1 /*
   2  * Copyright (c) 1999, 2012, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "jvm_bsd.h"
  34 #include "memory/allocation.inline.hpp"
  35 #include "memory/filemap.hpp"
  36 #include "mutex_bsd.inline.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "os_share_bsd.hpp"
  39 #include "prims/jniFastGetField.hpp"
  40 #include "prims/jvm.h"
  41 #include "prims/jvm_misc.hpp"
  42 #include "runtime/arguments.hpp"
  43 #include "runtime/extendedPC.hpp"
  44 #include "runtime/globals.hpp"
  45 #include "runtime/interfaceSupport.hpp"
  46 #include "runtime/java.hpp"
  47 #include "runtime/javaCalls.hpp"
  48 #include "runtime/mutexLocker.hpp"
  49 #include "runtime/objectMonitor.hpp"
  50 #include "runtime/osThread.hpp"
  51 #include "runtime/perfMemory.hpp"
  52 #include "runtime/sharedRuntime.hpp"
  53 #include "runtime/statSampler.hpp"
  54 #include "runtime/stubRoutines.hpp"
  55 #include "runtime/threadCritical.hpp"
  56 #include "runtime/timer.hpp"
  57 #include "services/attachListener.hpp"
  58 #include "services/runtimeService.hpp"
  59 #include "thread_bsd.inline.hpp"
  60 #include "utilities/decoder.hpp"
  61 #include "utilities/defaultStream.hpp"
  62 #include "utilities/events.hpp"
  63 #include "utilities/growableArray.hpp"
  64 #include "utilities/vmError.hpp"
  65 #ifdef TARGET_ARCH_x86
  66 # include "assembler_x86.inline.hpp"
  67 # include "nativeInst_x86.hpp"
  68 #endif
  69 #ifdef TARGET_ARCH_sparc
  70 # include "assembler_sparc.inline.hpp"
  71 # include "nativeInst_sparc.hpp"
  72 #endif
  73 #ifdef TARGET_ARCH_zero
  74 # include "assembler_zero.inline.hpp"
  75 # include "nativeInst_zero.hpp"
  76 #endif
  77 #ifdef TARGET_ARCH_arm
  78 # include "assembler_arm.inline.hpp"
  79 # include "nativeInst_arm.hpp"
  80 #endif
  81 #ifdef TARGET_ARCH_ppc
  82 # include "assembler_ppc.inline.hpp"
  83 # include "nativeInst_ppc.hpp"
  84 #endif
  85 #ifdef COMPILER1
  86 #include "c1/c1_Runtime1.hpp"
  87 #endif
  88 #ifdef COMPILER2
  89 #include "opto/runtime.hpp"
  90 #endif
  91 
  92 // put OS-includes here
  93 # include <sys/types.h>
  94 # include <sys/mman.h>
  95 # include <sys/stat.h>
  96 # include <sys/select.h>
  97 # include <pthread.h>
  98 # include <signal.h>
  99 # include <errno.h>
 100 # include <dlfcn.h>
 101 # include <stdio.h>
 102 # include <unistd.h>
 103 # include <sys/resource.h>
 104 # include <pthread.h>
 105 # include <sys/stat.h>
 106 # include <sys/time.h>
 107 # include <sys/times.h>
 108 # include <sys/utsname.h>
 109 # include <sys/socket.h>
 110 # include <sys/wait.h>
 111 # include <time.h>
 112 # include <pwd.h>
 113 # include <poll.h>
 114 # include <semaphore.h>
 115 # include <fcntl.h>
 116 # include <string.h>
 117 #ifdef _ALLBSD_SOURCE
 118 # include <sys/param.h>
 119 # include <sys/sysctl.h>
 120 #else
 121 # include <syscall.h>
 122 # include <sys/sysinfo.h>
 123 # include <gnu/libc-version.h>
 124 #endif
 125 # include <sys/ipc.h>
 126 # include <sys/shm.h>
 127 #ifndef __APPLE__
 128 # include <link.h>
 129 #endif
 130 # include <stdint.h>
 131 # include <inttypes.h>
 132 # include <sys/ioctl.h>
 133 
 134 #if defined(__FreeBSD__) || defined(__NetBSD__)
 135 # include <elf.h>
 136 #endif
 137 
 138 #ifdef __APPLE__
 139 # include <mach/mach.h> // semaphore_* API
 140 # include <mach-o/dyld.h>
 141 # include <sys/proc_info.h>
 142 # include <objc/objc-auto.h>
 143 #endif
 144 
 145 #ifndef MAP_ANONYMOUS
 146 #define MAP_ANONYMOUS MAP_ANON
 147 #endif
 148 
 149 #define MAX_PATH    (2 * K)
 150 
 151 // for timer info max values which include all bits
 152 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 153 
 154 #define LARGEPAGES_BIT (1 << 6)
 155 ////////////////////////////////////////////////////////////////////////////////
 156 // global variables
 157 julong os::Bsd::_physical_memory = 0;
 158 
 159 #ifndef _ALLBSD_SOURCE
 160 address   os::Bsd::_initial_thread_stack_bottom = NULL;
 161 uintptr_t os::Bsd::_initial_thread_stack_size   = 0;
 162 #endif
 163 
 164 int (*os::Bsd::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 165 #ifndef _ALLBSD_SOURCE
 166 int (*os::Bsd::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 167 Mutex* os::Bsd::_createThread_lock = NULL;
 168 #endif
 169 pthread_t os::Bsd::_main_thread;
 170 int os::Bsd::_page_size = -1;
 171 #ifndef _ALLBSD_SOURCE
 172 bool os::Bsd::_is_floating_stack = false;
 173 bool os::Bsd::_is_NPTL = false;
 174 bool os::Bsd::_supports_fast_thread_cpu_time = false;
 175 const char * os::Bsd::_glibc_version = NULL;
 176 const char * os::Bsd::_libpthread_version = NULL;
 177 #endif
 178 
 179 static jlong initial_time_count=0;
 180 
 181 static int clock_tics_per_sec = 100;
 182 
 183 // For diagnostics to print a message once. see run_periodic_checks
 184 static sigset_t check_signal_done;
 185 static bool check_signals = true;;
 186 
 187 static pid_t _initial_pid = 0;
 188 
 189 /* Signal number used to suspend/resume a thread */
 190 
 191 /* do not use any signal number less than SIGSEGV, see 4355769 */
 192 static int SR_signum = SIGUSR2;
 193 sigset_t SR_sigset;
 194 
 195 
 196 ////////////////////////////////////////////////////////////////////////////////
 197 // utility functions
 198 
 199 static int SR_initialize();
 200 static int SR_finalize();
 201 
 202 julong os::available_memory() {
 203   return Bsd::available_memory();
 204 }
 205 
 206 julong os::Bsd::available_memory() {
 207 #ifdef _ALLBSD_SOURCE
 208   // XXXBSD: this is just a stopgap implementation
 209   return physical_memory() >> 2;
 210 #else
 211   // values in struct sysinfo are "unsigned long"
 212   struct sysinfo si;
 213   sysinfo(&si);
 214 
 215   return (julong)si.freeram * si.mem_unit;
 216 #endif
 217 }
 218 
 219 julong os::physical_memory() {
 220   return Bsd::physical_memory();
 221 }
 222 
 223 julong os::allocatable_physical_memory(julong size) {
 224 #ifdef _LP64
 225   return size;
 226 #else
 227   julong result = MIN2(size, (julong)3800*M);
 228    if (!is_allocatable(result)) {
 229      // See comments under solaris for alignment considerations
 230      julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
 231      result =  MIN2(size, reasonable_size);
 232    }
 233    return result;
 234 #endif // _LP64
 235 }
 236 
 237 ////////////////////////////////////////////////////////////////////////////////
 238 // environment support
 239 
 240 bool os::getenv(const char* name, char* buf, int len) {
 241   const char* val = ::getenv(name);
 242   if (val != NULL && strlen(val) < (size_t)len) {
 243     strcpy(buf, val);
 244     return true;
 245   }
 246   if (len > 0) buf[0] = 0;  // return a null string
 247   return false;
 248 }
 249 
 250 
 251 // Return true if user is running as root.
 252 
 253 bool os::have_special_privileges() {
 254   static bool init = false;
 255   static bool privileges = false;
 256   if (!init) {
 257     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 258     init = true;
 259   }
 260   return privileges;
 261 }
 262 
 263 
 264 #ifndef _ALLBSD_SOURCE
 265 #ifndef SYS_gettid
 266 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 267 #ifdef __ia64__
 268 #define SYS_gettid 1105
 269 #elif __i386__
 270 #define SYS_gettid 224
 271 #elif __amd64__
 272 #define SYS_gettid 186
 273 #elif __sparc__
 274 #define SYS_gettid 143
 275 #else
 276 #error define gettid for the arch
 277 #endif
 278 #endif
 279 #endif
 280 
 281 // Cpu architecture string
 282 #if   defined(ZERO)
 283 static char cpu_arch[] = ZERO_LIBARCH;
 284 #elif defined(IA64)
 285 static char cpu_arch[] = "ia64";
 286 #elif defined(IA32)
 287 static char cpu_arch[] = "i386";
 288 #elif defined(AMD64)
 289 static char cpu_arch[] = "amd64";
 290 #elif defined(ARM)
 291 static char cpu_arch[] = "arm";
 292 #elif defined(PPC)
 293 static char cpu_arch[] = "ppc";
 294 #elif defined(SPARC)
 295 #  ifdef _LP64
 296 static char cpu_arch[] = "sparcv9";
 297 #  else
 298 static char cpu_arch[] = "sparc";
 299 #  endif
 300 #else
 301 #error Add appropriate cpu_arch setting
 302 #endif
 303 
 304 // Compiler variant
 305 #ifdef COMPILER2
 306 #define COMPILER_VARIANT "server"
 307 #else
 308 #define COMPILER_VARIANT "client"
 309 #endif
 310 
 311 #ifndef _ALLBSD_SOURCE
 312 // pid_t gettid()
 313 //
 314 // Returns the kernel thread id of the currently running thread. Kernel
 315 // thread id is used to access /proc.
 316 //
 317 // (Note that getpid() on BsdThreads returns kernel thread id too; but
 318 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
 319 //
 320 pid_t os::Bsd::gettid() {
 321   int rslt = syscall(SYS_gettid);
 322   if (rslt == -1) {
 323      // old kernel, no NPTL support
 324      return getpid();
 325   } else {
 326      return (pid_t)rslt;
 327   }
 328 }
 329 
 330 // Most versions of bsd have a bug where the number of processors are
 331 // determined by looking at the /proc file system.  In a chroot environment,
 332 // the system call returns 1.  This causes the VM to act as if it is
 333 // a single processor and elide locking (see is_MP() call).
 334 static bool unsafe_chroot_detected = false;
 335 static const char *unstable_chroot_error = "/proc file system not found.\n"
 336                      "Java may be unstable running multithreaded in a chroot "
 337                      "environment on Bsd when /proc filesystem is not mounted.";
 338 #endif
 339 
 340 #ifdef _ALLBSD_SOURCE
 341 void os::Bsd::initialize_system_info() {
 342   int mib[2];
 343   size_t len;
 344   int cpu_val;
 345   u_long mem_val;
 346 
 347   /* get processors count via hw.ncpus sysctl */
 348   mib[0] = CTL_HW;
 349   mib[1] = HW_NCPU;
 350   len = sizeof(cpu_val);
 351   if (sysctl(mib, 2, &cpu_val, &len, NULL, 0) != -1 && cpu_val >= 1) {
 352        set_processor_count(cpu_val);
 353   }
 354   else {
 355        set_processor_count(1);   // fallback
 356   }
 357 
 358   /* get physical memory via hw.usermem sysctl (hw.usermem is used
 359    * instead of hw.physmem because we need size of allocatable memory
 360    */
 361   mib[0] = CTL_HW;
 362   mib[1] = HW_USERMEM;
 363   len = sizeof(mem_val);
 364   if (sysctl(mib, 2, &mem_val, &len, NULL, 0) != -1)
 365        _physical_memory = mem_val;
 366   else
 367        _physical_memory = 256*1024*1024;       // fallback (XXXBSD?)
 368 
 369 #ifdef __OpenBSD__
 370   {
 371        // limit _physical_memory memory view on OpenBSD since
 372        // datasize rlimit restricts us anyway.
 373        struct rlimit limits;
 374        getrlimit(RLIMIT_DATA, &limits);
 375        _physical_memory = MIN2(_physical_memory, (julong)limits.rlim_cur);
 376   }
 377 #endif
 378 }
 379 #else
 380 void os::Bsd::initialize_system_info() {
 381   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 382   if (processor_count() == 1) {
 383     pid_t pid = os::Bsd::gettid();
 384     char fname[32];
 385     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 386     FILE *fp = fopen(fname, "r");
 387     if (fp == NULL) {
 388       unsafe_chroot_detected = true;
 389     } else {
 390       fclose(fp);
 391     }
 392   }
 393   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 394   assert(processor_count() > 0, "bsd error");
 395 }
 396 #endif
 397 
 398 #ifdef __APPLE__
 399 static const char *get_home() {
 400   const char *home_dir = ::getenv("HOME");
 401   if ((home_dir == NULL) || (*home_dir == '\0')) {
 402     struct passwd *passwd_info = getpwuid(geteuid());
 403     if (passwd_info != NULL) {
 404       home_dir = passwd_info->pw_dir;
 405     }
 406   }
 407 
 408   return home_dir;
 409 }
 410 #endif
 411 
 412 void os::init_system_properties_values() {
 413 //  char arch[12];
 414 //  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 415 
 416   // The next steps are taken in the product version:
 417   //
 418   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 419   // This library should be located at:
 420   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 421   //
 422   // If "/jre/lib/" appears at the right place in the path, then we
 423   // assume libjvm[_g].so is installed in a JDK and we use this path.
 424   //
 425   // Otherwise exit with message: "Could not create the Java virtual machine."
 426   //
 427   // The following extra steps are taken in the debugging version:
 428   //
 429   // If "/jre/lib/" does NOT appear at the right place in the path
 430   // instead of exit check for $JAVA_HOME environment variable.
 431   //
 432   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 433   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 434   // it looks like libjvm[_g].so is installed there
 435   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 436   //
 437   // Otherwise exit.
 438   //
 439   // Important note: if the location of libjvm.so changes this
 440   // code needs to be changed accordingly.
 441 
 442   // The next few definitions allow the code to be verbatim:
 443 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 444 #define getenv(n) ::getenv(n)
 445 
 446 /*
 447  * See ld(1):
 448  *      The linker uses the following search paths to locate required
 449  *      shared libraries:
 450  *        1: ...
 451  *        ...
 452  *        7: The default directories, normally /lib and /usr/lib.
 453  */
 454 #ifndef DEFAULT_LIBPATH
 455 #define DEFAULT_LIBPATH "/lib:/usr/lib"
 456 #endif
 457 
 458 #define EXTENSIONS_DIR  "/lib/ext"
 459 #define ENDORSED_DIR    "/lib/endorsed"
 460 #define REG_DIR         "/usr/java/packages"
 461 
 462 #ifdef __APPLE__
 463 #define SYS_EXTENSIONS_DIR   "/Library/Java/Extensions"
 464 #define SYS_EXTENSIONS_DIRS  SYS_EXTENSIONS_DIR ":/Network" SYS_EXTENSIONS_DIR ":/System" SYS_EXTENSIONS_DIR ":/usr/lib/java"
 465         const char *user_home_dir = get_home();
 466         // the null in SYS_EXTENSIONS_DIRS counts for the size of the colon after user_home_dir
 467         int system_ext_size = strlen(user_home_dir) + sizeof(SYS_EXTENSIONS_DIR) +
 468             sizeof(SYS_EXTENSIONS_DIRS);
 469 #endif
 470 
 471   {
 472     /* sysclasspath, java_home, dll_dir */
 473     {
 474         char *home_path;
 475         char *dll_path;
 476         char *pslash;
 477         char buf[MAXPATHLEN];
 478         os::jvm_path(buf, sizeof(buf));
 479 
 480         // Found the full path to libjvm.so.
 481         // Now cut the path to <java_home>/jre if we can.
 482         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 483         pslash = strrchr(buf, '/');
 484         if (pslash != NULL)
 485             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 486         dll_path = malloc(strlen(buf) + 1);
 487         if (dll_path == NULL)
 488             return;
 489         strcpy(dll_path, buf);
 490         Arguments::set_dll_dir(dll_path);
 491 
 492         if (pslash != NULL) {
 493             pslash = strrchr(buf, '/');
 494             if (pslash != NULL) {
 495                 *pslash = '\0';       /* get rid of /<arch> (/lib on macosx) */
 496 #ifndef __APPLE__
 497                 pslash = strrchr(buf, '/');
 498                 if (pslash != NULL)
 499                     *pslash = '\0';   /* get rid of /lib */
 500 #endif
 501             }
 502         }
 503 
 504         home_path = malloc(strlen(buf) + 1);
 505         if (home_path == NULL)
 506             return;
 507         strcpy(home_path, buf);
 508         Arguments::set_java_home(home_path);
 509 
 510         if (!set_boot_path('/', ':'))
 511             return;
 512     }
 513 
 514     /*
 515      * Where to look for native libraries
 516      *
 517      * Note: Due to a legacy implementation, most of the library path
 518      * is set in the launcher.  This was to accomodate linking restrictions
 519      * on legacy Bsd implementations (which are no longer supported).
 520      * Eventually, all the library path setting will be done here.
 521      *
 522      * However, to prevent the proliferation of improperly built native
 523      * libraries, the new path component /usr/java/packages is added here.
 524      * Eventually, all the library path setting will be done here.
 525      */
 526     {
 527         char *ld_library_path;
 528 
 529         /*
 530          * Construct the invariant part of ld_library_path. Note that the
 531          * space for the colon and the trailing null are provided by the
 532          * nulls included by the sizeof operator (so actually we allocate
 533          * a byte more than necessary).
 534          */
 535 #ifdef __APPLE__
 536         ld_library_path = (char *) malloc(system_ext_size);
 537         sprintf(ld_library_path, "%s" SYS_EXTENSIONS_DIR ":" SYS_EXTENSIONS_DIRS, user_home_dir);
 538 #else
 539         ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
 540             strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
 541         sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
 542 #endif
 543 
 544         /*
 545          * Get the user setting of LD_LIBRARY_PATH, and prepended it.  It
 546          * should always exist (until the legacy problem cited above is
 547          * addressed).
 548          */
 549 #ifdef __APPLE__
 550         // Prepend the default path with the JAVA_LIBRARY_PATH so that the app launcher code can specify a directory inside an app wrapper
 551         char *l = getenv("JAVA_LIBRARY_PATH");
 552         if (l != NULL) {
 553             char *t = ld_library_path;
 554             /* That's +1 for the colon and +1 for the trailing '\0' */
 555             ld_library_path = (char *) malloc(strlen(l) + 1 + strlen(t) + 1);
 556             sprintf(ld_library_path, "%s:%s", l, t);
 557             free(t);
 558         }
 559 
 560         char *v = getenv("DYLD_LIBRARY_PATH");
 561 #else
 562         char *v = getenv("LD_LIBRARY_PATH");
 563 #endif
 564         if (v != NULL) {
 565             char *t = ld_library_path;
 566             /* That's +1 for the colon and +1 for the trailing '\0' */
 567             ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
 568             sprintf(ld_library_path, "%s:%s", v, t);
 569             free(t);
 570         }
 571         Arguments::set_library_path(ld_library_path);
 572     }
 573 
 574     /*
 575      * Extensions directories.
 576      *
 577      * Note that the space for the colon and the trailing null are provided
 578      * by the nulls included by the sizeof operator (so actually one byte more
 579      * than necessary is allocated).
 580      */
 581     {
 582 #ifdef __APPLE__
 583         char *buf = malloc(strlen(Arguments::get_java_home()) +
 584             sizeof(EXTENSIONS_DIR) + system_ext_size);
 585         sprintf(buf, "%s" SYS_EXTENSIONS_DIR ":%s" EXTENSIONS_DIR ":"
 586             SYS_EXTENSIONS_DIRS, user_home_dir, Arguments::get_java_home());
 587 #else
 588         char *buf = malloc(strlen(Arguments::get_java_home()) +
 589             sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
 590         sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
 591             Arguments::get_java_home());
 592 #endif
 593 
 594         Arguments::set_ext_dirs(buf);
 595     }
 596 
 597     /* Endorsed standards default directory. */
 598     {
 599         char * buf;
 600         buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 601         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 602         Arguments::set_endorsed_dirs(buf);
 603     }
 604   }
 605 
 606 #ifdef __APPLE__
 607 #undef SYS_EXTENSIONS_DIR
 608 #endif
 609 #undef malloc
 610 #undef getenv
 611 #undef EXTENSIONS_DIR
 612 #undef ENDORSED_DIR
 613 
 614   // Done
 615   return;
 616 }
 617 
 618 ////////////////////////////////////////////////////////////////////////////////
 619 // breakpoint support
 620 
 621 void os::breakpoint() {
 622   BREAKPOINT;
 623 }
 624 
 625 extern "C" void breakpoint() {
 626   // use debugger to set breakpoint here
 627 }
 628 
 629 ////////////////////////////////////////////////////////////////////////////////
 630 // signal support
 631 
 632 debug_only(static bool signal_sets_initialized = false);
 633 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 634 
 635 bool os::Bsd::is_sig_ignored(int sig) {
 636       struct sigaction oact;
 637       sigaction(sig, (struct sigaction*)NULL, &oact);
 638       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 639                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 640       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
 641            return true;
 642       else
 643            return false;
 644 }
 645 
 646 void os::Bsd::signal_sets_init() {
 647   // Should also have an assertion stating we are still single-threaded.
 648   assert(!signal_sets_initialized, "Already initialized");
 649   // Fill in signals that are necessarily unblocked for all threads in
 650   // the VM. Currently, we unblock the following signals:
 651   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 652   //                         by -Xrs (=ReduceSignalUsage));
 653   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 654   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 655   // the dispositions or masks wrt these signals.
 656   // Programs embedding the VM that want to use the above signals for their
 657   // own purposes must, at this time, use the "-Xrs" option to prevent
 658   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 659   // (See bug 4345157, and other related bugs).
 660   // In reality, though, unblocking these signals is really a nop, since
 661   // these signals are not blocked by default.
 662   sigemptyset(&unblocked_sigs);
 663   sigemptyset(&allowdebug_blocked_sigs);
 664   sigaddset(&unblocked_sigs, SIGILL);
 665   sigaddset(&unblocked_sigs, SIGSEGV);
 666   sigaddset(&unblocked_sigs, SIGBUS);
 667   sigaddset(&unblocked_sigs, SIGFPE);
 668   sigaddset(&unblocked_sigs, SR_signum);
 669 
 670   if (!ReduceSignalUsage) {
 671    if (!os::Bsd::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 672       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 673       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 674    }
 675    if (!os::Bsd::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 676       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 677       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 678    }
 679    if (!os::Bsd::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 680       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 681       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 682    }
 683   }
 684   // Fill in signals that are blocked by all but the VM thread.
 685   sigemptyset(&vm_sigs);
 686   if (!ReduceSignalUsage)
 687     sigaddset(&vm_sigs, BREAK_SIGNAL);
 688   debug_only(signal_sets_initialized = true);
 689 
 690 }
 691 
 692 // These are signals that are unblocked while a thread is running Java.
 693 // (For some reason, they get blocked by default.)
 694 sigset_t* os::Bsd::unblocked_signals() {
 695   assert(signal_sets_initialized, "Not initialized");
 696   return &unblocked_sigs;
 697 }
 698 
 699 // These are the signals that are blocked while a (non-VM) thread is
 700 // running Java. Only the VM thread handles these signals.
 701 sigset_t* os::Bsd::vm_signals() {
 702   assert(signal_sets_initialized, "Not initialized");
 703   return &vm_sigs;
 704 }
 705 
 706 // These are signals that are blocked during cond_wait to allow debugger in
 707 sigset_t* os::Bsd::allowdebug_blocked_signals() {
 708   assert(signal_sets_initialized, "Not initialized");
 709   return &allowdebug_blocked_sigs;
 710 }
 711 
 712 void os::Bsd::hotspot_sigmask(Thread* thread) {
 713 
 714   //Save caller's signal mask before setting VM signal mask
 715   sigset_t caller_sigmask;
 716   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 717 
 718   OSThread* osthread = thread->osthread();
 719   osthread->set_caller_sigmask(caller_sigmask);
 720 
 721   pthread_sigmask(SIG_UNBLOCK, os::Bsd::unblocked_signals(), NULL);
 722 
 723   if (!ReduceSignalUsage) {
 724     if (thread->is_VM_thread()) {
 725       // Only the VM thread handles BREAK_SIGNAL ...
