1 /* 2 * Copyright (c) 1999, 2016, 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 "classfile/classLoader.hpp" 27 #include "classfile/systemDictionary.hpp" 28 #include "classfile/vmSymbols.hpp" 29 #include "code/icBuffer.hpp" 30 #include "code/vtableStubs.hpp" 31 #include "compiler/compileBroker.hpp" 32 #include "compiler/disassembler.hpp" 33 #include "interpreter/interpreter.hpp" 34 #include "jvm_linux.h" 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 "prims/jniFastGetField.hpp" 42 #include "prims/jvm.h" 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/timer.hpp" 63 #include "semaphore_posix.hpp" 64 #include "services/attachListener.hpp" 65 #include "services/memTracker.hpp" 66 #include "services/runtimeService.hpp" 67 #include "utilities/decoder.hpp" 68 #include "utilities/defaultStream.hpp" 69 #include "utilities/events.hpp" 70 #include "utilities/elfFile.hpp" 71 #include "utilities/growableArray.hpp" 72 #include "utilities/macros.hpp" 73 #include "utilities/vmError.hpp" 74 75 // put OS-includes here 76 # include <sys/types.h> 77 # include <sys/mman.h> 78 # include <sys/stat.h> 79 # include <sys/select.h> 80 # include <pthread.h> 81 # include <signal.h> 82 # include <errno.h> 83 # include <dlfcn.h> 84 # include <stdio.h> 85 # include <unistd.h> 86 # include <sys/resource.h> 87 # include <pthread.h> 88 # include <sys/stat.h> 89 # include <sys/time.h> 90 # include <sys/times.h> 91 # include <sys/utsname.h> 92 # include <sys/socket.h> 93 # include <sys/wait.h> 94 # include <pwd.h> 95 # include <poll.h> 96 # include <semaphore.h> 97 # include <fcntl.h> 98 # include <string.h> 99 # include <syscall.h> 100 # include <sys/sysinfo.h> 101 # include <gnu/libc-version.h> 102 # include <sys/ipc.h> 103 # include <sys/shm.h> 104 # include <link.h> 105 # include <stdint.h> 106 # include <inttypes.h> 107 # include <sys/ioctl.h> 108 109 #ifndef _GNU_SOURCE 110 #define _GNU_SOURCE 111 #include <sched.h> 112 #undef _GNU_SOURCE 113 #else 114 #include <sched.h> 115 #endif 116 117 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 118 // getrusage() is prepared to handle the associated failure. 119 #ifndef RUSAGE_THREAD 120 #define RUSAGE_THREAD (1) /* only the calling thread */ 121 #endif 122 123 #define MAX_PATH (2 * K) 124 125 #define MAX_SECS 100000000 126 127 // for timer info max values which include all bits 128 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 129 130 #define LARGEPAGES_BIT (1 << 6) 131 //////////////////////////////////////////////////////////////////////////////// 132 // global variables 133 julong os::Linux::_physical_memory = 0; 134 135 address os::Linux::_initial_thread_stack_bottom = NULL; 136 uintptr_t os::Linux::_initial_thread_stack_size = 0; 137 138 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 139 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 140 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL; 141 Mutex* os::Linux::_createThread_lock = NULL; 142 pthread_t os::Linux::_main_thread; 143 int os::Linux::_page_size = -1; 144 bool os::Linux::_supports_fast_thread_cpu_time = false; 145 uint32_t os::Linux::_os_version = 0; 146 const char * os::Linux::_glibc_version = NULL; 147 const char * os::Linux::_libpthread_version = NULL; 148 pthread_condattr_t os::Linux::_condattr[1]; 149 150 static jlong initial_time_count=0; 151 152 static int clock_tics_per_sec = 100; 153 154 // For diagnostics to print a message once. see run_periodic_checks 155 static sigset_t check_signal_done; 156 static bool check_signals = true; 157 158 // Signal number used to suspend/resume a thread 159 160 // do not use any signal number less than SIGSEGV, see 4355769 161 static int SR_signum = SIGUSR2; 162 sigset_t SR_sigset; 163 164 // Declarations 165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 166 167 // utility functions 168 169 static int SR_initialize(); 170 171 julong os::available_memory() { 172 return Linux::available_memory(); 173 } 174 175 julong os::Linux::available_memory() { 176 // values in struct sysinfo are "unsigned long" 177 struct sysinfo si; 178 sysinfo(&si); 179 180 return (julong)si.freeram * si.mem_unit; 181 } 182 183 julong os::physical_memory() { 184 return Linux::physical_memory(); 185 } 186 187 // Return true if user is running as root. 188 189 bool os::have_special_privileges() { 190 static bool init = false; 191 static bool privileges = false; 192 if (!init) { 193 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 194 init = true; 195 } 196 return privileges; 197 } 198 199 200 #ifndef SYS_gettid 201 // i386: 224, ia64: 1105, amd64: 186, sparc 143 202 #ifdef __ia64__ 203 #define SYS_gettid 1105 204 #else 205 #ifdef __i386__ 206 #define SYS_gettid 224 207 #else 208 #ifdef __amd64__ 209 #define SYS_gettid 186 210 #else 211 #ifdef __sparc__ 212 #define SYS_gettid 143 213 #else 214 #error define gettid for the arch 215 #endif 216 #endif 217 #endif 218 #endif 219 #endif 220 221 // Cpu architecture string 222 static char cpu_arch[] = HOTSPOT_LIB_ARCH; 223 224 225 // pid_t gettid() 226 // 227 // Returns the kernel thread id of the currently running thread. Kernel 228 // thread id is used to access /proc. 229 pid_t os::Linux::gettid() { 230 int rslt = syscall(SYS_gettid); 231 assert(rslt != -1, "must be."); // old linuxthreads implementation? 232 return (pid_t)rslt; 233 } 234 235 // Most versions of linux have a bug where the number of processors are 236 // determined by looking at the /proc file system. In a chroot environment, 237 // the system call returns 1. This causes the VM to act as if it is 238 // a single processor and elide locking (see is_MP() call). 239 static bool unsafe_chroot_detected = false; 240 static const char *unstable_chroot_error = "/proc file system not found.\n" 241 "Java may be unstable running multithreaded in a chroot " 242 "environment on Linux when /proc filesystem is not mounted."; 243 244 void os::Linux::initialize_system_info() { 245 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 246 if (processor_count() == 1) { 247 pid_t pid = os::Linux::gettid(); 248 char fname[32]; 249 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 250 FILE *fp = fopen(fname, "r"); 251 if (fp == NULL) { 252 unsafe_chroot_detected = true; 253 } else { 254 fclose(fp); 255 } 256 } 257 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 258 assert(processor_count() > 0, "linux error"); 259 } 260 261 void os::init_system_properties_values() { 262 // The next steps are taken in the product version: 263 // 264 // Obtain the JAVA_HOME value from the location of libjvm.so. 265 // This library should be located at: 266 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 267 // 268 // If "/jre/lib/" appears at the right place in the path, then we 269 // assume libjvm.so is installed in a JDK and we use this path. 270 // 271 // Otherwise exit with message: "Could not create the Java virtual machine." 272 // 273 // The following extra steps are taken in the debugging version: 274 // 275 // If "/jre/lib/" does NOT appear at the right place in the path 276 // instead of exit check for $JAVA_HOME environment variable. 277 // 278 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 279 // then we append a fake suffix "hotspot/libjvm.so" to this path so 280 // it looks like libjvm.so is installed there 281 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 282 // 283 // Otherwise exit. 284 // 285 // Important note: if the location of libjvm.so changes this 286 // code needs to be changed accordingly. 287 288 // See ld(1): 289 // The linker uses the following search paths to locate required 290 // shared libraries: 291 // 1: ... 292 // ... 293 // 7: The default directories, normally /lib and /usr/lib. 294 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 295 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 296 #else 297 #define DEFAULT_LIBPATH "/lib:/usr/lib" 298 #endif 299 300 // Base path of extensions installed on the system. 301 #define SYS_EXT_DIR "/usr/java/packages" 302 #define EXTENSIONS_DIR "/lib/ext" 303 304 // Buffer that fits several sprintfs. 305 // Note that the space for the colon and the trailing null are provided 306 // by the nulls included by the sizeof operator. 307 const size_t bufsize = 308 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 309 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 310 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 311 312 // sysclasspath, java_home, dll_dir 313 { 314 char *pslash; 315 os::jvm_path(buf, bufsize); 316 317 // Found the full path to libjvm.so. 318 // Now cut the path to <java_home>/jre if we can. 319 pslash = strrchr(buf, '/'); 320 if (pslash != NULL) { 321 *pslash = '\0'; // Get rid of /libjvm.so. 322 } 323 pslash = strrchr(buf, '/'); 324 if (pslash != NULL) { 325 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 326 } 327 Arguments::set_dll_dir(buf); 328 329 if (pslash != NULL) { 330 pslash = strrchr(buf, '/'); 331 if (pslash != NULL) { 332 *pslash = '\0'; // Get rid of /<arch>. 333 pslash = strrchr(buf, '/'); 334 if (pslash != NULL) { 335 *pslash = '\0'; // Get rid of /lib. 336 } 337 } 338 } 339 Arguments::set_java_home(buf); 340 set_boot_path('/', ':'); 341 } 342 343 // Where to look for native libraries. 344 // 345 // Note: Due to a legacy implementation, most of the library path 346 // is set in the launcher. This was to accomodate linking restrictions 347 // on legacy Linux implementations (which are no longer supported). 348 // Eventually, all the library path setting will be done here. 349 // 350 // However, to prevent the proliferation of improperly built native 351 // libraries, the new path component /usr/java/packages is added here. 352 // Eventually, all the library path setting will be done here. 353 { 354 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 355 // should always exist (until the legacy problem cited above is 356 // addressed). 357 const char *v = ::getenv("LD_LIBRARY_PATH"); 358 const char *v_colon = ":"; 359 if (v == NULL) { v = ""; v_colon = ""; } 360 // That's +1 for the colon and +1 for the trailing '\0'. 361 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 362 strlen(v) + 1 + 363 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1, 364 mtInternal); 365 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch); 366 Arguments::set_library_path(ld_library_path); 367 FREE_C_HEAP_ARRAY(char, ld_library_path); 368 } 369 370 // Extensions directories. 371 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 372 Arguments::set_ext_dirs(buf); 373 374 FREE_C_HEAP_ARRAY(char, buf); 375 376 #undef DEFAULT_LIBPATH 377 #undef SYS_EXT_DIR 378 #undef EXTENSIONS_DIR 379 } 380 381 //////////////////////////////////////////////////////////////////////////////// 382 // breakpoint support 383 384 void os::breakpoint() { 385 BREAKPOINT; 386 } 387 388 extern "C" void breakpoint() { 389 // use debugger to set breakpoint here 390 } 391 392 //////////////////////////////////////////////////////////////////////////////// 393 // signal support 394 395 debug_only(static bool signal_sets_initialized = false); 396 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 397 398 bool os::Linux::is_sig_ignored(int sig) { 399 struct sigaction oact; 400 sigaction(sig, (struct sigaction*)NULL, &oact); 401 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 402 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 403 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) { 404 return true; 405 } else { 406 return false; 407 } 408 } 409 410 void os::Linux::signal_sets_init() { 411 // Should also have an assertion stating we are still single-threaded. 412 assert(!signal_sets_initialized, "Already initialized"); 413 // Fill in signals that are necessarily unblocked for all threads in 414 // the VM. Currently, we unblock the following signals: 415 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 416 // by -Xrs (=ReduceSignalUsage)); 417 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 418 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 419 // the dispositions or masks wrt these signals. 420 // Programs embedding the VM that want to use the above signals for their 421 // own purposes must, at this time, use the "-Xrs" option to prevent 422 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 423 // (See bug 4345157, and other related bugs). 424 // In reality, though, unblocking these signals is really a nop, since 425 // these signals are not blocked by default. 426 sigemptyset(&unblocked_sigs); 427 sigemptyset(&allowdebug_blocked_sigs); 428 sigaddset(&unblocked_sigs, SIGILL); 429 sigaddset(&unblocked_sigs, SIGSEGV); 430 sigaddset(&unblocked_sigs, SIGBUS); 431 sigaddset(&unblocked_sigs, SIGFPE); 432 #if defined(PPC64) 433 sigaddset(&unblocked_sigs, SIGTRAP); 434 #endif 435 sigaddset(&unblocked_sigs, SR_signum); 436 437 if (!ReduceSignalUsage) { 438 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 439 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 440 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 441 } 442 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 443 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 444 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 445 } 446 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 447 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 448 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 449 } 450 } 451 // Fill in signals that are blocked by all but the VM thread. 452 sigemptyset(&vm_sigs); 453 if (!ReduceSignalUsage) { 454 sigaddset(&vm_sigs, BREAK_SIGNAL); 455 } 456 debug_only(signal_sets_initialized = true); 457 458 } 459 460 // These are signals that are unblocked while a thread is running Java. 461 // (For some reason, they get blocked by default.) 462 sigset_t* os::Linux::unblocked_signals() { 463 assert(signal_sets_initialized, "Not initialized"); 464 return &unblocked_sigs; 465 } 466 467 // These are the signals that are blocked while a (non-VM) thread is 468 // running Java. Only the VM thread handles these signals. 469 sigset_t* os::Linux::vm_signals() { 470 assert(signal_sets_initialized, "Not initialized"); 471 return &vm_sigs; 472 } 473 474 // These are signals that are blocked during cond_wait to allow debugger in 475 sigset_t* os::Linux::allowdebug_blocked_signals() { 476 assert(signal_sets_initialized, "Not initialized"); 477 return &allowdebug_blocked_sigs; 478 } 479 480 void os::Linux::hotspot_sigmask(Thread* thread) { 481 482 //Save caller's signal mask before setting VM signal mask 483 sigset_t caller_sigmask; 484 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 485 486 OSThread* osthread = thread->osthread(); 487 osthread->set_caller_sigmask(caller_sigmask); 488 489 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 490 491 if (!ReduceSignalUsage) { 492 if (thread->is_VM_thread()) { 493 // Only the VM thread handles BREAK_SIGNAL ... 494 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 495 } else { 496 // ... all other threads block BREAK_SIGNAL 497 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 498 } 499 } 500 } 501 502 ////////////////////////////////////////////////////////////////////////////// 503 // detecting pthread library 504 505 void os::Linux::libpthread_init() { 506 // Save glibc and pthread version strings. 507 #if !defined(_CS_GNU_LIBC_VERSION) || \ 508 !defined(_CS_GNU_LIBPTHREAD_VERSION) 509 #error "glibc too old (< 2.3.2)" 510 #endif 511 512 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 513 assert(n > 0, "cannot retrieve glibc version"); 514 char *str = (char *)malloc(n, mtInternal); 515 confstr(_CS_GNU_LIBC_VERSION, str, n); 516 os::Linux::set_glibc_version(str); 517 518 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 519 assert(n > 0, "cannot retrieve pthread version"); 520 str = (char *)malloc(n, mtInternal); 521 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 522 os::Linux::set_libpthread_version(str); 523 } 524 525 ///////////////////////////////////////////////////////////////////////////// 526 // thread stack expansion 527 528 // os::Linux::manually_expand_stack() takes care of expanding the thread 529 // stack. Note that this is normally not needed: pthread stacks allocate 530 // thread stack using mmap() without MAP_NORESERVE, so the stack is already 531 // committed. Therefore it is not necessary to expand the stack manually. 532 // 533 // Manually expanding the stack was historically needed on LinuxThreads 534 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays 535 // it is kept to deal with very rare corner cases: 536 // 537 // For one, user may run the VM on an own implementation of threads 538 // whose stacks are - like the old LinuxThreads - implemented using 539 // mmap(MAP_GROWSDOWN). 540 // 541 // Also, this coding may be needed if the VM is running on the primordial 542 // thread. Normally we avoid running on the primordial thread; however, 543 // user may still invoke the VM on the primordial thread. 544 // 545 // The following historical comment describes the details about running 546 // on a thread stack allocated with mmap(MAP_GROWSDOWN): 547 548 549 // Force Linux kernel to expand current thread stack. If "bottom" is close 550 // to the stack guard, caller should block all signals. 551 // 552 // MAP_GROWSDOWN: 553 // A special mmap() flag that is used to implement thread stacks. It tells 554 // kernel that the memory region should extend downwards when needed. This 555 // allows early versions of LinuxThreads to only mmap the first few pages 556 // when creating a new thread. Linux kernel will automatically expand thread 557 // stack as needed (on page faults). 558 // 559 // However, because the memory region of a MAP_GROWSDOWN stack can grow on 560 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 561 // region, it's hard to tell if the fault is due to a legitimate stack 562 // access or because of reading/writing non-exist memory (e.g. buffer 563 // overrun). As a rule, if the fault happens below current stack pointer, 564 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the 565 // application (see Linux kernel fault.c). 566 // 567 // This Linux feature can cause SIGSEGV when VM bangs thread stack for 568 // stack overflow detection. 569 // 570 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 571 // not use MAP_GROWSDOWN. 572 // 573 // To get around the problem and allow stack banging on Linux, we need to 574 // manually expand thread stack after receiving the SIGSEGV. 575 // 576 // There are two ways to expand thread stack to address "bottom", we used 577 // both of them in JVM before 1.5: 578 // 1. adjust stack pointer first so that it is below "bottom", and then 579 // touch "bottom" 580 // 2. mmap() the page in question 581 // 582 // Now alternate signal stack is gone, it's harder to use 2. For instance, 583 // if current sp is already near the lower end of page 101, and we need to 584 // call mmap() to map page 100, it is possible that part of the mmap() frame 585 // will be placed in page 100. When page 100 is mapped, it is zero-filled. 586 // That will destroy the mmap() frame and cause VM to crash. 587 // 588 // The following code works by adjusting sp first, then accessing the "bottom" 589 // page to force a page fault. Linux kernel will then automatically expand the 590 // stack mapping. 591 // 592 // _expand_stack_to() assumes its frame size is less than page size, which 593 // should always be true if the function is not inlined. 594 595 static void NOINLINE _expand_stack_to(address bottom) { 596 address sp; 597 size_t size; 598 volatile char *p; 599 600 // Adjust bottom to point to the largest address within the same page, it 601 // gives us a one-page buffer if alloca() allocates slightly more memory. 602 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 603 bottom += os::Linux::page_size() - 1; 604 605 // sp might be slightly above current stack pointer; if that's the case, we 606 // will alloca() a little more space than necessary, which is OK. Don't use 607 // os::current_stack_pointer(), as its result can be slightly below current 608 // stack pointer, causing us to not alloca enough to reach "bottom". 609 sp = (address)&sp; 610 611 if (sp > bottom) { 612 size = sp - bottom; 613 p = (volatile char *)alloca(size); 614 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 615 p[0] = '\0'; 616 } 617 } 618 619 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 620 assert(t!=NULL, "just checking"); 621 assert(t->osthread()->expanding_stack(), "expand should be set"); 622 assert(t->stack_base() != NULL, "stack_base was not initialized"); 623 624 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) { 625 sigset_t mask_all, old_sigset; 626 sigfillset(&mask_all); 627 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 628 _expand_stack_to(addr); 629 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 630 return true; 631 } 632 return false; 633 } 634 635 ////////////////////////////////////////////////////////////////////////////// 636 // create new thread 637 638 // Thread start routine for all newly created threads 639 static void *thread_native_entry(Thread *thread) { 640 // Try to randomize the cache line index of hot stack frames. 641 // This helps when threads of the same stack traces evict each other's 642 // cache lines. The threads can be either from the same JVM instance, or 643 // from different JVM instances. The benefit is especially true for 644 // processors with hyperthreading technology. 645 static int counter = 0; 646 int pid = os::current_process_id(); 647 alloca(((pid ^ counter++) & 7) * 128); 648 649 thread->initialize_thread_current(); 650 651 OSThread* osthread = thread->osthread(); 652 Monitor* sync = osthread->startThread_lock(); 653 654 osthread->set_thread_id(os::current_thread_id()); 655 656 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 657 os::current_thread_id(), (uintx) pthread_self()); 658 659 if (UseNUMA) { 660 int lgrp_id = os::numa_get_group_id(); 661 if (lgrp_id != -1) { 662 thread->set_lgrp_id(lgrp_id); 663 } 664 } 665 // initialize signal mask for this thread 666 os::Linux::hotspot_sigmask(thread); 667 668 // initialize floating point control register 669 os::Linux::init_thread_fpu_state(); 670 671 // handshaking with parent thread 672 { 673 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 674 675 // notify parent thread 676 osthread->set_state(INITIALIZED); 677 sync->notify_all(); 678 679 // wait until os::start_thread() 680 while (osthread->get_state() == INITIALIZED) { 681 sync->wait(Mutex::_no_safepoint_check_flag); 682 } 683 } 684 685 // call one more level start routine 686 thread->run(); 687 688 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 689 os::current_thread_id(), (uintx) pthread_self()); 690 691 // If a thread has not deleted itself ("delete this") as part of its 692 // termination sequence, we have to ensure thread-local-storage is 693 // cleared before we actually terminate. No threads should ever be 694 // deleted asynchronously with respect to their termination. 695 if (Thread::current_or_null_safe() != NULL) { 696 assert(Thread::current_or_null_safe() == thread, "current thread is wrong"); 697 thread->clear_thread_current(); 698 } 699 700 return 0; 701 } 702 703 bool os::create_thread(Thread* thread, ThreadType thr_type, 704 size_t req_stack_size) { 705 assert(thread->osthread() == NULL, "caller responsible"); 706 707 // Allocate the OSThread object 708 OSThread* osthread = new OSThread(NULL, NULL); 709 if (osthread == NULL) { 710 return false; 711 } 712 713 // set the correct thread state 714 osthread->set_thread_type(thr_type); 715 716 // Initial state is ALLOCATED but not INITIALIZED 717 osthread->set_state(ALLOCATED); 718 719 thread->set_osthread(osthread); 720 721 // init thread attributes 722 pthread_attr_t attr; 723 pthread_attr_init(&attr); 724 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 725 726 // calculate stack size if it's not specified by caller 727 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size); 728 pthread_attr_setstacksize(&attr, stack_size); 729 730 // glibc guard page 731 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 732 733 ThreadState state; 734 735 { 736 pthread_t tid; 737 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread); 738 739 char buf[64]; 740 if (ret == 0) { 741 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ", 742 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 743 } else { 744 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.", 745 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 746 } 747 748 pthread_attr_destroy(&attr); 749 750 if (ret != 0) { 751 // Need to clean up stuff we've allocated so far 752 thread->set_osthread(NULL); 753 delete osthread; 754 return false; 755 } 756 757 // Store pthread info into the OSThread 758 osthread->set_pthread_id(tid); 759 760 // Wait until child thread is either initialized or aborted 761 { 762 Monitor* sync_with_child = osthread->startThread_lock(); 763 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 764 while ((state = osthread->get_state()) == ALLOCATED) { 765 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 766 } 767 } 768 } 769 770 // Aborted due to thread limit being reached 771 if (state == ZOMBIE) { 772 thread->set_osthread(NULL); 773 delete osthread; 774 return false; 775 } 776 777 // The thread is returned suspended (in state INITIALIZED), 778 // and is started higher up in the call chain 779 assert(state == INITIALIZED, "race condition"); 780 return true; 781 } 782 783 ///////////////////////////////////////////////////////////////////////////// 784 // attach existing thread 785 786 // bootstrap the main thread 787 bool os::create_main_thread(JavaThread* thread) { 788 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 789 return create_attached_thread(thread); 790 } 791 792 bool os::create_attached_thread(JavaThread* thread) { 793 #ifdef ASSERT 794 thread->verify_not_published(); 795 #endif 796 797 // Allocate the OSThread object 798 OSThread* osthread = new OSThread(NULL, NULL); 799 800 if (osthread == NULL) { 801 return false; 802 } 803 804 // Store pthread info into the OSThread 805 osthread->set_thread_id(os::Linux::gettid()); 806 osthread->set_pthread_id(::pthread_self()); 807 808 // initialize floating point control register 809 os::Linux::init_thread_fpu_state(); 810 811 // Initial thread state is RUNNABLE 812 osthread->set_state(RUNNABLE); 813 814 thread->set_osthread(osthread); 815 816 if (UseNUMA) { 817 int lgrp_id = os::numa_get_group_id(); 818 if (lgrp_id != -1) { 819 thread->set_lgrp_id(lgrp_id); 820 } 821 } 822 823 if (os::Linux::is_initial_thread()) { 824 // If current thread is initial thread, its stack is mapped on demand, 825 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 826 // the entire stack region to avoid SEGV in stack banging. 