1 /* 2 * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 // no precompiled headers 26 #include "jvm.h" 27 #include "classfile/classLoader.hpp" 28 #include "classfile/systemDictionary.hpp" 29 #include "classfile/vmSymbols.hpp" 30 #include "code/icBuffer.hpp" 31 #include "code/vtableStubs.hpp" 32 #include "compiler/compileBroker.hpp" 33 #include "compiler/disassembler.hpp" 34 #include "interpreter/interpreter.hpp" 35 #include "logging/log.hpp" 36 #include "memory/allocation.inline.hpp" 37 #include "memory/filemap.hpp" 38 #include "oops/oop.inline.hpp" 39 #include "os_linux.inline.hpp" 40 #include "os_share_linux.hpp" 41 #include "osContainer_linux.hpp" 42 #include "prims/jniFastGetField.hpp" 43 #include "prims/jvm_misc.hpp" 44 #include "runtime/arguments.hpp" 45 #include "runtime/atomic.hpp" 46 #include "runtime/extendedPC.hpp" 47 #include "runtime/globals.hpp" 48 #include "runtime/interfaceSupport.inline.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.hpp" 55 #include "runtime/osThread.hpp" 56 #include "runtime/perfMemory.hpp" 57 #include "runtime/sharedRuntime.hpp" 58 #include "runtime/statSampler.hpp" 59 #include "runtime/stubRoutines.hpp" 60 #include "runtime/thread.inline.hpp" 61 #include "runtime/threadCritical.hpp" 62 #include "runtime/threadSMR.hpp" 63 #include "runtime/timer.hpp" 64 #include "semaphore_posix.hpp" 65 #include "services/attachListener.hpp" 66 #include "services/memTracker.hpp" 67 #include "services/runtimeService.hpp" 68 #include "utilities/align.hpp" 69 #include "utilities/decoder.hpp" 70 #include "utilities/defaultStream.hpp" 71 #include "utilities/events.hpp" 72 #include "utilities/elfFile.hpp" 73 #include "utilities/growableArray.hpp" 74 #include "utilities/macros.hpp" 75 #include "utilities/vmError.hpp" 76 77 // put OS-includes here 78 # include <sys/types.h> 79 # include <sys/mman.h> 80 # include <sys/stat.h> 81 # include <sys/select.h> 82 # include <pthread.h> 83 # include <signal.h> 84 # include <errno.h> 85 # include <dlfcn.h> 86 # include <stdio.h> 87 # include <unistd.h> 88 # include <sys/resource.h> 89 # include <pthread.h> 90 # include <sys/stat.h> 91 # include <sys/time.h> 92 # include <sys/times.h> 93 # include <sys/utsname.h> 94 # include <sys/socket.h> 95 # include <sys/wait.h> 96 # include <pwd.h> 97 # include <poll.h> 98 # include <fcntl.h> 99 # include <string.h> 100 # include <syscall.h> 101 # include <sys/sysinfo.h> 102 # include <gnu/libc-version.h> 103 # include <sys/ipc.h> 104 # include <sys/shm.h> 105 # include <link.h> 106 # include <stdint.h> 107 # include <inttypes.h> 108 # include <sys/ioctl.h> 109 110 #ifndef _GNU_SOURCE 111 #define _GNU_SOURCE 112 #include <sched.h> 113 #undef _GNU_SOURCE 114 #else 115 #include <sched.h> 116 #endif 117 118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 119 // getrusage() is prepared to handle the associated failure. 120 #ifndef RUSAGE_THREAD 121 #define RUSAGE_THREAD (1) /* only the calling thread */ 122 #endif 123 124 #define MAX_PATH (2 * K) 125 126 #define MAX_SECS 100000000 127 128 // for timer info max values which include all bits 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 130 131 #define LARGEPAGES_BIT (1 << 6) 132 #define DAX_SHARED_BIT (1 << 8) 133 //////////////////////////////////////////////////////////////////////////////// 134 // global variables 135 julong os::Linux::_physical_memory = 0; 136 137 address os::Linux::_initial_thread_stack_bottom = NULL; 138 uintptr_t os::Linux::_initial_thread_stack_size = 0; 139 140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 142 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL; 143 Mutex* os::Linux::_createThread_lock = NULL; 144 pthread_t os::Linux::_main_thread; 145 int os::Linux::_page_size = -1; 146 bool os::Linux::_supports_fast_thread_cpu_time = false; 147 uint32_t os::Linux::_os_version = 0; 148 const char * os::Linux::_glibc_version = NULL; 149 const char * os::Linux::_libpthread_version = NULL; 150 151 static jlong initial_time_count=0; 152 153 static int clock_tics_per_sec = 100; 154 155 // If the VM might have been created on the primordial thread, we need to resolve the 156 // primordial thread stack bounds and check if the current thread might be the 157 // primordial thread in places. If we know that the primordial thread is never used, 158 // such as when the VM was created by one of the standard java launchers, we can 159 // avoid this 160 static bool suppress_primordial_thread_resolution = false; 161 162 // For diagnostics to print a message once. see run_periodic_checks 163 static sigset_t check_signal_done; 164 static bool check_signals = true; 165 166 // Signal number used to suspend/resume a thread 167 168 // do not use any signal number less than SIGSEGV, see 4355769 169 static int SR_signum = SIGUSR2; 170 sigset_t SR_sigset; 171 172 // utility functions 173 174 static int SR_initialize(); 175 176 julong os::available_memory() { 177 return Linux::available_memory(); 178 } 179 180 julong os::Linux::available_memory() { 181 // values in struct sysinfo are "unsigned long" 182 struct sysinfo si; 183 julong avail_mem; 184 185 if (OSContainer::is_containerized()) { 186 jlong mem_limit, mem_usage; 187 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) { 188 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value", 189 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit); 190 } 191 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) { 192 log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage); 193 } 194 if (mem_limit > 0 && mem_usage > 0 ) { 195 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0; 196 log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem); 197 return avail_mem; 198 } 199 } 200 201 sysinfo(&si); 202 avail_mem = (julong)si.freeram * si.mem_unit; 203 log_trace(os)("available memory: " JULONG_FORMAT, avail_mem); 204 return avail_mem; 205 } 206 207 julong os::physical_memory() { 208 jlong phys_mem = 0; 209 if (OSContainer::is_containerized()) { 210 jlong mem_limit; 211 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) { 212 log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit); 213 return mem_limit; 214 } 215 log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value", 216 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit); 217 } 218 219 phys_mem = Linux::physical_memory(); 220 log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem); 221 return phys_mem; 222 } 223 224 // Return true if user is running as root. 225 226 bool os::have_special_privileges() { 227 static bool init = false; 228 static bool privileges = false; 229 if (!init) { 230 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 231 init = true; 232 } 233 return privileges; 234 } 235 236 237 #ifndef SYS_gettid 238 // i386: 224, ia64: 1105, amd64: 186, sparc 143 239 #ifdef __ia64__ 240 #define SYS_gettid 1105 241 #else 242 #ifdef __i386__ 243 #define SYS_gettid 224 244 #else 245 #ifdef __amd64__ 246 #define SYS_gettid 186 247 #else 248 #ifdef __sparc__ 249 #define SYS_gettid 143 250 #else 251 #error define gettid for the arch 252 #endif 253 #endif 254 #endif 255 #endif 256 #endif 257 258 259 // pid_t gettid() 260 // 261 // Returns the kernel thread id of the currently running thread. Kernel 262 // thread id is used to access /proc. 263 pid_t os::Linux::gettid() { 264 int rslt = syscall(SYS_gettid); 265 assert(rslt != -1, "must be."); // old linuxthreads implementation? 266 return (pid_t)rslt; 267 } 268 269 // Most versions of linux have a bug where the number of processors are 270 // determined by looking at the /proc file system. In a chroot environment, 271 // the system call returns 1. This causes the VM to act as if it is 272 // a single processor and elide locking (see is_MP() call). 273 static bool unsafe_chroot_detected = false; 274 static const char *unstable_chroot_error = "/proc file system not found.\n" 275 "Java may be unstable running multithreaded in a chroot " 276 "environment on Linux when /proc filesystem is not mounted."; 277 278 void os::Linux::initialize_system_info() { 279 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 280 if (processor_count() == 1) { 281 pid_t pid = os::Linux::gettid(); 282 char fname[32]; 283 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 284 FILE *fp = fopen(fname, "r"); 285 if (fp == NULL) { 286 unsafe_chroot_detected = true; 287 } else { 288 fclose(fp); 289 } 290 } 291 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 292 assert(processor_count() > 0, "linux error"); 293 } 294 295 void os::init_system_properties_values() { 296 // The next steps are taken in the product version: 297 // 298 // Obtain the JAVA_HOME value from the location of libjvm.so. 299 // This library should be located at: 300 // <JAVA_HOME>/lib/{client|server}/libjvm.so. 301 // 302 // If "/jre/lib/" appears at the right place in the path, then we 303 // assume libjvm.so is installed in a JDK and we use this path. 304 // 305 // Otherwise exit with message: "Could not create the Java virtual machine." 306 // 307 // The following extra steps are taken in the debugging version: 308 // 309 // If "/jre/lib/" does NOT appear at the right place in the path 310 // instead of exit check for $JAVA_HOME environment variable. 311 // 312 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 313 // then we append a fake suffix "hotspot/libjvm.so" to this path so 314 // it looks like libjvm.so is installed there 315 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 316 // 317 // Otherwise exit. 318 // 319 // Important note: if the location of libjvm.so changes this 320 // code needs to be changed accordingly. 321 322 // See ld(1): 323 // The linker uses the following search paths to locate required 324 // shared libraries: 325 // 1: ... 326 // ... 327 // 7: The default directories, normally /lib and /usr/lib. 328 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390) 329 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 330 #else 331 #define DEFAULT_LIBPATH "/lib:/usr/lib" 332 #endif 333 334 // Base path of extensions installed on the system. 335 #define SYS_EXT_DIR "/usr/java/packages" 336 #define EXTENSIONS_DIR "/lib/ext" 337 338 // Buffer that fits several sprintfs. 339 // Note that the space for the colon and the trailing null are provided 340 // by the nulls included by the sizeof operator. 341 const size_t bufsize = 342 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 343 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 344 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 345 346 // sysclasspath, java_home, dll_dir 347 { 348 char *pslash; 349 os::jvm_path(buf, bufsize); 350 351 // Found the full path to libjvm.so. 352 // Now cut the path to <java_home>/jre if we can. 353 pslash = strrchr(buf, '/'); 354 if (pslash != NULL) { 355 *pslash = '\0'; // Get rid of /libjvm.so. 356 } 357 pslash = strrchr(buf, '/'); 358 if (pslash != NULL) { 359 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 360 } 361 Arguments::set_dll_dir(buf); 362 363 if (pslash != NULL) { 364 pslash = strrchr(buf, '/'); 365 if (pslash != NULL) { 366 *pslash = '\0'; // Get rid of /lib. 367 } 368 } 369 Arguments::set_java_home(buf); 370 if (!set_boot_path('/', ':')) { 371 vm_exit_during_initialization("Failed setting boot class path.", NULL); 372 } 373 } 374 375 // Where to look for native libraries. 376 // 377 // Note: Due to a legacy implementation, most of the library path 378 // is set in the launcher. This was to accomodate linking restrictions 379 // on legacy Linux implementations (which are no longer supported). 380 // Eventually, all the library path setting will be done here. 381 // 382 // However, to prevent the proliferation of improperly built native 383 // libraries, the new path component /usr/java/packages is added here. 384 // Eventually, all the library path setting will be done here. 385 { 386 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 387 // should always exist (until the legacy problem cited above is 388 // addressed). 389 const char *v = ::getenv("LD_LIBRARY_PATH"); 390 const char *v_colon = ":"; 391 if (v == NULL) { v = ""; v_colon = ""; } 392 // That's +1 for the colon and +1 for the trailing '\0'. 393 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 394 strlen(v) + 1 + 395 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1, 396 mtInternal); 397 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon); 398 Arguments::set_library_path(ld_library_path); 399 FREE_C_HEAP_ARRAY(char, ld_library_path); 400 } 401 402 // Extensions directories. 403 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 404 Arguments::set_ext_dirs(buf); 405 406 FREE_C_HEAP_ARRAY(char, buf); 407 408 #undef DEFAULT_LIBPATH 409 #undef SYS_EXT_DIR 410 #undef EXTENSIONS_DIR 411 } 412 413 //////////////////////////////////////////////////////////////////////////////// 414 // breakpoint support 415 416 void os::breakpoint() { 417 BREAKPOINT; 418 } 419 420 extern "C" void breakpoint() { 421 // use debugger to set breakpoint here 422 } 423 424 //////////////////////////////////////////////////////////////////////////////// 425 // signal support 426 427 debug_only(static bool signal_sets_initialized = false); 428 static sigset_t unblocked_sigs, vm_sigs; 429 430 void os::Linux::signal_sets_init() { 431 // Should also have an assertion stating we are still single-threaded. 432 assert(!signal_sets_initialized, "Already initialized"); 433 // Fill in signals that are necessarily unblocked for all threads in 434 // the VM. Currently, we unblock the following signals: 435 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 436 // by -Xrs (=ReduceSignalUsage)); 437 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 438 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 439 // the dispositions or masks wrt these signals. 440 // Programs embedding the VM that want to use the above signals for their 441 // own purposes must, at this time, use the "-Xrs" option to prevent 442 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 443 // (See bug 4345157, and other related bugs). 444 // In reality, though, unblocking these signals is really a nop, since 445 // these signals are not blocked by default. 446 sigemptyset(&unblocked_sigs); 447 sigaddset(&unblocked_sigs, SIGILL); 448 sigaddset(&unblocked_sigs, SIGSEGV); 449 sigaddset(&unblocked_sigs, SIGBUS); 450 sigaddset(&unblocked_sigs, SIGFPE); 451 #if defined(PPC64) 452 sigaddset(&unblocked_sigs, SIGTRAP); 453 #endif 454 sigaddset(&unblocked_sigs, SR_signum); 455 456 if (!ReduceSignalUsage) { 457 if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 458 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 459 } 460 if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 461 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 462 } 463 if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 464 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 465 } 466 } 467 // Fill in signals that are blocked by all but the VM thread. 468 sigemptyset(&vm_sigs); 469 if (!ReduceSignalUsage) { 470 sigaddset(&vm_sigs, BREAK_SIGNAL); 471 } 472 debug_only(signal_sets_initialized = true); 473 474 } 475 476 // These are signals that are unblocked while a thread is running Java. 477 // (For some reason, they get blocked by default.) 478 sigset_t* os::Linux::unblocked_signals() { 479 assert(signal_sets_initialized, "Not initialized"); 480 return &unblocked_sigs; 481 } 482 483 // These are the signals that are blocked while a (non-VM) thread is 484 // running Java. Only the VM thread handles these signals. 485 sigset_t* os::Linux::vm_signals() { 486 assert(signal_sets_initialized, "Not initialized"); 487 return &vm_sigs; 488 } 489 490 void os::Linux::hotspot_sigmask(Thread* thread) { 491 492 //Save caller's signal mask before setting VM signal mask 493 sigset_t caller_sigmask; 494 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 495 496 OSThread* osthread = thread->osthread(); 497 osthread->set_caller_sigmask(caller_sigmask); 498 499 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 500 501 if (!ReduceSignalUsage) { 502 if (thread->is_VM_thread()) { 503 // Only the VM thread handles BREAK_SIGNAL ... 504 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 505 } else { 506 // ... all other threads block BREAK_SIGNAL 507 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 508 } 509 } 510 } 511 512 ////////////////////////////////////////////////////////////////////////////// 513 // detecting pthread library 514 515 void os::Linux::libpthread_init() { 516 // Save glibc and pthread version strings. 517 #if !defined(_CS_GNU_LIBC_VERSION) || \ 518 !defined(_CS_GNU_LIBPTHREAD_VERSION) 519 #error "glibc too old (< 2.3.2)" 520 #endif 521 522 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 523 assert(n > 0, "cannot retrieve glibc version"); 524 char *str = (char *)malloc(n, mtInternal); 525 confstr(_CS_GNU_LIBC_VERSION, str, n); 526 os::Linux::set_glibc_version(str); 527 528 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 529 assert(n > 0, "cannot retrieve pthread version"); 530 str = (char *)malloc(n, mtInternal); 531 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 532 os::Linux::set_libpthread_version(str); 533 } 534 535 ///////////////////////////////////////////////////////////////////////////// 536 // thread stack expansion 537 538 // os::Linux::manually_expand_stack() takes care of expanding the thread 539 // stack. Note that this is normally not needed: pthread stacks allocate 540 // thread stack using mmap() without MAP_NORESERVE, so the stack is already 541 // committed. Therefore it is not necessary to expand the stack manually. 542 // 543 // Manually expanding the stack was historically needed on LinuxThreads 544 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays 545 // it is kept to deal with very rare corner cases: 546 // 547 // For one, user may run the VM on an own implementation of threads 548 // whose stacks are - like the old LinuxThreads - implemented using 549 // mmap(MAP_GROWSDOWN). 550 // 551 // Also, this coding may be needed if the VM is running on the primordial 552 // thread. Normally we avoid running on the primordial thread; however, 553 // user may still invoke the VM on the primordial thread. 554 // 555 // The following historical comment describes the details about running 556 // on a thread stack allocated with mmap(MAP_GROWSDOWN): 557 558 559 // Force Linux kernel to expand current thread stack. If "bottom" is close 560 // to the stack guard, caller should block all signals. 561 // 562 // MAP_GROWSDOWN: 563 // A special mmap() flag that is used to implement thread stacks. It tells 564 // kernel that the memory region should extend downwards when needed. This 565 // allows early versions of LinuxThreads to only mmap the first few pages 566 // when creating a new thread. Linux kernel will automatically expand thread 567 // stack as needed (on page faults). 568 // 569 // However, because the memory region of a MAP_GROWSDOWN stack can grow on 570 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 571 // region, it's hard to tell if the fault is due to a legitimate stack 572 // access or because of reading/writing non-exist memory (e.g. buffer 573 // overrun). As a rule, if the fault happens below current stack pointer, 574 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the 575 // application (see Linux kernel fault.c). 576 // 577 // This Linux feature can cause SIGSEGV when VM bangs thread stack for 578 // stack overflow detection. 579 // 580 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 581 // not use MAP_GROWSDOWN. 582 // 583 // To get around the problem and allow stack banging on Linux, we need to 584 // manually expand thread stack after receiving the SIGSEGV. 585 // 586 // There are two ways to expand thread stack to address "bottom", we used 587 // both of them in JVM before 1.5: 588 // 1. adjust stack pointer first so that it is below "bottom", and then 589 // touch "bottom" 590 // 2. mmap() the page in question 591 // 592 // Now alternate signal stack is gone, it's harder to use 2. For instance, 593 // if current sp is already near the lower end of page 101, and we need to 594 // call mmap() to map page 100, it is possible that part of the mmap() frame 595 // will be placed in page 100. When page 100 is mapped, it is zero-filled. 596 // That will destroy the mmap() frame and cause VM to crash. 597 // 598 // The following code works by adjusting sp first, then accessing the "bottom" 599 // page to force a page fault. Linux kernel will then automatically expand the 600 // stack mapping. 601 // 602 // _expand_stack_to() assumes its frame size is less than page size, which 603 // should always be true if the function is not inlined. 604 605 static void NOINLINE _expand_stack_to(address bottom) { 606 address sp; 607 size_t size; 608 volatile char *p; 609 610 // Adjust bottom to point to the largest address within the same page, it 611 // gives us a one-page buffer if alloca() allocates slightly more memory. 612 bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size()); 613 bottom += os::Linux::page_size() - 1; 614 615 // sp might be slightly above current stack pointer; if that's the case, we 616 // will alloca() a little more space than necessary, which is OK. Don't use 617 // os::current_stack_pointer(), as its result can be slightly below current 618 // stack pointer, causing us to not alloca enough to reach "bottom". 619 sp = (address)&sp; 620 621 if (sp > bottom) { 622 size = sp - bottom; 623 p = (volatile char *)alloca(size); 624 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 625 p[0] = '\0'; 626 } 627 } 628 629 void os::Linux::expand_stack_to(address bottom) { 630 _expand_stack_to(bottom); 631 } 632 633 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 634 assert(t!=NULL, "just checking"); 635 assert(t->osthread()->expanding_stack(), "expand should be set"); 636 assert(t->stack_base() != NULL, "stack_base was not initialized"); 637 638 if (addr < t->stack_base() && addr >= t->stack_reserved_zone_base()) { 639 sigset_t mask_all, old_sigset; 640 sigfillset(&mask_all); 641 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 642 _expand_stack_to(addr); 643 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 644 return true; 645 } 646 return false; 647 } 648 649 ////////////////////////////////////////////////////////////////////////////// 650 // create new thread 651 652 // Thread start routine for all newly created threads 653 static void *thread_native_entry(Thread *thread) { 654 // Try to randomize the cache line index of hot stack frames. 655 // This helps when threads of the same stack traces evict each other's 656 // cache lines. The threads can be either from the same JVM instance, or 657 // from different JVM instances. The benefit is especially true for 658 // processors with hyperthreading technology. 659 static int counter = 0; 660 int pid = os::current_process_id(); 661 alloca(((pid ^ counter++) & 7) * 128); 662 663 thread->initialize_thread_current(); 664 665 OSThread* osthread = thread->osthread(); 666 Monitor* sync = osthread->startThread_lock(); 667 668 osthread->set_thread_id(os::current_thread_id()); 669 670 log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 671 os::current_thread_id(), (uintx) pthread_self()); 672 673 if (UseNUMA) { 674 int lgrp_id = os::numa_get_group_id(); 675 if (lgrp_id != -1) { 676 thread->set_lgrp_id(lgrp_id); 677 } 678 } 679 // initialize signal mask for this thread 680 os::Linux::hotspot_sigmask(thread); 681 682 // initialize floating point control register 683 os::Linux::init_thread_fpu_state(); 684 685 // handshaking with parent thread 686 { 687 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 688 689 // notify parent thread 690 osthread->set_state(INITIALIZED); 691 sync->notify_all(); 692 693 // wait until os::start_thread() 694 while (osthread->get_state() == INITIALIZED) { 695 sync->wait(Mutex::_no_safepoint_check_flag); 696 } 697 } 698 699 // call one more level start routine 700 thread->run(); 701 702 log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 703 os::current_thread_id(), (uintx) pthread_self()); 704 705 // If a thread has not deleted itself ("delete this") as part of its 706 // termination sequence, we have to ensure thread-local-storage is 707 // cleared before we actually terminate. No threads should ever be 708 // deleted asynchronously with respect to their termination. 709 if (Thread::current_or_null_safe() != NULL) { 710 assert(Thread::current_or_null_safe() == thread, "current thread is wrong"); 711 thread->clear_thread_current(); 712 } 713 714 return 0; 715 } 716 717 bool os::create_thread(Thread* thread, ThreadType thr_type, 718 size_t req_stack_size) { 719 assert(thread->osthread() == NULL, "caller responsible"); 720 721 // Allocate the OSThread object 722 OSThread* osthread = new OSThread(NULL, NULL); 723 if (osthread == NULL) { 724 return false; 725 } 726 727 // set the correct thread state 728 osthread->set_thread_type(thr_type); 729 730 // Initial state is ALLOCATED but not INITIALIZED 731 osthread->set_state(ALLOCATED); 732 733 thread->set_osthread(osthread); 734 735 // init thread attributes 736 pthread_attr_t attr; 737 pthread_attr_init(&attr); 738 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 739 740 // Calculate stack size if it's not specified by caller. 