 726       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 727     } else {
 728       // ... all other threads block BREAK_SIGNAL
 729       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 730     }
 731   }
 732 }
 733 
 734 #ifndef _ALLBSD_SOURCE
 735 //////////////////////////////////////////////////////////////////////////////
 736 // detecting pthread library
 737 
 738 void os::Bsd::libpthread_init() {
 739   // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
 740   // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
 741   // generic name for earlier versions.
 742   // Define macros here so we can build HotSpot on old systems.
 743 # ifndef _CS_GNU_LIBC_VERSION
 744 # define _CS_GNU_LIBC_VERSION 2
 745 # endif
 746 # ifndef _CS_GNU_LIBPTHREAD_VERSION
 747 # define _CS_GNU_LIBPTHREAD_VERSION 3
 748 # endif
 749 
 750   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 751   if (n > 0) {
 752      char *str = (char *)malloc(n);
 753      confstr(_CS_GNU_LIBC_VERSION, str, n);
 754      os::Bsd::set_glibc_version(str);
 755   } else {
 756      // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
 757      static char _gnu_libc_version[32];
 758      jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
 759               "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
 760      os::Bsd::set_glibc_version(_gnu_libc_version);
 761   }
 762 
 763   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 764   if (n > 0) {
 765      char *str = (char *)malloc(n);
 766      confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 767      // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
 768      // us "NPTL-0.29" even we are running with BsdThreads. Check if this
 769      // is the case. BsdThreads has a hard limit on max number of threads.
 770      // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
 771      // On the other hand, NPTL does not have such a limit, sysconf()
 772      // will return -1 and errno is not changed. Check if it is really NPTL.
 773      if (strcmp(os::Bsd::glibc_version(), "glibc 2.3.2") == 0 &&
 774          strstr(str, "NPTL") &&
 775          sysconf(_SC_THREAD_THREADS_MAX) > 0) {
 776        free(str);
 777        os::Bsd::set_libpthread_version("bsdthreads");
 778      } else {
 779        os::Bsd::set_libpthread_version(str);
 780      }
 781   } else {
 782     // glibc before 2.3.2 only has BsdThreads.
 783     os::Bsd::set_libpthread_version("bsdthreads");
 784   }
 785 
 786   if (strstr(libpthread_version(), "NPTL")) {
 787      os::Bsd::set_is_NPTL();
 788   } else {
 789      os::Bsd::set_is_BsdThreads();
 790   }
 791 
 792   // BsdThreads have two flavors: floating-stack mode, which allows variable
 793   // stack size; and fixed-stack mode. NPTL is always floating-stack.
 794   if (os::Bsd::is_NPTL() || os::Bsd::supports_variable_stack_size()) {
 795      os::Bsd::set_is_floating_stack();
 796   }
 797 }
 798 
 799 /////////////////////////////////////////////////////////////////////////////
 800 // thread stack
 801 
 802 // Force Bsd kernel to expand current thread stack. If "bottom" is close
 803 // to the stack guard, caller should block all signals.
 804 //
 805 // MAP_GROWSDOWN:
 806 //   A special mmap() flag that is used to implement thread stacks. It tells
 807 //   kernel that the memory region should extend downwards when needed. This
 808 //   allows early versions of BsdThreads to only mmap the first few pages
 809 //   when creating a new thread. Bsd kernel will automatically expand thread
 810 //   stack as needed (on page faults).
 811 //
 812 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 813 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 814 //   region, it's hard to tell if the fault is due to a legitimate stack
 815 //   access or because of reading/writing non-exist memory (e.g. buffer
 816 //   overrun). As a rule, if the fault happens below current stack pointer,
 817 //   Bsd kernel does not expand stack, instead a SIGSEGV is sent to the
 818 //   application (see Bsd kernel fault.c).
 819 //
 820 //   This Bsd feature can cause SIGSEGV when VM bangs thread stack for
 821 //   stack overflow detection.
 822 //
 823 //   Newer version of BsdThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 824 //   not use this flag. However, the stack of initial thread is not created
 825 //   by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
 826 //   unlikely) that user code can create a thread with MAP_GROWSDOWN stack
 827 //   and then attach the thread to JVM.
 828 //
 829 // To get around the problem and allow stack banging on Bsd, we need to
 830 // manually expand thread stack after receiving the SIGSEGV.
 831 //
 832 // There are two ways to expand thread stack to address "bottom", we used
 833 // both of them in JVM before 1.5:
 834 //   1. adjust stack pointer first so that it is below "bottom", and then
 835 //      touch "bottom"
 836 //   2. mmap() the page in question
 837 //
 838 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 839 // if current sp is already near the lower end of page 101, and we need to
 840 // call mmap() to map page 100, it is possible that part of the mmap() frame
 841 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 842 // That will destroy the mmap() frame and cause VM to crash.
 843 //
 844 // The following code works by adjusting sp first, then accessing the "bottom"
 845 // page to force a page fault. Bsd kernel will then automatically expand the
 846 // stack mapping.
 847 //
 848 // _expand_stack_to() assumes its frame size is less than page size, which
 849 // should always be true if the function is not inlined.
 850 
 851 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 852 #define NOINLINE
 853 #else
 854 #define NOINLINE __attribute__ ((noinline))
 855 #endif
 856 
 857 static void _expand_stack_to(address bottom) NOINLINE;
 858 
 859 static void _expand_stack_to(address bottom) {
 860   address sp;
 861   size_t size;
 862   volatile char *p;
 863 
 864   // Adjust bottom to point to the largest address within the same page, it
 865   // gives us a one-page buffer if alloca() allocates slightly more memory.
 866   bottom = (address)align_size_down((uintptr_t)bottom, os::Bsd::page_size());
 867   bottom += os::Bsd::page_size() - 1;
 868 
 869   // sp might be slightly above current stack pointer; if that's the case, we
 870   // will alloca() a little more space than necessary, which is OK. Don't use
 871   // os::current_stack_pointer(), as its result can be slightly below current
 872   // stack pointer, causing us to not alloca enough to reach "bottom".
 873   sp = (address)&sp;
 874 
 875   if (sp > bottom) {
 876     size = sp - bottom;
 877     p = (volatile char *)alloca(size);
 878     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 879     p[0] = '\0';
 880   }
 881 }
 882 
 883 bool os::Bsd::manually_expand_stack(JavaThread * t, address addr) {
 884   assert(t!=NULL, "just checking");
 885   assert(t->osthread()->expanding_stack(), "expand should be set");
 886   assert(t->stack_base() != NULL, "stack_base was not initialized");
 887 
 888   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 889     sigset_t mask_all, old_sigset;
 890     sigfillset(&mask_all);
 891     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 892     _expand_stack_to(addr);
 893     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 894     return true;
 895   }
 896   return false;
 897 }
 898 #endif
 899 
 900 //////////////////////////////////////////////////////////////////////////////
 901 // create new thread
 902 
 903 static address highest_vm_reserved_address();
 904 
 905 // check if it's safe to start a new thread
 906 static bool _thread_safety_check(Thread* thread) {
 907 #ifdef _ALLBSD_SOURCE
 908     return true;
 909 #else
 910   if (os::Bsd::is_BsdThreads() && !os::Bsd::is_floating_stack()) {
 911     // Fixed stack BsdThreads (SuSE Bsd/x86, and some versions of Redhat)
 912     //   Heap is mmap'ed at lower end of memory space. Thread stacks are
 913     //   allocated (MAP_FIXED) from high address space. Every thread stack
 914     //   occupies a fixed size slot (usually 2Mbytes, but user can change
 915     //   it to other values if they rebuild BsdThreads).
 916     //
 917     // Problem with MAP_FIXED is that mmap() can still succeed even part of
 918     // the memory region has already been mmap'ed. That means if we have too
 919     // many threads and/or very large heap, eventually thread stack will
 920     // collide with heap.
 921     //
 922     // Here we try to prevent heap/stack collision by comparing current
 923     // stack bottom with the highest address that has been mmap'ed by JVM
 924     // plus a safety margin for memory maps created by native code.
 925     //
 926     // This feature can be disabled by setting ThreadSafetyMargin to 0
 927     //
 928     if (ThreadSafetyMargin > 0) {
 929       address stack_bottom = os::current_stack_base() - os::current_stack_size();
 930 
 931       // not safe if our stack extends below the safety margin
 932       return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
 933     } else {
 934       return true;
 935     }
 936   } else {
 937     // Floating stack BsdThreads or NPTL:
 938     //   Unlike fixed stack BsdThreads, thread stacks are not MAP_FIXED. When
 939     //   there's not enough space left, pthread_create() will fail. If we come
 940     //   here, that means enough space has been reserved for stack.
 941     return true;
 942   }
 943 #endif
 944 }
 945 
 946 #ifdef __APPLE__
 947 // library handle for calling objc_registerThreadWithCollector()
 948 // without static linking to the libobjc library
 949 #define OBJC_LIB "/usr/lib/libobjc.dylib"
 950 #define OBJC_GCREGISTER "objc_registerThreadWithCollector"
 951 typedef void (*objc_registerThreadWithCollector_t)();
 952 extern "C" objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction;
 953 objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction = NULL;
 954 #endif
 955 
 956 // Thread start routine for all newly created threads
 957 static void *java_start(Thread *thread) {
 958   // Try to randomize the cache line index of hot stack frames.
 959   // This helps when threads of the same stack traces evict each other's
 960   // cache lines. The threads can be either from the same JVM instance, or
 961   // from different JVM instances. The benefit is especially true for
 962   // processors with hyperthreading technology.
 963   static int counter = 0;
 964   int pid = os::current_process_id();
 965   alloca(((pid ^ counter++) & 7) * 128);
 966 
 967   ThreadLocalStorage::set_thread(thread);
 968 
 969   OSThread* osthread = thread->osthread();
 970   Monitor* sync = osthread->startThread_lock();
 971 
 972   // non floating stack BsdThreads needs extra check, see above
 973   if (!_thread_safety_check(thread)) {
 974     // notify parent thread
 975     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 976     osthread->set_state(ZOMBIE);
 977     sync->notify_all();
 978     return NULL;
 979   }
 980 
 981 #ifdef _ALLBSD_SOURCE
 982   // thread_id is pthread_id on BSD
 983   osthread->set_thread_id(::pthread_self());
 984 #else
 985   // thread_id is kernel thread id (similar to Solaris LWP id)
 986   osthread->set_thread_id(os::Bsd::gettid());
 987 
 988   if (UseNUMA) {
 989     int lgrp_id = os::numa_get_group_id();
 990     if (lgrp_id != -1) {
 991       thread->set_lgrp_id(lgrp_id);
 992     }
 993   }
 994 #endif
 995   // initialize signal mask for this thread
 996   os::Bsd::hotspot_sigmask(thread);
 997 
 998   // initialize floating point control register
 999   os::Bsd::init_thread_fpu_state();
1000 
1001 #ifdef __APPLE__
1002   // register thread with objc gc
1003   if (objc_registerThreadWithCollectorFunction != NULL) {
1004     objc_registerThreadWithCollectorFunction();
1005   }
1006 #endif
1007 
1008   // handshaking with parent thread
1009   {
1010     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
1011 
1012     // notify parent thread
1013     osthread->set_state(INITIALIZED);
1014     sync->notify_all();
1015 
1016     // wait until os::start_thread()
1017     while (osthread->get_state() == INITIALIZED) {
1018       sync->wait(Mutex::_no_safepoint_check_flag);
1019     }
1020   }
1021 
1022   // call one more level start routine
1023   thread->run();
1024 
1025   return 0;
1026 }
1027 
1028 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1029   assert(thread->osthread() == NULL, "caller responsible");
1030 
1031   // Allocate the OSThread object
1032   OSThread* osthread = new OSThread(NULL, NULL);
1033   if (osthread == NULL) {
1034     return false;
1035   }
1036 
1037   // set the correct thread state
1038   osthread->set_thread_type(thr_type);
1039 
1040   // Initial state is ALLOCATED but not INITIALIZED
1041   osthread->set_state(ALLOCATED);
1042 
1043   thread->set_osthread(osthread);
1044 
1045   // init thread attributes
1046   pthread_attr_t attr;
1047   pthread_attr_init(&attr);
1048   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1049 
1050   // stack size
1051   if (os::Bsd::supports_variable_stack_size()) {
1052     // calculate stack size if it's not specified by caller
1053     if (stack_size == 0) {
1054       stack_size = os::Bsd::default_stack_size(thr_type);
1055 
1056       switch (thr_type) {
1057       case os::java_thread:
1058         // Java threads use ThreadStackSize which default value can be
1059         // changed with the flag -Xss
1060         assert (JavaThread::stack_size_at_create() > 0, "this should be set");
1061         stack_size = JavaThread::stack_size_at_create();
1062         break;
1063       case os::compiler_thread:
1064         if (CompilerThreadStackSize > 0) {
1065           stack_size = (size_t)(CompilerThreadStackSize * K);
1066           break;
1067         } // else fall through:
1068           // use VMThreadStackSize if CompilerThreadStackSize is not defined
1069       case os::vm_thread:
1070       case os::pgc_thread:
1071       case os::cgc_thread:
1072       case os::watcher_thread:
1073         if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1074         break;
1075       }
1076     }
1077 
1078     stack_size = MAX2(stack_size, os::Bsd::min_stack_allowed);
1079     pthread_attr_setstacksize(&attr, stack_size);
1080   } else {
1081     // let pthread_create() pick the default value.
1082   }
1083 
1084 #ifndef _ALLBSD_SOURCE
1085   // glibc guard page
1086   pthread_attr_setguardsize(&attr, os::Bsd::default_guard_size(thr_type));
1087 #endif
1088 
1089   ThreadState state;
1090 
1091   {
1092 
1093 #ifndef _ALLBSD_SOURCE
1094     // Serialize thread creation if we are running with fixed stack BsdThreads
1095     bool lock = os::Bsd::is_BsdThreads() && !os::Bsd::is_floating_stack();
1096     if (lock) {
1097       os::Bsd::createThread_lock()->lock_without_safepoint_check();
1098     }
1099 #endif
1100 
1101     pthread_t tid;
1102     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
1103 
1104     pthread_attr_destroy(&attr);
1105 
1106     if (ret != 0) {
1107       if (PrintMiscellaneous && (Verbose || WizardMode)) {
1108         perror("pthread_create()");
1109       }
1110       // Need to clean up stuff we've allocated so far
1111       thread->set_osthread(NULL);
1112       delete osthread;
1113 #ifndef _ALLBSD_SOURCE
1114       if (lock) os::Bsd::createThread_lock()->unlock();
1115 #endif
1116       return false;
1117     }
1118 
1119     // Store pthread info into the OSThread
1120     osthread->set_pthread_id(tid);
1121 
1122     // Wait until child thread is either initialized or aborted
1123     {
1124       Monitor* sync_with_child = osthread->startThread_lock();
1125       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1126       while ((state = osthread->get_state()) == ALLOCATED) {
1127         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
1128       }
1129     }
1130 
1131 #ifndef _ALLBSD_SOURCE
1132     if (lock) {
1133       os::Bsd::createThread_lock()->unlock();
1134     }
1135 #endif
1136   }
1137 
1138   // Aborted due to thread limit being reached
1139   if (state == ZOMBIE) {
1140       thread->set_osthread(NULL);
1141       delete osthread;
1142       return false;
1143   }
1144 
1145   // The thread is returned suspended (in state INITIALIZED),
1146   // and is started higher up in the call chain
1147   assert(state == INITIALIZED, "race condition");
1148   return true;
1149 }
1150 
1151 /////////////////////////////////////////////////////////////////////////////
1152 // attach existing thread
1153 
1154 // bootstrap the main thread
1155 bool os::create_main_thread(JavaThread* thread) {
1156   assert(os::Bsd::_main_thread == pthread_self(), "should be called inside main thread");
1157   return create_attached_thread(thread);
1158 }
1159 
1160 bool os::create_attached_thread(JavaThread* thread) {
1161 #ifdef ASSERT
1162     thread->verify_not_published();
1163 #endif
1164 
1165   // Allocate the OSThread object
1166   OSThread* osthread = new OSThread(NULL, NULL);
1167 
1168   if (osthread == NULL) {
1169     return false;
1170   }
1171 
1172   // Store pthread info into the OSThread
1173 #ifdef _ALLBSD_SOURCE
1174   osthread->set_thread_id(::pthread_self());
1175 #else
1176   osthread->set_thread_id(os::Bsd::gettid());
1177 #endif
1178   osthread->set_pthread_id(::pthread_self());
1179 
1180   // initialize floating point control register
1181   os::Bsd::init_thread_fpu_state();
1182 
1183   // Initial thread state is RUNNABLE
1184   osthread->set_state(RUNNABLE);
1185 
1186   thread->set_osthread(osthread);
1187 
1188 #ifndef _ALLBSD_SOURCE
1189   if (UseNUMA) {
1190     int lgrp_id = os::numa_get_group_id();
1191     if (lgrp_id != -1) {
1192       thread->set_lgrp_id(lgrp_id);
1193     }
1194   }
1195 
1196   if (os::Bsd::is_initial_thread()) {
1197     // If current thread is initial thread, its stack is mapped on demand,
1198     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1199     // the entire stack region to avoid SEGV in stack banging.
1200     // It is also useful to get around the heap-stack-gap problem on SuSE
1201     // kernel (see 4821821 for details). We first expand stack to the top
1202     // of yellow zone, then enable stack yellow zone (order is significant,
1203     // enabling yellow zone first will crash JVM on SuSE Bsd), so there
1204     // is no gap between the last two virtual memory regions.
1205 
1206     JavaThread *jt = (JavaThread *)thread;
1207     address addr = jt->stack_yellow_zone_base();
1208     assert(addr != NULL, "initialization problem?");
1209     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1210 
1211     osthread->set_expanding_stack();
1212     os::Bsd::manually_expand_stack(jt, addr);
1213     osthread->clear_expanding_stack();
1214   }
1215 #endif
1216 
1217   // initialize signal mask for this thread
1218   // and save the caller's signal mask
1219   os::Bsd::hotspot_sigmask(thread);
1220 
1221   return true;
1222 }
1223 
1224 void os::pd_start_thread(Thread* thread) {
1225   OSThread * osthread = thread->osthread();
1226   assert(osthread->get_state() != INITIALIZED, "just checking");
1227   Monitor* sync_with_child = osthread->startThread_lock();
1228   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1229   sync_with_child->notify();
1230 }
1231 
1232 // Free Bsd resources related to the OSThread
1233 void os::free_thread(OSThread* osthread) {
1234   assert(osthread != NULL, "osthread not set");
1235 
1236   if (Thread::current()->osthread() == osthread) {
1237     // Restore caller's signal mask
1238     sigset_t sigmask = osthread->caller_sigmask();
1239     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1240    }
1241 
1242   delete osthread;
1243 }
1244 
1245 //////////////////////////////////////////////////////////////////////////////
1246 // thread local storage
1247 
1248 int os::allocate_thread_local_storage() {
1249   pthread_key_t key;
1250   int rslt = pthread_key_create(&key, NULL);
1251   assert(rslt == 0, "cannot allocate thread local storage");
1252   return (int)key;
1253 }
1254 
1255 // Note: This is currently not used by VM, as we don't destroy TLS key
1256 // on VM exit.
1257 void os::free_thread_local_storage(int index) {
1258   int rslt = pthread_key_delete((pthread_key_t)index);
1259   assert(rslt == 0, "invalid index");
1260 }
1261 
1262 void os::thread_local_storage_at_put(int index, void* value) {
1263   int rslt = pthread_setspecific((pthread_key_t)index, value);
1264   assert(rslt == 0, "pthread_setspecific failed");
1265 }
1266 
1267 extern "C" Thread* get_thread() {
1268   return ThreadLocalStorage::thread();
1269 }
1270 
1271 //////////////////////////////////////////////////////////////////////////////
1272 // initial thread
1273 
1274 #ifndef _ALLBSD_SOURCE
1275 // Check if current thread is the initial thread, similar to Solaris thr_main.
1276 bool os::Bsd::is_initial_thread(void) {
1277   char dummy;
1278   // If called before init complete, thread stack bottom will be null.
1279   // Can be called if fatal error occurs before initialization.
1280   if (initial_thread_stack_bottom() == NULL) return false;
1281   assert(initial_thread_stack_bottom() != NULL &&
1282          initial_thread_stack_size()   != 0,
1283          "os::init did not locate initial thread's stack region");
1284   if ((address)&dummy >= initial_thread_stack_bottom() &&
1285       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1286        return true;
1287   else return false;
1288 }
1289 
1290 // Find the virtual memory area that contains addr
1291 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1292   FILE *fp = fopen("/proc/self/maps", "r");
1293   if (fp) {
1294     address low, high;
1295     while (!feof(fp)) {
1296       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1297         if (low <= addr && addr < high) {
1298            if (vma_low)  *vma_low  = low;
1299            if (vma_high) *vma_high = high;
1300            fclose (fp);
1301            return true;
1302         }
1303       }
1304       for (;;) {
1305         int ch = fgetc(fp);
1306         if (ch == EOF || ch == (int)'\n') break;
1307       }
1308     }
1309     fclose(fp);
1310   }
1311   return false;
1312 }
1313 
1314 // Locate initial thread stack. This special handling of initial thread stack
1315 // is needed because pthread_getattr_np() on most (all?) Bsd distros returns
1316 // bogus value for initial thread.
1317 void os::Bsd::capture_initial_stack(size_t max_size) {
1318   // stack size is the easy part, get it from RLIMIT_STACK
1319   size_t stack_size;
1320   struct rlimit rlim;
1321   getrlimit(RLIMIT_STACK, &rlim);
1322   stack_size = rlim.rlim_cur;
1323 
1324   // 6308388: a bug in ld.so will relocate its own .data section to the
1325   //   lower end of primordial stack; reduce ulimit -s value a little bit
1326   //   so we won't install guard page on ld.so's data section.
1327   stack_size -= 2 * page_size();
1328 
1329   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1330   //   7.1, in both cases we will get 2G in return value.
1331   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1332   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
1333   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
1334   //   in case other parts in glibc still assumes 2M max stack size.
1335   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1336 #ifndef IA64
1337   if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1338 #else
1339   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1340   if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1341 #endif
1342 
1343   // Try to figure out where the stack base (top) is. This is harder.
1344   //
1345   // When an application is started, glibc saves the initial stack pointer in
1346   // a global variable "__libc_stack_end", which is then used by system
1347   // libraries. __libc_stack_end should be pretty close to stack top. The
1348   // variable is available since the very early days. However, because it is
1349   // a private interface, it could disappear in the future.
1350   //
1351   // Bsd kernel saves start_stack information in /proc/<pid>/stat. Similar
1352   // to __libc_stack_end, it is very close to stack top, but isn't the real
1353   // stack top. Note that /proc may not exist if VM is running as a chroot
1354   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1355   // /proc/<pid>/stat could change in the future (though unlikely).
1356   //
1357   // We try __libc_stack_end first. If that doesn't work, look for
1358   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1359   // as a hint, which should work well in most cases.
1360 
1361   uintptr_t stack_start;
1362 
1363   // try __libc_stack_end first
1364   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1365   if (p && *p) {
1366     stack_start = *p;
1367   } else {
1368     // see if we can get the start_stack field from /proc/self/stat
1369     FILE *fp;
1370     int pid;
1371     char state;
1372     int ppid;
1373     int pgrp;
1374     int session;
1375     int nr;
1376     int tpgrp;
1377     unsigned long flags;
1378     unsigned long minflt;
1379     unsigned long cminflt;
1380     unsigned long majflt;
1381     unsigned long cmajflt;
1382     unsigned long utime;
1383     unsigned long stime;
1384     long cutime;
1385     long cstime;
1386     long prio;
1387     long nice;
1388     long junk;
1389     long it_real;
1390     uintptr_t start;
1391     uintptr_t vsize;
1392     intptr_t rss;
1393     uintptr_t rsslim;
1394     uintptr_t scodes;
1395     uintptr_t ecode;
1396     int i;
1397 
1398     // Figure what the primordial thread stack base is. Code is inspired
1399     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1400     // followed by command name surrounded by parentheses, state, etc.