827 // It is also useful to get around the heap-stack-gap problem on SuSE 828 // kernel (see 4821821 for details). We first expand stack to the top 829 // of yellow zone, then enable stack yellow zone (order is significant, 830 // enabling yellow zone first will crash JVM on SuSE Linux), so there 831 // is no gap between the last two virtual memory regions. 832 833 JavaThread *jt = (JavaThread *)thread; 834 address addr = jt->stack_reserved_zone_base(); 835 assert(addr != NULL, "initialization problem?"); 836 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 837 838 osthread->set_expanding_stack(); 839 os::Linux::manually_expand_stack(jt, addr); 840 osthread->clear_expanding_stack(); 841 } 842 843 // initialize signal mask for this thread 844 // and save the caller's signal mask 845 os::Linux::hotspot_sigmask(thread); 846 847 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 848 os::current_thread_id(), (uintx) pthread_self()); 849 850 return true; 851 } 852 853 void os::pd_start_thread(Thread* thread) { 854 OSThread * osthread = thread->osthread(); 855 assert(osthread->get_state() != INITIALIZED, "just checking"); 856 Monitor* sync_with_child = osthread->startThread_lock(); 857 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 858 sync_with_child->notify(); 859 } 860 861 // Free Linux resources related to the OSThread 862 void os::free_thread(OSThread* osthread) { 863 assert(osthread != NULL, "osthread not set"); 864 865 // We are told to free resources of the argument thread, 866 // but we can only really operate on the current thread. 867 assert(Thread::current()->osthread() == osthread, 868 "os::free_thread but not current thread"); 869 870 #ifdef ASSERT 871 sigset_t current; 872 sigemptyset(¤t); 873 pthread_sigmask(SIG_SETMASK, NULL, ¤t); 874 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!"); 875 #endif 876 877 // Restore caller's signal mask 878 sigset_t sigmask = osthread->caller_sigmask(); 879 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 880 881 delete osthread; 882 } 883 884 ////////////////////////////////////////////////////////////////////////////// 885 // initial thread 886 887 // Check if current thread is the initial thread, similar to Solaris thr_main. 888 bool os::Linux::is_initial_thread(void) { 889 char dummy; 890 // If called before init complete, thread stack bottom will be null. 891 // Can be called if fatal error occurs before initialization. 892 if (initial_thread_stack_bottom() == NULL) return false; 893 assert(initial_thread_stack_bottom() != NULL && 894 initial_thread_stack_size() != 0, 895 "os::init did not locate initial thread's stack region"); 896 if ((address)&dummy >= initial_thread_stack_bottom() && 897 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) { 898 return true; 899 } else { 900 return false; 901 } 902 } 903 904 // Find the virtual memory area that contains addr 905 static bool find_vma(address addr, address* vma_low, address* vma_high) { 906 FILE *fp = fopen("/proc/self/maps", "r"); 907 if (fp) { 908 address low, high; 909 while (!feof(fp)) { 910 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 911 if (low <= addr && addr < high) { 912 if (vma_low) *vma_low = low; 913 if (vma_high) *vma_high = high; 914 fclose(fp); 915 return true; 916 } 917 } 918 for (;;) { 919 int ch = fgetc(fp); 920 if (ch == EOF || ch == (int)'\n') break; 921 } 922 } 923 fclose(fp); 924 } 925 return false; 926 } 927 928 // Locate initial thread stack. This special handling of initial thread stack 929 // is needed because pthread_getattr_np() on most (all?) Linux distros returns 930 // bogus value for initial thread. 931 void os::Linux::capture_initial_stack(size_t max_size) { 932 // stack size is the easy part, get it from RLIMIT_STACK 933 struct rlimit rlim; 934 getrlimit(RLIMIT_STACK, &rlim); 935 size_t stack_size = rlim.rlim_cur; 936 937 // 6308388: a bug in ld.so will relocate its own .data section to the 938 // lower end of primordial stack; reduce ulimit -s value a little bit 939 // so we won't install guard page on ld.so's data section. 940 stack_size -= 2 * page_size(); 941 942 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 943 // 7.1, in both cases we will get 2G in return value. 944 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 945 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 946 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 947 // in case other parts in glibc still assumes 2M max stack size. 948 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 949 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 950 if (stack_size > 2 * K * K IA64_ONLY(*2)) { 951 stack_size = 2 * K * K IA64_ONLY(*2); 952 } 953 // Try to figure out where the stack base (top) is. This is harder. 954 // 955 // When an application is started, glibc saves the initial stack pointer in 956 // a global variable "__libc_stack_end", which is then used by system 957 // libraries. __libc_stack_end should be pretty close to stack top. The 958 // variable is available since the very early days. However, because it is 959 // a private interface, it could disappear in the future. 960 // 961 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 962 // to __libc_stack_end, it is very close to stack top, but isn't the real 963 // stack top. Note that /proc may not exist if VM is running as a chroot 964 // program, so reading /proc/<pid>/stat could fail. Also the contents of 965 // /proc/<pid>/stat could change in the future (though unlikely). 966 // 967 // We try __libc_stack_end first. If that doesn't work, look for 968 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 969 // as a hint, which should work well in most cases. 970 971 uintptr_t stack_start; 972 973 // try __libc_stack_end first 974 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 975 if (p && *p) { 976 stack_start = *p; 977 } else { 978 // see if we can get the start_stack field from /proc/self/stat 979 FILE *fp; 980 int pid; 981 char state; 982 int ppid; 983 int pgrp; 984 int session; 985 int nr; 986 int tpgrp; 987 unsigned long flags; 988 unsigned long minflt; 989 unsigned long cminflt; 990 unsigned long majflt; 991 unsigned long cmajflt; 992 unsigned long utime; 993 unsigned long stime; 994 long cutime; 995 long cstime; 996 long prio; 997 long nice; 998 long junk; 999 long it_real; 1000 uintptr_t start; 1001 uintptr_t vsize; 1002 intptr_t rss; 1003 uintptr_t rsslim; 1004 uintptr_t scodes; 1005 uintptr_t ecode; 1006 int i; 1007 1008 // Figure what the primordial thread stack base is. Code is inspired 1009 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1010 // followed by command name surrounded by parentheses, state, etc. 1011 char stat[2048]; 1012 int statlen; 1013 1014 fp = fopen("/proc/self/stat", "r"); 1015 if (fp) { 1016 statlen = fread(stat, 1, 2047, fp); 1017 stat[statlen] = '\0'; 1018 fclose(fp); 1019 1020 // Skip pid and the command string. Note that we could be dealing with 1021 // weird command names, e.g. user could decide to rename java launcher 1022 // to "java 1.4.2 :)", then the stat file would look like 1023 // 1234 (java 1.4.2 :)) R ... ... 1024 // We don't really need to know the command string, just find the last 1025 // occurrence of ")" and then start parsing from there. See bug 4726580. 1026 char * s = strrchr(stat, ')'); 1027 1028 i = 0; 1029 if (s) { 1030 // Skip blank chars 1031 do { s++; } while (s && isspace(*s)); 1032 1033 #define _UFM UINTX_FORMAT 1034 #define _DFM INTX_FORMAT 1035 1036 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1037 // 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 1038 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, 1039 &state, // 3 %c 1040 &ppid, // 4 %d 1041 &pgrp, // 5 %d 1042 &session, // 6 %d 1043 &nr, // 7 %d 1044 &tpgrp, // 8 %d 1045 &flags, // 9 %lu 1046 &minflt, // 10 %lu 1047 &cminflt, // 11 %lu 1048 &majflt, // 12 %lu 1049 &cmajflt, // 13 %lu 1050 &utime, // 14 %lu 1051 &stime, // 15 %lu 1052 &cutime, // 16 %ld 1053 &cstime, // 17 %ld 1054 &prio, // 18 %ld 1055 &nice, // 19 %ld 1056 &junk, // 20 %ld 1057 &it_real, // 21 %ld 1058 &start, // 22 UINTX_FORMAT 1059 &vsize, // 23 UINTX_FORMAT 1060 &rss, // 24 INTX_FORMAT 1061 &rsslim, // 25 UINTX_FORMAT 1062 &scodes, // 26 UINTX_FORMAT 1063 &ecode, // 27 UINTX_FORMAT 1064 &stack_start); // 28 UINTX_FORMAT 1065 } 1066 1067 #undef _UFM 1068 #undef _DFM 1069 1070 if (i != 28 - 2) { 1071 assert(false, "Bad conversion from /proc/self/stat"); 1072 // product mode - assume we are the initial thread, good luck in the 1073 // embedded case. 1074 warning("Can't detect initial thread stack location - bad conversion"); 1075 stack_start = (uintptr_t) &rlim; 1076 } 1077 } else { 1078 // For some reason we can't open /proc/self/stat (for example, running on 1079 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1080 // most cases, so don't abort: 1081 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1082 stack_start = (uintptr_t) &rlim; 1083 } 1084 } 1085 1086 // Now we have a pointer (stack_start) very close to the stack top, the 1087 // next thing to do is to figure out the exact location of stack top. We 1088 // can find out the virtual memory area that contains stack_start by 1089 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1090 // and its upper limit is the real stack top. (again, this would fail if 1091 // running inside chroot, because /proc may not exist.) 1092 1093 uintptr_t stack_top; 1094 address low, high; 1095 if (find_vma((address)stack_start, &low, &high)) { 1096 // success, "high" is the true stack top. (ignore "low", because initial 1097 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1098 stack_top = (uintptr_t)high; 1099 } else { 1100 // failed, likely because /proc/self/maps does not exist 1101 warning("Can't detect initial thread stack location - find_vma failed"); 1102 // best effort: stack_start is normally within a few pages below the real 1103 // stack top, use it as stack top, and reduce stack size so we won't put 1104 // guard page outside stack. 1105 stack_top = stack_start; 1106 stack_size -= 16 * page_size(); 1107 } 1108 1109 // stack_top could be partially down the page so align it 1110 stack_top = align_size_up(stack_top, page_size()); 1111 1112 if (max_size && stack_size > max_size) { 1113 _initial_thread_stack_size = max_size; 1114 } else { 1115 _initial_thread_stack_size = stack_size; 1116 } 1117 1118 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1119 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1120 } 1121 1122 //////////////////////////////////////////////////////////////////////////////// 1123 // time support 1124 1125 // Time since start-up in seconds to a fine granularity. 1126 // Used by VMSelfDestructTimer and the MemProfiler. 1127 double os::elapsedTime() { 1128 1129 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1130 } 1131 1132 jlong os::elapsed_counter() { 1133 return javaTimeNanos() - initial_time_count; 1134 } 1135 1136 jlong os::elapsed_frequency() { 1137 return NANOSECS_PER_SEC; // nanosecond resolution 1138 } 1139 1140 bool os::supports_vtime() { return true; } 1141 bool os::enable_vtime() { return false; } 1142 bool os::vtime_enabled() { return false; } 1143 1144 double os::elapsedVTime() { 1145 struct rusage usage; 1146 int retval = getrusage(RUSAGE_THREAD, &usage); 1147 if (retval == 0) { 1148 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); 1149 } else { 1150 // better than nothing, but not much 1151 return elapsedTime(); 1152 } 1153 } 1154 1155 jlong os::javaTimeMillis() { 1156 timeval time; 1157 int status = gettimeofday(&time, NULL); 1158 assert(status != -1, "linux error"); 1159 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1160 } 1161 1162 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) { 1163 timeval time; 1164 int status = gettimeofday(&time, NULL); 1165 assert(status != -1, "linux error"); 1166 seconds = jlong(time.tv_sec); 1167 nanos = jlong(time.tv_usec) * 1000; 1168 } 1169 1170 1171 #ifndef CLOCK_MONOTONIC 1172 #define CLOCK_MONOTONIC (1) 1173 #endif 1174 1175 void os::Linux::clock_init() { 1176 // we do dlopen's in this particular order due to bug in linux 1177 // dynamical loader (see 6348968) leading to crash on exit 1178 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1179 if (handle == NULL) { 1180 handle = dlopen("librt.so", RTLD_LAZY); 1181 } 1182 1183 if (handle) { 1184 int (*clock_getres_func)(clockid_t, struct timespec*) = 1185 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1186 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1187 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1188 if (clock_getres_func && clock_gettime_func) { 1189 // See if monotonic clock is supported by the kernel. Note that some 1190 // early implementations simply return kernel jiffies (updated every 1191 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1192 // for nano time (though the monotonic property is still nice to have). 1193 // It's fixed in newer kernels, however clock_getres() still returns 1194 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1195 // resolution for now. Hopefully as people move to new kernels, this 1196 // won't be a problem. 1197 struct timespec res; 1198 struct timespec tp; 1199 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1200 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1201 // yes, monotonic clock is supported 1202 _clock_gettime = clock_gettime_func; 1203 return; 1204 } else { 1205 // close librt if there is no monotonic clock 1206 dlclose(handle); 1207 } 1208 } 1209 } 1210 warning("No monotonic clock was available - timed services may " \ 1211 "be adversely affected if the time-of-day clock changes"); 1212 } 1213 1214 #ifndef SYS_clock_getres 1215 #if defined(IA32) || defined(AMD64) 1216 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1217 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1218 #else 1219 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1220 #define sys_clock_getres(x,y) -1 1221 #endif 1222 #else 1223 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1224 #endif 1225 1226 void os::Linux::fast_thread_clock_init() { 1227 if (!UseLinuxPosixThreadCPUClocks) { 1228 return; 1229 } 1230 clockid_t clockid; 1231 struct timespec tp; 1232 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1233 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1234 1235 // Switch to using fast clocks for thread cpu time if 1236 // the sys_clock_getres() returns 0 error code. 1237 // Note, that some kernels may support the current thread 1238 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1239 // returned by the pthread_getcpuclockid(). 1240 // If the fast Posix clocks are supported then the sys_clock_getres() 1241 // must return at least tp.tv_sec == 0 which means a resolution 1242 // better than 1 sec. This is extra check for reliability. 1243 1244 if (pthread_getcpuclockid_func && 1245 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1246 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1247 _supports_fast_thread_cpu_time = true; 1248 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1249 } 1250 } 1251 1252 jlong os::javaTimeNanos() { 1253 if (os::supports_monotonic_clock()) { 1254 struct timespec tp; 1255 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1256 assert(status == 0, "gettime error"); 1257 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1258 return result; 1259 } else { 1260 timeval time; 1261 int status = gettimeofday(&time, NULL); 1262 assert(status != -1, "linux error"); 1263 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1264 return 1000 * usecs; 1265 } 1266 } 1267 1268 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1269 if (os::supports_monotonic_clock()) { 1270 info_ptr->max_value = ALL_64_BITS; 1271 1272 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1273 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1274 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1275 } else { 1276 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1277 info_ptr->max_value = ALL_64_BITS; 1278 1279 // gettimeofday is a real time clock so it skips 1280 info_ptr->may_skip_backward = true; 1281 info_ptr->may_skip_forward = true; 1282 } 1283 1284 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1285 } 1286 1287 // Return the real, user, and system times in seconds from an 1288 // arbitrary fixed point in the past. 1289 bool os::getTimesSecs(double* process_real_time, 1290 double* process_user_time, 1291 double* process_system_time) { 1292 struct tms ticks; 1293 clock_t real_ticks = times(&ticks); 1294 1295 if (real_ticks == (clock_t) (-1)) { 1296 return false; 1297 } else { 1298 double ticks_per_second = (double) clock_tics_per_sec; 1299 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1300 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1301 *process_real_time = ((double) real_ticks) / ticks_per_second; 1302 1303 return true; 1304 } 1305 } 1306 1307 1308 char * os::local_time_string(char *buf, size_t buflen) { 1309 struct tm t; 1310 time_t long_time; 1311 time(&long_time); 1312 localtime_r(&long_time, &t); 1313 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1314 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1315 t.tm_hour, t.tm_min, t.tm_sec); 1316 return buf; 1317 } 1318 1319 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1320 return localtime_r(clock, res); 1321 } 1322 1323 //////////////////////////////////////////////////////////////////////////////// 1324 // runtime exit support 1325 1326 // Note: os::shutdown() might be called very early during initialization, or 1327 // called from signal handler. Before adding something to os::shutdown(), make 1328 // sure it is async-safe and can handle partially initialized VM. 1329 void os::shutdown() { 1330 1331 // allow PerfMemory to attempt cleanup of any persistent resources 1332 perfMemory_exit(); 1333 1334 // needs to remove object in file system 1335 AttachListener::abort(); 1336 1337 // flush buffered output, finish log files 1338 ostream_abort(); 1339 1340 // Check for abort hook 1341 abort_hook_t abort_hook = Arguments::abort_hook(); 1342 if (abort_hook != NULL) { 1343 abort_hook(); 1344 } 1345 1346 } 1347 1348 // Note: os::abort() might be called very early during initialization, or 1349 // called from signal handler. Before adding something to os::abort(), make 1350 // sure it is async-safe and can handle partially initialized VM. 1351 void os::abort(bool dump_core, void* siginfo, const void* context) { 1352 os::shutdown(); 1353 if (dump_core) { 1354 #ifndef PRODUCT 1355 fdStream out(defaultStream::output_fd()); 1356 out.print_raw("Current thread is "); 1357 char buf[16]; 1358 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1359 out.print_raw_cr(buf); 1360 out.print_raw_cr("Dumping core ..."); 1361 #endif 1362 ::abort(); // dump core 1363 } 1364 1365 ::exit(1); 1366 } 1367 1368 // Die immediately, no exit hook, no abort hook, no cleanup. 1369 void os::die() { 1370 ::abort(); 1371 } 1372 1373 1374 // This method is a copy of JDK's sysGetLastErrorString 1375 // from src/solaris/hpi/src/system_md.c 1376 1377 size_t os::lasterror(char *buf, size_t len) { 1378 if (errno == 0) return 0; 1379 1380 const char *s = os::strerror(errno); 1381 size_t n = ::strlen(s); 1382 if (n >= len) { 1383 n = len - 1; 1384 } 1385 ::strncpy(buf, s, n); 1386 buf[n] = '\0'; 1387 return n; 1388 } 1389 1390 // thread_id is kernel thread id (similar to Solaris LWP id) 1391 intx os::current_thread_id() { return os::Linux::gettid(); } 1392 int os::current_process_id() { 1393 return ::getpid(); 1394 } 1395 1396 // DLL functions 1397 1398 const char* os::dll_file_extension() { return ".so"; } 1399 1400 // This must be hard coded because it's the system's temporary 1401 // directory not the java application's temp directory, ala java.io.tmpdir. 1402 const char* os::get_temp_directory() { return "/tmp"; } 1403 1404 static bool file_exists(const char* filename) { 1405 struct stat statbuf; 1406 if (filename == NULL || strlen(filename) == 0) { 1407 return false; 1408 } 1409 return os::stat(filename, &statbuf) == 0; 1410 } 1411 1412 bool os::dll_build_name(char* buffer, size_t buflen, 1413 const char* pname, const char* fname) { 1414 bool retval = false; 1415 // Copied from libhpi 1416 const size_t pnamelen = pname ? strlen(pname) : 0; 1417 1418 // Return error on buffer overflow. 1419 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1420 return retval; 1421 } 1422 1423 if (pnamelen == 0) { 1424 snprintf(buffer, buflen, "lib%s.so", fname); 1425 retval = true; 1426 } else if (strchr(pname, *os::path_separator()) != NULL) { 1427 int n; 1428 char** pelements = split_path(pname, &n); 1429 if (pelements == NULL) { 1430 return false; 1431 } 1432 for (int i = 0; i < n; i++) { 1433 // Really shouldn't be NULL, but check can't hurt 1434 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1435 continue; // skip the empty path values 1436 } 1437 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1438 if (file_exists(buffer)) { 1439 retval = true; 1440 break; 1441 } 1442 } 1443 // release the storage 1444 for (int i = 0; i < n; i++) { 1445 if (pelements[i] != NULL) { 1446 FREE_C_HEAP_ARRAY(char, pelements[i]); 1447 } 1448 } 1449 if (pelements != NULL) { 1450 FREE_C_HEAP_ARRAY(char*, pelements); 1451 } 1452 } else { 1453 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1454 retval = true; 1455 } 1456 return retval; 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 ElfFile ef(filename); 1633 if (!ef.specifies_noexecstack()) { 1634 if (!is_init_completed()) { 1635 os::Linux::_stack_is_executable = true; 1636 // This is OK - No Java threads have been created yet, and hence no 1637 // stack guard pages to fix. 1638 // 1639 // This should happen only when you are building JDK7 using a very 1640 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1641 // 1642 // Dynamic loader will make all stacks executable after 1643 // this function returns, and will not do that again. 1644 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1645 } else { 1646 warning("You have loaded library %s which might have disabled stack guard. " 1647 "The VM will try to fix the stack guard now.\n" 1648 "It's highly recommended that you fix the library with " 1649 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1650 filename); 1651 1652 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1653 JavaThread *jt = JavaThread::current(); 1654 if (jt->thread_state() != _thread_in_native) { 1655 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1656 // that requires ExecStack. Cannot enter safe point. Let's give up. 1657 warning("Unable to fix stack guard. Giving up."); 1658 } else { 1659 if (!LoadExecStackDllInVMThread) { 1660 // This is for the case where the DLL has an static 1661 // constructor function that executes JNI code. We cannot 1662 // load such DLLs in the VMThread. 1663 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1664 } 1665 1666 ThreadInVMfromNative tiv(jt); 1667 debug_only(VMNativeEntryWrapper vew;) 1668 1669 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1670 VMThread::execute(&op); 1671 if (LoadExecStackDllInVMThread) { 1672 result = op.loaded_library(); 1673 } 1674 load_attempted = true; 1675 } 1676 } 1677 } 1678 } 1679 1680 if (!load_attempted) { 1681 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1682 } 1683 1684 if (result != NULL) { 1685 // Successful loading 1686 return result; 1687 } 1688 1689 Elf32_Ehdr elf_head; 1690 int diag_msg_max_length=ebuflen-strlen(ebuf); 1691 char* diag_msg_buf=ebuf+strlen(ebuf); 1692 1693 if (diag_msg_max_length==0) { 1694 // No more space in ebuf for additional diagnostics message 1695 return NULL; 1696 } 1697 1698 1699 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1700 1701 if (file_descriptor < 0) { 1702 // Can't open library, report dlerror() message 1703 return NULL; 1704 } 1705 1706 bool failed_to_read_elf_head= 1707 (sizeof(elf_head)!