1 /*
   2  * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "jvm.h"
  26 #include "logging/log.hpp"
  27 #include "memory/allocation.inline.hpp"
  28 #include "utilities/globalDefinitions.hpp"
  29 #include "runtime/frame.inline.hpp"
  30 #include "runtime/interfaceSupport.inline.hpp"
  31 #include "runtime/os.hpp"
  32 #include "services/memTracker.hpp"
  33 #include "utilities/align.hpp"
  34 #include "utilities/formatBuffer.hpp"
  35 #include "utilities/macros.hpp"
  36 #include "utilities/vmError.hpp"
  37 
  38 #include <dlfcn.h>
  39 #include <pthread.h>
  40 #include <signal.h>
  41 #include <sys/mman.h>
  42 #include <sys/resource.h>
  43 #include <sys/utsname.h>
  44 #include <time.h>
  45 #include <unistd.h>
  46 
  47 // Todo: provide a os::get_max_process_id() or similar. Number of processes
  48 // may have been configured, can be read more accurately from proc fs etc.
  49 #ifndef MAX_PID
  50 #define MAX_PID INT_MAX
  51 #endif
  52 #define IS_VALID_PID(p) (p > 0 && p < MAX_PID)
  53 
  54 #ifndef MAP_ANONYMOUS
  55   #define MAP_ANONYMOUS MAP_ANON
  56 #endif
  57 
  58 #define check_with_errno(check_type, cond, msg)                             \
  59   do {                                                                      \
  60     int err = errno;                                                        \
  61     check_type(cond, "%s; error='%s' (errno=%s)", msg, os::strerror(err),   \
  62                os::errno_name(err));                                        \
  63 } while (false)
  64 
  65 #define assert_with_errno(cond, msg)    check_with_errno(assert, cond, msg)
  66 #define guarantee_with_errno(cond, msg) check_with_errno(guarantee, cond, msg)
  67 
  68 // Check core dump limit and report possible place where core can be found
  69 void os::check_dump_limit(char* buffer, size_t bufferSize) {
  70   if (!FLAG_IS_DEFAULT(CreateCoredumpOnCrash) && !CreateCoredumpOnCrash) {
  71     jio_snprintf(buffer, bufferSize, "CreateCoredumpOnCrash is disabled from command line");
  72     VMError::record_coredump_status(buffer, false);
  73     return;
  74   }
  75 
  76   int n;
  77   struct rlimit rlim;
  78   bool success;
  79 
  80   char core_path[PATH_MAX];
  81   n = get_core_path(core_path, PATH_MAX);
  82 
  83   if (n <= 0) {
  84     jio_snprintf(buffer, bufferSize, "core.%d (may not exist)", current_process_id());
  85     success = true;
  86 #ifdef LINUX
  87   } else if (core_path[0] == '"') { // redirect to user process
  88     jio_snprintf(buffer, bufferSize, "Core dumps may be processed with %s", core_path);
  89     success = true;
  90 #endif
  91   } else if (getrlimit(RLIMIT_CORE, &rlim) != 0) {
  92     jio_snprintf(buffer, bufferSize, "%s (may not exist)", core_path);
  93     success = true;
  94   } else {
  95     switch(rlim.rlim_cur) {
  96       case RLIM_INFINITY:
  97         jio_snprintf(buffer, bufferSize, "%s", core_path);
  98         success = true;
  99         break;
 100       case 0:
 101         jio_snprintf(buffer, bufferSize, "Core dumps have been disabled. To enable core dumping, try \"ulimit -c unlimited\" before starting Java again");
 102         success = false;
 103         break;
 104       default:
 105         jio_snprintf(buffer, bufferSize, "%s (max size " UINT64_FORMAT " kB). To ensure a full core dump, try \"ulimit -c unlimited\" before starting Java again", core_path, uint64_t(rlim.rlim_cur) / 1024);
 106         success = true;
 107         break;
 108     }
 109   }
 110 
 111   VMError::record_coredump_status(buffer, success);
 112 }
 113 
 114 int os::get_native_stack(address* stack, int frames, int toSkip) {
 115   int frame_idx = 0;
 116   int num_of_frames;  // number of frames captured
 117   frame fr = os::current_frame();
 118   while (fr.pc() && frame_idx < frames) {
 119     if (toSkip > 0) {
 120       toSkip --;
 121     } else {
 122       stack[frame_idx ++] = fr.pc();
 123     }
 124     if (fr.fp() == NULL || fr.cb() != NULL ||
 125         fr.sender_pc() == NULL || os::is_first_C_frame(&fr)) break;
 126 
 127     if (fr.sender_pc() && !os::is_first_C_frame(&fr)) {
 128       fr = os::get_sender_for_C_frame(&fr);
 129     } else {
 130       break;
 131     }
 132   }
 133   num_of_frames = frame_idx;
 134   for (; frame_idx < frames; frame_idx ++) {
 135     stack[frame_idx] = NULL;
 136   }
 137 
 138   return num_of_frames;
 139 }
 140 
 141 
 142 bool os::unsetenv(const char* name) {
 143   assert(name != NULL, "Null pointer");
 144   return (::unsetenv(name) == 0);
 145 }
 146 
 147 int os::get_last_error() {
 148   return errno;
 149 }
 150 
 151 bool os::is_debugger_attached() {
 152   // not implemented
 153   return false;
 154 }
 155 
 156 void os::wait_for_keypress_at_exit(void) {
 157   // don't do anything on posix platforms
 158   return;
 159 }
 160 
 161 int os::create_file_for_heap(const char* dir) {
 162 
 163   const char name_template[] = "/jvmheap.XXXXXX";
 164 
 165   char *fullname = (char*)os::malloc((strlen(dir) + strlen(name_template) + 1), mtInternal);
 166   if (fullname == NULL) {
 167     vm_exit_during_initialization(err_msg("Malloc failed during creation of backing file for heap (%s)", os::strerror(errno)));
 168     return -1;
 169   }
 170   (void)strncpy(fullname, dir, strlen(dir)+1);
 171   (void)strncat(fullname, name_template, strlen(name_template));
 172 
 173   os::native_path(fullname);
 174 
 175   sigset_t set, oldset;
 176   int ret = sigfillset(&set);
 177   assert_with_errno(ret == 0, "sigfillset returned error");
 178 
 179   // set the file creation mask.
 180   mode_t file_mode = S_IRUSR | S_IWUSR;
 181 
 182   // create a new file.
 183   int fd = mkstemp(fullname);
 184 
 185   if (fd < 0) {
 186     warning("Could not create file for heap with template %s", fullname);
 187     os::free(fullname);
 188     return -1;
 189   }
 190 
 191   // delete the name from the filesystem. When 'fd' is closed, the file (and space) will be deleted.
 192   ret = unlink(fullname);
 193   assert_with_errno(ret == 0, "unlink returned error");
 194 
 195   os::free(fullname);
 196   return fd;
 197 }
 198 
 199 static char* reserve_mmapped_memory(size_t bytes, char* requested_addr) {
 200   char * addr;
 201   int flags = MAP_PRIVATE NOT_AIX( | MAP_NORESERVE ) | MAP_ANONYMOUS;
 202   if (requested_addr != NULL) {
 203     assert((uintptr_t)requested_addr % os::vm_page_size() == 0, "Requested address should be aligned to OS page size");
 204     flags |= MAP_FIXED;
 205   }
 206 
 207   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
 208   // touch an uncommitted page. Otherwise, the read/write might
 209   // succeed if we have enough swap space to back the physical page.
 210   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
 211                        flags, -1, 0);
 212 
 213   if (addr != MAP_FAILED) {
 214     MemTracker::record_virtual_memory_reserve((address)addr, bytes, CALLER_PC);
 215     return addr;
 216   }
 217   return NULL;
 218 }
 219 
 220 static int util_posix_fallocate(int fd, off_t offset, off_t len) {
 221 #ifdef __APPLE__
 222   fstore_t store = { F_ALLOCATECONTIG, F_PEOFPOSMODE, 0, len };
 223   // First we try to get a continuous chunk of disk space
 224   int ret = fcntl(fd, F_PREALLOCATE, &store);
 225   if (ret == -1) {
 226     // Maybe we are too fragmented, try to allocate non-continuous range
 227     store.fst_flags = F_ALLOCATEALL;
 228     ret = fcntl(fd, F_PREALLOCATE, &store);
 229   }
 230   if(ret != -1) {
 231     return ftruncate(fd, len);
 232   }
 233   return -1;
 234 #else
 235   return posix_fallocate(fd, offset, len);
 236 #endif
 237 }
 238 
 239 // Map the given address range to the provided file descriptor.
 240 char* os::map_memory_to_file(char* base, size_t size, int fd) {
 241   assert(fd != -1, "File descriptor is not valid");
 242 
 243   // allocate space for the file
 244   int ret = util_posix_fallocate(fd, 0, (off_t)size);
 245   if (ret != 0) {
 246     vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory. error(%d)", ret));
 247     return NULL;
 248   }
 249 
 250   int prot = PROT_READ | PROT_WRITE;
 251   int flags = MAP_SHARED;
 252   if (base != NULL) {
 253     flags |= MAP_FIXED;
 254   }
 255   char* addr = (char*)mmap(base, size, prot, flags, fd, 0);
 256 
 257   if (addr == MAP_FAILED) {
 258     warning("Failed mmap to file. (%s)", os::strerror(errno));
 259     return NULL;
 260   }
 261   if (base != NULL && addr != base) {
 262     if (!os::release_memory(addr, size)) {
 263       warning("Could not release memory on unsuccessful file mapping");
 264     }
 265     return NULL;
 266   }
 267   return addr;
 268 }
 269 
 270 char* os::replace_existing_mapping_with_file_mapping(char* base, size_t size, int fd) {
 271   assert(fd != -1, "File descriptor is not valid");
 272   assert(base != NULL, "Base cannot be NULL");
 273 
 274   return map_memory_to_file(base, size, fd);
 275 }
 276 
 277 // Multiple threads can race in this code, and can remap over each other with MAP_FIXED,
 278 // so on posix, unmap the section at the start and at the end of the chunk that we mapped
 279 // rather than unmapping and remapping the whole chunk to get requested alignment.
 280 char* os::reserve_memory_aligned(size_t size, size_t alignment, int file_desc) {
 281   assert((alignment & (os::vm_allocation_granularity() - 1)) == 0,
 282       "Alignment must be a multiple of allocation granularity (page size)");
 283   assert((size & (alignment -1)) == 0, "size must be 'alignment' aligned");
 284 
 285   size_t extra_size = size + alignment;
 286   assert(extra_size >= size, "overflow, size is too large to allow alignment");
 287 
 288   char* extra_base;
 289   if (file_desc != -1) {
 290     // For file mapping, we do not call os:reserve_memory(extra_size, NULL, alignment, file_desc) because
 291     // we need to deal with shrinking of the file space later when we release extra memory after alignment.
