(); 4576 } 4577 4578 set_signal_handler(SIGSEGV, true); 4579 set_signal_handler(SIGPIPE, true); 4580 set_signal_handler(SIGBUS, true); 4581 set_signal_handler(SIGILL, true); 4582 set_signal_handler(SIGFPE, true); 4583 #if defined(PPC64) 4584 set_signal_handler(SIGTRAP, true); 4585 #endif 4586 set_signal_handler(SIGXFSZ, true); 4587 4588 if (libjsig_is_loaded) { 4589 // Tell libjsig jvm finishes setting signal handlers 4590 (*end_signal_setting)(); 4591 } 4592 4593 // We don't activate signal checker if libjsig is in place, we trust ourselves 4594 // and if UserSignalHandler is installed all bets are off. 4595 // Log that signal checking is off only if -verbose:jni is specified. 4596 if (CheckJNICalls) { 4597 if (libjsig_is_loaded) { 4598 if (PrintJNIResolving) { 4599 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4600 } 4601 check_signals = false; 4602 } 4603 if (AllowUserSignalHandlers) { 4604 if (PrintJNIResolving) { 4605 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4606 } 4607 check_signals = false; 4608 } 4609 } 4610 } 4611 } 4612 4613 // This is the fastest way to get thread cpu time on Linux. 4614 // Returns cpu time (user+sys) for any thread, not only for current. 4615 // POSIX compliant clocks are implemented in the kernels 2.6.16+. 4616 // It might work on 2.6.10+ with a special kernel/glibc patch. 4617 // For reference, please, see IEEE Std 1003.1-2004: 4618 // http://www.unix.org/single_unix_specification 4619 4620 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4621 struct timespec tp; 4622 int rc = os::Linux::clock_gettime(clockid, &tp); 4623 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4624 4625 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4626 } 4627 4628 void os::Linux::initialize_os_info() { 4629 assert(_os_version == 0, "OS info already initialized"); 4630 4631 struct utsname _uname; 4632 4633 uint32_t major; 4634 uint32_t minor; 4635 uint32_t fix; 4636 4637 int rc; 4638 4639 // Kernel version is unknown if 4640 // verification below fails. 4641 _os_version = 0x01000000; 4642 4643 rc = uname(&_uname); 4644 if (rc != -1) { 4645 4646 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix); 4647 if (rc == 3) { 4648 4649 if (major < 256 && minor < 256 && fix < 256) { 4650 // Kernel version format is as expected, 4651 // set it overriding unknown state. 4652 _os_version = (major << 16) | 4653 (minor << 8 ) | 4654 (fix << 0 ) ; 4655 } 4656 } 4657 } 4658 } 4659 4660 uint32_t os::Linux::os_version() { 4661 assert(_os_version != 0, "not initialized"); 4662 return _os_version & 0x00FFFFFF; 4663 } 4664 4665 bool os::Linux::os_version_is_known() { 4666 assert(_os_version != 0, "not initialized"); 4667 return _os_version & 0x01000000 ? false : true; 4668 } 4669 4670 ///// 4671 // glibc on Linux platform uses non-documented flag 4672 // to indicate, that some special sort of signal 4673 // trampoline is used. 4674 // We will never set this flag, and we should 4675 // ignore this flag in our diagnostic 4676 #ifdef SIGNIFICANT_SIGNAL_MASK 4677 #undef SIGNIFICANT_SIGNAL_MASK 4678 #endif 4679 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4680 4681 static const char* get_signal_handler_name(address handler, 4682 char* buf, int buflen) { 4683 int offset = 0; 4684 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4685 if (found) { 4686 // skip directory names 4687 const char *p1, *p2; 4688 p1 = buf; 4689 size_t len = strlen(os::file_separator()); 4690 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4691 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4692 } else { 4693 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4694 } 4695 return buf; 4696 } 4697 4698 static void print_signal_handler(outputStream* st, int sig, 4699 char* buf, size_t buflen) { 4700 struct sigaction sa; 4701 4702 sigaction(sig, NULL, &sa); 4703 4704 // See comment for SIGNIFICANT_SIGNAL_MASK define 4705 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4706 4707 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4708 4709 address handler = (sa.sa_flags & SA_SIGINFO) 4710 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4711 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4712 4713 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4714 st->print("SIG_DFL"); 4715 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4716 st->print("SIG_IGN"); 4717 } else { 4718 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4719 } 4720 4721 st->print(", sa_mask[0]="); 4722 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4723 4724 address rh = VMError::get_resetted_sighandler(sig); 4725 // May be, handler was resetted by VMError? 4726 if (rh != NULL) { 4727 handler = rh; 4728 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4729 } 4730 4731 st->print(", sa_flags="); 4732 os::Posix::print_sa_flags(st, sa.sa_flags); 4733 4734 // Check: is it our handler? 4735 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4736 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4737 // It is our signal handler 4738 // check for flags, reset system-used one! 4739 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4740 st->print( 4741 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4742 os::Linux::get_our_sigflags(sig)); 4743 } 4744 } 4745 st->cr(); 4746 } 4747 4748 4749 #define DO_SIGNAL_CHECK(sig) \ 4750 do { \ 4751 if (!sigismember(&check_signal_done, sig)) { \ 4752 os::Linux::check_signal_handler(sig); \ 4753 } \ 4754 } while (0) 4755 4756 // This method is a periodic task to check for misbehaving JNI applications 4757 // under CheckJNI, we can add any periodic checks here 4758 4759 void os::run_periodic_checks() { 4760 if (check_signals == false) return; 4761 4762 // SEGV and BUS if overridden could potentially prevent 4763 // generation of hs*.log in the event of a crash, debugging 4764 // such a case can be very challenging, so we absolutely 4765 // check the following for a good measure: 4766 DO_SIGNAL_CHECK(SIGSEGV); 4767 DO_SIGNAL_CHECK(SIGILL); 