741 size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size); 742 // In the Linux NPTL pthread implementation the guard size mechanism 743 // is not implemented properly. The posix standard requires adding 744 // the size of the guard pages to the stack size, instead Linux 745 // takes the space out of 'stacksize'. Thus we adapt the requested 746 // stack_size by the size of the guard pages to mimick proper 747 // behaviour. However, be careful not to end up with a size 748 // of zero due to overflow. Don't add the guard page in that case. 749 size_t guard_size = os::Linux::default_guard_size(thr_type); 750 if (stack_size <= SIZE_MAX - guard_size) { 751 stack_size += guard_size; 752 } 753 assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned"); 754 755 int status = pthread_attr_setstacksize(&attr, stack_size); 756 assert_status(status == 0, status, "pthread_attr_setstacksize"); 757 758 // Configure glibc guard page. 759 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 760 761 ThreadState state; 762 763 { 764 pthread_t tid; 765 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread); 766 767 char buf[64]; 768 if (ret == 0) { 769 log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ", 770 (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 771 } else { 772 log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.", 773 os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr)); 774 } 775 776 pthread_attr_destroy(&attr); 777 778 if (ret != 0) { 779 // Need to clean up stuff we've allocated so far 780 thread->set_osthread(NULL); 781 delete osthread; 782 return false; 783 } 784 785 // Store pthread info into the OSThread 786 osthread->set_pthread_id(tid); 787 788 // Wait until child thread is either initialized or aborted 789 { 790 Monitor* sync_with_child = osthread->startThread_lock(); 791 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 792 while ((state = osthread->get_state()) == ALLOCATED) { 793 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 794 } 795 } 796 } 797 798 // Aborted due to thread limit being reached 799 if (state == ZOMBIE) { 800 thread->set_osthread(NULL); 801 delete osthread; 802 return false; 803 } 804 805 // The thread is returned suspended (in state INITIALIZED), 806 // and is started higher up in the call chain 807 assert(state == INITIALIZED, "race condition"); 808 return true; 809 } 810 811 ///////////////////////////////////////////////////////////////////////////// 812 // attach existing thread 813 814 // bootstrap the main thread 815 bool os::create_main_thread(JavaThread* thread) { 816 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 817 return create_attached_thread(thread); 818 } 819 820 bool os::create_attached_thread(JavaThread* thread) { 821 #ifdef ASSERT 822 thread->verify_not_published(); 823 #endif 824 825 // Allocate the OSThread object 826 OSThread* osthread = new OSThread(NULL, NULL); 827 828 if (osthread == NULL) { 829 return false; 830 } 831 832 // Store pthread info into the OSThread 833 osthread->set_thread_id(os::Linux::gettid()); 834 osthread->set_pthread_id(::pthread_self()); 835 836 // initialize floating point control register 837 os::Linux::init_thread_fpu_state(); 838 839 // Initial thread state is RUNNABLE 840 osthread->set_state(RUNNABLE); 841 842 thread->set_osthread(osthread); 843 844 if (UseNUMA) { 845 int lgrp_id = os::numa_get_group_id(); 846 if (lgrp_id != -1) { 847 thread->set_lgrp_id(lgrp_id); 848 } 849 } 850 851 if (os::is_primordial_thread()) { 852 // If current thread is primordial thread, its stack is mapped on demand, 853 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 854 // the entire stack region to avoid SEGV in stack banging. 855 // It is also useful to get around the heap-stack-gap problem on SuSE 856 // kernel (see 4821821 for details). We first expand stack to the top 857 // of yellow zone, then enable stack yellow zone (order is significant, 858 // enabling yellow zone first will crash JVM on SuSE Linux), so there 859 // is no gap between the last two virtual memory regions. 860 861 JavaThread *jt = (JavaThread *)thread; 862 address addr = jt->stack_reserved_zone_base(); 863 assert(addr != NULL, "initialization problem?"); 864 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 865 866 osthread->set_expanding_stack(); 867 os::Linux::manually_expand_stack(jt, addr); 868 osthread->clear_expanding_stack(); 869 } 870 871 // initialize signal mask for this thread 872 // and save the caller's signal mask 873 os::Linux::hotspot_sigmask(thread); 874 875 log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").", 876 os::current_thread_id(), (uintx) pthread_self()); 877 878 return true; 879 } 880 881 void os::pd_start_thread(Thread* thread) { 882 OSThread * osthread = thread->osthread(); 883 assert(osthread->get_state() != INITIALIZED, "just checking"); 884 Monitor* sync_with_child = osthread->startThread_lock(); 885 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 886 sync_with_child->notify(); 887 } 888 889 // Free Linux resources related to the OSThread 890 void os::free_thread(OSThread* osthread) { 891 assert(osthread != NULL, "osthread not set"); 892 893 // We are told to free resources of the argument thread, 894 // but we can only really operate on the current thread. 895 assert(Thread::current()->osthread() == osthread, 896 "os::free_thread but not current thread"); 897 898 #ifdef ASSERT 899 sigset_t current; 900 sigemptyset(¤t); 901 pthread_sigmask(SIG_SETMASK, NULL, ¤t); 902 assert(!sigismember(¤t, SR_signum), "SR signal should not be blocked!"); 903 #endif 904 905 // Restore caller's signal mask 906 sigset_t sigmask = osthread->caller_sigmask(); 907 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 908 909 delete osthread; 910 } 911 912 ////////////////////////////////////////////////////////////////////////////// 913 // primordial thread 914 915 // Check if current thread is the primordial thread, similar to Solaris thr_main. 916 bool os::is_primordial_thread(void) { 917 if (suppress_primordial_thread_resolution) { 918 return false; 919 } 920 char dummy; 921 // If called before init complete, thread stack bottom will be null. 922 // Can be called if fatal error occurs before initialization. 923 if (os::Linux::initial_thread_stack_bottom() == NULL) return false; 924 assert(os::Linux::initial_thread_stack_bottom() != NULL && 925 os::Linux::initial_thread_stack_size() != 0, 926 "os::init did not locate primordial thread's stack region"); 927 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() && 928 (address)&dummy < os::Linux::initial_thread_stack_bottom() + 929 os::Linux::initial_thread_stack_size()) { 930 return true; 931 } else { 932 return false; 933 } 934 } 935 936 // Find the virtual memory area that contains addr 937 static bool find_vma(address addr, address* vma_low, address* vma_high) { 938 FILE *fp = fopen("/proc/self/maps", "r"); 939 if (fp) { 940 address low, high; 941 while (!feof(fp)) { 942 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 943 if (low <= addr && addr < high) { 944 if (vma_low) *vma_low = low; 945 if (vma_high) *vma_high = high; 946 fclose(fp); 947 return true; 948 } 949 } 950 for (;;) { 951 int ch = fgetc(fp); 952 if (ch == EOF || ch == (int)'\n') break; 953 } 954 } 955 fclose(fp); 956 } 957 return false; 958 } 959 960 // Locate primordial thread stack. This special handling of primordial thread stack 961 // is needed because pthread_getattr_np() on most (all?) Linux distros returns 962 // bogus value for the primordial process thread. While the launcher has created 963 // the VM in a new thread since JDK 6, we still have to allow for the use of the 964 // JNI invocation API from a primordial thread. 965 void os::Linux::capture_initial_stack(size_t max_size) { 966 967 // max_size is either 0 (which means accept OS default for thread stacks) or 968 // a user-specified value known to be at least the minimum needed. If we 969 // are actually on the primordial thread we can make it appear that we have a 970 // smaller max_size stack by inserting the guard pages at that location. But we 971 // cannot do anything to emulate a larger stack than what has been provided by 972 // the OS or threading library. In fact if we try to use a stack greater than 973 // what is set by rlimit then we will crash the hosting process. 974 975 // Maximum stack size is the easy part, get it from RLIMIT_STACK. 976 // If this is "unlimited" then it will be a huge value. 977 struct rlimit rlim; 978 getrlimit(RLIMIT_STACK, &rlim); 979 size_t stack_size = rlim.rlim_cur; 980 981 // 6308388: a bug in ld.so will relocate its own .data section to the 982 // lower end of primordial stack; reduce ulimit -s value a little bit 983 // so we won't install guard page on ld.so's data section. 984 // But ensure we don't underflow the stack size - allow 1 page spare 985 if (stack_size >= (size_t)(3 * page_size())) { 986 stack_size -= 2 * page_size(); 987 } 988 989 // Try to figure out where the stack base (top) is. This is harder. 990 // 991 // When an application is started, glibc saves the initial stack pointer in 992 // a global variable "__libc_stack_end", which is then used by system 993 // libraries. __libc_stack_end should be pretty close to stack top. The 994 // variable is available since the very early days. However, because it is 995 // a private interface, it could disappear in the future. 996 // 997 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 998 // to __libc_stack_end, it is very close to stack top, but isn't the real 999 // stack top. Note that /proc may not exist if VM is running as a chroot 1000 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1001 // /proc/<pid>/stat could change in the future (though unlikely). 1002 // 1003 // We try __libc_stack_end first. If that doesn't work, look for 1004 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1005 // as a hint, which should work well in most cases. 1006 1007 uintptr_t stack_start; 1008 1009 // try __libc_stack_end first 1010 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1011 if (p && *p) { 1012 stack_start = *p; 1013 } else { 1014 // see if we can get the start_stack field from /proc/self/stat 1015 FILE *fp; 1016 int pid; 1017 char state; 1018 int ppid; 1019 int pgrp; 1020 int session; 1021 int nr; 1022 int tpgrp; 1023 unsigned long flags; 1024 unsigned long minflt; 1025 unsigned long cminflt; 1026 unsigned long majflt; 1027 unsigned long cmajflt; 1028 unsigned long utime; 1029 unsigned long stime; 1030 long cutime; 1031 long cstime; 1032 long prio; 1033 long nice; 1034 long junk; 1035 long it_real; 1036 uintptr_t start; 1037 uintptr_t vsize; 1038 intptr_t rss; 1039 uintptr_t rsslim; 1040 uintptr_t scodes; 1041 uintptr_t ecode; 1042 int i; 1043 1044 // Figure what the primordial thread stack base is. Code is inspired 1045 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1046 // followed by command name surrounded by parentheses, state, etc. 1047 char stat[2048]; 1048 int statlen; 1049 1050 fp = fopen("/proc/self/stat", "r"); 1051 if (fp) { 1052 statlen = fread(stat, 1, 2047, fp); 1053 stat[statlen] = '\0'; 1054 fclose(fp); 1055 1056 // Skip pid and the command string. Note that we could be dealing with 1057 // weird command names, e.g. user could decide to rename java launcher 1058 // to "java 1.4.2 :)", then the stat file would look like 1059 // 1234 (java 1.4.2 :)) R ... ... 1060 // We don't really need to know the command string, just find the last 1061 // occurrence of ")" and then start parsing from there. See bug 4726580. 1062 char * s = strrchr(stat, ')'); 1063 1064 i = 0; 1065 if (s) { 1066 // Skip blank chars 1067 do { s++; } while (s && isspace(*s)); 1068 1069 #define _UFM UINTX_FORMAT 1070 #define _DFM INTX_FORMAT 1071 1072 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1073 // 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 1074 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, 1075 &state, // 3 %c 1076 &ppid, // 4 %d 1077 &pgrp, // 5 %d 1078 &session, // 6 %d 1079 &nr, // 7 %d 1080 &tpgrp, // 8 %d 1081 &flags, // 9 %lu 1082 &minflt, // 10 %lu 1083 &cminflt, // 11 %lu 1084 &majflt, // 12 %lu 1085 &cmajflt, // 13 %lu 1086 &utime, // 14 %lu 1087 &stime, // 15 %lu 1088 &cutime, // 16 %ld 1089 &cstime, // 17 %ld 1090 &prio, // 18 %ld 1091 &nice, // 19 %ld 1092 &junk, // 20 %ld 1093 &it_real, // 21 %ld 1094 &start, // 22 UINTX_FORMAT 1095 &vsize, // 23 UINTX_FORMAT 1096 &rss, // 24 INTX_FORMAT 1097 &rsslim, // 25 UINTX_FORMAT 1098 &scodes, // 26 UINTX_FORMAT 1099 &ecode, // 27 UINTX_FORMAT 1100 &stack_start); // 28 UINTX_FORMAT 1101 } 1102 1103 #undef _UFM 1104 #undef _DFM 1105 1106 if (i != 28 - 2) { 1107 assert(false, "Bad conversion from /proc/self/stat"); 1108 // product mode - assume we are the primordial thread, good luck in the 1109 // embedded case. 1110 warning("Can't detect primordial thread stack location - bad conversion"); 1111 stack_start = (uintptr_t) &rlim; 1112 } 1113 } else { 1114 // For some reason we can't open /proc/self/stat (for example, running on 1115 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1116 // most cases, so don't abort: 1117 warning("Can't detect primordial thread stack location - no /proc/self/stat"); 1118 stack_start = (uintptr_t) &rlim; 1119 } 1120 } 1121 1122 // Now we have a pointer (stack_start) very close to the stack top, the 1123 // next thing to do is to figure out the exact location of stack top. We 1124 // can find out the virtual memory area that contains stack_start by 1125 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1126 // and its upper limit is the real stack top. (again, this would fail if 1127 // running inside chroot, because /proc may not exist.) 1128 1129 uintptr_t stack_top; 1130 address low, high; 1131 if (find_vma((address)stack_start, &low, &high)) { 1132 // success, "high" is the true stack top. (ignore "low", because initial 1133 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1134 stack_top = (uintptr_t)high; 1135 } else { 1136 // failed, likely because /proc/self/maps does not exist 1137 warning("Can't detect primordial thread stack location - find_vma failed"); 1138 // best effort: stack_start is normally within a few pages below the real 1139 // stack top, use it as stack top, and reduce stack size so we won't put 1140 // guard page outside stack. 1141 stack_top = stack_start; 1142 stack_size -= 16 * page_size(); 1143 } 1144 1145 // stack_top could be partially down the page so align it 1146 stack_top = align_up(stack_top, page_size()); 1147 1148 // Allowed stack value is minimum of max_size and what we derived from rlimit 1149 if (max_size > 0) { 1150 _initial_thread_stack_size = MIN2(max_size, stack_size); 1151 } else { 1152 // Accept the rlimit max, but if stack is unlimited then it will be huge, so 1153 // clamp it at 8MB as we do on Solaris 1154 _initial_thread_stack_size = MIN2(stack_size, 8*M); 1155 } 1156 _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size()); 1157 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1158 1159 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!"); 1160 1161 if (log_is_enabled(Info, os, thread)) { 1162 // See if we seem to be on primordial process thread 1163 bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) && 1164 uintptr_t(&rlim) < stack_top; 1165 1166 log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: " 1167 SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT, 1168 primordial ? "primordial" : "user", max_size / K, _initial_thread_stack_size / K, 1169 stack_top, intptr_t(_initial_thread_stack_bottom)); 1170 } 1171 } 1172 1173 //////////////////////////////////////////////////////////////////////////////// 1174 // time support 1175 1176 // Time since start-up in seconds to a fine granularity. 1177 // Used by VMSelfDestructTimer and the MemProfiler. 1178 double os::elapsedTime() { 1179 1180 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1181 } 1182 1183 jlong os::elapsed_counter() { 1184 return javaTimeNanos() - initial_time_count; 1185 } 1186 1187 jlong os::elapsed_frequency() { 1188 return NANOSECS_PER_SEC; // nanosecond resolution 1189 } 1190 1191 bool os::supports_vtime() { return true; } 1192 bool os::enable_vtime() { return false; } 1193 bool os::vtime_enabled() { return false; } 1194 1195 double os::elapsedVTime() { 1196 struct rusage usage; 1197 int retval = getrusage(RUSAGE_THREAD, &usage); 1198 if (retval == 0) { 1199 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); 1200 } else { 1201 // better than nothing, but not much 1202 return elapsedTime(); 1203 } 1204 } 1205 1206 jlong os::javaTimeMillis() { 1207 timeval time; 1208 int status = gettimeofday(&time, NULL); 1209 assert(status != -1, "linux error"); 1210 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1211 } 1212 1213 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) { 1214 timeval time; 1215 int status = gettimeofday(&time, NULL); 1216 assert(status != -1, "linux error"); 1217 seconds = jlong(time.tv_sec); 1218 nanos = jlong(time.tv_usec) * 1000; 1219 } 1220 1221 1222 #ifndef CLOCK_MONOTONIC 1223 #define CLOCK_MONOTONIC (1) 1224 #endif 1225 1226 void os::Linux::clock_init() { 1227 // we do dlopen's in this particular order due to bug in linux 1228 // dynamical loader (see 6348968) leading to crash on exit 1229 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1230 if (handle == NULL) { 1231 handle = dlopen("librt.so", RTLD_LAZY); 1232 } 1233 1234 if (handle) { 1235 int (*clock_getres_func)(clockid_t, struct timespec*) = 1236 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1237 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1238 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1239 if (clock_getres_func && clock_gettime_func) { 1240 // See if monotonic clock is supported by the kernel. Note that some 1241 // early implementations simply return kernel jiffies (updated every 1242 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1243 // for nano time (though the monotonic property is still nice to have). 1244 // It's fixed in newer kernels, however clock_getres() still returns 1245 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1246 // resolution for now. Hopefully as people move to new kernels, this 1247 // won't be a problem. 1248 struct timespec res; 1249 struct timespec tp; 1250 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1251 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1252 // yes, monotonic clock is supported 1253 _clock_gettime = clock_gettime_func; 1254 return; 1255 } else { 1256 // close librt if there is no monotonic clock 1257 dlclose(handle); 1258 } 1259 } 1260 } 1261 warning("No monotonic clock was available - timed services may " \ 1262 "be adversely affected if the time-of-day clock changes"); 1263 } 1264 1265 #ifndef SYS_clock_getres 1266 #if defined(X86) || defined(PPC64) || defined(S390) 1267 #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261) 1268 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1269 #else 1270 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1271 #define sys_clock_getres(x,y) -1 1272 #endif 1273 #else 1274 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1275 #endif 1276 1277 void os::Linux::fast_thread_clock_init() { 1278 if (!UseLinuxPosixThreadCPUClocks) { 1279 return; 1280 } 1281 clockid_t clockid; 1282 struct timespec tp; 1283 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1284 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1285 1286 // Switch to using fast clocks for thread cpu time if 1287 // the sys_clock_getres() returns 0 error code. 1288 // Note, that some kernels may support the current thread 1289 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1290 // returned by the pthread_getcpuclockid(). 1291 // If the fast Posix clocks are supported then the sys_clock_getres() 1292 // must return at least tp.tv_sec == 0 which means a resolution 1293 // better than 1 sec. This is extra check for reliability. 1294 1295 if (pthread_getcpuclockid_func && 1296 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1297 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1298 _supports_fast_thread_cpu_time = true; 1299 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1300 } 1301 } 1302 1303 jlong os::javaTimeNanos() { 1304 if (os::supports_monotonic_clock()) { 1305 struct timespec tp; 1306 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1307 assert(status == 0, "gettime error"); 1308 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1309 return result; 1310 } else { 1311 timeval time; 1312 int status = gettimeofday(&time, NULL); 1313 assert(status != -1, "linux error"); 1314 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1315 return 1000 * usecs; 1316 } 1317 } 1318 1319 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1320 if (os::supports_monotonic_clock()) { 1321 info_ptr->max_value = ALL_64_BITS; 1322 1323 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1324 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1325 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1326 } else { 1327 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1328 info_ptr->max_value = ALL_64_BITS; 1329 1330 // gettimeofday is a real time clock so it skips 1331 info_ptr->may_skip_backward = true; 1332 info_ptr->may_skip_forward = true; 1333 } 1334 1335 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1336 } 1337 1338 // Return the real, user, and system times in seconds from an 1339 // arbitrary fixed point in the past. 1340 bool os::getTimesSecs(double* process_real_time, 1341 double* process_user_time, 1342 double* process_system_time) { 1343 struct tms ticks; 1344 clock_t real_ticks = times(&ticks); 1345 1346 if (real_ticks == (clock_t) (-1)) { 1347 return false; 1348 } else { 1349 double ticks_per_second = (double) clock_tics_per_sec; 1350 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1351 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1352 *process_real_time = ((double) real_ticks) / ticks_per_second; 1353 1354 return true; 1355 } 1356 } 1357 1358 1359 char * os::local_time_string(char *buf, size_t buflen) { 1360 struct tm t; 1361 time_t long_time; 1362 time(&long_time); 1363 localtime_r(&long_time, &t); 1364 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1365 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1366 t.tm_hour, t.tm_min, t.tm_sec); 1367 return buf; 1368 } 1369 1370 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1371 return localtime_r(clock, res); 1372 } 1373 1374 //////////////////////////////////////////////////////////////////////////////// 1375 // runtime exit support 1376 1377 // Note: os::shutdown() might be called very early during initialization, or 1378 // called from signal handler. Before adding something to os::shutdown(), make 1379 // sure it is async-safe and can handle partially initialized VM. 1380 void os::shutdown() { 1381 1382 // allow PerfMemory to attempt cleanup of any persistent resources 1383 perfMemory_exit(); 1384 1385 // needs to remove object in file system 1386 AttachListener::abort(); 1387 1388 // flush buffered output, finish log files 1389 ostream_abort(); 1390 1391 // Check for abort hook 1392 abort_hook_t abort_hook = Arguments::abort_hook(); 1393 if (abort_hook != NULL) { 1394 abort_hook(); 1395 } 1396 1397 } 1398 1399 // Note: os::abort() might be called very early during initialization, or 1400 // called from signal handler. Before adding something to os::abort(), make 1401 // sure it is async-safe and can handle partially initialized VM. 1402 void os::abort(bool dump_core, void* siginfo, const void* context) { 1403 os::shutdown(); 1404 if (dump_core) { 1405 #ifndef PRODUCT 1406 fdStream out(defaultStream::output_fd()); 1407 out.print_raw("Current thread is "); 1408 char buf[16]; 1409 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1410 out.print_raw_cr(buf); 1411 out.print_raw_cr("Dumping core ..."); 1412 #endif 1413 ::abort(); // dump core 1414 } 1415 1416 ::exit(1); 1417 } 1418 1419 // Die immediately, no exit hook, no abort hook, no cleanup. 1420 void os::die() { 1421 ::abort(); 1422 } 1423 1424 // thread_id is kernel thread id (similar to Solaris LWP id) 1425 intx os::current_thread_id() { return os::Linux::gettid(); } 1426 int os::current_process_id() { 1427 return ::getpid(); 1428 } 1429 1430 // DLL functions 1431 1432 const char* os::dll_file_extension() { return ".so"; } 1433 1434 // This must be hard coded because it's the system's temporary 1435 // directory not the java application's temp directory, ala java.io.tmpdir. 1436 const char* os::get_temp_directory() { return "/tmp"; } 1437 1438 static bool file_exists(const char* filename) { 1439 struct stat statbuf; 1440 if (filename == NULL || strlen(filename) == 0) { 1441 return false; 1442 } 1443 return os::stat(filename, &statbuf) == 0; 1444 } 1445 1446 // check if addr is inside libjvm.so 1447 bool os::address_is_in_vm(address addr) { 1448 static address libjvm_base_addr; 1449 Dl_info dlinfo; 1450 1451 if (libjvm_base_addr == NULL) { 1452 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1453 libjvm_base_addr = (address)dlinfo.dli_fbase; 1454 } 1455 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1456 } 1457 1458 if (dladdr((void *)addr, &dlinfo) != 0) { 1459 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1460 } 1461 1462 return false; 1463 } 1464 1465 bool os::dll_address_to_function_name(address addr, char *buf, 1466 int buflen, int *offset, 1467 bool demangle) { 1468 // buf is not optional, but offset is optional 1469 assert(buf != NULL, "sanity check"); 1470 1471 Dl_info dlinfo; 1472 1473 if (dladdr((void*)addr, &dlinfo) != 0) { 1474 // see if we have a matching symbol 1475 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1476 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) { 1477 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1478 } 1479 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1480 return true; 1481 } 1482 // no matching symbol so try for just file info 1483 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1484 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1485 buf, buflen, offset, dlinfo.dli_fname, demangle)) { 1486 return true; 1487 } 1488 } 1489 } 1490 1491 buf[0] = '\0'; 1492 if (offset != NULL) *offset = -1; 1493 return false; 1494 } 1495 1496 struct _address_to_library_name { 1497 address addr; // input : memory address 1498 size_t buflen; // size of fname 1499 char* fname; // output: library name 1500 address base; // library base addr 1501 }; 1502 1503 static int address_to_library_name_callback(struct dl_phdr_info *info, 1504 size_t size, void *data) { 1505 int i; 1506 bool found = false; 1507 address libbase = NULL; 1508 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1509 1510 // iterate through all loadable segments 1511 for (i = 0; i < info->dlpi_phnum; i++) { 1512 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1513 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1514 // base address of a library is the lowest address of its loaded 1515 // segments. 