= 1708 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1709 1710 ::close(file_descriptor); 1711 if (failed_to_read_elf_head) { 1712 // file i/o error - report dlerror() msg 1713 return NULL; 1714 } 1715 1716 typedef struct { 1717 Elf32_Half code; // Actual value as defined in elf.h 1718 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1719 unsigned char elf_class; // 32 or 64 bit 1720 unsigned char endianess; // MSB or LSB 1721 char* name; // String representation 1722 } arch_t; 1723 1724 #ifndef EM_486 1725 #define EM_486 6 /* Intel 80486 */ 1726 #endif 1727 #ifndef EM_AARCH64 1728 #define EM_AARCH64 183 /* ARM AARCH64 */ 1729 #endif 1730 1731 static const arch_t arch_array[]={ 1732 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1733 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1734 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1735 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1736 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1737 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1738 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1739 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1740 #if defined(VM_LITTLE_ENDIAN) 1741 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"}, 1742 #else 1743 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"}, 1744 #endif 1745 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1746 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1747 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1748 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1749 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1750 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1751 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1752 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1753 }; 1754 1755 #if (defined IA32) 1756 static Elf32_Half running_arch_code=EM_386; 1757 #elif (defined AMD64) 1758 static Elf32_Half running_arch_code=EM_X86_64; 1759 #elif (defined IA64) 1760 static Elf32_Half running_arch_code=EM_IA_64; 1761 #elif (defined __sparc) && (defined _LP64) 1762 static Elf32_Half running_arch_code=EM_SPARCV9; 1763 #elif (defined __sparc) && (!defined _LP64) 1764 static Elf32_Half running_arch_code=EM_SPARC; 1765 #elif (defined __powerpc64__) 1766 static Elf32_Half running_arch_code=EM_PPC64; 1767 #elif (defined __powerpc__) 1768 static Elf32_Half running_arch_code=EM_PPC; 1769 #elif (defined ARM) 1770 static Elf32_Half running_arch_code=EM_ARM; 1771 #elif (defined S390) 1772 static Elf32_Half running_arch_code=EM_S390; 1773 #elif (defined ALPHA) 1774 static Elf32_Half running_arch_code=EM_ALPHA; 1775 #elif (defined MIPSEL) 1776 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1777 #elif (defined PARISC) 1778 static Elf32_Half running_arch_code=EM_PARISC; 1779 #elif (defined MIPS) 1780 static Elf32_Half running_arch_code=EM_MIPS; 1781 #elif (defined M68K) 1782 static Elf32_Half running_arch_code=EM_68K; 1783 #elif (defined AARCH64) 1784 static Elf32_Half running_arch_code=EM_AARCH64; 1785 #else 1786 #error Method os::dll_load requires that one of following is defined:\ 1787 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64 1788 #endif 1789 1790 // Identify compatability class for VM's architecture and library's architecture 1791 // Obtain string descriptions for architectures 1792 1793 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1794 int running_arch_index=-1; 1795 1796 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1797 if (running_arch_code == arch_array[i].code) { 1798 running_arch_index = i; 1799 } 1800 if (lib_arch.code == arch_array[i].code) { 1801 lib_arch.compat_class = arch_array[i].compat_class; 1802 lib_arch.name = arch_array[i].name; 1803 } 1804 } 1805 1806 assert(running_arch_index != -1, 1807 "Didn't find running architecture code (running_arch_code) in arch_array"); 1808 if (running_arch_index == -1) { 1809 // Even though running architecture detection failed 1810 // we may still continue with reporting dlerror() message 1811 return NULL; 1812 } 1813 1814 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1815 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1816 return NULL; 1817 } 1818 1819 #ifndef S390 1820 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1821 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1822 return NULL; 1823 } 1824 #endif // !S390 1825 1826 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1827 if (lib_arch.name!=NULL) { 1828 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1829 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1830 lib_arch.name, arch_array[running_arch_index].name); 1831 } else { 1832 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1833 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1834 lib_arch.code, 1835 arch_array[running_arch_index].name); 1836 } 1837 } 1838 1839 return NULL; 1840 } 1841 1842 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 1843 int ebuflen) { 1844 void * result = ::dlopen(filename, RTLD_LAZY); 1845 if (result == NULL) { 1846 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 1847 ebuf[ebuflen-1] = '\0'; 1848 } 1849 return result; 1850 } 1851 1852 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 1853 int ebuflen) { 1854 void * result = NULL; 1855 if (LoadExecStackDllInVMThread) { 1856 result = dlopen_helper(filename, ebuf, ebuflen); 1857 } 1858 1859 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 1860 // library that requires an executable stack, or which does not have this 1861 // stack attribute set, dlopen changes the stack attribute to executable. The 1862 // read protection of the guard pages gets lost. 1863 // 1864 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 1865 // may have been queued at the same time. 1866 1867 if (!_stack_is_executable) { 1868 JavaThread *jt = Threads::first(); 1869 1870 while (jt) { 1871 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 1872 jt->stack_guards_enabled()) { // No pending stack overflow exceptions 1873 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) { 1874 warning("Attempt to reguard stack yellow zone failed."); 1875 } 1876 } 1877 jt = jt->next(); 1878 } 1879 } 1880 1881 return result; 1882 } 1883 1884 void* os::dll_lookup(void* handle, const char* name) { 1885 void* res = dlsym(handle, name); 1886 return res; 1887 } 1888 1889 void* os::get_default_process_handle() { 1890 return (void*)::dlopen(NULL, RTLD_LAZY); 1891 } 1892 1893 static bool _print_ascii_file(const char* filename, outputStream* st) { 1894 int fd = ::open(filename, O_RDONLY); 1895 if (fd == -1) { 1896 return false; 1897 } 1898 1899 char buf[33]; 1900 int bytes; 1901 buf[32] = '\0'; 1902 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) { 1903 st->print_raw(buf, bytes); 1904 } 1905 1906 ::close(fd); 1907 1908 return true; 1909 } 1910 1911 void os::print_dll_info(outputStream *st) { 1912 st->print_cr("Dynamic libraries:"); 1913 1914 char fname[32]; 1915 pid_t pid = os::Linux::gettid(); 1916 1917 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1918 1919 if (!_print_ascii_file(fname, st)) { 1920 st->print("Can not get library information for pid = %d\n", pid); 1921 } 1922 } 1923 1924 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 1925 FILE *procmapsFile = NULL; 1926 1927 // Open the procfs maps file for the current process 1928 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 1929 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 1930 char line[PATH_MAX + 100]; 1931 1932 // Read line by line from 'file' 1933 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 1934 u8 base, top, offset, inode; 1935 char permissions[5]; 1936 char device[6]; 1937 char name[PATH_MAX + 1]; 1938 1939 // Parse fields from line 1940 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s", 1941 &base, &top, permissions, &offset, device, &inode, name); 1942 1943 // Filter by device id '00:00' so that we only get file system mapped files. 1944 if (strcmp(device, "00:00") != 0) { 1945 1946 // Call callback with the fields of interest 1947 if(callback(name, (address)base, (address)top, param)) { 1948 // Oops abort, callback aborted 1949 fclose(procmapsFile); 1950 return 1; 1951 } 1952 } 1953 } 1954 fclose(procmapsFile); 1955 } 1956 return 0; 1957 } 1958 1959 void os::print_os_info_brief(outputStream* st) { 1960 os::Linux::print_distro_info(st); 1961 1962 os::Posix::print_uname_info(st); 1963 1964 os::Linux::print_libversion_info(st); 1965 1966 } 1967 1968 void os::print_os_info(outputStream* st) { 1969 st->print("OS:"); 1970 1971 os::Linux::print_distro_info(st); 1972 1973 os::Posix::print_uname_info(st); 1974 1975 // Print warning if unsafe chroot environment detected 1976 if (unsafe_chroot_detected) { 1977 st->print("WARNING!! "); 1978 st->print_cr("%s", unstable_chroot_error); 1979 } 1980 1981 os::Linux::print_libversion_info(st); 1982 1983 os::Posix::print_rlimit_info(st); 1984 1985 os::Posix::print_load_average(st); 1986 1987 os::Linux::print_full_memory_info(st); 1988 } 1989 1990 // Try to identify popular distros. 1991 // Most Linux distributions have a /etc/XXX-release file, which contains 1992 // the OS version string. Newer Linux distributions have a /etc/lsb-release 1993 // file that also contains the OS version string. Some have more than one 1994 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 1995 // /etc/redhat-release.), so the order is important. 1996 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 1997 // their own specific XXX-release file as well as a redhat-release file. 1998 // Because of this the XXX-release file needs to be searched for before the 1999 // redhat-release file. 2000 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the 2001 // search for redhat-release / SuSE-release needs to be before lsb-release. 2002 // Since the lsb-release file is the new standard it needs to be searched 2003 // before the older style release files. 2004 // Searching system-release (Red Hat) and os-release (other Linuxes) are a 2005 // next to last resort. The os-release file is a new standard that contains 2006 // distribution information and the system-release file seems to be an old 2007 // standard that has been replaced by the lsb-release and os-release files. 2008 // Searching for the debian_version file is the last resort. It contains 2009 // an informative string like "6.0.6" or "wheezy/sid". Because of this 2010 // "Debian " is printed before the contents of the debian_version file. 2011 2012 const char* distro_files[] = { 2013 "/etc/oracle-release", 2014 "/etc/mandriva-release", 2015 "/etc/mandrake-release", 2016 "/etc/sun-release", 2017 "/etc/redhat-release", 2018 "/etc/SuSE-release", 2019 "/etc/lsb-release", 2020 "/etc/turbolinux-release", 2021 "/etc/gentoo-release", 2022 "/etc/ltib-release", 2023 "/etc/angstrom-version", 2024 "/etc/system-release", 2025 "/etc/os-release", 2026 NULL }; 2027 2028 void os::Linux::print_distro_info(outputStream* st) { 2029 for (int i = 0;; i++) { 2030 const char* file = distro_files[i]; 2031 if (file == NULL) { 2032 break; // done 2033 } 2034 // If file prints, we found it. 2035 if (_print_ascii_file(file, st)) { 2036 return; 2037 } 2038 } 2039 2040 if (file_exists("/etc/debian_version")) { 2041 st->print("Debian "); 2042 _print_ascii_file("/etc/debian_version", st); 2043 } else { 2044 st->print("Linux"); 2045 } 2046 st->cr(); 2047 } 2048 2049 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) { 2050 char buf[256]; 2051 while (fgets(buf, sizeof(buf), fp)) { 2052 // Edit out extra stuff in expected format 2053 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) { 2054 char* ptr = strstr(buf, "\""); // the name is in quotes 2055 if (ptr != NULL) { 2056 ptr++; // go beyond first quote 2057 char* nl = strchr(ptr, '\"'); 2058 if (nl != NULL) *nl = '\0'; 2059 strncpy(distro, ptr, length); 2060 } else { 2061 ptr = strstr(buf, "="); 2062 ptr++; // go beyond equals then 2063 char* nl = strchr(ptr, '\n'); 2064 if (nl != NULL) *nl = '\0'; 2065 strncpy(distro, ptr, length); 2066 } 2067 return; 2068 } else if (get_first_line) { 2069 char* nl = strchr(buf, '\n'); 2070 if (nl != NULL) *nl = '\0'; 2071 strncpy(distro, buf, length); 2072 return; 2073 } 2074 } 2075 // print last line and close 2076 char* nl = strchr(buf, '\n'); 2077 if (nl != NULL) *nl = '\0'; 2078 strncpy(distro, buf, length); 2079 } 2080 2081 static void parse_os_info(char* distro, size_t length, const char* file) { 2082 FILE* fp = fopen(file, "r"); 2083 if (fp != NULL) { 2084 // if suse format, print out first line 2085 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0); 2086 parse_os_info_helper(fp, distro, length, get_first_line); 2087 fclose(fp); 2088 } 2089 } 2090 2091 void os::get_summary_os_info(char* buf, size_t buflen) { 2092 for (int i = 0;; i++) { 2093 const char* file = distro_files[i]; 2094 if (file == NULL) { 2095 break; // ran out of distro_files 2096 } 2097 if (file_exists(file)) { 2098 parse_os_info(buf, buflen, file); 2099 return; 2100 } 2101 } 2102 // special case for debian 2103 if (file_exists("/etc/debian_version")) { 2104 strncpy(buf, "Debian ", buflen); 2105 parse_os_info(&buf[7], buflen-7, "/etc/debian_version"); 2106 } else { 2107 strncpy(buf, "Linux", buflen); 2108 } 2109 } 2110 2111 void os::Linux::print_libversion_info(outputStream* st) { 2112 // libc, pthread 2113 st->print("libc:"); 2114 st->print("%s ", os::Linux::glibc_version()); 2115 st->print("%s ", os::Linux::libpthread_version()); 2116 st->cr(); 2117 } 2118 2119 void os::Linux::print_full_memory_info(outputStream* st) { 2120 st->print("\n/proc/meminfo:\n"); 2121 _print_ascii_file("/proc/meminfo", st); 2122 st->cr(); 2123 } 2124 2125 void os::print_memory_info(outputStream* st) { 2126 2127 st->print("Memory:"); 2128 st->print(" %dk page", os::vm_page_size()>>10); 2129 2130 // values in struct sysinfo are "unsigned long" 2131 struct sysinfo si; 2132 sysinfo(&si); 2133 2134 st->print(", physical " UINT64_FORMAT "k", 2135 os::physical_memory() >> 10); 2136 st->print("(" UINT64_FORMAT "k free)", 2137 os::available_memory() >> 10); 2138 st->print(", swap " UINT64_FORMAT "k", 2139 ((jlong)si.totalswap * si.mem_unit) >> 10); 2140 st->print("(" UINT64_FORMAT "k free)", 2141 ((jlong)si.freeswap * si.mem_unit) >> 10); 2142 st->cr(); 2143 } 2144 2145 // Print the first "model name" line and the first "flags" line 2146 // that we find and nothing more. We assume "model name" comes 2147 // before "flags" so if we find a second "model name", then the 2148 // "flags" field is considered missing. 2149 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) { 2150 #if defined(IA32) || defined(AMD64) 2151 // Other platforms have less repetitive cpuinfo files 2152 FILE *fp = fopen("/proc/cpuinfo", "r"); 2153 if (fp) { 2154 while (!feof(fp)) { 2155 if (fgets(buf, buflen, fp)) { 2156 // Assume model name comes before flags 2157 bool model_name_printed = false; 2158 if (strstr(buf, "model name") != NULL) { 2159 if (!model_name_printed) { 2160 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n"); 2161 st->print_raw(buf); 2162 model_name_printed = true; 2163 } else { 2164 // model name printed but not flags? Odd, just return 2165 fclose(fp); 2166 return true; 2167 } 2168 } 2169 // print the flags line too 2170 if (strstr(buf, "flags") != NULL) { 2171 st->print_raw(buf); 2172 fclose(fp); 2173 return true; 2174 } 2175 } 2176 } 2177 fclose(fp); 2178 } 2179 #endif // x86 platforms 2180 return false; 2181 } 2182 2183 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) { 2184 // Only print the model name if the platform provides this as a summary 2185 if (!print_model_name_and_flags(st, buf, buflen)) { 2186 st->print("\n/proc/cpuinfo:\n"); 2187 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2188 st->print_cr(" <Not Available>"); 2189 } 2190 } 2191 } 2192 2193 #if defined(AMD64) || defined(IA32) || defined(X32) 2194 const char* search_string = "model name"; 2195 #elif defined(SPARC) 2196 const char* search_string = "cpu"; 2197 #elif defined(PPC64) 2198 const char* search_string = "cpu"; 2199 #else 2200 const char* search_string = "Processor"; 2201 #endif 2202 2203 // Parses the cpuinfo file for string representing the model name. 2204 void os::get_summary_cpu_info(char* cpuinfo, size_t length) { 2205 FILE* fp = fopen("/proc/cpuinfo", "r"); 2206 if (fp != NULL) { 2207 while (!feof(fp)) { 2208 char buf[256]; 2209 if (fgets(buf, sizeof(buf), fp)) { 2210 char* start = strstr(buf, search_string); 2211 if (start != NULL) { 2212 char *ptr = start + strlen(search_string); 2213 char *end = buf + strlen(buf); 2214 while (ptr != end) { 2215 // skip whitespace and colon for the rest of the name. 2216 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') { 2217 break; 2218 } 2219 ptr++; 2220 } 2221 if (ptr != end) { 2222 // reasonable string, get rid of newline and keep the rest 2223 char* nl = strchr(buf, '\n'); 2224 if (nl != NULL) *nl = '\0'; 2225 strncpy(cpuinfo, ptr, length); 2226 fclose(fp); 2227 return; 2228 } 2229 } 2230 } 2231 } 2232 fclose(fp); 2233 } 2234 // cpuinfo not found or parsing failed, just print generic string. The entire 2235 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86) 2236 #if defined(AMD64) 2237 strncpy(cpuinfo, "x86_64", length); 2238 #elif defined(IA32) 2239 strncpy(cpuinfo, "x86_32", length); 2240 #elif defined(IA64) 2241 strncpy(cpuinfo, "IA64", length); 2242 #elif defined(SPARC) 2243 strncpy(cpuinfo, "sparcv9", length); 2244 #elif defined(AARCH64) 2245 strncpy(cpuinfo, "AArch64", length); 2246 #elif defined(ARM) 2247 strncpy(cpuinfo, "ARM", length); 2248 #elif defined(PPC) 2249 strncpy(cpuinfo, "PPC64", length); 2250 #elif defined(ZERO_LIBARCH) 2251 strncpy(cpuinfo, ZERO_LIBARCH, length); 2252 #else 2253 strncpy(cpuinfo, "unknown", length); 2254 #endif 2255 } 2256 2257 static void print_signal_handler(outputStream* st, int sig, 2258 char* buf, size_t buflen); 2259 2260 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2261 st->print_cr("Signal Handlers:"); 2262 print_signal_handler(st, SIGSEGV, buf, buflen); 2263 print_signal_handler(st, SIGBUS , buf, buflen); 2264 print_signal_handler(st, SIGFPE , buf, buflen); 2265 print_signal_handler(st, SIGPIPE, buf, buflen); 2266 print_signal_handler(st, SIGXFSZ, buf, buflen); 2267 print_signal_handler(st, SIGILL , buf, buflen); 2268 print_signal_handler(st, SR_signum, buf, buflen); 2269 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2270 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2271 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2272 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2273 #if defined(PPC64) 2274 print_signal_handler(st, SIGTRAP, buf, buflen); 2275 #endif 2276 } 2277 2278 static char saved_jvm_path[MAXPATHLEN] = {0}; 2279 2280 // Find the full path to the current module, libjvm.so 2281 void os::jvm_path(char *buf, jint buflen) { 2282 // Error checking. 2283 if (buflen < MAXPATHLEN) { 2284 assert(false, "must use a large-enough buffer"); 2285 buf[0] = '\0'; 2286 return; 2287 } 2288 // Lazy resolve the path to current module. 2289 if (saved_jvm_path[0] != 0) { 2290 strcpy(buf, saved_jvm_path); 2291 return; 2292 } 2293 2294 char dli_fname[MAXPATHLEN]; 2295 bool ret = dll_address_to_library_name( 2296 CAST_FROM_FN_PTR(address, os::jvm_path), 2297 dli_fname, sizeof(dli_fname), NULL); 2298 assert(ret, "cannot locate libjvm"); 2299 char *rp = NULL; 2300 if (ret && dli_fname[0] != '\0') { 2301 rp = realpath(dli_fname, buf); 2302 } 2303 if (rp == NULL) { 2304 return; 2305 } 2306 2307 if (Arguments::sun_java_launcher_is_altjvm()) { 2308 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2309 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2310 // If "/jre/lib/" appears at the right place in the string, then 2311 // assume we are installed in a JDK and we're done. Otherwise, check 2312 // for a JAVA_HOME environment variable and fix up the path so it 2313 // looks like libjvm.so is installed there (append a fake suffix 2314 // hotspot/libjvm.so). 2315 const char *p = buf + strlen(buf) - 1; 2316 for (int count = 0; p > buf && count < 5; ++count) { 2317 for (--p; p > buf && *p != '/'; --p) 2318 /* empty */ ; 2319 } 2320 2321 if (strncmp(p, "/jre/lib/", 9) != 0) { 2322 // Look for JAVA_HOME in the environment. 2323 char* java_home_var = ::getenv("JAVA_HOME"); 2324 if (java_home_var != NULL && java_home_var[0] != 0) { 2325 char* jrelib_p; 2326 int len; 2327 2328 // Check the current module name "libjvm.so". 2329 p = strrchr(buf, '/'); 2330 if (p == NULL) { 2331 return; 2332 } 2333 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2334 2335 rp = realpath(java_home_var, buf); 2336 if (rp == NULL) { 2337 return; 2338 } 2339 2340 // determine if this is a legacy image or modules image 2341 // modules image doesn't have "jre" subdirectory 2342 len = strlen(buf); 2343 assert(len < buflen, "Ran out of buffer room"); 2344 jrelib_p = buf + len; 2345 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2346 if (0 != access(buf, F_OK)) { 2347 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2348 } 2349 2350 if (0 == access(buf, F_OK)) { 2351 // Use current module name "libjvm.so" 2352 len = strlen(buf); 2353 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2354 } else { 2355 // Go back to path of .so 2356 rp = realpath(dli_fname, buf); 2357 if (rp == NULL) { 2358 return; 2359 } 2360 } 2361 } 2362 } 2363 } 2364 2365 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2366 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2367 } 2368 2369 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2370 // no prefix required, not even "_" 2371 } 2372 2373 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2374 // no suffix required 2375 } 2376 2377 //////////////////////////////////////////////////////////////////////////////// 2378 // sun.misc.Signal support 2379 2380 static volatile jint sigint_count = 0; 2381 2382 static void UserHandler(int sig, void *siginfo, void *context) { 2383 // 4511530 - sem_post is serialized and handled by the manager thread. When 2384 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2385 // don't want to flood the manager thread with sem_post requests. 2386 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2387 return; 2388 } 2389 2390 // Ctrl-C is pressed during error reporting, likely because the error 2391 // handler fails to abort. Let VM die immediately. 2392 if (sig == SIGINT && is_error_reported()) { 2393 os::die(); 2394 } 2395 2396 os::signal_notify(sig); 2397 } 2398 2399 void* os::user_handler() { 2400 return CAST_FROM_FN_PTR(void*, UserHandler); 2401 } 2402 2403 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) { 2404 struct timespec ts; 2405 // Semaphore's are always associated with CLOCK_REALTIME 2406 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2407 // see unpackTime for discussion on overflow checking 2408 if (sec >= MAX_SECS) { 2409 ts.tv_sec += MAX_SECS; 2410 ts.tv_nsec = 0; 2411 } else { 2412 ts.tv_sec += sec; 2413 ts.tv_nsec += nsec; 2414 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2415 ts.tv_nsec -= NANOSECS_PER_SEC; 2416 ++ts.tv_sec; // note: this must be <= max_secs 2417 } 2418 } 2419 2420 return ts; 2421 } 2422 2423 extern "C" { 2424 typedef void (*sa_handler_t)(int); 2425 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2426 } 2427 2428 void* os::signal(int signal_number, void* handler) { 2429 struct sigaction sigAct, oldSigAct; 2430 2431 sigfillset(&(sigAct.sa_mask)); 2432 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2433 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2434 2435 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2436 // -1 means registration failed 2437 return (void *)-1; 2438 } 2439 2440 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2441 } 2442 2443 void os::signal_raise(int signal_number) { 2444 ::raise(signal_number); 2445 } 2446 2447 // The following code is moved from os.cpp for making this 2448 // code platform specific, which it is by its very nature. 2449 2450 // Will be modified when max signal is changed to be dynamic 2451 int os::sigexitnum_pd() { 2452 return NSIG; 2453 } 2454 2455 // a counter for each possible signal value 2456 static volatile jint pending_signals[NSIG+1] = { 0 }; 2457 2458 // Linux(POSIX) specific hand shaking semaphore. 2459 static sem_t sig_sem; 2460 static PosixSemaphore sr_semaphore; 2461 2462 void os::signal_init_pd() { 2463 // Initialize signal structures 2464 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2465 2466 // Initialize signal semaphore 2467 ::sem_init(&sig_sem, 0, 0); 2468 } 2469 2470 void os::signal_notify(int sig) { 2471 Atomic::inc(&pending_signals[sig]); 2472 ::sem_post(&sig_sem); 2473 } 2474 2475 static int check_pending_signals(bool wait) { 2476 Atomic::store(0, &sigint_count); 2477 for (;;) { 2478 for (int i = 0; i < NSIG + 1; i++) { 2479 jint n = pending_signals[i]; 2480 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2481 return i; 2482 } 2483 } 2484 if (!wait) { 2485 return -1; 2486 } 2487 JavaThread *thread = JavaThread::current(); 2488 ThreadBlockInVM tbivm(thread); 2489 2490 bool threadIsSuspended; 2491 do { 2492 thread->set_suspend_equivalent(); 2493 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2494 ::sem_wait(&sig_sem); 2495 2496 // were we externally suspended while we were waiting? 