1401     char stat[2048];
1402     int statlen;
1403 
1404     fp = fopen("/proc/self/stat", "r");
1405     if (fp) {
1406       statlen = fread(stat, 1, 2047, fp);
1407       stat[statlen] = '\0';
1408       fclose(fp);
1409 
1410       // Skip pid and the command string. Note that we could be dealing with
1411       // weird command names, e.g. user could decide to rename java launcher
1412       // to "java 1.4.2 :)", then the stat file would look like
1413       //                1234 (java 1.4.2 :)) R ... ...
1414       // We don't really need to know the command string, just find the last
1415       // occurrence of ")" and then start parsing from there. See bug 4726580.
1416       char * s = strrchr(stat, ')');
1417 
1418       i = 0;
1419       if (s) {
1420         // Skip blank chars
1421         do s++; while (isspace(*s));
1422 
1423 #define _UFM UINTX_FORMAT
1424 #define _DFM INTX_FORMAT
1425 
1426         /*                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2 */
1427         /*              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 */
1428         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,
1429              &state,          /* 3  %c  */
1430              &ppid,           /* 4  %d  */
1431              &pgrp,           /* 5  %d  */
1432              &session,        /* 6  %d  */
1433              &nr,             /* 7  %d  */
1434              &tpgrp,          /* 8  %d  */
1435              &flags,          /* 9  %lu  */
1436              &minflt,         /* 10 %lu  */
1437              &cminflt,        /* 11 %lu  */
1438              &majflt,         /* 12 %lu  */
1439              &cmajflt,        /* 13 %lu  */
1440              &utime,          /* 14 %lu  */
1441              &stime,          /* 15 %lu  */
1442              &cutime,         /* 16 %ld  */
1443              &cstime,         /* 17 %ld  */
1444              &prio,           /* 18 %ld  */
1445              &nice,           /* 19 %ld  */
1446              &junk,           /* 20 %ld  */
1447              &it_real,        /* 21 %ld  */
1448              &start,          /* 22 UINTX_FORMAT */
1449              &vsize,          /* 23 UINTX_FORMAT */
1450              &rss,            /* 24 INTX_FORMAT  */
1451              &rsslim,         /* 25 UINTX_FORMAT */
1452              &scodes,         /* 26 UINTX_FORMAT */
1453              &ecode,          /* 27 UINTX_FORMAT */
1454              &stack_start);   /* 28 UINTX_FORMAT */
1455       }
1456 
1457 #undef _UFM
1458 #undef _DFM
1459 
1460       if (i != 28 - 2) {
1461          assert(false, "Bad conversion from /proc/self/stat");
1462          // product mode - assume we are the initial thread, good luck in the
1463          // embedded case.
1464          warning("Can't detect initial thread stack location - bad conversion");
1465          stack_start = (uintptr_t) &rlim;
1466       }
1467     } else {
1468       // For some reason we can't open /proc/self/stat (for example, running on
1469       // FreeBSD with a Bsd emulator, or inside chroot), this should work for
1470       // most cases, so don't abort:
1471       warning("Can't detect initial thread stack location - no /proc/self/stat");
1472       stack_start = (uintptr_t) &rlim;
1473     }
1474   }
1475 
1476   // Now we have a pointer (stack_start) very close to the stack top, the
1477   // next thing to do is to figure out the exact location of stack top. We
1478   // can find out the virtual memory area that contains stack_start by
1479   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1480   // and its upper limit is the real stack top. (again, this would fail if
1481   // running inside chroot, because /proc may not exist.)
1482 
1483   uintptr_t stack_top;
1484   address low, high;
1485   if (find_vma((address)stack_start, &low, &high)) {
1486     // success, "high" is the true stack top. (ignore "low", because initial
1487     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1488     stack_top = (uintptr_t)high;
1489   } else {
1490     // failed, likely because /proc/self/maps does not exist
1491     warning("Can't detect initial thread stack location - find_vma failed");
1492     // best effort: stack_start is normally within a few pages below the real
1493     // stack top, use it as stack top, and reduce stack size so we won't put
1494     // guard page outside stack.
1495     stack_top = stack_start;
1496     stack_size -= 16 * page_size();
1497   }
1498 
1499   // stack_top could be partially down the page so align it
1500   stack_top = align_size_up(stack_top, page_size());
1501 
1502   if (max_size && stack_size > max_size) {
1503      _initial_thread_stack_size = max_size;
1504   } else {
1505      _initial_thread_stack_size = stack_size;
1506   }
1507 
1508   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1509   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1510 }
1511 #endif
1512 
1513 ////////////////////////////////////////////////////////////////////////////////
1514 // time support
1515 
1516 // Time since start-up in seconds to a fine granularity.
1517 // Used by VMSelfDestructTimer and the MemProfiler.
1518 double os::elapsedTime() {
1519 
1520   return (double)(os::elapsed_counter()) * 0.000001;
1521 }
1522 
1523 jlong os::elapsed_counter() {
1524   timeval time;
1525   int status = gettimeofday(&time, NULL);
1526   return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1527 }
1528 
1529 jlong os::elapsed_frequency() {
1530   return (1000 * 1000);
1531 }
1532 
1533 // XXX: For now, code this as if BSD does not support vtime.
1534 bool os::supports_vtime() { return false; }
1535 bool os::enable_vtime()   { return false; }
1536 bool os::vtime_enabled()  { return false; }
1537 double os::elapsedVTime() {
1538   // better than nothing, but not much
1539   return elapsedTime();
1540 }
1541 
1542 jlong os::javaTimeMillis() {
1543   timeval time;
1544   int status = gettimeofday(&time, NULL);
1545   assert(status != -1, "bsd error");
1546   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1547 }
1548 
1549 #ifndef CLOCK_MONOTONIC
1550 #define CLOCK_MONOTONIC (1)
1551 #endif
1552 
1553 #ifdef __APPLE__
1554 void os::Bsd::clock_init() {
1555         // XXXDARWIN: Investigate replacement monotonic clock
1556 }
1557 #elif defined(_ALLBSD_SOURCE)
1558 void os::Bsd::clock_init() {
1559   struct timespec res;
1560   struct timespec tp;
1561   if (::clock_getres(CLOCK_MONOTONIC, &res) == 0 &&
1562       ::clock_gettime(CLOCK_MONOTONIC, &tp)  == 0) {
1563     // yes, monotonic clock is supported
1564     _clock_gettime = ::clock_gettime;
1565   }
1566 }
1567 #else
1568 void os::Bsd::clock_init() {
1569   // we do dlopen's in this particular order due to bug in bsd
1570   // dynamical loader (see 6348968) leading to crash on exit
1571   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1572   if (handle == NULL) {
1573     handle = dlopen("librt.so", RTLD_LAZY);
1574   }
1575 
1576   if (handle) {
1577     int (*clock_getres_func)(clockid_t, struct timespec*) =
1578            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1579     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1580            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1581     if (clock_getres_func && clock_gettime_func) {
1582       // See if monotonic clock is supported by the kernel. Note that some
1583       // early implementations simply return kernel jiffies (updated every
1584       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1585       // for nano time (though the monotonic property is still nice to have).
1586       // It's fixed in newer kernels, however clock_getres() still returns
1587       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1588       // resolution for now. Hopefully as people move to new kernels, this
1589       // won't be a problem.
1590       struct timespec res;
1591       struct timespec tp;
1592       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1593           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1594         // yes, monotonic clock is supported
1595         _clock_gettime = clock_gettime_func;
1596       } else {
1597         // close librt if there is no monotonic clock
1598         dlclose(handle);
1599       }
1600     }
1601   }
1602 }
1603 #endif
1604 
1605 #ifndef _ALLBSD_SOURCE
1606 #ifndef SYS_clock_getres
1607 
1608 #if defined(IA32) || defined(AMD64)
1609 #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1610 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1611 #else
1612 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1613 #define sys_clock_getres(x,y)  -1
1614 #endif
1615 
1616 #else
1617 #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1618 #endif
1619 
1620 void os::Bsd::fast_thread_clock_init() {
1621   if (!UseBsdPosixThreadCPUClocks) {
1622     return;
1623   }
1624   clockid_t clockid;
1625   struct timespec tp;
1626   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1627       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1628 
1629   // Switch to using fast clocks for thread cpu time if
1630   // the sys_clock_getres() returns 0 error code.
1631   // Note, that some kernels may support the current thread
1632   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1633   // returned by the pthread_getcpuclockid().
1634   // If the fast Posix clocks are supported then the sys_clock_getres()
1635   // must return at least tp.tv_sec == 0 which means a resolution
1636   // better than 1 sec. This is extra check for reliability.
1637 
1638   if(pthread_getcpuclockid_func &&
1639      pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1640      sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1641 
1642     _supports_fast_thread_cpu_time = true;
1643     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1644   }
1645 }
1646 #endif
1647 
1648 jlong os::javaTimeNanos() {
1649   if (Bsd::supports_monotonic_clock()) {
1650     struct timespec tp;
1651     int status = Bsd::clock_gettime(CLOCK_MONOTONIC, &tp);
1652     assert(status == 0, "gettime error");
1653     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1654     return result;
1655   } else {
1656     timeval time;
1657     int status = gettimeofday(&time, NULL);
1658     assert(status != -1, "bsd error");
1659     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1660     return 1000 * usecs;
1661   }
1662 }
1663 
1664 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1665   if (Bsd::supports_monotonic_clock()) {
1666     info_ptr->max_value = ALL_64_BITS;
1667 
1668     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1669     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1670     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1671   } else {
1672     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1673     info_ptr->max_value = ALL_64_BITS;
1674 
1675     // gettimeofday is a real time clock so it skips
1676     info_ptr->may_skip_backward = true;
1677     info_ptr->may_skip_forward = true;
1678   }
1679 
1680   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1681 }
1682 
1683 // Return the real, user, and system times in seconds from an
1684 // arbitrary fixed point in the past.
1685 bool os::getTimesSecs(double* process_real_time,
1686                       double* process_user_time,
1687                       double* process_system_time) {
1688   struct tms ticks;
1689   clock_t real_ticks = times(&ticks);
1690 
1691   if (real_ticks == (clock_t) (-1)) {
1692     return false;
1693   } else {
1694     double ticks_per_second = (double) clock_tics_per_sec;
1695     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1696     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1697     *process_real_time = ((double) real_ticks) / ticks_per_second;
1698 
1699     return true;
1700   }
1701 }
1702 
1703 
1704 char * os::local_time_string(char *buf, size_t buflen) {
1705   struct tm t;
1706   time_t long_time;
1707   time(&long_time);
1708   localtime_r(&long_time, &t);
1709   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1710                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1711                t.tm_hour, t.tm_min, t.tm_sec);
1712   return buf;
1713 }
1714 
1715 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1716   return localtime_r(clock, res);
1717 }
1718 
1719 ////////////////////////////////////////////////////////////////////////////////
1720 // runtime exit support
1721 
1722 // Note: os::shutdown() might be called very early during initialization, or
1723 // called from signal handler. Before adding something to os::shutdown(), make
1724 // sure it is async-safe and can handle partially initialized VM.
1725 void os::shutdown() {
1726 
1727   // allow PerfMemory to attempt cleanup of any persistent resources
1728   perfMemory_exit();
1729 
1730   // needs to remove object in file system
1731   AttachListener::abort();
1732 
1733   // flush buffered output, finish log files
1734   ostream_abort();
1735 
1736   // Check for abort hook
1737   abort_hook_t abort_hook = Arguments::abort_hook();
1738   if (abort_hook != NULL) {
1739     abort_hook();
1740   }
1741 
1742 }
1743 
1744 // Note: os::abort() might be called very early during initialization, or
1745 // called from signal handler. Before adding something to os::abort(), make
1746 // sure it is async-safe and can handle partially initialized VM.
1747 void os::abort(bool dump_core) {
1748   os::shutdown();
1749   if (dump_core) {
1750 #ifndef PRODUCT
1751     fdStream out(defaultStream::output_fd());
1752     out.print_raw("Current thread is ");
1753     char buf[16];
1754     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1755     out.print_raw_cr(buf);
1756     out.print_raw_cr("Dumping core ...");
1757 #endif
1758     ::abort(); // dump core
1759   }
1760 
1761   ::exit(1);
1762 }
1763 
1764 // Die immediately, no exit hook, no abort hook, no cleanup.
1765 void os::die() {
1766   // _exit() on BsdThreads only kills current thread
1767   ::abort();
1768 }
1769 
1770 // unused on bsd for now.
1771 void os::set_error_file(const char *logfile) {}
1772 
1773 
1774 // This method is a copy of JDK's sysGetLastErrorString
1775 // from src/solaris/hpi/src/system_md.c
1776 
1777 size_t os::lasterror(char *buf, size_t len) {
1778 
1779   if (errno == 0)  return 0;
1780 
1781   const char *s = ::strerror(errno);
1782   size_t n = ::strlen(s);
1783   if (n >= len) {
1784     n = len - 1;
1785   }
1786   ::strncpy(buf, s, n);
1787   buf[n] = '\0';
1788   return n;
1789 }
1790 
1791 intx os::current_thread_id() { return (intx)pthread_self(); }
1792 int os::current_process_id() {
1793 
1794   // Under the old bsd thread library, bsd gives each thread
1795   // its own process id. Because of this each thread will return
1796   // a different pid if this method were to return the result
1797   // of getpid(2). Bsd provides no api that returns the pid
1798   // of the launcher thread for the vm. This implementation
1799   // returns a unique pid, the pid of the launcher thread
1800   // that starts the vm 'process'.
1801 
1802   // Under the NPTL, getpid() returns the same pid as the
1803   // launcher thread rather than a unique pid per thread.
1804   // Use gettid() if you want the old pre NPTL behaviour.
1805 
1806   // if you are looking for the result of a call to getpid() that
1807   // returns a unique pid for the calling thread, then look at the
1808   // OSThread::thread_id() method in osThread_bsd.hpp file
1809 
1810   return (int)(_initial_pid ? _initial_pid : getpid());
1811 }
1812 
1813 // DLL functions
1814 
1815 #define JNI_LIB_PREFIX "lib"
1816 #ifdef __APPLE__
1817 #define JNI_LIB_SUFFIX ".dylib"
1818 #else
1819 #define JNI_LIB_SUFFIX ".so"
1820 #endif
1821 
1822 const char* os::dll_file_extension() { return JNI_LIB_SUFFIX; }
1823 
1824 // This must be hard coded because it's the system's temporary
1825 // directory not the java application's temp directory, ala java.io.tmpdir.
1826 #ifdef __APPLE__
1827 // macosx has a secure per-user temporary directory
1828 char temp_path_storage[PATH_MAX];
1829 const char* os::get_temp_directory() {
1830   static char *temp_path = NULL;
1831   if (temp_path == NULL) {
1832     int pathSize = confstr(_CS_DARWIN_USER_TEMP_DIR, temp_path_storage, PATH_MAX);
1833     if (pathSize == 0 || pathSize > PATH_MAX) {
1834       strlcpy(temp_path_storage, "/tmp/", sizeof(temp_path_storage));
1835     }
1836     temp_path = temp_path_storage;
1837   }
1838   return temp_path;
1839 }
1840 #else /* __APPLE__ */
1841 const char* os::get_temp_directory() { return "/tmp"; }
1842 #endif /* __APPLE__ */
1843 
1844 static bool file_exists(const char* filename) {
1845   struct stat statbuf;
1846   if (filename == NULL || strlen(filename) == 0) {
1847     return false;
1848   }
1849   return os::stat(filename, &statbuf) == 0;
1850 }
1851 
1852 void os::dll_build_name(char* buffer, size_t buflen,
1853                         const char* pname, const char* fname) {
1854   // Copied from libhpi
1855   const size_t pnamelen = pname ? strlen(pname) : 0;
1856 
1857   // Quietly truncate on buffer overflow.  Should be an error.
1858   if (pnamelen + strlen(fname) + strlen(JNI_LIB_PREFIX) + strlen(JNI_LIB_SUFFIX) + 2 > buflen) {
1859       *buffer = '\0';
1860       return;
1861   }
1862 
1863   if (pnamelen == 0) {
1864     snprintf(buffer, buflen, JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, fname);
1865   } else if (strchr(pname, *os::path_separator()) != NULL) {
1866     int n;
1867     char** pelements = split_path(pname, &n);
1868     for (int i = 0 ; i < n ; i++) {
1869       // Really shouldn't be NULL, but check can't hurt
1870       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1871         continue; // skip the empty path values
1872       }
1873       snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX,
1874           pelements[i], fname);
1875       if (file_exists(buffer)) {
1876         break;
1877       }
1878     }
1879     // release the storage
1880     for (int i = 0 ; i < n ; i++) {
1881       if (pelements[i] != NULL) {
1882         FREE_C_HEAP_ARRAY(char, pelements[i]);
1883       }
1884     }
1885     if (pelements != NULL) {
1886       FREE_C_HEAP_ARRAY(char*, pelements);
1887     }
1888   } else {
1889     snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, pname, fname);
1890   }
1891 }
1892 
1893 const char* os::get_current_directory(char *buf, int buflen) {
1894   return getcwd(buf, buflen);
1895 }
1896 
1897 // check if addr is inside libjvm[_g].so
1898 bool os::address_is_in_vm(address addr) {
1899   static address libjvm_base_addr;
1900   Dl_info dlinfo;
1901 
1902   if (libjvm_base_addr == NULL) {
1903     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1904     libjvm_base_addr = (address)dlinfo.dli_fbase;
1905     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1906   }
1907 
1908   if (dladdr((void *)addr, &dlinfo)) {
1909     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1910   }
1911 
1912   return false;
1913 }
1914 
1915 bool os::dll_address_to_function_name(address addr, char *buf,
1916                                       int buflen, int *offset) {
1917   Dl_info dlinfo;
1918 
1919   if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1920     if (buf != NULL) {
1921       if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1922         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1923       }
1924     }
1925     if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1926     return true;
1927   } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1928     if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1929        buf, buflen, offset, dlinfo.dli_fname)) {
1930        return true;
1931     }
1932   }
1933 
1934   if (buf != NULL) buf[0] = '\0';
1935   if (offset != NULL) *offset = -1;
1936   return false;
1937 }
1938 
1939 #ifdef _ALLBSD_SOURCE
1940 // ported from solaris version
1941 bool os::dll_address_to_library_name(address addr, char* buf,
1942                                      int buflen, int* offset) {
1943   Dl_info dlinfo;
1944 
1945   if (dladdr((void*)addr, &dlinfo)){
1946      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1947      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1948      return true;
1949   } else {
1950      if (buf) buf[0] = '\0';
1951      if (offset) *offset = -1;
1952      return false;
1953   }
1954 }
1955 #else
1956 struct _address_to_library_name {
1957   address addr;          // input : memory address
1958   size_t  buflen;        //         size of fname
1959   char*   fname;         // output: library name
1960   address base;          //         library base addr
1961 };
1962 
1963 static int address_to_library_name_callback(struct dl_phdr_info *info,
1964                                             size_t size, void *data) {
1965   int i;
1966   bool found = false;
1967   address libbase = NULL;
1968   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1969 
1970   // iterate through all loadable segments
1971   for (i = 0; i < info->dlpi_phnum; i++) {
1972     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1973     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1974       // base address of a library is the lowest address of its loaded
1975       // segments.
1976       if (libbase == NULL || libbase > segbase) {
1977         libbase = segbase;
1978       }
1979       // see if 'addr' is within current segment
1980       if (segbase <= d->addr &&
1981           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1982         found = true;
1983       }
1984     }
1985   }
1986 
1987   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1988   // so dll_address_to_library_name() can fall through to use dladdr() which
1989   // can figure out executable name from argv[0].
1990   if (found && info->dlpi_name && info->dlpi_name[0]) {
1991     d->base = libbase;
1992     if (d->fname) {
1993       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1994     }
1995     return 1;
1996   }
1997   return 0;
1998 }
1999 
2000 bool os::dll_address_to_library_name(address addr, char* buf,
2001                                      int buflen, int* offset) {
2002   Dl_info dlinfo;
2003   struct _address_to_library_name data;
2004 
2005   // There is a bug in old glibc dladdr() implementation that it could resolve
2006   // to wrong library name if the .so file has a base address != NULL. Here
2007   // we iterate through the program headers of all loaded libraries to find
2008   // out which library 'addr' really belongs to. This workaround can be
2009   // removed once the minimum requirement for glibc is moved to 2.3.x.
2010   data.addr = addr;
2011   data.fname = buf;
2012   data.buflen = buflen;
2013   data.base = NULL;
2014   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
2015 
2016   if (rslt) {
2017      // buf already contains library name
2018      if (offset) *offset = addr - data.base;
2019      return true;
2020   } else if (dladdr((void*)addr, &dlinfo)){
2021      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
2022      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
2023      return true;
2024   } else {
2025      if (buf) buf[0] = '\0';
2026      if (offset) *offset = -1;
2027      return false;
2028   }
2029 }
2030 #endif
2031 
2032   // Loads .dll/.so and
2033   // in case of error it checks if .dll/.so was built for the
2034   // same architecture as Hotspot is running on
2035 
2036 #ifdef __APPLE__
2037 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
2038   void * result= ::dlopen(filename, RTLD_LAZY);
2039   if (result != NULL) {
2040     // Successful loading
2041     return result;
2042   }
2043 
2044   // Read system error message into ebuf
2045   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2046   ebuf[ebuflen-1]='\0';
2047 
2048   return NULL;
2049 }
2050 #else
2051 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
2052 {
2053   void * result= ::dlopen(filename, RTLD_LAZY);
2054   if (result != NULL) {
2055     // Successful loading
2056     return result;
2057   }
2058 
2059   Elf32_Ehdr elf_head;
2060 
2061   // Read system error message into ebuf
2062   // It may or may not be overwritten below
2063   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2064   ebuf[ebuflen-1]='\0';
2065   int diag_msg_max_length=ebuflen-strlen(ebuf);
2066   char* diag_msg_buf=ebuf+strlen(ebuf);
2067 
2068   if (diag_msg_max_length==0) {
2069     // No more space in ebuf for additional diagnostics message
2070     return NULL;
2071   }
2072 
2073 
2074   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2075 
2076   if (file_descriptor < 0) {
2077     // Can't open library, report dlerror() message
2078     return NULL;
2079   }
2080 
2081   bool failed_to_read_elf_head=
2082     (sizeof(elf_head)!=
2083         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2084 
2085   ::close(file_descriptor);
2086   if (failed_to_read_elf_head) {
2087     // file i/o error - report dlerror() msg
2088     return NULL;
2089   }
2090 
2091   typedef struct {
2092     Elf32_Half  code;         // Actual value as defined in elf.h
2093     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
2094     char        elf_class;    // 32 or 64 bit
2095     char        endianess;    // MSB or LSB
2096     char*       name;         // String representation
2097   } arch_t;
2098 
2099   #ifndef EM_486
2100   #define EM_486          6               /* Intel 80486 */
2101   #endif
2102 
2103   #ifndef EM_MIPS_RS3_LE
2104   #define EM_MIPS_RS3_LE  10              /* MIPS */
2105   #endif
2106 
2107   #ifndef EM_PPC64
2108   #define EM_PPC64        21              /* PowerPC64 */
2109   #endif
2110 
2111   #ifndef EM_S390
2112   #define EM_S390         22              /* IBM System/390 */
2113   #endif
2114 
2115   #ifndef EM_IA_64
2116   #define EM_IA_64        50              /* HP/Intel IA-64 */
2117   #endif
2118 
2119   #ifndef EM_X86_64
2120   #define EM_X86_64       62              /* AMD x86-64 */
2121   #endif
2122 
2123   static const arch_t arch_array[]={
2124     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2125     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2126     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2127     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2128     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2129     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2130     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2131     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2132     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
2133     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
2134     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
2135     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
2136     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
2137     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
2138     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
2139     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
2140   };
2141 
2142   #if  (defined IA32)
2143     static  Elf32_Half running_arch_code=EM_386;
2144   #elif   (defined AMD64)
2145     static  Elf32_Half running_arch_code=EM_X86_64;
2146   #elif  (defined IA64)
2147     static  Elf32_Half running_arch_code=EM_IA_64;
2148   #elif  (defined __sparc) && (defined _LP64)
2149     static  Elf32_Half running_arch_code=EM_SPARCV9;
2150   #elif  (defined __sparc) && (!defined _LP64)
2151     static  Elf32_Half running_arch_code=EM_SPARC;
2152   #elif  (defined __powerpc64__)
2153     static  Elf32_Half running_arch_code=EM_PPC64;
2154   #elif  (defined __powerpc__)
2155     static  Elf32_Half running_arch_code=EM_PPC;
2156   #elif  (defined ARM)
2157     static  Elf32_Half running_arch_code=EM_ARM;
2158   #elif  (defined S390)
2159     static  Elf32_Half running_arch_code=EM_S390;
2160   #elif  (defined ALPHA)
2161     static  Elf32_Half running_arch_code=EM_ALPHA;
2162   #elif  (defined MIPSEL)
2163     static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
2164   #elif  (defined PARISC)
2165     static  Elf32_Half running_arch_code=EM_PARISC;
2166   #elif  (defined MIPS)
2167     static  Elf32_Half running_arch_code=EM_MIPS;
2168   #elif  (defined M68K)
2169     static  Elf32_Half running_arch_code=EM_68K;
2170   #else
2171     #error Method os::dll_load requires that one of following is defined:\
2172          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2173   #endif
2174 
2175   // Identify compatability class for VM's architecture and library's architecture
2176   // Obtain string descriptions for architectures
2177 
2178   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2179   int running_arch_index=-1;
2180 
2181   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2182     if (running_arch_code == arch_array[i].code) {
2183       running_arch_index    = i;
2184     }
2185     if (lib_arch.code == arch_array[i].code) {
2186       lib_arch.compat_class = arch_array[i].compat_class;
2187       lib_arch.name         = arch_array[i].name;
2188     }
2189   }
2190 
2191   assert(running_arch_index != -1,
2192     "Didn't find running architecture code (running_arch_code) in arch_array");
2193   if (running_arch_index == -1) {
2194     // Even though running architecture detection failed
2195     // we may still continue with reporting dlerror() message
2196     return NULL;
2197   }
2198 
2199   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2200     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2201     return NULL;
2202   }
2203 
2204 #ifndef S390
2205   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2206     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2207     return NULL;
2208   }
2209 #endif // !S390
2210 
2211   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2212     if ( lib_arch.name!=NULL ) {
2213       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2214         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2215         lib_arch.name, arch_array[running_arch_index].name);
2216     } else {
2217       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2218       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2219         lib_arch.code,
2220         arch_array[running_arch_index].name);
2221     }
2222   }
2223 
2224   return NULL;
2225 }
2226 #endif /* !__APPLE__ */
2227 
2228 // XXX: Do we need a lock around this as per Linux?