1516 if (libbase == NULL || libbase > segbase) { 1517 libbase = segbase; 1518 } 1519 // see if 'addr' is within current segment 1520 if (segbase <= d->addr && 1521 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1522 found = true; 1523 } 1524 } 1525 } 1526 1527 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1528 // so dll_address_to_library_name() can fall through to use dladdr() which 1529 // can figure out executable name from argv[0]. 1530 if (found && info->dlpi_name && info->dlpi_name[0]) { 1531 d->base = libbase; 1532 if (d->fname) { 1533 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1534 } 1535 return 1; 1536 } 1537 return 0; 1538 } 1539 1540 bool os::dll_address_to_library_name(address addr, char* buf, 1541 int buflen, int* offset) { 1542 // buf is not optional, but offset is optional 1543 assert(buf != NULL, "sanity check"); 1544 1545 Dl_info dlinfo; 1546 struct _address_to_library_name data; 1547 1548 // There is a bug in old glibc dladdr() implementation that it could resolve 1549 // to wrong library name if the .so file has a base address != NULL. Here 1550 // we iterate through the program headers of all loaded libraries to find 1551 // out which library 'addr' really belongs to. This workaround can be 1552 // removed once the minimum requirement for glibc is moved to 2.3.x. 1553 data.addr = addr; 1554 data.fname = buf; 1555 data.buflen = buflen; 1556 data.base = NULL; 1557 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1558 1559 if (rslt) { 1560 // buf already contains library name 1561 if (offset) *offset = addr - data.base; 1562 return true; 1563 } 1564 if (dladdr((void*)addr, &dlinfo) != 0) { 1565 if (dlinfo.dli_fname != NULL) { 1566 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1567 } 1568 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1569 *offset = addr - (address)dlinfo.dli_fbase; 1570 } 1571 return true; 1572 } 1573 1574 buf[0] = '\0'; 1575 if (offset) *offset = -1; 1576 return false; 1577 } 1578 1579 // Loads .dll/.so and 1580 // in case of error it checks if .dll/.so was built for the 1581 // same architecture as Hotspot is running on 1582 1583 1584 // Remember the stack's state. The Linux dynamic linker will change 1585 // the stack to 'executable' at most once, so we must safepoint only once. 1586 bool os::Linux::_stack_is_executable = false; 1587 1588 // VM operation that loads a library. This is necessary if stack protection 1589 // of the Java stacks can be lost during loading the library. If we 1590 // do not stop the Java threads, they can stack overflow before the stacks 1591 // are protected again. 1592 class VM_LinuxDllLoad: public VM_Operation { 1593 private: 1594 const char *_filename; 1595 char *_ebuf; 1596 int _ebuflen; 1597 void *_lib; 1598 public: 1599 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1600 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1601 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1602 void doit() { 1603 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1604 os::Linux::_stack_is_executable = true; 1605 } 1606 void* loaded_library() { return _lib; } 1607 }; 1608 1609 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1610 void * result = NULL; 1611 bool load_attempted = false; 1612 1613 // Check whether the library to load might change execution rights 1614 // of the stack. If they are changed, the protection of the stack 1615 // guard pages will be lost. We need a safepoint to fix this. 1616 // 1617 // See Linux man page execstack(8) for more info. 1618 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1619 if (!ElfFile::specifies_noexecstack(filename)) { 1620 if (!is_init_completed()) { 1621 os::Linux::_stack_is_executable = true; 1622 // This is OK - No Java threads have been created yet, and hence no 1623 // stack guard pages to fix. 1624 // 1625 // This should happen only when you are building JDK7 using a very 1626 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1627 // 1628 // Dynamic loader will make all stacks executable after 1629 // this function returns, and will not do that again. 1630 assert(Threads::number_of_threads() == 0, "no Java threads should exist yet."); 1631 } else { 1632 warning("You have loaded library %s which might have disabled stack guard. " 1633 "The VM will try to fix the stack guard now.\n" 1634 "It's highly recommended that you fix the library with " 1635 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1636 filename); 1637 1638 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1639 JavaThread *jt = JavaThread::current(); 1640 if (jt->thread_state() != _thread_in_native) { 1641 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1642 // that requires ExecStack. Cannot enter safe point. Let's give up. 1643 warning("Unable to fix stack guard. Giving up."); 1644 } else { 1645 if (!LoadExecStackDllInVMThread) { 1646 // This is for the case where the DLL has an static 1647 // constructor function that executes JNI code. We cannot 1648 // load such DLLs in the VMThread. 1649 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1650 } 1651 1652 ThreadInVMfromNative tiv(jt); 1653 debug_only(VMNativeEntryWrapper vew;) 1654 1655 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1656 VMThread::execute(&op); 1657 if (LoadExecStackDllInVMThread) { 1658 result = op.loaded_library(); 1659 } 1660 load_attempted = true; 1661 } 1662 } 1663 } 1664 } 1665 1666 if (!load_attempted) { 1667 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1668 } 1669 1670 if (result != NULL) { 1671 // Successful loading 1672 return result; 1673 } 1674 1675 Elf32_Ehdr elf_head; 1676 int diag_msg_max_length=ebuflen-strlen(ebuf); 1677 char* diag_msg_buf=ebuf+strlen(ebuf); 1678 1679 if (diag_msg_max_length==0) { 1680 // No more space in ebuf for additional diagnostics message 1681 return NULL; 1682 } 1683 1684 1685 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1686 1687 if (file_descriptor < 0) { 1688 // Can't open library, report dlerror() message 1689 return NULL; 1690 } 1691 1692 bool failed_to_read_elf_head= 1693 (sizeof(elf_head)!= 1694 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1695 1696 ::close(file_descriptor); 1697 if (failed_to_read_elf_head) { 1698 // file i/o error - report dlerror() msg 1699 return NULL; 1700 } 1701 1702 typedef struct { 1703 Elf32_Half code; // Actual value as defined in elf.h 1704 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1705 unsigned char elf_class; // 32 or 64 bit 1706 unsigned char endianess; // MSB or LSB 1707 char* name; // String representation 1708 } arch_t; 1709 1710 #ifndef EM_486 1711 #define EM_486 6 /* Intel 80486 */ 1712 #endif 1713 #ifndef EM_AARCH64 1714 #define EM_AARCH64 183 /* ARM AARCH64 */ 1715 #endif 1716 1717 static const arch_t arch_array[]={ 1718 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1719 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1720 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1721 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1722 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1723 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1724 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1725 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1726 #if defined(VM_LITTLE_ENDIAN) 1727 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"}, 1728 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"}, 1729 #else 1730 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1731 {EM_SH, EM_SH, ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"}, 1732 #endif 1733 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1734 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1735 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1736 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1737 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1738 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1739 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1740 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1741 }; 1742 1743 #if (defined IA32) 1744 static Elf32_Half running_arch_code=EM_386; 1745 #elif (defined AMD64) || (defined X32) 1746 static Elf32_Half running_arch_code=EM_X86_64; 1747 #elif (defined IA64) 1748 static Elf32_Half running_arch_code=EM_IA_64; 1749 #elif (defined __sparc) && (defined _LP64) 1750 static Elf32_Half running_arch_code=EM_SPARCV9; 1751 #elif (defined __sparc) && (!defined _LP64) 1752 static Elf32_Half running_arch_code=EM_SPARC; 1753 #elif (defined __powerpc64__) 1754 static Elf32_Half running_arch_code=EM_PPC64; 1755 #elif (defined __powerpc__) 1756 static Elf32_Half running_arch_code=EM_PPC; 1757 #elif (defined AARCH64) 1758 static Elf32_Half running_arch_code=EM_AARCH64; 1759 #elif (defined ARM) 1760 static Elf32_Half running_arch_code=EM_ARM; 1761 #elif (defined S390) 1762 static Elf32_Half running_arch_code=EM_S390; 1763 #elif (defined ALPHA) 1764 static Elf32_Half running_arch_code=EM_ALPHA; 1765 #elif (defined MIPSEL) 1766 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1767 #elif (defined PARISC) 1768 static Elf32_Half running_arch_code=EM_PARISC; 1769 #elif (defined MIPS) 1770 static Elf32_Half running_arch_code=EM_MIPS; 1771 #elif (defined M68K) 1772 static Elf32_Half running_arch_code=EM_68K; 1773 #elif (defined SH) 1774 static Elf32_Half running_arch_code=EM_SH; 1775 #else 1776 #error Method os::dll_load requires that one of following is defined:\ 1777 AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc 1778 #endif 1779 1780 // Identify compatability class for VM's architecture and library's architecture 1781 // Obtain string descriptions for architectures 1782 1783 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1784 int running_arch_index=-1; 1785 1786 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1787 if (running_arch_code == arch_array[i].code) { 1788 running_arch_index = i; 1789 } 1790 if (lib_arch.code == arch_array[i].code) { 1791 lib_arch.compat_class = arch_array[i].compat_class; 1792 lib_arch.name = arch_array[i].name; 1793 } 1794 } 1795 1796 assert(running_arch_index != -1, 1797 "Didn't find running architecture code (running_arch_code) in arch_array"); 1798 if (running_arch_index == -1) { 1799 // Even though running architecture detection failed 1800 // we may still continue with reporting dlerror() message 1801 return NULL; 1802 } 1803 1804 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1805 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1806 return NULL; 1807 } 1808 1809 #ifndef S390 1810 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1811 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1812 return NULL; 1813 } 1814 #endif // !S390 1815 1816 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1817 if (lib_arch.name!=NULL) { 1818 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1819 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1820 lib_arch.name, arch_array[running_arch_index].name); 1821 } else { 1822 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1823 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1824 lib_arch.code, 1825 arch_array[running_arch_index].name); 1826 } 1827 } 1828 1829 return NULL; 1830 } 1831 1832 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 1833 int ebuflen) { 1834 void * result = ::dlopen(filename, RTLD_LAZY); 1835 if (result == NULL) { 1836 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 1837 ebuf[ebuflen-1] = '\0'; 1838 } 1839 return result; 1840 } 1841 1842 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 1843 int ebuflen) { 1844 void * result = NULL; 1845 if (LoadExecStackDllInVMThread) { 1846 result = dlopen_helper(filename, ebuf, ebuflen); 1847 } 1848 1849 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 1850 // library that requires an executable stack, or which does not have this 1851 // stack attribute set, dlopen changes the stack attribute to executable. The 1852 // read protection of the guard pages gets lost. 1853 // 1854 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 1855 // may have been queued at the same time. 1856 1857 if (!_stack_is_executable) { 1858 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) { 1859 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 1860 jt->stack_guards_enabled()) { // No pending stack overflow exceptions 1861 if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) { 1862 warning("Attempt to reguard stack yellow zone failed."); 1863 } 1864 } 1865 } 1866 } 1867 1868 return result; 1869 } 1870 1871 void* os::dll_lookup(void* handle, const char* name) { 1872 void* res = dlsym(handle, name); 1873 return res; 1874 } 1875 1876 void* os::get_default_process_handle() { 1877 return (void*)::dlopen(NULL, RTLD_LAZY); 1878 } 1879 1880 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) { 1881 int fd = ::open(filename, O_RDONLY); 1882 if (fd == -1) { 1883 return false; 1884 } 1885 1886 if (hdr != NULL) { 1887 st->print_cr("%s", hdr); 1888 } 1889 1890 char buf[33]; 1891 int bytes; 1892 buf[32] = '\0'; 1893 while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) { 1894 st->print_raw(buf, bytes); 1895 } 1896 1897 ::close(fd); 1898 1899 return true; 1900 } 1901 1902 void os::print_dll_info(outputStream *st) { 1903 st->print_cr("Dynamic libraries:"); 1904 1905 char fname[32]; 1906 pid_t pid = os::Linux::gettid(); 1907 1908 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1909 1910 if (!_print_ascii_file(fname, st)) { 1911 st->print("Can not get library information for pid = %d\n", pid); 1912 } 1913 } 1914 1915 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 1916 FILE *procmapsFile = NULL; 1917 1918 // Open the procfs maps file for the current process 1919 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 1920 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 1921 char line[PATH_MAX + 100]; 1922 1923 // Read line by line from 'file' 1924 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 1925 u8 base, top, offset, inode; 1926 char permissions[5]; 1927 char device[6]; 1928 char name[PATH_MAX + 1]; 1929 1930 // Parse fields from line 1931 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s", 1932 &base, &top, permissions, &offset, device, &inode, name); 1933 1934 // Filter by device id '00:00' so that we only get file system mapped files. 1935 if (strcmp(device, "00:00") != 0) { 1936 1937 // Call callback with the fields of interest 1938 if(callback(name, (address)base, (address)top, param)) { 1939 // Oops abort, callback aborted 1940 fclose(procmapsFile); 1941 return 1; 1942 } 1943 } 1944 } 1945 fclose(procmapsFile); 1946 } 1947 return 0; 1948 } 1949 1950 void os::print_os_info_brief(outputStream* st) { 1951 os::Linux::print_distro_info(st); 1952 1953 os::Posix::print_uname_info(st); 1954 1955 os::Linux::print_libversion_info(st); 1956 1957 } 1958 1959 void os::print_os_info(outputStream* st) { 1960 st->print("OS:"); 1961 1962 os::Linux::print_distro_info(st); 1963 1964 os::Posix::print_uname_info(st); 1965 1966 // Print warning if unsafe chroot environment detected 1967 if (unsafe_chroot_detected) { 1968 st->print("WARNING!! "); 1969 st->print_cr("%s", unstable_chroot_error); 1970 } 1971 1972 os::Linux::print_libversion_info(st); 1973 1974 os::Posix::print_rlimit_info(st); 1975 1976 os::Posix::print_load_average(st); 1977 1978 os::Linux::print_full_memory_info(st); 1979 1980 os::Linux::print_proc_sys_info(st); 1981 1982 os::Linux::print_ld_preload_file(st); 1983 1984 os::Linux::print_container_info(st); 1985 } 1986 1987 // Try to identify popular distros. 1988 // Most Linux distributions have a /etc/XXX-release file, which contains 1989 // the OS version string. Newer Linux distributions have a /etc/lsb-release 1990 // file that also contains the OS version string. Some have more than one 1991 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 1992 // /etc/redhat-release.), so the order is important. 1993 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 1994 // their own specific XXX-release file as well as a redhat-release file. 1995 // Because of this the XXX-release file needs to be searched for before the 1996 // redhat-release file. 1997 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the 1998 // search for redhat-release / SuSE-release needs to be before lsb-release. 1999 // Since the lsb-release file is the new standard it needs to be searched 2000 // before the older style release files. 2001 // Searching system-release (Red Hat) and os-release (other Linuxes) are a 2002 // next to last resort. The os-release file is a new standard that contains 2003 // distribution information and the system-release file seems to be an old 2004 // standard that has been replaced by the lsb-release and os-release files. 2005 // Searching for the debian_version file is the last resort. It contains 2006 // an informative string like "6.0.6" or "wheezy/sid". Because of this 2007 // "Debian " is printed before the contents of the debian_version file. 2008 2009 const char* distro_files[] = { 2010 "/etc/oracle-release", 2011 "/etc/mandriva-release", 2012 "/etc/mandrake-release", 2013 "/etc/sun-release", 2014 "/etc/redhat-release", 2015 "/etc/SuSE-release", 2016 "/etc/lsb-release", 2017 "/etc/turbolinux-release", 2018 "/etc/gentoo-release", 2019 "/etc/ltib-release", 2020 "/etc/angstrom-version", 2021 "/etc/system-release", 2022 "/etc/os-release", 2023 NULL }; 2024 2025 void os::Linux::print_distro_info(outputStream* st) { 2026 for (int i = 0;; i++) { 2027 const char* file = distro_files[i]; 2028 if (file == NULL) { 2029 break; // done 2030 } 2031 // If file prints, we found it. 2032 if (_print_ascii_file(file, st)) { 2033 return; 2034 } 2035 } 2036 2037 if (file_exists("/etc/debian_version")) { 2038 st->print("Debian "); 2039 _print_ascii_file("/etc/debian_version", st); 2040 } else { 2041 st->print("Linux"); 2042 } 2043 st->cr(); 2044 } 2045 2046 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) { 2047 char buf[256]; 2048 while (fgets(buf, sizeof(buf), fp)) { 2049 // Edit out extra stuff in expected format 2050 if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) { 2051 char* ptr = strstr(buf, "\""); // the name is in quotes 2052 if (ptr != NULL) { 2053 ptr++; // go beyond first quote 2054 char* nl = strchr(ptr, '\"'); 2055 if (nl != NULL) *nl = '\0'; 2056 strncpy(distro, ptr, length); 2057 } else { 2058 ptr = strstr(buf, "="); 2059 ptr++; // go beyond equals then 2060 char* nl = strchr(ptr, '\n'); 2061 if (nl != NULL) *nl = '\0'; 2062 strncpy(distro, ptr, length); 2063 } 2064 return; 2065 } else if (get_first_line) { 2066 char* nl = strchr(buf, '\n'); 2067 if (nl != NULL) *nl = '\0'; 2068 strncpy(distro, buf, length); 2069 return; 2070 } 2071 } 2072 // print last line and close 2073 char* nl = strchr(buf, '\n'); 2074 if (nl != NULL) *nl = '\0'; 2075 strncpy(distro, buf, length); 2076 } 2077 2078 static void parse_os_info(char* distro, size_t length, const char* file) { 2079 FILE* fp = fopen(file, "r"); 2080 if (fp != NULL) { 2081 // if suse format, print out first line 2082 bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0); 2083 parse_os_info_helper(fp, distro, length, get_first_line); 2084 fclose(fp); 2085 } 2086 } 2087 2088 void os::get_summary_os_info(char* buf, size_t buflen) { 2089 for (int i = 0;; i++) { 2090 const char* file = distro_files[i]; 2091 if (file == NULL) { 2092 break; // ran out of distro_files 2093 } 2094 if (file_exists(file)) { 2095 parse_os_info(buf, buflen, file); 2096 return; 2097 } 2098 } 2099 // special case for debian 2100 if (file_exists("/etc/debian_version")) { 2101 strncpy(buf, "Debian ", buflen); 2102 if (buflen > 7) { 2103 parse_os_info(&buf[7], buflen-7, "/etc/debian_version"); 2104 } 2105 } else { 2106 strncpy(buf, "Linux", buflen); 2107 } 2108 } 2109 2110 void os::Linux::print_libversion_info(outputStream* st) { 2111 // libc, pthread 2112 st->print("libc:"); 2113 st->print("%s ", os::Linux::glibc_version()); 2114 st->print("%s ", os::Linux::libpthread_version()); 2115 st->cr(); 2116 } 2117 2118 void os::Linux::print_proc_sys_info(outputStream* st) { 2119 st->cr(); 2120 st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):"); 2121 _print_ascii_file("/proc/sys/kernel/threads-max", st); 2122 st->cr(); 2123 st->cr(); 2124 2125 st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):"); 2126 _print_ascii_file("/proc/sys/vm/max_map_count", st); 2127 st->cr(); 2128 st->cr(); 2129 2130 st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):"); 2131 _print_ascii_file("/proc/sys/kernel/pid_max", st); 2132 st->cr(); 2133 st->cr(); 2134 } 2135 2136 void os::Linux::print_full_memory_info(outputStream* st) { 2137 st->print("\n/proc/meminfo:\n"); 2138 _print_ascii_file("/proc/meminfo", st); 2139 st->cr(); 2140 } 2141 2142 void os::Linux::print_ld_preload_file(outputStream* st) { 2143 _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:"); 2144 st->cr(); 2145 } 2146 2147 void os::Linux::print_container_info(outputStream* st) { 2148 if (!OSContainer::is_containerized()) { 2149 return; 2150 } 2151 2152 st->print("container (cgroup) information:\n"); 2153 2154 const char *p_ct = OSContainer::container_type(); 2155 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed"); 2156 2157 char *p = OSContainer::cpu_cpuset_cpus(); 2158 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed"); 2159 free(p); 2160 2161 p = OSContainer::cpu_cpuset_memory_nodes(); 2162 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed"); 2163 free(p); 2164 2165 int i = OSContainer::active_processor_count(); 2166 if (i > 0) { 2167 st->print("active_processor_count: %d\n", i); 2168 } else { 2169 st->print("active_processor_count: failed\n"); 2170 } 2171 2172 i = OSContainer::cpu_quota(); 2173 st->print("cpu_quota: %d\n", i); 2174 2175 i = OSContainer::cpu_period(); 2176 st->print("cpu_period: %d\n", i); 2177 2178 i = OSContainer::cpu_shares(); 2179 st->print("cpu_shares: %d\n", i); 2180 2181 jlong j = OSContainer::memory_limit_in_bytes(); 2182 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j); 2183 2184 j = OSContainer::memory_and_swap_limit_in_bytes(); 2185 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j); 2186 2187 j = OSContainer::memory_soft_limit_in_bytes(); 2188 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j); 2189 2190 j = OSContainer::OSContainer::memory_usage_in_bytes(); 2191 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j); 2192 2193 j = OSContainer::OSContainer::memory_max_usage_in_bytes(); 2194 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j); 2195 st->cr(); 2196 } 2197 2198 void os::print_memory_info(outputStream* st) { 2199 2200 st->print("Memory:"); 2201 st->print(" %dk page", os::vm_page_size()>>10); 2202 2203 // values in struct sysinfo are "unsigned long" 2204 struct sysinfo si; 2205 sysinfo(&si); 2206 2207 st->print(", physical " UINT64_FORMAT "k", 2208 os::physical_memory() >> 10); 2209 st->print("(" UINT64_FORMAT "k free)", 2210 os::available_memory() >> 10); 2211 st->print(", swap " UINT64_FORMAT "k", 2212 ((jlong)si.totalswap * si.mem_unit) >> 10); 2213 st->print("(" UINT64_FORMAT "k free)", 2214 ((jlong)si.freeswap * si.mem_unit) >> 10); 2215 st->cr(); 2216 } 2217 2218 // Print the first "model name" line and the first "flags" line 2219 // that we find and nothing more. We assume "model name" comes 2220 // before "flags" so if we find a second "model name", then the 2221 // "flags" field is considered missing. 2222 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) { 2223 #if defined(IA32) || defined(AMD64) 2224 // Other platforms have less repetitive cpuinfo files 2225 FILE *fp = fopen("/proc/cpuinfo", "r"); 2226 if (fp) { 2227 while (!feof(fp)) { 2228 if (fgets(buf, buflen, fp)) { 2229 // Assume model name comes before flags 2230 bool model_name_printed = false; 2231 if (strstr(buf, "model name") != NULL) { 2232 if (!model_name_printed) { 2233 st->print_raw("CPU Model and flags from /proc/cpuinfo:\n"); 2234 st->print_raw(buf); 2235 model_name_printed = true; 2236 } else { 2237 // model name printed but not flags? Odd, just return 2238 fclose(fp); 2239 return true; 2240 } 2241 } 2242 // print the flags line too 2243 if (strstr(buf, "flags") != NULL) { 2244 st->print_raw(buf); 2245 fclose(fp); 2246 return true; 2247 } 2248 } 2249 } 2250 fclose(fp); 2251 } 2252 #endif // x86 platforms 2253 return false; 2254 } 2255 2256 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) { 2257 // Only print the model name if the platform provides this as a summary 2258 if (!print_model_name_and_flags(st, buf, buflen)) { 2259 st->print("\n/proc/cpuinfo:\n"); 2260 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2261 st->print_cr(" <Not Available>"); 2262 } 2263 } 2264 } 2265 2266 #if defined(AMD64) || defined(IA32) || defined(X32) 2267 const char* search_string = "model name"; 2268 #elif defined(M68K) 2269 const char* search_string = "CPU"; 2270 #elif defined(PPC64) 2271 const char* search_string = "cpu"; 2272 #elif defined(S390) 2273 const char* search_string = "processor"; 2274 #elif defined(SPARC) 2275 const char* search_string = "cpu"; 2276 #else 2277 const char* search_string = "Processor"; 2278 #endif 2279 2280 // Parses the cpuinfo file for string representing the model name. 2281 void os::get_summary_cpu_info(char* cpuinfo, size_t length) { 2282 FILE* fp = fopen("/proc/cpuinfo", "r"); 2283 if (fp != NULL) { 2284 while (!feof(fp)) { 2285 char buf[256]; 2286 if (fgets(buf, sizeof(buf), fp)) { 2287 char* start = strstr(buf, search_string); 2288 if (start != NULL) { 2289 char *ptr = start + strlen(search_string); 2290 char *end = buf + strlen(buf); 2291 while (ptr != end) { 2292 // skip whitespace and colon for the rest of the name. 2293 if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') { 2294 break; 2295 } 2296 ptr++; 2297 } 2298 if (ptr != end) { 2299 // reasonable string, get rid of newline and keep the rest 2300 char* nl = strchr(buf, '\n'); 2301 if (nl != NULL) *nl = '\0'; 2302 strncpy(cpuinfo, ptr, length); 2303 fclose(fp); 2304 return; 2305 } 2306 } 2307 } 2308 } 2309 fclose(fp); 2310 } 2311 // cpuinfo not found or parsing failed, just print generic string. The entire 2312 // /proc/cpuinfo file will be printed later in the file (or enough of it for x86) 2313 #if defined(AARCH64) 2314 strncpy(cpuinfo, "AArch64", length); 2315 #elif defined(AMD64) 2316 strncpy(cpuinfo, "x86_64", length); 2317 #elif defined(ARM) // Order wrt. AARCH64 is relevant! 