1 /*
   2  * Copyright (c) 1999, 2017, 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 "utilities/globalDefinitions.hpp"
  26 #include "prims/jvm.h"
  27 #include "runtime/frame.inline.hpp"
  28 #include "runtime/interfaceSupport.hpp"
  29 #include "runtime/os.hpp"
  30 #include "utilities/align.hpp"
  31 #include "utilities/macros.hpp"
  32 #include "utilities/vmError.hpp"
  33 
  34 #ifndef __APPLE__
  35 // POSIX unamed semaphores are not supported on OS X.
  36 #include "semaphore_posix.hpp"
  37 #endif
  38 
  39 #include <dlfcn.h>
  40 #include <pthread.h>
  41 #include <semaphore.h>
  42 #include <signal.h>
  43 #include <sys/mman.h>
  44 #include <sys/resource.h>
  45 #include <sys/utsname.h>
  46 #include <time.h>
  47 #include <unistd.h>
  48 
  49 // Todo: provide a os::get_max_process_id() or similar. Number of processes
  50 // may have been configured, can be read more accurately from proc fs etc.
  51 #ifndef MAX_PID
  52 #define MAX_PID INT_MAX
  53 #endif
  54 #define IS_VALID_PID(p) (p > 0 && p < MAX_PID)
  55 
  56 #ifndef MAP_ANONYMOUS
  57   #define MAP_ANONYMOUS MAP_ANON
  58 #endif
  59 
  60 #define check_with_errno(check_type, cond, msg)                             \
  61   do {                                                                      \
  62     int err = errno;                                                        \
  63     check_type(cond, "%s; error='%s' (errno=%s)", msg, os::strerror(err),   \
  64                os::errno_name(err));                                        \
  65 } while (false)
  66 
  67 #define assert_with_errno(cond, msg)    check_with_errno(assert, cond, msg)
  68 #define guarantee_with_errno(cond, msg) check_with_errno(guarantee, cond, msg)
  69 
  70 // Check core dump limit and report possible place where core can be found
  71 void os::check_dump_limit(char* buffer, size_t bufferSize) {
  72   if (!FLAG_IS_DEFAULT(CreateCoredumpOnCrash) && !CreateCoredumpOnCrash) {
  73     jio_snprintf(buffer, bufferSize, "CreateCoredumpOnCrash is disabled from command line");
  74     VMError::record_coredump_status(buffer, false);
  75     return;
  76   }
  77 
  78   int n;
  79   struct rlimit rlim;
  80   bool success;
  81 
  82   char core_path[PATH_MAX];
  83   n = get_core_path(core_path, PATH_MAX);
  84 
  85   if (n <= 0) {
  86     jio_snprintf(buffer, bufferSize, "core.%d (may not exist)", current_process_id());
  87     success = true;
  88 #ifdef LINUX
  89   } else if (core_path[0] == '"') { // redirect to user process
  90     jio_snprintf(buffer, bufferSize, "Core dumps may be processed with %s", core_path);
  91     success = true;
  92 #endif
  93   } else if (getrlimit(RLIMIT_CORE, &rlim) != 0) {
  94     jio_snprintf(buffer, bufferSize, "%s (may not exist)", core_path);
  95     success = true;
  96   } else {
  97     switch(rlim.rlim_cur) {
  98       case RLIM_INFINITY:
  99         jio_snprintf(buffer, bufferSize, "%s", core_path);
 100         success = true;
 101         break;
 102       case 0:
 103         jio_snprintf(buffer, bufferSize, "Core dumps have been disabled. To enable core dumping, try \"ulimit -c unlimited\" before starting Java again");
 104         success = false;
 105         break;
 106       default:
 107         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);
 108         success = true;
 109         break;
 110     }
 111   }
 112 
 113   VMError::record_coredump_status(buffer, success);
 114 }
 115 
 116 int os::get_native_stack(address* stack, int frames, int toSkip) {
 117   int frame_idx = 0;
 118   int num_of_frames;  // number of frames captured
 119   frame fr = os::current_frame();
 120   while (fr.pc() && frame_idx < frames) {
 121     if (toSkip > 0) {
 122       toSkip --;
 123     } else {
 124       stack[frame_idx ++] = fr.pc();