2229 void* os::dll_lookup(void* handle, const char* name) {
2230   return dlsym(handle, name);
2231 }
2232 
2233 
2234 static bool _print_ascii_file(const char* filename, outputStream* st) {
2235   int fd = ::open(filename, O_RDONLY);
2236   if (fd == -1) {
2237      return false;
2238   }
2239 
2240   char buf[32];
2241   int bytes;
2242   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2243     st->print_raw(buf, bytes);
2244   }
2245 
2246   ::close(fd);
2247 
2248   return true;
2249 }
2250 
2251 void os::print_dll_info(outputStream *st) {
2252    st->print_cr("Dynamic libraries:");
2253 #ifdef _ALLBSD_SOURCE
2254 #ifdef RTLD_DI_LINKMAP
2255     Dl_info dli;
2256     void *handle;
2257     Link_map *map;
2258     Link_map *p;
2259 
2260     if (!dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli)) {
2261         st->print_cr("Error: Cannot print dynamic libraries.");
2262         return;
2263     }
2264     handle = dlopen(dli.dli_fname, RTLD_LAZY);
2265     if (handle == NULL) {
2266         st->print_cr("Error: Cannot print dynamic libraries.");
2267         return;
2268     }
2269     dlinfo(handle, RTLD_DI_LINKMAP, &map);
2270     if (map == NULL) {
2271         st->print_cr("Error: Cannot print dynamic libraries.");
2272         return;
2273     }
2274 
2275     while (map->l_prev != NULL)
2276         map = map->l_prev;
2277 
2278     while (map != NULL) {
2279         st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
2280         map = map->l_next;
2281     }
2282 
2283     dlclose(handle);
2284 #elif defined(__APPLE__)
2285     uint32_t count;
2286     uint32_t i;
2287 
2288     count = _dyld_image_count();
2289     for (i = 1; i < count; i++) {
2290         const char *name = _dyld_get_image_name(i);
2291         intptr_t slide = _dyld_get_image_vmaddr_slide(i);
2292         st->print_cr(PTR_FORMAT " \t%s", slide, name);
2293     }
2294 #else
2295    st->print_cr("Error: Cannot print dynamic libraries.");
2296 #endif
2297 #else
2298    char fname[32];
2299    pid_t pid = os::Bsd::gettid();
2300 
2301    jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2302 
2303    if (!_print_ascii_file(fname, st)) {
2304      st->print("Can not get library information for pid = %d\n", pid);
2305    }
2306 #endif
2307 }
2308 
2309 
2310 void os::print_os_info(outputStream* st) {
2311   st->print("OS:");
2312 
2313   // Try to identify popular distros.
2314   // Most Bsd distributions have /etc/XXX-release file, which contains
2315   // the OS version string. Some have more than one /etc/XXX-release file
2316   // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2317   // so the order is important.
2318   if (!_print_ascii_file("/etc/mandrake-release", st) &&
2319       !_print_ascii_file("/etc/sun-release", st) &&
2320       !_print_ascii_file("/etc/redhat-release", st) &&
2321       !_print_ascii_file("/etc/SuSE-release", st) &&
2322       !_print_ascii_file("/etc/turbobsd-release", st) &&
2323       !_print_ascii_file("/etc/gentoo-release", st) &&
2324       !_print_ascii_file("/etc/debian_version", st) &&
2325       !_print_ascii_file("/etc/ltib-release", st) &&
2326       !_print_ascii_file("/etc/angstrom-version", st)) {
2327       st->print("Bsd");
2328   }
2329   st->cr();
2330 
2331   // kernel
2332   st->print("uname:");
2333   struct utsname name;
2334   uname(&name);
2335   st->print(name.sysname); st->print(" ");
2336   st->print(name.release); st->print(" ");
2337   st->print(name.version); st->print(" ");
2338   st->print(name.machine);
2339   st->cr();
2340 
2341 #ifndef _ALLBSD_SOURCE
2342   // Print warning if unsafe chroot environment detected
2343   if (unsafe_chroot_detected) {
2344     st->print("WARNING!! ");
2345     st->print_cr(unstable_chroot_error);
2346   }
2347 
2348   // libc, pthread
2349   st->print("libc:");
2350   st->print(os::Bsd::glibc_version()); st->print(" ");
2351   st->print(os::Bsd::libpthread_version()); st->print(" ");
2352   if (os::Bsd::is_BsdThreads()) {
2353      st->print("(%s stack)", os::Bsd::is_floating_stack() ? "floating" : "fixed");
2354   }
2355   st->cr();
2356 #endif
2357 
2358   // rlimit
2359   st->print("rlimit:");
2360   struct rlimit rlim;
2361 
2362   st->print(" STACK ");
2363   getrlimit(RLIMIT_STACK, &rlim);
2364   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2365   else st->print("%uk", rlim.rlim_cur >> 10);
2366 
2367   st->print(", CORE ");
2368   getrlimit(RLIMIT_CORE, &rlim);
2369   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2370   else st->print("%uk", rlim.rlim_cur >> 10);
2371 
2372   st->print(", NPROC ");
2373   getrlimit(RLIMIT_NPROC, &rlim);
2374   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2375   else st->print("%d", rlim.rlim_cur);
2376 
2377   st->print(", NOFILE ");
2378   getrlimit(RLIMIT_NOFILE, &rlim);
2379   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2380   else st->print("%d", rlim.rlim_cur);
2381 
2382 #ifndef _ALLBSD_SOURCE
2383   st->print(", AS ");
2384   getrlimit(RLIMIT_AS, &rlim);
2385   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2386   else st->print("%uk", rlim.rlim_cur >> 10);
2387   st->cr();
2388 
2389   // load average
2390   st->print("load average:");
2391   double loadavg[3];
2392   os::loadavg(loadavg, 3);
2393   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2394   st->cr();
2395 #endif
2396 }
2397 
2398 void os::pd_print_cpu_info(outputStream* st) {
2399   // Nothing to do for now.
2400 }
2401 
2402 void os::print_memory_info(outputStream* st) {
2403 
2404   st->print("Memory:");
2405   st->print(" %dk page", os::vm_page_size()>>10);
2406 
2407 #ifndef _ALLBSD_SOURCE
2408   // values in struct sysinfo are "unsigned long"
2409   struct sysinfo si;
2410   sysinfo(&si);
2411 #endif
2412 
2413   st->print(", physical " UINT64_FORMAT "k",
2414             os::physical_memory() >> 10);
2415   st->print("(" UINT64_FORMAT "k free)",
2416             os::available_memory() >> 10);
2417 #ifndef _ALLBSD_SOURCE
2418   st->print(", swap " UINT64_FORMAT "k",
2419             ((jlong)si.totalswap * si.mem_unit) >> 10);
2420   st->print("(" UINT64_FORMAT "k free)",
2421             ((jlong)si.freeswap * si.mem_unit) >> 10);
2422 #endif
2423   st->cr();
2424 
2425   // meminfo
2426   st->print("\n/proc/meminfo:\n");
2427   _print_ascii_file("/proc/meminfo", st);
2428   st->cr();
2429 }
2430 
2431 // Taken from /usr/include/bits/siginfo.h  Supposed to be architecture specific
2432 // but they're the same for all the bsd arch that we support
2433 // and they're the same for solaris but there's no common place to put this.
2434 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2435                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2436                           "ILL_COPROC", "ILL_BADSTK" };
2437 
2438 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2439                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2440                           "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2441 
2442 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2443 
2444 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2445 
2446 void os::print_siginfo(outputStream* st, void* siginfo) {
2447   st->print("siginfo:");
2448 
2449   const int buflen = 100;
2450   char buf[buflen];
2451   siginfo_t *si = (siginfo_t*)siginfo;
2452   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2453   if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2454     st->print("si_errno=%s", buf);
2455   } else {
2456     st->print("si_errno=%d", si->si_errno);
2457   }
2458   const int c = si->si_code;
2459   assert(c > 0, "unexpected si_code");
2460   switch (si->si_signo) {
2461   case SIGILL:
2462     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2463     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2464     break;
2465   case SIGFPE:
2466     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2467     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2468     break;
2469   case SIGSEGV:
2470     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2471     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2472     break;
2473   case SIGBUS:
2474     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2475     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2476     break;
2477   default:
2478     st->print(", si_code=%d", si->si_code);
2479     // no si_addr
2480   }
2481 
2482   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2483       UseSharedSpaces) {
2484     FileMapInfo* mapinfo = FileMapInfo::current_info();
2485     if (mapinfo->is_in_shared_space(si->si_addr)) {
2486       st->print("\n\nError accessing class data sharing archive."   \
2487                 " Mapped file inaccessible during execution, "      \
2488                 " possible disk/network problem.");
2489     }
2490   }
2491   st->cr();
2492 }
2493 
2494 
2495 static void print_signal_handler(outputStream* st, int sig,
2496                                  char* buf, size_t buflen);
2497 
2498 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2499   st->print_cr("Signal Handlers:");
2500   print_signal_handler(st, SIGSEGV, buf, buflen);
2501   print_signal_handler(st, SIGBUS , buf, buflen);
2502   print_signal_handler(st, SIGFPE , buf, buflen);
2503   print_signal_handler(st, SIGPIPE, buf, buflen);
2504   print_signal_handler(st, SIGXFSZ, buf, buflen);
2505   print_signal_handler(st, SIGILL , buf, buflen);
2506   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2507   print_signal_handler(st, SR_signum, buf, buflen);
2508   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2509   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2510   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2511   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2512 }
2513 
2514 static char saved_jvm_path[MAXPATHLEN] = {0};
2515 
2516 // Find the full path to the current module, libjvm or libjvm_g
2517 void os::jvm_path(char *buf, jint buflen) {
2518   // Error checking.
2519   if (buflen < MAXPATHLEN) {
2520     assert(false, "must use a large-enough buffer");
2521     buf[0] = '\0';
2522     return;
2523   }
2524   // Lazy resolve the path to current module.
2525   if (saved_jvm_path[0] != 0) {
2526     strcpy(buf, saved_jvm_path);
2527     return;
2528   }
2529 
2530   char dli_fname[MAXPATHLEN];
2531   bool ret = dll_address_to_library_name(
2532                 CAST_FROM_FN_PTR(address, os::jvm_path),
2533                 dli_fname, sizeof(dli_fname), NULL);
2534   assert(ret != 0, "cannot locate libjvm");
2535   char *rp = realpath(dli_fname, buf);
2536   if (rp == NULL)
2537     return;
2538 
2539   if (Arguments::created_by_gamma_launcher()) {
2540     // Support for the gamma launcher.  Typical value for buf is
2541     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm".  If "/jre/lib/" appears at
2542     // the right place in the string, then assume we are installed in a JDK and
2543     // we're done.  Otherwise, check for a JAVA_HOME environment variable and
2544     // construct a path to the JVM being overridden.
2545 
2546     const char *p = buf + strlen(buf) - 1;
2547     for (int count = 0; p > buf && count < 5; ++count) {
2548       for (--p; p > buf && *p != '/'; --p)
2549         /* empty */ ;
2550     }
2551 
2552     if (strncmp(p, "/jre/lib/", 9) != 0) {
2553       // Look for JAVA_HOME in the environment.
2554       char* java_home_var = ::getenv("JAVA_HOME");
2555       if (java_home_var != NULL && java_home_var[0] != 0) {
2556         char* jrelib_p;
2557         int len;
2558 
2559         // Check the current module name "libjvm" or "libjvm_g".
2560         p = strrchr(buf, '/');
2561         assert(strstr(p, "/libjvm") == p, "invalid library name");
2562         p = strstr(p, "_g") ? "_g" : "";
2563 
2564         rp = realpath(java_home_var, buf);
2565         if (rp == NULL)
2566           return;
2567 
2568         // determine if this is a legacy image or modules image
2569         // modules image doesn't have "jre" subdirectory
2570         len = strlen(buf);
2571         jrelib_p = buf + len;
2572 
2573         // Add the appropriate library subdir
2574         snprintf(jrelib_p, buflen-len, "/jre/lib");
2575         if (0 != access(buf, F_OK)) {
2576           snprintf(jrelib_p, buflen-len, "/lib");
2577         }
2578 
2579         // Add the appropriate client or server subdir
2580         len = strlen(buf);
2581         jrelib_p = buf + len;
2582         snprintf(jrelib_p, buflen-len, "/%s", COMPILER_VARIANT);
2583         if (0 != access(buf, F_OK)) {
2584           snprintf(jrelib_p, buflen-len, "");
2585         }
2586 
2587         // If the path exists within JAVA_HOME, add the JVM library name
2588         // to complete the path to JVM being overridden.  Otherwise fallback
2589         // to the path to the current library.
2590         if (0 == access(buf, F_OK)) {
2591           // Use current module name "libjvm[_g]" instead of
2592           // "libjvm"debug_only("_g")"" since for fastdebug version
2593           // we should have "libjvm" but debug_only("_g") adds "_g"!
2594           len = strlen(buf);
2595           snprintf(buf + len, buflen-len, "/libjvm%s%s", p, JNI_LIB_SUFFIX);
2596         } else {
2597           // Fall back to path of current library
2598           rp = realpath(dli_fname, buf);
2599           if (rp == NULL)
2600             return;
2601         }
2602       }
2603     }
2604   }
2605 
2606   strcpy(saved_jvm_path, buf);
2607 }
2608 
2609 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2610   // no prefix required, not even "_"
2611 }
2612 
2613 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2614   // no suffix required
2615 }
2616 
2617 ////////////////////////////////////////////////////////////////////////////////
2618 // sun.misc.Signal support
2619 
2620 static volatile jint sigint_count = 0;
2621 
2622 static void
2623 UserHandler(int sig, void *siginfo, void *context) {
2624   // 4511530 - sem_post is serialized and handled by the manager thread. When
2625   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2626   // don't want to flood the manager thread with sem_post requests.
2627   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2628       return;
2629 
2630   // Ctrl-C is pressed during error reporting, likely because the error
2631   // handler fails to abort. Let VM die immediately.
2632   if (sig == SIGINT && is_error_reported()) {
2633      os::die();
2634   }
2635 
2636   os::signal_notify(sig);
2637 }
2638 
2639 void* os::user_handler() {
2640   return CAST_FROM_FN_PTR(void*, UserHandler);
2641 }
2642 
2643 extern "C" {
2644   typedef void (*sa_handler_t)(int);
2645   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2646 }
2647 
2648 void* os::signal(int signal_number, void* handler) {
2649   struct sigaction sigAct, oldSigAct;
2650 
2651   sigfillset(&(sigAct.sa_mask));
2652   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2653   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2654 
2655   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2656     // -1 means registration failed
2657     return (void *)-1;
2658   }
2659 
2660   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2661 }
2662 
2663 void os::signal_raise(int signal_number) {
2664   ::raise(signal_number);
2665 }
2666 
2667 /*
2668  * The following code is moved from os.cpp for making this
2669  * code platform specific, which it is by its very nature.
2670  */
2671 
2672 // Will be modified when max signal is changed to be dynamic
2673 int os::sigexitnum_pd() {
2674   return NSIG;
2675 }
2676 
2677 // a counter for each possible signal value
2678 static volatile jint pending_signals[NSIG+1] = { 0 };
2679 
2680 // Bsd(POSIX) specific hand shaking semaphore.
2681 #ifdef __APPLE__
2682 static semaphore_t sig_sem;
2683 #define SEM_INIT(sem, value)    semaphore_create(mach_task_self(), &sem, SYNC_POLICY_FIFO, value)
2684 #define SEM_WAIT(sem)           semaphore_wait(sem);
2685 #define SEM_POST(sem)           semaphore_signal(sem);
2686 #else
2687 static sem_t sig_sem;
2688 #define SEM_INIT(sem, value)    sem_init(&sem, 0, value)
2689 #define SEM_WAIT(sem)           sem_wait(&sem);
2690 #define SEM_POST(sem)           sem_post(&sem);
2691 #endif
2692 
2693 void os::signal_init_pd() {
2694   // Initialize signal structures
2695   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2696 
2697   // Initialize signal semaphore
2698   ::SEM_INIT(sig_sem, 0);
2699 }
2700 
2701 void os::signal_notify(int sig) {
2702   Atomic::inc(&pending_signals[sig]);
2703   ::SEM_POST(sig_sem);
2704 }
2705 
2706 static int check_pending_signals(bool wait) {
2707   Atomic::store(0, &sigint_count);
2708   for (;;) {
2709     for (int i = 0; i < NSIG + 1; i++) {
2710       jint n = pending_signals[i];
2711       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2712         return i;
2713       }
2714     }
2715     if (!wait) {
2716       return -1;
2717     }
2718     JavaThread *thread = JavaThread::current();
2719     ThreadBlockInVM tbivm(thread);
2720 
2721     bool threadIsSuspended;
2722     do {
2723       thread->set_suspend_equivalent();
2724       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2725       ::SEM_WAIT(sig_sem);
2726 
2727       // were we externally suspended while we were waiting?
2728       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2729       if (threadIsSuspended) {
2730         //
2731         // The semaphore has been incremented, but while we were waiting
2732         // another thread suspended us. We don't want to continue running
2733         // while suspended because that would surprise the thread that
2734         // suspended us.
2735         //
2736         ::SEM_POST(sig_sem);
2737 
2738         thread->java_suspend_self();
2739       }
2740     } while (threadIsSuspended);
2741   }
2742 }
2743 
2744 int os::signal_lookup() {
2745   return check_pending_signals(false);
2746 }
2747 
2748 int os::signal_wait() {
2749   return check_pending_signals(true);
2750 }
2751 
2752 ////////////////////////////////////////////////////////////////////////////////
2753 // Virtual Memory
2754 
2755 int os::vm_page_size() {
2756   // Seems redundant as all get out
2757   assert(os::Bsd::page_size() != -1, "must call os::init");
2758   return os::Bsd::page_size();
2759 }
2760 
2761 // Solaris allocates memory by pages.
2762 int os::vm_allocation_granularity() {
2763   assert(os::Bsd::page_size() != -1, "must call os::init");
2764   return os::Bsd::page_size();
2765 }
2766 
2767 // Rationale behind this function:
2768 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2769 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2770 //  samples for JITted code. Here we create private executable mapping over the code cache
2771 //  and then we can use standard (well, almost, as mapping can change) way to provide
2772 //  info for the reporting script by storing timestamp and location of symbol
2773 void bsd_wrap_code(char* base, size_t size) {
2774   static volatile jint cnt = 0;
2775 
2776   if (!UseOprofile) {
2777     return;
2778   }
2779 
2780   char buf[PATH_MAX + 1];
2781   int num = Atomic::add(1, &cnt);
2782 
2783   snprintf(buf, PATH_MAX + 1, "%s/hs-vm-%d-%d",
2784            os::get_temp_directory(), os::current_process_id(), num);
2785   unlink(buf);
2786 
2787   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2788 
2789   if (fd != -1) {
2790     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2791     if (rv != (off_t)-1) {
2792       if (::write(fd, "", 1) == 1) {
2793         mmap(base, size,
2794              PROT_READ|PROT_WRITE|PROT_EXEC,
2795              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2796       }
2797     }
2798     ::close(fd);
2799     unlink(buf);
2800   }
2801 }
2802 
2803 // NOTE: Bsd kernel does not really reserve the pages for us.
2804 //       All it does is to check if there are enough free pages
2805 //       left at the time of mmap(). This could be a potential
2806 //       problem.
2807 bool os::commit_memory(char* addr, size_t size, bool exec) {
2808   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2809 #ifdef __OpenBSD__
2810   // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
2811   return ::mprotect(addr, size, prot) == 0;
2812 #else
2813   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2814                                    MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2815   return res != (uintptr_t) MAP_FAILED;
2816 #endif
2817 }
2818 
2819 #ifndef _ALLBSD_SOURCE
2820 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2821 #ifndef MAP_HUGETLB
2822 #define MAP_HUGETLB 0x40000
2823 #endif
2824 
2825 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2826 #ifndef MADV_HUGEPAGE
2827 #define MADV_HUGEPAGE 14
2828 #endif
2829 #endif
2830 
2831 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2832                        bool exec) {
2833 #ifndef _ALLBSD_SOURCE
2834   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2835     int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2836     uintptr_t res =
2837       (uintptr_t) ::mmap(addr, size, prot,
2838                          MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2839                          -1, 0);
2840     return res != (uintptr_t) MAP_FAILED;
2841   }
2842 #endif
2843 
2844   return commit_memory(addr, size, exec);
2845 }
2846 
2847 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2848 #ifndef _ALLBSD_SOURCE
2849   if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2850     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2851     // be supported or the memory may already be backed by huge pages.
2852     ::madvise(addr, bytes, MADV_HUGEPAGE);
2853   }
2854 #endif
2855 }
2856 
2857 void os::free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2858   ::madvise(addr, bytes, MADV_DONTNEED);
2859 }
2860 
2861 void os::numa_make_global(char *addr, size_t bytes) {
2862 }
2863 
2864 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2865 }
2866 
2867 bool os::numa_topology_changed()   { return false; }
2868 
2869 size_t os::numa_get_groups_num() {
2870   return 1;
2871 }
2872 
2873 int os::numa_get_group_id() {
2874   return 0;
2875 }
2876 
2877 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2878   if (size > 0) {
2879     ids[0] = 0;
2880     return 1;
2881   }
2882   return 0;
2883 }
2884 
2885 bool os::get_page_info(char *start, page_info* info) {
2886   return false;
2887 }
2888 
2889 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2890   return end;
2891 }
2892 
2893 #ifndef _ALLBSD_SOURCE
2894 // Something to do with the numa-aware allocator needs these symbols
2895 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2896 extern "C" JNIEXPORT void numa_error(char *where) { }
2897 extern "C" JNIEXPORT int fork1() { return fork(); }
2898 
2899 
2900 // If we are running with libnuma version > 2, then we should
2901 // be trying to use symbols with versions 1.1
2902 // If we are running with earlier version, which did not have symbol versions,
2903 // we should use the base version.