2497 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2498 if (threadIsSuspended) { 2499 // The semaphore has been incremented, but while we were waiting 2500 // another thread suspended us. We don't want to continue running 2501 // while suspended because that would surprise the thread that 2502 // suspended us. 2503 ::sem_post(&sig_sem); 2504 2505 thread->java_suspend_self(); 2506 } 2507 } while (threadIsSuspended); 2508 } 2509 } 2510 2511 int os::signal_lookup() { 2512 return check_pending_signals(false); 2513 } 2514 2515 int os::signal_wait() { 2516 return check_pending_signals(true); 2517 } 2518 2519 //////////////////////////////////////////////////////////////////////////////// 2520 // Virtual Memory 2521 2522 int os::vm_page_size() { 2523 // Seems redundant as all get out 2524 assert(os::Linux::page_size() != -1, "must call os::init"); 2525 return os::Linux::page_size(); 2526 } 2527 2528 // Solaris allocates memory by pages. 2529 int os::vm_allocation_granularity() { 2530 assert(os::Linux::page_size() != -1, "must call os::init"); 2531 return os::Linux::page_size(); 2532 } 2533 2534 // Rationale behind this function: 2535 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2536 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2537 // samples for JITted code. Here we create private executable mapping over the code cache 2538 // and then we can use standard (well, almost, as mapping can change) way to provide 2539 // info for the reporting script by storing timestamp and location of symbol 2540 void linux_wrap_code(char* base, size_t size) { 2541 static volatile jint cnt = 0; 2542 2543 if (!UseOprofile) { 2544 return; 2545 } 2546 2547 char buf[PATH_MAX+1]; 2548 int num = Atomic::add(1, &cnt); 2549 2550 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2551 os::get_temp_directory(), os::current_process_id(), num); 2552 unlink(buf); 2553 2554 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2555 2556 if (fd != -1) { 2557 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2558 if (rv != (off_t)-1) { 2559 if (::write(fd, "", 1) == 1) { 2560 mmap(base, size, 2561 PROT_READ|PROT_WRITE|PROT_EXEC, 2562 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2563 } 2564 } 2565 ::close(fd); 2566 unlink(buf); 2567 } 2568 } 2569 2570 static bool recoverable_mmap_error(int err) { 2571 // See if the error is one we can let the caller handle. This 2572 // list of errno values comes from JBS-6843484. I can't find a 2573 // Linux man page that documents this specific set of errno 2574 // values so while this list currently matches Solaris, it may 2575 // change as we gain experience with this failure mode. 2576 switch (err) { 2577 case EBADF: 2578 case EINVAL: 2579 case ENOTSUP: 2580 // let the caller deal with these errors 2581 return true; 2582 2583 default: 2584 // Any remaining errors on this OS can cause our reserved mapping 2585 // to be lost. That can cause confusion where different data 2586 // structures think they have the same memory mapped. The worst 2587 // scenario is if both the VM and a library think they have the 2588 // same memory mapped. 2589 return false; 2590 } 2591 } 2592 2593 static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2594 int err) { 2595 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2596 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec, 2597 os::strerror(err), err); 2598 } 2599 2600 static void warn_fail_commit_memory(char* addr, size_t size, 2601 size_t alignment_hint, bool exec, 2602 int err) { 2603 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2604 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, 2605 alignment_hint, exec, os::strerror(err), err); 2606 } 2607 2608 // NOTE: Linux kernel does not really reserve the pages for us. 2609 // All it does is to check if there are enough free pages 2610 // left at the time of mmap(). This could be a potential 2611 // problem. 2612 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2613 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2614 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2615 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2616 if (res != (uintptr_t) MAP_FAILED) { 2617 if (UseNUMAInterleaving) { 2618 numa_make_global(addr, size); 2619 } 2620 return 0; 2621 } 2622 2623 int err = errno; // save errno from mmap() call above 2624 2625 if (!recoverable_mmap_error(err)) { 2626 warn_fail_commit_memory(addr, size, exec, err); 2627 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2628 } 2629 2630 return err; 2631 } 2632 2633 bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2634 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2635 } 2636 2637 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2638 const char* mesg) { 2639 assert(mesg != NULL, "mesg must be specified"); 2640 int err = os::Linux::commit_memory_impl(addr, size, exec); 2641 if (err != 0) { 2642 // the caller wants all commit errors to exit with the specified mesg: 2643 warn_fail_commit_memory(addr, size, exec, err); 2644 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2645 } 2646 } 2647 2648 // Define MAP_HUGETLB here so we can build HotSpot on old systems. 2649 #ifndef MAP_HUGETLB 2650 #define MAP_HUGETLB 0x40000 2651 #endif 2652 2653 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2654 #ifndef MADV_HUGEPAGE 2655 #define MADV_HUGEPAGE 14 2656 #endif 2657 2658 int os::Linux::commit_memory_impl(char* addr, size_t size, 2659 size_t alignment_hint, bool exec) { 2660 int err = os::Linux::commit_memory_impl(addr, size, exec); 2661 if (err == 0) { 2662 realign_memory(addr, size, alignment_hint); 2663 } 2664 return err; 2665 } 2666 2667 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2668 bool exec) { 2669 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2670 } 2671 2672 void os::pd_commit_memory_or_exit(char* addr, size_t size, 2673 size_t alignment_hint, bool exec, 2674 const char* mesg) { 2675 assert(mesg != NULL, "mesg must be specified"); 2676 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2677 if (err != 0) { 2678 // the caller wants all commit errors to exit with the specified mesg: 2679 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2680 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2681 } 2682 } 2683 2684 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2685 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2686 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2687 // be supported or the memory may already be backed by huge pages. 2688 ::madvise(addr, bytes, MADV_HUGEPAGE); 2689 } 2690 } 2691 2692 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2693 // This method works by doing an mmap over an existing mmaping and effectively discarding 2694 // the existing pages. However it won't work for SHM-based large pages that cannot be 2695 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2696 // small pages on top of the SHM segment. This method always works for small pages, so we 2697 // allow that in any case. 2698 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2699 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2700 } 2701 } 2702 2703 void os::numa_make_global(char *addr, size_t bytes) { 2704 Linux::numa_interleave_memory(addr, bytes); 2705 } 2706 2707 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2708 // bind policy to MPOL_PREFERRED for the current thread. 2709 #define USE_MPOL_PREFERRED 0 2710 2711 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2712 // To make NUMA and large pages more robust when both enabled, we need to ease 2713 // the requirements on where the memory should be allocated. MPOL_BIND is the 2714 // default policy and it will force memory to be allocated on the specified 2715 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2716 // the specified node, but will not force it. Using this policy will prevent 2717 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2718 // free large pages. 2719 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2720 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2721 } 2722 2723 bool os::numa_topology_changed() { return false; } 2724 2725 size_t os::numa_get_groups_num() { 2726 int max_node = Linux::numa_max_node(); 2727 return max_node > 0 ? max_node + 1 : 1; 2728 } 2729 2730 int os::numa_get_group_id() { 2731 int cpu_id = Linux::sched_getcpu(); 2732 if (cpu_id != -1) { 2733 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2734 if (lgrp_id != -1) { 2735 return lgrp_id; 2736 } 2737 } 2738 return 0; 2739 } 2740 2741 size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2742 for (size_t i = 0; i < size; i++) { 2743 ids[i] = i; 2744 } 2745 return size; 2746 } 2747 2748 bool os::get_page_info(char *start, page_info* info) { 2749 return false; 2750 } 2751 2752 char *os::scan_pages(char *start, char* end, page_info* page_expected, 2753 page_info* page_found) { 2754 return end; 2755 } 2756 2757 2758 int os::Linux::sched_getcpu_syscall(void) { 2759 unsigned int cpu = 0; 2760 int retval = -1; 2761 2762 #if defined(IA32) 2763 #ifndef SYS_getcpu 2764 #define SYS_getcpu 318 2765 #endif 2766 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2767 #elif defined(AMD64) 2768 // Unfortunately we have to bring all these macros here from vsyscall.h 2769 // to be able to compile on old linuxes. 2770 #define __NR_vgetcpu 2 2771 #define VSYSCALL_START (-10UL << 20) 2772 #define VSYSCALL_SIZE 1024 2773 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2774 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2775 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2776 retval = vgetcpu(&cpu, NULL, NULL); 2777 #endif 2778 2779 return (retval == -1) ? retval : cpu; 2780 } 2781 2782 // Something to do with the numa-aware allocator needs these symbols 2783 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2784 extern "C" JNIEXPORT void numa_error(char *where) { } 2785 2786 2787 // If we are running with libnuma version > 2, then we should 2788 // be trying to use symbols with versions 1.1 2789 // If we are running with earlier version, which did not have symbol versions, 2790 // we should use the base version. 2791 void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2792 void *f = dlvsym(handle, name, "libnuma_1.1"); 2793 if (f == NULL) { 2794 f = dlsym(handle, name); 2795 } 2796 return f; 2797 } 2798 2799 bool os::Linux::libnuma_init() { 2800 // sched_getcpu() should be in libc. 2801 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2802 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2803 2804 // If it's not, try a direct syscall. 2805 if (sched_getcpu() == -1) { 2806 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2807 (void*)&sched_getcpu_syscall)); 2808 } 2809 2810 if (sched_getcpu() != -1) { // Does it work? 2811 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2812 if (handle != NULL) { 2813 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2814 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2815 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2816 libnuma_dlsym(handle, "numa_max_node"))); 2817 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2818 libnuma_dlsym(handle, "numa_available"))); 2819 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2820 libnuma_dlsym(handle, "numa_tonode_memory"))); 2821 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2822 libnuma_dlsym(handle, "numa_interleave_memory"))); 2823 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2824 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2825 2826 2827 if (numa_available() != -1) { 2828 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2829 // Create a cpu -> node mapping 2830 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2831 rebuild_cpu_to_node_map(); 2832 return true; 2833 } 2834 } 2835 } 2836 return false; 2837 } 2838 2839 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2840 // The table is later used in get_node_by_cpu(). 2841 void os::Linux::rebuild_cpu_to_node_map() { 2842 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2843 // in libnuma (possible values are starting from 16, 2844 // and continuing up with every other power of 2, but less 2845 // than the maximum number of CPUs supported by kernel), and 2846 // is a subject to change (in libnuma version 2 the requirements 2847 // are more reasonable) we'll just hardcode the number they use 2848 // in the library. 2849 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2850 2851 size_t cpu_num = os::active_processor_count(); 2852 size_t cpu_map_size = NCPUS / BitsPerCLong; 2853 size_t cpu_map_valid_size = 2854 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2855 2856 cpu_to_node()->clear(); 2857 cpu_to_node()->at_grow(cpu_num - 1); 2858 size_t node_num = numa_get_groups_num(); 2859 2860 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2861 for (size_t i = 0; i < node_num; i++) { 2862 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2863 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2864 if (cpu_map[j] != 0) { 2865 for (size_t k = 0; k < BitsPerCLong; k++) { 2866 if (cpu_map[j] & (1UL << k)) { 2867 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2868 } 2869 } 2870 } 2871 } 2872 } 2873 } 2874 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2875 } 2876 2877 int os::Linux::get_node_by_cpu(int cpu_id) { 2878 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2879 return cpu_to_node()->at(cpu_id); 2880 } 2881 return -1; 2882 } 2883 2884 GrowableArray<int>* os::Linux::_cpu_to_node; 2885 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2886 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2887 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2888 os::Linux::numa_available_func_t os::Linux::_numa_available; 2889 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2890 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2891 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2892 unsigned long* os::Linux::_numa_all_nodes; 2893 2894 bool os::pd_uncommit_memory(char* addr, size_t size) { 2895 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2896 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2897 return res != (uintptr_t) MAP_FAILED; 2898 } 2899 2900 static address get_stack_commited_bottom(address bottom, size_t size) { 2901 address nbot = bottom; 2902 address ntop = bottom + size; 2903 2904 size_t page_sz = os::vm_page_size(); 2905 unsigned pages = size / page_sz; 2906 2907 unsigned char vec[1]; 2908 unsigned imin = 1, imax = pages + 1, imid; 2909 int mincore_return_value = 0; 2910 2911 assert(imin <= imax, "Unexpected page size"); 2912 2913 while (imin < imax) { 2914 imid = (imax + imin) / 2; 2915 nbot = ntop - (imid * page_sz); 2916 2917 // Use a trick with mincore to check whether the page is mapped or not. 2918 // mincore sets vec to 1 if page resides in memory and to 0 if page 2919 // is swapped output but if page we are asking for is unmapped 2920 // it returns -1,ENOMEM 2921 mincore_return_value = mincore(nbot, page_sz, vec); 2922 2923 if (mincore_return_value == -1) { 2924 // Page is not mapped go up 2925 // to find first mapped page 2926 if (errno != EAGAIN) { 2927 assert(errno == ENOMEM, "Unexpected mincore errno"); 2928 imax = imid; 2929 } 2930 } else { 2931 // Page is mapped go down 2932 // to find first not mapped page 2933 imin = imid + 1; 2934 } 2935 } 2936 2937 nbot = nbot + page_sz; 2938 2939 // Adjust stack bottom one page up if last checked page is not mapped 2940 if (mincore_return_value == -1) { 2941 nbot = nbot + page_sz; 2942 } 2943 2944 return nbot; 2945 } 2946 2947 2948 // Linux uses a growable mapping for the stack, and if the mapping for 2949 // the stack guard pages is not removed when we detach a thread the 2950 // stack cannot grow beyond the pages where the stack guard was 2951 // mapped. If at some point later in the process the stack expands to 2952 // that point, the Linux kernel cannot expand the stack any further 2953 // because the guard pages are in the way, and a segfault occurs. 2954 // 2955 // However, it's essential not to split the stack region by unmapping 2956 // a region (leaving a hole) that's already part of the stack mapping, 2957 // so if the stack mapping has already grown beyond the guard pages at 2958 // the time we create them, we have to truncate the stack mapping. 2959 // So, we need to know the extent of the stack mapping when 2960 // create_stack_guard_pages() is called. 2961 2962 // We only need this for stacks that are growable: at the time of 2963 // writing thread stacks don't use growable mappings (i.e. those 2964 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 2965 // only applies to the main thread. 2966 2967 // If the (growable) stack mapping already extends beyond the point 2968 // where we're going to put our guard pages, truncate the mapping at 2969 // that point by munmap()ping it. This ensures that when we later 2970 // munmap() the guard pages we don't leave a hole in the stack 2971 // mapping. This only affects the main/initial thread 2972 2973 bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 2974 if (os::Linux::is_initial_thread()) { 2975 // As we manually grow stack up to bottom inside create_attached_thread(), 2976 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 2977 // we don't need to do anything special. 2978 // Check it first, before calling heavy function. 2979 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 2980 unsigned char vec[1]; 2981 2982 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 2983 // Fallback to slow path on all errors, including EAGAIN 2984 stack_extent = (uintptr_t) get_stack_commited_bottom( 2985 os::Linux::initial_thread_stack_bottom(), 2986 (size_t)addr - stack_extent); 2987 } 2988 2989 if (stack_extent < (uintptr_t)addr) { 2990 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 2991 } 2992 } 2993 2994 return os::commit_memory(addr, size, !ExecMem); 2995 } 2996 2997 // If this is a growable mapping, remove the guard pages entirely by 2998 // munmap()ping them. If not, just call uncommit_memory(). This only 2999 // affects the main/initial thread, but guard against future OS changes 3000 // It's safe to always unmap guard pages for initial thread because we 3001 // always place it right after end of the mapped region 3002 3003 bool os::remove_stack_guard_pages(char* addr, size_t size) { 3004 uintptr_t stack_extent, stack_base; 3005 3006 if (os::Linux::is_initial_thread()) { 3007 return ::munmap(addr, size) == 0; 3008 } 3009 3010 return os::uncommit_memory(addr, size); 3011 } 3012 3013 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3014 // at 'requested_addr'. If there are existing memory mappings at the same 3015 // location, however, they will be overwritten. If 'fixed' is false, 3016 // 'requested_addr' is only treated as a hint, the return value may or 3017 // may not start from the requested address. Unlike Linux mmap(), this 3018 // function returns NULL to indicate failure. 3019 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3020 char * addr; 3021 int flags; 3022 3023 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3024 if (fixed) { 3025 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3026 flags |= MAP_FIXED; 3027 } 3028 3029 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3030 // touch an uncommitted page. Otherwise, the read/write might 3031 // succeed if we have enough swap space to back the physical page. 3032 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3033 flags, -1, 0); 3034 3035 return addr == MAP_FAILED ? NULL : addr; 3036 } 3037 3038 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address 3039 // (req_addr != NULL) or with a given alignment. 3040 // - bytes shall be a multiple of alignment. 3041 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3042 // - alignment sets the alignment at which memory shall be allocated. 3043 // It must be a multiple of allocation granularity. 3044 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3045 // req_addr or NULL. 3046 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) { 3047 3048 size_t extra_size = bytes; 3049 if (req_addr == NULL && alignment > 0) { 3050 extra_size += alignment; 3051 } 3052 3053 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE, 3054 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 3055 -1, 0); 3056 if (start == MAP_FAILED) { 3057 start = NULL; 3058 } else { 3059 if (req_addr != NULL) { 3060 if (start != req_addr) { 3061 ::munmap(start, extra_size); 3062 start = NULL; 3063 } 3064 } else { 3065 char* const start_aligned = (char*) align_ptr_up(start, alignment); 3066 char* const end_aligned = start_aligned + bytes; 3067 char* const end = start + extra_size; 3068 if (start_aligned > start) { 3069 ::munmap(start, start_aligned - start); 3070 } 3071 if (end_aligned < end) { 3072 ::munmap(end_aligned, end - end_aligned); 3073 } 3074 start = start_aligned; 3075 } 3076 } 3077 return start; 3078 } 3079 3080 static int anon_munmap(char * addr, size_t size) { 3081 return ::munmap(addr, size) == 0; 3082 } 3083 3084 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3085 size_t alignment_hint) { 3086 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3087 } 3088 3089 bool os::pd_release_memory(char* addr, size_t size) { 3090 return anon_munmap(addr, size); 3091 } 3092 3093 static bool linux_mprotect(char* addr, size_t size, int prot) { 3094 // Linux wants the mprotect address argument to be page aligned. 3095 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3096 3097 // According to SUSv3, mprotect() should only be used with mappings 3098 // established by mmap(), and mmap() always maps whole pages. Unaligned 3099 // 'addr' likely indicates problem in the VM (e.g. trying to change 3100 // protection of malloc'ed or statically allocated memory). Check the 3101 // caller if you hit this assert. 3102 assert(addr == bottom, "sanity check"); 3103 3104 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3105 return ::mprotect(bottom, size, prot) == 0; 3106 } 3107 3108 // Set protections specified 3109 bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3110 bool is_committed) { 3111 unsigned int p = 0; 3112 switch (prot) { 3113 case MEM_PROT_NONE: p = PROT_NONE; break; 3114 case MEM_PROT_READ: p = PROT_READ; break; 3115 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3116 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3117 default: 3118 ShouldNotReachHere(); 3119 } 3120 // is_committed is unused. 3121 return linux_mprotect(addr, bytes, p); 3122 } 3123 3124 bool os::guard_memory(char* addr, size_t size) { 3125 return linux_mprotect(addr, size, PROT_NONE); 3126 } 3127 3128 bool os::unguard_memory(char* addr, size_t size) { 3129 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3130 } 3131 3132 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3133 size_t page_size) { 3134 bool result = false; 3135 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3136 MAP_ANONYMOUS|MAP_PRIVATE, 3137 -1, 0); 3138 if (p != MAP_FAILED) { 3139 void *aligned_p = align_ptr_up(p, page_size); 3140 3141 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3142 3143 munmap(p, page_size * 2); 3144 } 3145 3146 if (warn && !result) { 3147 warning("TransparentHugePages is not supported by the operating system."); 3148 } 3149 3150 return result; 3151 } 3152 3153 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3154 bool result = false; 3155 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3156 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3157 -1, 0); 3158 3159 if (p != MAP_FAILED) { 3160 // We don't know if this really is a huge page or not. 3161 FILE *fp = fopen("/proc/self/maps", "r"); 3162 if (fp) { 3163 while (!feof(fp)) { 3164 char chars[257]; 3165 long x = 0; 3166 if (fgets(chars, sizeof(chars), fp)) { 3167 if (sscanf(chars, "%lx-%*x", &x) == 1 3168 && x == (long)p) { 3169 if (strstr (chars, "hugepage")) { 3170 result = true; 3171 break; 3172 } 3173 } 3174 } 3175 } 3176 fclose(fp); 3177 } 3178 munmap(p, page_size); 3179 } 3180 3181 if (warn && !result) { 3182 warning("HugeTLBFS is not supported by the operating system."); 3183 } 3184 3185 return result; 3186 } 3187 3188 // Set the coredump_filter bits to include largepages in core dump (bit 6) 3189 // 3190 // From the coredump_filter documentation: 3191 // 3192 // - (bit 0) anonymous private memory 3193 // - (bit 1) anonymous shared memory 3194 // - (bit 2) file-backed private memory 3195 // - (bit 3) file-backed shared memory 3196 // - (bit 4) ELF header pages in file-backed private memory areas (it is 3197 // effective only if the bit 2 is cleared) 3198 // - (bit 5) hugetlb private memory 3199 // - (bit 6) hugetlb shared memory 3200 // 3201 static void set_coredump_filter(void) { 3202 FILE *f; 3203 long cdm; 3204 3205 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3206 return; 3207 } 3208 3209 if (fscanf(f, "%lx", &cdm) != 1) { 3210 fclose(f); 3211 return; 3212 } 3213 3214 rewind(f); 3215 3216 if ((cdm & LARGEPAGES_BIT) == 0) { 3217 cdm |= LARGEPAGES_BIT; 3218 fprintf(f, "%#lx", cdm); 3219 } 3220 3221 fclose(f); 3222 } 3223 3224 // Large page support 3225 3226 static size_t _large_page_size = 0; 3227 3228 size_t os::Linux::find_large_page_size() { 3229 size_t large_page_size = 0; 3230 3231 // large_page_size on Linux is used to round up heap size. x86 uses either 3232 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3233 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3234 // page as large as 256M. 3235 // 3236 // Here we try to figure out page size by parsing /proc/meminfo and looking 3237 // for a line with the following format: 3238 // Hugepagesize: 2048 kB 3239 // 3240 // If we can't determine the value (e.g. /proc is not mounted, or the text 3241 // format has been changed), we'll use the largest page size supported by 3242 // the processor. 