 292     // We also cannot called os:reserve_memory() with file_desc set to -1 because on aix we might get SHM memory.
 293     // So here to call a helper function while reserve memory for us. After we have a aligned base,
 294     // we will replace anonymous mapping with file mapping.
 295     extra_base = reserve_mmapped_memory(extra_size, NULL);
 296     if (extra_base != NULL) {
 297       MemTracker::record_virtual_memory_reserve((address)extra_base, extra_size, CALLER_PC);
 298     }
 299   } else {
 300     extra_base = os::reserve_memory(extra_size, NULL, alignment);
 301   }
 302 
 303   if (extra_base == NULL) {
 304     return NULL;
 305   }
 306 
 307   // Do manual alignment
 308   char* aligned_base = align_up(extra_base, alignment);
 309 
 310   // [  |                                       |  ]
 311   // ^ extra_base
 312   //    ^ extra_base + begin_offset == aligned_base
 313   //     extra_base + begin_offset + size       ^
 314   //                       extra_base + extra_size ^
 315   // |<>| == begin_offset
 316   //                              end_offset == |<>|
 317   size_t begin_offset = aligned_base - extra_base;
 318   size_t end_offset = (extra_base + extra_size) - (aligned_base + size);
 319 
 320   if (begin_offset > 0) {
 321       os::release_memory(extra_base, begin_offset);
 322   }
 323 
 324   if (end_offset > 0) {
 325       os::release_memory(extra_base + begin_offset + size, end_offset);
 326   }
 327 
 328   if (file_desc != -1) {
 329     // After we have an aligned address, we can replace anonymous mapping with file mapping
 330     if (replace_existing_mapping_with_file_mapping(aligned_base, size, file_desc) == NULL) {
 331       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
 332     }
 333     MemTracker::record_virtual_memory_commit((address)aligned_base, size, CALLER_PC);
 334   }
 335   return aligned_base;
 336 }
 337 
 338 int os::vsnprintf(char* buf, size_t len, const char* fmt, va_list args) {
 339   // All supported POSIX platforms provide C99 semantics.
 340   int result = ::vsnprintf(buf, len, fmt, args);
 341   // If an encoding error occurred (result < 0) then it's not clear
 342   // whether the buffer is NUL terminated, so ensure it is.
 343   if ((result < 0) && (len > 0)) {
 344     buf[len - 1] = '\0';
 345   }
 346   return result;
 347 }
 348 
 349 int os::get_fileno(FILE* fp) {
 350   return NOT_AIX(::)fileno(fp);
 351 }
 352 
 353 struct tm* os::gmtime_pd(const time_t* clock, struct tm*  res) {
 354   return gmtime_r(clock, res);
 355 }
 356 
 357 void os::Posix::print_load_average(outputStream* st) {
 358   st->print("load average:");
 359   double loadavg[3];
 360   os::loadavg(loadavg, 3);
 361   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
 362   st->cr();
 363 }
 364 
 365 void os::Posix::print_rlimit_info(outputStream* st) {
 366   st->print("rlimit:");
 367   struct rlimit rlim;
 368 
 369   st->print(" STACK ");
 370   getrlimit(RLIMIT_STACK, &rlim);
 371   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 372   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 373 
 374   st->print(", CORE ");
 375   getrlimit(RLIMIT_CORE, &rlim);
 376   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 377   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 378 
 379   // Isn't there on solaris
 380 #if defined(AIX)
 381   st->print(", NPROC ");
 382   st->print("%d", sysconf(_SC_CHILD_MAX));
 383 #elif !defined(SOLARIS)
 384   st->print(", NPROC ");
 385   getrlimit(RLIMIT_NPROC, &rlim);
 386   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 387   else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
 388 #endif
 389 
 390   st->print(", NOFILE ");
 391   getrlimit(RLIMIT_NOFILE, &rlim);
 392   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 393   else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
 394 
 395   st->print(", AS ");
 396   getrlimit(RLIMIT_AS, &rlim);
 397   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 398   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 399 
 400   st->print(", DATA ");
 401   getrlimit(RLIMIT_DATA, &rlim);
 402   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 403   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 404 
 405   st->print(", FSIZE ");
 406   getrlimit(RLIMIT_FSIZE, &rlim);
 407   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 408   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 409 
 410   st->cr();
 411 }
 412 
 413 void os::Posix::print_uname_info(outputStream* st) {
 414   // kernel
 415   st->print("uname:");
 416   struct utsname name;
 417   uname(&name);
 418   st->print("%s ", name.sysname);
 419 #ifdef ASSERT
 420   st->print("%s ", name.nodename);
 421 #endif
 422   st->print("%s ", name.release);
 423   st->print("%s ", name.version);
 424   st->print("%s", name.machine);
 425   st->cr();
 426 }
 427 
 428 bool os::get_host_name(char* buf, size_t buflen) {
 429   struct utsname name;
 430   uname(&name);
 431   jio_snprintf(buf, buflen, "%s", name.nodename);
 432   return true;
 433 }
 434 
 435 bool os::has_allocatable_memory_limit(julong* limit) {
 436   struct rlimit rlim;
 437   int getrlimit_res = getrlimit(RLIMIT_AS, &rlim);
 438   // if there was an error when calling getrlimit, assume that there is no limitation
 439   // on virtual memory.
 440   bool result;
 441   if ((getrlimit_res != 0) || (rlim.rlim_cur == RLIM_INFINITY)) {
 442     result = false;
 443   } else {
 444     *limit = (julong)rlim.rlim_cur;
 445     result = true;
 446   }
 447 #ifdef _LP64
 448   return result;
 449 #else
 450   // arbitrary virtual space limit for 32 bit Unices found by testing. If
 451   // getrlimit above returned a limit, bound it with this limit. Otherwise
 452   // directly use it.
 453   const julong max_virtual_limit = (julong)3800*M;
 454   if (result) {
 455     *limit = MIN2(*limit, max_virtual_limit);
 456   } else {
 457     *limit = max_virtual_limit;
 458   }
 459 
 460   // bound by actually allocatable memory. The algorithm uses two bounds, an
 461   // upper and a lower limit. The upper limit is the current highest amount of
 462   // memory that could not be allocated, the lower limit is the current highest
 463   // amount of memory that could be allocated.
 464   // The algorithm iteratively refines the result by halving the difference
 465   // between these limits, updating either the upper limit (if that value could
 466   // not be allocated) or the lower limit (if the that value could be allocated)
 467   // until the difference between these limits is "small".
 468 
 469   // the minimum amount of memory we care about allocating.
 470   const julong min_allocation_size = M;
 471 
 472   julong upper_limit = *limit;
 473 
 474   // first check a few trivial cases
 475   if (is_allocatable(upper_limit) || (upper_limit <= min_allocation_size)) {
 476     *limit = upper_limit;
 477   } else if (!is_allocatable(min_allocation_size)) {
 478     // we found that not even min_allocation_size is allocatable. Return it
 479     // anyway. There is no point to search for a better value any more.
 480     *limit = min_allocation_size;
 481   } else {
 482     // perform the binary search.
 483     julong lower_limit = min_allocation_size;
 484     while ((upper_limit - lower_limit) > min_allocation_size) {
 485       julong temp_limit = ((upper_limit - lower_limit) / 2) + lower_limit;
 486       temp_limit = align_down(temp_limit, min_allocation_size);
 487       if (is_allocatable(temp_limit)) {
 488         lower_limit = temp_limit;
 489       } else {
 490         upper_limit = temp_limit;
 491       }
 492     }
 493     *limit = lower_limit;
 494   }
 495   return true;
 496 #endif
 497 }
 498 
 499 const char* os::get_current_directory(char *buf, size_t buflen) {
 500   return getcwd(buf, buflen);
 501 }
 502 
 503 FILE* os::open(int fd, const char* mode) {
 504   return ::fdopen(fd, mode);
 505 }
 506 
 507 void os::flockfile(FILE* fp) {
 508   ::flockfile(fp);
 509 }
 510 
 511 void os::funlockfile(FILE* fp) {
 512   ::funlockfile(fp);
 513 }
 514 
 515 // Builds a platform dependent Agent_OnLoad_<lib_name> function name
 516 // which is used to find statically linked in agents.
 517 // Parameters:
 518 //            sym_name: Symbol in library we are looking for
 519 //            lib_name: Name of library to look in, NULL for shared libs.
 520 //            is_absolute_path == true if lib_name is absolute path to agent
 521 //                                     such as "/a/b/libL.so"
 522 //            == false if only the base name of the library is passed in
 523 //               such as "L"
 524 char* os::build_agent_function_name(const char *sym_name, const char *lib_name,
 525                                     bool is_absolute_path) {
 526   char *agent_entry_name;
 527   size_t len;
 528   size_t name_len;
 529   size_t prefix_len = strlen(JNI_LIB_PREFIX);
 530   size_t suffix_len = strlen(JNI_LIB_SUFFIX);
 531   const char *start;
 532 
 533   if (lib_name != NULL) {
 534     name_len = strlen(lib_name);
 535     if (is_absolute_path) {
 536       // Need to strip path, prefix and suffix
 537       if ((start = strrchr(lib_name, *os::file_separator())) != NULL) {
 538         lib_name = ++start;
 539       }
 540       if (strlen(lib_name) <= (prefix_len + suffix_len)) {
 541         return NULL;
 542       }
 543       lib_name += prefix_len;
 544       name_len = strlen(lib_name) - suffix_len;
 545     }
 546   }
 547   len = (lib_name != NULL ? name_len : 0) + strlen(sym_name) + 2;
 548   agent_entry_name = NEW_C_HEAP_ARRAY_RETURN_NULL(char, len, mtThread);
 549   if (agent_entry_name == NULL) {
 550     return NULL;
 551   }
 552   strcpy(agent_entry_name, sym_name);
 553   if (lib_name != NULL) {
 554     strcat(agent_entry_name, "_");
 555     strncat(agent_entry_name, lib_name, name_len);
 556   }
 557   return agent_entry_name;
 558 }
 559 
 560 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
 561   assert(thread == Thread::current(),  "thread consistency check");
 562 
 563   ParkEvent * const slp = thread->_SleepEvent ;
 564   slp->reset() ;
 565   OrderAccess::fence() ;
 566 
 567   if (interruptible) {
 568     jlong prevtime = javaTimeNanos();
 569 
 570     for (;;) {
 571       if (os::is_interrupted(thread, true)) {
 572         return OS_INTRPT;
 573       }
 574 
 575       jlong newtime = javaTimeNanos();
 576 
 577       if (newtime - prevtime < 0) {
 578         // time moving backwards, should only happen if no monotonic clock
 579         // not a guarantee() because JVM should not abort on kernel/glibc bugs
 580         assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected in os::sleep(interruptible)");
 581       } else {
 582         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
 583       }
 584 
 585       if (millis <= 0) {
 586         return OS_OK;
 587       }
 588 
 589       prevtime = newtime;
 590 
 591       {
 592         assert(thread->is_Java_thread(), "sanity check");
 593         JavaThread *jt = (JavaThread *) thread;
 594         ThreadBlockInVM tbivm(jt);
 595         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
 596 
 597         jt->set_suspend_equivalent();
 598         // cleared by handle_special_suspend_equivalent_condition() or
 599         // java_suspend_self() via check_and_wait_while_suspended()
 600 
 601         slp->park(millis);
 602 
 603         // were we externally suspended while we were waiting?