 125     }
 126     if (fr.fp() == NULL || fr.cb() != NULL ||
 127         fr.sender_pc() == NULL || os::is_first_C_frame(&fr)) break;
 128 
 129     if (fr.sender_pc() && !os::is_first_C_frame(&fr)) {
 130       fr = os::get_sender_for_C_frame(&fr);
 131     } else {
 132       break;
 133     }
 134   }
 135   num_of_frames = frame_idx;
 136   for (; frame_idx < frames; frame_idx ++) {
 137     stack[frame_idx] = NULL;
 138   }
 139 
 140   return num_of_frames;
 141 }
 142 
 143 
 144 bool os::unsetenv(const char* name) {
 145   assert(name != NULL, "Null pointer");
 146   return (::unsetenv(name) == 0);
 147 }
 148 
 149 int os::get_last_error() {
 150   return errno;
 151 }
 152 
 153 bool os::is_debugger_attached() {
 154   // not implemented
 155   return false;
 156 }
 157 
 158 void os::wait_for_keypress_at_exit(void) {
 159   // don't do anything on posix platforms
 160   return;
 161 }
 162 
 163 int os::create_file_for_heap(const char* dir) {
 164 
 165   const char name_template[] = "/jvmheap.XXXXXX";
 166 
 167   char *fullname = (char*)os::malloc((strlen(dir) + strlen(name_template) + 1), mtInternal);
 168   if (fullname == NULL) {
 169     vm_exit_during_initialization(err_msg("Malloc failed during creation of backing file for heap (%s)", os::strerror(errno)));
 170     return -1;
 171   }
 172   (void)strncpy(fullname, dir, strlen(dir)+1);
 173   (void)strncat(fullname, name_template, strlen(name_template));
 174 
 175   os::native_path(fullname);
 176 
 177   sigset_t set, oldset;
 178   int ret = sigfillset(&set);
 179   assert_with_errno(ret == 0, "sigfillset returned error");
 180 
 181   // set the file creation mask.
 182   mode_t file_mode = S_IRUSR | S_IWUSR;
 183 
 184   // create a new file.
 185   int fd = mkstemp(fullname);
 186 
 187   if (fd < 0) {
 188     warning("Could not create file for heap with template %s", fullname);
 189     os::free(fullname);
 190     return -1;
 191   }
 192 
 193   // delete the name from the filesystem. When 'fd' is closed, the file (and space) will be deleted.
 194   ret = unlink(fullname);
 195   assert_with_errno(ret == 0, "unlink returned error");
 196 
 197   os::free(fullname);
 198   return fd;
 199 }
 200 
 201 static char* reserve_mmapped_memory(size_t bytes, char* requested_addr) {
 202   char * addr;
 203   int flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
 204   if (requested_addr != NULL) {
 205     assert((uintptr_t)requested_addr % os::vm_page_size() == 0, "Requested address should be aligned to OS page size");
 206     flags |= MAP_FIXED;
 207   }
 208 
 209   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
 210   // touch an uncommitted page. Otherwise, the read/write might
 211   // succeed if we have enough swap space to back the physical page.
 212   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
 213                        flags, -1, 0);
 214 
 215   if (addr != MAP_FAILED) {
 216     MemTracker::record_virtual_memory_reserve((address)addr, bytes, CALLER_PC);
 217     return addr;
 218   }
 219   return NULL;
 220 }
 221 
 222 static int util_posix_fallocate(int fd, off_t offset, off_t len) {
 223 #ifdef __APPLE__
 224   fstore_t store = { F_ALLOCATECONTIG, F_PEOFPOSMODE, 0, len };
 225   // First we try to get a continuous chunk of disk space
 226   int ret = fcntl(fd, F_PREALLOCATE, &store);
 227   if (ret == -1) {
 228     // Maybe we are too fragmented, try to allocate non-continuous range
 229     store.fst_flags = F_ALLOCATEALL;
 230     ret = fcntl(fd, F_PREALLOCATE, &store);
 231   }
 232   if(ret != -1) {
 233     return ftruncate(fd, len);
 234   }
 235   return -1;
 236 #else
 237   return posix_fallocate(fd, offset, len);
 238 #endif
 239 }
 240 
 241 // Map the given address range to the provided file descriptor.