4768 DO_SIGNAL_CHECK(SIGFPE); 4769 DO_SIGNAL_CHECK(SIGBUS); 4770 DO_SIGNAL_CHECK(SIGPIPE); 4771 DO_SIGNAL_CHECK(SIGXFSZ); 4772 #if defined(PPC64) 4773 DO_SIGNAL_CHECK(SIGTRAP); 4774 #endif 4775 4776 // ReduceSignalUsage allows the user to override these handlers 4777 // see comments at the very top and jvm_md.h 4778 if (!ReduceSignalUsage) { 4779 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4780 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4781 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4782 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4783 } 4784 4785 DO_SIGNAL_CHECK(SR_signum); 4786 } 4787 4788 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4789 4790 static os_sigaction_t os_sigaction = NULL; 4791 4792 void os::Linux::check_signal_handler(int sig) { 4793 char buf[O_BUFLEN]; 4794 address jvmHandler = NULL; 4795 4796 4797 struct sigaction act; 4798 if (os_sigaction == NULL) { 4799 // only trust the default sigaction, in case it has been interposed 4800 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4801 if (os_sigaction == NULL) return; 4802 } 4803 4804 os_sigaction(sig, (struct sigaction*)NULL, &act); 4805 4806 4807 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4808 4809 address thisHandler = (act.sa_flags & SA_SIGINFO) 4810 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4811 : CAST_FROM_FN_PTR(address, act.sa_handler); 4812 4813 4814 switch (sig) { 4815 case SIGSEGV: 4816 case SIGBUS: 4817 case SIGFPE: 4818 case SIGPIPE: 4819 case SIGILL: 4820 case SIGXFSZ: 4821 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4822 break; 4823 4824 case SHUTDOWN1_SIGNAL: 4825 case SHUTDOWN2_SIGNAL: 4826 case SHUTDOWN3_SIGNAL: 4827 case BREAK_SIGNAL: 4828 jvmHandler = (address)user_handler(); 4829 break; 4830 4831 default: 4832 if (sig == SR_signum) { 4833 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4834 } else { 4835 return; 4836 } 4837 break; 4838 } 4839 4840 if (thisHandler != jvmHandler) { 4841 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4842 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4843 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4844 // No need to check this sig any longer 4845 sigaddset(&check_signal_done, sig); 4846 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4847 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4848 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4849 exception_name(sig, buf, O_BUFLEN)); 4850 } 4851 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4852 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4853 tty->print("expected:"); 4854 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig)); 4855 tty->cr(); 4856 tty->print(" found:"); 4857 os::Posix::print_sa_flags(tty, act.sa_flags); 4858 tty->cr(); 4859 // No need to check this sig any longer 4860 sigaddset(&check_signal_done, sig); 4861 } 4862 4863 // Dump all the signal 4864 if (sigismember(&check_signal_done, sig)) { 4865 print_signal_handlers(tty, buf, O_BUFLEN); 4866 } 4867 } 4868 4869 extern void report_error(char* file_name, int line_no, char* title, 4870 char* format, ...); 4871 4872 // this is called _before_ most of the global arguments have been parsed 4873 void os::init(void) { 4874 char dummy; // used to get a guess on initial stack address 4875 4876 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4877 4878 init_random(1234567); 4879 4880 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4881 if (Linux::page_size() == -1) { 4882 fatal("os_linux.cpp: os::init: sysconf failed (%s)", 4883 os::strerror(errno)); 4884 } 4885 init_page_sizes((size_t) Linux::page_size()); 4886 4887 Linux::initialize_system_info(); 4888 4889 Linux::initialize_os_info(); 4890 4891 // _main_thread points to the thread that created/loaded the JVM. 4892 Linux::_main_thread = pthread_self(); 4893 4894 Linux::clock_init(); 4895 initial_time_count = javaTimeNanos(); 4896 4897 // retrieve entry point for pthread_setname_np 4898 Linux::_pthread_setname_np = 4899 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4900 4901 os::Posix::init(); 4902 } 4903 4904 // To install functions for atexit system call 4905 extern "C" { 4906 static void perfMemory_exit_helper() { 4907 perfMemory_exit(); 4908 } 4909 } 4910 4911 void os::pd_init_container_support() { 4912 OSContainer::init(); 4913 } 4914 4915 // this is called _after_ the global arguments have been parsed 4916 jint os::init_2(void) { 4917 4918 os::Posix::init_2(); 4919 4920 Linux::fast_thread_clock_init(); 4921 4922 // initialize suspend/resume support - must do this before signal_sets_init() 4923 if (SR_initialize() != 0) { 4924 perror("SR_initialize failed"); 4925 return JNI_ERR; 4926 } 4927 4928 Linux::signal_sets_init(); 4929 Linux::install_signal_handlers(); 4930 4931 // Check and sets minimum stack sizes against command line options 4932 if (Posix::set_minimum_stack_sizes() == JNI_ERR) { 4933 return JNI_ERR; 4934 } 4935 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4936 4937 #if defined(IA32) 4938 workaround_expand_exec_shield_cs_limit(); 4939 #endif 4940 4941 Linux::libpthread_init(); 4942 Linux::sched_getcpu_init(); 4943 log_info(os)("HotSpot is running with %s, %s", 4944 Linux::glibc_version(), Linux::libpthread_version()); 4945 4946 if (UseNUMA) { 4947 if (!Linux::libnuma_init()) { 4948 UseNUMA = false; 4949 } else { 4950 if ((Linux::numa_max_node() < 1)) { 4951 // There's only one node(they start from 0), disable NUMA. 4952 UseNUMA = false; 4953 } 4954 } 4955 4956 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4957 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4958 // we can make the adaptive lgrp chunk resizing work. If the user specified both 4959 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn 4960 // and disable adaptive resizing. 