 604         jt->check_and_wait_while_suspended();
 605       }
 606     }
 607   } else {
 608     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
 609     jlong prevtime = javaTimeNanos();
 610 
 611     for (;;) {
 612       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
 613       // the 1st iteration ...
 614       jlong newtime = javaTimeNanos();
 615 
 616       if (newtime - prevtime < 0) {
 617         // time moving backwards, should only happen if no monotonic clock
 618         // not a guarantee() because JVM should not abort on kernel/glibc bugs
 619         assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected on os::sleep(!interruptible)");
 620       } else {
 621         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
 622       }
 623 
 624       if (millis <= 0) break ;
 625 
 626       prevtime = newtime;
 627       slp->park(millis);
 628     }
 629     return OS_OK ;
 630   }
 631 }
 632 
 633 ////////////////////////////////////////////////////////////////////////////////
 634 // interrupt support
 635 
 636 void os::interrupt(Thread* thread) {
 637   debug_only(Thread::check_for_dangling_thread_pointer(thread);)
 638 
 639   OSThread* osthread = thread->osthread();
 640 
 641   if (!osthread->interrupted()) {
 642     osthread->set_interrupted(true);
 643     // More than one thread can get here with the same value of osthread,
 644     // resulting in multiple notifications.  We do, however, want the store
 645     // to interrupted() to be visible to other threads before we execute unpark().
 646     OrderAccess::fence();
 647     ParkEvent * const slp = thread->_SleepEvent ;
 648     if (slp != NULL) slp->unpark() ;
 649   }
 650 
 651   // For JSR166. Unpark even if interrupt status already was set
 652   if (thread->is_Java_thread())
 653     ((JavaThread*)thread)->parker()->unpark();
 654 
 655   ParkEvent * ev = thread->_ParkEvent ;
 656   if (ev != NULL) ev->unpark() ;
 657 }
 658 
 659 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
 660   debug_only(Thread::check_for_dangling_thread_pointer(thread);)
 661 
 662   OSThread* osthread = thread->osthread();
 663 
 664   bool interrupted = osthread->interrupted();
 665 
 666   // NOTE that since there is no "lock" around the interrupt and
 667   // is_interrupted operations, there is the possibility that the
 668   // interrupted flag (in osThread) will be "false" but that the
 669   // low-level events will be in the signaled state. This is
 670   // intentional. The effect of this is that Object.wait() and
 671   // LockSupport.park() will appear to have a spurious wakeup, which
 672   // is allowed and not harmful, and the possibility is so rare that
 673   // it is not worth the added complexity to add yet another lock.
 674   // For the sleep event an explicit reset is performed on entry
 675   // to os::sleep, so there is no early return. It has also been
 676   // recommended not to put the interrupted flag into the "event"
 677   // structure because it hides the issue.
 678   if (interrupted && clear_interrupted) {
 679     osthread->set_interrupted(false);
 680     // consider thread->_SleepEvent->reset() ... optional optimization
 681   }
 682 
 683   return interrupted;
 684 }
 685 
 686 
 687 
 688 static const struct {
 689   int sig; const char* name;
 690 }
 691  g_signal_info[] =
 692   {
 693   {  SIGABRT,     "SIGABRT" },
 694 #ifdef SIGAIO
 695   {  SIGAIO,      "SIGAIO" },
 696 #endif
 697   {  SIGALRM,     "SIGALRM" },
 698 #ifdef SIGALRM1
 699   {  SIGALRM1,    "SIGALRM1" },
 700 #endif
 701   {  SIGBUS,      "SIGBUS" },
 702 #ifdef SIGCANCEL
 703   {  SIGCANCEL,   "SIGCANCEL" },
 704 #endif
 705   {  SIGCHLD,     "SIGCHLD" },
 706 #ifdef SIGCLD
 707   {  SIGCLD,      "SIGCLD" },
 708 #endif
 709   {  SIGCONT,     "SIGCONT" },
 710 #ifdef SIGCPUFAIL
 711   {  SIGCPUFAIL,  "SIGCPUFAIL" },
 712 #endif
 713 #ifdef SIGDANGER
 714   {  SIGDANGER,   "SIGDANGER" },
 715 #endif
 716 #ifdef SIGDIL
 717   {  SIGDIL,      "SIGDIL" },
 718 #endif
 719 #ifdef SIGEMT
 720   {  SIGEMT,      "SIGEMT" },
 721 #endif
 722   {  SIGFPE,      "SIGFPE" },
 723 #ifdef SIGFREEZE
 724   {  SIGFREEZE,   "SIGFREEZE" },
 725 #endif
 726 #ifdef SIGGFAULT
 727   {  SIGGFAULT,   "SIGGFAULT" },
 728 #endif
 729 #ifdef SIGGRANT
 730   {  SIGGRANT,    "SIGGRANT" },
 731 #endif
 732   {  SIGHUP,      "SIGHUP" },
 733   {  SIGILL,      "SIGILL" },
 734   {  SIGINT,      "SIGINT" },
 735 #ifdef SIGIO
 736   {  SIGIO,       "SIGIO" },
 737 #endif
 738 #ifdef SIGIOINT
 739   {  SIGIOINT,    "SIGIOINT" },
 740 #endif
 741 #ifdef SIGIOT
 742 // SIGIOT is there for BSD compatibility, but on most Unices just a
 743 // synonym for SIGABRT. The result should be "SIGABRT", not
 744 // "SIGIOT".
 745 #if (SIGIOT != SIGABRT )
 746   {  SIGIOT,      "SIGIOT" },
 747 #endif
 748 #endif
 749 #ifdef SIGKAP
 750   {  SIGKAP,      "SIGKAP" },
 751 #endif
 752   {  SIGKILL,     "SIGKILL" },
 753 #ifdef SIGLOST
 754   {  SIGLOST,     "SIGLOST" },
 755 #endif
 756 #ifdef SIGLWP
 757   {  SIGLWP,      "SIGLWP" },
 758 #endif
 759 #ifdef SIGLWPTIMER
 760   {  SIGLWPTIMER, "SIGLWPTIMER" },
 761 #endif
 762 #ifdef SIGMIGRATE
 763   {  SIGMIGRATE,  "SIGMIGRATE" },
 764 #endif
 765 #ifdef SIGMSG
 766   {  SIGMSG,      "SIGMSG" },
 767 #endif
 768   {  SIGPIPE,     "SIGPIPE" },
 769 #ifdef SIGPOLL
 770   {  SIGPOLL,     "SIGPOLL" },
 771 #endif
 772 #ifdef SIGPRE
 773   {  SIGPRE,      "SIGPRE" },
 774 #endif
 775   {  SIGPROF,     "SIGPROF" },
 776 #ifdef SIGPTY
 777   {  SIGPTY,      "SIGPTY" },
 778 #endif
 779 #ifdef SIGPWR
 780   {  SIGPWR,      "SIGPWR" },
 781 #endif
 782   {  SIGQUIT,     "SIGQUIT" },
 783 #ifdef SIGRECONFIG
 784   {  SIGRECONFIG, "SIGRECONFIG" },
 785 #endif
 786 #ifdef SIGRECOVERY
 787   {  SIGRECOVERY, "SIGRECOVERY" },
 788 #endif
 789 #ifdef SIGRESERVE
 790   {  SIGRESERVE,  "SIGRESERVE" },
 791 #endif
 792 #ifdef SIGRETRACT
 793   {  SIGRETRACT,  "SIGRETRACT" },
 794 #endif
 795 #ifdef SIGSAK
 796   {  SIGSAK,      "SIGSAK" },
 797 #endif
 798   {  SIGSEGV,     "SIGSEGV" },
 799 #ifdef SIGSOUND
 800   {  SIGSOUND,    "SIGSOUND" },
 801 #endif
 802 #ifdef SIGSTKFLT
 803   {  SIGSTKFLT,    "SIGSTKFLT" },
 804 #endif
 805   {  SIGSTOP,     "SIGSTOP" },
 806   {  SIGSYS,      "SIGSYS" },
 807 #ifdef SIGSYSERROR
 808   {  SIGSYSERROR, "SIGSYSERROR" },
 809 #endif
 810 #ifdef SIGTALRM
 811   {  SIGTALRM,    "SIGTALRM" },
 812 #endif
 813   {  SIGTERM,     "SIGTERM" },
 814 #ifdef SIGTHAW
 815   {  SIGTHAW,     "SIGTHAW" },
 816 #endif
 817   {  SIGTRAP,     "SIGTRAP" },
 818 #ifdef SIGTSTP
 819   {  SIGTSTP,     "SIGTSTP" },
 820 #endif
 821   {  SIGTTIN,     "SIGTTIN" },
 822   {  SIGTTOU,     "SIGTTOU" },
 823 #ifdef SIGURG
 824   {  SIGURG,      "SIGURG" },
 825 #endif
 826   {  SIGUSR1,     "SIGUSR1" },
 827   {  SIGUSR2,     "SIGUSR2" },
 828 #ifdef SIGVIRT
 829   {  SIGVIRT,     "SIGVIRT" },
 830 #endif
 831   {  SIGVTALRM,   "SIGVTALRM" },
 832 #ifdef SIGWAITING
 833   {  SIGWAITING,  "SIGWAITING" },
 834 #endif
 835 #ifdef SIGWINCH
 836   {  SIGWINCH,    "SIGWINCH" },
 837 #endif
 838 #ifdef SIGWINDOW
 839   {  SIGWINDOW,   "SIGWINDOW" },
 840 #endif
 841   {  SIGXCPU,     "SIGXCPU" },
 842   {  SIGXFSZ,     "SIGXFSZ" },
 843 #ifdef SIGXRES
 844   {  SIGXRES,     "SIGXRES" },
 845 #endif
 846   { -1, NULL }
 847 };
 848 
 849 // Returned string is a constant. For unknown signals "UNKNOWN" is returned.