 242 char* os::map_memory_to_file(char* base, size_t size, int fd) {
 243   assert(fd != -1, "File descriptor is not valid");
 244 
 245   // allocate space for the file
 246   if (util_posix_fallocate(fd, 0, (off_t)size) != 0) {
 247     vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory."));
 248     return NULL;
 249   }
 250 
 251   int prot = PROT_READ | PROT_WRITE;
 252   int flags = MAP_SHARED;
 253   if (base != NULL) {
 254     flags |= MAP_FIXED;
 255   }
 256   char* addr = (char*)mmap(base, size, prot, flags, fd, 0);
 257 
 258   if (addr == MAP_FAILED) {
 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   }
 300   else {
 301     extra_base = os::reserve_memory(extra_size, NULL, alignment);
 302   }
 303 
 304   if (extra_base == NULL) {
 305     return NULL;
 306   }
 307 
 308   // Do manual alignment
 309   char* aligned_base = align_up(extra_base, alignment);
 310 
 311   // [  |                                       |  ]
 312   // ^ extra_base
 313   //    ^ extra_base + begin_offset == aligned_base
 314   //     extra_base + begin_offset + size       ^
 315   //                       extra_base + extra_size ^
 316   // |<>| == begin_offset
 317   //                              end_offset == |<>|
 318   size_t begin_offset = aligned_base - extra_base;
 319   size_t end_offset = (extra_base + extra_size) - (aligned_base + size);
 320 
 321   if (begin_offset > 0) {
 322       os::release_memory(extra_base, begin_offset);
 323   }
 324 
 325   if (end_offset > 0) {
 326       os::release_memory(extra_base + begin_offset + size, end_offset);
 327   }
 328 
 329   if (file_desc != -1) {
 330     // After we have an aligned address, we can replace anonymous mapping with file mapping
 331     if (replace_existing_mapping_with_file_mapping(aligned_base, size, file_desc) == NULL) {
 332       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
 333     }
 334     MemTracker::record_virtual_memory_commit((address)aligned_base, size, CALLER_PC);
 335   }
 336   return aligned_base;
 337 }
 338 
 339 int os::log_vsnprintf(char* buf, size_t len, const char* fmt, va_list args) {
 340     return vsnprintf(buf, len, fmt, args);
 341 }
 342 
 343 int os::get_fileno(FILE* fp) {
 344   return NOT_AIX(::)fileno(fp);
 345 }
 346 
 347 struct tm* os::gmtime_pd(const time_t* clock, struct tm*  res) {
 348   return gmtime_r(clock, res);
 349 }
 350 
 351 void os::Posix::print_load_average(outputStream* st) {
 352   st->print("load average:");
 353   double loadavg[3];
 354   os::loadavg(loadavg, 3);
 355   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
 356   st->cr();
 357 }
 358 
 359 void os::Posix::print_rlimit_info(outputStream* st) {
 360   st->print("rlimit:");
 361   struct rlimit rlim;
 362 
 363   st->print(" STACK ");
 364   getrlimit(RLIMIT_STACK, &rlim);
 365   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 366   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 367 
 368   st->print(", CORE ");
 369   getrlimit(RLIMIT_CORE, &rlim);
 370   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 371   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 372 
 373   // Isn't there on solaris
 374 #if defined(AIX)
 375   st->print(", NPROC ");
 376   st->print("%d", sysconf(_SC_CHILD_MAX));
 377 #elif !defined(SOLARIS)
 378   st->print(", NPROC ");
 379   getrlimit(RLIMIT_NPROC, &rlim);
 380   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 381   else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
 382 #endif
 383 
 384   st->print(", NOFILE ");
 385   getrlimit(RLIMIT_NOFILE, &rlim);
 386   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 387   else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
 388 
 389   st->print(", AS ");
 390   getrlimit(RLIMIT_AS, &rlim);
 391   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 392   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 393 
 394   st->print(", DATA ");
 395   getrlimit(RLIMIT_DATA, &rlim);
 396   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 397   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 398 
 399   st->print(", FSIZE ");
 400   getrlimit(RLIMIT_FSIZE, &rlim);
 401   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 402   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 403 
 404   st->cr();
 405 }
 406 
 407 void os::Posix::print_uname_info(outputStream* st) {
 408   // kernel
 409   st->print("uname:");
 410   struct utsname name;
 411   uname(&name);
 412   st->print("%s ", name.sysname);
 413 #ifdef ASSERT
 414   st->print("%s ", name.nodename);
 415 #endif
 416   st->print("%s ", name.release);
 417   st->print("%s ", name.version);
 418   st->print("%s", name.machine);
 419   st->cr();
 420 }
 421 
 422 bool os::get_host_name(char* buf, size_t buflen) {
 423   struct utsname name;
 424   uname(&name);
 425   jio_snprintf(buf, buflen, "%s", name.nodename);
 426   return true;
 427 }
 428 
 429 bool os::has_allocatable_memory_limit(julong* limit) {
 430   struct rlimit rlim;
 431   int getrlimit_res = getrlimit(RLIMIT_AS, &rlim);
 432   // if there was an error when calling getrlimit, assume that there is no limitation
 433   // on virtual memory.