2904 void* os::Bsd::libnuma_dlsym(void* handle, const char *name) {
2905   void *f = dlvsym(handle, name, "libnuma_1.1");
2906   if (f == NULL) {
2907     f = dlsym(handle, name);
2908   }
2909   return f;
2910 }
2911 
2912 bool os::Bsd::libnuma_init() {
2913   // sched_getcpu() should be in libc.
2914   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2915                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2916 
2917   if (sched_getcpu() != -1) { // Does it work?
2918     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2919     if (handle != NULL) {
2920       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2921                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2922       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2923                                        libnuma_dlsym(handle, "numa_max_node")));
2924       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2925                                         libnuma_dlsym(handle, "numa_available")));
2926       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2927                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2928       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2929                                             libnuma_dlsym(handle, "numa_interleave_memory")));
2930 
2931 
2932       if (numa_available() != -1) {
2933         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2934         // Create a cpu -> node mapping
2935         _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2936         rebuild_cpu_to_node_map();
2937         return true;
2938       }
2939     }
2940   }
2941   return false;
2942 }
2943 
2944 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2945 // The table is later used in get_node_by_cpu().
2946 void os::Bsd::rebuild_cpu_to_node_map() {
2947   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2948                               // in libnuma (possible values are starting from 16,
2949                               // and continuing up with every other power of 2, but less
2950                               // than the maximum number of CPUs supported by kernel), and
2951                               // is a subject to change (in libnuma version 2 the requirements
2952                               // are more reasonable) we'll just hardcode the number they use
2953                               // in the library.
2954   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2955 
2956   size_t cpu_num = os::active_processor_count();
2957   size_t cpu_map_size = NCPUS / BitsPerCLong;
2958   size_t cpu_map_valid_size =
2959     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2960 
2961   cpu_to_node()->clear();
2962   cpu_to_node()->at_grow(cpu_num - 1);
2963   size_t node_num = numa_get_groups_num();
2964 
2965   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2966   for (size_t i = 0; i < node_num; i++) {
2967     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2968       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2969         if (cpu_map[j] != 0) {
2970           for (size_t k = 0; k < BitsPerCLong; k++) {
2971             if (cpu_map[j] & (1UL << k)) {
2972               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2973             }
2974           }
2975         }
2976       }
2977     }
2978   }
2979   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2980 }
2981 
2982 int os::Bsd::get_node_by_cpu(int cpu_id) {
2983   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2984     return cpu_to_node()->at(cpu_id);
2985   }
2986   return -1;
2987 }
2988 
2989 GrowableArray<int>* os::Bsd::_cpu_to_node;
2990 os::Bsd::sched_getcpu_func_t os::Bsd::_sched_getcpu;
2991 os::Bsd::numa_node_to_cpus_func_t os::Bsd::_numa_node_to_cpus;
2992 os::Bsd::numa_max_node_func_t os::Bsd::_numa_max_node;
2993 os::Bsd::numa_available_func_t os::Bsd::_numa_available;
2994 os::Bsd::numa_tonode_memory_func_t os::Bsd::_numa_tonode_memory;
2995 os::Bsd::numa_interleave_memory_func_t os::Bsd::_numa_interleave_memory;
2996 unsigned long* os::Bsd::_numa_all_nodes;
2997 #endif
2998 
2999 bool os::uncommit_memory(char* addr, size_t size) {
3000 #ifdef __OpenBSD__
3001   // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
3002   return ::mprotect(addr, size, PROT_NONE) == 0;
3003 #else
3004   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3005                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3006   return res  != (uintptr_t) MAP_FAILED;
3007 #endif
3008 }
3009 
3010 bool os::create_stack_guard_pages(char* addr, size_t size) {
3011   return os::commit_memory(addr, size);
3012 }
3013 
3014 // If this is a growable mapping, remove the guard pages entirely by
3015 // munmap()ping them.  If not, just call uncommit_memory().
3016 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3017   return os::uncommit_memory(addr, size);
3018 }
3019 
3020 static address _highest_vm_reserved_address = NULL;
3021 
3022 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3023 // at 'requested_addr'. If there are existing memory mappings at the same
3024 // location, however, they will be overwritten. If 'fixed' is false,
3025 // 'requested_addr' is only treated as a hint, the return value may or
3026 // may not start from the requested address. Unlike Bsd mmap(), this
3027 // function returns NULL to indicate failure.
3028 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3029   char * addr;
3030   int flags;
3031 
3032   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3033   if (fixed) {
3034     assert((uintptr_t)requested_addr % os::Bsd::page_size() == 0, "unaligned address");
3035     flags |= MAP_FIXED;
3036   }
3037 
3038   // Map uncommitted pages PROT_READ and PROT_WRITE, change access
3039   // to PROT_EXEC if executable when we commit the page.
3040   addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
3041                        flags, -1, 0);
3042 
3043   if (addr != MAP_FAILED) {
3044     // anon_mmap() should only get called during VM initialization,
3045     // don't need lock (actually we can skip locking even it can be called
3046     // from multiple threads, because _highest_vm_reserved_address is just a
3047     // hint about the upper limit of non-stack memory regions.)
3048     if ((address)addr + bytes > _highest_vm_reserved_address) {
3049       _highest_vm_reserved_address = (address)addr + bytes;
3050     }
3051   }
3052 
3053   return addr == MAP_FAILED ? NULL : addr;
3054 }
3055 
3056 // Don't update _highest_vm_reserved_address, because there might be memory
3057 // regions above addr + size. If so, releasing a memory region only creates
3058 // a hole in the address space, it doesn't help prevent heap-stack collision.
3059 //
3060 static int anon_munmap(char * addr, size_t size) {
3061   return ::munmap(addr, size) == 0;
3062 }
3063 
3064 char* os::reserve_memory(size_t bytes, char* requested_addr,
3065                          size_t alignment_hint) {
3066   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3067 }
3068 
3069 bool os::release_memory(char* addr, size_t size) {
3070   return anon_munmap(addr, size);
3071 }
3072 
3073 static address highest_vm_reserved_address() {
3074   return _highest_vm_reserved_address;
3075 }
3076 
3077 static bool bsd_mprotect(char* addr, size_t size, int prot) {
3078   // Bsd wants the mprotect address argument to be page aligned.
3079   char* bottom = (char*)align_size_down((intptr_t)addr, os::Bsd::page_size());
3080 
3081   // According to SUSv3, mprotect() should only be used with mappings
3082   // established by mmap(), and mmap() always maps whole pages. Unaligned
3083   // 'addr' likely indicates problem in the VM (e.g. trying to change
3084   // protection of malloc'ed or statically allocated memory). Check the
3085   // caller if you hit this assert.
3086   assert(addr == bottom, "sanity check");
3087 
3088   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Bsd::page_size());
3089   return ::mprotect(bottom, size, prot) == 0;
3090 }
3091 
3092 // Set protections specified
3093 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3094                         bool is_committed) {
3095   unsigned int p = 0;
3096   switch (prot) {
3097   case MEM_PROT_NONE: p = PROT_NONE; break;
3098   case MEM_PROT_READ: p = PROT_READ; break;
3099   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3100   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3101   default:
3102     ShouldNotReachHere();
3103   }
3104   // is_committed is unused.
3105   return bsd_mprotect(addr, bytes, p);
3106 }
3107 
3108 bool os::guard_memory(char* addr, size_t size) {
3109   return bsd_mprotect(addr, size, PROT_NONE);
3110 }
3111 
3112 bool os::unguard_memory(char* addr, size_t size) {
3113   return bsd_mprotect(addr, size, PROT_READ|PROT_WRITE);
3114 }
3115 
3116 bool os::Bsd::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3117   bool result = false;
3118 #ifndef _ALLBSD_SOURCE
3119   void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
3120                   MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3121                   -1, 0);
3122 
3123   if (p != (void *) -1) {
3124     // We don't know if this really is a huge page or not.
3125     FILE *fp = fopen("/proc/self/maps", "r");
3126     if (fp) {
3127       while (!feof(fp)) {
3128         char chars[257];
3129         long x = 0;
3130         if (fgets(chars, sizeof(chars), fp)) {
3131           if (sscanf(chars, "%lx-%*x", &x) == 1
3132               && x == (long)p) {
3133             if (strstr (chars, "hugepage")) {
3134               result = true;
3135               break;
3136             }
3137           }
3138         }
3139       }
3140       fclose(fp);
3141     }
3142     munmap (p, page_size);
3143     if (result)
3144       return true;
3145   }
3146 
3147   if (warn) {
3148     warning("HugeTLBFS is not supported by the operating system.");
3149   }
3150 #endif
3151 
3152   return result;
3153 }
3154 
3155 /*
3156 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3157 *
3158 * From the coredump_filter documentation:
3159 *
3160 * - (bit 0) anonymous private memory
3161 * - (bit 1) anonymous shared memory
3162 * - (bit 2) file-backed private memory
3163 * - (bit 3) file-backed shared memory
3164 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3165 *           effective only if the bit 2 is cleared)
3166 * - (bit 5) hugetlb private memory
3167 * - (bit 6) hugetlb shared memory
3168 */
3169 static void set_coredump_filter(void) {
3170   FILE *f;
3171   long cdm;
3172 
3173   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3174     return;
3175   }
3176 
3177   if (fscanf(f, "%lx", &cdm) != 1) {
3178     fclose(f);
3179     return;
3180   }
3181 
3182   rewind(f);
3183 
3184   if ((cdm & LARGEPAGES_BIT) == 0) {
3185     cdm |= LARGEPAGES_BIT;
3186     fprintf(f, "%#lx", cdm);
3187   }
3188 
3189   fclose(f);
3190 }
3191 
3192 // Large page support
3193 
3194 static size_t _large_page_size = 0;
3195 
3196 void os::large_page_init() {
3197 #ifndef _ALLBSD_SOURCE
3198   if (!UseLargePages) {
3199     UseHugeTLBFS = false;
3200     UseSHM = false;
3201     return;
3202   }
3203 
3204   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3205     // If UseLargePages is specified on the command line try both methods,
3206     // if it's default, then try only HugeTLBFS.
3207     if (FLAG_IS_DEFAULT(UseLargePages)) {
3208       UseHugeTLBFS = true;
3209     } else {
3210       UseHugeTLBFS = UseSHM = true;
3211     }
3212   }
3213 
3214   if (LargePageSizeInBytes) {
3215     _large_page_size = LargePageSizeInBytes;
3216   } else {
3217     // large_page_size on Bsd is used to round up heap size. x86 uses either
3218     // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3219     // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3220     // page as large as 256M.
3221     //
3222     // Here we try to figure out page size by parsing /proc/meminfo and looking
3223     // for a line with the following format:
3224     //    Hugepagesize:     2048 kB
3225     //
3226     // If we can't determine the value (e.g. /proc is not mounted, or the text
3227     // format has been changed), we'll use the largest page size supported by
3228     // the processor.
3229 
3230 #ifndef ZERO
3231     _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3232                        ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3233 #endif // ZERO
3234 
3235     FILE *fp = fopen("/proc/meminfo", "r");
3236     if (fp) {
3237       while (!feof(fp)) {
3238         int x = 0;
3239         char buf[16];
3240         if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3241           if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3242             _large_page_size = x * K;
3243             break;
3244           }
3245         } else {
3246           // skip to next line
3247           for (;;) {
3248             int ch = fgetc(fp);
3249             if (ch == EOF || ch == (int)'\n') break;
3250           }
3251         }
3252       }
3253       fclose(fp);
3254     }
3255   }
3256 
3257   // print a warning if any large page related flag is specified on command line
3258   bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3259 
3260   const size_t default_page_size = (size_t)Bsd::page_size();
3261   if (_large_page_size > default_page_size) {
3262     _page_sizes[0] = _large_page_size;
3263     _page_sizes[1] = default_page_size;
3264     _page_sizes[2] = 0;
3265   }
3266   UseHugeTLBFS = UseHugeTLBFS &&
3267                  Bsd::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3268 
3269   if (UseHugeTLBFS)
3270     UseSHM = false;
3271 
3272   UseLargePages = UseHugeTLBFS || UseSHM;
3273 
3274   set_coredump_filter();
3275 #endif
3276 }
3277 
3278 #ifndef _ALLBSD_SOURCE
3279 #ifndef SHM_HUGETLB
3280 #define SHM_HUGETLB 04000
3281 #endif
3282 #endif
3283 
3284 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3285   // "exec" is passed in but not used.  Creating the shared image for
3286   // the code cache doesn't have an SHM_X executable permission to check.
3287   assert(UseLargePages && UseSHM, "only for SHM large pages");
3288 
3289   key_t key = IPC_PRIVATE;
3290   char *addr;
3291 
3292   bool warn_on_failure = UseLargePages &&
3293                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3294                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3295                         );
3296   char msg[128];
3297 
3298   // Create a large shared memory region to attach to based on size.
3299   // Currently, size is the total size of the heap
3300 #ifndef _ALLBSD_SOURCE
3301   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3302 #else
3303   int shmid = shmget(key, bytes, IPC_CREAT|SHM_R|SHM_W);
3304 #endif
3305   if (shmid == -1) {
3306      // Possible reasons for shmget failure:
3307      // 1. shmmax is too small for Java heap.
3308      //    > check shmmax value: cat /proc/sys/kernel/shmmax
3309      //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3310      // 2. not enough large page memory.
3311      //    > check available large pages: cat /proc/meminfo
3312      //    > increase amount of large pages:
3313      //          echo new_value > /proc/sys/vm/nr_hugepages
3314      //      Note 1: different Bsd may use different name for this property,
3315      //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3316      //      Note 2: it's possible there's enough physical memory available but
3317      //            they are so fragmented after a long run that they can't
3318      //            coalesce into large pages. Try to reserve large pages when
3319      //            the system is still "fresh".
3320      if (warn_on_failure) {
3321        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3322        warning(msg);
3323      }
3324      return NULL;
3325   }
3326 
3327   // attach to the region
3328   addr = (char*)shmat(shmid, req_addr, 0);
3329   int err = errno;
3330 
3331   // Remove shmid. If shmat() is successful, the actual shared memory segment
3332   // will be deleted when it's detached by shmdt() or when the process
3333   // terminates. If shmat() is not successful this will remove the shared
3334   // segment immediately.
3335   shmctl(shmid, IPC_RMID, NULL);
3336 
3337   if ((intptr_t)addr == -1) {
3338      if (warn_on_failure) {
3339        jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3340        warning(msg);
3341      }
3342      return NULL;
3343   }
3344 
3345   return addr;
3346 }
3347 
3348 bool os::release_memory_special(char* base, size_t bytes) {
3349   // detaching the SHM segment will also delete it, see reserve_memory_special()
3350   int rslt = shmdt(base);
3351   return rslt == 0;
3352 }
3353 
3354 size_t os::large_page_size() {
3355   return _large_page_size;
3356 }
3357 
3358 // HugeTLBFS allows application to commit large page memory on demand;
3359 // with SysV SHM the entire memory region must be allocated as shared
3360 // memory.
3361 bool os::can_commit_large_page_memory() {
3362   return UseHugeTLBFS;
3363 }
3364 
3365 bool os::can_execute_large_page_memory() {
3366   return UseHugeTLBFS;
3367 }
3368 
3369 // Reserve memory at an arbitrary address, only if that area is
3370 // available (and not reserved for something else).
3371 
3372 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3373   const int max_tries = 10;
3374   char* base[max_tries];
3375   size_t size[max_tries];
3376   const size_t gap = 0x000000;
3377 
3378   // Assert only that the size is a multiple of the page size, since
3379   // that's all that mmap requires, and since that's all we really know
3380   // about at this low abstraction level.  If we need higher alignment,
3381   // we can either pass an alignment to this method or verify alignment
3382   // in one of the methods further up the call chain.  See bug 5044738.
3383   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3384 
3385   // Repeatedly allocate blocks until the block is allocated at the
3386   // right spot. Give up after max_tries. Note that reserve_memory() will
3387   // automatically update _highest_vm_reserved_address if the call is
3388   // successful. The variable tracks the highest memory address every reserved
3389   // by JVM. It is used to detect heap-stack collision if running with
3390   // fixed-stack BsdThreads. Because here we may attempt to reserve more
3391   // space than needed, it could confuse the collision detecting code. To
3392   // solve the problem, save current _highest_vm_reserved_address and
3393   // calculate the correct value before return.
3394   address old_highest = _highest_vm_reserved_address;
3395 
3396   // Bsd mmap allows caller to pass an address as hint; give it a try first,
3397   // if kernel honors the hint then we can return immediately.
3398   char * addr = anon_mmap(requested_addr, bytes, false);
3399   if (addr == requested_addr) {
3400      return requested_addr;
3401   }
3402 
3403   if (addr != NULL) {
3404      // mmap() is successful but it fails to reserve at the requested address
3405      anon_munmap(addr, bytes);
3406   }
3407 
3408   int i;
3409   for (i = 0; i < max_tries; ++i) {
3410     base[i] = reserve_memory(bytes);
3411 
3412     if (base[i] != NULL) {
3413       // Is this the block we wanted?
3414       if (base[i] == requested_addr) {
3415         size[i] = bytes;
3416         break;
3417       }
3418 
3419       // Does this overlap the block we wanted? Give back the overlapped
3420       // parts and try again.
3421 
3422       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3423       if (top_overlap >= 0 && top_overlap < bytes) {
3424         unmap_memory(base[i], top_overlap);
3425         base[i] += top_overlap;
3426         size[i] = bytes - top_overlap;
3427       } else {
3428         size_t bottom_overlap = base[i] + bytes - requested_addr;
3429         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3430           unmap_memory(requested_addr, bottom_overlap);
3431           size[i] = bytes - bottom_overlap;
3432         } else {
3433           size[i] = bytes;
3434         }
3435       }
3436     }
3437   }
3438 
3439   // Give back the unused reserved pieces.
3440 
3441   for (int j = 0; j < i; ++j) {
3442     if (base[j] != NULL) {
3443       unmap_memory(base[j], size[j]);
3444     }
3445   }
3446 
3447   if (i < max_tries) {
3448     _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3449     return requested_addr;
3450   } else {
3451     _highest_vm_reserved_address = old_highest;
3452     return NULL;
3453   }
3454 }
3455 
3456 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3457   RESTARTABLE_RETURN_INT(::read(fd, buf, nBytes));
3458 }
3459 
3460 // TODO-FIXME: reconcile Solaris' os::sleep with the bsd variation.
3461 // Solaris uses poll(), bsd uses park().
3462 // Poll() is likely a better choice, assuming that Thread.interrupt()
3463 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3464 // SIGSEGV, see 4355769.
3465 
3466 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3467   assert(thread == Thread::current(),  "thread consistency check");
3468 
3469   ParkEvent * const slp = thread->_SleepEvent ;
3470   slp->reset() ;
3471   OrderAccess::fence() ;
3472 
3473   if (interruptible) {
3474     jlong prevtime = javaTimeNanos();
3475 
3476     for (;;) {
3477       if (os::is_interrupted(thread, true)) {
3478         return OS_INTRPT;
3479       }
3480 
3481       jlong newtime = javaTimeNanos();
3482 
3483       if (newtime - prevtime < 0) {
3484         // time moving backwards, should only happen if no monotonic clock
3485         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3486         assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
3487       } else {
3488         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3489       }
3490 
3491       if(millis <= 0) {
3492         return OS_OK;
3493       }
3494 
3495       prevtime = newtime;
3496 
3497       {
3498         assert(thread->is_Java_thread(), "sanity check");
3499         JavaThread *jt = (JavaThread *) thread;
3500         ThreadBlockInVM tbivm(jt);
3501         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3502 
3503         jt->set_suspend_equivalent();
3504         // cleared by handle_special_suspend_equivalent_condition() or
3505         // java_suspend_self() via check_and_wait_while_suspended()
3506 
3507         slp->park(millis);
3508 
3509         // were we externally suspended while we were waiting?
3510         jt->check_and_wait_while_suspended();
3511       }
3512     }
3513   } else {
3514     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3515     jlong prevtime = javaTimeNanos();
3516 
3517     for (;;) {
3518       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3519       // the 1st iteration ...
3520       jlong newtime = javaTimeNanos();
3521 
3522       if (newtime - prevtime < 0) {
3523         // time moving backwards, should only happen if no monotonic clock
3524         // not a guarantee() because JVM should not abort on kernel/glibc bugs
3525         assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
3526       } else {
3527         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3528       }
3529 
3530       if(millis <= 0) break ;
3531 
3532       prevtime = newtime;
3533       slp->park(millis);
3534     }
3535     return OS_OK ;
3536   }
3537 }
3538 
3539 int os::naked_sleep() {
3540   // %% make the sleep time an integer flag. for now use 1 millisec.
3541   return os::sleep(Thread::current(), 1, false);
3542 }
3543 
3544 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3545 void os::infinite_sleep() {
3546   while (true) {    // sleep forever ...
3547     ::sleep(100);   // ... 100 seconds at a time
3548   }
3549 }
3550 
3551 // Used to convert frequent JVM_Yield() to nops
3552 bool os::dont_yield() {
3553   return DontYieldALot;
3554 }
3555 
3556 void os::yield() {
3557   sched_yield();
3558 }
3559 
3560 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3561 
3562 void os::yield_all(int attempts) {
3563   // Yields to all threads, including threads with lower priorities
3564   // Threads on Bsd are all with same priority. The Solaris style
3565   // os::yield_all() with nanosleep(1ms) is not necessary.
3566   sched_yield();
3567 }
3568 
3569 // Called from the tight loops to possibly influence time-sharing heuristics
3570 void os::loop_breaker(int attempts) {
3571   os::yield_all(attempts);
3572 }
3573 
3574 ////////////////////////////////////////////////////////////////////////////////
3575 // thread priority support
3576 
3577 // Note: Normal Bsd applications are run with SCHED_OTHER policy. SCHED_OTHER
3578 // only supports dynamic priority, static priority must be zero. For real-time
3579 // applications, Bsd supports SCHED_RR which allows static priority (1-99).
3580 // However, for large multi-threaded applications, SCHED_RR is not only slower
3581 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3582 // of 5 runs - Sep 2005).
3583 //
3584 // The following code actually changes the niceness of kernel-thread/LWP. It
3585 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3586 // not the entire user process, and user level threads are 1:1 mapped to kernel
3587 // threads. It has always been the case, but could change in the future. For
3588 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3589 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3590 
3591 #if defined(_ALLBSD_SOURCE) && !defined(__APPLE__)
3592 int os::java_to_os_priority[CriticalPriority + 1] = {
3593   19,              // 0 Entry should never be used
3594 
3595    0,              // 1 MinPriority
3596    3,              // 2
3597    6,              // 3
3598 
3599   10,              // 4
3600   15,              // 5 NormPriority
3601   18,              // 6
3602 
3603   21,              // 7
3604   25,              // 8
3605   28,              // 9 NearMaxPriority
3606 
3607   31,              // 10 MaxPriority
3608 
3609   31               // 11 CriticalPriority
3610 };
3611 #elif defined(__APPLE__)
3612 /* Using Mach high-level priority assignments */
3613 int os::java_to_os_priority[CriticalPriority + 1] = {
3614    0,              // 0 Entry should never be used (MINPRI_USER)
3615 
3616   27,              // 1 MinPriority
3617   28,              // 2
3618   29,              // 3
3619 
3620   30,              // 4
3621   31,              // 5 NormPriority (BASEPRI_DEFAULT)
3622   32,              // 6
3623 
3624   33,              // 7
3625   34,              // 8
3626   35,              // 9 NearMaxPriority
3627 
3628   36,              // 10 MaxPriority
3629 
3630   36               // 11 CriticalPriority
3631 };
3632 #else
3633 int os::java_to_os_priority[CriticalPriority + 1] = {
3634   19,              // 0 Entry should never be used
3635 
3636    4,              // 1 MinPriority
3637    3,              // 2
3638    2,              // 3
3639 
3640    1,              // 4
3641    0,              // 5 NormPriority
3642   -1,              // 6
3643 
3644   -2,              // 7
3645   -3,              // 8
3646   -4,              // 9 NearMaxPriority
3647 
3648   -5,              // 10 MaxPriority
3649 
3650   -5               // 11 CriticalPriority
3651 };
3652 #endif
3653 
3654 static int prio_init() {
3655   if (ThreadPriorityPolicy == 1) {
3656     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3657     // if effective uid is not root. Perhaps, a more elegant way of doing
3658     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3659     if (geteuid() != 0) {
3660       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3661         warning("-XX:ThreadPriorityPolicy requires root privilege on Bsd");
3662       }
3663       ThreadPriorityPolicy = 0;
3664     }
3665   }
3666   if (UseCriticalJavaThreadPriority) {
3667     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3668   }
3669   return 0;
3670 }
3671 
3672 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3673   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3674 
3675 #ifdef __OpenBSD__
3676   // OpenBSD pthread_setprio starves low priority threads
3677   return OS_OK;
3678 #elif defined(__FreeBSD__)
3679   int ret = pthread_setprio(thread->osthread()->pthread_id(), newpri);
3680 #elif defined(__APPLE__) || defined(__NetBSD__)
3681   struct sched_param sp;
3682   int policy;
3683   pthread_t self = pthread_self();
3684 
3685   if (pthread_getschedparam(self, &policy, &sp) != 0)
3686     return OS_ERR;
3687 
3688   sp.sched_priority = newpri;
3689   if (pthread_setschedparam(self, policy, &sp) != 0)
3690     return OS_ERR;
3691 
3692   return OS_OK;
3693 #else
3694   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3695   return (ret == 0) ? OS_OK : OS_ERR;
3696 #endif
3697 }
3698 
3699 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3700   if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3701     *priority_ptr = java_to_os_priority[NormPriority];
3702     return OS_OK;
3703   }
3704 
3705   errno = 0;
3706 #if defined(__OpenBSD__) || defined(__FreeBSD__)
3707   *priority_ptr = pthread_getprio(thread->osthread()->pthread_id());
3708 #elif defined(__APPLE__) || defined(__NetBSD__)
3709   int policy;
3710   struct sched_param sp;
3711 
3712   pthread_getschedparam(pthread_self(), &policy, &sp);
3713   *priority_ptr = sp.sched_priority;
3714 #else
3715   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3716 #endif
3717   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3718 }
3719 
3720 // Hint to the underlying OS that a task switch would not be good.