 850 const char* os::Posix::get_signal_name(int sig, char* out, size_t outlen) {
 851 
 852   const char* ret = NULL;
 853 
 854 #ifdef SIGRTMIN
 855   if (sig >= SIGRTMIN && sig <= SIGRTMAX) {
 856     if (sig == SIGRTMIN) {
 857       ret = "SIGRTMIN";
 858     } else if (sig == SIGRTMAX) {
 859       ret = "SIGRTMAX";
 860     } else {
 861       jio_snprintf(out, outlen, "SIGRTMIN+%d", sig - SIGRTMIN);
 862       return out;
 863     }
 864   }
 865 #endif
 866 
 867   if (sig > 0) {
 868     for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
 869       if (g_signal_info[idx].sig == sig) {
 870         ret = g_signal_info[idx].name;
 871         break;
 872       }
 873     }
 874   }
 875 
 876   if (!ret) {
 877     if (!is_valid_signal(sig)) {
 878       ret = "INVALID";
 879     } else {
 880       ret = "UNKNOWN";
 881     }
 882   }
 883 
 884   if (out && outlen > 0) {
 885     strncpy(out, ret, outlen);
 886     out[outlen - 1] = '\0';
 887   }
 888   return out;
 889 }
 890 
 891 int os::Posix::get_signal_number(const char* signal_name) {
 892   char tmp[30];
 893   const char* s = signal_name;
 894   if (s[0] != 'S' || s[1] != 'I' || s[2] != 'G') {
 895     jio_snprintf(tmp, sizeof(tmp), "SIG%s", signal_name);
 896     s = tmp;
 897   }
 898   for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
 899     if (strcmp(g_signal_info[idx].name, s) == 0) {
 900       return g_signal_info[idx].sig;
 901     }
 902   }
 903   return -1;
 904 }
 905 
 906 int os::get_signal_number(const char* signal_name) {
 907   return os::Posix::get_signal_number(signal_name);
 908 }
 909 
 910 // Returns true if signal number is valid.
 911 bool os::Posix::is_valid_signal(int sig) {
 912   // MacOS not really POSIX compliant: sigaddset does not return
 913   // an error for invalid signal numbers. However, MacOS does not
 914   // support real time signals and simply seems to have just 33
 915   // signals with no holes in the signal range.
 916 #ifdef __APPLE__
 917   return sig >= 1 && sig < NSIG;
 918 #else
 919   // Use sigaddset to check for signal validity.
 920   sigset_t set;
 921   sigemptyset(&set);
 922   if (sigaddset(&set, sig) == -1 && errno == EINVAL) {
 923     return false;
 924   }
 925   return true;
 926 #endif
 927 }
 928 
 929 bool os::Posix::is_sig_ignored(int sig) {
 930   struct sigaction oact;
 931   sigaction(sig, (struct sigaction*)NULL, &oact);
 932   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 933                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 934   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 935     return true;
 936   } else {
 937     return false;
 938   }
 939 }
 940 
 941 // Returns:
 942 // NULL for an invalid signal number
 943 // "SIG<num>" for a valid but unknown signal number
 944 // signal name otherwise.
 945 const char* os::exception_name(int sig, char* buf, size_t size) {
 946   if (!os::Posix::is_valid_signal(sig)) {
 947     return NULL;
 948   }
 949   const char* const name = os::Posix::get_signal_name(sig, buf, size);
 950   if (strcmp(name, "UNKNOWN") == 0) {
 951     jio_snprintf(buf, size, "SIG%d", sig);
 952   }
 953   return buf;
 954 }
 955 
 956 #define NUM_IMPORTANT_SIGS 32
 957 // Returns one-line short description of a signal set in a user provided buffer.
 958 const char* os::Posix::describe_signal_set_short(const sigset_t* set, char* buffer, size_t buf_size) {
 959   assert(buf_size == (NUM_IMPORTANT_SIGS + 1), "wrong buffer size");
 960   // Note: for shortness, just print out the first 32. That should
 961   // cover most of the useful ones, apart from realtime signals.
 962   for (int sig = 1; sig <= NUM_IMPORTANT_SIGS; sig++) {
 963     const int rc = sigismember(set, sig);
 964     if (rc == -1 && errno == EINVAL) {
 965       buffer[sig-1] = '?';
 966     } else {
 967       buffer[sig-1] = rc == 0 ? '0' : '1';
 968     }
 969   }
 970   buffer[NUM_IMPORTANT_SIGS] = 0;
 971   return buffer;
 972 }
 973 
 974 // Prints one-line description of a signal set.
 975 void os::Posix::print_signal_set_short(outputStream* st, const sigset_t* set) {
 976   char buf[NUM_IMPORTANT_SIGS + 1];
 977   os::Posix::describe_signal_set_short(set, buf, sizeof(buf));
 978   st->print("%s", buf);
 979 }
 980 
 981 // Writes one-line description of a combination of sigaction.sa_flags into a user
 982 // provided buffer. Returns that buffer.
 983 const char* os::Posix::describe_sa_flags(int flags, char* buffer, size_t size) {
 984   char* p = buffer;
 985   size_t remaining = size;
 986   bool first = true;
 987   int idx = 0;
 988 
 989   assert(buffer, "invalid argument");
 990 
 991   if (size == 0) {
 992     return buffer;
 993   }
 994 
 995   strncpy(buffer, "none", size);
 996 
 997   const struct {
 998     // NB: i is an unsigned int here because SA_RESETHAND is on some
 999     // systems 0x80000000, which is implicitly unsigned.  Assignining
1000     // it to an int field would be an overflow in unsigned-to-signed
1001     // conversion.
1002     unsigned int i;
1003     const char* s;
1004   } flaginfo [] = {
1005     { SA_NOCLDSTOP, "SA_NOCLDSTOP" },
1006     { SA_ONSTACK,   "SA_ONSTACK"   },
1007     { SA_RESETHAND, "SA_RESETHAND" },
1008     { SA_RESTART,   "SA_RESTART"   },
1009     { SA_SIGINFO,   "SA_SIGINFO"   },
1010     { SA_NOCLDWAIT, "SA_NOCLDWAIT" },
1011     { SA_NODEFER,   "SA_NODEFER"   },
1012 #ifdef AIX
1013     { SA_ONSTACK,   "SA_ONSTACK"   },
1014     { SA_OLDSTYLE,  "SA_OLDSTYLE"  },
1015 #endif
1016     { 0, NULL }
1017   };
1018 
1019   for (idx = 0; flaginfo[idx].s && remaining > 1; idx++) {
1020     if (flags & flaginfo[idx].i) {
1021       if (first) {
1022         jio_snprintf(p, remaining, "%s", flaginfo[idx].s);
1023         first = false;
1024       } else {
1025         jio_snprintf(p, remaining, "|%s", flaginfo[idx].s);
1026       }
1027       const size_t len = strlen(p);
1028       p += len;
1029       remaining -= len;
1030     }
1031   }
1032 
1033   buffer[size - 1] = '\0';
1034 
1035   return buffer;
1036 }
1037 
1038 // Prints one-line description of a combination of sigaction.sa_flags.
1039 void os::Posix::print_sa_flags(outputStream* st, int flags) {
1040   char buffer[0x100];
1041   os::Posix::describe_sa_flags(flags, buffer, sizeof(buffer));
1042   st->print("%s", buffer);
1043 }
1044 
1045 // Helper function for os::Posix::print_siginfo_...():
1046 // return a textual description for signal code.
1047 struct enum_sigcode_desc_t {
1048   const char* s_name;
1049   const char* s_desc;
1050 };
1051 
1052 static bool get_signal_code_description(const siginfo_t* si, enum_sigcode_desc_t* out) {
1053 
1054   const struct {
1055     int sig; int code; const char* s_code; const char* s_desc;
1056   } t1 [] = {
1057     { SIGILL,  ILL_ILLOPC,   "ILL_ILLOPC",   "Illegal opcode." },
1058     { SIGILL,  ILL_ILLOPN,   "ILL_ILLOPN",   "Illegal operand." },
1059     { SIGILL,  ILL_ILLADR,   "ILL_ILLADR",   "Illegal addressing mode." },
1060     { SIGILL,  ILL_ILLTRP,   "ILL_ILLTRP",   "Illegal trap." },
1061     { SIGILL,  ILL_PRVOPC,   "ILL_PRVOPC",   "Privileged opcode." },
1062     { SIGILL,  ILL_PRVREG,   "ILL_PRVREG",   "Privileged register." },
1063     { SIGILL,  ILL_COPROC,   "ILL_COPROC",   "Coprocessor error." },
1064     { SIGILL,  ILL_BADSTK,   "ILL_BADSTK",   "Internal stack error." },
1065 #if defined(IA64) && defined(LINUX)
1066     { SIGILL,  ILL_BADIADDR, "ILL_BADIADDR", "Unimplemented instruction address" },
1067     { SIGILL,  ILL_BREAK,    "ILL_BREAK",    "Application Break instruction" },
1068 #endif
1069     { SIGFPE,  FPE_INTDIV,   "FPE_INTDIV",   "Integer divide by zero." },
1070     { SIGFPE,  FPE_INTOVF,   "FPE_INTOVF",   "Integer overflow." },
1071     { SIGFPE,  FPE_FLTDIV,   "FPE_FLTDIV",   "Floating-point divide by zero." },
1072     { SIGFPE,  FPE_FLTOVF,   "FPE_FLTOVF",   "Floating-point overflow." },
1073     { SIGFPE,  FPE_FLTUND,   "FPE_FLTUND",   "Floating-point underflow." },
1074     { SIGFPE,  FPE_FLTRES,   "FPE_FLTRES",   "Floating-point inexact result." },
1075     { SIGFPE,  FPE_FLTINV,   "FPE_FLTINV",   "Invalid floating-point operation." },
1076     { SIGFPE,  FPE_FLTSUB,   "FPE_FLTSUB",   "Subscript out of range." },
1077     { SIGSEGV, SEGV_MAPERR,  "SEGV_MAPERR",  "Address not mapped to object." },
1078     { SIGSEGV, SEGV_ACCERR,  "SEGV_ACCERR",  "Invalid permissions for mapped object." },
1079 #ifdef AIX
1080     // no explanation found what keyerr would be
1081     { SIGSEGV, SEGV_KEYERR,  "SEGV_KEYERR",  "key error" },
1082 #endif
1083 #if defined(IA64) && !defined(AIX)
1084     { SIGSEGV, SEGV_PSTKOVF, "SEGV_PSTKOVF", "Paragraph stack overflow" },
1085 #endif
1086 #if defined(__sparc) && defined(SOLARIS)
1087 // define Solaris Sparc M7 ADI SEGV signals
1088 #if !defined(SEGV_ACCADI)
1089 #define SEGV_ACCADI 3
1090 #endif
1091     { SIGSEGV, SEGV_ACCADI,  "SEGV_ACCADI",  "ADI not enabled for mapped object." },
1092 #if !defined(SEGV_ACCDERR)
1093 #define SEGV_ACCDERR 4
1094 #endif
1095     { SIGSEGV, SEGV_ACCDERR, "SEGV_ACCDERR", "ADI disrupting exception." },
1096 #if !defined(SEGV_ACCPERR)
1097 #define SEGV_ACCPERR 5
1098 #endif
1099     { SIGSEGV, SEGV_ACCPERR, "SEGV_ACCPERR", "ADI precise exception." },
1100 #endif // defined(__sparc) && defined(SOLARIS)
1101     { SIGBUS,  BUS_ADRALN,   "BUS_ADRALN",   "Invalid address alignment." },
1102     { SIGBUS,  BUS_ADRERR,   "BUS_ADRERR",   "Nonexistent physical address." },
1103     { SIGBUS,  BUS_OBJERR,   "BUS_OBJERR",   "Object-specific hardware error." },
1104     { SIGTRAP, TRAP_BRKPT,   "TRAP_BRKPT",   "Process breakpoint." },
1105     { SIGTRAP, TRAP_TRACE,   "TRAP_TRACE",   "Process trace trap." },
1106     { SIGCHLD, CLD_EXITED,   "CLD_EXITED",   "Child has exited." },
1107     { SIGCHLD, CLD_KILLED,   "CLD_KILLED",   "Child has terminated abnormally and did not create a core file." },
1108     { SIGCHLD, CLD_DUMPED,   "CLD_DUMPED",   "Child has terminated abnormally and created a core file." },
1109     { SIGCHLD, CLD_TRAPPED,  "CLD_TRAPPED",  "Traced child has trapped." },
1110     { SIGCHLD, CLD_STOPPED,  "CLD_STOPPED",  "Child has stopped." },
1111     { SIGCHLD, CLD_CONTINUED,"CLD_CONTINUED","Stopped child has continued." },
1112 #ifdef SIGPOLL
1113     { SIGPOLL, POLL_OUT,     "POLL_OUT",     "Output buffers available." },
1114     { SIGPOLL, POLL_MSG,     "POLL_MSG",     "Input message available." },
1115     { SIGPOLL, POLL_ERR,     "POLL_ERR",     "I/O error." },
1116     { SIGPOLL, POLL_PRI,     "POLL_PRI",     "High priority input available." },
1117     { SIGPOLL, POLL_HUP,     "POLL_HUP",     "Device disconnected. [Option End]" },
1118 #endif
1119     { -1, -1, NULL, NULL }
1120   };
1121 
1122   // Codes valid in any signal context.