 434   bool result;
 435   if ((getrlimit_res != 0) || (rlim.rlim_cur == RLIM_INFINITY)) {
 436     result = false;
 437   } else {
 438     *limit = (julong)rlim.rlim_cur;
 439     result = true;
 440   }
 441 #ifdef _LP64
 442   return result;
 443 #else
 444   // arbitrary virtual space limit for 32 bit Unices found by testing. If
 445   // getrlimit above returned a limit, bound it with this limit. Otherwise
 446   // directly use it.
 447   const julong max_virtual_limit = (julong)3800*M;
 448   if (result) {
 449     *limit = MIN2(*limit, max_virtual_limit);
 450   } else {
 451     *limit = max_virtual_limit;
 452   }
 453 
 454   // bound by actually allocatable memory. The algorithm uses two bounds, an
 455   // upper and a lower limit. The upper limit is the current highest amount of
 456   // memory that could not be allocated, the lower limit is the current highest
 457   // amount of memory that could be allocated.
 458   // The algorithm iteratively refines the result by halving the difference
 459   // between these limits, updating either the upper limit (if that value could
 460   // not be allocated) or the lower limit (if the that value could be allocated)
 461   // until the difference between these limits is "small".
 462 
 463   // the minimum amount of memory we care about allocating.
 464   const julong min_allocation_size = M;
 465 
 466   julong upper_limit = *limit;
 467 
 468   // first check a few trivial cases
 469   if (is_allocatable(upper_limit) || (upper_limit <= min_allocation_size)) {
 470     *limit = upper_limit;
 471   } else if (!is_allocatable(min_allocation_size)) {
 472     // we found that not even min_allocation_size is allocatable. Return it
 473     // anyway. There is no point to search for a better value any more.
 474     *limit = min_allocation_size;
 475   } else {
 476     // perform the binary search.
 477     julong lower_limit = min_allocation_size;
 478     while ((upper_limit - lower_limit) > min_allocation_size) {
 479       julong temp_limit = ((upper_limit - lower_limit) / 2) + lower_limit;
 480       temp_limit = align_down(temp_limit, min_allocation_size);
 481       if (is_allocatable(temp_limit)) {
 482         lower_limit = temp_limit;
 483       } else {
 484         upper_limit = temp_limit;
 485       }
 486     }
 487     *limit = lower_limit;
 488   }
 489   return true;
 490 #endif
 491 }
 492 
 493 const char* os::get_current_directory(char *buf, size_t buflen) {
 494   return getcwd(buf, buflen);
 495 }
 496 
 497 FILE* os::open(int fd, const char* mode) {
 498   return ::fdopen(fd, mode);
 499 }
 500 
 501 void os::flockfile(FILE* fp) {
 502   ::flockfile(fp);
 503 }
 504 
 505 void os::funlockfile(FILE* fp) {
 506   ::funlockfile(fp);
 507 }
 508 
 509 // Builds a platform dependent Agent_OnLoad_<lib_name> function name
 510 // which is used to find statically linked in agents.
 511 // Parameters:
 512 //            sym_name: Symbol in library we are looking for
 513 //            lib_name: Name of library to look in, NULL for shared libs.