3721 // Void return because it's a hint and can fail.
3722 void os::hint_no_preempt() {}
3723 
3724 ////////////////////////////////////////////////////////////////////////////////
3725 // suspend/resume support
3726 
3727 //  the low-level signal-based suspend/resume support is a remnant from the
3728 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3729 //  within hotspot. Now there is a single use-case for this:
3730 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3731 //      that runs in the watcher thread.
3732 //  The remaining code is greatly simplified from the more general suspension
3733 //  code that used to be used.
3734 //
3735 //  The protocol is quite simple:
3736 //  - suspend:
3737 //      - sends a signal to the target thread
3738 //      - polls the suspend state of the osthread using a yield loop
3739 //      - target thread signal handler (SR_handler) sets suspend state
3740 //        and blocks in sigsuspend until continued
3741 //  - resume:
3742 //      - sets target osthread state to continue
3743 //      - sends signal to end the sigsuspend loop in the SR_handler
3744 //
3745 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3746 //
3747 
3748 static void resume_clear_context(OSThread *osthread) {
3749   osthread->set_ucontext(NULL);
3750   osthread->set_siginfo(NULL);
3751 
3752   // notify the suspend action is completed, we have now resumed
3753   osthread->sr.clear_suspended();
3754 }
3755 
3756 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3757   osthread->set_ucontext(context);
3758   osthread->set_siginfo(siginfo);
3759 }
3760 
3761 //
3762 // Handler function invoked when a thread's execution is suspended or
3763 // resumed. We have to be careful that only async-safe functions are
3764 // called here (Note: most pthread functions are not async safe and
3765 // should be avoided.)
3766 //
3767 // Note: sigwait() is a more natural fit than sigsuspend() from an
3768 // interface point of view, but sigwait() prevents the signal hander
3769 // from being run. libpthread would get very confused by not having
3770 // its signal handlers run and prevents sigwait()'s use with the
3771 // mutex granting granting signal.
3772 //
3773 // Currently only ever called on the VMThread
3774 //
3775 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3776   // Save and restore errno to avoid confusing native code with EINTR
3777   // after sigsuspend.
3778   int old_errno = errno;
3779 
3780   Thread* thread = Thread::current();
3781   OSThread* osthread = thread->osthread();
3782   assert(thread->is_VM_thread(), "Must be VMThread");
3783   // read current suspend action
3784   int action = osthread->sr.suspend_action();
3785   if (action == SR_SUSPEND) {
3786     suspend_save_context(osthread, siginfo, context);
3787 
3788     // Notify the suspend action is about to be completed. do_suspend()
3789     // waits until SR_SUSPENDED is set and then returns. We will wait
3790     // here for a resume signal and that completes the suspend-other
3791     // action. do_suspend/do_resume is always called as a pair from
3792     // the same thread - so there are no races
3793 
3794     // notify the caller
3795     osthread->sr.set_suspended();
3796 
3797     sigset_t suspend_set;  // signals for sigsuspend()
3798 
3799     // get current set of blocked signals and unblock resume signal
3800     pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3801     sigdelset(&suspend_set, SR_signum);
3802 
3803     // wait here until we are resumed
3804     do {
3805       sigsuspend(&suspend_set);
3806       // ignore all returns until we get a resume signal
3807     } while (osthread->sr.suspend_action() != SR_CONTINUE);
3808 
3809     resume_clear_context(osthread);
3810 
3811   } else {
3812     assert(action == SR_CONTINUE, "unexpected sr action");
3813     // nothing special to do - just leave the handler
3814   }
3815 
3816   errno = old_errno;
3817 }
3818 
3819 
3820 static int SR_initialize() {
3821   struct sigaction act;
3822   char *s;
3823   /* Get signal number to use for suspend/resume */
3824   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3825     int sig = ::strtol(s, 0, 10);
3826     if (sig > 0 || sig < NSIG) {
3827         SR_signum = sig;
3828     }
3829   }
3830 
3831   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3832         "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3833 
3834   sigemptyset(&SR_sigset);
3835   sigaddset(&SR_sigset, SR_signum);
3836 
3837   /* Set up signal handler for suspend/resume */
3838   act.sa_flags = SA_RESTART|SA_SIGINFO;
3839   act.sa_handler = (void (*)(int)) SR_handler;
3840 
3841   // SR_signum is blocked by default.
3842   // 4528190 - We also need to block pthread restart signal (32 on all
3843   // supported Bsd platforms). Note that BsdThreads need to block
3844   // this signal for all threads to work properly. So we don't have
3845   // to use hard-coded signal number when setting up the mask.
3846   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3847 
3848   if (sigaction(SR_signum, &act, 0) == -1) {
3849     return -1;
3850   }
3851 
3852   // Save signal flag
3853   os::Bsd::set_our_sigflags(SR_signum, act.sa_flags);
3854   return 0;
3855 }
3856 
3857 static int SR_finalize() {
3858   return 0;
3859 }
3860 
3861 
3862 // returns true on success and false on error - really an error is fatal
3863 // but this seems the normal response to library errors
3864 static bool do_suspend(OSThread* osthread) {
3865   // mark as suspended and send signal
3866   osthread->sr.set_suspend_action(SR_SUSPEND);
3867   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3868   assert_status(status == 0, status, "pthread_kill");
3869 
3870   // check status and wait until notified of suspension
3871   if (status == 0) {
3872     for (int i = 0; !osthread->sr.is_suspended(); i++) {
3873       os::yield_all(i);
3874     }
3875     osthread->sr.set_suspend_action(SR_NONE);
3876     return true;
3877   }
3878   else {
3879     osthread->sr.set_suspend_action(SR_NONE);
3880     return false;
3881   }
3882 }
3883 
3884 static void do_resume(OSThread* osthread) {
3885   assert(osthread->sr.is_suspended(), "thread should be suspended");
3886   osthread->sr.set_suspend_action(SR_CONTINUE);
3887 
3888   int status = pthread_kill(osthread->pthread_id(), SR_signum);
3889   assert_status(status == 0, status, "pthread_kill");
3890   // check status and wait unit notified of resumption
3891   if (status == 0) {
3892     for (int i = 0; osthread->sr.is_suspended(); i++) {
3893       os::yield_all(i);
3894     }
3895   }
3896   osthread->sr.set_suspend_action(SR_NONE);
3897 }
3898 
3899 ////////////////////////////////////////////////////////////////////////////////
3900 // interrupt support
3901 
3902 void os::interrupt(Thread* thread) {
3903   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3904     "possibility of dangling Thread pointer");
3905 
3906   OSThread* osthread = thread->osthread();
3907 
3908   if (!osthread->interrupted()) {
3909     osthread->set_interrupted(true);
3910     // More than one thread can get here with the same value of osthread,
3911     // resulting in multiple notifications.  We do, however, want the store
3912     // to interrupted() to be visible to other threads before we execute unpark().
3913     OrderAccess::fence();
3914     ParkEvent * const slp = thread->_SleepEvent ;
3915     if (slp != NULL) slp->unpark() ;
3916   }
3917 
3918   // For JSR166. Unpark even if interrupt status already was set
3919   if (thread->is_Java_thread())
3920     ((JavaThread*)thread)->parker()->unpark();
3921 
3922   ParkEvent * ev = thread->_ParkEvent ;
3923   if (ev != NULL) ev->unpark() ;
3924 
3925 }
3926 
3927 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3928   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3929     "possibility of dangling Thread pointer");
3930 
3931   OSThread* osthread = thread->osthread();
3932 
3933   bool interrupted = osthread->interrupted();
3934 
3935   if (interrupted && clear_interrupted) {
3936     osthread->set_interrupted(false);
3937     // consider thread->_SleepEvent->reset() ... optional optimization
3938   }
3939 
3940   return interrupted;
3941 }
3942 
3943 ///////////////////////////////////////////////////////////////////////////////////
3944 // signal handling (except suspend/resume)
3945 
3946 // This routine may be used by user applications as a "hook" to catch signals.
3947 // The user-defined signal handler must pass unrecognized signals to this
3948 // routine, and if it returns true (non-zero), then the signal handler must
3949 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3950 // routine will never retun false (zero), but instead will execute a VM panic
3951 // routine kill the process.
3952 //
3953 // If this routine returns false, it is OK to call it again.  This allows
3954 // the user-defined signal handler to perform checks either before or after
3955 // the VM performs its own checks.  Naturally, the user code would be making
3956 // a serious error if it tried to handle an exception (such as a null check
3957 // or breakpoint) that the VM was generating for its own correct operation.
3958 //
3959 // This routine may recognize any of the following kinds of signals:
3960 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3961 // It should be consulted by handlers for any of those signals.
3962 //
3963 // The caller of this routine must pass in the three arguments supplied
3964 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3965 // field of the structure passed to sigaction().  This routine assumes that
3966 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3967 //
3968 // Note that the VM will print warnings if it detects conflicting signal
3969 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3970 //
3971 extern "C" JNIEXPORT int
3972 JVM_handle_bsd_signal(int signo, siginfo_t* siginfo,
3973                         void* ucontext, int abort_if_unrecognized);
3974 
3975 void signalHandler(int sig, siginfo_t* info, void* uc) {
3976   assert(info != NULL && uc != NULL, "it must be old kernel");
3977   JVM_handle_bsd_signal(sig, info, uc, true);
3978 }
3979 
3980 
3981 // This boolean allows users to forward their own non-matching signals
3982 // to JVM_handle_bsd_signal, harmlessly.
3983 bool os::Bsd::signal_handlers_are_installed = false;
3984 
3985 // For signal-chaining
3986 struct sigaction os::Bsd::sigact[MAXSIGNUM];
3987 unsigned int os::Bsd::sigs = 0;
3988 bool os::Bsd::libjsig_is_loaded = false;
3989 typedef struct sigaction *(*get_signal_t)(int);
3990 get_signal_t os::Bsd::get_signal_action = NULL;
3991 
3992 struct sigaction* os::Bsd::get_chained_signal_action(int sig) {
3993   struct sigaction *actp = NULL;
3994 
3995   if (libjsig_is_loaded) {
3996     // Retrieve the old signal handler from libjsig
3997     actp = (*get_signal_action)(sig);
3998   }
3999   if (actp == NULL) {
4000     // Retrieve the preinstalled signal handler from jvm
4001     actp = get_preinstalled_handler(sig);
4002   }
4003 
4004   return actp;
4005 }
4006 
4007 static bool call_chained_handler(struct sigaction *actp, int sig,
4008                                  siginfo_t *siginfo, void *context) {
4009   // Call the old signal handler
4010   if (actp->sa_handler == SIG_DFL) {
4011     // It's more reasonable to let jvm treat it as an unexpected exception
4012     // instead of taking the default action.
4013     return false;
4014   } else if (actp->sa_handler != SIG_IGN) {
4015     if ((actp->sa_flags & SA_NODEFER) == 0) {
4016       // automaticlly block the signal
4017       sigaddset(&(actp->sa_mask), sig);
4018     }
4019 
4020     sa_handler_t hand;
4021     sa_sigaction_t sa;
4022     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4023     // retrieve the chained handler
4024     if (siginfo_flag_set) {
4025       sa = actp->sa_sigaction;
4026     } else {
4027       hand = actp->sa_handler;
4028     }
4029 
4030     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4031       actp->sa_handler = SIG_DFL;
4032     }
4033 
4034     // try to honor the signal mask
4035     sigset_t oset;
4036     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4037 
4038     // call into the chained handler
4039     if (siginfo_flag_set) {
4040       (*sa)(sig, siginfo, context);
4041     } else {
4042       (*hand)(sig);
4043     }
4044 
4045     // restore the signal mask
4046     pthread_sigmask(SIG_SETMASK, &oset, 0);
4047   }
4048   // Tell jvm's signal handler the signal is taken care of.
4049   return true;
4050 }
4051 
4052 bool os::Bsd::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4053   bool chained = false;
4054   // signal-chaining
4055   if (UseSignalChaining) {
4056     struct sigaction *actp = get_chained_signal_action(sig);
4057     if (actp != NULL) {
4058       chained = call_chained_handler(actp, sig, siginfo, context);
4059     }
4060   }
4061   return chained;
4062 }
4063 
4064 struct sigaction* os::Bsd::get_preinstalled_handler(int sig) {
4065   if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4066     return &sigact[sig];
4067   }
4068   return NULL;
4069 }
4070 
4071 void os::Bsd::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4072   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4073   sigact[sig] = oldAct;
4074   sigs |= (unsigned int)1 << sig;
4075 }
4076 
4077 // for diagnostic
4078 int os::Bsd::sigflags[MAXSIGNUM];
4079 
4080 int os::Bsd::get_our_sigflags(int sig) {
4081   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4082   return sigflags[sig];
4083 }
4084 
4085 void os::Bsd::set_our_sigflags(int sig, int flags) {
4086   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4087   sigflags[sig] = flags;
4088 }
4089 
4090 void os::Bsd::set_signal_handler(int sig, bool set_installed) {
4091   // Check for overwrite.
4092   struct sigaction oldAct;
4093   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4094 
4095   void* oldhand = oldAct.sa_sigaction
4096                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4097                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4098   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4099       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4100       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4101     if (AllowUserSignalHandlers || !set_installed) {
4102       // Do not overwrite; user takes responsibility to forward to us.
4103       return;
4104     } else if (UseSignalChaining) {
4105       // save the old handler in jvm
4106       save_preinstalled_handler(sig, oldAct);
4107       // libjsig also interposes the sigaction() call below and saves the
4108       // old sigaction on it own.
4109     } else {
4110       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4111                     "%#lx for signal %d.", (long)oldhand, sig));
4112     }
4113   }
4114 
4115   struct sigaction sigAct;
4116   sigfillset(&(sigAct.sa_mask));
4117   sigAct.sa_handler = SIG_DFL;
4118   if (!set_installed) {
4119     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4120   } else {
4121     sigAct.sa_sigaction = signalHandler;
4122     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4123   }
4124   // Save flags, which are set by ours
4125   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4126   sigflags[sig] = sigAct.sa_flags;
4127 
4128   int ret = sigaction(sig, &sigAct, &oldAct);
4129   assert(ret == 0, "check");
4130 
4131   void* oldhand2  = oldAct.sa_sigaction
4132                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4133                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4134   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4135 }
4136 
4137 // install signal handlers for signals that HotSpot needs to
4138 // handle in order to support Java-level exception handling.
4139 
4140 void os::Bsd::install_signal_handlers() {
4141   if (!signal_handlers_are_installed) {
4142     signal_handlers_are_installed = true;
4143 
4144     // signal-chaining
4145     typedef void (*signal_setting_t)();
4146     signal_setting_t begin_signal_setting = NULL;
4147     signal_setting_t end_signal_setting = NULL;
4148     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4149                              dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4150     if (begin_signal_setting != NULL) {
4151       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4152                              dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4153       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4154                             dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4155       libjsig_is_loaded = true;
4156       assert(UseSignalChaining, "should enable signal-chaining");
4157     }
4158     if (libjsig_is_loaded) {
4159       // Tell libjsig jvm is setting signal handlers
4160       (*begin_signal_setting)();
4161     }
4162 
4163     set_signal_handler(SIGSEGV, true);
4164     set_signal_handler(SIGPIPE, true);
4165     set_signal_handler(SIGBUS, true);
4166     set_signal_handler(SIGILL, true);
4167     set_signal_handler(SIGFPE, true);
4168     set_signal_handler(SIGXFSZ, true);
4169 
4170 #if defined(__APPLE__)
4171     // In Mac OS X 10.4, CrashReporter will write a crash log for all 'fatal' signals, including
4172     // signals caught and handled by the JVM. To work around this, we reset the mach task
4173     // signal handler that's placed on our process by CrashReporter. This disables
4174     // CrashReporter-based reporting.
4175     //
4176     // This work-around is not necessary for 10.5+, as CrashReporter no longer intercedes
4177     // on caught fatal signals.
4178     //
4179     // Additionally, gdb installs both standard BSD signal handlers, and mach exception
4180     // handlers. By replacing the existing task exception handler, we disable gdb's mach
4181     // exception handling, while leaving the standard BSD signal handlers functional.
4182     kern_return_t kr;
4183     kr = task_set_exception_ports(mach_task_self(),
4184         EXC_MASK_BAD_ACCESS | EXC_MASK_ARITHMETIC,
4185         MACH_PORT_NULL,
4186         EXCEPTION_STATE_IDENTITY,
4187         MACHINE_THREAD_STATE);
4188 
4189     assert(kr == KERN_SUCCESS, "could not set mach task signal handler");
4190 #endif
4191 
4192     if (libjsig_is_loaded) {
4193       // Tell libjsig jvm finishes setting signal handlers
4194       (*end_signal_setting)();
4195     }
4196 
4197     // We don't activate signal checker if libjsig is in place, we trust ourselves
4198     // and if UserSignalHandler is installed all bets are off
4199     if (CheckJNICalls) {
4200       if (libjsig_is_loaded) {
4201         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4202         check_signals = false;
4203       }
4204       if (AllowUserSignalHandlers) {
4205         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4206         check_signals = false;
4207       }
4208     }
4209   }
4210 }
4211 
4212 #ifndef _ALLBSD_SOURCE
4213 // This is the fastest way to get thread cpu time on Bsd.
4214 // Returns cpu time (user+sys) for any thread, not only for current.
4215 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4216 // It might work on 2.6.10+ with a special kernel/glibc patch.
4217 // For reference, please, see IEEE Std 1003.1-2004:
4218 //   http://www.unix.org/single_unix_specification
4219 
4220 jlong os::Bsd::fast_thread_cpu_time(clockid_t clockid) {
4221   struct timespec tp;
4222   int rc = os::Bsd::clock_gettime(clockid, &tp);
4223   assert(rc == 0, "clock_gettime is expected to return 0 code");
4224 
4225   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4226 }
4227 #endif
4228 
4229 /////
4230 // glibc on Bsd platform uses non-documented flag
4231 // to indicate, that some special sort of signal
4232 // trampoline is used.
4233 // We will never set this flag, and we should
4234 // ignore this flag in our diagnostic
4235 #ifdef SIGNIFICANT_SIGNAL_MASK
4236 #undef SIGNIFICANT_SIGNAL_MASK
4237 #endif
4238 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4239 
4240 static const char* get_signal_handler_name(address handler,
4241                                            char* buf, int buflen) {
4242   int offset;
4243   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4244   if (found) {
4245     // skip directory names
4246     const char *p1, *p2;
4247     p1 = buf;
4248     size_t len = strlen(os::file_separator());
4249     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4250     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4251   } else {
4252     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4253   }
4254   return buf;
4255 }
4256 
4257 static void print_signal_handler(outputStream* st, int sig,
4258                                  char* buf, size_t buflen) {
4259   struct sigaction sa;
4260 
4261   sigaction(sig, NULL, &sa);
4262 
4263   // See comment for SIGNIFICANT_SIGNAL_MASK define
4264   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4265 
4266   st->print("%s: ", os::exception_name(sig, buf, buflen));
4267 
4268   address handler = (sa.sa_flags & SA_SIGINFO)
4269     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4270     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4271 
4272   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4273     st->print("SIG_DFL");
4274   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4275     st->print("SIG_IGN");
4276   } else {
4277     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4278   }
4279 
4280   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4281 
4282   address rh = VMError::get_resetted_sighandler(sig);
4283   // May be, handler was resetted by VMError?
4284   if(rh != NULL) {
4285     handler = rh;
4286     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4287   }
4288 
4289   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
4290 
4291   // Check: is it our handler?
4292   if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4293      handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4294     // It is our signal handler
4295     // check for flags, reset system-used one!
4296     if((int)sa.sa_flags != os::Bsd::get_our_sigflags(sig)) {
4297       st->print(
4298                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4299                 os::Bsd::get_our_sigflags(sig));
4300     }
4301   }
4302   st->cr();
4303 }
4304 
4305 
4306 #define DO_SIGNAL_CHECK(sig) \
4307   if (!sigismember(&check_signal_done, sig)) \
4308     os::Bsd::check_signal_handler(sig)
4309 
4310 // This method is a periodic task to check for misbehaving JNI applications
4311 // under CheckJNI, we can add any periodic checks here
4312 
4313 void os::run_periodic_checks() {
4314 
4315   if (check_signals == false) return;
4316 
4317   // SEGV and BUS if overridden could potentially prevent
4318   // generation of hs*.log in the event of a crash, debugging
4319   // such a case can be very challenging, so we absolutely
4320   // check the following for a good measure:
4321   DO_SIGNAL_CHECK(SIGSEGV);
4322   DO_SIGNAL_CHECK(SIGILL);
4323   DO_SIGNAL_CHECK(SIGFPE);
4324   DO_SIGNAL_CHECK(SIGBUS);
4325   DO_SIGNAL_CHECK(SIGPIPE);
4326   DO_SIGNAL_CHECK(SIGXFSZ);
4327 
4328 
4329   // ReduceSignalUsage allows the user to override these handlers
4330   // see comments at the very top and jvm_solaris.h
4331   if (!ReduceSignalUsage) {
4332     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4333     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4334     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4335     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4336   }
4337 
4338   DO_SIGNAL_CHECK(SR_signum);
4339   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4340 }
4341 
4342 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4343 
4344 static os_sigaction_t os_sigaction = NULL;
4345 
4346 void os::Bsd::check_signal_handler(int sig) {
4347   char buf[O_BUFLEN];
4348   address jvmHandler = NULL;
4349 
4350 
4351   struct sigaction act;
4352   if (os_sigaction == NULL) {
4353     // only trust the default sigaction, in case it has been interposed
4354     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4355     if (os_sigaction == NULL) return;
4356   }
4357 
4358   os_sigaction(sig, (struct sigaction*)NULL, &act);
4359 
4360 
4361   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4362 
4363   address thisHandler = (act.sa_flags & SA_SIGINFO)
4364     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4365     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4366 
4367 
4368   switch(sig) {
4369   case SIGSEGV:
4370   case SIGBUS:
4371   case SIGFPE:
4372   case SIGPIPE:
4373   case SIGILL:
4374   case SIGXFSZ:
4375     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4376     break;
4377 
4378   case SHUTDOWN1_SIGNAL:
4379   case SHUTDOWN2_SIGNAL:
4380   case SHUTDOWN3_SIGNAL:
4381   case BREAK_SIGNAL:
4382     jvmHandler = (address)user_handler();
4383     break;
4384 
4385   case INTERRUPT_SIGNAL:
4386     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4387     break;
4388 
4389   default:
4390     if (sig == SR_signum) {
4391       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4392     } else {
4393       return;
4394     }
4395     break;
4396   }
4397 
4398   if (thisHandler != jvmHandler) {
4399     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4400     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4401     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4402     // No need to check this sig any longer
4403     sigaddset(&check_signal_done, sig);
4404   } else if(os::Bsd::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Bsd::get_our_sigflags(sig)) {
4405     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4406     tty->print("expected:" PTR32_FORMAT, os::Bsd::get_our_sigflags(sig));
4407     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4408     // No need to check this sig any longer
4409     sigaddset(&check_signal_done, sig);
4410   }
4411 
4412   // Dump all the signal
4413   if (sigismember(&check_signal_done, sig)) {
4414     print_signal_handlers(tty, buf, O_BUFLEN);
4415   }
4416 }
4417 
4418 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4419 
4420 extern bool signal_name(int signo, char* buf, size_t len);
4421 
4422 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4423   if (0 < exception_code && exception_code <= SIGRTMAX) {
4424     // signal
4425     if (!signal_name(exception_code, buf, size)) {
4426       jio_snprintf(buf, size, "SIG%d", exception_code);
4427     }
4428     return buf;
4429   } else {
4430     return NULL;
4431   }
4432 }
4433 
4434 // this is called _before_ the most of global arguments have been parsed
4435 void os::init(void) {
4436   char dummy;   /* used to get a guess on initial stack address */
4437 //  first_hrtime = gethrtime();
4438 
4439   // With BsdThreads the JavaMain thread pid (primordial thread)
4440   // is different than the pid of the java launcher thread.