2318 strncpy(cpuinfo, "ARM", length); 2319 #elif defined(IA32) 2320 strncpy(cpuinfo, "x86_32", length); 2321 #elif defined(IA64) 2322 strncpy(cpuinfo, "IA64", length); 2323 #elif defined(PPC) 2324 strncpy(cpuinfo, "PPC64", length); 2325 #elif defined(S390) 2326 strncpy(cpuinfo, "S390", length); 2327 #elif defined(SPARC) 2328 strncpy(cpuinfo, "sparcv9", length); 2329 #elif defined(ZERO_LIBARCH) 2330 strncpy(cpuinfo, ZERO_LIBARCH, length); 2331 #else 2332 strncpy(cpuinfo, "unknown", length); 2333 #endif 2334 } 2335 2336 static void print_signal_handler(outputStream* st, int sig, 2337 char* buf, size_t buflen); 2338 2339 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2340 st->print_cr("Signal Handlers:"); 2341 print_signal_handler(st, SIGSEGV, buf, buflen); 2342 print_signal_handler(st, SIGBUS , buf, buflen); 2343 print_signal_handler(st, SIGFPE , buf, buflen); 2344 print_signal_handler(st, SIGPIPE, buf, buflen); 2345 print_signal_handler(st, SIGXFSZ, buf, buflen); 2346 print_signal_handler(st, SIGILL , buf, buflen); 2347 print_signal_handler(st, SR_signum, buf, buflen); 2348 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2349 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2350 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2351 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2352 #if defined(PPC64) 2353 print_signal_handler(st, SIGTRAP, buf, buflen); 2354 #endif 2355 } 2356 2357 static char saved_jvm_path[MAXPATHLEN] = {0}; 2358 2359 // Find the full path to the current module, libjvm.so 2360 void os::jvm_path(char *buf, jint buflen) { 2361 // Error checking. 2362 if (buflen < MAXPATHLEN) { 2363 assert(false, "must use a large-enough buffer"); 2364 buf[0] = '\0'; 2365 return; 2366 } 2367 // Lazy resolve the path to current module. 2368 if (saved_jvm_path[0] != 0) { 2369 strcpy(buf, saved_jvm_path); 2370 return; 2371 } 2372 2373 char dli_fname[MAXPATHLEN]; 2374 bool ret = dll_address_to_library_name( 2375 CAST_FROM_FN_PTR(address, os::jvm_path), 2376 dli_fname, sizeof(dli_fname), NULL); 2377 assert(ret, "cannot locate libjvm"); 2378 char *rp = NULL; 2379 if (ret && dli_fname[0] != '\0') { 2380 rp = os::Posix::realpath(dli_fname, buf, buflen); 2381 } 2382 if (rp == NULL) { 2383 return; 2384 } 2385 2386 if (Arguments::sun_java_launcher_is_altjvm()) { 2387 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2388 // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so". 2389 // If "/jre/lib/" appears at the right place in the string, then 2390 // assume we are installed in a JDK and we're done. Otherwise, check 2391 // for a JAVA_HOME environment variable and fix up the path so it 2392 // looks like libjvm.so is installed there (append a fake suffix 2393 // hotspot/libjvm.so). 2394 const char *p = buf + strlen(buf) - 1; 2395 for (int count = 0; p > buf && count < 5; ++count) { 2396 for (--p; p > buf && *p != '/'; --p) 2397 /* empty */ ; 2398 } 2399 2400 if (strncmp(p, "/jre/lib/", 9) != 0) { 2401 // Look for JAVA_HOME in the environment. 2402 char* java_home_var = ::getenv("JAVA_HOME"); 2403 if (java_home_var != NULL && java_home_var[0] != 0) { 2404 char* jrelib_p; 2405 int len; 2406 2407 // Check the current module name "libjvm.so". 2408 p = strrchr(buf, '/'); 2409 if (p == NULL) { 2410 return; 2411 } 2412 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2413 2414 rp = os::Posix::realpath(java_home_var, buf, buflen); 2415 if (rp == NULL) { 2416 return; 2417 } 2418 2419 // determine if this is a legacy image or modules image 2420 // modules image doesn't have "jre" subdirectory 2421 len = strlen(buf); 2422 assert(len < buflen, "Ran out of buffer room"); 2423 jrelib_p = buf + len; 2424 snprintf(jrelib_p, buflen-len, "/jre/lib"); 2425 if (0 != access(buf, F_OK)) { 2426 snprintf(jrelib_p, buflen-len, "/lib"); 2427 } 2428 2429 if (0 == access(buf, F_OK)) { 2430 // Use current module name "libjvm.so" 2431 len = strlen(buf); 2432 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2433 } else { 2434 // Go back to path of .so 2435 rp = os::Posix::realpath(dli_fname, buf, buflen); 2436 if (rp == NULL) { 2437 return; 2438 } 2439 } 2440 } 2441 } 2442 } 2443 2444 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2445 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2446 } 2447 2448 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2449 // no prefix required, not even "_" 2450 } 2451 2452 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2453 // no suffix required 2454 } 2455 2456 //////////////////////////////////////////////////////////////////////////////// 2457 // sun.misc.Signal support 2458 2459 static volatile jint sigint_count = 0; 2460 2461 static void UserHandler(int sig, void *siginfo, void *context) { 2462 // 4511530 - sem_post is serialized and handled by the manager thread. When 2463 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2464 // don't want to flood the manager thread with sem_post requests. 2465 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2466 return; 2467 } 2468 2469 // Ctrl-C is pressed during error reporting, likely because the error 2470 // handler fails to abort. Let VM die immediately. 2471 if (sig == SIGINT && VMError::is_error_reported()) { 2472 os::die(); 2473 } 2474 2475 os::signal_notify(sig); 2476 } 2477 2478 void* os::user_handler() { 2479 return CAST_FROM_FN_PTR(void*, UserHandler); 2480 } 2481 2482 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) { 2483 struct timespec ts; 2484 // Semaphore's are always associated with CLOCK_REALTIME 2485 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2486 // see os_posix.cpp for discussion on overflow checking 2487 if (sec >= MAX_SECS) { 2488 ts.tv_sec += MAX_SECS; 2489 ts.tv_nsec = 0; 2490 } else { 2491 ts.tv_sec += sec; 2492 ts.tv_nsec += nsec; 2493 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2494 ts.tv_nsec -= NANOSECS_PER_SEC; 2495 ++ts.tv_sec; // note: this must be <= max_secs 2496 } 2497 } 2498 2499 return ts; 2500 } 2501 2502 extern "C" { 2503 typedef void (*sa_handler_t)(int); 2504 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2505 } 2506 2507 void* os::signal(int signal_number, void* handler) { 2508 struct sigaction sigAct, oldSigAct; 2509 2510 sigfillset(&(sigAct.sa_mask)); 2511 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2512 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2513 2514 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2515 // -1 means registration failed 2516 return (void *)-1; 2517 } 2518 2519 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2520 } 2521 2522 void os::signal_raise(int signal_number) { 2523 ::raise(signal_number); 2524 } 2525 2526 // The following code is moved from os.cpp for making this 2527 // code platform specific, which it is by its very nature. 2528 2529 // Will be modified when max signal is changed to be dynamic 2530 int os::sigexitnum_pd() { 2531 return NSIG; 2532 } 2533 2534 // a counter for each possible signal value 2535 static volatile jint pending_signals[NSIG+1] = { 0 }; 2536 2537 // Linux(POSIX) specific hand shaking semaphore. 2538 static Semaphore* sig_sem = NULL; 2539 static PosixSemaphore sr_semaphore; 2540 2541 static void jdk_misc_signal_init() { 2542 // Initialize signal structures 2543 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2544 2545 // Initialize signal semaphore 2546 sig_sem = new Semaphore(); 2547 } 2548 2549 void os::signal_notify(int sig) { 2550 if (sig_sem != NULL) { 2551 Atomic::inc(&pending_signals[sig]); 2552 sig_sem->signal(); 2553 } else { 2554 // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init 2555 // initialization isn't called. 2556 assert(ReduceSignalUsage, "signal semaphore should be created"); 2557 } 2558 } 2559 2560 static int check_pending_signals() { 2561 Atomic::store(0, &sigint_count); 2562 for (;;) { 2563 for (int i = 0; i < NSIG + 1; i++) { 2564 jint n = pending_signals[i]; 2565 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2566 return i; 2567 } 2568 } 2569 JavaThread *thread = JavaThread::current(); 2570 ThreadBlockInVM tbivm(thread); 2571 2572 bool threadIsSuspended; 2573 do { 2574 thread->set_suspend_equivalent(); 2575 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2576 sig_sem->wait(); 2577 2578 // were we externally suspended while we were waiting? 2579 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2580 if (threadIsSuspended) { 2581 // The semaphore has been incremented, but while we were waiting 2582 // another thread suspended us. We don't want to continue running 2583 // while suspended because that would surprise the thread that 2584 // suspended us. 2585 sig_sem->signal(); 2586 2587 thread->java_suspend_self(); 2588 } 2589 } while (threadIsSuspended); 2590 } 2591 } 2592 2593 int os::signal_wait() { 2594 return check_pending_signals(); 2595 } 2596 2597 //////////////////////////////////////////////////////////////////////////////// 2598 // Virtual Memory 2599 2600 int os::vm_page_size() { 2601 // Seems redundant as all get out 2602 assert(os::Linux::page_size() != -1, "must call os::init"); 2603 return os::Linux::page_size(); 2604 } 2605 2606 // Solaris allocates memory by pages. 2607 int os::vm_allocation_granularity() { 2608 assert(os::Linux::page_size() != -1, "must call os::init"); 2609 return os::Linux::page_size(); 2610 } 2611 2612 // Rationale behind this function: 2613 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2614 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2615 // samples for JITted code. Here we create private executable mapping over the code cache 2616 // and then we can use standard (well, almost, as mapping can change) way to provide 2617 // info for the reporting script by storing timestamp and location of symbol 2618 void linux_wrap_code(char* base, size_t size) { 2619 static volatile jint cnt = 0; 2620 2621 if (!UseOprofile) { 2622 return; 2623 } 2624 2625 char buf[PATH_MAX+1]; 2626 int num = Atomic::add(1, &cnt); 2627 2628 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2629 os::get_temp_directory(), os::current_process_id(), num); 2630 unlink(buf); 2631 2632 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2633 2634 if (fd != -1) { 2635 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2636 if (rv != (off_t)-1) { 2637 if (::write(fd, "", 1) == 1) { 2638 mmap(base, size, 2639 PROT_READ|PROT_WRITE|PROT_EXEC, 2640 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2641 } 2642 } 2643 ::close(fd); 2644 unlink(buf); 2645 } 2646 } 2647 2648 static bool recoverable_mmap_error(int err) { 2649 // See if the error is one we can let the caller handle. This 2650 // list of errno values comes from JBS-6843484. I can't find a 2651 // Linux man page that documents this specific set of errno 2652 // values so while this list currently matches Solaris, it may 2653 // change as we gain experience with this failure mode. 2654 switch (err) { 2655 case EBADF: 2656 case EINVAL: 2657 case ENOTSUP: 2658 // let the caller deal with these errors 2659 return true; 2660 2661 default: 2662 // Any remaining errors on this OS can cause our reserved mapping 2663 // to be lost. That can cause confusion where different data 2664 // structures think they have the same memory mapped. The worst 2665 // scenario is if both the VM and a library think they have the 2666 // same memory mapped. 2667 return false; 2668 } 2669 } 2670 2671 static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2672 int err) { 2673 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2674 ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec, 2675 os::strerror(err), err); 2676 } 2677 2678 static void warn_fail_commit_memory(char* addr, size_t size, 2679 size_t alignment_hint, bool exec, 2680 int err) { 2681 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2682 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, 2683 alignment_hint, exec, os::strerror(err), err); 2684 } 2685 2686 // NOTE: Linux kernel does not really reserve the pages for us. 2687 // All it does is to check if there are enough free pages 2688 // left at the time of mmap(). This could be a potential 2689 // problem. 2690 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2691 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2692 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2693 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2694 if (res != (uintptr_t) MAP_FAILED) { 2695 if (UseNUMAInterleaving) { 2696 numa_make_global(addr, size); 2697 } 2698 return 0; 2699 } 2700 2701 int err = errno; // save errno from mmap() call above 2702 2703 if (!recoverable_mmap_error(err)) { 2704 warn_fail_commit_memory(addr, size, exec, err); 2705 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2706 } 2707 2708 return err; 2709 } 2710 2711 bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2712 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2713 } 2714 2715 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2716 const char* mesg) { 2717 assert(mesg != NULL, "mesg must be specified"); 2718 int err = os::Linux::commit_memory_impl(addr, size, exec); 2719 if (err != 0) { 2720 // the caller wants all commit errors to exit with the specified mesg: 2721 warn_fail_commit_memory(addr, size, exec, err); 2722 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2723 } 2724 } 2725 2726 // Define MAP_HUGETLB here so we can build HotSpot on old systems. 2727 #ifndef MAP_HUGETLB 2728 #define MAP_HUGETLB 0x40000 2729 #endif 2730 2731 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2732 #ifndef MADV_HUGEPAGE 2733 #define MADV_HUGEPAGE 14 2734 #endif 2735 2736 int os::Linux::commit_memory_impl(char* addr, size_t size, 2737 size_t alignment_hint, bool exec) { 2738 int err = os::Linux::commit_memory_impl(addr, size, exec); 2739 if (err == 0) { 2740 realign_memory(addr, size, alignment_hint); 2741 } 2742 return err; 2743 } 2744 2745 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2746 bool exec) { 2747 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2748 } 2749 2750 void os::pd_commit_memory_or_exit(char* addr, size_t size, 2751 size_t alignment_hint, bool exec, 2752 const char* mesg) { 2753 assert(mesg != NULL, "mesg must be specified"); 2754 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2755 if (err != 0) { 2756 // the caller wants all commit errors to exit with the specified mesg: 2757 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2758 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg); 2759 } 2760 } 2761 2762 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2763 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2764 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2765 // be supported or the memory may already be backed by huge pages. 2766 ::madvise(addr, bytes, MADV_HUGEPAGE); 2767 } 2768 } 2769 2770 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2771 // This method works by doing an mmap over an existing mmaping and effectively discarding 2772 // the existing pages. However it won't work for SHM-based large pages that cannot be 2773 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2774 // small pages on top of the SHM segment. This method always works for small pages, so we 2775 // allow that in any case. 2776 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2777 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2778 } 2779 } 2780 2781 void os::numa_make_global(char *addr, size_t bytes) { 2782 Linux::numa_interleave_memory(addr, bytes); 2783 } 2784 2785 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2786 // bind policy to MPOL_PREFERRED for the current thread. 2787 #define USE_MPOL_PREFERRED 0 2788 2789 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2790 // To make NUMA and large pages more robust when both enabled, we need to ease 2791 // the requirements on where the memory should be allocated. MPOL_BIND is the 2792 // default policy and it will force memory to be allocated on the specified 2793 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2794 // the specified node, but will not force it. Using this policy will prevent 2795 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2796 // free large pages. 2797 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2798 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2799 } 2800 2801 bool os::numa_topology_changed() { return false; } 2802 2803 size_t os::numa_get_groups_num() { 2804 // Return just the number of nodes in which it's possible to allocate memory 2805 // (in numa terminology, configured nodes). 2806 return Linux::numa_num_configured_nodes(); 2807 } 2808 2809 int os::numa_get_group_id() { 2810 int cpu_id = Linux::sched_getcpu(); 2811 if (cpu_id != -1) { 2812 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2813 if (lgrp_id != -1) { 2814 return lgrp_id; 2815 } 2816 } 2817 return 0; 2818 } 2819 2820 int os::Linux::get_existing_num_nodes() { 2821 int node; 2822 int highest_node_number = Linux::numa_max_node(); 2823 int num_nodes = 0; 2824 2825 // Get the total number of nodes in the system including nodes without memory. 2826 for (node = 0; node <= highest_node_number; node++) { 2827 if (isnode_in_existing_nodes(node)) { 2828 num_nodes++; 2829 } 2830 } 2831 return num_nodes; 2832 } 2833 2834 size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2835 int highest_node_number = Linux::numa_max_node(); 2836 size_t i = 0; 2837 2838 // Map all node ids in which it is possible to allocate memory. Also nodes are 2839 // not always consecutively available, i.e. available from 0 to the highest 2840 // node number. If the nodes have been bound explicitly using numactl membind, 2841 // then allocate memory from those nodes only. 2842 for (int node = 0; node <= highest_node_number; node++) { 2843 if (Linux::isnode_in_bound_nodes((unsigned int)node)) { 2844 ids[i++] = node; 2845 } 2846 } 2847 return i; 2848 } 2849 2850 bool os::get_page_info(char *start, page_info* info) { 2851 return false; 2852 } 2853 2854 char *os::scan_pages(char *start, char* end, page_info* page_expected, 2855 page_info* page_found) { 2856 return end; 2857 } 2858 2859 2860 int os::Linux::sched_getcpu_syscall(void) { 2861 unsigned int cpu = 0; 2862 int retval = -1; 2863 2864 #if defined(IA32) 2865 #ifndef SYS_getcpu 2866 #define SYS_getcpu 318 2867 #endif 2868 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2869 #elif defined(AMD64) 2870 // Unfortunately we have to bring all these macros here from vsyscall.h 2871 // to be able to compile on old linuxes. 2872 #define __NR_vgetcpu 2 2873 #define VSYSCALL_START (-10UL << 20) 2874 #define VSYSCALL_SIZE 1024 2875 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2876 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2877 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2878 retval = vgetcpu(&cpu, NULL, NULL); 2879 #endif 2880 2881 return (retval == -1) ? retval : cpu; 2882 } 2883 2884 void os::Linux::sched_getcpu_init() { 2885 // sched_getcpu() should be in libc. 2886 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2887 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2888 2889 // If it's not, try a direct syscall. 2890 if (sched_getcpu() == -1) { 2891 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2892 (void*)&sched_getcpu_syscall)); 2893 } 2894 2895 if (sched_getcpu() == -1) { 2896 vm_exit_during_initialization("getcpu(2) system call not supported by kernel"); 2897 } 2898 } 2899 2900 // Something to do with the numa-aware allocator needs these symbols 2901 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2902 extern "C" JNIEXPORT void numa_error(char *where) { } 2903 2904 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails 2905 // load symbol from base version instead. 2906 void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2907 void *f = dlvsym(handle, name, "libnuma_1.1"); 2908 if (f == NULL) { 2909 f = dlsym(handle, name); 2910 } 2911 return f; 2912 } 2913 2914 // Handle request to load libnuma symbol version 1.2 (API v2) only. 2915 // Return NULL if the symbol is not defined in this particular version. 2916 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) { 2917 return dlvsym(handle, name, "libnuma_1.2"); 2918 } 2919 2920 bool os::Linux::libnuma_init() { 2921 if (sched_getcpu() != -1) { // Requires sched_getcpu() support 2922 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2923 if (handle != NULL) { 2924 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2925 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2926 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2927 libnuma_dlsym(handle, "numa_max_node"))); 2928 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t, 2929 libnuma_dlsym(handle, "numa_num_configured_nodes"))); 2930 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2931 libnuma_dlsym(handle, "numa_available"))); 2932 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2933 libnuma_dlsym(handle, "numa_tonode_memory"))); 2934 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2935 libnuma_dlsym(handle, "numa_interleave_memory"))); 2936 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t, 2937 libnuma_v2_dlsym(handle, "numa_interleave_memory"))); 2938 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2939 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2940 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t, 2941 libnuma_dlsym(handle, "numa_bitmask_isbitset"))); 2942 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t, 2943 libnuma_dlsym(handle, "numa_distance"))); 2944 set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t, 2945 libnuma_v2_dlsym(handle, "numa_get_membind"))); 2946 2947 if (numa_available() != -1) { 2948 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2949 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr")); 2950 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr")); 2951 // Create an index -> node mapping, since nodes are not always consecutive 2952 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2953 rebuild_nindex_to_node_map(); 2954 // Create a cpu -> node mapping 2955 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2956 rebuild_cpu_to_node_map(); 2957 return true; 2958 } 2959 } 2960 } 2961 return false; 2962 } 2963 2964 size_t os::Linux::default_guard_size(os::ThreadType thr_type) { 2965 // Creating guard page is very expensive. Java thread has HotSpot 2966 // guard pages, only enable glibc guard page for non-Java threads. 2967 // (Remember: compiler thread is a Java thread, too!) 2968 return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size()); 2969 } 2970 2971 void os::Linux::rebuild_nindex_to_node_map() { 2972 int highest_node_number = Linux::numa_max_node(); 2973 2974 nindex_to_node()->clear(); 2975 for (int node = 0; node <= highest_node_number; node++) { 2976 if (Linux::isnode_in_existing_nodes(node)) { 2977 nindex_to_node()->append(node); 2978 } 2979 } 2980 } 2981 2982 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2983 // The table is later used in get_node_by_cpu(). 2984 void os::Linux::rebuild_cpu_to_node_map() { 2985 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2986 // in libnuma (possible values are starting from 16, 2987 // and continuing up with every other power of 2, but less 2988 // than the maximum number of CPUs supported by kernel), and 2989 // is a subject to change (in libnuma version 2 the requirements 2990 // are more reasonable) we'll just hardcode the number they use 2991 // in the library. 2992 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2993 2994 size_t cpu_num = processor_count(); 2995 size_t cpu_map_size = NCPUS / BitsPerCLong; 2996 size_t cpu_map_valid_size = 2997 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2998 2999 cpu_to_node()->clear(); 3000 cpu_to_node()->at_grow(cpu_num - 1); 3001 3002 size_t node_num = get_existing_num_nodes(); 3003 3004 int distance = 0; 3005 int closest_distance = INT_MAX; 3006 int closest_node = 0; 3007 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 3008 for (size_t i = 0; i < node_num; i++) { 3009 // Check if node is configured (not a memory-less node). If it is not, find 3010 // the closest configured node. Check also if node is bound, i.e. it's allowed 3011 // to allocate memory from the node. If it's not allowed, map cpus in that node 3012 // to the closest node from which memory allocation is allowed. 3013 if (!isnode_in_configured_nodes(nindex_to_node()->at(i)) || 3014 !isnode_in_bound_nodes(nindex_to_node()->at(i))) { 3015 closest_distance = INT_MAX; 3016 // Check distance from all remaining nodes in the system. Ignore distance 3017 // from itself, from another non-configured node, and from another non-bound 3018 // node. 3019 for (size_t m = 0; m < node_num; m++) { 3020 if (m != i && 3021 isnode_in_configured_nodes(nindex_to_node()->at(m)) && 3022 isnode_in_bound_nodes(nindex_to_node()->at(m))) { 3023 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m)); 3024 // If a closest node is found, update. There is always at least one 3025 // configured and bound node in the system so there is always at least 3026 // one node close. 