4961 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) { 4962 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, " 4963 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)"); 4964 UseAdaptiveSizePolicy = false; 4965 UseAdaptiveNUMAChunkSizing = false; 4966 } 4967 } 4968 4969 if (!UseNUMA && ForceNUMA) { 4970 UseNUMA = true; 4971 } 4972 } 4973 4974 if (MaxFDLimit) { 4975 // set the number of file descriptors to max. print out error 4976 // if getrlimit/setrlimit fails but continue regardless. 4977 struct rlimit nbr_files; 4978 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4979 if (status != 0) { 4980 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno)); 4981 } else { 4982 nbr_files.rlim_cur = nbr_files.rlim_max; 4983 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4984 if (status != 0) { 4985 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno)); 4986 } 4987 } 4988 } 4989 4990 // Initialize lock used to serialize thread creation (see os::create_thread) 4991 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4992 4993 // at-exit methods are called in the reverse order of their registration. 4994 // atexit functions are called on return from main or as a result of a 4995 // call to exit(3C). There can be only 32 of these functions registered 4996 // and atexit() does not set errno. 4997 4998 if (PerfAllowAtExitRegistration) { 4999 // only register atexit functions if PerfAllowAtExitRegistration is set. 5000 // atexit functions can be delayed until process exit time, which 5001 // can be problematic for embedded VM situations. Embedded VMs should 5002 // call DestroyJavaVM() to assure that VM resources are released. 5003 5004 // note: perfMemory_exit_helper atexit function may be removed in 5005 // the future if the appropriate cleanup code can be added to the 5006 // VM_Exit VMOperation's doit method. 5007 if (atexit(perfMemory_exit_helper) != 0) { 5008 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 5009 } 5010 } 5011 5012 // initialize thread priority policy 5013 prio_init(); 5014 5015 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) { 5016 set_coredump_filter(false /*largepages*/, true /*dax_shared*/); 5017 } 5018 return JNI_OK; 5019 } 5020 5021 // Mark the polling page as unreadable 5022 void os::make_polling_page_unreadable(void) { 5023 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 5024 fatal("Could not disable polling page"); 5025 } 5026 } 5027 5028 // Mark the polling page as readable 5029 void os::make_polling_page_readable(void) { 5030 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 5031 fatal("Could not enable polling page"); 5032 } 5033 } 5034 5035 // older glibc versions don't have this macro (which expands to 5036 // an optimized bit-counting function) so we have to roll our own 5037 #ifndef CPU_COUNT 5038 5039 static int _cpu_count(const cpu_set_t* cpus) { 5040 int count = 0; 5041 // only look up to the number of configured processors 5042 for (int i = 0; i < os::processor_count(); i++) { 5043 if (CPU_ISSET(i, cpus)) { 5044 count++; 5045 } 5046 } 5047 return count; 5048 } 5049 5050 #define CPU_COUNT(cpus) _cpu_count(cpus) 5051 5052 #endif // CPU_COUNT 5053 5054 // Get the current number of available processors for this process. 5055 // This value can change at any time during a process's lifetime. 5056 // sched_getaffinity gives an accurate answer as it accounts for cpusets. 5057 // If it appears there may be more than 1024 processors then we do a 5058 // dynamic check - see 6515172 for details. 5059 // If anything goes wrong we fallback to returning the number of online 5060 // processors - which can be greater than the number available to the process. 5061 int os::Linux::active_processor_count() { 5062 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors 5063 cpu_set_t* cpus_p = &cpus; 5064 int cpus_size = sizeof(cpu_set_t); 5065 5066 int configured_cpus = os::processor_count(); // upper bound on available cpus 5067 int cpu_count = 0; 5068 5069 // old build platforms may not support dynamic cpu sets 5070 #ifdef CPU_ALLOC 5071 5072 // To enable easy testing of the dynamic path on different platforms we 5073 // introduce a diagnostic flag: UseCpuAllocPath 5074 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) { 5075 // kernel may use a mask bigger than cpu_set_t 5076 log_trace(os)("active_processor_count: using dynamic path %s" 5077 "- configured processors: %d", 5078 UseCpuAllocPath ? "(forced) " : "", 5079 configured_cpus); 5080 cpus_p = CPU_ALLOC(configured_cpus); 5081 if (cpus_p != NULL) { 5082 cpus_size = CPU_ALLOC_SIZE(configured_cpus); 5083 // zero it just to be safe 5084 CPU_ZERO_S(cpus_size, cpus_p); 5085 } 5086 else { 5087 // failed to allocate so fallback to online cpus 5088 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 5089 log_trace(os)("active_processor_count: " 5090 "CPU_ALLOC failed (%s) - using " 5091 "online processor count: %d", 5092 os::strerror(errno), online_cpus); 5093 return online_cpus; 5094 } 5095 } 5096 else { 5097 log_trace(os)("active_processor_count: using static path - configured processors: %d", 5098 configured_cpus); 5099 } 5100 #else // CPU_ALLOC 5101 // these stubs won't be executed 5102 #define CPU_COUNT_S(size, cpus) -1 5103 #define CPU_FREE(cpus) 5104 5105 log_trace(os)("active_processor_count: only static path available - configured processors: %d", 5106 configured_cpus); 5107 #endif // CPU_ALLOC 5108 5109 // pid 0 means the current thread - which we have to assume represents the process 5110 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) { 5111 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5112 cpu_count = CPU_COUNT_S(cpus_size, cpus_p); 5113 } 5114 else { 5115 cpu_count = CPU_COUNT(cpus_p); 5116 } 5117 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count); 5118 } 5119 else { 5120 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN); 5121 warning("sched_getaffinity failed (%s)- using online processor count (%d) " 5122 "which may exceed available processors", os::strerror(errno), cpu_count); 5123 } 5124 5125 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5126 CPU_FREE(cpus_p); 5127 } 5128 5129 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check"); 5130 return cpu_count; 5131 } 5132 5133 // Determine the active processor count from one of 5134 // three different sources: 5135 // 5136 // 1. User option -XX:ActiveProcessorCount 5137 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN) 5138 // 3. extracted from cgroup cpu subsystem (shares and quotas) 5139 // 5140 // Option 1, if specified, will always override. 