4441   // So, on Bsd, the launcher thread pid is passed to the VM
4442   // via the sun.java.launcher.pid property.
4443   // Use this property instead of getpid() if it was correctly passed.
4444   // See bug 6351349.
4445   pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4446 
4447   _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4448 
4449   clock_tics_per_sec = CLK_TCK;
4450 
4451   init_random(1234567);
4452 
4453   ThreadCritical::initialize();
4454 
4455   Bsd::set_page_size(getpagesize());
4456   if (Bsd::page_size() == -1) {
4457     fatal(err_msg("os_bsd.cpp: os::init: sysconf failed (%s)",
4458                   strerror(errno)));
4459   }
4460   init_page_sizes((size_t) Bsd::page_size());
4461 
4462   Bsd::initialize_system_info();
4463 
4464   // main_thread points to the aboriginal thread
4465   Bsd::_main_thread = pthread_self();
4466 
4467   Bsd::clock_init();
4468   initial_time_count = os::elapsed_counter();
4469 
4470 #ifdef __APPLE__
4471   // XXXDARWIN
4472   // Work around the unaligned VM callbacks in hotspot's
4473   // sharedRuntime. The callbacks don't use SSE2 instructions, and work on
4474   // Linux, Solaris, and FreeBSD. On Mac OS X, dyld (rightly so) enforces
4475   // alignment when doing symbol lookup. To work around this, we force early
4476   // binding of all symbols now, thus binding when alignment is known-good.
4477   _dyld_bind_fully_image_containing_address((const void *) &os::init);
4478 #endif
4479 }
4480 
4481 // To install functions for atexit system call
4482 extern "C" {
4483   static void perfMemory_exit_helper() {
4484     perfMemory_exit();
4485   }
4486 }
4487 
4488 // this is called _after_ the global arguments have been parsed
4489 jint os::init_2(void)
4490 {
4491 #ifndef _ALLBSD_SOURCE
4492   Bsd::fast_thread_clock_init();
4493 #endif
4494 
4495   // Allocate a single page and mark it as readable for safepoint polling
4496   address polling_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4497   guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4498 
4499   os::set_polling_page( polling_page );
4500 
4501 #ifndef PRODUCT
4502   if(Verbose && PrintMiscellaneous)
4503     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4504 #endif
4505 
4506   if (!UseMembar) {
4507     address mem_serialize_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4508     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4509     os::set_memory_serialize_page( mem_serialize_page );
4510 
4511 #ifndef PRODUCT
4512     if(Verbose && PrintMiscellaneous)
4513       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4514 #endif
4515   }
4516 
4517   os::large_page_init();
4518 
4519   // initialize suspend/resume support - must do this before signal_sets_init()
4520   if (SR_initialize() != 0) {
4521     perror("SR_initialize failed");
4522     return JNI_ERR;
4523   }
4524 
4525   Bsd::signal_sets_init();
4526   Bsd::install_signal_handlers();
4527 
4528   // Check minimum allowable stack size for thread creation and to initialize
4529   // the java system classes, including StackOverflowError - depends on page
4530   // size.  Add a page for compiler2 recursion in main thread.
4531   // Add in 2*BytesPerWord times page size to account for VM stack during
4532   // class initialization depending on 32 or 64 bit VM.
4533   os::Bsd::min_stack_allowed = MAX2(os::Bsd::min_stack_allowed,
4534             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4535                     2*BytesPerWord COMPILER2_PRESENT(+1)) * Bsd::page_size());
4536 
4537   size_t threadStackSizeInBytes = ThreadStackSize * K;
4538   if (threadStackSizeInBytes != 0 &&
4539       threadStackSizeInBytes < os::Bsd::min_stack_allowed) {
4540         tty->print_cr("\nThe stack size specified is too small, "
4541                       "Specify at least %dk",
4542                       os::Bsd::min_stack_allowed/ K);
4543         return JNI_ERR;
4544   }
4545 
4546   // Make the stack size a multiple of the page size so that
4547   // the yellow/red zones can be guarded.
4548   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4549         vm_page_size()));
4550 
4551 #ifndef _ALLBSD_SOURCE
4552   Bsd::capture_initial_stack(JavaThread::stack_size_at_create());
4553 
4554   Bsd::libpthread_init();
4555   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4556      tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4557           Bsd::glibc_version(), Bsd::libpthread_version(),
4558           Bsd::is_floating_stack() ? "floating stack" : "fixed stack");
4559   }
4560 
4561   if (UseNUMA) {
4562     if (!Bsd::libnuma_init()) {
4563       UseNUMA = false;
4564     } else {
4565       if ((Bsd::numa_max_node() < 1)) {
4566         // There's only one node(they start from 0), disable NUMA.
4567         UseNUMA = false;
4568       }
4569     }
4570     // With SHM large pages we cannot uncommit a page, so there's not way
4571     // we can make the adaptive lgrp chunk resizing work. If the user specified
4572     // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4573     // disable adaptive resizing.
4574     if (UseNUMA && UseLargePages && UseSHM) {
4575       if (!FLAG_IS_DEFAULT(UseNUMA)) {
4576         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4577           UseLargePages = false;
4578         } else {
4579           warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4580           UseAdaptiveSizePolicy = false;
4581           UseAdaptiveNUMAChunkSizing = false;
4582         }
4583       } else {
4584         UseNUMA = false;
4585       }
4586     }
4587     if (!UseNUMA && ForceNUMA) {
4588       UseNUMA = true;
4589     }
4590   }
4591 #endif
4592 
4593   if (MaxFDLimit) {
4594     // set the number of file descriptors to max. print out error
4595     // if getrlimit/setrlimit fails but continue regardless.
4596     struct rlimit nbr_files;
4597     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4598     if (status != 0) {
4599       if (PrintMiscellaneous && (Verbose || WizardMode))
4600         perror("os::init_2 getrlimit failed");
4601     } else {
4602       nbr_files.rlim_cur = nbr_files.rlim_max;
4603 
4604 #ifdef __APPLE__
4605       // Darwin returns RLIM_INFINITY for rlim_max, but fails with EINVAL if
4606       // you attempt to use RLIM_INFINITY. As per setrlimit(2), OPEN_MAX must
4607       // be used instead
4608       nbr_files.rlim_cur = MIN(OPEN_MAX, nbr_files.rlim_cur);
4609 #endif
4610 
4611       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4612       if (status != 0) {
4613         if (PrintMiscellaneous && (Verbose || WizardMode))
4614           perror("os::init_2 setrlimit failed");
4615       }
4616     }
4617   }
4618 
4619 #ifndef _ALLBSD_SOURCE
4620   // Initialize lock used to serialize thread creation (see os::create_thread)
4621   Bsd::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4622 #endif
4623 
4624   // at-exit methods are called in the reverse order of their registration.
4625   // atexit functions are called on return from main or as a result of a
4626   // call to exit(3C). There can be only 32 of these functions registered
4627   // and atexit() does not set errno.
4628 
4629   if (PerfAllowAtExitRegistration) {
4630     // only register atexit functions if PerfAllowAtExitRegistration is set.
4631     // atexit functions can be delayed until process exit time, which
4632     // can be problematic for embedded VM situations. Embedded VMs should
4633     // call DestroyJavaVM() to assure that VM resources are released.
4634 
4635     // note: perfMemory_exit_helper atexit function may be removed in
4636     // the future if the appropriate cleanup code can be added to the
4637     // VM_Exit VMOperation's doit method.
4638     if (atexit(perfMemory_exit_helper) != 0) {
4639       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4640     }
4641   }
4642 
4643   // initialize thread priority policy
4644   prio_init();
4645 
4646 #ifdef __APPLE__
4647   // dynamically link to objective c gc registration
4648   void *handleLibObjc = dlopen(OBJC_LIB, RTLD_LAZY);
4649   if (handleLibObjc != NULL) {
4650     objc_registerThreadWithCollectorFunction = (objc_registerThreadWithCollector_t) dlsym(handleLibObjc, OBJC_GCREGISTER);
4651   }
4652 #endif
4653 
4654   return JNI_OK;
4655 }
4656 
4657 // this is called at the end of vm_initialization
4658 void os::init_3(void) { }
4659 
4660 // Mark the polling page as unreadable
4661 void os::make_polling_page_unreadable(void) {
4662   if( !guard_memory((char*)_polling_page, Bsd::page_size()) )
4663     fatal("Could not disable polling page");
4664 };
4665 
4666 // Mark the polling page as readable
4667 void os::make_polling_page_readable(void) {
4668   if( !bsd_mprotect((char *)_polling_page, Bsd::page_size(), PROT_READ)) {
4669     fatal("Could not enable polling page");
4670   }
4671 };
4672 
4673 int os::active_processor_count() {
4674 #ifdef _ALLBSD_SOURCE
4675   return _processor_count;
4676 #else
4677   // Bsd doesn't yet have a (official) notion of processor sets,
4678   // so just return the number of online processors.
4679   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4680   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4681   return online_cpus;
4682 #endif
4683 }
4684 
4685 void os::set_native_thread_name(const char *name) {
4686 #if defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED > MAC_OS_X_VERSION_10_5
4687   // This is only supported in Snow Leopard and beyond
4688   if (name != NULL) {
4689     // Add a "Java: " prefix to the name
4690     char buf[MAXTHREADNAMESIZE];
4691     snprintf(buf, sizeof(buf), "Java: %s", name);
4692     pthread_setname_np(buf);
4693   }
4694 #endif
4695 }
4696 
4697 bool os::distribute_processes(uint length, uint* distribution) {
4698   // Not yet implemented.
4699   return false;
4700 }
4701 
4702 bool os::bind_to_processor(uint processor_id) {
4703   // Not yet implemented.
4704   return false;
4705 }
4706 
4707 ///
4708 
4709 // Suspends the target using the signal mechanism and then grabs the PC before
4710 // resuming the target. Used by the flat-profiler only
4711 ExtendedPC os::get_thread_pc(Thread* thread) {
4712   // Make sure that it is called by the watcher for the VMThread
4713   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4714   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4715 
4716   ExtendedPC epc;
4717 
4718   OSThread* osthread = thread->osthread();
4719   if (do_suspend(osthread)) {
4720     if (osthread->ucontext() != NULL) {
4721       epc = os::Bsd::ucontext_get_pc(osthread->ucontext());
4722     } else {
4723       // NULL context is unexpected, double-check this is the VMThread
4724       guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4725     }
4726     do_resume(osthread);
4727   }
4728   // failure means pthread_kill failed for some reason - arguably this is
4729   // a fatal problem, but such problems are ignored elsewhere
4730 
4731   return epc;
4732 }
4733 
4734 int os::Bsd::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4735 {
4736 #ifdef _ALLBSD_SOURCE
4737   return pthread_cond_timedwait(_cond, _mutex, _abstime);
4738 #else
4739    if (is_NPTL()) {
4740       return pthread_cond_timedwait(_cond, _mutex, _abstime);
4741    } else {
4742 #ifndef IA64
4743       // 6292965: BsdThreads pthread_cond_timedwait() resets FPU control
4744       // word back to default 64bit precision if condvar is signaled. Java
4745       // wants 53bit precision.  Save and restore current value.
4746       int fpu = get_fpu_control_word();
4747 #endif // IA64
4748       int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4749 #ifndef IA64
4750       set_fpu_control_word(fpu);
4751 #endif // IA64
4752       return status;
4753    }
4754 #endif
4755 }
4756 
4757 ////////////////////////////////////////////////////////////////////////////////
4758 // debug support
4759 
4760 static address same_page(address x, address y) {
4761   int page_bits = -os::vm_page_size();
4762   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4763     return x;
4764   else if (x > y)
4765     return (address)(intptr_t(y) | ~page_bits) + 1;
4766   else
4767     return (address)(intptr_t(y) & page_bits);
4768 }
4769 
4770 bool os::find(address addr, outputStream* st) {
4771   Dl_info dlinfo;
4772   memset(&dlinfo, 0, sizeof(dlinfo));
4773   if (dladdr(addr, &dlinfo)) {
4774     st->print(PTR_FORMAT ": ", addr);
4775     if (dlinfo.dli_sname != NULL) {
4776       st->print("%s+%#x", dlinfo.dli_sname,
4777                  addr - (intptr_t)dlinfo.dli_saddr);
4778     } else if (dlinfo.dli_fname) {
4779       st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4780     } else {
4781       st->print("<absolute address>");
4782     }
4783     if (dlinfo.dli_fname) {
4784       st->print(" in %s", dlinfo.dli_fname);
4785     }
4786     if (dlinfo.dli_fbase) {
4787       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4788     }
4789     st->cr();
4790 
4791     if (Verbose) {
4792       // decode some bytes around the PC
4793       address begin = same_page(addr-40, addr);
4794       address end   = same_page(addr+40, addr);
4795       address       lowest = (address) dlinfo.dli_sname;
4796       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4797       if (begin < lowest)  begin = lowest;
4798       Dl_info dlinfo2;
4799       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4800           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4801         end = (address) dlinfo2.dli_saddr;
4802       Disassembler::decode(begin, end, st);
4803     }
4804     return true;
4805   }
4806   return false;
4807 }
4808 
4809 ////////////////////////////////////////////////////////////////////////////////
4810 // misc
4811 
4812 // This does not do anything on Bsd. This is basically a hook for being
4813 // able to use structured exception handling (thread-local exception filters)
4814 // on, e.g., Win32.
4815 void
4816 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4817                          JavaCallArguments* args, Thread* thread) {
4818   f(value, method, args, thread);
4819 }
4820 
4821 void os::print_statistics() {
4822 }
4823 
4824 int os::message_box(const char* title, const char* message) {
4825   int i;
4826   fdStream err(defaultStream::error_fd());
4827   for (i = 0; i < 78; i++) err.print_raw("=");
4828   err.cr();
4829   err.print_raw_cr(title);
4830   for (i = 0; i < 78; i++) err.print_raw("-");
4831   err.cr();
4832   err.print_raw_cr(message);
4833   for (i = 0; i < 78; i++) err.print_raw("=");
4834   err.cr();
4835 
4836   char buf[16];
4837   // Prevent process from exiting upon "read error" without consuming all CPU
4838   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4839 
4840   return buf[0] == 'y' || buf[0] == 'Y';
4841 }
4842 
4843 int os::stat(const char *path, struct stat *sbuf) {
4844   char pathbuf[MAX_PATH];
4845   if (strlen(path) > MAX_PATH - 1) {
4846     errno = ENAMETOOLONG;
4847     return -1;
4848   }
4849   os::native_path(strcpy(pathbuf, path));
4850   return ::stat(pathbuf, sbuf);
4851 }
4852 
4853 bool os::check_heap(bool force) {
4854   return true;
4855 }
4856 
4857 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4858   return ::vsnprintf(buf, count, format, args);
4859 }
4860 
4861 // Is a (classpath) directory empty?
4862 bool os::dir_is_empty(const char* path) {
4863   DIR *dir = NULL;
4864   struct dirent *ptr;
4865 
4866   dir = opendir(path);
4867   if (dir == NULL) return true;
4868 
4869   /* Scan the directory */
4870   bool result = true;
4871   char buf[sizeof(struct dirent) + MAX_PATH];
4872   while (result && (ptr = ::readdir(dir)) != NULL) {
4873     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4874       result = false;
4875     }
4876   }
4877   closedir(dir);
4878   return result;
4879 }
4880 
4881 // This code originates from JDK's sysOpen and open64_w
4882 // from src/solaris/hpi/src/system_md.c
4883 
4884 #ifndef O_DELETE
4885 #define O_DELETE 0x10000
4886 #endif
4887 
4888 // Open a file. Unlink the file immediately after open returns
4889 // if the specified oflag has the O_DELETE flag set.
4890 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4891 
4892 int os::open(const char *path, int oflag, int mode) {
4893 
4894   if (strlen(path) > MAX_PATH - 1) {
4895     errno = ENAMETOOLONG;
4896     return -1;
4897   }
4898   int fd;
4899   int o_delete = (oflag & O_DELETE);
4900   oflag = oflag & ~O_DELETE;
4901 
4902   fd = ::open(path, oflag, mode);
4903   if (fd == -1) return -1;
4904 
4905   //If the open succeeded, the file might still be a directory
4906   {
4907     struct stat buf;
4908     int ret = ::fstat(fd, &buf);
4909     int st_mode = buf.st_mode;
4910 
4911     if (ret != -1) {
4912       if ((st_mode & S_IFMT) == S_IFDIR) {
4913         errno = EISDIR;
4914         ::close(fd);
4915         return -1;
4916       }
4917     } else {
4918       ::close(fd);
4919       return -1;
4920     }
4921   }
4922 
4923     /*
4924      * All file descriptors that are opened in the JVM and not
4925      * specifically destined for a subprocess should have the
4926      * close-on-exec flag set.  If we don't set it, then careless 3rd
4927      * party native code might fork and exec without closing all
4928      * appropriate file descriptors (e.g. as we do in closeDescriptors in
4929      * UNIXProcess.c), and this in turn might:
4930      *
4931      * - cause end-of-file to fail to be detected on some file
4932      *   descriptors, resulting in mysterious hangs, or
4933      *
4934      * - might cause an fopen in the subprocess to fail on a system
4935      *   suffering from bug 1085341.
4936      *
4937      * (Yes, the default setting of the close-on-exec flag is a Unix
4938      * design flaw)
4939      *
4940      * See:
4941      * 1085341: 32-bit stdio routines should support file descriptors >255
4942      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4943      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4944      */
4945 #ifdef FD_CLOEXEC
4946     {
4947         int flags = ::fcntl(fd, F_GETFD);
4948         if (flags != -1)
4949             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4950     }
4951 #endif
4952 
4953   if (o_delete != 0) {
4954     ::unlink(path);
4955   }
4956   return fd;
4957 }
4958 
4959 
4960 // create binary file, rewriting existing file if required
4961 int os::create_binary_file(const char* path, bool rewrite_existing) {
4962   int oflags = O_WRONLY | O_CREAT;
4963   if (!rewrite_existing) {
4964     oflags |= O_EXCL;
4965   }
4966   return ::open(path, oflags, S_IREAD | S_IWRITE);
4967 }
4968 
4969 // return current position of file pointer
4970 jlong os::current_file_offset(int fd) {
4971   return (jlong)::lseek(fd, (off_t)0, SEEK_CUR);
4972 }
4973 
4974 // move file pointer to the specified offset
4975 jlong os::seek_to_file_offset(int fd, jlong offset) {
4976   return (jlong)::lseek(fd, (off_t)offset, SEEK_SET);
4977 }
4978 
4979 // This code originates from JDK's sysAvailable
4980 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4981 
4982 int os::available(int fd, jlong *bytes) {
4983   jlong cur, end;
4984   int mode;
4985   struct stat buf;
4986 
4987   if (::fstat(fd, &buf) >= 0) {
4988     mode = buf.st_mode;
4989     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4990       /*
4991       * XXX: is the following call interruptible? If so, this might
4992       * need to go through the INTERRUPT_IO() wrapper as for other
4993       * blocking, interruptible calls in this file.
4994       */
4995       int n;
4996       if (::ioctl(fd, FIONREAD, &n) >= 0) {
4997         *bytes = n;
4998         return 1;
4999       }
5000     }
5001   }
5002   if ((cur = ::lseek(fd, 0L, SEEK_CUR)) == -1) {
5003     return 0;
5004   } else if ((end = ::lseek(fd, 0L, SEEK_END)) == -1) {
5005     return 0;
5006   } else if (::lseek(fd, cur, SEEK_SET) == -1) {
5007     return 0;
5008   }
5009   *bytes = end - cur;
5010   return 1;
5011 }
5012 
5013 int os::socket_available(int fd, jint *pbytes) {
5014    if (fd < 0)
5015      return OS_OK;
5016 
5017    int ret;
5018 
5019    RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
5020 
5021    //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5022    // is expected to return 0 on failure and 1 on success to the jdk.
5023 
5024    return (ret == OS_ERR) ? 0 : 1;
5025 }
5026 
5027 // Map a block of memory.
5028 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
5029                      char *addr, size_t bytes, bool read_only,
5030                      bool allow_exec) {
5031   int prot;
5032   int flags;
5033 
5034   if (read_only) {
5035     prot = PROT_READ;
5036     flags = MAP_SHARED;
5037   } else {
5038     prot = PROT_READ | PROT_WRITE;
5039     flags = MAP_PRIVATE;
5040   }
5041 
5042   if (allow_exec) {
5043     prot |= PROT_EXEC;
5044   }
5045 
5046   if (addr != NULL) {
5047     flags |= MAP_FIXED;
5048   }
5049 
5050   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5051                                      fd, file_offset);
5052   if (mapped_address == MAP_FAILED) {
5053     return NULL;
5054   }
5055   return mapped_address;
5056 }
5057 
5058 
5059 // Remap a block of memory.
5060 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
5061                        char *addr, size_t bytes, bool read_only,
5062                        bool allow_exec) {
5063   // same as map_memory() on this OS
5064   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5065                         allow_exec);
5066 }
5067 
5068 
5069 // Unmap a block of memory.
5070 bool os::unmap_memory(char* addr, size_t bytes) {
5071   return munmap(addr, bytes) == 0;
5072 }
5073 
5074 #ifndef _ALLBSD_SOURCE
5075 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5076 
5077 static clockid_t thread_cpu_clockid(Thread* thread) {
5078   pthread_t tid = thread->osthread()->pthread_id();
5079   clockid_t clockid;
5080 
5081   // Get thread clockid
5082   int rc = os::Bsd::pthread_getcpuclockid(tid, &clockid);
5083   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5084   return clockid;
5085 }
5086 #endif
5087 
5088 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5089 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5090 // of a thread.
5091 //
5092 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5093 // the fast estimate available on the platform.