3027 if (distance != 0 && distance < closest_distance) { 3028 closest_distance = distance; 3029 closest_node = nindex_to_node()->at(m); 3030 } 3031 } 3032 } 3033 } else { 3034 // Current node is already a configured node. 3035 closest_node = nindex_to_node()->at(i); 3036 } 3037 3038 // Get cpus from the original node and map them to the closest node. If node 3039 // is a configured node (not a memory-less node), then original node and 3040 // closest node are the same. 3041 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 3042 for (size_t j = 0; j < cpu_map_valid_size; j++) { 3043 if (cpu_map[j] != 0) { 3044 for (size_t k = 0; k < BitsPerCLong; k++) { 3045 if (cpu_map[j] & (1UL << k)) { 3046 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node); 3047 } 3048 } 3049 } 3050 } 3051 } 3052 } 3053 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 3054 } 3055 3056 int os::Linux::get_node_by_cpu(int cpu_id) { 3057 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 3058 return cpu_to_node()->at(cpu_id); 3059 } 3060 return -1; 3061 } 3062 3063 GrowableArray<int>* os::Linux::_cpu_to_node; 3064 GrowableArray<int>* os::Linux::_nindex_to_node; 3065 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 3066 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 3067 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 3068 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes; 3069 os::Linux::numa_available_func_t os::Linux::_numa_available; 3070 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 3071 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 3072 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2; 3073 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 3074 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset; 3075 os::Linux::numa_distance_func_t os::Linux::_numa_distance; 3076 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind; 3077 unsigned long* os::Linux::_numa_all_nodes; 3078 struct bitmask* os::Linux::_numa_all_nodes_ptr; 3079 struct bitmask* os::Linux::_numa_nodes_ptr; 3080 3081 bool os::pd_uncommit_memory(char* addr, size_t size) { 3082 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 3083 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 3084 return res != (uintptr_t) MAP_FAILED; 3085 } 3086 3087 static address get_stack_commited_bottom(address bottom, size_t size) { 3088 address nbot = bottom; 3089 address ntop = bottom + size; 3090 3091 size_t page_sz = os::vm_page_size(); 3092 unsigned pages = size / page_sz; 3093 3094 unsigned char vec[1]; 3095 unsigned imin = 1, imax = pages + 1, imid; 3096 int mincore_return_value = 0; 3097 3098 assert(imin <= imax, "Unexpected page size"); 3099 3100 while (imin < imax) { 3101 imid = (imax + imin) / 2; 3102 nbot = ntop - (imid * page_sz); 3103 3104 // Use a trick with mincore to check whether the page is mapped or not. 3105 // mincore sets vec to 1 if page resides in memory and to 0 if page 3106 // is swapped output but if page we are asking for is unmapped 3107 // it returns -1,ENOMEM 3108 mincore_return_value = mincore(nbot, page_sz, vec); 3109 3110 if (mincore_return_value == -1) { 3111 // Page is not mapped go up 3112 // to find first mapped page 3113 if (errno != EAGAIN) { 3114 assert(errno == ENOMEM, "Unexpected mincore errno"); 3115 imax = imid; 3116 } 3117 } else { 3118 // Page is mapped go down 3119 // to find first not mapped page 3120 imin = imid + 1; 3121 } 3122 } 3123 3124 nbot = nbot + page_sz; 3125 3126 // Adjust stack bottom one page up if last checked page is not mapped 3127 if (mincore_return_value == -1) { 3128 nbot = nbot + page_sz; 3129 } 3130 3131 return nbot; 3132 } 3133 3134 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) { 3135 int mincore_return_value; 3136 const size_t stripe = 1024; // query this many pages each time 3137 unsigned char vec[stripe + 1]; 3138 // set a guard 3139 vec[stripe] = 'X'; 3140 3141 const size_t page_sz = os::vm_page_size(); 3142 size_t pages = size / page_sz; 3143 3144 assert(is_aligned(start, page_sz), "Start address must be page aligned"); 3145 assert(is_aligned(size, page_sz), "Size must be page aligned"); 3146 3147 committed_start = NULL; 3148 3149 int loops = (pages + stripe - 1) / stripe; 3150 int committed_pages = 0; 3151 address loop_base = start; 3152 bool found_range = false; 3153 3154 for (int index = 0; index < loops && !found_range; index ++) { 3155 assert(pages > 0, "Nothing to do"); 3156 int pages_to_query = (pages >= stripe) ? stripe : pages; 3157 pages -= pages_to_query; 3158 3159 // Get stable read 3160 while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN); 3161 3162 // During shutdown, some memory goes away without properly notifying NMT, 3163 // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object. 3164 // Bailout and return as not committed for now. 3165 if (mincore_return_value == -1 && errno == ENOMEM) { 3166 return false; 3167 } 3168 3169 assert(vec[stripe] == 'X', "overflow guard"); 3170 assert(mincore_return_value == 0, "Range must be valid"); 3171 // Process this stripe 3172 for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) { 3173 if ((vec[vecIdx] & 0x01) == 0) { // not committed 3174 // End of current contiguous region 3175 if (committed_start != NULL) { 3176 found_range = true; 3177 break; 3178 } 3179 } else { // committed 3180 // Start of region 3181 if (committed_start == NULL) { 3182 committed_start = loop_base + page_sz * vecIdx; 3183 } 3184 committed_pages ++; 3185 } 3186 } 3187 3188 loop_base += pages_to_query * page_sz; 3189 } 3190 3191 if (committed_start != NULL) { 3192 assert(committed_pages > 0, "Must have committed region"); 3193 assert(committed_pages <= int(size / page_sz), "Can not commit more than it has"); 3194 assert(committed_start >= start && committed_start < start + size, "Out of range"); 3195 committed_size = page_sz * committed_pages; 3196 return true; 3197 } else { 3198 assert(committed_pages == 0, "Should not have committed region"); 3199 return false; 3200 } 3201 } 3202 3203 3204 // Linux uses a growable mapping for the stack, and if the mapping for 3205 // the stack guard pages is not removed when we detach a thread the 3206 // stack cannot grow beyond the pages where the stack guard was 3207 // mapped. If at some point later in the process the stack expands to 3208 // that point, the Linux kernel cannot expand the stack any further 3209 // because the guard pages are in the way, and a segfault occurs. 3210 // 3211 // However, it's essential not to split the stack region by unmapping 3212 // a region (leaving a hole) that's already part of the stack mapping, 3213 // so if the stack mapping has already grown beyond the guard pages at 3214 // the time we create them, we have to truncate the stack mapping. 3215 // So, we need to know the extent of the stack mapping when 3216 // create_stack_guard_pages() is called. 3217 3218 // We only need this for stacks that are growable: at the time of 3219 // writing thread stacks don't use growable mappings (i.e. those 3220 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3221 // only applies to the main thread. 3222 3223 // If the (growable) stack mapping already extends beyond the point 3224 // where we're going to put our guard pages, truncate the mapping at 3225 // that point by munmap()ping it. This ensures that when we later 3226 // munmap() the guard pages we don't leave a hole in the stack 3227 // mapping. This only affects the main/primordial thread 3228 3229 bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3230 if (os::is_primordial_thread()) { 3231 // As we manually grow stack up to bottom inside create_attached_thread(), 3232 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3233 // we don't need to do anything special. 3234 // Check it first, before calling heavy function. 3235 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3236 unsigned char vec[1]; 3237 3238 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3239 // Fallback to slow path on all errors, including EAGAIN 3240 stack_extent = (uintptr_t) get_stack_commited_bottom( 3241 os::Linux::initial_thread_stack_bottom(), 3242 (size_t)addr - stack_extent); 3243 } 3244 3245 if (stack_extent < (uintptr_t)addr) { 3246 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3247 } 3248 } 3249 3250 return os::commit_memory(addr, size, !ExecMem); 3251 } 3252 3253 // If this is a growable mapping, remove the guard pages entirely by 3254 // munmap()ping them. If not, just call uncommit_memory(). This only 3255 // affects the main/primordial thread, but guard against future OS changes. 3256 // It's safe to always unmap guard pages for primordial thread because we 3257 // always place it right after end of the mapped region. 3258 3259 bool os::remove_stack_guard_pages(char* addr, size_t size) { 3260 uintptr_t stack_extent, stack_base; 3261 3262 if (os::is_primordial_thread()) { 3263 return ::munmap(addr, size) == 0; 3264 } 3265 3266 return os::uncommit_memory(addr, size); 3267 } 3268 3269 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3270 // at 'requested_addr'. If there are existing memory mappings at the same 3271 // location, however, they will be overwritten. If 'fixed' is false, 3272 // 'requested_addr' is only treated as a hint, the return value may or 3273 // may not start from the requested address. Unlike Linux mmap(), this 3274 // function returns NULL to indicate failure. 3275 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3276 char * addr; 3277 int flags; 3278 3279 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3280 if (fixed) { 3281 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3282 flags |= MAP_FIXED; 3283 } 3284 3285 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3286 // touch an uncommitted page. Otherwise, the read/write might 3287 // succeed if we have enough swap space to back the physical page. 3288 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3289 flags, -1, 0); 3290 3291 return addr == MAP_FAILED ? NULL : addr; 3292 } 3293 3294 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address 3295 // (req_addr != NULL) or with a given alignment. 3296 // - bytes shall be a multiple of alignment. 3297 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3298 // - alignment sets the alignment at which memory shall be allocated. 3299 // It must be a multiple of allocation granularity. 3300 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3301 // req_addr or NULL. 3302 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) { 3303 3304 size_t extra_size = bytes; 3305 if (req_addr == NULL && alignment > 0) { 3306 extra_size += alignment; 3307 } 3308 3309 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE, 3310 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 3311 -1, 0); 3312 if (start == MAP_FAILED) { 3313 start = NULL; 3314 } else { 3315 if (req_addr != NULL) { 3316 if (start != req_addr) { 3317 ::munmap(start, extra_size); 3318 start = NULL; 3319 } 3320 } else { 3321 char* const start_aligned = align_up(start, alignment); 3322 char* const end_aligned = start_aligned + bytes; 3323 char* const end = start + extra_size; 3324 if (start_aligned > start) { 3325 ::munmap(start, start_aligned - start); 3326 } 3327 if (end_aligned < end) { 3328 ::munmap(end_aligned, end - end_aligned); 3329 } 3330 start = start_aligned; 3331 } 3332 } 3333 return start; 3334 } 3335 3336 static int anon_munmap(char * addr, size_t size) { 3337 return ::munmap(addr, size) == 0; 3338 } 3339 3340 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3341 size_t alignment_hint) { 3342 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3343 } 3344 3345 bool os::pd_release_memory(char* addr, size_t size) { 3346 return anon_munmap(addr, size); 3347 } 3348 3349 static bool linux_mprotect(char* addr, size_t size, int prot) { 3350 // Linux wants the mprotect address argument to be page aligned. 3351 char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size()); 3352 3353 // According to SUSv3, mprotect() should only be used with mappings 3354 // established by mmap(), and mmap() always maps whole pages. Unaligned 3355 // 'addr' likely indicates problem in the VM (e.g. trying to change 3356 // protection of malloc'ed or statically allocated memory). Check the 3357 // caller if you hit this assert. 3358 assert(addr == bottom, "sanity check"); 3359 3360 size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3361 return ::mprotect(bottom, size, prot) == 0; 3362 } 3363 3364 // Set protections specified 3365 bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3366 bool is_committed) { 3367 unsigned int p = 0; 3368 switch (prot) { 3369 case MEM_PROT_NONE: p = PROT_NONE; break; 3370 case MEM_PROT_READ: p = PROT_READ; break; 3371 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3372 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3373 default: 3374 ShouldNotReachHere(); 3375 } 3376 // is_committed is unused. 3377 return linux_mprotect(addr, bytes, p); 3378 } 3379 3380 bool os::guard_memory(char* addr, size_t size) { 3381 return linux_mprotect(addr, size, PROT_NONE); 3382 } 3383 3384 bool os::unguard_memory(char* addr, size_t size) { 3385 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3386 } 3387 3388 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3389 size_t page_size) { 3390 bool result = false; 3391 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3392 MAP_ANONYMOUS|MAP_PRIVATE, 3393 -1, 0); 3394 if (p != MAP_FAILED) { 3395 void *aligned_p = align_up(p, page_size); 3396 3397 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3398 3399 munmap(p, page_size * 2); 3400 } 3401 3402 if (warn && !result) { 3403 warning("TransparentHugePages is not supported by the operating system."); 3404 } 3405 3406 return result; 3407 } 3408 3409 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3410 bool result = false; 3411 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3412 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3413 -1, 0); 3414 3415 if (p != MAP_FAILED) { 3416 // We don't know if this really is a huge page or not. 3417 FILE *fp = fopen("/proc/self/maps", "r"); 3418 if (fp) { 3419 while (!feof(fp)) { 3420 char chars[257]; 3421 long x = 0; 3422 if (fgets(chars, sizeof(chars), fp)) { 3423 if (sscanf(chars, "%lx-%*x", &x) == 1 3424 && x == (long)p) { 3425 if (strstr (chars, "hugepage")) { 3426 result = true; 3427 break; 3428 } 3429 } 3430 } 3431 } 3432 fclose(fp); 3433 } 3434 munmap(p, page_size); 3435 } 3436 3437 if (warn && !result) { 3438 warning("HugeTLBFS is not supported by the operating system."); 3439 } 3440 3441 return result; 3442 } 3443 3444 // Set the coredump_filter bits to include largepages in core dump (bit 6) 3445 // 3446 // From the coredump_filter documentation: 3447 // 3448 // - (bit 0) anonymous private memory 3449 // - (bit 1) anonymous shared memory 3450 // - (bit 2) file-backed private memory 3451 // - (bit 3) file-backed shared memory 3452 // - (bit 4) ELF header pages in file-backed private memory areas (it is 3453 // effective only if the bit 2 is cleared) 3454 // - (bit 5) hugetlb private memory 3455 // - (bit 6) hugetlb shared memory 3456 // - (bit 7) dax private memory 3457 // - (bit 8) dax shared memory 3458 // 3459 static void set_coredump_filter(bool largepages, bool dax_shared) { 3460 FILE *f; 3461 long cdm; 3462 bool filter_changed = false; 3463 3464 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3465 return; 3466 } 3467 3468 if (fscanf(f, "%lx", &cdm) != 1) { 3469 fclose(f); 3470 return; 3471 } 3472 3473 rewind(f); 3474 3475 if (largepages && (cdm & LARGEPAGES_BIT) == 0) { 3476 cdm |= LARGEPAGES_BIT; 3477 filter_changed = true; 3478 } 3479 if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) { 3480 cdm |= DAX_SHARED_BIT; 3481 filter_changed = true; 3482 } 3483 if (filter_changed) { 3484 fprintf(f, "%#lx", cdm); 3485 } 3486 3487 fclose(f); 3488 } 3489 3490 // Large page support 3491 3492 static size_t _large_page_size = 0; 3493 3494 size_t os::Linux::find_large_page_size() { 3495 size_t large_page_size = 0; 3496 3497 // large_page_size on Linux is used to round up heap size. x86 uses either 3498 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3499 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3500 // page as large as 256M. 3501 // 3502 // Here we try to figure out page size by parsing /proc/meminfo and looking 3503 // for a line with the following format: 3504 // Hugepagesize: 2048 kB 3505 // 3506 // If we can't determine the value (e.g. /proc is not mounted, or the text 3507 // format has been changed), we'll use the largest page size supported by 3508 // the processor. 3509 3510 #ifndef ZERO 3511 large_page_size = 3512 AARCH64_ONLY(2 * M) 3513 AMD64_ONLY(2 * M) 3514 ARM32_ONLY(2 * M) 3515 IA32_ONLY(4 * M) 3516 IA64_ONLY(256 * M) 3517 PPC_ONLY(4 * M) 3518 S390_ONLY(1 * M) 3519 SPARC_ONLY(4 * M); 3520 #endif // ZERO 3521 3522 FILE *fp = fopen("/proc/meminfo", "r"); 3523 if (fp) { 3524 while (!feof(fp)) { 3525 int x = 0; 3526 char buf[16]; 3527 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3528 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3529 large_page_size = x * K; 3530 break; 3531 } 3532 } else { 3533 // skip to next line 3534 for (;;) { 3535 int ch = fgetc(fp); 3536 if (ch == EOF || ch == (int)'\n') break; 3537 } 3538 } 3539 } 3540 fclose(fp); 3541 } 3542 3543 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3544 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3545 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3546 proper_unit_for_byte_size(large_page_size)); 3547 } 3548 3549 return large_page_size; 3550 } 3551 3552 size_t os::Linux::setup_large_page_size() { 3553 _large_page_size = Linux::find_large_page_size(); 3554 const size_t default_page_size = (size_t)Linux::page_size(); 3555 if (_large_page_size > default_page_size) { 3556 _page_sizes[0] = _large_page_size; 3557 _page_sizes[1] = default_page_size; 3558 _page_sizes[2] = 0; 3559 } 3560 3561 return _large_page_size; 3562 } 3563 3564 bool os::Linux::setup_large_page_type(size_t page_size) { 3565 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3566 FLAG_IS_DEFAULT(UseSHM) && 3567 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3568 3569 // The type of large pages has not been specified by the user. 3570 3571 // Try UseHugeTLBFS and then UseSHM. 3572 UseHugeTLBFS = UseSHM = true; 3573 3574 // Don't try UseTransparentHugePages since there are known 3575 // performance issues with it turned on. This might change in the future. 3576 UseTransparentHugePages = false; 3577 } 3578 3579 if (UseTransparentHugePages) { 3580 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3581 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3582 UseHugeTLBFS = false; 3583 UseSHM = false; 3584 return true; 3585 } 3586 UseTransparentHugePages = false; 3587 } 3588 3589 if (UseHugeTLBFS) { 3590 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3591 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3592 UseSHM = false; 3593 return true; 3594 } 3595 UseHugeTLBFS = false; 3596 } 3597 3598 return UseSHM; 3599 } 3600 3601 void os::large_page_init() { 3602 if (!UseLargePages && 3603 !UseTransparentHugePages && 3604 !UseHugeTLBFS && 3605 !UseSHM) { 3606 // Not using large pages. 3607 return; 3608 } 3609 3610 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3611 // The user explicitly turned off large pages. 3612 // Ignore the rest of the large pages flags. 3613 UseTransparentHugePages = false; 3614 UseHugeTLBFS = false; 3615 UseSHM = false; 3616 return; 3617 } 3618 3619 size_t large_page_size = Linux::setup_large_page_size(); 3620 UseLargePages = Linux::setup_large_page_type(large_page_size); 3621 3622 set_coredump_filter(true /*largepages*/, false /*dax_shared*/); 3623 } 3624 3625 #ifndef SHM_HUGETLB 3626 #define SHM_HUGETLB 04000 3627 #endif 3628 3629 #define shm_warning_format(format, ...) \ 3630 do { \ 3631 if (UseLargePages && \ 3632 (!FLAG_IS_DEFAULT(UseLargePages) || \ 3633 !FLAG_IS_DEFAULT(UseSHM) || \ 3634 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \ 3635 warning(format, __VA_ARGS__); \ 3636 } \ 3637 } while (0) 3638 3639 #define shm_warning(str) shm_warning_format("%s", str) 3640 3641 #define shm_warning_with_errno(str) \ 3642 do { \ 3643 int err = errno; \ 3644 shm_warning_format(str " (error = %d)", err); \ 3645 } while (0) 3646 3647 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) { 3648 assert(is_aligned(bytes, alignment), "Must be divisible by the alignment"); 3649 3650 if (!is_aligned(alignment, SHMLBA)) { 3651 assert(false, "Code below assumes that alignment is at least SHMLBA aligned"); 3652 return NULL; 3653 } 3654 3655 // To ensure that we get 'alignment' aligned memory from shmat, 3656 // we pre-reserve aligned virtual memory and then attach to that. 3657 3658 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL); 3659 if (pre_reserved_addr == NULL) { 3660 // Couldn't pre-reserve aligned memory. 3661 shm_warning("Failed to pre-reserve aligned memory for shmat."); 3662 return NULL; 3663 } 3664 3665 // SHM_REMAP is needed to allow shmat to map over an existing mapping. 3666 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP); 3667 3668 if ((intptr_t)addr == -1) { 3669 int err = errno; 3670 shm_warning_with_errno("Failed to attach shared memory."); 3671 3672 assert(err != EACCES, "Unexpected error"); 3673 assert(err != EIDRM, "Unexpected error"); 3674 assert(err != EINVAL, "Unexpected error"); 3675 3676 // Since we don't know if the kernel unmapped the pre-reserved memory area 3677 // we can't unmap it, since that would potentially unmap memory that was 3678 // mapped from other threads. 3679 return NULL; 3680 } 3681 3682 return addr; 3683 } 3684 3685 static char* shmat_at_address(int shmid, char* req_addr) { 3686 if (!is_aligned(req_addr, SHMLBA)) { 3687 assert(false, "Requested address needs to be SHMLBA aligned"); 3688 return NULL; 3689 } 3690 3691 char* addr = (char*)shmat(shmid, req_addr, 0); 3692 3693 if ((intptr_t)addr == -1) { 3694 shm_warning_with_errno("Failed to attach shared memory."); 3695 return NULL; 3696 } 3697 3698 return addr; 3699 } 3700 3701 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) { 3702 // If a req_addr has been provided, we assume that the caller has already aligned the address. 3703 if (req_addr != NULL) { 3704 assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size"); 3705 assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment"); 3706 return shmat_at_address(shmid, req_addr); 3707 } 3708 3709 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically 3710 // return large page size aligned memory addresses when req_addr == NULL. 3711 // However, if the alignment is larger than the large page size, we have 3712 // to manually ensure that the memory returned is 'alignment' aligned. 3713 if (alignment > os::large_page_size()) { 3714 assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size"); 3715 return shmat_with_alignment(shmid, bytes, alignment); 3716 } else { 3717 return shmat_at_address(shmid, NULL); 3718 } 3719 } 3720 3721 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3722 char* req_addr, bool exec) { 3723 // "exec" is passed in but not used. Creating the shared image for 3724 // the code cache doesn't have an SHM_X executable permission to check. 3725 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3726 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3727 assert(is_aligned(req_addr, alignment), "Unaligned address"); 3728 3729 if (!is_aligned(bytes, os::large_page_size())) { 3730 return NULL; // Fallback to small pages. 3731 } 3732 3733 // Create a large shared memory region to attach to based on size. 3734 // Currently, size is the total size of the heap. 3735 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3736 if (shmid == -1) { 3737 // Possible reasons for shmget failure: 3738 // 1. shmmax is too small for Java heap. 3739 // > check shmmax value: cat /proc/sys/kernel/shmmax 3740 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3741 // 2. not enough large page memory. 3742 // > check available large pages: cat /proc/meminfo 3743 // > increase amount of large pages: 3744 // echo new_value > /proc/sys/vm/nr_hugepages 3745 // Note 1: different Linux may use different name for this property, 3746 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3747 // Note 2: it's possible there's enough physical memory available but 3748 // they are so fragmented after a long run that they can't 3749 // coalesce into large pages. Try to reserve large pages when 3750 // the system is still "fresh". 3751 shm_warning_with_errno("Failed to reserve shared memory."); 3752 return NULL; 3753 } 3754 3755 // Attach to the region. 3756 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr); 3757 3758 // Remove shmid. If shmat() is successful, the actual shared memory segment 3759 // will be deleted when it's detached by shmdt() or when the process 3760 // terminates. If shmat() is not successful this will remove the shared 3761 // segment immediately. 3762 shmctl(shmid, IPC_RMID, NULL); 3763 3764 return addr; 3765 } 3766 3767 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3768 int error) { 3769 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3770 3771 bool warn_on_failure = UseLargePages && 3772 (!FLAG_IS_DEFAULT(UseLargePages) || 3773 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3774 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3775 3776 if (warn_on_failure) { 3777 char msg[128]; 3778 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3779 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3780 warning("%s", msg); 3781 } 3782 } 3783 3784 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3785 char* req_addr, 3786 bool exec) { 3787 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3788 assert(is_aligned(bytes, os::large_page_size()), "Unaligned size"); 3789 assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3790 3791 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3792 char* addr = (char*)::mmap(req_addr, bytes, prot, 3793 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3794 -1, 0); 3795 3796 if (addr == MAP_FAILED) { 3797 warn_on_large_pages_failure(req_addr, bytes, errno); 3798 return NULL; 3799 } 3800 3801 assert(is_aligned(addr, os::large_page_size()), "Must be"); 3802 3803 return addr; 3804 } 3805 3806 // Reserve memory using mmap(MAP_HUGETLB). 3807 // - bytes shall be a multiple of alignment. 