3243 3244 #ifndef ZERO 3245 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3246 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M); 3247 #endif // ZERO 3248 3249 FILE *fp = fopen("/proc/meminfo", "r"); 3250 if (fp) { 3251 while (!feof(fp)) { 3252 int x = 0; 3253 char buf[16]; 3254 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3255 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3256 large_page_size = x * K; 3257 break; 3258 } 3259 } else { 3260 // skip to next line 3261 for (;;) { 3262 int ch = fgetc(fp); 3263 if (ch == EOF || ch == (int)'\n') break; 3264 } 3265 } 3266 } 3267 fclose(fp); 3268 } 3269 3270 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3271 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3272 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3273 proper_unit_for_byte_size(large_page_size)); 3274 } 3275 3276 return large_page_size; 3277 } 3278 3279 size_t os::Linux::setup_large_page_size() { 3280 _large_page_size = Linux::find_large_page_size(); 3281 const size_t default_page_size = (size_t)Linux::page_size(); 3282 if (_large_page_size > default_page_size) { 3283 _page_sizes[0] = _large_page_size; 3284 _page_sizes[1] = default_page_size; 3285 _page_sizes[2] = 0; 3286 } 3287 3288 return _large_page_size; 3289 } 3290 3291 bool os::Linux::setup_large_page_type(size_t page_size) { 3292 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3293 FLAG_IS_DEFAULT(UseSHM) && 3294 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3295 3296 // The type of large pages has not been specified by the user. 3297 3298 // Try UseHugeTLBFS and then UseSHM. 3299 UseHugeTLBFS = UseSHM = true; 3300 3301 // Don't try UseTransparentHugePages since there are known 3302 // performance issues with it turned on. This might change in the future. 3303 UseTransparentHugePages = false; 3304 } 3305 3306 if (UseTransparentHugePages) { 3307 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3308 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3309 UseHugeTLBFS = false; 3310 UseSHM = false; 3311 return true; 3312 } 3313 UseTransparentHugePages = false; 3314 } 3315 3316 if (UseHugeTLBFS) { 3317 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3318 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3319 UseSHM = false; 3320 return true; 3321 } 3322 UseHugeTLBFS = false; 3323 } 3324 3325 return UseSHM; 3326 } 3327 3328 void os::large_page_init() { 3329 if (!UseLargePages && 3330 !UseTransparentHugePages && 3331 !UseHugeTLBFS && 3332 !UseSHM) { 3333 // Not using large pages. 3334 return; 3335 } 3336 3337 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3338 // The user explicitly turned off large pages. 3339 // Ignore the rest of the large pages flags. 3340 UseTransparentHugePages = false; 3341 UseHugeTLBFS = false; 3342 UseSHM = false; 3343 return; 3344 } 3345 3346 size_t large_page_size = Linux::setup_large_page_size(); 3347 UseLargePages = Linux::setup_large_page_type(large_page_size); 3348 3349 set_coredump_filter(); 3350 } 3351 3352 #ifndef SHM_HUGETLB 3353 #define SHM_HUGETLB 04000 3354 #endif 3355 3356 #define shm_warning_format(format, ...) \ 3357 do { \ 3358 if (UseLargePages && \ 3359 (!FLAG_IS_DEFAULT(UseLargePages) || \ 3360 !FLAG_IS_DEFAULT(UseSHM) || \ 3361 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \ 3362 warning(format, __VA_ARGS__); \ 3363 } \ 3364 } while (0) 3365 3366 #define shm_warning(str) shm_warning_format("%s", str) 3367 3368 #define shm_warning_with_errno(str) \ 3369 do { \ 3370 int err = errno; \ 3371 shm_warning_format(str " (error = %d)", err); \ 3372 } while (0) 3373 3374 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) { 3375 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment"); 3376 3377 if (!is_size_aligned(alignment, SHMLBA)) { 3378 assert(false, "Code below assumes that alignment is at least SHMLBA aligned"); 3379 return NULL; 3380 } 3381 3382 // To ensure that we get 'alignment' aligned memory from shmat, 3383 // we pre-reserve aligned virtual memory and then attach to that. 3384 3385 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL); 3386 if (pre_reserved_addr == NULL) { 3387 // Couldn't pre-reserve aligned memory. 3388 shm_warning("Failed to pre-reserve aligned memory for shmat."); 3389 return NULL; 3390 } 3391 3392 // SHM_REMAP is needed to allow shmat to map over an existing mapping. 3393 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP); 3394 3395 if ((intptr_t)addr == -1) { 3396 int err = errno; 3397 shm_warning_with_errno("Failed to attach shared memory."); 3398 3399 assert(err != EACCES, "Unexpected error"); 3400 assert(err != EIDRM, "Unexpected error"); 3401 assert(err != EINVAL, "Unexpected error"); 3402 3403 // Since we don't know if the kernel unmapped the pre-reserved memory area 3404 // we can't unmap it, since that would potentially unmap memory that was 3405 // mapped from other threads. 3406 return NULL; 3407 } 3408 3409 return addr; 3410 } 3411 3412 static char* shmat_at_address(int shmid, char* req_addr) { 3413 if (!is_ptr_aligned(req_addr, SHMLBA)) { 3414 assert(false, "Requested address needs to be SHMLBA aligned"); 3415 return NULL; 3416 } 3417 3418 char* addr = (char*)shmat(shmid, req_addr, 0); 3419 3420 if ((intptr_t)addr == -1) { 3421 shm_warning_with_errno("Failed to attach shared memory."); 3422 return NULL; 3423 } 3424 3425 return addr; 3426 } 3427 3428 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) { 3429 // If a req_addr has been provided, we assume that the caller has already aligned the address. 3430 if (req_addr != NULL) { 3431 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size"); 3432 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment"); 3433 return shmat_at_address(shmid, req_addr); 3434 } 3435 3436 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically 3437 // return large page size aligned memory addresses when req_addr == NULL. 3438 // However, if the alignment is larger than the large page size, we have 3439 // to manually ensure that the memory returned is 'alignment' aligned. 3440 if (alignment > os::large_page_size()) { 3441 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size"); 3442 return shmat_with_alignment(shmid, bytes, alignment); 3443 } else { 3444 return shmat_at_address(shmid, NULL); 3445 } 3446 } 3447 3448 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3449 char* req_addr, bool exec) { 3450 // "exec" is passed in but not used. Creating the shared image for 3451 // the code cache doesn't have an SHM_X executable permission to check. 3452 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3453 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3454 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address"); 3455 3456 if (!is_size_aligned(bytes, os::large_page_size())) { 3457 return NULL; // Fallback to small pages. 3458 } 3459 3460 // Create a large shared memory region to attach to based on size. 3461 // Currently, size is the total size of the heap. 3462 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3463 if (shmid == -1) { 3464 // Possible reasons for shmget failure: 3465 // 1. shmmax is too small for Java heap. 3466 // > check shmmax value: cat /proc/sys/kernel/shmmax 3467 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3468 // 2. not enough large page memory. 3469 // > check available large pages: cat /proc/meminfo 3470 // > increase amount of large pages: 3471 // echo new_value > /proc/sys/vm/nr_hugepages 3472 // Note 1: different Linux may use different name for this property, 3473 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3474 // Note 2: it's possible there's enough physical memory available but 3475 // they are so fragmented after a long run that they can't 3476 // coalesce into large pages. Try to reserve large pages when 3477 // the system is still "fresh". 3478 shm_warning_with_errno("Failed to reserve shared memory."); 3479 return NULL; 3480 } 3481 3482 // Attach to the region. 3483 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr); 3484 3485 // Remove shmid. If shmat() is successful, the actual shared memory segment 3486 // will be deleted when it's detached by shmdt() or when the process 3487 // terminates. If shmat() is not successful this will remove the shared 3488 // segment immediately. 3489 shmctl(shmid, IPC_RMID, NULL); 3490 3491 return addr; 3492 } 3493 3494 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3495 int error) { 3496 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3497 3498 bool warn_on_failure = UseLargePages && 3499 (!FLAG_IS_DEFAULT(UseLargePages) || 3500 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3501 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3502 3503 if (warn_on_failure) { 3504 char msg[128]; 3505 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3506 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3507 warning("%s", msg); 3508 } 3509 } 3510 3511 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3512 char* req_addr, 3513 bool exec) { 3514 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3515 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3516 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3517 3518 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3519 char* addr = (char*)::mmap(req_addr, bytes, prot, 3520 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3521 -1, 0); 3522 3523 if (addr == MAP_FAILED) { 3524 warn_on_large_pages_failure(req_addr, bytes, errno); 3525 return NULL; 3526 } 3527 3528 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3529 3530 return addr; 3531 } 3532 3533 // Reserve memory using mmap(MAP_HUGETLB). 3534 // - bytes shall be a multiple of alignment. 3535 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3536 // - alignment sets the alignment at which memory shall be allocated. 3537 // It must be a multiple of allocation granularity. 3538 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3539 // req_addr or NULL. 3540 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3541 size_t alignment, 3542 char* req_addr, 3543 bool exec) { 3544 size_t large_page_size = os::large_page_size(); 3545 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3546 3547 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3548 assert(is_size_aligned(bytes, alignment), "Must be"); 3549 3550 // First reserve - but not commit - the address range in small pages. 3551 char* const start = anon_mmap_aligned(bytes, alignment, req_addr); 3552 3553 if (start == NULL) { 3554 return NULL; 3555 } 3556 3557 assert(is_ptr_aligned(start, alignment), "Must be"); 3558 3559 char* end = start + bytes; 3560 3561 // Find the regions of the allocated chunk that can be promoted to large pages. 3562 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3563 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3564 3565 size_t lp_bytes = lp_end - lp_start; 3566 3567 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3568 3569 if (lp_bytes == 0) { 3570 // The mapped region doesn't even span the start and the end of a large page. 3571 // Fall back to allocate a non-special area. 3572 ::munmap(start, end - start); 3573 return NULL; 3574 } 3575 3576 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3577 3578 void* result; 3579 3580 // Commit small-paged leading area. 3581 if (start != lp_start) { 3582 result = ::mmap(start, lp_start - start, prot, 3583 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3584 -1, 0); 3585 if (result == MAP_FAILED) { 3586 ::munmap(lp_start, end - lp_start); 3587 return NULL; 3588 } 3589 } 3590 3591 // Commit large-paged area. 3592 result = ::mmap(lp_start, lp_bytes, prot, 3593 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3594 -1, 0); 3595 if (result == MAP_FAILED) { 3596 warn_on_large_pages_failure(lp_start, lp_bytes, errno); 3597 // If the mmap above fails, the large pages region will be unmapped and we 3598 // have regions before and after with small pages. Release these regions. 3599 // 3600 // | mapped | unmapped | mapped | 3601 // ^ ^ ^ ^ 3602 // start lp_start lp_end end 3603 // 3604 ::munmap(start, lp_start - start); 3605 ::munmap(lp_end, end - lp_end); 3606 return NULL; 3607 } 3608 3609 // Commit small-paged trailing area. 3610 if (lp_end != end) { 3611 result = ::mmap(lp_end, end - lp_end, prot, 3612 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3613 -1, 0); 3614 if (result == MAP_FAILED) { 3615 ::munmap(start, lp_end - start); 3616 return NULL; 3617 } 3618 } 3619 3620 return start; 3621 } 3622 3623 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3624 size_t alignment, 3625 char* req_addr, 3626 bool exec) { 3627 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3628 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3629 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be"); 3630 assert(is_power_of_2(os::large_page_size()), "Must be"); 3631 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3632 3633 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3634 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3635 } else { 3636 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3637 } 3638 } 3639 3640 char* os::reserve_memory_special(size_t bytes, size_t alignment, 3641 char* req_addr, bool exec) { 3642 assert(UseLargePages, "only for large pages"); 3643 3644 char* addr; 3645 if (UseSHM) { 3646 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3647 } else { 3648 assert(UseHugeTLBFS, "must be"); 3649 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3650 } 3651 3652 if (addr != NULL) { 3653 if (UseNUMAInterleaving) { 3654 numa_make_global(addr, bytes); 3655 } 3656 3657 // The memory is committed 3658 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3659 } 3660 3661 return addr; 3662 } 3663 3664 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3665 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3666 return shmdt(base) == 0; 3667 } 3668 3669 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3670 return pd_release_memory(base, bytes); 3671 } 3672 3673 bool os::release_memory_special(char* base, size_t bytes) { 3674 bool res; 3675 if (MemTracker::tracking_level() > NMT_minimal) { 3676 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3677 res = os::Linux::release_memory_special_impl(base, bytes); 3678 if (res) { 3679 tkr.record((address)base, bytes); 3680 } 3681 3682 } else { 3683 res = os::Linux::release_memory_special_impl(base, bytes); 3684 } 3685 return res; 3686 } 3687 3688 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3689 assert(UseLargePages, "only for large pages"); 3690 bool res; 3691 3692 if (UseSHM) { 3693 res = os::Linux::release_memory_special_shm(base, bytes); 3694 } else { 3695 assert(UseHugeTLBFS, "must be"); 3696 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3697 } 3698 return res; 3699 } 3700 3701 size_t os::large_page_size() { 3702 return _large_page_size; 3703 } 3704 3705 // With SysV SHM the entire memory region must be allocated as shared 3706 // memory. 3707 // HugeTLBFS allows application to commit large page memory on demand. 3708 // However, when committing memory with HugeTLBFS fails, the region 3709 // that was supposed to be committed will lose the old reservation 3710 // and allow other threads to steal that memory region. Because of this 3711 // behavior we can't commit HugeTLBFS memory. 3712 bool os::can_commit_large_page_memory() { 3713 return UseTransparentHugePages; 3714 } 3715 3716 bool os::can_execute_large_page_memory() { 3717 return UseTransparentHugePages || UseHugeTLBFS; 3718 } 3719 3720 // Reserve memory at an arbitrary address, only if that area is 3721 // available (and not reserved for something else). 3722 3723 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3724 const int max_tries = 10; 3725 char* base[max_tries]; 3726 size_t size[max_tries]; 3727 const size_t gap = 0x000000; 3728 3729 // Assert only that the size is a multiple of the page size, since 3730 // that's all that mmap requires, and since that's all we really know 3731 // about at this low abstraction level. If we need higher alignment, 3732 // we can either pass an alignment to this method or verify alignment 3733 // in one of the methods further up the call chain. See bug 5044738. 3734 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3735 3736 // Repeatedly allocate blocks until the block is allocated at the 3737 // right spot. 3738 3739 // Linux mmap allows caller to pass an address as hint; give it a try first, 3740 // if kernel honors the hint then we can return immediately. 3741 char * addr = anon_mmap(requested_addr, bytes, false); 3742 if (addr == requested_addr) { 3743 return requested_addr; 3744 } 3745 3746 if (addr != NULL) { 3747 // mmap() is successful but it fails to reserve at the requested address 3748 anon_munmap(addr, bytes); 3749 } 3750 3751 int i; 3752 for (i = 0; i < max_tries; ++i) { 3753 base[i] = reserve_memory(bytes); 3754 3755 if (base[i] != NULL) { 3756 // Is this the block we wanted? 3757 if (base[i] == requested_addr) { 3758 size[i] = bytes; 3759 break; 3760 } 3761 3762 // Does this overlap the block we wanted? Give back the overlapped 3763 // parts and try again. 3764 3765 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3766 if (top_overlap >= 0 && (size_t)top_overlap < bytes) { 3767 unmap_memory(base[i], top_overlap); 3768 base[i] += top_overlap; 3769 size[i] = bytes - top_overlap; 3770 } else { 3771 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr; 3772 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) { 3773 unmap_memory(requested_addr, bottom_overlap); 3774 size[i] = bytes - bottom_overlap; 3775 } else { 3776 size[i] = bytes; 3777 } 3778 } 3779 } 3780 } 3781 3782 // Give back the unused reserved pieces. 3783 3784 for (int j = 0; j < i; ++j) { 3785 if (base[j] != NULL) { 3786 unmap_memory(base[j], size[j]); 3787 } 3788 } 3789 3790 if (i < max_tries) { 3791 return requested_addr; 3792 } else { 3793 return NULL; 3794 } 3795 } 3796 3797 size_t os::read(int fd, void *buf, unsigned int nBytes) { 3798 return ::read(fd, buf, nBytes); 3799 } 3800 3801 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) { 3802 return ::pread(fd, buf, nBytes, offset); 3803 } 3804 3805 // Short sleep, direct OS call. 3806 // 3807 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3808 // sched_yield(2) will actually give up the CPU: 3809 // 3810 // * Alone on this pariticular CPU, keeps running. 3811 // * Before the introduction of "skip_buddy" with "compat_yield" disabled 3812 // (pre 2.6.39). 3813 // 3814 // So calling this with 0 is an alternative. 3815 // 3816 void os::naked_short_sleep(jlong ms) { 3817 struct timespec req; 3818 3819 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3820 req.tv_sec = 0; 3821 if (ms > 0) { 3822 req.tv_nsec = (ms % 1000) * 1000000; 3823 } else { 3824 req.tv_nsec = 1; 3825 } 3826 3827 nanosleep(&req, NULL); 3828 3829 return; 3830 } 3831 3832 // Sleep forever; naked call to OS-specific sleep; use with CAUTION 3833 void os::infinite_sleep() { 3834 while (true) { // sleep forever ... 3835 ::sleep(100); // ... 100 seconds at a time 3836 } 3837 } 3838 3839 // Used to convert frequent JVM_Yield() to nops 3840 bool os::dont_yield() { 3841 return DontYieldALot; 3842 } 3843 3844 void os::naked_yield() { 3845 sched_yield(); 3846 } 3847 3848 //////////////////////////////////////////////////////////////////////////////// 3849 // thread priority support 3850 3851 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3852 // only supports dynamic priority, static priority must be zero. For real-time 3853 // applications, Linux supports SCHED_RR which allows static priority (1-99). 3854 // However, for large multi-threaded applications, SCHED_RR is not only slower 3855 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3856 // of 5 runs - Sep 2005). 3857 // 3858 // The following code actually changes the niceness of kernel-thread/LWP. It 3859 // has an assumption that setpriority() only modifies one kernel-thread/LWP, 3860 // not the entire user process, and user level threads are 1:1 mapped to kernel 3861 // threads. It has always been the case, but could change in the future. For 3862 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3863 // It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3864 3865 int os::java_to_os_priority[CriticalPriority + 1] = { 3866 19, // 0 Entry should never be used 3867 3868 4, // 1 MinPriority 3869 3, // 2 3870 2, // 3 3871 3872 1, // 4 3873 0, // 5 NormPriority 3874 -1, // 6 3875 3876 -2, // 7 3877 -3, // 8 3878 -4, // 9 NearMaxPriority 3879 3880 -5, // 10 MaxPriority 3881 3882 -5 // 11 CriticalPriority 3883 }; 3884 3885 static int prio_init() { 3886 if (ThreadPriorityPolicy == 1) { 3887 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3888 // if effective uid is not root. Perhaps, a more elegant way of doing 3889 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3890 if (geteuid() != 0) { 3891 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3892 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3893 } 3894 ThreadPriorityPolicy = 0; 3895 } 3896 } 3897 if (UseCriticalJavaThreadPriority) { 3898 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3899 } 3900 return 0; 3901 } 3902 3903 OSReturn os::set_native_priority(Thread* thread, int newpri) { 3904 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 3905 3906 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3907 return (ret == 0) ? OS_OK : OS_ERR; 3908 } 3909 3910 OSReturn os::get_native_priority(const Thread* const thread, 3911 int *priority_ptr) { 3912 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 3913 *priority_ptr = java_to_os_priority[NormPriority]; 3914 return OS_OK; 3915 } 3916 3917 errno = 0; 3918 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3919 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3920 } 3921 3922 // Hint to the underlying OS that a task switch would not be good. 3923 // Void return because it's a hint and can fail. 3924 void os::hint_no_preempt() {} 3925 3926 //////////////////////////////////////////////////////////////////////////////// 3927 // suspend/resume support 3928 3929 // the low-level signal-based suspend/resume support is a remnant from the 3930 // old VM-suspension that used to be for java-suspension, safepoints etc, 3931 // within hotspot. Now there is a single use-case for this: 3932 // - calling get_thread_pc() on the VMThread by the flat-profiler task 3933 // that runs in the watcher thread. 3934 // The remaining code is greatly simplified from the more general suspension 3935 // code that used to be used. 3936 // 3937 // The protocol is quite simple: 3938 // - suspend: 3939 // - sends a signal to the target thread 3940 // - polls the suspend state of the osthread using a yield loop 3941 // - target thread signal handler (SR_handler) sets suspend state 3942 // and blocks in sigsuspend until continued 3943 // - resume: 3944 // - sets target osthread state to continue 3945 // - sends signal to end the sigsuspend loop in the SR_handler 3946 // 3947 // Note that the SR_lock plays no role in this suspend/resume protocol, 3948 // but is checked for NULL in SR_handler as a thread termination indicator. 3949 3950 static void resume_clear_context(OSThread *osthread) { 3951 osthread->set_ucontext(NULL); 3952 osthread->set_siginfo(NULL); 3953 } 3954 3955 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 3956 ucontext_t* context) { 3957 osthread->set_ucontext(context); 3958 osthread->set_siginfo(siginfo); 3959 } 3960 3961 // Handler function invoked when a thread's execution is suspended or 3962 // resumed. We have to be careful that only async-safe functions are 3963 // called here (Note: most pthread functions are not async safe and 3964 // should be avoided.) 3965 // 3966 // Note: sigwait() is a more natural fit than sigsuspend() from an 3967 // interface point of view, but sigwait() prevents the signal hander 3968 // from being run. libpthread would get very confused by not having 3969 // its signal handlers run and prevents sigwait()'s use with the 3970 // mutex granting granting signal. 3971 // 3972 // Currently only ever called on the VMThread and JavaThreads (PC sampling) 3973 // 3974 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3975 // Save and restore errno to avoid confusing native code with EINTR 3976 // after sigsuspend. 3977 int old_errno = errno; 3978 3979 Thread* thread = Thread::current_or_null_safe(); 3980 assert(thread != NULL, "Missing current thread in SR_handler"); 3981 3982 // On some systems we have seen signal delivery get "stuck" until the signal 3983 // mask is changed as part of thread termination. Check that the current thread 3984 // has not already terminated (via SR_lock()) - else the following assertion 3985 // will fail because the thread is no longer a JavaThread as the ~JavaThread 3986 // destructor has completed. 