 514 //            is_absolute_path == true if lib_name is absolute path to agent
 515 //                                     such as "/a/b/libL.so"
 516 //            == false if only the base name of the library is passed in
 517 //               such as "L"
 518 char* os::build_agent_function_name(const char *sym_name, const char *lib_name,
 519                                     bool is_absolute_path) {
 520   char *agent_entry_name;
 521   size_t len;
 522   size_t name_len;
 523   size_t prefix_len = strlen(JNI_LIB_PREFIX);
 524   size_t suffix_len = strlen(JNI_LIB_SUFFIX);
 525   const char *start;
 526 
 527   if (lib_name != NULL) {
 528     name_len = strlen(lib_name);
 529     if (is_absolute_path) {
 530       // Need to strip path, prefix and suffix
 531       if ((start = strrchr(lib_name, *os::file_separator())) != NULL) {
 532         lib_name = ++start;
 533       }
 534       if (strlen(lib_name) <= (prefix_len + suffix_len)) {
 535         return NULL;
 536       }
 537       lib_name += prefix_len;
 538       name_len = strlen(lib_name) - suffix_len;
 539     }
 540   }
 541   len = (lib_name != NULL ? name_len : 0) + strlen(sym_name) + 2;
 542   agent_entry_name = NEW_C_HEAP_ARRAY_RETURN_NULL(char, len, mtThread);
 543   if (agent_entry_name == NULL) {
 544     return NULL;
 545   }
 546   strcpy(agent_entry_name, sym_name);
 547   if (lib_name != NULL) {
 548     strcat(agent_entry_name, "_");
 549     strncat(agent_entry_name, lib_name, name_len);
 550   }
 551   return agent_entry_name;
 552 }
 553 
 554 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
 555   assert(thread == Thread::current(),  "thread consistency check");
 556 
 557   ParkEvent * const slp = thread->_SleepEvent ;
 558   slp->reset() ;
 559   OrderAccess::fence() ;
 560 
 561   if (interruptible) {
 562     jlong prevtime = javaTimeNanos();
 563 
 564     for (;;) {
 565       if (os::is_interrupted(thread, true)) {
 566         return OS_INTRPT;
 567       }
 568 
 569       jlong newtime = javaTimeNanos();
 570 
 571       if (newtime - prevtime < 0) {
 572         // time moving backwards, should only happen if no monotonic clock
 573         // not a guarantee() because JVM should not abort on kernel/glibc bugs
 574         assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected in os::sleep(interruptible)");
 575       } else {
 576         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
 577       }
 578 
 579       if (millis <= 0) {
 580         return OS_OK;
 581       }
 582 
 583       prevtime = newtime;
 584 
 585       {
 586         assert(thread->is_Java_thread(), "sanity check");
 587         JavaThread *jt = (JavaThread *) thread;
 588         ThreadBlockInVM tbivm(jt);
 589         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
 590 
 591         jt->set_suspend_equivalent();
 592         // cleared by handle_special_suspend_equivalent_condition() or
 593         // java_suspend_self() via check_and_wait_while_suspended()
 594 
 595         slp->park(millis);
 596 
 597         // were we externally suspended while we were waiting?
 598         jt->check_and_wait_while_suspended();
 599       }
 600     }
 601   } else {
 602     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
 603     jlong prevtime = javaTimeNanos();
 604 
 605     for (;;) {
 606       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
 607       // the 1st iteration ...
 608       jlong newtime = javaTimeNanos();
 609 
 610       if (newtime - prevtime < 0) {
 611         // time moving backwards, should only happen if no monotonic clock
 612         // not a guarantee() because JVM should not abort on kernel/glibc bugs
 613         assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected on os::sleep(!interruptible)");
 614       } else {
 615         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
 616       }
 617 
 618       if (millis <= 0) break ;
 619 
 620       prevtime = newtime;
 621       slp->park(millis);
 622     }
 623     return OS_OK ;
 624   }
 625 }
 626 
 627 ////////////////////////////////////////////////////////////////////////////////
 628 // interrupt support
 629 
 630 void os::interrupt(Thread* thread) {
 631   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
 632     "possibility of dangling Thread pointer");
 633 
 634   OSThread* osthread = thread->osthread();
 635 
 636   if (!osthread->interrupted()) {
 637     osthread->set_interrupted(true);
 638     // More than one thread can get here with the same value of osthread,
 639     // resulting in multiple notifications.  We do, however, want the store
 640     // to interrupted() to be visible to other threads before we execute unpark().