5141 // If the cgroup subsystem is active and configured, we 5142 // will return the min of the cgroup and option 2 results. 5143 // This is required since tools, such as numactl, that 5144 // alter cpu affinity do not update cgroup subsystem 5145 // cpuset configuration files. 5146 int os::active_processor_count() { 5147 // User has overridden the number of active processors 5148 if (ActiveProcessorCount > 0) { 5149 log_trace(os)("active_processor_count: " 5150 "active processor count set by user : %d", 5151 ActiveProcessorCount); 5152 return ActiveProcessorCount; 5153 } 5154 5155 int active_cpus; 5156 if (OSContainer::is_containerized()) { 5157 active_cpus = OSContainer::active_processor_count(); 5158 log_trace(os)("active_processor_count: determined by OSContainer: %d", 5159 active_cpus); 5160 } else { 5161 active_cpus = os::Linux::active_processor_count(); 5162 } 5163 5164 return active_cpus; 5165 } 5166 5167 void os::set_native_thread_name(const char *name) { 5168 if (Linux::_pthread_setname_np) { 5169 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 5170 snprintf(buf, sizeof(buf), "%s", name); 5171 buf[sizeof(buf) - 1] = '\0'; 5172 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 5173 // ERANGE should not happen; all other errors should just be ignored. 5174 assert(rc != ERANGE, "pthread_setname_np failed"); 5175 } 5176 } 5177 5178 bool os::distribute_processes(uint length, uint* distribution) { 5179 // Not yet implemented. 5180 return false; 5181 } 5182 5183 bool os::bind_to_processor(uint processor_id) { 5184 // Not yet implemented. 5185 return false; 5186 } 5187 5188 /// 5189 5190 void os::SuspendedThreadTask::internal_do_task() { 5191 if (do_suspend(_thread->osthread())) { 5192 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 5193 do_task(context); 5194 do_resume(_thread->osthread()); 5195 } 5196 } 5197 5198 //////////////////////////////////////////////////////////////////////////////// 5199 // debug support 5200 5201 bool os::find(address addr, outputStream* st) { 5202 Dl_info dlinfo; 5203 memset(&dlinfo, 0, sizeof(dlinfo)); 5204 if (dladdr(addr, &dlinfo) != 0) { 5205 st->print(PTR_FORMAT ": ", p2i(addr)); 5206 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5207 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname, 5208 p2i(addr) - p2i(dlinfo.dli_saddr)); 5209 } else if (dlinfo.dli_fbase != NULL) { 5210 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase)); 5211 } else { 5212 st->print("<absolute address>"); 5213 } 5214 if (dlinfo.dli_fname != NULL) { 5215 st->print(" in %s", dlinfo.dli_fname); 5216 } 5217 if (dlinfo.dli_fbase != NULL) { 5218 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase)); 5219 } 5220 st->cr(); 5221 5222 if (Verbose) { 5223 // decode some bytes around the PC 5224 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5225 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5226 address lowest = (address) dlinfo.dli_sname; 5227 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5228 if (begin < lowest) begin = lowest; 5229 Dl_info dlinfo2; 5230 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5231 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 5232 end = (address) dlinfo2.dli_saddr; 5233 } 5234 Disassembler::decode(begin, end, st); 5235 } 5236 return true; 5237 } 5238 return false; 5239 } 5240 5241 //////////////////////////////////////////////////////////////////////////////// 5242 // misc 5243 5244 // This does not do anything on Linux. This is basically a hook for being 5245 // able to use structured exception handling (thread-local exception filters) 5246 // on, e.g., Win32. 5247 void 5248 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method, 5249 JavaCallArguments* args, Thread* thread) { 5250 f(value, method, args, thread); 5251 } 5252 5253 void os::print_statistics() { 5254 } 5255 5256 bool os::message_box(const char* title, const char* message) { 5257 int i; 5258 fdStream err(defaultStream::error_fd()); 5259 for (i = 0; i < 78; i++) err.print_raw("="); 5260 err.cr(); 5261 err.print_raw_cr(title); 5262 for (i = 0; i < 78; i++) err.print_raw("-"); 5263 err.cr(); 5264 err.print_raw_cr(message); 5265 for (i = 0; i < 78; i++) err.print_raw("="); 5266 err.cr(); 5267 5268 char buf[16]; 5269 // Prevent process from exiting upon "read error" without consuming all CPU 5270 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5271 5272 return buf[0] == 'y' || buf[0] == 'Y'; 5273 } 5274 5275 int os::stat(const char *path, struct stat *sbuf) { 5276 char pathbuf[MAX_PATH]; 5277 if (strlen(path) > MAX_PATH - 1) { 5278 errno = ENAMETOOLONG; 5279 return -1; 5280 } 5281 os::native_path(strcpy(pathbuf, path)); 5282 return ::stat(pathbuf, sbuf); 5283 } 5284 5285 // Is a (classpath) directory empty? 5286 bool os::dir_is_empty(const char* path) { 5287 DIR *dir = NULL; 5288 struct dirent *ptr; 5289 5290 dir = opendir(path); 5291 if (dir == NULL) return true; 5292 5293 // Scan the directory 5294 bool result = true; 5295 char buf[sizeof(struct dirent) + MAX_PATH]; 5296 while (result && (ptr = ::readdir(dir)) != NULL) { 5297 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5298 result = false; 5299 } 5300 } 5301 closedir(dir); 5302 return result; 5303 } 5304 5305 // This code originates from JDK's sysOpen and open64_w 5306 // from src/solaris/hpi/src/system_md.c 5307 5308 int os::open(const char *path, int oflag, int mode) { 5309 if (strlen(path) > MAX_PATH - 1) { 5310 errno = ENAMETOOLONG; 5311 return -1; 5312 } 5313 5314 // All file descriptors that are opened in the Java process and not 5315 // specifically destined for a subprocess should have the close-on-exec 5316 // flag set. If we don't set it, then careless 3rd party native code 5317 // might fork and exec without closing all appropriate file descriptors 5318 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5319 // turn might: 5320 // 5321 // - cause end-of-file to fail to be detected on some file 5322 // descriptors, resulting in mysterious hangs, or 5323 // 5324 // - might cause an fopen in the subprocess to fail on a system 5325 // suffering from bug 1085341. 