1123   const struct {
1124     int code; const char* s_code; const char* s_desc;
1125   } t2 [] = {
1126     { SI_USER,      "SI_USER",     "Signal sent by kill()." },
1127     { SI_QUEUE,     "SI_QUEUE",    "Signal sent by the sigqueue()." },
1128     { SI_TIMER,     "SI_TIMER",    "Signal generated by expiration of a timer set by timer_settime()." },
1129     { SI_ASYNCIO,   "SI_ASYNCIO",  "Signal generated by completion of an asynchronous I/O request." },
1130     { SI_MESGQ,     "SI_MESGQ",    "Signal generated by arrival of a message on an empty message queue." },
1131     // Linux specific
1132 #ifdef SI_TKILL
1133     { SI_TKILL,     "SI_TKILL",    "Signal sent by tkill (pthread_kill)" },
1134 #endif
1135 #ifdef SI_DETHREAD
1136     { SI_DETHREAD,  "SI_DETHREAD", "Signal sent by execve() killing subsidiary threads" },
1137 #endif
1138 #ifdef SI_KERNEL
1139     { SI_KERNEL,    "SI_KERNEL",   "Signal sent by kernel." },
1140 #endif
1141 #ifdef SI_SIGIO
1142     { SI_SIGIO,     "SI_SIGIO",    "Signal sent by queued SIGIO" },
1143 #endif
1144 
1145 #ifdef AIX
1146     { SI_UNDEFINED, "SI_UNDEFINED","siginfo contains partial information" },
1147     { SI_EMPTY,     "SI_EMPTY",    "siginfo contains no useful information" },
1148 #endif
1149 
1150 #ifdef __sun
1151     { SI_NOINFO,    "SI_NOINFO",   "No signal information" },
1152     { SI_RCTL,      "SI_RCTL",     "kernel generated signal via rctl action" },
1153     { SI_LWP,       "SI_LWP",      "Signal sent via lwp_kill" },
1154 #endif
1155 
1156     { -1, NULL, NULL }
1157   };
1158 
1159   const char* s_code = NULL;
1160   const char* s_desc = NULL;
1161 
1162   for (int i = 0; t1[i].sig != -1; i ++) {
1163     if (t1[i].sig == si->si_signo && t1[i].code == si->si_code) {
1164       s_code = t1[i].s_code;
1165       s_desc = t1[i].s_desc;
1166       break;
1167     }
1168   }
1169 
1170   if (s_code == NULL) {
1171     for (int i = 0; t2[i].s_code != NULL; i ++) {
1172       if (t2[i].code == si->si_code) {
1173         s_code = t2[i].s_code;
1174         s_desc = t2[i].s_desc;
1175       }
1176     }
1177   }
1178 
1179   if (s_code == NULL) {
1180     out->s_name = "unknown";
1181     out->s_desc = "unknown";
1182     return false;
1183   }
1184 
1185   out->s_name = s_code;
1186   out->s_desc = s_desc;
1187 
1188   return true;
1189 }
1190 
1191 void os::print_siginfo(outputStream* os, const void* si0) {
1192 
1193   const siginfo_t* const si = (const siginfo_t*) si0;
1194 
1195   char buf[20];
1196   os->print("siginfo:");
1197 
1198   if (!si) {
1199     os->print(" <null>");
1200     return;
1201   }
1202 
1203   const int sig = si->si_signo;
1204 
1205   os->print(" si_signo: %d (%s)", sig, os::Posix::get_signal_name(sig, buf, sizeof(buf)));
1206 
1207   enum_sigcode_desc_t ed;
1208   get_signal_code_description(si, &ed);
1209   os->print(", si_code: %d (%s)", si->si_code, ed.s_name);
1210 
1211   if (si->si_errno) {
1212     os->print(", si_errno: %d", si->si_errno);
1213   }
1214 
1215   // Output additional information depending on the signal code.
1216 
1217   // Note: Many implementations lump si_addr, si_pid, si_uid etc. together as unions,
1218   // so it depends on the context which member to use. For synchronous error signals,
1219   // we print si_addr, unless the signal was sent by another process or thread, in
1220   // which case we print out pid or tid of the sender.
1221   if (si->si_code == SI_USER || si->si_code == SI_QUEUE) {
1222     const pid_t pid = si->si_pid;
1223     os->print(", si_pid: %ld", (long) pid);
1224     if (IS_VALID_PID(pid)) {
1225       const pid_t me = getpid();
1226       if (me == pid) {
1227         os->print(" (current process)");
1228       }
1229     } else {
1230       os->print(" (invalid)");
1231     }
1232     os->print(", si_uid: %ld", (long) si->si_uid);
1233     if (sig == SIGCHLD) {
1234       os->print(", si_status: %d", si->si_status);
1235     }
1236   } else if (sig == SIGSEGV || sig == SIGBUS || sig == SIGILL ||
1237              sig == SIGTRAP || sig == SIGFPE) {
1238     os->print(", si_addr: " PTR_FORMAT, p2i(si->si_addr));
1239 #ifdef SIGPOLL
1240   } else if (sig == SIGPOLL) {
1241     os->print(", si_band: %ld", si->si_band);
1242 #endif
1243   }
1244 
1245 }
1246 
1247 int os::Posix::unblock_thread_signal_mask(const sigset_t *set) {
1248   return pthread_sigmask(SIG_UNBLOCK, set, NULL);
1249 }
1250 
1251 address os::Posix::ucontext_get_pc(const ucontext_t* ctx) {
1252 #if defined(AIX)
1253    return Aix::ucontext_get_pc(ctx);
1254 #elif defined(BSD)
1255    return Bsd::ucontext_get_pc(ctx);
1256 #elif defined(LINUX)
1257    return Linux::ucontext_get_pc(ctx);
1258 #elif defined(SOLARIS)
1259    return Solaris::ucontext_get_pc(ctx);
1260 #else
1261    VMError::report_and_die("unimplemented ucontext_get_pc");
1262 #endif
1263 }
1264 
1265 void os::Posix::ucontext_set_pc(ucontext_t* ctx, address pc) {
1266 #if defined(AIX)
1267    Aix::ucontext_set_pc(ctx, pc);
1268 #elif defined(BSD)
1269    Bsd::ucontext_set_pc(ctx, pc);
1270 #elif defined(LINUX)
1271    Linux::ucontext_set_pc(ctx, pc);
1272 #elif defined(SOLARIS)
1273    Solaris::ucontext_set_pc(ctx, pc);
1274 #else
1275    VMError::report_and_die("unimplemented ucontext_get_pc");
1276 #endif
1277 }
1278 
1279 char* os::Posix::describe_pthread_attr(char* buf, size_t buflen, const pthread_attr_t* attr) {
1280   size_t stack_size = 0;
1281   size_t guard_size = 0;
1282   int detachstate = 0;
1283   pthread_attr_getstacksize(attr, &stack_size);
1284   pthread_attr_getguardsize(attr, &guard_size);
1285   // Work around linux NPTL implementation error, see also os::create_thread() in os_linux.cpp.
1286   LINUX_ONLY(stack_size -= guard_size);
1287   pthread_attr_getdetachstate(attr, &detachstate);
1288   jio_snprintf(buf, buflen, "stacksize: " SIZE_FORMAT "k, guardsize: " SIZE_FORMAT "k, %s",
1289     stack_size / 1024, guard_size / 1024,
1290     (detachstate == PTHREAD_CREATE_DETACHED ? "detached" : "joinable"));
1291   return buf;
1292 }
1293 
1294 char* os::Posix::realpath(const char* filename, char* outbuf, size_t outbuflen) {
1295 
1296   if (filename == NULL || outbuf == NULL || outbuflen < 1) {
1297     assert(false, "os::Posix::realpath: invalid arguments.");
1298     errno = EINVAL;
1299     return NULL;
1300   }
1301 
1302   char* result = NULL;
1303 
1304   // This assumes platform realpath() is implemented according to POSIX.1-2008.
1305   // POSIX.1-2008 allows to specify NULL for the output buffer, in which case
1306   // output buffer is dynamically allocated and must be ::free()'d by the caller.
1307   char* p = ::realpath(filename, NULL);
1308   if (p != NULL) {
1309     if (strlen(p) < outbuflen) {
1310       strcpy(outbuf, p);
1311       result = outbuf;
1312     } else {
1313       errno = ENAMETOOLONG;
1314     }
1315     ::free(p); // *not* os::free
1316   } else {
1317     // Fallback for platforms struggling with modern Posix standards (AIX 5.3, 6.1). If realpath
1318     // returns EINVAL, this may indicate that realpath is not POSIX.1-2008 compatible and
1319     // that it complains about the NULL we handed down as user buffer.