3987 3988 if (thread->SR_lock() == NULL) { 3989 return; 3990 } 3991 3992 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3993 3994 OSThread* osthread = thread->osthread(); 3995 3996 os::SuspendResume::State current = osthread->sr.state(); 3997 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3998 suspend_save_context(osthread, siginfo, context); 3999 4000 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4001 os::SuspendResume::State state = osthread->sr.suspended(); 4002 if (state == os::SuspendResume::SR_SUSPENDED) { 4003 sigset_t suspend_set; // signals for sigsuspend() 4004 sigemptyset(&suspend_set); 4005 // get current set of blocked signals and unblock resume signal 4006 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4007 sigdelset(&suspend_set, SR_signum); 4008 4009 sr_semaphore.signal(); 4010 // wait here until we are resumed 4011 while (1) { 4012 sigsuspend(&suspend_set); 4013 4014 os::SuspendResume::State result = osthread->sr.running(); 4015 if (result == os::SuspendResume::SR_RUNNING) { 4016 sr_semaphore.signal(); 4017 break; 4018 } 4019 } 4020 4021 } else if (state == os::SuspendResume::SR_RUNNING) { 4022 // request was cancelled, continue 4023 } else { 4024 ShouldNotReachHere(); 4025 } 4026 4027 resume_clear_context(osthread); 4028 } else if (current == os::SuspendResume::SR_RUNNING) { 4029 // request was cancelled, continue 4030 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4031 // ignore 4032 } else { 4033 // ignore 4034 } 4035 4036 errno = old_errno; 4037 } 4038 4039 static int SR_initialize() { 4040 struct sigaction act; 4041 char *s; 4042 4043 // Get signal number to use for suspend/resume 4044 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4045 int sig = ::strtol(s, 0, 10); 4046 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769. 4047 sig < NSIG) { // Must be legal signal and fit into sigflags[]. 4048 SR_signum = sig; 4049 } else { 4050 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.", 4051 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum); 4052 } 4053 } 4054 4055 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4056 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4057 4058 sigemptyset(&SR_sigset); 4059 sigaddset(&SR_sigset, SR_signum); 4060 4061 // Set up signal handler for suspend/resume 4062 act.sa_flags = SA_RESTART|SA_SIGINFO; 4063 act.sa_handler = (void (*)(int)) SR_handler; 4064 4065 // SR_signum is blocked by default. 4066 // 4528190 - We also need to block pthread restart signal (32 on all 4067 // supported Linux platforms). Note that LinuxThreads need to block 4068 // this signal for all threads to work properly. So we don't have 4069 // to use hard-coded signal number when setting up the mask. 4070 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4071 4072 if (sigaction(SR_signum, &act, 0) == -1) { 4073 return -1; 4074 } 4075 4076 // Save signal flag 4077 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4078 return 0; 4079 } 4080 4081 static int sr_notify(OSThread* osthread) { 4082 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4083 assert_status(status == 0, status, "pthread_kill"); 4084 return status; 4085 } 4086 4087 // "Randomly" selected value for how long we want to spin 4088 // before bailing out on suspending a thread, also how often 4089 // we send a signal to a thread we want to resume 4090 static const int RANDOMLY_LARGE_INTEGER = 1000000; 4091 static const int RANDOMLY_LARGE_INTEGER2 = 100; 4092 4093 // returns true on success and false on error - really an error is fatal 4094 // but this seems the normal response to library errors 4095 static bool do_suspend(OSThread* osthread) { 4096 assert(osthread->sr.is_running(), "thread should be running"); 4097 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4098 4099 // mark as suspended and send signal 4100 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4101 // failed to switch, state wasn't running? 4102 ShouldNotReachHere(); 4103 return false; 4104 } 4105 4106 if (sr_notify(osthread) != 0) { 4107 ShouldNotReachHere(); 4108 } 4109 4110 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4111 while (true) { 4112 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4113 break; 4114 } else { 4115 // timeout 4116 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4117 if (cancelled == os::SuspendResume::SR_RUNNING) { 4118 return false; 4119 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4120 // make sure that we consume the signal on the semaphore as well 4121 sr_semaphore.wait(); 4122 break; 4123 } else { 4124 ShouldNotReachHere(); 4125 return false; 4126 } 4127 } 4128 } 4129 4130 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4131 return true; 4132 } 4133 4134 static void do_resume(OSThread* osthread) { 4135 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4136 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4137 4138 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4139 // failed to switch to WAKEUP_REQUEST 4140 ShouldNotReachHere(); 4141 return; 4142 } 4143 4144 while (true) { 4145 if (sr_notify(osthread) == 0) { 4146 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4147 if (osthread->sr.is_running()) { 4148 return; 4149 } 4150 } 4151 } else { 4152 ShouldNotReachHere(); 4153 } 4154 } 4155 4156 guarantee(osthread->sr.is_running(), "Must be running!"); 4157 } 4158 4159 /////////////////////////////////////////////////////////////////////////////////// 4160 // signal handling (except suspend/resume) 4161 4162 // This routine may be used by user applications as a "hook" to catch signals. 4163 // The user-defined signal handler must pass unrecognized signals to this 4164 // routine, and if it returns true (non-zero), then the signal handler must 4165 // return immediately. If the flag "abort_if_unrecognized" is true, then this 4166 // routine will never retun false (zero), but instead will execute a VM panic 4167 // routine kill the process. 4168 // 4169 // If this routine returns false, it is OK to call it again. This allows 4170 // the user-defined signal handler to perform checks either before or after 4171 // the VM performs its own checks. Naturally, the user code would be making 4172 // a serious error if it tried to handle an exception (such as a null check 4173 // or breakpoint) that the VM was generating for its own correct operation. 4174 // 4175 // This routine may recognize any of the following kinds of signals: 4176 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4177 // It should be consulted by handlers for any of those signals. 4178 // 4179 // The caller of this routine must pass in the three arguments supplied 4180 // to the function referred to in the "sa_sigaction" (not the "sa_handler") 4181 // field of the structure passed to sigaction(). This routine assumes that 4182 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4183 // 4184 // Note that the VM will print warnings if it detects conflicting signal 4185 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4186 // 4187 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4188 siginfo_t* siginfo, 4189 void* ucontext, 4190 int abort_if_unrecognized); 4191 4192 void signalHandler(int sig, siginfo_t* info, void* uc) { 4193 assert(info != NULL && uc != NULL, "it must be old kernel"); 4194 int orig_errno = errno; // Preserve errno value over signal handler. 4195 JVM_handle_linux_signal(sig, info, uc, true); 4196 errno = orig_errno; 4197 } 4198 4199 4200 // This boolean allows users to forward their own non-matching signals 4201 // to JVM_handle_linux_signal, harmlessly. 4202 bool os::Linux::signal_handlers_are_installed = false; 4203 4204 // For signal-chaining 4205 struct sigaction sigact[NSIG]; 4206 uint64_t sigs = 0; 4207 #if (64 < NSIG-1) 4208 #error "Not all signals can be encoded in sigs. Adapt its type!" 4209 #endif 4210 bool os::Linux::libjsig_is_loaded = false; 4211 typedef struct sigaction *(*get_signal_t)(int); 4212 get_signal_t os::Linux::get_signal_action = NULL; 4213 4214 struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4215 struct sigaction *actp = NULL; 4216 4217 if (libjsig_is_loaded) { 4218 // Retrieve the old signal handler from libjsig 4219 actp = (*get_signal_action)(sig); 4220 } 4221 if (actp == NULL) { 4222 // Retrieve the preinstalled signal handler from jvm 4223 actp = get_preinstalled_handler(sig); 4224 } 4225 4226 return actp; 4227 } 4228 4229 static bool call_chained_handler(struct sigaction *actp, int sig, 4230 siginfo_t *siginfo, void *context) { 4231 // Call the old signal handler 4232 if (actp->sa_handler == SIG_DFL) { 4233 // It's more reasonable to let jvm treat it as an unexpected exception 4234 // instead of taking the default action. 4235 return false; 4236 } else if (actp->sa_handler != SIG_IGN) { 4237 if ((actp->sa_flags & SA_NODEFER) == 0) { 4238 // automaticlly block the signal 4239 sigaddset(&(actp->sa_mask), sig); 4240 } 4241 4242 sa_handler_t hand = NULL; 4243 sa_sigaction_t sa = NULL; 4244 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4245 // retrieve the chained handler 4246 if (siginfo_flag_set) { 4247 sa = actp->sa_sigaction; 4248 } else { 4249 hand = actp->sa_handler; 4250 } 4251 4252 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4253 actp->sa_handler = SIG_DFL; 4254 } 4255 4256 // try to honor the signal mask 4257 sigset_t oset; 4258 sigemptyset(&oset); 4259 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4260 4261 // call into the chained handler 4262 if (siginfo_flag_set) { 4263 (*sa)(sig, siginfo, context); 4264 } else { 4265 (*hand)(sig); 4266 } 4267 4268 // restore the signal mask 4269 pthread_sigmask(SIG_SETMASK, &oset, NULL); 4270 } 4271 // Tell jvm's signal handler the signal is taken care of. 4272 return true; 4273 } 4274 4275 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4276 bool chained = false; 4277 // signal-chaining 4278 if (UseSignalChaining) { 4279 struct sigaction *actp = get_chained_signal_action(sig); 4280 if (actp != NULL) { 4281 chained = call_chained_handler(actp, sig, siginfo, context); 4282 } 4283 } 4284 return chained; 4285 } 4286 4287 struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4288 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) { 4289 return &sigact[sig]; 4290 } 4291 return NULL; 4292 } 4293 4294 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4295 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4296 sigact[sig] = oldAct; 4297 sigs |= (uint64_t)1 << (sig-1); 4298 } 4299 4300 // for diagnostic 4301 int sigflags[NSIG]; 4302 4303 int os::Linux::get_our_sigflags(int sig) { 4304 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4305 return sigflags[sig]; 4306 } 4307 4308 void os::Linux::set_our_sigflags(int sig, int flags) { 4309 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4310 if (sig > 0 && sig < NSIG) { 4311 sigflags[sig] = flags; 4312 } 4313 } 4314 4315 void os::Linux::set_signal_handler(int sig, bool set_installed) { 4316 // Check for overwrite. 4317 struct sigaction oldAct; 4318 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4319 4320 void* oldhand = oldAct.sa_sigaction 4321 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4322 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4323 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4324 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4325 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4326 if (AllowUserSignalHandlers || !set_installed) { 4327 // Do not overwrite; user takes responsibility to forward to us. 4328 return; 4329 } else if (UseSignalChaining) { 4330 // save the old handler in jvm 4331 save_preinstalled_handler(sig, oldAct); 4332 // libjsig also interposes the sigaction() call below and saves the 4333 // old sigaction on it own. 4334 } else { 4335 fatal("Encountered unexpected pre-existing sigaction handler " 4336 "%#lx for signal %d.", (long)oldhand, sig); 4337 } 4338 } 4339 4340 struct sigaction sigAct; 4341 sigfillset(&(sigAct.sa_mask)); 4342 sigAct.sa_handler = SIG_DFL; 4343 if (!set_installed) { 4344 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4345 } else { 4346 sigAct.sa_sigaction = signalHandler; 4347 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4348 } 4349 // Save flags, which are set by ours 4350 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4351 sigflags[sig] = sigAct.sa_flags; 4352 4353 int ret = sigaction(sig, &sigAct, &oldAct); 4354 assert(ret == 0, "check"); 4355 4356 void* oldhand2 = oldAct.sa_sigaction 4357 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4358 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4359 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4360 } 4361 4362 // install signal handlers for signals that HotSpot needs to 4363 // handle in order to support Java-level exception handling. 4364 4365 void os::Linux::install_signal_handlers() { 4366 if (!signal_handlers_are_installed) { 4367 signal_handlers_are_installed = true; 4368 4369 // signal-chaining 4370 typedef void (*signal_setting_t)(); 4371 signal_setting_t begin_signal_setting = NULL; 4372 signal_setting_t end_signal_setting = NULL; 4373 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4374 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4375 if (begin_signal_setting != NULL) { 4376 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4377 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4378 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4379 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4380 libjsig_is_loaded = true; 4381 assert(UseSignalChaining, "should enable signal-chaining"); 4382 } 4383 if (libjsig_is_loaded) { 4384 // Tell libjsig jvm is setting signal handlers 4385 (*begin_signal_setting)(); 4386 } 4387 4388 set_signal_handler(SIGSEGV, true); 4389 set_signal_handler(SIGPIPE, true); 4390 set_signal_handler(SIGBUS, true); 4391 set_signal_handler(SIGILL, true); 4392 set_signal_handler(SIGFPE, true); 4393 #if defined(PPC64) 4394 set_signal_handler(SIGTRAP, true); 4395 #endif 4396 set_signal_handler(SIGXFSZ, true); 4397 4398 if (libjsig_is_loaded) { 4399 // Tell libjsig jvm finishes setting signal handlers 4400 (*end_signal_setting)(); 4401 } 4402 4403 // We don't activate signal checker if libjsig is in place, we trust ourselves 4404 // and if UserSignalHandler is installed all bets are off. 4405 // Log that signal checking is off only if -verbose:jni is specified. 4406 if (CheckJNICalls) { 4407 if (libjsig_is_loaded) { 4408 if (PrintJNIResolving) { 4409 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4410 } 4411 check_signals = false; 4412 } 4413 if (AllowUserSignalHandlers) { 4414 if (PrintJNIResolving) { 4415 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4416 } 4417 check_signals = false; 4418 } 4419 } 4420 } 4421 } 4422 4423 // This is the fastest way to get thread cpu time on Linux. 4424 // Returns cpu time (user+sys) for any thread, not only for current. 4425 // POSIX compliant clocks are implemented in the kernels 2.6.16+. 4426 // It might work on 2.6.10+ with a special kernel/glibc patch. 4427 // For reference, please, see IEEE Std 1003.1-2004: 4428 // http://www.unix.org/single_unix_specification 4429 4430 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4431 struct timespec tp; 4432 int rc = os::Linux::clock_gettime(clockid, &tp); 4433 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4434 4435 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4436 } 4437 4438 void os::Linux::initialize_os_info() { 4439 assert(_os_version == 0, "OS info already initialized"); 4440 4441 struct utsname _uname; 4442 4443 uint32_t major; 4444 uint32_t minor; 4445 uint32_t fix; 4446 4447 int rc; 4448 4449 // Kernel version is unknown if 4450 // verification below fails. 4451 _os_version = 0x01000000; 4452 4453 rc = uname(&_uname); 4454 if (rc != -1) { 4455 4456 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix); 4457 if (rc == 3) { 4458 4459 if (major < 256 && minor < 256 && fix < 256) { 4460 // Kernel version format is as expected, 4461 // set it overriding unknown state. 4462 _os_version = (major << 16) | 4463 (minor << 8 ) | 4464 (fix << 0 ) ; 4465 } 4466 } 4467 } 4468 } 4469 4470 uint32_t os::Linux::os_version() { 4471 assert(_os_version != 0, "not initialized"); 4472 return _os_version & 0x00FFFFFF; 4473 } 4474 4475 bool os::Linux::os_version_is_known() { 4476 assert(_os_version != 0, "not initialized"); 4477 return _os_version & 0x01000000 ? false : true; 4478 } 4479 4480 ///// 4481 // glibc on Linux platform uses non-documented flag 4482 // to indicate, that some special sort of signal 4483 // trampoline is used. 4484 // We will never set this flag, and we should 4485 // ignore this flag in our diagnostic 4486 #ifdef SIGNIFICANT_SIGNAL_MASK 4487 #undef SIGNIFICANT_SIGNAL_MASK 4488 #endif 4489 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4490 4491 static const char* get_signal_handler_name(address handler, 4492 char* buf, int buflen) { 4493 int offset = 0; 4494 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4495 if (found) { 4496 // skip directory names 4497 const char *p1, *p2; 4498 p1 = buf; 4499 size_t len = strlen(os::file_separator()); 4500 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4501 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4502 } else { 4503 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4504 } 4505 return buf; 4506 } 4507 4508 static void print_signal_handler(outputStream* st, int sig, 4509 char* buf, size_t buflen) { 4510 struct sigaction sa; 4511 4512 sigaction(sig, NULL, &sa); 4513 4514 // See comment for SIGNIFICANT_SIGNAL_MASK define 4515 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4516 4517 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4518 4519 address handler = (sa.sa_flags & SA_SIGINFO) 4520 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4521 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4522 4523 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4524 st->print("SIG_DFL"); 4525 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4526 st->print("SIG_IGN"); 4527 } else { 4528 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4529 } 4530 4531 st->print(", sa_mask[0]="); 4532 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4533 4534 address rh = VMError::get_resetted_sighandler(sig); 4535 // May be, handler was resetted by VMError? 4536 if (rh != NULL) { 4537 handler = rh; 4538 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4539 } 4540 4541 st->print(", sa_flags="); 4542 os::Posix::print_sa_flags(st, sa.sa_flags); 4543 4544 // Check: is it our handler? 4545 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4546 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4547 // It is our signal handler 4548 // check for flags, reset system-used one! 4549 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4550 st->print( 4551 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4552 os::Linux::get_our_sigflags(sig)); 4553 } 4554 } 4555 st->cr(); 4556 } 4557 4558 4559 #define DO_SIGNAL_CHECK(sig) \ 4560 do { \ 4561 if (!sigismember(&check_signal_done, sig)) { \ 4562 os::Linux::check_signal_handler(sig); \ 4563 } \ 4564 } while (0) 4565 4566 // This method is a periodic task to check for misbehaving JNI applications 4567 // under CheckJNI, we can add any periodic checks here 4568 4569 void os::run_periodic_checks() { 4570 if (check_signals == false) return; 4571 4572 // SEGV and BUS if overridden could potentially prevent 4573 // generation of hs*.log in the event of a crash, debugging 4574 // such a case can be very challenging, so we absolutely 4575 // check the following for a good measure: 4576 DO_SIGNAL_CHECK(SIGSEGV); 4577 DO_SIGNAL_CHECK(SIGILL); 4578 DO_SIGNAL_CHECK(SIGFPE); 4579 DO_SIGNAL_CHECK(SIGBUS); 4580 DO_SIGNAL_CHECK(SIGPIPE); 4581 DO_SIGNAL_CHECK(SIGXFSZ); 4582 #if defined(PPC64) 4583 DO_SIGNAL_CHECK(SIGTRAP); 4584 #endif 4585 4586 // ReduceSignalUsage allows the user to override these handlers 4587 // see comments at the very top and jvm_solaris.h 4588 if (!ReduceSignalUsage) { 4589 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4590 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4591 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4592 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4593 } 4594 4595 DO_SIGNAL_CHECK(SR_signum); 4596 } 4597 4598 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4599 4600 static os_sigaction_t os_sigaction = NULL; 4601 4602 void os::Linux::check_signal_handler(int sig) { 4603 char buf[O_BUFLEN]; 4604 address jvmHandler = NULL; 4605 4606 4607 struct sigaction act; 4608 if (os_sigaction == NULL) { 4609 // only trust the default sigaction, in case it has been interposed 4610 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4611 if (os_sigaction == NULL) return; 4612 } 4613 4614 os_sigaction(sig, (struct sigaction*)NULL, &act); 4615 4616 4617 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4618 4619 address thisHandler = (act.sa_flags & SA_SIGINFO) 4620 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4621 : CAST_FROM_FN_PTR(address, act.sa_handler); 4622 4623 4624 switch (sig) { 4625 case SIGSEGV: 4626 case SIGBUS: 4627 case SIGFPE: 4628 case SIGPIPE: 4629 case SIGILL: 4630 case SIGXFSZ: 4631 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4632 break; 4633 4634 case SHUTDOWN1_SIGNAL: 4635 case SHUTDOWN2_SIGNAL: 4636 case SHUTDOWN3_SIGNAL: 4637 case BREAK_SIGNAL: 4638 jvmHandler = (address)user_handler(); 4639 break; 4640 4641 default: 4642 if (sig == SR_signum) { 4643 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4644 } else { 4645 return; 4646 } 4647 break; 4648 } 4649 4650 if (thisHandler != jvmHandler) { 4651 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4652 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4653 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4654 // No need to check this sig any longer 4655 sigaddset(&check_signal_done, sig); 4656 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4657 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4658 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4659 exception_name(sig, buf, O_BUFLEN)); 4660 } 4661 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4662 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4663 tty->print("expected:"); 4664 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig)); 4665 tty->cr(); 4666 tty->print(" found:"); 4667 os::Posix::print_sa_flags(tty, act.sa_flags); 4668 tty->cr(); 4669 // No need to check this sig any longer 4670 sigaddset(&check_signal_done, sig); 4671 } 4672 4673 // Dump all the signal 4674 if (sigismember(&check_signal_done, sig)) { 4675 print_signal_handlers(tty, buf, O_BUFLEN); 4676 } 4677 } 4678 4679 extern void report_error(char* file_name, int line_no, char* title, 4680 char* format, ...); 4681 4682 // this is called _before_ the most of global arguments have been parsed 4683 void os::init(void) { 4684 char dummy; // used to get a guess on initial stack address 4685 // first_hrtime = gethrtime(); 4686 4687 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4688 4689 init_random(1234567); 4690 4691 ThreadCritical::initialize(); 4692 4693 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4694 if (Linux::page_size() == -1) { 4695 fatal("os_linux.cpp: os::init: sysconf failed (%s)", 4696 os::strerror(errno)); 4697 } 4698 init_page_sizes((size_t) Linux::page_size()); 4699 4700 Linux::initialize_system_info(); 4701 4702 Linux::initialize_os_info(); 4703 4704 // main_thread points to the aboriginal thread 4705 Linux::_main_thread = pthread_self(); 4706 4707 Linux::clock_init(); 4708 initial_time_count = javaTimeNanos(); 4709 4710 // pthread_condattr initialization for monotonic clock 4711 int status; 4712 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4713 if ((status = pthread_condattr_init(_condattr)) != 0) { 4714 fatal("pthread_condattr_init: %s", os::strerror(status)); 4715 } 4716 // Only set the clock if CLOCK_MONOTONIC is available 4717 if (os::supports_monotonic_clock()) { 4718 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4719 if (status == EINVAL) { 4720 warning("Unable to use monotonic clock with relative timed-waits" \ 4721 " - changes to the time-of-day clock may have adverse affects"); 4722 } else { 4723 fatal("pthread_condattr_setclock: %s", os::strerror(status)); 4724 } 4725 } 4726 } 4727 // else it defaults to CLOCK_REALTIME 4728 4729 // retrieve entry point for pthread_setname_np 4730 Linux::_pthread_setname_np = 4731 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4732 4733 } 4734 4735 // To install functions for atexit system call 4736 extern "C" { 4737 static void perfMemory_exit_helper() { 4738 perfMemory_exit(); 4739 } 4740 } 4741 4742 // this is called _after_ the global arguments have been parsed 4743 jint os::init_2(void) { 4744 Linux::fast_thread_clock_init(); 4745 4746 // Allocate a single page and mark it as readable for safepoint polling 4747 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4748 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page"); 4749 4750 os::set_polling_page(polling_page); 4751 log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page)); 4752 4753 if (!UseMembar) { 4754 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4755 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4756 os::set_memory_serialize_page(mem_serialize_page); 4757 log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page)); 4758 } 4759 4760 // initialize suspend/resume support - must do this before signal_sets_init() 4761 if (SR_initialize() != 0) { 4762 perror("SR_initialize failed"); 4763 return JNI_ERR; 4764 } 4765 4766 Linux::signal_sets_init(); 4767 Linux::install_signal_handlers(); 4768 4769 // Check and sets minimum stack sizes against command line options 4770 if (Posix::set_minimum_stack_sizes() == JNI_ERR) { 4771 return JNI_ERR; 4772 } 4773 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4774 4775 #if defined(IA32) 4776 workaround_expand_exec_shield_cs_limit(); 4777 #endif 4778 4779 Linux::libpthread_init(); 4780 log_info(os)("HotSpot is running with %s, %s", 4781 Linux::glibc_version(), Linux::libpthread_version()); 4782 4783 if (UseNUMA) { 4784 if (!Linux::libnuma_init()) { 4785 UseNUMA = false; 4786 } else { 4787 if ((Linux::numa_max_node() < 1)) { 4788 // There's only one node(they start from 0), disable NUMA. 4789 UseNUMA = false; 4790 } 4791 } 4792 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4793 // we can make the adaptive lgrp chunk resizing work. If the user specified 4794 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4795 // disable adaptive resizing. 