5326 // 5327 // (Yes, the default setting of the close-on-exec flag is a Unix 5328 // design flaw) 5329 // 5330 // See: 5331 // 1085341: 32-bit stdio routines should support file descriptors >255 5332 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5333 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5334 // 5335 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5336 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5337 // because it saves a system call and removes a small window where the flag 5338 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5339 // and we fall back to using FD_CLOEXEC (see below). 5340 #ifdef O_CLOEXEC 5341 oflag |= O_CLOEXEC; 5342 #endif 5343 5344 int fd = ::open64(path, oflag, mode); 5345 if (fd == -1) return -1; 5346 5347 //If the open succeeded, the file might still be a directory 5348 { 5349 struct stat64 buf64; 5350 int ret = ::fstat64(fd, &buf64); 5351 int st_mode = buf64.st_mode; 5352 5353 if (ret != -1) { 5354 if ((st_mode & S_IFMT) == S_IFDIR) { 5355 errno = EISDIR; 5356 ::close(fd); 5357 return -1; 5358 } 5359 } else { 5360 ::close(fd); 5361 return -1; 5362 } 5363 } 5364 5365 #ifdef FD_CLOEXEC 5366 // Validate that the use of the O_CLOEXEC flag on open above worked. 5367 // With recent kernels, we will perform this check exactly once. 5368 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5369 if (!O_CLOEXEC_is_known_to_work) { 5370 int flags = ::fcntl(fd, F_GETFD); 5371 if (flags != -1) { 5372 if ((flags & FD_CLOEXEC) != 0) 5373 O_CLOEXEC_is_known_to_work = 1; 5374 else 5375 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5376 } 5377 } 5378 #endif 5379 5380 return fd; 5381 } 5382 5383 5384 // create binary file, rewriting existing file if required 5385 int os::create_binary_file(const char* path, bool rewrite_existing) { 5386 int oflags = O_WRONLY | O_CREAT; 5387 if (!rewrite_existing) { 5388 oflags |= O_EXCL; 5389 } 5390 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5391 } 5392 5393 // return current position of file pointer 5394 jlong os::current_file_offset(int fd) { 5395 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5396 } 5397 5398 // move file pointer to the specified offset 5399 jlong os::seek_to_file_offset(int fd, jlong offset) { 5400 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5401 } 5402 5403 // This code originates from JDK's sysAvailable 5404 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5405 5406 int os::available(int fd, jlong *bytes) { 5407 jlong cur, end; 5408 int mode; 5409 struct stat64 buf64; 5410 5411 if (::fstat64(fd, &buf64) >= 0) { 5412 mode = buf64.st_mode; 5413 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5414 int n; 5415 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5416 *bytes = n; 5417 return 1; 5418 } 5419 } 5420 } 5421 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5422 return 0; 5423 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5424 return 0; 5425 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5426 return 0; 5427 } 5428 *bytes = end - cur; 5429 return 1; 5430 } 5431 5432 // Map a block of memory. 5433 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5434 char *addr, size_t bytes, bool read_only, 5435 bool allow_exec) { 5436 int prot; 5437 int flags = MAP_PRIVATE; 5438 5439 if (read_only) { 5440 prot = PROT_READ; 5441 } else { 5442 prot = PROT_READ | PROT_WRITE; 5443 } 5444 5445 if (allow_exec) { 5446 prot |= PROT_EXEC; 5447 } 5448 5449 if (addr != NULL) { 5450 flags |= MAP_FIXED; 5451 } 5452 5453 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5454 fd, file_offset); 5455 if (mapped_address == MAP_FAILED) { 5456 return NULL; 5457 } 5458 return mapped_address; 5459 } 5460 5461 5462 // Remap a block of memory. 5463 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5464 char *addr, size_t bytes, bool read_only, 5465 bool allow_exec) { 5466 // same as map_memory() on this OS 5467 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5468 allow_exec); 5469 } 5470 5471 5472 // Unmap a block of memory. 5473 bool os::pd_unmap_memory(char* addr, size_t bytes) { 5474 return munmap(addr, bytes) == 0; 5475 } 5476 5477 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5478 5479 static clockid_t thread_cpu_clockid(Thread* thread) { 5480 pthread_t tid = thread->osthread()->pthread_id(); 5481 clockid_t clockid; 5482 5483 // Get thread clockid 5484 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5485 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5486 return clockid; 5487 } 5488 5489 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5490 // are used by JVM M&M and JVMTI to get user+sys or user CPU time 5491 // of a thread. 5492 // 5493 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5494 // the fast estimate available on the platform. 5495 5496 jlong os::current_thread_cpu_time() { 5497 if (os::Linux::supports_fast_thread_cpu_time()) { 5498 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5499 } else { 5500 // return user + sys since the cost is the same 5501 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5502 } 5503 } 5504 5505 jlong os::thread_cpu_time(Thread* thread) { 5506 // consistent with what current_thread_cpu_time() returns 5507 if (os::Linux::supports_fast_thread_cpu_time()) { 5508 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5509 } else { 5510 return slow_thread_cpu_time(thread, true /* user + sys */); 5511 } 5512 } 5513 5514 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5515 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5516 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5517 } else { 5518 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5519 } 5520 } 5521 5522 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5523 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5524 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5525 } else { 5526 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5527 } 5528 } 5529 5530 // -1 on error. 