1320     // In this case, use the user provided buffer but at least check whether realpath caused
1321     // a memory overwrite.
1322     if (errno == EINVAL) {
1323       outbuf[outbuflen - 1] = '\0';
1324       p = ::realpath(filename, outbuf);
1325       if (p != NULL) {
1326         guarantee(outbuf[outbuflen - 1] == '\0', "realpath buffer overwrite detected.");
1327         result = p;
1328       }
1329     }
1330   }
1331   return result;
1332 
1333 }
1334 
1335 
1336 // Check minimum allowable stack sizes for thread creation and to initialize
1337 // the java system classes, including StackOverflowError - depends on page
1338 // size.
1339 // The space needed for frames during startup is platform dependent. It
1340 // depends on word size, platform calling conventions, C frame layout and
1341 // interpreter/C1/C2 design decisions. Therefore this is given in a
1342 // platform (os/cpu) dependent constant.
1343 // To this, space for guard mechanisms is added, which depends on the
1344 // page size which again depends on the concrete system the VM is running
1345 // on. Space for libc guard pages is not included in this size.
1346 jint os::Posix::set_minimum_stack_sizes() {
1347   size_t os_min_stack_allowed = SOLARIS_ONLY(thr_min_stack()) NOT_SOLARIS(PTHREAD_STACK_MIN);
1348 
1349   _java_thread_min_stack_allowed = _java_thread_min_stack_allowed +
1350                                    JavaThread::stack_guard_zone_size() +
1351                                    JavaThread::stack_shadow_zone_size();
1352 
1353   _java_thread_min_stack_allowed = align_up(_java_thread_min_stack_allowed, vm_page_size());
1354   _java_thread_min_stack_allowed = MAX2(_java_thread_min_stack_allowed, os_min_stack_allowed);
1355 
1356   size_t stack_size_in_bytes = ThreadStackSize * K;
1357   if (stack_size_in_bytes != 0 &&
1358       stack_size_in_bytes < _java_thread_min_stack_allowed) {
1359     // The '-Xss' and '-XX:ThreadStackSize=N' options both set
1360     // ThreadStackSize so we go with "Java thread stack size" instead
1361     // of "ThreadStackSize" to be more friendly.
1362     tty->print_cr("\nThe Java thread stack size specified is too small. "
1363                   "Specify at least " SIZE_FORMAT "k",
1364                   _java_thread_min_stack_allowed / K);
1365     return JNI_ERR;
1366   }
1367 
1368   // Make the stack size a multiple of the page size so that
1369   // the yellow/red zones can be guarded.
1370   JavaThread::set_stack_size_at_create(align_up(stack_size_in_bytes, vm_page_size()));
1371 
1372   // Reminder: a compiler thread is a Java thread.
1373   _compiler_thread_min_stack_allowed = _compiler_thread_min_stack_allowed +
1374                                        JavaThread::stack_guard_zone_size() +
1375                                        JavaThread::stack_shadow_zone_size();
1376 
1377   _compiler_thread_min_stack_allowed = align_up(_compiler_thread_min_stack_allowed, vm_page_size());
1378   _compiler_thread_min_stack_allowed = MAX2(_compiler_thread_min_stack_allowed, os_min_stack_allowed);
1379 
1380   stack_size_in_bytes = CompilerThreadStackSize * K;
1381   if (stack_size_in_bytes != 0 &&
1382       stack_size_in_bytes < _compiler_thread_min_stack_allowed) {
1383     tty->print_cr("\nThe CompilerThreadStackSize specified is too small. "
1384                   "Specify at least " SIZE_FORMAT "k",
1385                   _compiler_thread_min_stack_allowed / K);
1386     return JNI_ERR;
1387   }
1388 
1389   _vm_internal_thread_min_stack_allowed = align_up(_vm_internal_thread_min_stack_allowed, vm_page_size());
1390   _vm_internal_thread_min_stack_allowed = MAX2(_vm_internal_thread_min_stack_allowed, os_min_stack_allowed);
1391 
1392   stack_size_in_bytes = VMThreadStackSize * K;
1393   if (stack_size_in_bytes != 0 &&
1394       stack_size_in_bytes < _vm_internal_thread_min_stack_allowed) {
1395     tty->print_cr("\nThe VMThreadStackSize specified is too small. "
1396                   "Specify at least " SIZE_FORMAT "k",
1397                   _vm_internal_thread_min_stack_allowed / K);
1398     return JNI_ERR;
1399   }
1400   return JNI_OK;
1401 }
1402 
1403 // Called when creating the thread.  The minimum stack sizes have already been calculated
1404 size_t os::Posix::get_initial_stack_size(ThreadType thr_type, size_t req_stack_size) {
1405   size_t stack_size;
1406   if (req_stack_size == 0) {
1407     stack_size = default_stack_size(thr_type);
1408   } else {
1409     stack_size = req_stack_size;
1410   }
1411 
1412   switch (thr_type) {
1413   case os::java_thread:
1414     // Java threads use ThreadStackSize which default value can be
1415     // changed with the flag -Xss
1416     if (req_stack_size == 0 && JavaThread::stack_size_at_create() > 0) {
1417       // no requested size and we have a more specific default value
1418       stack_size = JavaThread::stack_size_at_create();
1419     }
1420     stack_size = MAX2(stack_size,
1421                       _java_thread_min_stack_allowed);
1422     break;
1423   case os::compiler_thread:
1424     if (req_stack_size == 0 && CompilerThreadStackSize > 0) {
1425       // no requested size and we have a more specific default value
1426       stack_size = (size_t)(CompilerThreadStackSize * K);
1427     }
1428     stack_size = MAX2(stack_size,
1429                       _compiler_thread_min_stack_allowed);
1430     break;
1431   case os::vm_thread:
1432   case os::pgc_thread:
1433   case os::cgc_thread:
1434   case os::watcher_thread:
1435   default:  // presume the unknown thr_type is a VM internal
1436     if (req_stack_size == 0 && VMThreadStackSize > 0) {
1437       // no requested size and we have a more specific default value
1438       stack_size = (size_t)(VMThreadStackSize * K);
1439     }
1440 
1441     stack_size = MAX2(stack_size,
1442                       _vm_internal_thread_min_stack_allowed);
1443     break;
1444   }
1445 
1446   // pthread_attr_setstacksize() may require that the size be rounded up to the OS page size.
1447   // Be careful not to round up to 0. Align down in that case.
1448   if (stack_size <= SIZE_MAX - vm_page_size()) {
1449     stack_size = align_up(stack_size, vm_page_size());
1450   } else {
1451     stack_size = align_down(stack_size, vm_page_size());
1452   }
1453 
1454   return stack_size;
1455 }
1456 
1457 Thread* os::ThreadCrashProtection::_protected_thread = NULL;
1458 os::ThreadCrashProtection* os::ThreadCrashProtection::_crash_protection = NULL;
1459 volatile intptr_t os::ThreadCrashProtection::_crash_mux = 0;
1460 
1461 os::ThreadCrashProtection::ThreadCrashProtection() {
1462 }
1463 
1464 /*
1465  * See the caveats for this class in os_posix.hpp
1466  * Protects the callback call so that SIGSEGV / SIGBUS jumps back into this
1467  * method and returns false. If none of the signals are raised, returns true.
1468  * The callback is supposed to provide the method that should be protected.
1469  */
1470 bool os::ThreadCrashProtection::call(os::CrashProtectionCallback& cb) {
1471   sigset_t saved_sig_mask;
1472 
1473   Thread::muxAcquire(&_crash_mux, "CrashProtection");
1474 
1475   _protected_thread = Thread::current_or_null();
1476   assert(_protected_thread != NULL, "Cannot crash protect a NULL thread");
1477 
1478   // we cannot rely on sigsetjmp/siglongjmp to save/restore the signal mask
1479   // since on at least some systems (OS X) siglongjmp will restore the mask
1480   // for the process, not the thread
1481   pthread_sigmask(0, NULL, &saved_sig_mask);
1482   if (sigsetjmp(_jmpbuf, 0) == 0) {
1483     // make sure we can see in the signal handler that we have crash protection
1484     // installed
1485     _crash_protection = this;
1486     cb.call();
1487     // and clear the crash protection
1488     _crash_protection = NULL;
1489     _protected_thread = NULL;
1490     Thread::muxRelease(&_crash_mux);
1491     return true;
1492   }
1493   // this happens when we siglongjmp() back
1494   pthread_sigmask(SIG_SETMASK, &saved_sig_mask, NULL);
1495   _crash_protection = NULL;
1496   _protected_thread = NULL;
1497   Thread::muxRelease(&_crash_mux);
1498   return false;
1499 }
1500 
1501 void os::ThreadCrashProtection::restore() {
1502   assert(_crash_protection != NULL, "must have crash protection");
1503   siglongjmp(_jmpbuf, 1);
1504 }
1505 
1506 void os::ThreadCrashProtection::check_crash_protection(int sig,
1507     Thread* thread) {
1508 
1509   if (thread != NULL &&
1510       thread == _protected_thread &&
1511       _crash_protection != NULL) {
1512 
1513     if (sig == SIGSEGV || sig == SIGBUS) {
1514       _crash_protection->restore();
1515     }
1516   }
1517 }
1518 
1519 
1520 // Shared pthread_mutex/cond based PlatformEvent implementation.
1521 // Not currently usable by Solaris.
1522 
1523 #ifndef SOLARIS
1524 
1525 // Shared condattr object for use with relative timed-waits. Will be associated
1526 // with CLOCK_MONOTONIC if available to avoid issues with time-of-day changes,
1527 // but otherwise whatever default is used by the platform - generally the
1528 // time-of-day clock.
1529 static pthread_condattr_t _condAttr[1];
1530 
1531 // Shared mutexattr to explicitly set the type to PTHREAD_MUTEX_NORMAL as not
1532 // all systems (e.g. FreeBSD) map the default to "normal".
1533 static pthread_mutexattr_t _mutexAttr[1];
1534 
1535 // common basic initialization that is always supported
1536 static void pthread_init_common(void) {
1537   int status;
1538   if ((status = pthread_condattr_init(_condAttr)) != 0) {
1539     fatal("pthread_condattr_init: %s", os::strerror(status));
1540   }
1541   if ((status = pthread_mutexattr_init(_mutexAttr)) != 0) {
1542     fatal("pthread_mutexattr_init: %s", os::strerror(status));
1543   }
1544   if ((status = pthread_mutexattr_settype(_mutexAttr, PTHREAD_MUTEX_NORMAL)) != 0) {
1545     fatal("pthread_mutexattr_settype: %s", os::strerror(status));
1546   }
1547 }
1548 
1549 // Not all POSIX types and API's are available on all notionally "posix"
1550 // platforms. If we have build-time support then we will check for actual
1551 // runtime support via dlopen/dlsym lookup. This allows for running on an
1552 // older OS version compared to the build platform. But if there is no
1553 // build time support then there cannot be any runtime support as we do not
1554 // know what the runtime types would be (for example clockid_t might be an
1555 // int or int64_t).