3808 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3809 // - alignment sets the alignment at which memory shall be allocated. 3810 // It must be a multiple of allocation granularity. 3811 // Returns address of memory or NULL. If req_addr was not NULL, will only return 3812 // req_addr or NULL. 3813 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3814 size_t alignment, 3815 char* req_addr, 3816 bool exec) { 3817 size_t large_page_size = os::large_page_size(); 3818 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3819 3820 assert(is_aligned(req_addr, alignment), "Must be"); 3821 assert(is_aligned(bytes, alignment), "Must be"); 3822 3823 // First reserve - but not commit - the address range in small pages. 3824 char* const start = anon_mmap_aligned(bytes, alignment, req_addr); 3825 3826 if (start == NULL) { 3827 return NULL; 3828 } 3829 3830 assert(is_aligned(start, alignment), "Must be"); 3831 3832 char* end = start + bytes; 3833 3834 // Find the regions of the allocated chunk that can be promoted to large pages. 3835 char* lp_start = align_up(start, large_page_size); 3836 char* lp_end = align_down(end, large_page_size); 3837 3838 size_t lp_bytes = lp_end - lp_start; 3839 3840 assert(is_aligned(lp_bytes, large_page_size), "Must be"); 3841 3842 if (lp_bytes == 0) { 3843 // The mapped region doesn't even span the start and the end of a large page. 3844 // Fall back to allocate a non-special area. 3845 ::munmap(start, end - start); 3846 return NULL; 3847 } 3848 3849 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3850 3851 void* result; 3852 3853 // Commit small-paged leading area. 3854 if (start != lp_start) { 3855 result = ::mmap(start, lp_start - start, prot, 3856 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3857 -1, 0); 3858 if (result == MAP_FAILED) { 3859 ::munmap(lp_start, end - lp_start); 3860 return NULL; 3861 } 3862 } 3863 3864 // Commit large-paged area. 3865 result = ::mmap(lp_start, lp_bytes, prot, 3866 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3867 -1, 0); 3868 if (result == MAP_FAILED) { 3869 warn_on_large_pages_failure(lp_start, lp_bytes, errno); 3870 // If the mmap above fails, the large pages region will be unmapped and we 3871 // have regions before and after with small pages. Release these regions. 3872 // 3873 // | mapped | unmapped | mapped | 3874 // ^ ^ ^ ^ 3875 // start lp_start lp_end end 3876 // 3877 ::munmap(start, lp_start - start); 3878 ::munmap(lp_end, end - lp_end); 3879 return NULL; 3880 } 3881 3882 // Commit small-paged trailing area. 3883 if (lp_end != end) { 3884 result = ::mmap(lp_end, end - lp_end, prot, 3885 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3886 -1, 0); 3887 if (result == MAP_FAILED) { 3888 ::munmap(start, lp_end - start); 3889 return NULL; 3890 } 3891 } 3892 3893 return start; 3894 } 3895 3896 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3897 size_t alignment, 3898 char* req_addr, 3899 bool exec) { 3900 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3901 assert(is_aligned(req_addr, alignment), "Must be"); 3902 assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be"); 3903 assert(is_power_of_2(os::large_page_size()), "Must be"); 3904 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3905 3906 if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3907 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3908 } else { 3909 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3910 } 3911 } 3912 3913 char* os::reserve_memory_special(size_t bytes, size_t alignment, 3914 char* req_addr, bool exec) { 3915 assert(UseLargePages, "only for large pages"); 3916 3917 char* addr; 3918 if (UseSHM) { 3919 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3920 } else { 3921 assert(UseHugeTLBFS, "must be"); 3922 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3923 } 3924 3925 if (addr != NULL) { 3926 if (UseNUMAInterleaving) { 3927 numa_make_global(addr, bytes); 3928 } 3929 3930 // The memory is committed 3931 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3932 } 3933 3934 return addr; 3935 } 3936 3937 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3938 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3939 return shmdt(base) == 0; 3940 } 3941 3942 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3943 return pd_release_memory(base, bytes); 3944 } 3945 3946 bool os::release_memory_special(char* base, size_t bytes) { 3947 bool res; 3948 if (MemTracker::tracking_level() > NMT_minimal) { 3949 Tracker tkr(Tracker::release); 3950 res = os::Linux::release_memory_special_impl(base, bytes); 3951 if (res) { 3952 tkr.record((address)base, bytes); 3953 } 3954 3955 } else { 3956 res = os::Linux::release_memory_special_impl(base, bytes); 3957 } 3958 return res; 3959 } 3960 3961 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3962 assert(UseLargePages, "only for large pages"); 3963 bool res; 3964 3965 if (UseSHM) { 3966 res = os::Linux::release_memory_special_shm(base, bytes); 3967 } else { 3968 assert(UseHugeTLBFS, "must be"); 3969 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3970 } 3971 return res; 3972 } 3973 3974 size_t os::large_page_size() { 3975 return _large_page_size; 3976 } 3977 3978 // With SysV SHM the entire memory region must be allocated as shared 3979 // memory. 3980 // HugeTLBFS allows application to commit large page memory on demand. 3981 // However, when committing memory with HugeTLBFS fails, the region 3982 // that was supposed to be committed will lose the old reservation 3983 // and allow other threads to steal that memory region. Because of this 3984 // behavior we can't commit HugeTLBFS memory. 3985 bool os::can_commit_large_page_memory() { 3986 return UseTransparentHugePages; 3987 } 3988 3989 bool os::can_execute_large_page_memory() { 3990 return UseTransparentHugePages || UseHugeTLBFS; 3991 } 3992 3993 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) { 3994 assert(file_desc >= 0, "file_desc is not valid"); 3995 char* result = pd_attempt_reserve_memory_at(bytes, requested_addr); 3996 if (result != NULL) { 3997 if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) { 3998 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory")); 3999 } 4000 } 4001 return result; 4002 } 4003 4004 // Reserve memory at an arbitrary address, only if that area is 4005 // available (and not reserved for something else). 4006 4007 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 4008 const int max_tries = 10; 4009 char* base[max_tries]; 4010 size_t size[max_tries]; 4011 const size_t gap = 0x000000; 4012 4013 // Assert only that the size is a multiple of the page size, since 4014 // that's all that mmap requires, and since that's all we really know 4015 // about at this low abstraction level. If we need higher alignment, 4016 // we can either pass an alignment to this method or verify alignment 4017 // in one of the methods further up the call chain. See bug 5044738. 4018 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 4019 4020 // Repeatedly allocate blocks until the block is allocated at the 4021 // right spot. 4022 4023 // Linux mmap allows caller to pass an address as hint; give it a try first, 4024 // if kernel honors the hint then we can return immediately. 4025 char * addr = anon_mmap(requested_addr, bytes, false); 4026 if (addr == requested_addr) { 4027 return requested_addr; 4028 } 4029 4030 if (addr != NULL) { 4031 // mmap() is successful but it fails to reserve at the requested address 4032 anon_munmap(addr, bytes); 4033 } 4034 4035 int i; 4036 for (i = 0; i < max_tries; ++i) { 4037 base[i] = reserve_memory(bytes); 4038 4039 if (base[i] != NULL) { 4040 // Is this the block we wanted? 4041 if (base[i] == requested_addr) { 4042 size[i] = bytes; 4043 break; 4044 } 4045 4046 // Does this overlap the block we wanted? Give back the overlapped 4047 // parts and try again. 4048 4049 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i]; 4050 if (top_overlap >= 0 && (size_t)top_overlap < bytes) { 4051 unmap_memory(base[i], top_overlap); 4052 base[i] += top_overlap; 4053 size[i] = bytes - top_overlap; 4054 } else { 4055 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr; 4056 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) { 4057 unmap_memory(requested_addr, bottom_overlap); 4058 size[i] = bytes - bottom_overlap; 4059 } else { 4060 size[i] = bytes; 4061 } 4062 } 4063 } 4064 } 4065 4066 // Give back the unused reserved pieces. 4067 4068 for (int j = 0; j < i; ++j) { 4069 if (base[j] != NULL) { 4070 unmap_memory(base[j], size[j]); 4071 } 4072 } 4073 4074 if (i < max_tries) { 4075 return requested_addr; 4076 } else { 4077 return NULL; 4078 } 4079 } 4080 4081 size_t os::read(int fd, void *buf, unsigned int nBytes) { 4082 return ::read(fd, buf, nBytes); 4083 } 4084 4085 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) { 4086 return ::pread(fd, buf, nBytes, offset); 4087 } 4088 4089 // Short sleep, direct OS call. 4090 // 4091 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 4092 // sched_yield(2) will actually give up the CPU: 4093 // 4094 // * Alone on this pariticular CPU, keeps running. 4095 // * Before the introduction of "skip_buddy" with "compat_yield" disabled 4096 // (pre 2.6.39). 4097 // 4098 // So calling this with 0 is an alternative. 4099 // 4100 void os::naked_short_sleep(jlong ms) { 4101 struct timespec req; 4102 4103 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 4104 req.tv_sec = 0; 4105 if (ms > 0) { 4106 req.tv_nsec = (ms % 1000) * 1000000; 4107 } else { 4108 req.tv_nsec = 1; 4109 } 4110 4111 nanosleep(&req, NULL); 4112 4113 return; 4114 } 4115 4116 // Sleep forever; naked call to OS-specific sleep; use with CAUTION 4117 void os::infinite_sleep() { 4118 while (true) { // sleep forever ... 4119 ::sleep(100); // ... 100 seconds at a time 4120 } 4121 } 4122 4123 // Used to convert frequent JVM_Yield() to nops 4124 bool os::dont_yield() { 4125 return DontYieldALot; 4126 } 4127 4128 void os::naked_yield() { 4129 sched_yield(); 4130 } 4131 4132 //////////////////////////////////////////////////////////////////////////////// 4133 // thread priority support 4134 4135 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 4136 // only supports dynamic priority, static priority must be zero. For real-time 4137 // applications, Linux supports SCHED_RR which allows static priority (1-99). 4138 // However, for large multi-threaded applications, SCHED_RR is not only slower 4139 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 4140 // of 5 runs - Sep 2005). 4141 // 4142 // The following code actually changes the niceness of kernel-thread/LWP. It 4143 // has an assumption that setpriority() only modifies one kernel-thread/LWP, 4144 // not the entire user process, and user level threads are 1:1 mapped to kernel 4145 // threads. It has always been the case, but could change in the future. For 4146 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). 4147 // It is only used when ThreadPriorityPolicy=1 and requires root privilege. 4148 4149 int os::java_to_os_priority[CriticalPriority + 1] = { 4150 19, // 0 Entry should never be used 4151 4152 4, // 1 MinPriority 4153 3, // 2 4154 2, // 3 4155 4156 1, // 4 4157 0, // 5 NormPriority 4158 -1, // 6 4159 4160 -2, // 7 4161 -3, // 8 4162 -4, // 9 NearMaxPriority 4163 4164 -5, // 10 MaxPriority 4165 4166 -5 // 11 CriticalPriority 4167 }; 4168 4169 static int prio_init() { 4170 if (ThreadPriorityPolicy == 1) { 4171 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 4172 // if effective uid is not root. Perhaps, a more elegant way of doing 4173 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 4174 if (geteuid() != 0) { 4175 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 4176 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 4177 } 4178 ThreadPriorityPolicy = 0; 4179 } 4180 } 4181 if (UseCriticalJavaThreadPriority) { 4182 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 4183 } 4184 return 0; 4185 } 4186 4187 OSReturn os::set_native_priority(Thread* thread, int newpri) { 4188 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 4189 4190 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 4191 return (ret == 0) ? OS_OK : OS_ERR; 4192 } 4193 4194 OSReturn os::get_native_priority(const Thread* const thread, 4195 int *priority_ptr) { 4196 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 4197 *priority_ptr = java_to_os_priority[NormPriority]; 4198 return OS_OK; 4199 } 4200 4201 errno = 0; 4202 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 4203 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 4204 } 4205 4206 //////////////////////////////////////////////////////////////////////////////// 4207 // suspend/resume support 4208 4209 // The low-level signal-based suspend/resume support is a remnant from the 4210 // old VM-suspension that used to be for java-suspension, safepoints etc, 4211 // within hotspot. Currently used by JFR's OSThreadSampler 4212 // 4213 // The remaining code is greatly simplified from the more general suspension 4214 // code that used to be used. 4215 // 4216 // The protocol is quite simple: 4217 // - suspend: 4218 // - sends a signal to the target thread 4219 // - polls the suspend state of the osthread using a yield loop 4220 // - target thread signal handler (SR_handler) sets suspend state 4221 // and blocks in sigsuspend until continued 4222 // - resume: 4223 // - sets target osthread state to continue 4224 // - sends signal to end the sigsuspend loop in the SR_handler 4225 // 4226 // Note that the SR_lock plays no role in this suspend/resume protocol, 4227 // but is checked for NULL in SR_handler as a thread termination indicator. 4228 // The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs. 4229 // 4230 // Note that resume_clear_context() and suspend_save_context() are needed 4231 // by SR_handler(), so that fetch_frame_from_ucontext() works, 4232 // which in part is used by: 4233 // - Forte Analyzer: AsyncGetCallTrace() 4234 // - StackBanging: get_frame_at_stack_banging_point() 4235 4236 static void resume_clear_context(OSThread *osthread) { 4237 osthread->set_ucontext(NULL); 4238 osthread->set_siginfo(NULL); 4239 } 4240 4241 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 4242 ucontext_t* context) { 4243 osthread->set_ucontext(context); 4244 osthread->set_siginfo(siginfo); 4245 } 4246 4247 // Handler function invoked when a thread's execution is suspended or 4248 // resumed. We have to be careful that only async-safe functions are 4249 // called here (Note: most pthread functions are not async safe and 4250 // should be avoided.) 4251 // 4252 // Note: sigwait() is a more natural fit than sigsuspend() from an 4253 // interface point of view, but sigwait() prevents the signal hander 4254 // from being run. libpthread would get very confused by not having 4255 // its signal handlers run and prevents sigwait()'s use with the 4256 // mutex granting granting signal. 4257 // 4258 // Currently only ever called on the VMThread and JavaThreads (PC sampling) 4259 // 4260 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 4261 // Save and restore errno to avoid confusing native code with EINTR 4262 // after sigsuspend. 4263 int old_errno = errno; 4264 4265 Thread* thread = Thread::current_or_null_safe(); 4266 assert(thread != NULL, "Missing current thread in SR_handler"); 4267 4268 // On some systems we have seen signal delivery get "stuck" until the signal 4269 // mask is changed as part of thread termination. Check that the current thread 4270 // has not already terminated (via SR_lock()) - else the following assertion 4271 // will fail because the thread is no longer a JavaThread as the ~JavaThread 4272 // destructor has completed. 4273 4274 if (thread->SR_lock() == NULL) { 4275 return; 4276 } 4277 4278 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 4279 4280 OSThread* osthread = thread->osthread(); 4281 4282 os::SuspendResume::State current = osthread->sr.state(); 4283 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 4284 suspend_save_context(osthread, siginfo, context); 4285 4286 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4287 os::SuspendResume::State state = osthread->sr.suspended(); 4288 if (state == os::SuspendResume::SR_SUSPENDED) { 4289 sigset_t suspend_set; // signals for sigsuspend() 4290 sigemptyset(&suspend_set); 4291 // get current set of blocked signals and unblock resume signal 4292 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4293 sigdelset(&suspend_set, SR_signum); 4294 4295 sr_semaphore.signal(); 4296 // wait here until we are resumed 4297 while (1) { 4298 sigsuspend(&suspend_set); 4299 4300 os::SuspendResume::State result = osthread->sr.running(); 4301 if (result == os::SuspendResume::SR_RUNNING) { 4302 sr_semaphore.signal(); 4303 break; 4304 } 4305 } 4306 4307 } else if (state == os::SuspendResume::SR_RUNNING) { 4308 // request was cancelled, continue 4309 } else { 4310 ShouldNotReachHere(); 4311 } 4312 4313 resume_clear_context(osthread); 4314 } else if (current == os::SuspendResume::SR_RUNNING) { 4315 // request was cancelled, continue 4316 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4317 // ignore 4318 } else { 4319 // ignore 4320 } 4321 4322 errno = old_errno; 4323 } 4324 4325 static int SR_initialize() { 4326 struct sigaction act; 4327 char *s; 4328 4329 // Get signal number to use for suspend/resume 4330 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4331 int sig = ::strtol(s, 0, 10); 4332 if (sig > MAX2(SIGSEGV, SIGBUS) && // See 4355769. 4333 sig < NSIG) { // Must be legal signal and fit into sigflags[]. 4334 SR_signum = sig; 4335 } else { 4336 warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.", 4337 sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum); 4338 } 4339 } 4340 4341 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4342 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4343 4344 sigemptyset(&SR_sigset); 4345 sigaddset(&SR_sigset, SR_signum); 4346 4347 // Set up signal handler for suspend/resume 4348 act.sa_flags = SA_RESTART|SA_SIGINFO; 4349 act.sa_handler = (void (*)(int)) SR_handler; 4350 4351 // SR_signum is blocked by default. 4352 // 4528190 - We also need to block pthread restart signal (32 on all 4353 // supported Linux platforms). Note that LinuxThreads need to block 4354 // this signal for all threads to work properly. So we don't have 4355 // to use hard-coded signal number when setting up the mask. 4356 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4357 4358 if (sigaction(SR_signum, &act, 0) == -1) { 4359 return -1; 4360 } 4361 4362 // Save signal flag 4363 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4364 return 0; 4365 } 4366 4367 static int sr_notify(OSThread* osthread) { 4368 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4369 assert_status(status == 0, status, "pthread_kill"); 4370 return status; 4371 } 4372 4373 // "Randomly" selected value for how long we want to spin 4374 // before bailing out on suspending a thread, also how often 4375 // we send a signal to a thread we want to resume 4376 static const int RANDOMLY_LARGE_INTEGER = 1000000; 4377 static const int RANDOMLY_LARGE_INTEGER2 = 100; 4378 4379 // returns true on success and false on error - really an error is fatal 4380 // but this seems the normal response to library errors 4381 static bool do_suspend(OSThread* osthread) { 4382 assert(osthread->sr.is_running(), "thread should be running"); 4383 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4384 4385 // mark as suspended and send signal 4386 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4387 // failed to switch, state wasn't running? 4388 ShouldNotReachHere(); 4389 return false; 4390 } 4391 4392 if (sr_notify(osthread) != 0) { 4393 ShouldNotReachHere(); 4394 } 4395 4396 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4397 while (true) { 4398 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) { 4399 break; 4400 } else { 4401 // timeout 4402 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4403 if (cancelled == os::SuspendResume::SR_RUNNING) { 4404 return false; 4405 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4406 // make sure that we consume the signal on the semaphore as well 4407 sr_semaphore.wait(); 4408 break; 4409 } else { 4410 ShouldNotReachHere(); 4411 return false; 4412 } 4413 } 4414 } 4415 4416 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4417 return true; 4418 } 4419 4420 static void do_resume(OSThread* osthread) { 4421 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4422 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4423 4424 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4425 // failed to switch to WAKEUP_REQUEST 4426 ShouldNotReachHere(); 4427 return; 4428 } 4429 4430 while (true) { 4431 if (sr_notify(osthread) == 0) { 4432 if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) { 4433 if (osthread->sr.is_running()) { 4434 return; 4435 } 4436 } 4437 } else { 4438 ShouldNotReachHere(); 4439 } 4440 } 4441 4442 guarantee(osthread->sr.is_running(), "Must be running!"); 4443 } 4444 4445 /////////////////////////////////////////////////////////////////////////////////// 4446 // signal handling (except suspend/resume) 4447 4448 // This routine may be used by user applications as a "hook" to catch signals. 4449 // The user-defined signal handler must pass unrecognized signals to this 4450 // routine, and if it returns true (non-zero), then the signal handler must 4451 // return immediately. If the flag "abort_if_unrecognized" is true, then this 4452 // routine will never retun false (zero), but instead will execute a VM panic 4453 // routine kill the process. 4454 // 4455 // If this routine returns false, it is OK to call it again. This allows 4456 // the user-defined signal handler to perform checks either before or after 4457 // the VM performs its own checks. Naturally, the user code would be making 4458 // a serious error if it tried to handle an exception (such as a null check 4459 // or breakpoint) that the VM was generating for its own correct operation. 4460 // 4461 // This routine may recognize any of the following kinds of signals: 4462 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4463 // It should be consulted by handlers for any of those signals. 4464 // 4465 // The caller of this routine must pass in the three arguments supplied 4466 // to the function referred to in the "sa_sigaction" (not the "sa_handler") 4467 // field of the structure passed to sigaction(). This routine assumes that 4468 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4469 // 4470 // Note that the VM will print warnings if it detects conflicting signal 4471 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4472 // 4473 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4474 siginfo_t* siginfo, 4475 void* ucontext, 4476 int abort_if_unrecognized); 4477 4478 static void signalHandler(int sig, siginfo_t* info, void* uc) { 4479 assert(info != NULL && uc != NULL, "it must be old kernel"); 4480 int orig_errno = errno; // Preserve errno value over signal handler. 4481 JVM_handle_linux_signal(sig, info, uc, true); 4482 errno = orig_errno; 4483 } 4484 4485 4486 // This boolean allows users to forward their own non-matching signals 4487 // to JVM_handle_linux_signal, harmlessly. 4488 bool os::Linux::signal_handlers_are_installed = false; 4489 4490 // For signal-chaining 4491 struct sigaction sigact[NSIG]; 4492 uint64_t sigs = 0; 4493 #if (64 < NSIG-1) 4494 #error "Not all signals can be encoded in sigs. Adapt its type!" 4495 #endif 4496 bool os::Linux::libjsig_is_loaded = false; 4497 typedef struct sigaction *(*get_signal_t)(int); 4498 get_signal_t os::Linux::get_signal_action = NULL; 4499 4500 struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4501 struct sigaction *actp = NULL; 4502 4503 if (libjsig_is_loaded) { 4504 // Retrieve the old signal handler from libjsig 4505 actp = (*get_signal_action)(sig); 4506 } 4507 if (actp == NULL) { 4508 // Retrieve the preinstalled signal handler from jvm 4509 actp = get_preinstalled_handler(sig); 4510 } 4511 4512 return actp; 4513 } 4514 4515 static bool call_chained_handler(struct sigaction *actp, int sig, 4516 siginfo_t *siginfo, void *context) { 4517 // Call the old signal handler 4518 if (actp->sa_handler == SIG_DFL) { 4519 // It's more reasonable to let jvm treat it as an unexpected exception 4520 // instead of taking the default action. 4521 return false; 4522 } else if (actp->sa_handler != SIG_IGN) { 4523 if ((actp->sa_flags & SA_NODEFER) == 0) { 4524 // automaticlly block the signal 4525 sigaddset(&(actp->sa_mask), sig); 4526 } 4527 4528 sa_handler_t hand = NULL; 4529 sa_sigaction_t sa = NULL; 4530 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4531 // retrieve the chained handler 4532 if (siginfo_flag_set) { 4533 sa = actp->sa_sigaction; 4534 } else { 4535 hand = actp->sa_handler; 4536 } 4537 4538 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4539 actp->sa_handler = SIG_DFL; 4540 } 4541 4542 // try to honor the signal mask 4543 sigset_t oset; 4544 sigemptyset(&oset); 4545 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4546 4547 // call into the chained handler 4548 if (siginfo_flag_set) { 4549 (*sa)(sig, siginfo, context); 4550 } else { 4551 (*hand)(sig); 4552 } 4553 4554 // restore the signal mask 4555 pthread_sigmask(SIG_SETMASK, &oset, NULL); 4556 } 4557 // Tell jvm's signal handler the signal is taken care of. 4558 return true; 4559 } 4560 4561 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4562 bool chained = false; 4563 // signal-chaining 4564 if (UseSignalChaining) { 4565 struct sigaction *actp = get_chained_signal_action(sig); 4566 if (actp != NULL) { 4567 chained = call_chained_handler(actp, sig, siginfo, context); 4568 } 4569 } 4570 return chained; 4571 } 4572 4573 struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4574 if ((((uint64_t)1 << (sig-1)) & sigs) != 0) { 4575 return &sigact[sig]; 4576 } 4577 return NULL; 4578 } 4579 4580 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4581 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4582 sigact[sig] = oldAct; 4583 sigs |= (uint64_t)1 << (sig-1); 4584 } 4585 4586 // for diagnostic 4587 int sigflags[NSIG]; 4588 4589 int os::Linux::get_our_sigflags(int sig) { 4590 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4591 return sigflags[sig]; 4592 } 4593 4594 void os::Linux::set_our_sigflags(int sig, int flags) { 4595 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4596 if (sig > 0 && sig < NSIG) { 4597 sigflags[sig] = flags; 4598 } 4599 } 4600 4601 void os::Linux::set_signal_handler(int sig, bool set_installed) { 4602 // Check for overwrite. 