5531 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5532 pid_t tid = thread->osthread()->thread_id(); 5533 char *s; 5534 char stat[2048]; 5535 int statlen; 5536 char proc_name[64]; 5537 int count; 5538 long sys_time, user_time; 5539 char cdummy; 5540 int idummy; 5541 long ldummy; 5542 FILE *fp; 5543 5544 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5545 fp = fopen(proc_name, "r"); 5546 if (fp == NULL) return -1; 5547 statlen = fread(stat, 1, 2047, fp); 5548 stat[statlen] = '\0'; 5549 fclose(fp); 5550 5551 // Skip pid and the command string. Note that we could be dealing with 5552 // weird command names, e.g. user could decide to rename java launcher 5553 // to "java 1.4.2 :)", then the stat file would look like 5554 // 1234 (java 1.4.2 :)) R ... ... 5555 // We don't really need to know the command string, just find the last 5556 // occurrence of ")" and then start parsing from there. See bug 4726580. 5557 s = strrchr(stat, ')'); 5558 if (s == NULL) return -1; 5559 5560 // Skip blank chars 5561 do { s++; } while (s && isspace(*s)); 5562 5563 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5564 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5565 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5566 &user_time, &sys_time); 5567 if (count != 13) return -1; 5568 if (user_sys_cpu_time) { 5569 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5570 } else { 5571 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5572 } 5573 } 5574 5575 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5576 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5577 info_ptr->may_skip_backward = false; // elapsed time not wall time 5578 info_ptr->may_skip_forward = false; // elapsed time not wall time 5579 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5580 } 5581 5582 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5583 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5584 info_ptr->may_skip_backward = false; // elapsed time not wall time 5585 info_ptr->may_skip_forward = false; // elapsed time not wall time 5586 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5587 } 5588 5589 bool os::is_thread_cpu_time_supported() { 5590 return true; 5591 } 5592 5593 // System loadavg support. Returns -1 if load average cannot be obtained. 5594 // Linux doesn't yet have a (official) notion of processor sets, 5595 // so just return the system wide load average. 5596 int os::loadavg(double loadavg[], int nelem) { 5597 return ::getloadavg(loadavg, nelem); 5598 } 5599 5600 void os::pause() { 5601 char filename[MAX_PATH]; 5602 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5603 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile); 5604 } else { 5605 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5606 } 5607 5608 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5609 if (fd != -1) { 5610 struct stat buf; 5611 ::close(fd); 5612 while (::stat(filename, &buf) == 0) { 5613 (void)::poll(NULL, 0, 100); 5614 } 5615 } else { 5616 jio_fprintf(stderr, 5617 "Could not open pause file '%s', continuing immediately.\n", filename); 5618 } 5619 } 5620 5621 extern char** environ; 5622 5623 // Run the specified command in a separate process. Return its exit value, 5624 // or -1 on failure (e.g. can't fork a new process). 5625 // Unlike system(), this function can be called from signal handler. It 5626 // doesn't block SIGINT et al. 5627 int os::fork_and_exec(char* cmd) { 5628 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5629 5630 pid_t pid = fork(); 5631 5632 if (pid < 0) { 5633 // fork failed 5634 return -1; 5635 5636 } else if (pid == 0) { 5637 // child process 5638 5639 execve("/bin/sh", (char* const*)argv, environ); 5640 5641 // execve failed 5642 _exit(-1); 5643 5644 } else { 5645 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5646 // care about the actual exit code, for now. 5647 5648 int status; 5649 5650 // Wait for the child process to exit. This returns immediately if 5651 // the child has already exited. */ 5652 while (waitpid(pid, &status, 0) < 0) { 5653 switch (errno) { 5654 case ECHILD: return 0; 5655 case EINTR: break; 5656 default: return -1; 5657 } 5658 } 5659 5660 if (WIFEXITED(status)) { 5661 // The child exited normally; get its exit code. 5662 return WEXITSTATUS(status); 5663 } else if (WIFSIGNALED(status)) { 5664 // The child exited because of a signal 5665 // The best value to return is 0x80 + signal number, 5666 // because that is what all Unix shells do, and because 5667 // it allows callers to distinguish between process exit and 5668 // process death by signal. 5669 return 0x80 + WTERMSIG(status); 5670 } else { 5671 // Unknown exit code; pass it through 5672 return status; 5673 } 5674 } 5675 } 5676 5677 // Get the default path to the core file 5678 // Returns the length of the string 5679 int os::get_core_path(char* buffer, size_t bufferSize) { 5680 /* 5681 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 5682 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 5683 */ 5684 const int core_pattern_len = 129; 5685 char core_pattern[core_pattern_len] = {0}; 5686 5687 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 5688 if (core_pattern_file == -1) { 5689 return -1; 5690 } 5691 5692 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 5693 ::close(core_pattern_file); 5694 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') { 5695 return -1; 5696 } 5697 if (core_pattern[ret-1] == '\n') { 5698 core_pattern[ret-1] = '\0'; 5699 } else { 5700 core_pattern[ret] = '\0'; 5701 } 5702 5703 char *pid_pos = strstr(core_pattern, "%p"); 5704 int written; 5705 5706 if (core_pattern[0] == '/') { 5707 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern); 5708 } else { 5709 char cwd[PATH_MAX]; 5710 5711 const char* p = get_current_directory(cwd, PATH_MAX); 5712 if (p == NULL) { 5713 return -1; 5714 } 5715 5716 if (core_pattern[0] == '|') { 5717 written = jio_snprintf(buffer, bufferSize, 5718 "\"%s\" (or dumping to %s/core.