1556 //
1557 #ifdef SUPPORTS_CLOCK_MONOTONIC
1558 
1559 // This means we have clockid_t, clock_gettime et al and CLOCK_MONOTONIC
1560 
1561 static int (*_clock_gettime)(clockid_t, struct timespec *);
1562 static int (*_pthread_condattr_setclock)(pthread_condattr_t *, clockid_t);
1563 
1564 static bool _use_clock_monotonic_condattr;
1565 
1566 // Determine what POSIX API's are present and do appropriate
1567 // configuration.
1568 void os::Posix::init(void) {
1569 
1570   // NOTE: no logging available when this is called. Put logging
1571   // statements in init_2().
1572 
1573   // Copied from os::Linux::clock_init(). The duplication is temporary.
1574 
1575   // 1. Check for CLOCK_MONOTONIC support.
1576 
1577   void* handle = NULL;
1578 
1579   // For linux we need librt, for other OS we can find
1580   // this function in regular libc.
1581 #ifdef NEEDS_LIBRT
1582   // We do dlopen's in this particular order due to bug in linux
1583   // dynamic loader (see 6348968) leading to crash on exit.
1584   handle = dlopen("librt.so.1", RTLD_LAZY);
1585   if (handle == NULL) {
1586     handle = dlopen("librt.so", RTLD_LAZY);
1587   }
1588 #endif
1589 
1590   if (handle == NULL) {
1591     handle = RTLD_DEFAULT;
1592   }
1593 
1594   _clock_gettime = NULL;
1595 
1596   int (*clock_getres_func)(clockid_t, struct timespec*) =
1597     (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1598   int (*clock_gettime_func)(clockid_t, struct timespec*) =
1599     (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1600   if (clock_getres_func != NULL && clock_gettime_func != NULL) {
1601     // We assume that if both clock_gettime and clock_getres support
1602     // CLOCK_MONOTONIC then the OS provides true high-res monotonic clock.
1603     struct timespec res;
1604     struct timespec tp;
1605     if (clock_getres_func(CLOCK_MONOTONIC, &res) == 0 &&
1606         clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1607       // Yes, monotonic clock is supported.
1608       _clock_gettime = clock_gettime_func;
1609     } else {
1610 #ifdef NEEDS_LIBRT
1611       // Close librt if there is no monotonic clock.
1612       if (handle != RTLD_DEFAULT) {
1613         dlclose(handle);
1614       }
1615 #endif
1616     }
1617   }
1618 
1619   // 2. Check for pthread_condattr_setclock support.
1620 
1621   _pthread_condattr_setclock = NULL;
1622 
1623   // libpthread is already loaded.
1624   int (*condattr_setclock_func)(pthread_condattr_t*, clockid_t) =
1625     (int (*)(pthread_condattr_t*, clockid_t))dlsym(RTLD_DEFAULT,
1626                                                    "pthread_condattr_setclock");
1627   if (condattr_setclock_func != NULL) {
1628     _pthread_condattr_setclock = condattr_setclock_func;
1629   }
1630 
1631   // Now do general initialization.
1632 
1633   pthread_init_common();
1634 
1635   int status;
1636   if (_pthread_condattr_setclock != NULL && _clock_gettime != NULL) {
1637     if ((status = _pthread_condattr_setclock(_condAttr, CLOCK_MONOTONIC)) != 0) {
1638       if (status == EINVAL) {
1639         _use_clock_monotonic_condattr = false;
1640         warning("Unable to use monotonic clock with relative timed-waits" \
1641                 " - changes to the time-of-day clock may have adverse affects");
1642       } else {
1643         fatal("pthread_condattr_setclock: %s", os::strerror(status));
1644       }
1645     } else {
1646       _use_clock_monotonic_condattr = true;
1647     }
1648   } else {
1649     _use_clock_monotonic_condattr = false;
1650   }
1651 }
1652 
1653 void os::Posix::init_2(void) {
1654   log_info(os)("Use of CLOCK_MONOTONIC is%s supported",
1655                (_clock_gettime != NULL ? "" : " not"));
1656   log_info(os)("Use of pthread_condattr_setclock is%s supported",
1657                (_pthread_condattr_setclock != NULL ? "" : " not"));
1658   log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with %s",
1659                _use_clock_monotonic_condattr ? "CLOCK_MONOTONIC" : "the default clock");
1660 }
1661 
1662 #else // !SUPPORTS_CLOCK_MONOTONIC
1663 
1664 void os::Posix::init(void) {
1665   pthread_init_common();
1666 }
1667 
1668 void os::Posix::init_2(void) {
1669   log_info(os)("Use of CLOCK_MONOTONIC is not supported");
1670   log_info(os)("Use of pthread_condattr_setclock is not supported");
1671   log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with the default clock");
1672 }
1673 
1674 #endif // SUPPORTS_CLOCK_MONOTONIC
1675 
1676 os::PlatformEvent::PlatformEvent() {
1677   int status = pthread_cond_init(_cond, _condAttr);
1678   assert_status(status == 0, status, "cond_init");
1679   status = pthread_mutex_init(_mutex, _mutexAttr);
1680   assert_status(status == 0, status, "mutex_init");
1681   _event   = 0;
1682   _nParked = 0;
1683 }
1684 
1685 // Utility to convert the given timeout to an absolute timespec
1686 // (based on the appropriate clock) to use with pthread_cond_timewait.
1687 // The clock queried here must be the clock used to manage the
1688 // timeout of the condition variable.
1689 //
1690 // The passed in timeout value is either a relative time in nanoseconds
1691 // or an absolute time in milliseconds. A relative timeout will be
1692 // associated with CLOCK_MONOTONIC if available; otherwise, or if absolute,
1693 // the default time-of-day clock will be used.
1694 
1695 // Given time is a 64-bit value and the time_t used in the timespec is
1696 // sometimes a signed-32-bit value we have to watch for overflow if times
1697 // way in the future are given. Further on Solaris versions
1698 // prior to 10 there is a restriction (see cond_timedwait) that the specified
1699 // number of seconds, in abstime, is less than current_time + 100000000.
1700 // As it will be over 20 years before "now + 100000000" will overflow we can
1701 // ignore overflow and just impose a hard-limit on seconds using the value
1702 // of "now + 100000000". This places a limit on the timeout of about 3.17
1703 // years from "now".
1704 //
1705 #define MAX_SECS 100000000
1706 
1707 // Calculate a new absolute time that is "timeout" nanoseconds from "now".
1708 // "unit" indicates the unit of "now_part_sec" (may be nanos or micros depending
1709 // on which clock is being used).
1710 static void calc_rel_time(timespec* abstime, jlong timeout, jlong now_sec,
1711                           jlong now_part_sec, jlong unit) {
1712   time_t max_secs = now_sec + MAX_SECS;
1713 
1714   jlong seconds = timeout / NANOUNITS;
1715   timeout %= NANOUNITS; // remaining nanos
1716 
1717   if (seconds >= MAX_SECS) {
1718     // More seconds than we can add, so pin to max_secs.
1719     abstime->tv_sec = max_secs;
1720     abstime->tv_nsec = 0;
1721   } else {
1722     abstime->tv_sec = now_sec  + seconds;
1723     long nanos = (now_part_sec * (NANOUNITS / unit)) + timeout;
1724     if (nanos >= NANOUNITS) { // overflow
1725       abstime->tv_sec += 1;
1726       nanos -= NANOUNITS;
1727     }
1728     abstime->tv_nsec = nanos;
1729   }
1730 }
1731 
1732 // Unpack the given deadline in milliseconds since the epoch, into the given timespec.
1733 // The current time in seconds is also passed in to enforce an upper bound as discussed above.
1734 static void unpack_abs_time(timespec* abstime, jlong deadline, jlong now_sec) {
1735   time_t max_secs = now_sec + MAX_SECS;
1736 
1737   jlong seconds = deadline / MILLIUNITS;
1738   jlong millis = deadline % MILLIUNITS;
1739 
1740   if (seconds >= max_secs) {
1741     // Absolute seconds exceeds allowed max, so pin to max_secs.
1742     abstime->tv_sec = max_secs;
1743     abstime->tv_nsec = 0;
1744   } else {
1745     abstime->tv_sec = seconds;
1746     abstime->tv_nsec = millis * (NANOUNITS / MILLIUNITS);
1747   }
1748 }
1749 
1750 static void to_abstime(timespec* abstime, jlong timeout, bool isAbsolute) {
1751   DEBUG_ONLY(int max_secs = MAX_SECS;)
1752 
1753   if (timeout < 0) {
1754     timeout = 0;
1755   }
1756 
1757 #ifdef SUPPORTS_CLOCK_MONOTONIC
1758 
1759   if (_use_clock_monotonic_condattr && !isAbsolute) {
1760     struct timespec now;
1761     int status = _clock_gettime(CLOCK_MONOTONIC, &now);
1762     assert_status(status == 0, status, "clock_gettime");
1763     calc_rel_time(abstime, timeout, now.tv_sec, now.tv_nsec, NANOUNITS);
1764     DEBUG_ONLY(max_secs += now.tv_sec;)
1765   } else {
1766 
1767 #else
1768 
1769   { // Match the block scope.
1770 
1771 #endif // SUPPORTS_CLOCK_MONOTONIC
1772 
1773     // Time-of-day clock is all we can reliably use.
1774     struct timeval now;
1775     int status = gettimeofday(&now, NULL);
1776     assert_status(status == 0, errno, "gettimeofday");
1777     if (isAbsolute) {
1778       unpack_abs_time(abstime, timeout, now.tv_sec);
1779     } else {
1780       calc_rel_time(abstime, timeout, now.tv_sec, now.tv_usec, MICROUNITS);
1781     }
1782     DEBUG_ONLY(max_secs += now.tv_sec;)
1783   }
1784 
1785   assert(abstime->tv_sec >= 0, "tv_sec < 0");
1786   assert(abstime->tv_sec <= max_secs, "tv_sec > max_secs");
1787   assert(abstime->tv_nsec >= 0, "tv_nsec < 0");
1788   assert(abstime->tv_nsec < NANOUNITS, "tv_nsec >= NANOUNITS");
1789 }
1790 
1791 // PlatformEvent
1792 //
1793 // Assumption:
1794 //    Only one parker can exist on an event, which is why we allocate
1795 //    them per-thread. Multiple unparkers can coexist.
1796 //
1797 // _event serves as a restricted-range semaphore.
1798 //   -1 : thread is blocked, i.e. there is a waiter
1799 //    0 : neutral: thread is running or ready,
1800 //        could have been signaled after a wait started
1801 //    1 : signaled - thread is running or ready
1802 //
1803 //    Having three states allows for some detection of bad usage - see
1804 //    comments on unpark().
