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