%d)", 5719 &core_pattern[1], p, current_process_id()); 5720 } else { 5721 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 5722 } 5723 } 5724 5725 if (written < 0) { 5726 return -1; 5727 } 5728 5729 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 5730 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 5731 5732 if (core_uses_pid_file != -1) { 5733 char core_uses_pid = 0; 5734 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 5735 ::close(core_uses_pid_file); 5736 5737 if (core_uses_pid == '1') { 5738 jio_snprintf(buffer + written, bufferSize - written, 5739 ".%d", current_process_id()); 5740 } 5741 } 5742 } 5743 5744 return strlen(buffer); 5745 } 5746 5747 bool os::start_debugging(char *buf, int buflen) { 5748 int len = (int)strlen(buf); 5749 char *p = &buf[len]; 5750 5751 jio_snprintf(p, buflen-len, 5752 "\n\n" 5753 "Do you want to debug the problem?\n\n" 5754 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n" 5755 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n" 5756 "Otherwise, press RETURN to abort...", 5757 os::current_process_id(), os::current_process_id(), 5758 os::current_thread_id(), os::current_thread_id()); 5759 5760 bool yes = os::message_box("Unexpected Error", buf); 5761 5762 if (yes) { 5763 // yes, user asked VM to launch debugger 5764 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d", 5765 os::current_process_id(), os::current_process_id()); 5766 5767 os::fork_and_exec(buf); 5768 yes = false; 5769 } 5770 return yes; 5771 } 5772 5773 5774 // Java/Compiler thread: 5775 // 5776 // Low memory addresses 5777 // P0 +------------------------+ 5778 // | |\ Java thread created by VM does not have glibc 5779 // | glibc guard page | - guard page, attached Java thread usually has 5780 // | |/ 1 glibc guard page. 5781 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 5782 // | |\ 5783 // | HotSpot Guard Pages | - red, yellow and reserved pages 5784 // | |/ 5785 // +------------------------+ JavaThread::stack_reserved_zone_base() 5786 // | |\ 5787 // | Normal Stack | - 5788 // | |/ 5789 // P2 +------------------------+ Thread::stack_base() 5790 // 5791 // Non-Java thread: 5792 // 5793 // Low memory addresses 5794 // P0 +------------------------+ 5795 // | |\ 5796 // | glibc guard page | - usually 1 page 5797 // | |/ 5798 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 5799 // | |\ 5800 // | Normal Stack | - 5801 // | |/ 5802 // P2 +------------------------+ Thread::stack_base() 5803 // 5804 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size 5805 // returned from pthread_attr_getstack(). 5806 // ** Due to NPTL implementation error, linux takes the glibc guard page out 5807 // of the stack size given in pthread_attr. We work around this for 5808 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.) 5809 // 5810 #ifndef ZERO 5811 static void current_stack_region(address * bottom, size_t * size) { 5812 if (os::is_primordial_thread()) { 5813 // primordial thread needs special handling because pthread_getattr_np() 5814 // may return bogus value. 5815 *bottom = os::Linux::initial_thread_stack_bottom(); 5816 *size = os::Linux::initial_thread_stack_size(); 5817 } else { 5818 pthread_attr_t attr; 5819 5820 int rslt = pthread_getattr_np(pthread_self(), &attr); 5821 5822 // JVM needs to know exact stack location, abort if it fails 5823 if (rslt != 0) { 5824 if (rslt == ENOMEM) { 5825 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np"); 5826 } else { 5827 fatal("pthread_getattr_np failed with error = %d", rslt); 5828 } 5829 } 5830 5831 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) { 5832 fatal("Cannot locate current stack attributes!"); 5833 } 5834 5835 // Work around NPTL stack guard error. 5836 size_t guard_size = 0; 5837 rslt = pthread_attr_getguardsize(&attr, &guard_size); 5838 if (rslt != 0) { 5839 fatal("pthread_attr_getguardsize failed with error = %d", rslt); 5840 } 5841 *bottom += guard_size; 5842 *size -= guard_size; 5843 5844 pthread_attr_destroy(&attr); 5845 5846 } 5847 assert(os::current_stack_pointer() >= *bottom && 5848 os::current_stack_pointer() < *bottom + *size, "just checking"); 5849 } 5850 5851 address os::current_stack_base() { 5852 address bottom; 5853 size_t size; 5854 current_stack_region(&bottom, &size); 5855 return (bottom + size); 5856 } 5857 5858 size_t os::current_stack_size() { 5859 // This stack size includes the usable stack and HotSpot guard pages 5860 // (for the threads that have Hotspot guard pages). 5861 address bottom; 5862 size_t size; 5863 current_stack_region(&bottom, &size); 5864 return size; 5865 } 5866 #endif 5867 5868 static inline struct timespec get_mtime(const char* filename) { 5869 struct stat st; 5870 int ret = os::stat(filename, &st); 5871 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno)); 5872 return st.st_mtim; 5873 } 5874 5875 int os::compare_file_modified_times(const char* file1, const char* file2) { 5876 struct timespec filetime1 = get_mtime(file1); 5877 struct timespec filetime2 = get_mtime(file2); 5878 int diff = filetime1.tv_sec - filetime2.tv_sec; 5879 if (diff == 0) { 5880 return filetime1.tv_nsec - filetime2.tv_nsec; 5881 } 5882 return diff; 5883 } 5884 5885 /////////////// Unit tests /////////////// 5886 5887 #ifndef PRODUCT 5888 5889 #define test_log(...) \ 5890 do { \ 5891 if (VerboseInternalVMTests) { \ 5892 tty->print_cr(__VA_ARGS__); \ 5893 tty->flush(); \ 5894 } \ 5895 } while (false) 5896 