5094 
5095 jlong os::current_thread_cpu_time() {
5096 #ifdef __APPLE__
5097   return os::thread_cpu_time(Thread::current(), true /* user + sys */);
5098 #elif !defined(_ALLBSD_SOURCE)
5099   if (os::Bsd::supports_fast_thread_cpu_time()) {
5100     return os::Bsd::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5101   } else {
5102     // return user + sys since the cost is the same
5103     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5104   }
5105 #endif
5106 }
5107 
5108 jlong os::thread_cpu_time(Thread* thread) {
5109 #ifndef _ALLBSD_SOURCE
5110   // consistent with what current_thread_cpu_time() returns
5111   if (os::Bsd::supports_fast_thread_cpu_time()) {
5112     return os::Bsd::fast_thread_cpu_time(thread_cpu_clockid(thread));
5113   } else {
5114     return slow_thread_cpu_time(thread, true /* user + sys */);
5115   }
5116 #endif
5117 }
5118 
5119 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5120 #ifdef __APPLE__
5121   return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5122 #elif !defined(_ALLBSD_SOURCE)
5123   if (user_sys_cpu_time && os::Bsd::supports_fast_thread_cpu_time()) {
5124     return os::Bsd::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5125   } else {
5126     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5127   }
5128 #endif
5129 }
5130 
5131 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5132 #ifdef __APPLE__
5133   struct thread_basic_info tinfo;
5134   mach_msg_type_number_t tcount = THREAD_INFO_MAX;
5135   kern_return_t kr;
5136   mach_port_t mach_thread;
5137 
5138   mach_thread = pthread_mach_thread_np(thread->osthread()->thread_id());
5139   kr = thread_info(mach_thread, THREAD_BASIC_INFO, (thread_info_t)&tinfo, &tcount);
5140   if (kr != KERN_SUCCESS)
5141     return -1;
5142 
5143   if (user_sys_cpu_time) {
5144     jlong nanos;
5145     nanos = ((jlong) tinfo.system_time.seconds + tinfo.user_time.seconds) * (jlong)1000000000;
5146     nanos += ((jlong) tinfo.system_time.microseconds + (jlong) tinfo.user_time.microseconds) * (jlong)1000;
5147     return nanos;
5148   } else {
5149     return ((jlong)tinfo.user_time.seconds * 1000000000) + ((jlong)tinfo.user_time.microseconds * (jlong)1000);
5150   }
5151 #elif !defined(_ALLBSD_SOURCE)
5152   if (user_sys_cpu_time && os::Bsd::supports_fast_thread_cpu_time()) {
5153     return os::Bsd::fast_thread_cpu_time(thread_cpu_clockid(thread));
5154   } else {
5155     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5156   }
5157 #endif
5158 }
5159 
5160 #ifndef _ALLBSD_SOURCE
5161 //
5162 //  -1 on error.
5163 //
5164 
5165 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5166   static bool proc_pid_cpu_avail = true;
5167   static bool proc_task_unchecked = true;
5168   static const char *proc_stat_path = "/proc/%d/stat";
5169   pid_t  tid = thread->osthread()->thread_id();
5170   int i;
5171   char *s;
5172   char stat[2048];
5173   int statlen;
5174   char proc_name[64];
5175   int count;
5176   long sys_time, user_time;
5177   char string[64];
5178   char cdummy;
5179   int idummy;
5180   long ldummy;
5181   FILE *fp;
5182 
5183   // We first try accessing /proc/<pid>/cpu since this is faster to
5184   // process.  If this file is not present (bsd kernels 2.5 and above)
5185   // then we open /proc/<pid>/stat.
5186   if ( proc_pid_cpu_avail ) {
5187     sprintf(proc_name, "/proc/%d/cpu", tid);
5188     fp =  fopen(proc_name, "r");
5189     if ( fp != NULL ) {
5190       count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
5191       fclose(fp);
5192       if ( count != 3 ) return -1;
5193 
5194       if (user_sys_cpu_time) {
5195         return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5196       } else {
5197         return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5198       }
5199     }
5200     else proc_pid_cpu_avail = false;
5201   }
5202 
5203   // The /proc/<tid>/stat aggregates per-process usage on
5204   // new Bsd kernels 2.6+ where NPTL is supported.
5205   // The /proc/self/task/<tid>/stat still has the per-thread usage.
5206   // See bug 6328462.
5207   // There can be no directory /proc/self/task on kernels 2.4 with NPTL
5208   // and possibly in some other cases, so we check its availability.
5209   if (proc_task_unchecked && os::Bsd::is_NPTL()) {
5210     // This is executed only once
5211     proc_task_unchecked = false;
5212     fp = fopen("/proc/self/task", "r");
5213     if (fp != NULL) {
5214       proc_stat_path = "/proc/self/task/%d/stat";
5215       fclose(fp);
5216     }
5217   }
5218 
5219   sprintf(proc_name, proc_stat_path, tid);
5220   fp = fopen(proc_name, "r");
5221   if ( fp == NULL ) return -1;
5222   statlen = fread(stat, 1, 2047, fp);
5223   stat[statlen] = '\0';
5224   fclose(fp);
5225 
5226   // Skip pid and the command string. Note that we could be dealing with
5227   // weird command names, e.g. user could decide to rename java launcher
5228   // to "java 1.4.2 :)", then the stat file would look like
5229   //                1234 (java 1.4.2 :)) R ... ...
5230   // We don't really need to know the command string, just find the last
5231   // occurrence of ")" and then start parsing from there. See bug 4726580.
5232   s = strrchr(stat, ')');
5233   i = 0;
5234   if (s == NULL ) return -1;
5235 
5236   // Skip blank chars
5237   do s++; while (isspace(*s));
5238 
5239   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5240                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5241                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5242                  &user_time, &sys_time);
5243   if ( count != 13 ) return -1;
5244   if (user_sys_cpu_time) {
5245     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5246   } else {
5247     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5248   }
5249 }
5250 #endif
5251 
5252 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5253   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5254   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5255   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5256   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5257 }
5258 
5259 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5260   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5261   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5262   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5263   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5264 }
5265 
5266 bool os::is_thread_cpu_time_supported() {
5267 #ifdef __APPLE__
5268   return true;
5269 #elif defined(_ALLBSD_SOURCE)
5270   return false;
5271 #else
5272   return true;
5273 #endif
5274 }
5275 
5276 // System loadavg support.  Returns -1 if load average cannot be obtained.
5277 // Bsd doesn't yet have a (official) notion of processor sets,
5278 // so just return the system wide load average.
5279 int os::loadavg(double loadavg[], int nelem) {
5280   return ::getloadavg(loadavg, nelem);
5281 }
5282 
5283 void os::pause() {
5284   char filename[MAX_PATH];
5285   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5286     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5287   } else {
5288     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5289   }
5290 
5291   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5292   if (fd != -1) {
5293     struct stat buf;
5294     ::close(fd);
5295     while (::stat(filename, &buf) == 0) {
5296       (void)::poll(NULL, 0, 100);
5297     }
5298   } else {
5299     jio_fprintf(stderr,
5300       "Could not open pause file '%s', continuing immediately.\n", filename);
5301   }
5302 }
5303 
5304 
5305 // Refer to the comments in os_solaris.cpp park-unpark.
5306 //
5307 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5308 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5309 // For specifics regarding the bug see GLIBC BUGID 261237 :
5310 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5311 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5312 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5313 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5314 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5315 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5316 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5317 // of libpthread avoids the problem, but isn't practical.
5318 //
5319 // Possible remedies:
5320 //
5321 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5322 //      This is palliative and probabilistic, however.  If the thread is preempted
5323 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5324 //      than the minimum period may have passed, and the abstime may be stale (in the
5325 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5326 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5327 //
5328 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5329 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5330 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5331 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5332 //      thread.
5333 //
5334 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5335 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5336 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5337 //      This also works well.  In fact it avoids kernel-level scalability impediments
5338 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5339 //      timers in a graceful fashion.
5340 //
5341 // 4.   When the abstime value is in the past it appears that control returns
5342 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5343 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5344 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5345 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5346 //      It may be possible to avoid reinitialization by checking the return
5347 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5348 //      condvar we must establish the invariant that cond_signal() is only called
5349 //      within critical sections protected by the adjunct mutex.  This prevents
5350 //      cond_signal() from "seeing" a condvar that's in the midst of being
5351 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5352 //      desirable signal-after-unlock optimization that avoids futile context switching.
5353 //
5354 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5355 //      structure when a condvar is used or initialized.  cond_destroy()  would
5356 //      release the helper structure.  Our reinitialize-after-timedwait fix
5357 //      put excessive stress on malloc/free and locks protecting the c-heap.
5358 //
5359 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5360 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5361 // and only enabling the work-around for vulnerable environments.
5362 
5363 // utility to compute the abstime argument to timedwait:
5364 // millis is the relative timeout time
5365 // abstime will be the absolute timeout time
5366 // TODO: replace compute_abstime() with unpackTime()
5367 
5368 static struct timespec* compute_abstime(struct timespec* abstime, jlong millis) {
5369   if (millis < 0)  millis = 0;
5370   struct timeval now;
5371   int status = gettimeofday(&now, NULL);
5372   assert(status == 0, "gettimeofday");
5373   jlong seconds = millis / 1000;
5374   millis %= 1000;
5375   if (seconds > 50000000) { // see man cond_timedwait(3T)
5376     seconds = 50000000;
5377   }
5378   abstime->tv_sec = now.tv_sec  + seconds;
5379   long       usec = now.tv_usec + millis * 1000;
5380   if (usec >= 1000000) {
5381     abstime->tv_sec += 1;
5382     usec -= 1000000;
5383   }
5384   abstime->tv_nsec = usec * 1000;
5385   return abstime;
5386 }
5387 
5388 
5389 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5390 // Conceptually TryPark() should be equivalent to park(0).
5391 
5392 int os::PlatformEvent::TryPark() {
5393   for (;;) {
5394     const int v = _Event ;
5395     guarantee ((v == 0) || (v == 1), "invariant") ;
5396     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5397   }
5398 }
5399 
5400 void os::PlatformEvent::park() {       // AKA "down()"
5401   // Invariant: Only the thread associated with the Event/PlatformEvent
5402   // may call park().
5403   // TODO: assert that _Assoc != NULL or _Assoc == Self
5404   int v ;
5405   for (;;) {
5406       v = _Event ;
5407       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5408   }
5409   guarantee (v >= 0, "invariant") ;
5410   if (v == 0) {
5411      // Do this the hard way by blocking ...
5412      int status = pthread_mutex_lock(_mutex);
5413      assert_status(status == 0, status, "mutex_lock");
5414      guarantee (_nParked == 0, "invariant") ;
5415      ++ _nParked ;
5416      while (_Event < 0) {
5417         status = pthread_cond_wait(_cond, _mutex);
5418         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5419         // Treat this the same as if the wait was interrupted
5420         if (status == ETIMEDOUT) { status = EINTR; }
5421         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5422      }
5423      -- _nParked ;
5424 
5425     // In theory we could move the ST of 0 into _Event past the unlock(),
5426     // but then we'd need a MEMBAR after the ST.
5427     _Event = 0 ;
5428      status = pthread_mutex_unlock(_mutex);
5429      assert_status(status == 0, status, "mutex_unlock");
5430   }
5431   guarantee (_Event >= 0, "invariant") ;
5432 }
5433 
5434 int os::PlatformEvent::park(jlong millis) {
5435   guarantee (_nParked == 0, "invariant") ;
5436 
5437   int v ;
5438   for (;;) {
5439       v = _Event ;
5440       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5441   }
5442   guarantee (v >= 0, "invariant") ;
5443   if (v != 0) return OS_OK ;
5444 
5445   // We do this the hard way, by blocking the thread.
5446   // Consider enforcing a minimum timeout value.
5447   struct timespec abst;
5448   compute_abstime(&abst, millis);
5449 
5450   int ret = OS_TIMEOUT;
5451   int status = pthread_mutex_lock(_mutex);
5452   assert_status(status == 0, status, "mutex_lock");
5453   guarantee (_nParked == 0, "invariant") ;
5454   ++_nParked ;
5455 
5456   // Object.wait(timo) will return because of
5457   // (a) notification
5458   // (b) timeout
5459   // (c) thread.interrupt
5460   //
5461   // Thread.interrupt and object.notify{All} both call Event::set.
5462   // That is, we treat thread.interrupt as a special case of notification.
5463   // The underlying Solaris implementation, cond_timedwait, admits
5464   // spurious/premature wakeups, but the JLS/JVM spec prevents the
5465   // JVM from making those visible to Java code.  As such, we must
5466   // filter out spurious wakeups.  We assume all ETIME returns are valid.
5467   //
5468   // TODO: properly differentiate simultaneous notify+interrupt.
5469   // In that case, we should propagate the notify to another waiter.
5470 
5471   while (_Event < 0) {
5472     status = os::Bsd::safe_cond_timedwait(_cond, _mutex, &abst);
5473     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5474       pthread_cond_destroy (_cond);
5475       pthread_cond_init (_cond, NULL) ;
5476     }
5477     assert_status(status == 0 || status == EINTR ||
5478                   status == ETIMEDOUT,
5479                   status, "cond_timedwait");
5480     if (!FilterSpuriousWakeups) break ;                 // previous semantics
5481     if (status == ETIMEDOUT) break ;
5482     // We consume and ignore EINTR and spurious wakeups.
5483   }
5484   --_nParked ;
5485   if (_Event >= 0) {
5486      ret = OS_OK;
5487   }
5488   _Event = 0 ;
5489   status = pthread_mutex_unlock(_mutex);
5490   assert_status(status == 0, status, "mutex_unlock");
5491   assert (_nParked == 0, "invariant") ;
5492   return ret;
5493 }
5494 
5495 void os::PlatformEvent::unpark() {
5496   int v, AnyWaiters ;
5497   for (;;) {
5498       v = _Event ;
5499       if (v > 0) {
5500          // The LD of _Event could have reordered or be satisfied
5501          // by a read-aside from this processor's write buffer.
5502          // To avoid problems execute a barrier and then
5503          // ratify the value.
5504          OrderAccess::fence() ;
5505          if (_Event == v) return ;
5506          continue ;
5507       }
5508       if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5509   }
5510   if (v < 0) {
5511      // Wait for the thread associated with the event to vacate
5512      int status = pthread_mutex_lock(_mutex);
5513      assert_status(status == 0, status, "mutex_lock");
5514      AnyWaiters = _nParked ;
5515      assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5516      if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5517         AnyWaiters = 0 ;
5518         pthread_cond_signal (_cond);
5519      }
5520      status = pthread_mutex_unlock(_mutex);
5521      assert_status(status == 0, status, "mutex_unlock");
5522      if (AnyWaiters != 0) {
5523         status = pthread_cond_signal(_cond);
5524         assert_status(status == 0, status, "cond_signal");
5525      }
5526   }
5527 
5528   // Note that we signal() _after dropping the lock for "immortal" Events.
5529   // This is safe and avoids a common class of  futile wakeups.  In rare
5530   // circumstances this can cause a thread to return prematurely from
5531   // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5532   // simply re-test the condition and re-park itself.
5533 }
5534 
5535 
5536 // JSR166
5537 // -------------------------------------------------------
5538 
5539 /*
5540  * The solaris and bsd implementations of park/unpark are fairly
5541  * conservative for now, but can be improved. They currently use a
5542  * mutex/condvar pair, plus a a count.
5543  * Park decrements count if > 0, else does a condvar wait.  Unpark
5544  * sets count to 1 and signals condvar.  Only one thread ever waits
5545  * on the condvar. Contention seen when trying to park implies that someone
5546  * is unparking you, so don't wait. And spurious returns are fine, so there
5547  * is no need to track notifications.
5548  */
5549 
5550 #define MAX_SECS 100000000
5551 /*
5552  * This code is common to bsd and solaris and will be moved to a
5553  * common place in dolphin.
5554  *
5555  * The passed in time value is either a relative time in nanoseconds
5556  * or an absolute time in milliseconds. Either way it has to be unpacked
5557  * into suitable seconds and nanoseconds components and stored in the
5558  * given timespec structure.
5559  * Given time is a 64-bit value and the time_t used in the timespec is only
5560  * a signed-32-bit value (except on 64-bit Bsd) we have to watch for
5561  * overflow if times way in the future are given. Further on Solaris versions
5562  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5563  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5564  * As it will be 28 years before "now + 100000000" will overflow we can
5565  * ignore overflow and just impose a hard-limit on seconds using the value
5566  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5567  * years from "now".
5568  */
5569 
5570 static void unpackTime(struct timespec* absTime, bool isAbsolute, jlong time) {
5571   assert (time > 0, "convertTime");
5572 
5573   struct timeval now;
5574   int status = gettimeofday(&now, NULL);
5575   assert(status == 0, "gettimeofday");
5576 
5577   time_t max_secs = now.tv_sec + MAX_SECS;
5578 
5579   if (isAbsolute) {
5580     jlong secs = time / 1000;
5581     if (secs > max_secs) {
5582       absTime->tv_sec = max_secs;
5583     }
5584     else {
5585       absTime->tv_sec = secs;
5586     }
5587     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5588   }
5589   else {
5590     jlong secs = time / NANOSECS_PER_SEC;
5591     if (secs >= MAX_SECS) {
5592       absTime->tv_sec = max_secs;
5593       absTime->tv_nsec = 0;
5594     }
5595     else {
5596       absTime->tv_sec = now.tv_sec + secs;
5597       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5598       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5599         absTime->tv_nsec -= NANOSECS_PER_SEC;
5600         ++absTime->tv_sec; // note: this must be <= max_secs
5601       }
5602     }
5603   }
5604   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5605   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5606   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5607   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5608 }
5609 
5610 void Parker::park(bool isAbsolute, jlong time) {
5611   // Optional fast-path check:
5612   // Return immediately if a permit is available.
5613   if (_counter > 0) {
5614       _counter = 0 ;
5615       OrderAccess::fence();
5616       return ;
5617   }
5618 
5619   Thread* thread = Thread::current();
5620   assert(thread->is_Java_thread(), "Must be JavaThread");
5621   JavaThread *jt = (JavaThread *)thread;
5622 
5623   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5624   // Check interrupt before trying to wait
5625   if (Thread::is_interrupted(thread, false)) {
5626     return;
5627   }
5628 
5629   // Next, demultiplex/decode time arguments
5630   struct timespec absTime;
5631   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5632     return;
5633   }
5634   if (time > 0) {
5635     unpackTime(&absTime, isAbsolute, time);
5636   }
5637 
5638 
5639   // Enter safepoint region
5640   // Beware of deadlocks such as 6317397.
5641   // The per-thread Parker:: mutex is a classic leaf-lock.
5642   // In particular a thread must never block on the Threads_lock while
5643   // holding the Parker:: mutex.  If safepoints are pending both the
5644   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5645   ThreadBlockInVM tbivm(jt);
5646 
5647   // Don't wait if cannot get lock since interference arises from
5648   // unblocking.  Also. check interrupt before trying wait
5649   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5650     return;
5651   }
5652 
5653   int status ;
5654   if (_counter > 0)  { // no wait needed
5655     _counter = 0;
5656     status = pthread_mutex_unlock(_mutex);
5657     assert (status == 0, "invariant") ;
5658     OrderAccess::fence();
5659     return;
5660   }
5661 
5662 #ifdef ASSERT
5663   // Don't catch signals while blocked; let the running threads have the signals.
5664   // (This allows a debugger to break into the running thread.)
5665   sigset_t oldsigs;
5666   sigset_t* allowdebug_blocked = os::Bsd::allowdebug_blocked_signals();
5667   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5668 #endif
5669 
5670   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5671   jt->set_suspend_equivalent();
5672   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5673 
5674   if (time == 0) {
5675     status = pthread_cond_wait (_cond, _mutex) ;
5676   } else {
5677     status = os::Bsd::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5678     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5679       pthread_cond_destroy (_cond) ;
5680       pthread_cond_init    (_cond, NULL);
5681     }
5682   }
5683   assert_status(status == 0 || status == EINTR ||
5684                 status == ETIMEDOUT,
5685                 status, "cond_timedwait");
5686 
5687 #ifdef ASSERT
5688   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5689 #endif
5690 
5691   _counter = 0 ;
5692   status = pthread_mutex_unlock(_mutex) ;
5693   assert_status(status == 0, status, "invariant") ;
5694   // If externally suspended while waiting, re-suspend
5695   if (jt->handle_special_suspend_equivalent_condition()) {
5696     jt->java_suspend_self();
5697   }
5698 
5699   OrderAccess::fence();
5700 }
5701 
5702 void Parker::unpark() {
5703   int s, status ;
5704   status = pthread_mutex_lock(_mutex);
5705   assert (status == 0, "invariant") ;
5706   s = _counter;
5707   _counter = 1;
5708   if (s < 1) {
5709      if (WorkAroundNPTLTimedWaitHang) {
5710         status = pthread_cond_signal (_cond) ;
5711         assert (status == 0, "invariant") ;
5712         status = pthread_mutex_unlock(_mutex);
5713         assert (status == 0, "invariant") ;
5714      } else {
5715         status = pthread_mutex_unlock(_mutex);
5716         assert (status == 0, "invariant") ;
5717         status = pthread_cond_signal (_cond) ;
5718         assert (status == 0, "invariant") ;
5719      }
5720   } else {
5721     pthread_mutex_unlock(_mutex);
5722     assert (status == 0, "invariant") ;
5723   }
5724 }
5725 
5726 
5727 /* Darwin has no "environ" in a dynamic library. */
5728 #ifdef __APPLE__
5729 #include <crt_externs.h>
5730 #define environ (*_NSGetEnviron())
5731 #else
5732 extern char** environ;
5733 #endif
5734 
5735 // Run the specified command in a separate process. Return its exit value,
5736 // or -1 on failure (e.g. can't fork a new process).
5737 // Unlike system(), this function can be called from signal handler. It
5738 // doesn't block SIGINT et al.
5739 int os::fork_and_exec(char* cmd) {
5740   const char * argv[4] = {"sh", "-c", cmd, NULL};
5741 
5742   // fork() in BsdThreads/NPTL is not async-safe. It needs to run
5743   // pthread_atfork handlers and reset pthread library. All we need is a
5744   // separate process to execve. Make a direct syscall to fork process.
5745   // On IA64 there's no fork syscall, we have to use fork() and hope for
5746   // the best...
5747   pid_t pid = fork();
5748 
5749   if (pid < 0) {
5750     // fork failed
5751     return -1;
5752 
5753   } else if (pid == 0) {
5754     // child process
5755 
5756     // execve() in BsdThreads will call pthread_kill_other_threads_np()
5757     // first to kill every thread on the thread list. Because this list is
5758     // not reset by fork() (see notes above), execve() will instead kill
5759     // every thread in the parent process. We know this is the only thread
5760     // in the new process, so make a system call directly.
5761     // IA64 should use normal execve() from glibc to match the glibc fork()
5762     // above.
5763     execve("/bin/sh", (char* const*)argv, environ);
5764 
5765     // execve failed
5766     _exit(-1);
5767 
5768   } else  {
5769     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5770     // care about the actual exit code, for now.
5771 
5772     int status;
5773 
5774     // Wait for the child process to exit.  This returns immediately if
5775     // the child has already exited. */
5776     while (waitpid(pid, &status, 0) < 0) {
5777         switch (errno) {
5778         case ECHILD: return 0;
5779         case EINTR: break;
5780         default: return -1;
5781         }
5782     }
5783 
5784     if (WIFEXITED(status)) {
5785        // The child exited normally; get its exit code.
5786        return WEXITSTATUS(status);
5787     } else if (WIFSIGNALED(status)) {
5788        // The child exited because of a signal
5789        // The best value to return is 0x80 + signal number,
5790        // because that is what all Unix shells do, and because
5791        // it allows callers to distinguish between process exit and
5792        // process death by signal.
5793        return 0x80 + WTERMSIG(status);
5794     } else {
5795        // Unknown exit code; pass it through
5796        return status;
5797     }
5798   }
5799 }
5800 
5801 // is_headless_jre()
5802 //
5803 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5804 // in order to report if we are running in a headless jre
5805 //
5806 // Since JDK8 xawt/libmawt.so was moved into the same directory
5807 // as libawt.so, and renamed libawt_xawt.so
5808 //
5809 bool os::is_headless_jre() {
5810     struct stat statbuf;
5811     char buf[MAXPATHLEN];
5812     char libmawtpath[MAXPATHLEN];
5813     const char *xawtstr  = "/xawt/libmawt" JNI_LIB_SUFFIX;
5814     const char *new_xawtstr = "/libawt_xawt" JNI_LIB_SUFFIX;
5815     char *p;
5816 
5817     // Get path to libjvm.so
5818     os::jvm_path(buf, sizeof(buf));
5819 
5820     // Get rid of libjvm.so
5821     p = strrchr(buf, '/');
5822     if (p == NULL) return false;
5823     else *p = '\0';
5824 
5825     // Get rid of client or server
5826     p = strrchr(buf, '/');
5827     if (p == NULL) return false;
5828     else *p = '\0';
5829 
5830     // check xawt/libmawt.so
5831     strcpy(libmawtpath, buf);
5832     strcat(libmawtpath, xawtstr);
5833     if (::stat(libmawtpath, &statbuf) == 0) return false;
5834 
5835     // check libawt_xawt.so
5836     strcpy(libmawtpath, buf);
5837     strcat(libmawtpath, new_xawtstr);
5838     if (::stat(libmawtpath, &statbuf) == 0) return false;
5839 
5840     return true;
5841 }