1805 
1806 void os::PlatformEvent::park() {       // AKA "down()"
1807   // Transitions for _event:
1808   //   -1 => -1 : illegal
1809   //    1 =>  0 : pass - return immediately
1810   //    0 => -1 : block; then set _event to 0 before returning
1811 
1812   // Invariant: Only the thread associated with the PlatformEvent
1813   // may call park().
1814   assert(_nParked == 0, "invariant");
1815 
1816   int v;
1817 
1818   // atomically decrement _event
1819   for (;;) {
1820     v = _event;
1821     if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1822   }
1823   guarantee(v >= 0, "invariant");
1824 
1825   if (v == 0) { // Do this the hard way by blocking ...
1826     int status = pthread_mutex_lock(_mutex);
1827     assert_status(status == 0, status, "mutex_lock");
1828     guarantee(_nParked == 0, "invariant");
1829     ++_nParked;
1830     while (_event < 0) {
1831       // OS-level "spurious wakeups" are ignored
1832       status = pthread_cond_wait(_cond, _mutex);
1833       assert_status(status == 0, status, "cond_wait");
1834     }
1835     --_nParked;
1836 
1837     _event = 0;
1838     status = pthread_mutex_unlock(_mutex);
1839     assert_status(status == 0, status, "mutex_unlock");
1840     // Paranoia to ensure our locked and lock-free paths interact
1841     // correctly with each other.
1842     OrderAccess::fence();
1843   }
1844   guarantee(_event >= 0, "invariant");
1845 }
1846 
1847 int os::PlatformEvent::park(jlong millis) {
1848   // Transitions for _event:
1849   //   -1 => -1 : illegal
1850   //    1 =>  0 : pass - return immediately
1851   //    0 => -1 : block; then set _event to 0 before returning
1852 
1853   // Invariant: Only the thread associated with the Event/PlatformEvent
1854   // may call park().
1855   assert(_nParked == 0, "invariant");
1856 
1857   int v;
1858   // atomically decrement _event
1859   for (;;) {
1860     v = _event;
1861     if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1862   }
1863   guarantee(v >= 0, "invariant");
1864 
1865   if (v == 0) { // Do this the hard way by blocking ...
1866     struct timespec abst;
1867     // We have to watch for overflow when converting millis to nanos,
1868     // but if millis is that large then we will end up limiting to
1869     // MAX_SECS anyway, so just do that here.
1870     if (millis / MILLIUNITS > MAX_SECS) {
1871       millis = jlong(MAX_SECS) * MILLIUNITS;
1872     }
1873     to_abstime(&abst, millis * (NANOUNITS / MILLIUNITS), false);
1874 
1875     int ret = OS_TIMEOUT;
1876     int status = pthread_mutex_lock(_mutex);
1877     assert_status(status == 0, status, "mutex_lock");
1878     guarantee(_nParked == 0, "invariant");
1879     ++_nParked;
1880 
1881     while (_event < 0) {
1882       status = pthread_cond_timedwait(_cond, _mutex, &abst);
1883       assert_status(status == 0 || status == ETIMEDOUT,
1884                     status, "cond_timedwait");
1885       // OS-level "spurious wakeups" are ignored unless the archaic
1886       // FilterSpuriousWakeups is set false. That flag should be obsoleted.
1887       if (!FilterSpuriousWakeups) break;
1888       if (status == ETIMEDOUT) break;
1889     }
1890     --_nParked;
1891 
1892     if (_event >= 0) {
1893       ret = OS_OK;
1894     }
1895 
1896     _event = 0;
1897     status = pthread_mutex_unlock(_mutex);
1898     assert_status(status == 0, status, "mutex_unlock");
1899     // Paranoia to ensure our locked and lock-free paths interact
1900     // correctly with each other.
1901     OrderAccess::fence();
1902     return ret;
1903   }
1904   return OS_OK;
1905 }
1906 
1907 void os::PlatformEvent::unpark() {
1908   // Transitions for _event:
1909   //    0 => 1 : just return
1910   //    1 => 1 : just return
1911   //   -1 => either 0 or 1; must signal target thread
1912   //         That is, we can safely transition _event from -1 to either
1913   //         0 or 1.
1914   // See also: "Semaphores in Plan 9" by Mullender & Cox
1915   //
1916   // Note: Forcing a transition from "-1" to "1" on an unpark() means
1917   // that it will take two back-to-back park() calls for the owning
1918   // thread to block. This has the benefit of forcing a spurious return
1919   // from the first park() call after an unpark() call which will help
1920   // shake out uses of park() and unpark() without checking state conditions
1921   // properly. This spurious return doesn't manifest itself in any user code
1922   // but only in the correctly written condition checking loops of ObjectMonitor,
1923   // Mutex/Monitor, Thread::muxAcquire and os::sleep
1924 
1925   if (Atomic::xchg(1, &_event) >= 0) return;
1926 
1927   int status = pthread_mutex_lock(_mutex);
1928   assert_status(status == 0, status, "mutex_lock");
1929   int anyWaiters = _nParked;
1930   assert(anyWaiters == 0 || anyWaiters == 1, "invariant");
1931   status = pthread_mutex_unlock(_mutex);
1932   assert_status(status == 0, status, "mutex_unlock");
1933 
1934   // Note that we signal() *after* dropping the lock for "immortal" Events.
1935   // This is safe and avoids a common class of futile wakeups.  In rare
1936   // circumstances this can cause a thread to return prematurely from
1937   // cond_{timed}wait() but the spurious wakeup is benign and the victim
1938   // will simply re-test the condition and re-park itself.
1939   // This provides particular benefit if the underlying platform does not
1940   // provide wait morphing.
1941 
1942   if (anyWaiters != 0) {
1943     status = pthread_cond_signal(_cond);
1944     assert_status(status == 0, status, "cond_signal");
1945   }
1946 }
1947 
1948 // JSR166 support
1949 
1950  os::PlatformParker::PlatformParker() {
1951   int status;
1952   status = pthread_cond_init(&_cond[REL_INDEX], _condAttr);
1953   assert_status(status == 0, status, "cond_init rel");
1954   status = pthread_cond_init(&_cond[ABS_INDEX], NULL);
1955   assert_status(status == 0, status, "cond_init abs");
1956   status = pthread_mutex_init(_mutex, _mutexAttr);
1957   assert_status(status == 0, status, "mutex_init");
1958   _cur_index = -1; // mark as unused
1959 }
1960 
1961 // Parker::park decrements count if > 0, else does a condvar wait.  Unpark
1962 // sets count to 1 and signals condvar.  Only one thread ever waits
1963 // on the condvar. Contention seen when trying to park implies that someone
1964 // is unparking you, so don't wait. And spurious returns are fine, so there
1965 // is no need to track notifications.
1966 
1967 void Parker::park(bool isAbsolute, jlong time) {
1968 
1969   // Optional fast-path check:
1970   // Return immediately if a permit is available.
1971   // We depend on Atomic::xchg() having full barrier semantics
1972   // since we are doing a lock-free update to _counter.
1973   if (Atomic::xchg(0, &_counter) > 0) return;
1974 
1975   Thread* thread = Thread::current();
1976   assert(thread->is_Java_thread(), "Must be JavaThread");
1977   JavaThread *jt = (JavaThread *)thread;
1978 
1979   // Optional optimization -- avoid state transitions if there's
1980   // an interrupt pending.
1981   if (Thread::is_interrupted(thread, false)) {
1982     return;
1983   }
1984 
1985   // Next, demultiplex/decode time arguments
1986   struct timespec absTime;
1987   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
1988     return;
1989   }
1990   if (time > 0) {
1991     to_abstime(&absTime, time, isAbsolute);
1992   }
1993 
1994   // Enter safepoint region
1995   // Beware of deadlocks such as 6317397.
1996   // The per-thread Parker:: mutex is a classic leaf-lock.
1997   // In particular a thread must never block on the Threads_lock while
1998   // holding the Parker:: mutex.  If safepoints are pending both the
1999   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
2000   ThreadBlockInVM tbivm(jt);
2001 
2002   // Don't wait if cannot get lock since interference arises from
2003   // unparking. Also re-check interrupt before trying wait.
2004   if (Thread::is_interrupted(thread, false) ||
2005       pthread_mutex_trylock(_mutex) != 0) {
2006     return;
2007   }
2008 
2009   int status;
2010   if (_counter > 0)  { // no wait needed
2011     _counter = 0;
2012     status = pthread_mutex_unlock(_mutex);
2013     assert_status(status == 0, status, "invariant");
2014     // Paranoia to ensure our locked and lock-free paths interact
2015     // correctly with each other and Java-level accesses.
2016     OrderAccess::fence();
2017     return;
2018   }
2019 
2020   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2021   jt->set_suspend_equivalent();
2022   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2023 
2024   assert(_cur_index == -1, "invariant");
2025   if (time == 0) {
2026     _cur_index = REL_INDEX; // arbitrary choice when not timed
2027     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
2028     assert_status(status == 0, status, "cond_timedwait");
2029   }
2030   else {
2031     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
2032     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
2033     assert_status(status == 0 || status == ETIMEDOUT,
2034                   status, "cond_timedwait");
2035   }
2036   _cur_index = -1;
2037 
2038   _counter = 0;
2039   status = pthread_mutex_unlock(_mutex);
2040   assert_status(status == 0, status, "invariant");
2041   // Paranoia to ensure our locked and lock-free paths interact
2042   // correctly with each other and Java-level accesses.
2043   OrderAccess::fence();
2044 
2045   // If externally suspended while waiting, re-suspend
2046   if (jt->handle_special_suspend_equivalent_condition()) {
2047     jt->java_suspend_self();
2048   }
2049 }
2050 
2051 void Parker::unpark() {
2052   int status = pthread_mutex_lock(_mutex);
2053   assert_status(status == 0, status, "invariant");
2054   const int s = _counter;
2055   _counter = 1;
2056   // must capture correct index before unlocking
2057   int index = _cur_index;
2058   status = pthread_mutex_unlock(_mutex);
2059   assert_status(status == 0, status, "invariant");
2060 
2061   // Note that we signal() *after* dropping the lock for "immortal" Events.
2062   // This is safe and avoids a common class of futile wakeups.  In rare
2063   // circumstances this can cause a thread to return prematurely from
2064   // cond_{timed}wait() but the spurious wakeup is benign and the victim
2065   // will simply re-test the condition and re-park itself.
2066   // This provides particular benefit if the underlying platform does not
2067   // provide wait morphing.
2068 
2069   if (s < 1 && index != -1) {
2070     // thread is definitely parked
2071     status = pthread_cond_signal(&_cond[index]);
2072     assert_status(status == 0, status, "invariant");
2073   }
2074 }
2075 
2076 
2077 #endif // !SOLARIS