 641     OrderAccess::fence();
 642     ParkEvent * const slp = thread->_SleepEvent ;
 643     if (slp != NULL) slp->unpark() ;
 644   }
 645 
 646   // For JSR166. Unpark even if interrupt status already was set
 647   if (thread->is_Java_thread())
 648     ((JavaThread*)thread)->parker()->unpark();
 649 
 650   ParkEvent * ev = thread->_ParkEvent ;
 651   if (ev != NULL) ev->unpark() ;
 652 
 653 }
 654 
 655 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
 656   assert(Thread::current() == thread || Threads_lock->owned_by_self(),
 657     "possibility of dangling Thread pointer");
 658 
 659   OSThread* osthread = thread->osthread();
 660 
 661   bool interrupted = osthread->interrupted();
 662 
 663   // NOTE that since there is no "lock" around the interrupt and
 664   // is_interrupted operations, there is the possibility that the
 665   // interrupted flag (in osThread) will be "false" but that the
 666   // low-level events will be in the signaled state. This is
 667   // intentional. The effect of this is that Object.wait() and
 668   // LockSupport.park() will appear to have a spurious wakeup, which
 669   // is allowed and not harmful, and the possibility is so rare that
 670   // it is not worth the added complexity to add yet another lock.
 671   // For the sleep event an explicit reset is performed on entry
 672   // to os::sleep, so there is no early return. It has also been
 673   // recommended not to put the interrupted flag into the "event"
 674   // structure because it hides the issue.
 675   if (interrupted && clear_interrupted) {
 676     osthread->set_interrupted(false);
 677     // consider thread->_SleepEvent->reset() ... optional optimization
 678   }
 679 
 680   return interrupted;
 681 }
 682 
 683 
 684 
 685 static const struct {
 686   int sig; const char* name;
 687 }
 688  g_signal_info[] =
 689   {
 690   {  SIGABRT,     "SIGABRT" },
 691 #ifdef SIGAIO
 692   {  SIGAIO,      "SIGAIO" },
 693 #endif
 694   {  SIGALRM,     "SIGALRM" },
 695 #ifdef SIGALRM1
 696   {  SIGALRM1,    "SIGALRM1" },
 697 #endif
 698   {  SIGBUS,      "SIGBUS" },
 699 #ifdef SIGCANCEL
 700   {  SIGCANCEL,   "SIGCANCEL" },
 701 #endif
 702   {  SIGCHLD,     "SIGCHLD" },
 703 #ifdef SIGCLD
 704   {  SIGCLD,      "SIGCLD" },
 705 #endif
 706   {  SIGCONT,     "SIGCONT" },
 707 #ifdef SIGCPUFAIL
 708   {  SIGCPUFAIL,  "SIGCPUFAIL" },
 709 #endif
 710 #ifdef SIGDANGER
 711   {  SIGDANGER,   "SIGDANGER" },
 712 #endif
 713 #ifdef SIGDIL
 714   {  SIGDIL,      "SIGDIL" },
 715 #endif
 716 #ifdef SIGEMT
 717   {  SIGEMT,      "SIGEMT" },
 718 #endif
 719   {  SIGFPE,      "SIGFPE" },
 720 #ifdef SIGFREEZE
 721   {  SIGFREEZE,   "SIGFREEZE" },
 722 #endif
 723 #ifdef SIGGFAULT
 724   {  SIGGFAULT,   "SIGGFAULT" },
 725 #endif
 726 #ifdef SIGGRANT
 727   {  SIGGRANT,    "SIGGRANT" },
 728 #endif
 729   {  SIGHUP,      "SIGHUP" },
 730   {  SIGILL,      "SIGILL" },
 731   {  SIGINT,      "SIGINT" },
 732 #ifdef SIGIO
 733   {  SIGIO,       "SIGIO" },
 734 #endif
 735 #ifdef SIGIOINT
 736   {  SIGIOINT,    "SIGIOINT" },
 737 #endif
 738 #ifdef SIGIOT
 739 // SIGIOT is there for BSD compatibility, but on most Unices just a
 740 // synonym for SIGABRT. The result should be "SIGABRT", not
 741 // "SIGIOT".