4603 struct sigaction oldAct; 4604 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4605 4606 void* oldhand = oldAct.sa_sigaction 4607 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4608 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4609 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4610 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4611 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4612 if (AllowUserSignalHandlers || !set_installed) { 4613 // Do not overwrite; user takes responsibility to forward to us. 4614 return; 4615 } else if (UseSignalChaining) { 4616 // save the old handler in jvm 4617 save_preinstalled_handler(sig, oldAct); 4618 // libjsig also interposes the sigaction() call below and saves the 4619 // old sigaction on it own. 4620 } else { 4621 fatal("Encountered unexpected pre-existing sigaction handler " 4622 "%#lx for signal %d.", (long)oldhand, sig); 4623 } 4624 } 4625 4626 struct sigaction sigAct; 4627 sigfillset(&(sigAct.sa_mask)); 4628 sigAct.sa_handler = SIG_DFL; 4629 if (!set_installed) { 4630 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4631 } else { 4632 sigAct.sa_sigaction = signalHandler; 4633 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4634 } 4635 // Save flags, which are set by ours 4636 assert(sig > 0 && sig < NSIG, "vm signal out of expected range"); 4637 sigflags[sig] = sigAct.sa_flags; 4638 4639 int ret = sigaction(sig, &sigAct, &oldAct); 4640 assert(ret == 0, "check"); 4641 4642 void* oldhand2 = oldAct.sa_sigaction 4643 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4644 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4645 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4646 } 4647 4648 // install signal handlers for signals that HotSpot needs to 4649 // handle in order to support Java-level exception handling. 4650 4651 void os::Linux::install_signal_handlers() { 4652 if (!signal_handlers_are_installed) { 4653 signal_handlers_are_installed = true; 4654 4655 // signal-chaining 4656 typedef void (*signal_setting_t)(); 4657 signal_setting_t begin_signal_setting = NULL; 4658 signal_setting_t end_signal_setting = NULL; 4659 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4660 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4661 if (begin_signal_setting != NULL) { 4662 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4663 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4664 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4665 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4666 libjsig_is_loaded = true; 4667 assert(UseSignalChaining, "should enable signal-chaining"); 4668 } 4669 if (libjsig_is_loaded) { 4670 // Tell libjsig jvm is setting signal handlers 4671 (*begin_signal_setting)(); 4672 } 4673 4674 set_signal_handler(SIGSEGV, true); 4675 set_signal_handler(SIGPIPE, true); 4676 set_signal_handler(SIGBUS, true); 4677 set_signal_handler(SIGILL, true); 4678 set_signal_handler(SIGFPE, true); 4679 #if defined(PPC64) 4680 set_signal_handler(SIGTRAP, true); 4681 #endif 4682 set_signal_handler(SIGXFSZ, true); 4683 4684 if (libjsig_is_loaded) { 4685 // Tell libjsig jvm finishes setting signal handlers 4686 (*end_signal_setting)(); 4687 } 4688 4689 // We don't activate signal checker if libjsig is in place, we trust ourselves 4690 // and if UserSignalHandler is installed all bets are off. 4691 // Log that signal checking is off only if -verbose:jni is specified. 4692 if (CheckJNICalls) { 4693 if (libjsig_is_loaded) { 4694 if (PrintJNIResolving) { 4695 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4696 } 4697 check_signals = false; 4698 } 4699 if (AllowUserSignalHandlers) { 4700 if (PrintJNIResolving) { 4701 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4702 } 4703 check_signals = false; 4704 } 4705 } 4706 } 4707 } 4708 4709 // This is the fastest way to get thread cpu time on Linux. 4710 // Returns cpu time (user+sys) for any thread, not only for current. 4711 // POSIX compliant clocks are implemented in the kernels 2.6.16+. 4712 // It might work on 2.6.10+ with a special kernel/glibc patch. 4713 // For reference, please, see IEEE Std 1003.1-2004: 4714 // http://www.unix.org/single_unix_specification 4715 4716 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4717 struct timespec tp; 4718 int rc = os::Linux::clock_gettime(clockid, &tp); 4719 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4720 4721 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4722 } 4723 4724 void os::Linux::initialize_os_info() { 4725 assert(_os_version == 0, "OS info already initialized"); 4726 4727 struct utsname _uname; 4728 4729 uint32_t major; 4730 uint32_t minor; 4731 uint32_t fix; 4732 4733 int rc; 4734 4735 // Kernel version is unknown if 4736 // verification below fails. 4737 _os_version = 0x01000000; 4738 4739 rc = uname(&_uname); 4740 if (rc != -1) { 4741 4742 rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix); 4743 if (rc == 3) { 4744 4745 if (major < 256 && minor < 256 && fix < 256) { 4746 // Kernel version format is as expected, 4747 // set it overriding unknown state. 4748 _os_version = (major << 16) | 4749 (minor << 8 ) | 4750 (fix << 0 ) ; 4751 } 4752 } 4753 } 4754 } 4755 4756 uint32_t os::Linux::os_version() { 4757 assert(_os_version != 0, "not initialized"); 4758 return _os_version & 0x00FFFFFF; 4759 } 4760 4761 bool os::Linux::os_version_is_known() { 4762 assert(_os_version != 0, "not initialized"); 4763 return _os_version & 0x01000000 ? false : true; 4764 } 4765 4766 ///// 4767 // glibc on Linux platform uses non-documented flag 4768 // to indicate, that some special sort of signal 4769 // trampoline is used. 4770 // We will never set this flag, and we should 4771 // ignore this flag in our diagnostic 4772 #ifdef SIGNIFICANT_SIGNAL_MASK 4773 #undef SIGNIFICANT_SIGNAL_MASK 4774 #endif 4775 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4776 4777 static const char* get_signal_handler_name(address handler, 4778 char* buf, int buflen) { 4779 int offset = 0; 4780 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4781 if (found) { 4782 // skip directory names 4783 const char *p1, *p2; 4784 p1 = buf; 4785 size_t len = strlen(os::file_separator()); 4786 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4787 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4788 } else { 4789 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4790 } 4791 return buf; 4792 } 4793 4794 static void print_signal_handler(outputStream* st, int sig, 4795 char* buf, size_t buflen) { 4796 struct sigaction sa; 4797 4798 sigaction(sig, NULL, &sa); 4799 4800 // See comment for SIGNIFICANT_SIGNAL_MASK define 4801 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4802 4803 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4804 4805 address handler = (sa.sa_flags & SA_SIGINFO) 4806 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4807 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4808 4809 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4810 st->print("SIG_DFL"); 4811 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4812 st->print("SIG_IGN"); 4813 } else { 4814 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4815 } 4816 4817 st->print(", sa_mask[0]="); 4818 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4819 4820 address rh = VMError::get_resetted_sighandler(sig); 4821 // May be, handler was resetted by VMError? 4822 if (rh != NULL) { 4823 handler = rh; 4824 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4825 } 4826 4827 st->print(", sa_flags="); 4828 os::Posix::print_sa_flags(st, sa.sa_flags); 4829 4830 // Check: is it our handler? 4831 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4832 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4833 // It is our signal handler 4834 // check for flags, reset system-used one! 4835 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4836 st->print( 4837 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4838 os::Linux::get_our_sigflags(sig)); 4839 } 4840 } 4841 st->cr(); 4842 } 4843 4844 4845 #define DO_SIGNAL_CHECK(sig) \ 4846 do { \ 4847 if (!sigismember(&check_signal_done, sig)) { \ 4848 os::Linux::check_signal_handler(sig); \ 4849 } \ 4850 } while (0) 4851 4852 // This method is a periodic task to check for misbehaving JNI applications 4853 // under CheckJNI, we can add any periodic checks here 4854 4855 void os::run_periodic_checks() { 4856 if (check_signals == false) return; 4857 4858 // SEGV and BUS if overridden could potentially prevent 4859 // generation of hs*.log in the event of a crash, debugging 4860 // such a case can be very challenging, so we absolutely 4861 // check the following for a good measure: 4862 DO_SIGNAL_CHECK(SIGSEGV); 4863 DO_SIGNAL_CHECK(SIGILL); 4864 DO_SIGNAL_CHECK(SIGFPE); 4865 DO_SIGNAL_CHECK(SIGBUS); 4866 DO_SIGNAL_CHECK(SIGPIPE); 4867 DO_SIGNAL_CHECK(SIGXFSZ); 4868 #if defined(PPC64) 4869 DO_SIGNAL_CHECK(SIGTRAP); 4870 #endif 4871 4872 // ReduceSignalUsage allows the user to override these handlers 4873 // see comments at the very top and jvm_md.h 4874 if (!ReduceSignalUsage) { 4875 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4876 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4877 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4878 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4879 } 4880 4881 DO_SIGNAL_CHECK(SR_signum); 4882 } 4883 4884 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4885 4886 static os_sigaction_t os_sigaction = NULL; 4887 4888 void os::Linux::check_signal_handler(int sig) { 4889 char buf[O_BUFLEN]; 4890 address jvmHandler = NULL; 4891 4892 4893 struct sigaction act; 4894 if (os_sigaction == NULL) { 4895 // only trust the default sigaction, in case it has been interposed 4896 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4897 if (os_sigaction == NULL) return; 4898 } 4899 4900 os_sigaction(sig, (struct sigaction*)NULL, &act); 4901 4902 4903 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4904 4905 address thisHandler = (act.sa_flags & SA_SIGINFO) 4906 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4907 : CAST_FROM_FN_PTR(address, act.sa_handler); 4908 4909 4910 switch (sig) { 4911 case SIGSEGV: 4912 case SIGBUS: 4913 case SIGFPE: 4914 case SIGPIPE: 4915 case SIGILL: 4916 case SIGXFSZ: 4917 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4918 break; 4919 4920 case SHUTDOWN1_SIGNAL: 4921 case SHUTDOWN2_SIGNAL: 4922 case SHUTDOWN3_SIGNAL: 4923 case BREAK_SIGNAL: 4924 jvmHandler = (address)user_handler(); 4925 break; 4926 4927 default: 4928 if (sig == SR_signum) { 4929 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4930 } else { 4931 return; 4932 } 4933 break; 4934 } 4935 4936 if (thisHandler != jvmHandler) { 4937 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4938 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4939 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4940 // No need to check this sig any longer 4941 sigaddset(&check_signal_done, sig); 4942 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4943 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4944 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4945 exception_name(sig, buf, O_BUFLEN)); 4946 } 4947 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4948 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4949 tty->print("expected:"); 4950 os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig)); 4951 tty->cr(); 4952 tty->print(" found:"); 4953 os::Posix::print_sa_flags(tty, act.sa_flags); 4954 tty->cr(); 4955 // No need to check this sig any longer 4956 sigaddset(&check_signal_done, sig); 4957 } 4958 4959 // Dump all the signal 4960 if (sigismember(&check_signal_done, sig)) { 4961 print_signal_handlers(tty, buf, O_BUFLEN); 4962 } 4963 } 4964 4965 extern void report_error(char* file_name, int line_no, char* title, 4966 char* format, ...); 4967 4968 // this is called _before_ most of the global arguments have been parsed 4969 void os::init(void) { 4970 char dummy; // used to get a guess on initial stack address 4971 4972 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4973 4974 init_random(1234567); 4975 4976 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4977 if (Linux::page_size() == -1) { 4978 fatal("os_linux.cpp: os::init: sysconf failed (%s)", 4979 os::strerror(errno)); 4980 } 4981 init_page_sizes((size_t) Linux::page_size()); 4982 4983 Linux::initialize_system_info(); 4984 4985 Linux::initialize_os_info(); 4986 4987 // _main_thread points to the thread that created/loaded the JVM. 4988 Linux::_main_thread = pthread_self(); 4989 4990 Linux::clock_init(); 4991 initial_time_count = javaTimeNanos(); 4992 4993 // retrieve entry point for pthread_setname_np 4994 Linux::_pthread_setname_np = 4995 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4996 4997 os::Posix::init(); 4998 } 4999 5000 // To install functions for atexit system call 5001 extern "C" { 5002 static void perfMemory_exit_helper() { 5003 perfMemory_exit(); 5004 } 5005 } 5006 5007 void os::pd_init_container_support() { 5008 OSContainer::init(); 5009 } 5010 5011 // this is called _after_ the global arguments have been parsed 5012 jint os::init_2(void) { 5013 5014 os::Posix::init_2(); 5015 5016 Linux::fast_thread_clock_init(); 5017 5018 // initialize suspend/resume support - must do this before signal_sets_init() 5019 if (SR_initialize() != 0) { 5020 perror("SR_initialize failed"); 5021 return JNI_ERR; 5022 } 5023 5024 Linux::signal_sets_init(); 5025 Linux::install_signal_handlers(); 5026 // Initialize data for jdk.internal.misc.Signal 5027 if (!ReduceSignalUsage) { 5028 jdk_misc_signal_init(); 5029 } 5030 5031 // Check and sets minimum stack sizes against command line options 5032 if (Posix::set_minimum_stack_sizes() == JNI_ERR) { 5033 return JNI_ERR; 5034 } 5035 5036 suppress_primordial_thread_resolution = Arguments::created_by_java_launcher(); 5037 if (!suppress_primordial_thread_resolution) { 5038 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 5039 } 5040 5041 #if defined(IA32) 5042 workaround_expand_exec_shield_cs_limit(); 5043 #endif 5044 5045 Linux::libpthread_init(); 5046 Linux::sched_getcpu_init(); 5047 log_info(os)("HotSpot is running with %s, %s", 5048 Linux::glibc_version(), Linux::libpthread_version()); 5049 5050 if (UseNUMA) { 5051 if (!Linux::libnuma_init()) { 5052 UseNUMA = false; 5053 } else { 5054 if ((Linux::numa_max_node() < 1) || Linux::isbound_to_single_node()) { 5055 // If there's only one node (they start from 0) or if the process 5056 // is bound explicitly to a single node using membind, disable NUMA. 5057 UseNUMA = false; 5058 } 5059 } 5060 5061 if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 5062 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 5063 // we can make the adaptive lgrp chunk resizing work. If the user specified both 5064 // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn 5065 // and disable adaptive resizing. 5066 if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) { 5067 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, " 5068 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)"); 5069 UseAdaptiveSizePolicy = false; 5070 UseAdaptiveNUMAChunkSizing = false; 5071 } 5072 } 5073 5074 if (!UseNUMA && ForceNUMA) { 5075 UseNUMA = true; 5076 } 5077 } 5078 5079 if (MaxFDLimit) { 5080 // set the number of file descriptors to max. print out error 5081 // if getrlimit/setrlimit fails but continue regardless. 5082 struct rlimit nbr_files; 5083 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 5084 if (status != 0) { 5085 log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno)); 5086 } else { 5087 nbr_files.rlim_cur = nbr_files.rlim_max; 5088 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 5089 if (status != 0) { 5090 log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno)); 5091 } 5092 } 5093 } 5094 5095 // Initialize lock used to serialize thread creation (see os::create_thread) 5096 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 5097 5098 // at-exit methods are called in the reverse order of their registration. 5099 // atexit functions are called on return from main or as a result of a 5100 // call to exit(3C). There can be only 32 of these functions registered 5101 // and atexit() does not set errno. 5102 5103 if (PerfAllowAtExitRegistration) { 5104 // only register atexit functions if PerfAllowAtExitRegistration is set. 5105 // atexit functions can be delayed until process exit time, which 5106 // can be problematic for embedded VM situations. Embedded VMs should 5107 // call DestroyJavaVM() to assure that VM resources are released. 5108 5109 // note: perfMemory_exit_helper atexit function may be removed in 5110 // the future if the appropriate cleanup code can be added to the 5111 // VM_Exit VMOperation's doit method. 5112 if (atexit(perfMemory_exit_helper) != 0) { 5113 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 5114 } 5115 } 5116 5117 // initialize thread priority policy 5118 prio_init(); 5119 5120 if (!FLAG_IS_DEFAULT(AllocateHeapAt)) { 5121 set_coredump_filter(false /*largepages*/, true /*dax_shared*/); 5122 } 5123 return JNI_OK; 5124 } 5125 5126 // Mark the polling page as unreadable 5127 void os::make_polling_page_unreadable(void) { 5128 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 5129 fatal("Could not disable polling page"); 5130 } 5131 } 5132 5133 // Mark the polling page as readable 5134 void os::make_polling_page_readable(void) { 5135 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 5136 fatal("Could not enable polling page"); 5137 } 5138 } 5139 5140 // older glibc versions don't have this macro (which expands to 5141 // an optimized bit-counting function) so we have to roll our own 5142 #ifndef CPU_COUNT 5143 5144 static int _cpu_count(const cpu_set_t* cpus) { 5145 int count = 0; 5146 // only look up to the number of configured processors 5147 for (int i = 0; i < os::processor_count(); i++) { 5148 if (CPU_ISSET(i, cpus)) { 5149 count++; 5150 } 5151 } 5152 return count; 5153 } 5154 5155 #define CPU_COUNT(cpus) _cpu_count(cpus) 5156 5157 #endif // CPU_COUNT 5158 5159 // Get the current number of available processors for this process. 5160 // This value can change at any time during a process's lifetime. 5161 // sched_getaffinity gives an accurate answer as it accounts for cpusets. 5162 // If it appears there may be more than 1024 processors then we do a 5163 // dynamic check - see 6515172 for details. 5164 // If anything goes wrong we fallback to returning the number of online 5165 // processors - which can be greater than the number available to the process. 5166 int os::Linux::active_processor_count() { 5167 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors 5168 cpu_set_t* cpus_p = &cpus; 5169 int cpus_size = sizeof(cpu_set_t); 5170 5171 int configured_cpus = os::processor_count(); // upper bound on available cpus 5172 int cpu_count = 0; 5173 5174 // old build platforms may not support dynamic cpu sets 5175 #ifdef CPU_ALLOC 5176 5177 // To enable easy testing of the dynamic path on different platforms we 5178 // introduce a diagnostic flag: UseCpuAllocPath 5179 if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) { 5180 // kernel may use a mask bigger than cpu_set_t 5181 log_trace(os)("active_processor_count: using dynamic path %s" 5182 "- configured processors: %d", 5183 UseCpuAllocPath ? "(forced) " : "", 5184 configured_cpus); 5185 cpus_p = CPU_ALLOC(configured_cpus); 5186 if (cpus_p != NULL) { 5187 cpus_size = CPU_ALLOC_SIZE(configured_cpus); 5188 // zero it just to be safe 5189 CPU_ZERO_S(cpus_size, cpus_p); 5190 } 5191 else { 5192 // failed to allocate so fallback to online cpus 5193 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 5194 log_trace(os)("active_processor_count: " 5195 "CPU_ALLOC failed (%s) - using " 5196 "online processor count: %d", 5197 os::strerror(errno), online_cpus); 5198 return online_cpus; 5199 } 5200 } 5201 else { 5202 log_trace(os)("active_processor_count: using static path - configured processors: %d", 5203 configured_cpus); 5204 } 5205 #else // CPU_ALLOC 5206 // these stubs won't be executed 5207 #define CPU_COUNT_S(size, cpus) -1 5208 #define CPU_FREE(cpus) 5209 5210 log_trace(os)("active_processor_count: only static path available - configured processors: %d", 5211 configured_cpus); 5212 #endif // CPU_ALLOC 5213 5214 // pid 0 means the current thread - which we have to assume represents the process 5215 if (sched_getaffinity(0, cpus_size, cpus_p) == 0) { 5216 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5217 cpu_count = CPU_COUNT_S(cpus_size, cpus_p); 5218 } 5219 else { 5220 cpu_count = CPU_COUNT(cpus_p); 5221 } 5222 log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count); 5223 } 5224 else { 5225 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN); 5226 warning("sched_getaffinity failed (%s)- using online processor count (%d) " 5227 "which may exceed available processors", os::strerror(errno), cpu_count); 5228 } 5229 5230 if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used 5231 CPU_FREE(cpus_p); 5232 } 5233 5234 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check"); 5235 return cpu_count; 5236 } 5237 5238 // Determine the active processor count from one of 5239 // three different sources: 5240 // 5241 // 1. User option -XX:ActiveProcessorCount 5242 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN) 5243 // 3. extracted from cgroup cpu subsystem (shares and quotas) 5244 // 5245 // Option 1, if specified, will always override. 5246 // If the cgroup subsystem is active and configured, we 5247 // will return the min of the cgroup and option 2 results. 5248 // This is required since tools, such as numactl, that 5249 // alter cpu affinity do not update cgroup subsystem 5250 // cpuset configuration files. 5251 int os::active_processor_count() { 5252 // User has overridden the number of active processors 5253 if (ActiveProcessorCount > 0) { 5254 log_trace(os)("active_processor_count: " 5255 "active processor count set by user : %d", 5256 ActiveProcessorCount); 5257 return ActiveProcessorCount; 5258 } 5259 5260 int active_cpus; 5261 if (OSContainer::is_containerized()) { 5262 active_cpus = OSContainer::active_processor_count(); 5263 log_trace(os)("active_processor_count: determined by OSContainer: %d", 5264 active_cpus); 5265 } else { 5266 active_cpus = os::Linux::active_processor_count(); 5267 } 5268 5269 return active_cpus; 5270 } 5271 5272 uint os::processor_id() { 5273 const int id = Linux::sched_getcpu(); 5274 assert(id >= 0 && id < _processor_count, "Invalid processor id"); 5275 return (uint)id; 5276 } 5277 5278 void os::set_native_thread_name(const char *name) { 5279 if (Linux::_pthread_setname_np) { 5280 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 5281 snprintf(buf, sizeof(buf), "%s", name); 5282 buf[sizeof(buf) - 1] = '\0'; 5283 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 5284 // ERANGE should not happen; all other errors should just be ignored. 5285 assert(rc != ERANGE, "pthread_setname_np failed"); 5286 } 5287 } 5288 5289 bool os::distribute_processes(uint length, uint* distribution) { 5290 // Not yet implemented. 5291 return false; 5292 } 5293 5294 bool os::bind_to_processor(uint processor_id) { 5295 // Not yet implemented. 5296 return false; 5297 } 5298 5299 /// 5300 5301 void os::SuspendedThreadTask::internal_do_task() { 5302 if (do_suspend(_thread->osthread())) { 5303 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 5304 do_task(context); 5305 do_resume(_thread->osthread()); 5306 } 5307 } 5308 5309 //////////////////////////////////////////////////////////////////////////////// 5310 // debug support 5311 5312 bool os::find(address addr, outputStream* st) { 5313 Dl_info dlinfo; 5314 memset(&dlinfo, 0, sizeof(dlinfo)); 5315 if (dladdr(addr, &dlinfo) != 0) { 5316 st->print(PTR_FORMAT ": ", p2i(addr)); 5317 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5318 st->print("%s+" PTR_FORMAT, dlinfo.dli_sname, 5319 p2i(addr) - p2i(dlinfo.dli_saddr)); 5320 } else if (dlinfo.dli_fbase != NULL) { 5321 st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase)); 5322 } else { 5323 st->print("<absolute address>"); 5324 } 5325 if (dlinfo.dli_fname != NULL) { 5326 st->print(" in %s", dlinfo.dli_fname); 5327 } 5328 if (dlinfo.dli_fbase != NULL) { 5329 st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase)); 5330 } 5331 st->cr(); 5332 5333 if (Verbose) { 5334 // decode some bytes around the PC 5335 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5336 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5337 address lowest = (address) dlinfo.dli_sname; 5338 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5339 if (begin < lowest) begin = lowest; 5340 Dl_info dlinfo2; 5341 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5342 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 5343 end = (address) dlinfo2.dli_saddr; 5344 } 5345 Disassembler::decode(begin, end, st); 5346 } 5347 return true; 5348 } 5349 return false; 5350 } 5351 5352 //////////////////////////////////////////////////////////////////////////////// 5353 // misc 5354 5355 // This does not do anything on Linux. This is basically a hook for being 5356 // able to use structured exception handling (thread-local exception filters) 5357 // on, e.g., Win32. 