5897 class TestReserveMemorySpecial : AllStatic { 5898 public: 5899 static void small_page_write(void* addr, size_t size) { 5900 size_t page_size = os::vm_page_size(); 5901 5902 char* end = (char*)addr + size; 5903 for (char* p = (char*)addr; p < end; p += page_size) { 5904 *p = 1; 5905 } 5906 } 5907 5908 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 5909 if (!UseHugeTLBFS) { 5910 return; 5911 } 5912 5913 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 5914 5915 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 5916 5917 if (addr != NULL) { 5918 small_page_write(addr, size); 5919 5920 os::Linux::release_memory_special_huge_tlbfs(addr, size); 5921 } 5922 } 5923 5924 static void test_reserve_memory_special_huge_tlbfs_only() { 5925 if (!UseHugeTLBFS) { 5926 return; 5927 } 5928 5929 size_t lp = os::large_page_size(); 5930 5931 for (size_t size = lp; size <= lp * 10; size += lp) { 5932 test_reserve_memory_special_huge_tlbfs_only(size); 5933 } 5934 } 5935 5936 static void test_reserve_memory_special_huge_tlbfs_mixed() { 5937 size_t lp = os::large_page_size(); 5938 size_t ag = os::vm_allocation_granularity(); 5939 5940 // sizes to test 5941 const size_t sizes[] = { 5942 lp, lp + ag, lp + lp / 2, lp * 2, 5943 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2, 5944 lp * 10, lp * 10 + lp / 2 5945 }; 5946 const int num_sizes = sizeof(sizes) / sizeof(size_t); 5947 5948 // For each size/alignment combination, we test three scenarios: 5949 // 1) with req_addr == NULL 5950 // 2) with a non-null req_addr at which we expect to successfully allocate 5951 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we 5952 // expect the allocation to either fail or to ignore req_addr 5953 5954 // Pre-allocate two areas; they shall be as large as the largest allocation 5955 // and aligned to the largest alignment we will be testing. 5956 const size_t mapping_size = sizes[num_sizes - 1] * 2; 5957 char* const mapping1 = (char*) ::mmap(NULL, mapping_size, 5958 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 5959 -1, 0); 5960 assert(mapping1 != MAP_FAILED, "should work"); 5961 5962 char* const mapping2 = (char*) ::mmap(NULL, mapping_size, 5963 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 5964 -1, 0); 5965 assert(mapping2 != MAP_FAILED, "should work"); 5966 5967 // Unmap the first mapping, but leave the second mapping intact: the first 5968 // mapping will serve as a value for a "good" req_addr (case 2). The second 5969 // mapping, still intact, as "bad" req_addr (case 3). 5970 ::munmap(mapping1, mapping_size); 5971 5972 // Case 1 5973 test_log("%s, req_addr NULL:", __FUNCTION__); 5974 test_log("size align result"); 5975 5976 for (int i = 0; i < num_sizes; i++) { 5977 const size_t size = sizes[i]; 5978 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 5979 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 5980 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s", 5981 size, alignment, p2i(p), (p != NULL ? "" : "(failed)")); 5982 if (p != NULL) { 5983 assert(is_aligned(p, alignment), "must be"); 5984 small_page_write(p, size); 5985 os::Linux::release_memory_special_huge_tlbfs(p, size); 5986 } 5987 } 5988 } 5989 5990 // Case 2 5991 test_log("%s, req_addr non-NULL:", __FUNCTION__); 5992 test_log("size align req_addr result"); 5993 5994 for (int i = 0; i < num_sizes; i++) { 5995 const size_t size = sizes[i]; 5996 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 5997 char* const req_addr = align_up(mapping1, alignment); 5998 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 5999 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6000 size, alignment, p2i(req_addr), p2i(p), 6001 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)"))); 6002 if (p != NULL) { 6003 assert(p == req_addr, "must be"); 6004 small_page_write(p, size); 6005 os::Linux::release_memory_special_huge_tlbfs(p, size); 6006 } 6007 } 6008 } 6009 6010 // Case 3 6011 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__); 6012 test_log("size align req_addr result"); 6013 6014 for (int i = 0; i < num_sizes; i++) { 6015 const size_t size = sizes[i]; 6016 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6017 char* const req_addr = align_up(mapping2, alignment); 6018 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6019 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6020 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)"))); 6021 // as the area around req_addr contains already existing mappings, the API should always 6022 // return NULL (as per contract, it cannot return another address) 6023 assert(p == NULL, "must be"); 6024 } 6025 } 6026 6027 ::munmap(mapping2, mapping_size); 6028 6029 } 6030 6031 static void test_reserve_memory_special_huge_tlbfs() { 6032 if (!UseHugeTLBFS) { 6033 return; 6034 } 6035 6036 test_reserve_memory_special_huge_tlbfs_only(); 6037 test_reserve_memory_special_huge_tlbfs_mixed(); 6038 } 6039 6040 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6041 if (!UseSHM) { 6042 return; 6043 } 6044 6045 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6046 6047 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6048 6049 if (addr != NULL) { 6050 assert(is_aligned(addr, alignment), "Check"); 6051 assert(is_aligned(addr, os::large_page_size()), "Check"); 6052 6053 small_page_write(addr, size); 6054 6055 os::Linux::release_memory_special_shm(addr, size); 6056 } 6057 } 6058 6059 static void test_reserve_memory_special_shm() { 6060 size_t lp = os::large_page_size(); 6061 size_t ag = os::vm_allocation_granularity(); 6062 6063 for (size_t size = ag; size < lp * 3; size += ag) { 6064 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6065 test_reserve_memory_special_shm(size, alignment); 6066 } 6067 } 6068 } 6069 6070 static void test() { 6071 test_reserve_memory_special_huge_tlbfs(); 6072 test_reserve_memory_special_shm(); 6073 } 6074 }; 6075 6076 void TestReserveMemorySpecial_test() { 6077 TestReserveMemorySpecial::test(); 6078 } 6079 6080 #endif