4796 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4797 if (FLAG_IS_DEFAULT(UseNUMA)) { 4798 UseNUMA = false; 4799 } else { 4800 if (FLAG_IS_DEFAULT(UseLargePages) && 4801 FLAG_IS_DEFAULT(UseSHM) && 4802 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4803 UseLargePages = false; 4804 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) { 4805 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)"); 4806 UseAdaptiveSizePolicy = false; 4807 UseAdaptiveNUMAChunkSizing = false; 4808 } 4809 } 4810 } 4811 if (!UseNUMA && ForceNUMA) { 4812 UseNUMA = true; 4813 } 4814 } 4815 4816 if (MaxFDLimit) { 4817 // set the number of file descriptors to max. print out error 4818 // if getrlimit/setrlimit fails but continue regardless. 4819 struct rlimit nbr_files; 4820 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4821 if (status != 0) { 4822 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno)); 4823 } else { 4824 nbr_files.rlim_cur = nbr_files.rlim_max; 4825 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4826 if (status != 0) { 4827 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno)); 4828 } 4829 } 4830 } 4831 4832 // Initialize lock used to serialize thread creation (see os::create_thread) 4833 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4834 4835 // at-exit methods are called in the reverse order of their registration. 4836 // atexit functions are called on return from main or as a result of a 4837 // call to exit(3C). There can be only 32 of these functions registered 4838 // and atexit() does not set errno. 4839 4840 if (PerfAllowAtExitRegistration) { 4841 // only register atexit functions if PerfAllowAtExitRegistration is set. 4842 // atexit functions can be delayed until process exit time, which 4843 // can be problematic for embedded VM situations. Embedded VMs should 4844 // call DestroyJavaVM() to assure that VM resources are released. 4845 4846 // note: perfMemory_exit_helper atexit function may be removed in 4847 // the future if the appropriate cleanup code can be added to the 4848 // VM_Exit VMOperation's doit method. 4849 if (atexit(perfMemory_exit_helper) != 0) { 4850 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4851 } 4852 } 4853 4854 // initialize thread priority policy 4855 prio_init(); 4856 4857 return JNI_OK; 4858 } 4859 4860 // Mark the polling page as unreadable 4861 void os::make_polling_page_unreadable(void) { 4862 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 4863 fatal("Could not disable polling page"); 4864 } 4865 } 4866 4867 // Mark the polling page as readable 4868 void os::make_polling_page_readable(void) { 4869 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4870 fatal("Could not enable polling page"); 4871 } 4872 } 4873 4874 // older glibc versions don't have this macro (which expands to 4875 // an optimized bit-counting function) so we have to roll our own 4876 #ifndef CPU_COUNT 4877 4878 static int _cpu_count(const cpu_set_t* cpus) { 4879 int count = 0; 4880 // only look up to the number of configured processors 4881 for (int i = 0; i < os::processor_count(); i++) { 4882 if (CPU_ISSET(i, cpus)) { 4883 count++; 4884 } 4885 } 4886 return count; 4887 } 4888 4889 #define CPU_COUNT(cpus) _cpu_count(cpus) 4890 4891 #endif // CPU_COUNT 4892 4893 // Get the current number of available processors for this process. 4894 // This value can change at any time during a process's lifetime. 4895 // sched_getaffinity gives an accurate answer as it accounts for cpusets. 4896 // If it appears there may be more than 1024 processors then we do a 4897 // dynamic check - see 6515172 for details. 4898 // If anything goes wrong we fallback to returning the number of online 4899 // processors - which can be greater than the number available to the process. 4900 int os::active_processor_count() { 4901 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors 4902 cpu_set_t* cpus_p = &cpus; 4903 int cpus_size = sizeof(cpu_set_t); 4904 4905 int configured_cpus = processor_count(); // upper bound on available cpus 4906 int cpu_count = 0; 4907 4908 // old build platforms may not support dynamic cpu sets 4909 #ifdef CPU_ALLOC 4910 4911 // To enable easy testing of the dynamic path on different platforms we 4912 // introduce a diagnostic flag: UseCpuAllocPath 4913 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) { 4914 // kernel may use a mask bigger than cpu_set_t 4915 log_trace(os)("active_processor_count: using dynamic path %s" 4916 "- configured processors: %d", 4917 UseCpuAllocPath ? "(forced) " : "", 4918 configured_cpus); 4919 cpus_p = CPU_ALLOC(configured_cpus); 4920 if (cpus_p != NULL) { 4921 cpus_size = CPU_ALLOC_SIZE(configured_cpus); 4922 // zero it just to be safe 4923 CPU_ZERO_S(cpus_size, cpus_p); 4924 } 4925 else { 4926 // failed to allocate so fallback to online cpus 4927 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4928 log_trace(os)("active_processor_count: " 4929 "CPU_ALLOC failed (%s) - using " 4930 "online processor count: %d", 4931 os::strerror(errno), online_cpus); 4932 return online_cpus; 4933 } 4934 } 4935 else { 4936 log_trace(os)("active_processor_count: using static path - configured processors: %d", 4937 configured_cpus); 4938 } 4939 #else // CPU_ALLOC 4940 // these stubs won't be executed 4941 #define CPU_COUNT_S(size, cpus) -1 4942 #define CPU_FREE(cpus) 4943 4944 log_trace(os)("active_processor_count: only static path available - configured processors: %d", 4945 configured_cpus); 4946 #endif // CPU_ALLOC 4947 4948 // pid 0 means the current thread - which we have to assume represents the process 4949 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) { 4950 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 4951 cpu_count = CPU_COUNT_S(cpus_size, cpus_p); 4952 } 4953 else { 4954 cpu_count = CPU_COUNT(cpus_p); 4955 } 4956 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count); 4957 } 4958 else { 4959 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN); 4960 warning("sched_getaffinity failed (%s)- using online processor count (%d) " 4961 "which may exceed available processors", os::strerror(errno), cpu_count); 4962 } 4963 4964 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 4965 CPU_FREE(cpus_p); 4966 } 4967 4968 assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check"); 4969 return cpu_count; 4970 } 4971 4972 void os::set_native_thread_name(const char *name) { 4973 if (Linux::_pthread_setname_np) { 4974 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 4975 snprintf(buf, sizeof(buf), "%s", name); 4976 buf[sizeof(buf) - 1] = '\0'; 4977 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 4978 // ERANGE should not happen; all other errors should just be ignored. 4979 assert(rc != ERANGE, "pthread_setname_np failed"); 4980 } 4981 } 4982 4983 bool os::distribute_processes(uint length, uint* distribution) { 4984 // Not yet implemented. 4985 return false; 4986 } 4987 4988 bool os::bind_to_processor(uint processor_id) { 4989 // Not yet implemented. 4990 return false; 4991 } 4992 4993 /// 4994 4995 void os::SuspendedThreadTask::internal_do_task() { 4996 if (do_suspend(_thread->osthread())) { 4997 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4998 do_task(context); 4999 do_resume(_thread->osthread()); 5000 } 5001 } 5002 5003 class PcFetcher : public os::SuspendedThreadTask { 5004 public: 5005 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 5006 ExtendedPC result(); 5007 protected: 5008 void do_task(const os::SuspendedThreadTaskContext& context); 5009 private: 5010 ExtendedPC _epc; 5011 }; 5012 5013 ExtendedPC PcFetcher::result() { 5014 guarantee(is_done(), "task is not done yet."); 5015 return _epc; 5016 } 5017 5018 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 5019 Thread* thread = context.thread(); 5020 OSThread* osthread = thread->osthread(); 5021 if (osthread->ucontext() != NULL) { 5022 _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext()); 5023 } else { 5024 // NULL context is unexpected, double-check this is the VMThread 5025 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 5026 } 5027 } 5028 5029 // Suspends the target using the signal mechanism and then grabs the PC before 5030 // resuming the target. Used by the flat-profiler only 5031 ExtendedPC os::get_thread_pc(Thread* thread) { 5032 // Make sure that it is called by the watcher for the VMThread 5033 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 5034 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 5035 5036 PcFetcher fetcher(thread); 5037 fetcher.run(); 5038 return fetcher.result(); 5039 } 5040 5041 //////////////////////////////////////////////////////////////////////////////// 5042 // debug support 5043 5044 bool os::find(address addr, outputStream* st) { 5045 Dl_info dlinfo; 5046 memset(&dlinfo, 0, sizeof(dlinfo)); 5047 if (dladdr(addr, &dlinfo) != 0) { 5048 st->print(PTR_FORMAT ": ", p2i(addr)); 5049 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5050 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname, 5051 p2i(addr) - p2i(dlinfo.dli_saddr)); 5052 } else if (dlinfo.dli_fbase != NULL) { 5053 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase)); 5054 } else { 5055 st->print("<absolute address>"); 5056 } 5057 if (dlinfo.dli_fname != NULL) { 5058 st->print(" in %s", dlinfo.dli_fname); 5059 } 5060 if (dlinfo.dli_fbase != NULL) { 5061 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase)); 5062 } 5063 st->cr(); 5064 5065 if (Verbose) { 5066 // decode some bytes around the PC 5067 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5068 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5069 address lowest = (address) dlinfo.dli_sname; 5070 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5071 if (begin < lowest) begin = lowest; 5072 Dl_info dlinfo2; 5073 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5074 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 5075 end = (address) dlinfo2.dli_saddr; 5076 } 5077 Disassembler::decode(begin, end, st); 5078 } 5079 return true; 5080 } 5081 return false; 5082 } 5083 5084 //////////////////////////////////////////////////////////////////////////////// 5085 // misc 5086 5087 // This does not do anything on Linux. This is basically a hook for being 5088 // able to use structured exception handling (thread-local exception filters) 5089 // on, e.g., Win32. 5090 void 5091 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method, 5092 JavaCallArguments* args, Thread* thread) { 5093 f(value, method, args, thread); 5094 } 5095 5096 void os::print_statistics() { 5097 } 5098 5099 bool os::message_box(const char* title, const char* message) { 5100 int i; 5101 fdStream err(defaultStream::error_fd()); 5102 for (i = 0; i < 78; i++) err.print_raw("="); 5103 err.cr(); 5104 err.print_raw_cr(title); 5105 for (i = 0; i < 78; i++) err.print_raw("-"); 5106 err.cr(); 5107 err.print_raw_cr(message); 5108 for (i = 0; i < 78; i++) err.print_raw("="); 5109 err.cr(); 5110 5111 char buf[16]; 5112 // Prevent process from exiting upon "read error" without consuming all CPU 5113 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5114 5115 return buf[0] == 'y' || buf[0] == 'Y'; 5116 } 5117 5118 int os::stat(const char *path, struct stat *sbuf) { 5119 char pathbuf[MAX_PATH]; 5120 if (strlen(path) > MAX_PATH - 1) { 5121 errno = ENAMETOOLONG; 5122 return -1; 5123 } 5124 os::native_path(strcpy(pathbuf, path)); 5125 return ::stat(pathbuf, sbuf); 5126 } 5127 5128 // Is a (classpath) directory empty? 5129 bool os::dir_is_empty(const char* path) { 5130 DIR *dir = NULL; 5131 struct dirent *ptr; 5132 5133 dir = opendir(path); 5134 if (dir == NULL) return true; 5135 5136 // Scan the directory 5137 bool result = true; 5138 char buf[sizeof(struct dirent) + MAX_PATH]; 5139 while (result && (ptr = ::readdir(dir)) != NULL) { 5140 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5141 result = false; 5142 } 5143 } 5144 closedir(dir); 5145 return result; 5146 } 5147 5148 // This code originates from JDK's sysOpen and open64_w 5149 // from src/solaris/hpi/src/system_md.c 5150 5151 int os::open(const char *path, int oflag, int mode) { 5152 if (strlen(path) > MAX_PATH - 1) { 5153 errno = ENAMETOOLONG; 5154 return -1; 5155 } 5156 5157 // All file descriptors that are opened in the Java process and not 5158 // specifically destined for a subprocess should have the close-on-exec 5159 // flag set. If we don't set it, then careless 3rd party native code 5160 // might fork and exec without closing all appropriate file descriptors 5161 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5162 // turn might: 5163 // 5164 // - cause end-of-file to fail to be detected on some file 5165 // descriptors, resulting in mysterious hangs, or 5166 // 5167 // - might cause an fopen in the subprocess to fail on a system 5168 // suffering from bug 1085341. 5169 // 5170 // (Yes, the default setting of the close-on-exec flag is a Unix 5171 // design flaw) 5172 // 5173 // See: 5174 // 1085341: 32-bit stdio routines should support file descriptors >255 5175 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5176 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5177 // 5178 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5179 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5180 // because it saves a system call and removes a small window where the flag 5181 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5182 // and we fall back to using FD_CLOEXEC (see below). 5183 #ifdef O_CLOEXEC 5184 oflag |= O_CLOEXEC; 5185 #endif 5186 5187 int fd = ::open64(path, oflag, mode); 5188 if (fd == -1) return -1; 5189 5190 //If the open succeeded, the file might still be a directory 5191 { 5192 struct stat64 buf64; 5193 int ret = ::fstat64(fd, &buf64); 5194 int st_mode = buf64.st_mode; 5195 5196 if (ret != -1) { 5197 if ((st_mode & S_IFMT) == S_IFDIR) { 5198 errno = EISDIR; 5199 ::close(fd); 5200 return -1; 5201 } 5202 } else { 5203 ::close(fd); 5204 return -1; 5205 } 5206 } 5207 5208 #ifdef FD_CLOEXEC 5209 // Validate that the use of the O_CLOEXEC flag on open above worked. 5210 // With recent kernels, we will perform this check exactly once. 5211 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5212 if (!O_CLOEXEC_is_known_to_work) { 5213 int flags = ::fcntl(fd, F_GETFD); 5214 if (flags != -1) { 5215 if ((flags & FD_CLOEXEC) != 0) 5216 O_CLOEXEC_is_known_to_work = 1; 5217 else 5218 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5219 } 5220 } 5221 #endif 5222 5223 return fd; 5224 } 5225 5226 5227 // create binary file, rewriting existing file if required 5228 int os::create_binary_file(const char* path, bool rewrite_existing) { 5229 int oflags = O_WRONLY | O_CREAT; 5230 if (!rewrite_existing) { 5231 oflags |= O_EXCL; 5232 } 5233 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5234 } 5235 5236 // return current position of file pointer 5237 jlong os::current_file_offset(int fd) { 5238 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5239 } 5240 5241 // move file pointer to the specified offset 5242 jlong os::seek_to_file_offset(int fd, jlong offset) { 5243 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5244 } 5245 5246 // This code originates from JDK's sysAvailable 5247 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5248 5249 int os::available(int fd, jlong *bytes) { 5250 jlong cur, end; 5251 int mode; 5252 struct stat64 buf64; 5253 5254 if (::fstat64(fd, &buf64) >= 0) { 5255 mode = buf64.st_mode; 5256 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5257 int n; 5258 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5259 *bytes = n; 5260 return 1; 5261 } 5262 } 5263 } 5264 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5265 return 0; 5266 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5267 return 0; 5268 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5269 return 0; 5270 } 5271 *bytes = end - cur; 5272 return 1; 5273 } 5274 5275 // Map a block of memory. 5276 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5277 char *addr, size_t bytes, bool read_only, 5278 bool allow_exec) { 5279 int prot; 5280 int flags = MAP_PRIVATE; 5281 5282 if (read_only) { 5283 prot = PROT_READ; 5284 } else { 5285 prot = PROT_READ | PROT_WRITE; 5286 } 5287 5288 if (allow_exec) { 5289 prot |= PROT_EXEC; 5290 } 5291 5292 if (addr != NULL) { 5293 flags |= MAP_FIXED; 5294 } 5295 5296 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5297 fd, file_offset); 5298 if (mapped_address == MAP_FAILED) { 5299 return NULL; 5300 } 5301 return mapped_address; 5302 } 5303 5304 5305 // Remap a block of memory. 5306 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5307 char *addr, size_t bytes, bool read_only, 5308 bool allow_exec) { 5309 // same as map_memory() on this OS 5310 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5311 allow_exec); 5312 } 5313 5314 5315 // Unmap a block of memory. 5316 bool os::pd_unmap_memory(char* addr, size_t bytes) { 5317 return munmap(addr, bytes) == 0; 5318 } 5319 5320 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5321 5322 static clockid_t thread_cpu_clockid(Thread* thread) { 5323 pthread_t tid = thread->osthread()->pthread_id(); 5324 clockid_t clockid; 5325 5326 // Get thread clockid 5327 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5328 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5329 return clockid; 5330 } 5331 5332 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5333 // are used by JVM M&M and JVMTI to get user+sys or user CPU time 5334 // of a thread. 5335 // 5336 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5337 // the fast estimate available on the platform. 5338 5339 jlong os::current_thread_cpu_time() { 5340 if (os::Linux::supports_fast_thread_cpu_time()) { 5341 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5342 } else { 5343 // return user + sys since the cost is the same 5344 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5345 } 5346 } 5347 5348 jlong os::thread_cpu_time(Thread* thread) { 5349 // consistent with what current_thread_cpu_time() returns 5350 if (os::Linux::supports_fast_thread_cpu_time()) { 5351 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5352 } else { 5353 return slow_thread_cpu_time(thread, true /* user + sys */); 5354 } 5355 } 5356 5357 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5358 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5359 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5360 } else { 5361 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5362 } 5363 } 5364 5365 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5366 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5367 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5368 } else { 5369 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5370 } 5371 } 5372 5373 // -1 on error. 5374 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5375 pid_t tid = thread->osthread()->thread_id(); 5376 char *s; 5377 char stat[2048]; 5378 int statlen; 5379 char proc_name[64]; 5380 int count; 5381 long sys_time, user_time; 5382 char cdummy; 5383 int idummy; 5384 long ldummy; 5385 FILE *fp; 5386 5387 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5388 fp = fopen(proc_name, "r"); 5389 if (fp == NULL) return -1; 5390 statlen = fread(stat, 1, 2047, fp); 5391 stat[statlen] = '\0'; 5392 fclose(fp); 5393 5394 // Skip pid and the command string. Note that we could be dealing with 5395 // weird command names, e.g. user could decide to rename java launcher 5396 // to "java 1.4.2 :)", then the stat file would look like 5397 // 1234 (java 1.4.2 :)) R ... ... 5398 // We don't really need to know the command string, just find the last 5399 // occurrence of ")" and then start parsing from there. See bug 4726580. 5400 s = strrchr(stat, ')'); 5401 if (s == NULL) return -1; 5402 5403 // Skip blank chars 5404 do { s++; } while (s && isspace(*s)); 5405 5406 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5407 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5408 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5409 &user_time, &sys_time); 5410 if (count != 13) return -1; 5411 if (user_sys_cpu_time) { 5412 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5413 } else { 5414 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5415 } 5416 } 5417 5418 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5419 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5420 info_ptr->may_skip_backward = false; // elapsed time not wall time 5421 info_ptr->may_skip_forward = false; // elapsed time not wall time 5422 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5423 } 5424 5425 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5426 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5427 info_ptr->may_skip_backward = false; // elapsed time not wall time 5428 info_ptr->may_skip_forward = false; // elapsed time not wall time 5429 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5430 } 5431 5432 bool os::is_thread_cpu_time_supported() { 5433 return true; 5434 } 5435 5436 // System loadavg support. Returns -1 if load average cannot be obtained. 5437 // Linux doesn't yet have a (official) notion of processor sets, 5438 // so just return the system wide load average. 5439 int os::loadavg(double loadavg[], int nelem) { 5440 return ::getloadavg(loadavg, nelem); 5441 } 5442 5443 void os::pause() { 5444 char filename[MAX_PATH]; 5445 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5446 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile); 5447 } else { 5448 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5449 } 5450 5451 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5452 if (fd != -1) { 5453 struct stat buf; 5454 ::close(fd); 5455 while (::stat(filename, &buf) == 0) { 5456 (void)::poll(NULL, 0, 100); 5457 } 5458 } else { 5459 jio_fprintf(stderr, 5460 "Could not open pause file '%s', continuing immediately.\n", filename); 5461 } 5462 } 5463 5464 5465 // Refer to the comments in os_solaris.cpp park-unpark. The next two 5466 // comment paragraphs are worth repeating here: 5467 // 5468 // Assumption: 5469 // Only one parker can exist on an event, which is why we allocate 5470 // them per-thread. Multiple unparkers can coexist. 5471 // 5472 // _Event serves as a restricted-range semaphore. 5473 // -1 : thread is blocked, i.e. there is a waiter 5474 // 0 : neutral: thread is running or ready, 5475 // could have been signaled after a wait started 5476 // 1 : signaled - thread is running or ready 5477 // 5478 5479 // utility to compute the abstime argument to timedwait: 5480 // millis is the relative timeout time 5481 // abstime will be the absolute timeout time 5482 // TODO: replace compute_abstime() with unpackTime() 5483 5484 static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5485 if (millis < 0) millis = 0; 5486 5487 jlong seconds = millis / 1000; 5488 millis %= 1000; 5489 if (seconds > 50000000) { // see man cond_timedwait(3T) 5490 seconds = 50000000; 5491 } 5492 5493 if (os::supports_monotonic_clock()) { 5494 struct timespec now; 5495 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5496 assert_status(status == 0, status, "clock_gettime"); 5497 abstime->tv_sec = now.tv_sec + seconds; 5498 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5499 if (nanos >= NANOSECS_PER_SEC) { 5500 abstime->tv_sec += 1; 5501 nanos -= NANOSECS_PER_SEC; 5502 } 5503 abstime->tv_nsec = nanos; 5504 } else { 5505 struct timeval now; 5506 int status = gettimeofday(&now, NULL); 5507 assert(status == 0, "gettimeofday"); 5508 abstime->tv_sec = now.tv_sec + seconds; 5509 long usec = now.tv_usec + millis * 1000; 5510 if (usec >= 1000000) { 5511 abstime->tv_sec += 1; 5512 usec -= 1000000; 5513 } 5514 abstime->tv_nsec = usec * 1000; 5515 } 5516 return abstime; 5517 } 5518 5519 void os::PlatformEvent::park() { // AKA "down()" 5520 // Transitions for _Event: 5521 // -1 => -1 : illegal 5522 // 1 => 0 : pass - return immediately 5523 // 0 => -1 : block; then set _Event to 0 before returning 5524 5525 // Invariant: Only the thread associated with the Event/PlatformEvent 5526 // may call park(). 