5358 void 5359 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method, 5360 JavaCallArguments* args, Thread* thread) { 5361 f(value, method, args, thread); 5362 } 5363 5364 void os::print_statistics() { 5365 } 5366 5367 bool os::message_box(const char* title, const char* message) { 5368 int i; 5369 fdStream err(defaultStream::error_fd()); 5370 for (i = 0; i < 78; i++) err.print_raw("="); 5371 err.cr(); 5372 err.print_raw_cr(title); 5373 for (i = 0; i < 78; i++) err.print_raw("-"); 5374 err.cr(); 5375 err.print_raw_cr(message); 5376 for (i = 0; i < 78; i++) err.print_raw("="); 5377 err.cr(); 5378 5379 char buf[16]; 5380 // Prevent process from exiting upon "read error" without consuming all CPU 5381 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5382 5383 return buf[0] == 'y' || buf[0] == 'Y'; 5384 } 5385 5386 // Is a (classpath) directory empty? 5387 bool os::dir_is_empty(const char* path) { 5388 DIR *dir = NULL; 5389 struct dirent *ptr; 5390 5391 dir = opendir(path); 5392 if (dir == NULL) return true; 5393 5394 // Scan the directory 5395 bool result = true; 5396 while (result && (ptr = readdir(dir)) != NULL) { 5397 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5398 result = false; 5399 } 5400 } 5401 closedir(dir); 5402 return result; 5403 } 5404 5405 // This code originates from JDK's sysOpen and open64_w 5406 // from src/solaris/hpi/src/system_md.c 5407 5408 int os::open(const char *path, int oflag, int mode) { 5409 if (strlen(path) > MAX_PATH - 1) { 5410 errno = ENAMETOOLONG; 5411 return -1; 5412 } 5413 5414 // All file descriptors that are opened in the Java process and not 5415 // specifically destined for a subprocess should have the close-on-exec 5416 // flag set. If we don't set it, then careless 3rd party native code 5417 // might fork and exec without closing all appropriate file descriptors 5418 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5419 // turn might: 5420 // 5421 // - cause end-of-file to fail to be detected on some file 5422 // descriptors, resulting in mysterious hangs, or 5423 // 5424 // - might cause an fopen in the subprocess to fail on a system 5425 // suffering from bug 1085341. 5426 // 5427 // (Yes, the default setting of the close-on-exec flag is a Unix 5428 // design flaw) 5429 // 5430 // See: 5431 // 1085341: 32-bit stdio routines should support file descriptors >255 5432 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5433 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5434 // 5435 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5436 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5437 // because it saves a system call and removes a small window where the flag 5438 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5439 // and we fall back to using FD_CLOEXEC (see below). 5440 #ifdef O_CLOEXEC 5441 oflag |= O_CLOEXEC; 5442 #endif 5443 5444 int fd = ::open64(path, oflag, mode); 5445 if (fd == -1) return -1; 5446 5447 //If the open succeeded, the file might still be a directory 5448 { 5449 struct stat64 buf64; 5450 int ret = ::fstat64(fd, &buf64); 5451 int st_mode = buf64.st_mode; 5452 5453 if (ret != -1) { 5454 if ((st_mode & S_IFMT) == S_IFDIR) { 5455 errno = EISDIR; 5456 ::close(fd); 5457 return -1; 5458 } 5459 } else { 5460 ::close(fd); 5461 return -1; 5462 } 5463 } 5464 5465 #ifdef FD_CLOEXEC 5466 // Validate that the use of the O_CLOEXEC flag on open above worked. 5467 // With recent kernels, we will perform this check exactly once. 5468 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5469 if (!O_CLOEXEC_is_known_to_work) { 5470 int flags = ::fcntl(fd, F_GETFD); 5471 if (flags != -1) { 5472 if ((flags & FD_CLOEXEC) != 0) 5473 O_CLOEXEC_is_known_to_work = 1; 5474 else 5475 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5476 } 5477 } 5478 #endif 5479 5480 return fd; 5481 } 5482 5483 5484 // create binary file, rewriting existing file if required 5485 int os::create_binary_file(const char* path, bool rewrite_existing) { 5486 int oflags = O_WRONLY | O_CREAT; 5487 if (!rewrite_existing) { 5488 oflags |= O_EXCL; 5489 } 5490 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5491 } 5492 5493 // return current position of file pointer 5494 jlong os::current_file_offset(int fd) { 5495 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5496 } 5497 5498 // move file pointer to the specified offset 5499 jlong os::seek_to_file_offset(int fd, jlong offset) { 5500 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5501 } 5502 5503 // This code originates from JDK's sysAvailable 5504 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5505 5506 int os::available(int fd, jlong *bytes) { 5507 jlong cur, end; 5508 int mode; 5509 struct stat64 buf64; 5510 5511 if (::fstat64(fd, &buf64) >= 0) { 5512 mode = buf64.st_mode; 5513 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5514 int n; 5515 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5516 *bytes = n; 5517 return 1; 5518 } 5519 } 5520 } 5521 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5522 return 0; 5523 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5524 return 0; 5525 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5526 return 0; 5527 } 5528 *bytes = end - cur; 5529 return 1; 5530 } 5531 5532 // Map a block of memory. 5533 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5534 char *addr, size_t bytes, bool read_only, 5535 bool allow_exec) { 5536 int prot; 5537 int flags = MAP_PRIVATE; 5538 5539 if (read_only) { 5540 prot = PROT_READ; 5541 } else { 5542 prot = PROT_READ | PROT_WRITE; 5543 } 5544 5545 if (allow_exec) { 5546 prot |= PROT_EXEC; 5547 } 5548 5549 if (addr != NULL) { 5550 flags |= MAP_FIXED; 5551 } 5552 5553 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5554 fd, file_offset); 5555 if (mapped_address == MAP_FAILED) { 5556 return NULL; 5557 } 5558 return mapped_address; 5559 } 5560 5561 5562 // Remap a block of memory. 5563 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5564 char *addr, size_t bytes, bool read_only, 5565 bool allow_exec) { 5566 // same as map_memory() on this OS 5567 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5568 allow_exec); 5569 } 5570 5571 5572 // Unmap a block of memory. 5573 bool os::pd_unmap_memory(char* addr, size_t bytes) { 5574 return munmap(addr, bytes) == 0; 5575 } 5576 5577 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5578 5579 static jlong fast_cpu_time(Thread *thread) { 5580 clockid_t clockid; 5581 int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(), 5582 &clockid); 5583 if (rc == 0) { 5584 return os::Linux::fast_thread_cpu_time(clockid); 5585 } else { 5586 // It's possible to encounter a terminated native thread that failed 5587 // to detach itself from the VM - which should result in ESRCH. 5588 assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed"); 5589 return -1; 5590 } 5591 } 5592 5593 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5594 // are used by JVM M&M and JVMTI to get user+sys or user CPU time 5595 // of a thread. 5596 // 5597 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5598 // the fast estimate available on the platform. 5599 5600 jlong os::current_thread_cpu_time() { 5601 if (os::Linux::supports_fast_thread_cpu_time()) { 5602 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5603 } else { 5604 // return user + sys since the cost is the same 5605 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5606 } 5607 } 5608 5609 jlong os::thread_cpu_time(Thread* thread) { 5610 // consistent with what current_thread_cpu_time() returns 5611 if (os::Linux::supports_fast_thread_cpu_time()) { 5612 return fast_cpu_time(thread); 5613 } else { 5614 return slow_thread_cpu_time(thread, true /* user + sys */); 5615 } 5616 } 5617 5618 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5619 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5620 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5621 } else { 5622 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5623 } 5624 } 5625 5626 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5627 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5628 return fast_cpu_time(thread); 5629 } else { 5630 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5631 } 5632 } 5633 5634 // -1 on error. 5635 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5636 pid_t tid = thread->osthread()->thread_id(); 5637 char *s; 5638 char stat[2048]; 5639 int statlen; 5640 char proc_name[64]; 5641 int count; 5642 long sys_time, user_time; 5643 char cdummy; 5644 int idummy; 5645 long ldummy; 5646 FILE *fp; 5647 5648 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5649 fp = fopen(proc_name, "r"); 5650 if (fp == NULL) return -1; 5651 statlen = fread(stat, 1, 2047, fp); 5652 stat[statlen] = '\0'; 5653 fclose(fp); 5654 5655 // Skip pid and the command string. Note that we could be dealing with 5656 // weird command names, e.g. user could decide to rename java launcher 5657 // to "java 1.4.2 :)", then the stat file would look like 5658 // 1234 (java 1.4.2 :)) R ... ... 5659 // We don't really need to know the command string, just find the last 5660 // occurrence of ")" and then start parsing from there. See bug 4726580. 5661 s = strrchr(stat, ')'); 5662 if (s == NULL) return -1; 5663 5664 // Skip blank chars 5665 do { s++; } while (s && isspace(*s)); 5666 5667 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5668 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5669 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5670 &user_time, &sys_time); 5671 if (count != 13) return -1; 5672 if (user_sys_cpu_time) { 5673 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5674 } else { 5675 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5676 } 5677 } 5678 5679 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5680 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5681 info_ptr->may_skip_backward = false; // elapsed time not wall time 5682 info_ptr->may_skip_forward = false; // elapsed time not wall time 5683 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5684 } 5685 5686 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5687 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5688 info_ptr->may_skip_backward = false; // elapsed time not wall time 5689 info_ptr->may_skip_forward = false; // elapsed time not wall time 5690 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5691 } 5692 5693 bool os::is_thread_cpu_time_supported() { 5694 return true; 5695 } 5696 5697 // System loadavg support. Returns -1 if load average cannot be obtained. 5698 // Linux doesn't yet have a (official) notion of processor sets, 5699 // so just return the system wide load average. 5700 int os::loadavg(double loadavg[], int nelem) { 5701 return ::getloadavg(loadavg, nelem); 5702 } 5703 5704 void os::pause() { 5705 char filename[MAX_PATH]; 5706 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5707 jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile); 5708 } else { 5709 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5710 } 5711 5712 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5713 if (fd != -1) { 5714 struct stat buf; 5715 ::close(fd); 5716 while (::stat(filename, &buf) == 0) { 5717 (void)::poll(NULL, 0, 100); 5718 } 5719 } else { 5720 jio_fprintf(stderr, 5721 "Could not open pause file '%s', continuing immediately.\n", filename); 5722 } 5723 } 5724 5725 extern char** environ; 5726 5727 // Run the specified command in a separate process. Return its exit value, 5728 // or -1 on failure (e.g. can't fork a new process). 5729 // Unlike system(), this function can be called from signal handler. It 5730 // doesn't block SIGINT et al. 5731 int os::fork_and_exec(char* cmd) { 5732 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5733 5734 pid_t pid = fork(); 5735 5736 if (pid < 0) { 5737 // fork failed 5738 return -1; 5739 5740 } else if (pid == 0) { 5741 // child process 5742 5743 execve("/bin/sh", (char* const*)argv, environ); 5744 5745 // execve failed 5746 _exit(-1); 5747 5748 } else { 5749 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5750 // care about the actual exit code, for now. 5751 5752 int status; 5753 5754 // Wait for the child process to exit. This returns immediately if 5755 // the child has already exited. */ 5756 while (waitpid(pid, &status, 0) < 0) { 5757 switch (errno) { 5758 case ECHILD: return 0; 5759 case EINTR: break; 5760 default: return -1; 5761 } 5762 } 5763 5764 if (WIFEXITED(status)) { 5765 // The child exited normally; get its exit code. 5766 return WEXITSTATUS(status); 5767 } else if (WIFSIGNALED(status)) { 5768 // The child exited because of a signal 5769 // The best value to return is 0x80 + signal number, 5770 // because that is what all Unix shells do, and because 5771 // it allows callers to distinguish between process exit and 5772 // process death by signal. 5773 return 0x80 + WTERMSIG(status); 5774 } else { 5775 // Unknown exit code; pass it through 5776 return status; 5777 } 5778 } 5779 } 5780 5781 // Get the default path to the core file 5782 // Returns the length of the string 5783 int os::get_core_path(char* buffer, size_t bufferSize) { 5784 /* 5785 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 5786 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 5787 */ 5788 const int core_pattern_len = 129; 5789 char core_pattern[core_pattern_len] = {0}; 5790 5791 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 5792 if (core_pattern_file == -1) { 5793 return -1; 5794 } 5795 5796 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 5797 ::close(core_pattern_file); 5798 if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') { 5799 return -1; 5800 } 5801 if (core_pattern[ret-1] == '\n') { 5802 core_pattern[ret-1] = '\0'; 5803 } else { 5804 core_pattern[ret] = '\0'; 5805 } 5806 5807 // Replace the %p in the core pattern with the process id. NOTE: we do this 5808 // only if the pattern doesn't start with "|", and we support only one %p in 5809 // the pattern. 5810 char *pid_pos = strstr(core_pattern, "%p"); 5811 const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : ""; // skip over the "%p" 5812 int written; 5813 5814 if (core_pattern[0] == '/') { 5815 if (pid_pos != NULL) { 5816 *pid_pos = '\0'; 5817 written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern, 5818 current_process_id(), tail); 5819 } else { 5820 written = jio_snprintf(buffer, bufferSize, "%s", core_pattern); 5821 } 5822 } else { 5823 char cwd[PATH_MAX]; 5824 5825 const char* p = get_current_directory(cwd, PATH_MAX); 5826 if (p == NULL) { 5827 return -1; 5828 } 5829 5830 if (core_pattern[0] == '|') { 5831 written = jio_snprintf(buffer, bufferSize, 5832 "\"%s\" (or dumping to %s/core.%d)", 5833 &core_pattern[1], p, current_process_id()); 5834 } else if (pid_pos != NULL) { 5835 *pid_pos = '\0'; 5836 written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern, 5837 current_process_id(), tail); 5838 } else { 5839 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 5840 } 5841 } 5842 5843 if (written < 0) { 5844 return -1; 5845 } 5846 5847 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 5848 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 5849 5850 if (core_uses_pid_file != -1) { 5851 char core_uses_pid = 0; 5852 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 5853 ::close(core_uses_pid_file); 5854 5855 if (core_uses_pid == '1') { 5856 jio_snprintf(buffer + written, bufferSize - written, 5857 ".%d", current_process_id()); 5858 } 5859 } 5860 } 5861 5862 return strlen(buffer); 5863 } 5864 5865 bool os::start_debugging(char *buf, int buflen) { 5866 int len = (int)strlen(buf); 5867 char *p = &buf[len]; 5868 5869 jio_snprintf(p, buflen-len, 5870 "\n\n" 5871 "Do you want to debug the problem?\n\n" 5872 "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n" 5873 "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n" 5874 "Otherwise, press RETURN to abort...", 5875 os::current_process_id(), os::current_process_id(), 5876 os::current_thread_id(), os::current_thread_id()); 5877 5878 bool yes = os::message_box("Unexpected Error", buf); 5879 5880 if (yes) { 5881 // yes, user asked VM to launch debugger 5882 jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d", 5883 os::current_process_id(), os::current_process_id()); 5884 5885 os::fork_and_exec(buf); 5886 yes = false; 5887 } 5888 return yes; 5889 } 5890 5891 5892 // Java/Compiler thread: 5893 // 5894 // Low memory addresses 5895 // P0 +------------------------+ 5896 // | |\ Java thread created by VM does not have glibc 5897 // | glibc guard page | - guard page, attached Java thread usually has 5898 // | |/ 1 glibc guard page. 5899 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 5900 // | |\ 5901 // | HotSpot Guard Pages | - red, yellow and reserved pages 5902 // | |/ 5903 // +------------------------+ JavaThread::stack_reserved_zone_base() 5904 // | |\ 5905 // | Normal Stack | - 5906 // | |/ 5907 // P2 +------------------------+ Thread::stack_base() 5908 // 5909 // Non-Java thread: 5910 // 5911 // Low memory addresses 5912 // P0 +------------------------+ 5913 // | |\ 5914 // | glibc guard page | - usually 1 page 5915 // | |/ 5916 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size() 5917 // | |\ 5918 // | Normal Stack | - 5919 // | |/ 5920 // P2 +------------------------+ Thread::stack_base() 5921 // 5922 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size 5923 // returned from pthread_attr_getstack(). 5924 // ** Due to NPTL implementation error, linux takes the glibc guard page out 5925 // of the stack size given in pthread_attr. We work around this for 5926 // threads created by the VM. (We adapt bottom to be P1 and size accordingly.) 5927 // 5928 #ifndef ZERO 5929 static void current_stack_region(address * bottom, size_t * size) { 5930 if (os::is_primordial_thread()) { 5931 // primordial thread needs special handling because pthread_getattr_np() 5932 // may return bogus value. 5933 *bottom = os::Linux::initial_thread_stack_bottom(); 5934 *size = os::Linux::initial_thread_stack_size(); 5935 } else { 5936 pthread_attr_t attr; 5937 5938 int rslt = pthread_getattr_np(pthread_self(), &attr); 5939 5940 // JVM needs to know exact stack location, abort if it fails 5941 if (rslt != 0) { 5942 if (rslt == ENOMEM) { 5943 vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np"); 5944 } else { 5945 fatal("pthread_getattr_np failed with error = %d", rslt); 5946 } 5947 } 5948 5949 if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) { 5950 fatal("Cannot locate current stack attributes!"); 5951 } 5952 5953 // Work around NPTL stack guard error. 5954 size_t guard_size = 0; 5955 rslt = pthread_attr_getguardsize(&attr, &guard_size); 5956 if (rslt != 0) { 5957 fatal("pthread_attr_getguardsize failed with error = %d", rslt); 5958 } 5959 *bottom += guard_size; 5960 *size -= guard_size; 5961 5962 pthread_attr_destroy(&attr); 5963 5964 } 5965 assert(os::current_stack_pointer() >= *bottom && 5966 os::current_stack_pointer() < *bottom + *size, "just checking"); 5967 } 5968 5969 address os::current_stack_base() { 5970 address bottom; 5971 size_t size; 5972 current_stack_region(&bottom, &size); 5973 return (bottom + size); 5974 } 5975 5976 size_t os::current_stack_size() { 5977 // This stack size includes the usable stack and HotSpot guard pages 5978 // (for the threads that have Hotspot guard pages). 5979 address bottom; 5980 size_t size; 5981 current_stack_region(&bottom, &size); 5982 return size; 5983 } 5984 #endif 5985 5986 static inline struct timespec get_mtime(const char* filename) { 5987 struct stat st; 5988 int ret = os::stat(filename, &st); 5989 assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno)); 5990 return st.st_mtim; 5991 } 5992 5993 int os::compare_file_modified_times(const char* file1, const char* file2) { 5994 struct timespec filetime1 = get_mtime(file1); 5995 struct timespec filetime2 = get_mtime(file2); 5996 int diff = filetime1.tv_sec - filetime2.tv_sec; 5997 if (diff == 0) { 5998 return filetime1.tv_nsec - filetime2.tv_nsec; 5999 } 6000 return diff; 6001 } 6002 6003 /////////////// Unit tests /////////////// 6004 6005 #ifndef PRODUCT 6006 6007 #define test_log(...) \ 6008 do { \ 6009 if (VerboseInternalVMTests) { \ 6010 tty->print_cr(__VA_ARGS__); \ 6011 tty->flush(); \ 6012 } \ 6013 } while (false) 6014 6015 class TestReserveMemorySpecial : AllStatic { 6016 public: 6017 static void small_page_write(void* addr, size_t size) { 6018 size_t page_size = os::vm_page_size(); 6019 6020 char* end = (char*)addr + size; 6021 for (char* p = (char*)addr; p < end; p += page_size) { 6022 *p = 1; 6023 } 6024 } 6025 6026 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6027 if (!UseHugeTLBFS) { 6028 return; 6029 } 6030 6031 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6032 6033 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6034 6035 if (addr != NULL) { 6036 small_page_write(addr, size); 6037 6038 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6039 } 6040 } 6041 6042 static void test_reserve_memory_special_huge_tlbfs_only() { 6043 if (!UseHugeTLBFS) { 6044 return; 6045 } 6046 6047 size_t lp = os::large_page_size(); 6048 6049 for (size_t size = lp; size <= lp * 10; size += lp) { 6050 test_reserve_memory_special_huge_tlbfs_only(size); 6051 } 6052 } 6053 6054 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6055 size_t lp = os::large_page_size(); 6056 size_t ag = os::vm_allocation_granularity(); 6057 6058 // sizes to test 6059 const size_t sizes[] = { 6060 lp, lp + ag, lp + lp / 2, lp * 2, 6061 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2, 6062 lp * 10, lp * 10 + lp / 2 6063 }; 6064 const int num_sizes = sizeof(sizes) / sizeof(size_t); 6065 6066 // For each size/alignment combination, we test three scenarios: 6067 // 1) with req_addr == NULL 6068 // 2) with a non-null req_addr at which we expect to successfully allocate 6069 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we 6070 // expect the allocation to either fail or to ignore req_addr 6071 6072 // Pre-allocate two areas; they shall be as large as the largest allocation 6073 // and aligned to the largest alignment we will be testing. 6074 const size_t mapping_size = sizes[num_sizes - 1] * 2; 6075 char* const mapping1 = (char*) ::mmap(NULL, mapping_size, 6076 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6077 -1, 0); 6078 assert(mapping1 != MAP_FAILED, "should work"); 6079 6080 char* const mapping2 = (char*) ::mmap(NULL, mapping_size, 6081 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6082 -1, 0); 6083 assert(mapping2 != MAP_FAILED, "should work"); 6084 6085 // Unmap the first mapping, but leave the second mapping intact: the first 6086 // mapping will serve as a value for a "good" req_addr (case 2). The second 6087 // mapping, still intact, as "bad" req_addr (case 3). 6088 ::munmap(mapping1, mapping_size); 6089 6090 // Case 1 6091 test_log("%s, req_addr NULL:", __FUNCTION__); 6092 test_log("size align result"); 6093 6094 for (int i = 0; i < num_sizes; i++) { 6095 const size_t size = sizes[i]; 6096 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6097 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6098 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s", 6099 size, alignment, p2i(p), (p != NULL ? "" : "(failed)")); 6100 if (p != NULL) { 6101 assert(is_aligned(p, alignment), "must be"); 6102 small_page_write(p, size); 6103 os::Linux::release_memory_special_huge_tlbfs(p, size); 6104 } 6105 } 6106 } 6107 6108 // Case 2 6109 test_log("%s, req_addr non-NULL:", __FUNCTION__); 6110 test_log("size align req_addr result"); 6111 6112 for (int i = 0; i < num_sizes; i++) { 6113 const size_t size = sizes[i]; 6114 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6115 char* const req_addr = align_up(mapping1, alignment); 6116 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6117 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6118 size, alignment, p2i(req_addr), p2i(p), 6119 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)"))); 6120 if (p != NULL) { 6121 assert(p == req_addr, "must be"); 6122 small_page_write(p, size); 6123 os::Linux::release_memory_special_huge_tlbfs(p, size); 6124 } 6125 } 6126 } 6127 6128 // Case 3 6129 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__); 6130 test_log("size align req_addr result"); 6131 6132 for (int i = 0; i < num_sizes; i++) { 6133 const size_t size = sizes[i]; 6134 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6135 char* const req_addr = align_up(mapping2, alignment); 6136 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6137 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6138 size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)"))); 6139 // as the area around req_addr contains already existing mappings, the API should always 6140 // return NULL (as per contract, it cannot return another address) 6141 assert(p == NULL, "must be"); 6142 } 6143 } 6144 6145 ::munmap(mapping2, mapping_size); 6146 6147 } 6148 6149 static void test_reserve_memory_special_huge_tlbfs() { 6150 if (!UseHugeTLBFS) { 6151 return; 6152 } 6153 6154 test_reserve_memory_special_huge_tlbfs_only(); 6155 test_reserve_memory_special_huge_tlbfs_mixed(); 6156 } 6157 6158 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6159 if (!UseSHM) { 6160 return; 6161 } 6162 6163 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6164 6165 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6166 6167 if (addr != NULL) { 6168 assert(is_aligned(addr, alignment), "Check"); 6169 assert(is_aligned(addr, os::large_page_size()), "Check"); 6170 6171 small_page_write(addr, size); 6172 6173 os::Linux::release_memory_special_shm(addr, size); 6174 } 6175 } 6176 6177 static void test_reserve_memory_special_shm() { 6178 size_t lp = os::large_page_size(); 6179 size_t ag = os::vm_allocation_granularity(); 6180 6181 for (size_t size = ag; size < lp * 3; size += ag) { 6182 for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) { 6183 test_reserve_memory_special_shm(size, alignment); 6184 } 6185 } 6186 } 6187 6188 static void test() { 6189 test_reserve_memory_special_huge_tlbfs(); 6190 test_reserve_memory_special_shm(); 6191 } 6192 }; 6193 6194 void TestReserveMemorySpecial_test() { 6195 TestReserveMemorySpecial::test(); 6196 } 6197 6198 #endif