5527 // TODO: assert that _Assoc != NULL or _Assoc == Self 5528 assert(_nParked == 0, "invariant"); 5529 5530 int v; 5531 for (;;) { 5532 v = _Event; 5533 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5534 } 5535 guarantee(v >= 0, "invariant"); 5536 if (v == 0) { 5537 // Do this the hard way by blocking ... 5538 int status = pthread_mutex_lock(_mutex); 5539 assert_status(status == 0, status, "mutex_lock"); 5540 guarantee(_nParked == 0, "invariant"); 5541 ++_nParked; 5542 while (_Event < 0) { 5543 status = pthread_cond_wait(_cond, _mutex); 5544 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5545 // Treat this the same as if the wait was interrupted 5546 if (status == ETIME) { status = EINTR; } 5547 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5548 } 5549 --_nParked; 5550 5551 _Event = 0; 5552 status = pthread_mutex_unlock(_mutex); 5553 assert_status(status == 0, status, "mutex_unlock"); 5554 // Paranoia to ensure our locked and lock-free paths interact 5555 // correctly with each other. 5556 OrderAccess::fence(); 5557 } 5558 guarantee(_Event >= 0, "invariant"); 5559 } 5560 5561 int os::PlatformEvent::park(jlong millis) { 5562 // Transitions for _Event: 5563 // -1 => -1 : illegal 5564 // 1 => 0 : pass - return immediately 5565 // 0 => -1 : block; then set _Event to 0 before returning 5566 5567 guarantee(_nParked == 0, "invariant"); 5568 5569 int v; 5570 for (;;) { 5571 v = _Event; 5572 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5573 } 5574 guarantee(v >= 0, "invariant"); 5575 if (v != 0) return OS_OK; 5576 5577 // We do this the hard way, by blocking the thread. 5578 // Consider enforcing a minimum timeout value. 5579 struct timespec abst; 5580 compute_abstime(&abst, millis); 5581 5582 int ret = OS_TIMEOUT; 5583 int status = pthread_mutex_lock(_mutex); 5584 assert_status(status == 0, status, "mutex_lock"); 5585 guarantee(_nParked == 0, "invariant"); 5586 ++_nParked; 5587 5588 // Object.wait(timo) will return because of 5589 // (a) notification 5590 // (b) timeout 5591 // (c) thread.interrupt 5592 // 5593 // Thread.interrupt and object.notify{All} both call Event::set. 5594 // That is, we treat thread.interrupt as a special case of notification. 5595 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false. 5596 // We assume all ETIME returns are valid. 5597 // 5598 // TODO: properly differentiate simultaneous notify+interrupt. 5599 // In that case, we should propagate the notify to another waiter. 5600 5601 while (_Event < 0) { 5602 status = pthread_cond_timedwait(_cond, _mutex, &abst); 5603 assert_status(status == 0 || status == EINTR || 5604 status == ETIME || status == ETIMEDOUT, 5605 status, "cond_timedwait"); 5606 if (!FilterSpuriousWakeups) break; // previous semantics 5607 if (status == ETIME || status == ETIMEDOUT) break; 5608 // We consume and ignore EINTR and spurious wakeups. 5609 } 5610 --_nParked; 5611 if (_Event >= 0) { 5612 ret = OS_OK; 5613 } 5614 _Event = 0; 5615 status = pthread_mutex_unlock(_mutex); 5616 assert_status(status == 0, status, "mutex_unlock"); 5617 assert(_nParked == 0, "invariant"); 5618 // Paranoia to ensure our locked and lock-free paths interact 5619 // correctly with each other. 5620 OrderAccess::fence(); 5621 return ret; 5622 } 5623 5624 void os::PlatformEvent::unpark() { 5625 // Transitions for _Event: 5626 // 0 => 1 : just return 5627 // 1 => 1 : just return 5628 // -1 => either 0 or 1; must signal target thread 5629 // That is, we can safely transition _Event from -1 to either 5630 // 0 or 1. 5631 // See also: "Semaphores in Plan 9" by Mullender & Cox 5632 // 5633 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5634 // that it will take two back-to-back park() calls for the owning 5635 // thread to block. This has the benefit of forcing a spurious return 5636 // from the first park() call after an unpark() call which will help 5637 // shake out uses of park() and unpark() without condition variables. 5638 5639 if (Atomic::xchg(1, &_Event) >= 0) return; 5640 5641 // Wait for the thread associated with the event to vacate 5642 int status = pthread_mutex_lock(_mutex); 5643 assert_status(status == 0, status, "mutex_lock"); 5644 int AnyWaiters = _nParked; 5645 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5646 status = pthread_mutex_unlock(_mutex); 5647 assert_status(status == 0, status, "mutex_unlock"); 5648 if (AnyWaiters != 0) { 5649 // Note that we signal() *after* dropping the lock for "immortal" Events. 5650 // This is safe and avoids a common class of futile wakeups. In rare 5651 // circumstances this can cause a thread to return prematurely from 5652 // cond_{timed}wait() but the spurious wakeup is benign and the victim 5653 // will simply re-test the condition and re-park itself. 5654 // This provides particular benefit if the underlying platform does not 5655 // provide wait morphing. 5656 status = pthread_cond_signal(_cond); 5657 assert_status(status == 0, status, "cond_signal"); 5658 } 5659 } 5660 5661 5662 // JSR166 5663 // ------------------------------------------------------- 5664 5665 // The solaris and linux implementations of park/unpark are fairly 5666 // conservative for now, but can be improved. They currently use a 5667 // mutex/condvar pair, plus a a count. 5668 // Park decrements count if > 0, else does a condvar wait. Unpark 5669 // sets count to 1 and signals condvar. Only one thread ever waits 5670 // on the condvar. Contention seen when trying to park implies that someone 5671 // is unparking you, so don't wait. And spurious returns are fine, so there 5672 // is no need to track notifications. 5673 5674 // This code is common to linux and solaris and will be moved to a 5675 // common place in dolphin. 5676 // 5677 // The passed in time value is either a relative time in nanoseconds 5678 // or an absolute time in milliseconds. Either way it has to be unpacked 5679 // into suitable seconds and nanoseconds components and stored in the 5680 // given timespec structure. 5681 // Given time is a 64-bit value and the time_t used in the timespec is only 5682 // a signed-32-bit value (except on 64-bit Linux) we have to watch for 5683 // overflow if times way in the future are given. Further on Solaris versions 5684 // prior to 10 there is a restriction (see cond_timedwait) that the specified 5685 // number of seconds, in abstime, is less than current_time + 100,000,000. 5686 // As it will be 28 years before "now + 100000000" will overflow we can 5687 // ignore overflow and just impose a hard-limit on seconds using the value 5688 // of "now + 100,000,000". This places a limit on the timeout of about 3.17 5689 // years from "now". 5690 5691 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5692 assert(time > 0, "convertTime"); 5693 time_t max_secs = 0; 5694 5695 if (!os::supports_monotonic_clock() || isAbsolute) { 5696 struct timeval now; 5697 int status = gettimeofday(&now, NULL); 5698 assert(status == 0, "gettimeofday"); 5699 5700 max_secs = now.tv_sec + MAX_SECS; 5701 5702 if (isAbsolute) { 5703 jlong secs = time / 1000; 5704 if (secs > max_secs) { 5705 absTime->tv_sec = max_secs; 5706 } else { 5707 absTime->tv_sec = secs; 5708 } 5709 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5710 } else { 5711 jlong secs = time / NANOSECS_PER_SEC; 5712 if (secs >= MAX_SECS) { 5713 absTime->tv_sec = max_secs; 5714 absTime->tv_nsec = 0; 5715 } else { 5716 absTime->tv_sec = now.tv_sec + secs; 5717 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5718 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5719 absTime->tv_nsec -= NANOSECS_PER_SEC; 5720 ++absTime->tv_sec; // note: this must be <= max_secs 5721 } 5722 } 5723 } 5724 } else { 5725 // must be relative using monotonic clock 5726 struct timespec now; 5727 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5728 assert_status(status == 0, status, "clock_gettime"); 5729 max_secs = now.tv_sec + MAX_SECS; 5730 jlong secs = time / NANOSECS_PER_SEC; 5731 if (secs >= MAX_SECS) { 5732 absTime->tv_sec = max_secs; 5733 absTime->tv_nsec = 0; 5734 } else { 5735 absTime->tv_sec = now.tv_sec + secs; 5736 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5737 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5738 absTime->tv_nsec -= NANOSECS_PER_SEC; 5739 ++absTime->tv_sec; // note: this must be <= max_secs 5740 } 5741 } 5742 } 5743 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5744 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5745 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5746 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5747 } 5748 5749 void Parker::park(bool isAbsolute, jlong time) { 5750 // Ideally we'd do something useful while spinning, such 5751 // as calling unpackTime(). 5752 5753 // Optional fast-path check: 5754 // Return immediately if a permit is available. 5755 // We depend on Atomic::xchg() having full barrier semantics 5756 // since we are doing a lock-free update to _counter. 5757 if (Atomic::xchg(0, &_counter) > 0) return; 5758 5759 Thread* thread = Thread::current(); 5760 assert(thread->is_Java_thread(), "Must be JavaThread"); 5761 JavaThread *jt = (JavaThread *)thread; 5762 5763 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5764 // Check interrupt before trying to wait 5765 if (Thread::is_interrupted(thread, false)) { 5766 return; 5767 } 5768 5769 // Next, demultiplex/decode time arguments 5770 timespec absTime; 5771 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all 5772 return; 5773 } 5774 if (time > 0) { 5775 unpackTime(&absTime, isAbsolute, time); 5776 } 5777 5778 5779 // Enter safepoint region 5780 // Beware of deadlocks such as 6317397. 5781 // The per-thread Parker:: mutex is a classic leaf-lock. 5782 // In particular a thread must never block on the Threads_lock while 5783 // holding the Parker:: mutex. If safepoints are pending both the 5784 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5785 ThreadBlockInVM tbivm(jt); 5786 5787 // Don't wait if cannot get lock since interference arises from 5788 // unblocking. Also. check interrupt before trying wait 5789 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5790 return; 5791 } 5792 5793 int status; 5794 if (_counter > 0) { // no wait needed 5795 _counter = 0; 5796 status = pthread_mutex_unlock(_mutex); 5797 assert_status(status == 0, status, "invariant"); 5798 // Paranoia to ensure our locked and lock-free paths interact 5799 // correctly with each other and Java-level accesses. 5800 OrderAccess::fence(); 5801 return; 5802 } 5803 5804 #ifdef ASSERT 5805 // Don't catch signals while blocked; let the running threads have the signals. 5806 // (This allows a debugger to break into the running thread.) 5807 sigset_t oldsigs; 5808 sigemptyset(&oldsigs); 5809 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5810 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5811 #endif 5812 5813 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5814 jt->set_suspend_equivalent(); 5815 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5816 5817 assert(_cur_index == -1, "invariant"); 5818 if (time == 0) { 5819 _cur_index = REL_INDEX; // arbitrary choice when not timed 5820 status = pthread_cond_wait(&_cond[_cur_index], _mutex); 5821 } else { 5822 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5823 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime); 5824 } 5825 _cur_index = -1; 5826 assert_status(status == 0 || status == EINTR || 5827 status == ETIME || status == ETIMEDOUT, 5828 status, "cond_timedwait"); 5829 5830 #ifdef ASSERT 5831 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5832 #endif 5833 5834 _counter = 0; 5835 status = pthread_mutex_unlock(_mutex); 5836 assert_status(status == 0, status, "invariant"); 5837 // Paranoia to ensure our locked and lock-free paths interact 5838 // correctly with each other and Java-level accesses. 5839 OrderAccess::fence(); 5840 5841 // If externally suspended while waiting, re-suspend 5842 if (jt->handle_special_suspend_equivalent_condition()) { 5843 jt->java_suspend_self(); 5844 } 5845 } 5846 5847 void Parker::unpark() { 5848 int status = pthread_mutex_lock(_mutex); 5849 assert_status(status == 0, status, "invariant"); 5850 const int s = _counter; 5851 _counter = 1; 5852 // must capture correct index before unlocking 5853 int index = _cur_index; 5854 status = pthread_mutex_unlock(_mutex); 5855 assert_status(status == 0, status, "invariant"); 5856 if (s < 1 && index != -1) { 5857 // thread is definitely parked 5858 status = pthread_cond_signal(&_cond[index]); 5859 assert_status(status == 0, status, "invariant"); 5860 } 5861 } 5862 5863 5864 extern char** environ; 5865 5866 // Run the specified command in a separate process. Return its exit value, 5867 // or -1 on failure (e.g. can't fork a new process). 5868 // Unlike system(), this function can be called from signal handler. It 5869 // doesn't block SIGINT et al. 5870 int os::fork_and_exec(char* cmd) { 5871 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5872 5873 pid_t pid = fork(); 5874 5875 if (pid < 0) { 5876 // fork failed 5877 return -1; 5878 5879 } else if (pid == 0) { 5880 // child process 5881 5882 execve("/bin/sh", (char* const*)argv, environ); 5883 5884 // execve failed 5885 _exit(-1); 5886 5887 } else { 5888 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5889 // care about the actual exit code, for now. 5890 5891 int status; 5892 5893 // Wait for the child process to exit. This returns immediately if 5894 // the child has already exited. */ 5895 while (waitpid(pid, &status, 0) < 0) { 5896 switch (errno) { 5897 case ECHILD: return 0; 5898 case EINTR: break; 5899 default: return -1; 5900 } 5901 } 5902 5903 if (WIFEXITED(status)) { 5904 // The child exited normally; get its exit code. 5905 return WEXITSTATUS(status); 5906 } else if (WIFSIGNALED(status)) { 5907 // The child exited because of a signal 5908 // The best value to return is 0x80 + signal number, 5909 // because that is what all Unix shells do, and because 5910 // it allows callers to distinguish between process exit and 5911 // process death by signal. 5912 return 0x80 + WTERMSIG(status); 5913 } else { 5914 // Unknown exit code; pass it through 5915 return status; 5916 } 5917 } 5918 } 5919 5920 // is_headless_jre() 5921 // 5922 // Test for the existence of xawt/libmawt.so or libawt_xawt.so 5923 // in order to report if we are running in a headless jre 5924 // 5925 // Since JDK8 xawt/libmawt.so was moved into the same directory 5926 // as libawt.so, and renamed libawt_xawt.so 5927 // 5928 bool os::is_headless_jre() { 5929 struct stat statbuf; 5930 char buf[MAXPATHLEN]; 5931 char libmawtpath[MAXPATHLEN]; 5932 const char *xawtstr = "/xawt/libmawt.so"; 5933 const char *new_xawtstr = "/libawt_xawt.so"; 5934 char *p; 5935 5936 // Get path to libjvm.so 5937 os::jvm_path(buf, sizeof(buf)); 5938 5939 // Get rid of libjvm.so 5940 p = strrchr(buf, '/'); 5941 if (p == NULL) { 5942 return false; 5943 } else { 5944 *p = '\0'; 5945 } 5946 5947 // Get rid of client or server 5948 p = strrchr(buf, '/'); 5949 if (p == NULL) { 5950 return false; 5951 } else { 5952 *p = '\0'; 5953 } 5954 5955 // check xawt/libmawt.so 5956 strcpy(libmawtpath, buf); 5957 strcat(libmawtpath, xawtstr); 5958 if (::stat(libmawtpath, &statbuf) == 0) return false; 5959 5960 // check libawt_xawt.so 5961 strcpy(libmawtpath, buf); 5962 strcat(libmawtpath, new_xawtstr); 5963 if (::stat(libmawtpath, &statbuf) == 0) return false; 5964 5965 return true; 5966 } 5967 5968 // Get the default path to the core file 5969 // Returns the length of the string 5970 int os::get_core_path(char* buffer, size_t bufferSize) { 5971 /* 5972 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 5973 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 5974 */ 5975 const int core_pattern_len = 129; 5976 char core_pattern[core_pattern_len] = {0}; 5977 5978 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 5979 if (core_pattern_file == -1) { 5980 return -1; 5981 } 5982 5983 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 5984 ::close(core_pattern_file); 5985 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') { 5986 return -1; 5987 } 5988 if (core_pattern[ret-1] == '\n') { 5989 core_pattern[ret-1] = '\0'; 5990 } else { 5991 core_pattern[ret] = '\0'; 5992 } 5993 5994 char *pid_pos = strstr(core_pattern, "%p"); 5995 int written; 5996 5997 if (core_pattern[0] == '/') { 5998 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern); 5999 } else { 6000 char cwd[PATH_MAX]; 6001 6002 const char* p = get_current_directory(cwd, PATH_MAX); 6003 if (p == NULL) { 6004 return -1; 6005 } 6006 6007 if (core_pattern[0] == '|') { 6008 written = jio_snprintf(buffer, bufferSize, 6009 "\"%s\" (or dumping to %s/core.%d)", 6010 &core_pattern[1], p, current_process_id()); 6011 } else { 6012 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 6013 } 6014 } 6015 6016 if (written < 0) { 6017 return -1; 6018 } 6019 6020 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 6021 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 6022 6023 if (core_uses_pid_file != -1) { 6024 char core_uses_pid = 0; 6025 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 6026 ::close(core_uses_pid_file); 6027 6028 if (core_uses_pid == '1') { 6029 jio_snprintf(buffer + written, bufferSize - written, 6030 ".%d", current_process_id()); 6031 } 6032 } 6033 } 6034 6035 return strlen(buffer); 6036 } 6037 6038 bool os::start_debugging(char *buf, int buflen) { 6039 int len = (int)strlen(buf); 6040 char *p = &buf[len]; 6041 6042 jio_snprintf(p, buflen-len, 6043 "\n\n" 6044 "Do you want to debug the problem?\n\n" 6045 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n" 6046 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n" 6047 "Otherwise, press RETURN to abort...", 6048 os::current_process_id(), os::current_process_id(), 6049 os::current_thread_id(), os::current_thread_id()); 6050 6051 bool yes = os::message_box("Unexpected Error", buf); 6052 6053 if (yes) { 6054 // yes, user asked VM to launch debugger 6055 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d", 6056 os::current_process_id(), os::current_process_id()); 6057 6058 os::fork_and_exec(buf); 6059 yes = false; 6060 } 6061 return yes; 6062 } 6063 6064 static inline struct timespec get_mtime(const char* filename) { 6065 struct stat st; 6066 int ret = os::stat(filename, &st); 6067 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno)); 6068 return st.st_mtim; 6069 } 6070 6071 int os::compare_file_modified_times(const char* file1, const char* file2) { 6072 struct timespec filetime1 = get_mtime(file1); 6073 struct timespec filetime2 = get_mtime(file2); 6074 int diff = filetime1.tv_sec - filetime2.tv_sec; 6075 if (diff == 0) { 6076 return filetime1.tv_nsec - filetime2.tv_nsec; 6077 } 6078 return diff; 6079 } 6080 6081 /////////////// Unit tests /////////////// 6082 6083 #ifndef PRODUCT 6084 6085 #define test_log(...) \ 6086 do { \ 6087 if (VerboseInternalVMTests) { \ 6088 tty->print_cr(__VA_ARGS__); \ 6089 tty->flush(); \ 6090 } \ 6091 } while (false) 6092 6093 class TestReserveMemorySpecial : AllStatic { 6094 public: 6095 static void small_page_write(void* addr, size_t size) { 6096 size_t page_size = os::vm_page_size(); 6097 6098 char* end = (char*)addr + size; 6099 for (char* p = (char*)addr; p < end; p += page_size) { 6100 *p = 1; 6101 } 6102 } 6103 6104 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6105 if (!UseHugeTLBFS) { 6106 return; 6107 } 6108 6109 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6110 6111 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6112 6113 if (addr != NULL) { 6114 small_page_write(addr, size); 6115 6116 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6117 } 6118 } 6119 6120 static void test_reserve_memory_special_huge_tlbfs_only() { 6121 if (!UseHugeTLBFS) { 6122 return; 6123 } 6124 6125 size_t lp = os::large_page_size(); 6126 6127 for (size_t size = lp; size <= lp * 10; size += lp) { 6128 test_reserve_memory_special_huge_tlbfs_only(size); 6129 } 6130 } 6131 6132 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6133 size_t lp = os::large_page_size(); 6134 size_t ag = os::vm_allocation_granularity(); 6135 6136 // sizes to test 6137 const size_t sizes[] = { 6138 lp, lp + ag, lp + lp / 2, lp * 2, 6139 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2, 6140 lp * 10, lp * 10 + lp / 2 6141 }; 6142 const int num_sizes = sizeof(sizes) / sizeof(size_t); 6143 6144 // For each size/alignment combination, we test three scenarios: 6145 // 1) with req_addr == NULL 6146 // 2) with a non-null req_addr at which we expect to successfully allocate 6147 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we 6148 // expect the allocation to either fail or to ignore req_addr 6149 6150 // Pre-allocate two areas; they shall be as large as the largest allocation 6151 // and aligned to the largest alignment we will be testing. 6152 const size_t mapping_size = sizes[num_sizes - 1] * 2; 6153 char* const mapping1 = (char*) ::mmap(NULL, mapping_size, 6154 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6155 -1, 0); 6156 assert(mapping1 != MAP_FAILED, "should work"); 6157 6158 char* const mapping2 = (char*) ::mmap(NULL, mapping_size, 6159 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6160 -1, 0); 6161 assert(mapping2 != MAP_FAILED, "should work"); 6162 6163 // Unmap the first mapping, but leave the second mapping intact: the first 6164 // mapping will serve as a value for a "good" req_addr (case 2). The second 6165 // mapping, still intact, as "bad" req_addr (case 3). 6166 ::munmap(mapping1, mapping_size); 6167 6168 // Case 1 6169 test_log("%s, req_addr NULL:", __FUNCTION__); 6170 test_log("size align result"); 6171 6172 for (int i = 0; i < num_sizes; i++) { 6173 const size_t size = sizes[i]; 6174 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6175 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6176 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s", 6177 size, alignment, p2i(p), (p != NULL ? "" : "(failed)")); 6178 if (p != NULL) { 6179 assert(is_ptr_aligned(p, alignment), "must be"); 6180 small_page_write(p, size); 6181 os::Linux::release_memory_special_huge_tlbfs(p, size); 6182 } 6183 } 6184 } 6185 6186 // Case 2 6187 test_log("%s, req_addr non-NULL:", __FUNCTION__); 6188 test_log("size align req_addr result"); 6189 6190 for (int i = 0; i < num_sizes; i++) { 6191 const size_t size = sizes[i]; 6192 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6193 char* const req_addr = (char*) align_ptr_up(mapping1, alignment); 6194 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6195 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6196 size, alignment, p2i(req_addr), p2i(p), 6197 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)"))); 6198 if (p != NULL) { 6199 assert(p == req_addr, "must be"); 6200 small_page_write(p, size); 6201 os::Linux::release_memory_special_huge_tlbfs(p, size); 6202 } 6203 } 6204 } 6205 6206 // Case 3 6207 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__); 6208 test_log("size align req_addr result"); 6209 6210 for (int i = 0; i < num_sizes; i++) { 6211 const size_t size = sizes[i]; 6212 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6213 char* const req_addr = (char*) align_ptr_up(mapping2, alignment); 6214 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6215 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6216 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)"))); 6217 // as the area around req_addr contains already existing mappings, the API should always 6218 // return NULL (as per contract, it cannot return another address) 6219 assert(p == NULL, "must be"); 6220 } 6221 } 6222 6223 ::munmap(mapping2, mapping_size); 6224 6225 } 6226 6227 static void test_reserve_memory_special_huge_tlbfs() { 6228 if (!UseHugeTLBFS) { 6229 return; 6230 } 6231 6232 test_reserve_memory_special_huge_tlbfs_only(); 6233 test_reserve_memory_special_huge_tlbfs_mixed(); 6234 } 6235 6236 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6237 if (!UseSHM) { 6238 return; 6239 } 6240 6241 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6242 6243 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6244 6245 if (addr != NULL) { 6246 assert(is_ptr_aligned(addr, alignment), "Check"); 6247 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6248 6249 small_page_write(addr, size); 6250 6251 os::Linux::release_memory_special_shm(addr, size); 6252 } 6253 } 6254 6255 static void test_reserve_memory_special_shm() { 6256 size_t lp = os::large_page_size(); 6257 size_t ag = os::vm_allocation_granularity(); 6258 6259 for (size_t size = ag; size < lp * 3; size += ag) { 6260 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6261 test_reserve_memory_special_shm(size, alignment); 6262 } 6263 } 6264 } 6265 6266 static void test() { 6267 test_reserve_memory_special_huge_tlbfs(); 6268 test_reserve_memory_special_shm(); 6269 } 6270 }; 6271 6272 void TestReserveMemorySpecial_test() { 6273 TestReserveMemorySpecial::test(); 6274 } 6275 6276 #endif