 742 #if (SIGIOT != SIGABRT )
 743   {  SIGIOT,      "SIGIOT" },
 744 #endif
 745 #endif
 746 #ifdef SIGKAP
 747   {  SIGKAP,      "SIGKAP" },
 748 #endif
 749   {  SIGKILL,     "SIGKILL" },
 750 #ifdef SIGLOST
 751   {  SIGLOST,     "SIGLOST" },
 752 #endif
 753 #ifdef SIGLWP
 754   {  SIGLWP,      "SIGLWP" },
 755 #endif
 756 #ifdef SIGLWPTIMER
 757   {  SIGLWPTIMER, "SIGLWPTIMER" },
 758 #endif
 759 #ifdef SIGMIGRATE
 760   {  SIGMIGRATE,  "SIGMIGRATE" },
 761 #endif
 762 #ifdef SIGMSG
 763   {  SIGMSG,      "SIGMSG" },
 764 #endif
 765   {  SIGPIPE,     "SIGPIPE" },
 766 #ifdef SIGPOLL
 767   {  SIGPOLL,     "SIGPOLL" },
 768 #endif
 769 #ifdef SIGPRE
 770   {  SIGPRE,      "SIGPRE" },
 771 #endif
 772   {  SIGPROF,     "SIGPROF" },
 773 #ifdef SIGPTY
 774   {  SIGPTY,      "SIGPTY" },
 775 #endif
 776 #ifdef SIGPWR
 777   {  SIGPWR,      "SIGPWR" },
 778 #endif
 779   {  SIGQUIT,     "SIGQUIT" },
 780 #ifdef SIGRECONFIG
 781   {  SIGRECONFIG, "SIGRECONFIG" },
 782 #endif
 783 #ifdef SIGRECOVERY
 784   {  SIGRECOVERY, "SIGRECOVERY" },
 785 #endif
 786 #ifdef SIGRESERVE
 787   {  SIGRESERVE,  "SIGRESERVE" },
 788 #endif
 789 #ifdef SIGRETRACT
 790   {  SIGRETRACT,  "SIGRETRACT" },
 791 #endif
 792 #ifdef SIGSAK
 793   {  SIGSAK,      "SIGSAK" },
 794 #endif
 795   {  SIGSEGV,     "SIGSEGV" },
 796 #ifdef SIGSOUND
 797   {  SIGSOUND,    "SIGSOUND" },
 798 #endif
 799 #ifdef SIGSTKFLT
 800   {  SIGSTKFLT,    "SIGSTKFLT" },
 801 #endif
 802   {  SIGSTOP,     "SIGSTOP" },
 803   {  SIGSYS,      "SIGSYS" },
 804 #ifdef SIGSYSERROR
 805   {  SIGSYSERROR, "SIGSYSERROR" },
 806 #endif
 807 #ifdef SIGTALRM
 808   {  SIGTALRM,    "SIGTALRM" },
 809 #endif
 810   {  SIGTERM,     "SIGTERM" },
 811 #ifdef SIGTHAW
 812   {  SIGTHAW,     "SIGTHAW" },
 813 #endif
 814   {  SIGTRAP,     "SIGTRAP" },
 815 #ifdef SIGTSTP
 816   {  SIGTSTP,     "SIGTSTP" },
 817 #endif
 818   {  SIGTTIN,     "SIGTTIN" },
 819   {  SIGTTOU,     "SIGTTOU" },
 820 #ifdef SIGURG
 821   {  SIGURG,      "SIGURG" },
 822 #endif
 823   {  SIGUSR1,     "SIGUSR1" },
 824   {  SIGUSR2,     "SIGUSR2" },
 825 #ifdef SIGVIRT
 826   {  SIGVIRT,     "SIGVIRT" },
 827 #endif
 828   {  SIGVTALRM,   "SIGVTALRM" },
 829 #ifdef SIGWAITING
 830   {  SIGWAITING,  "SIGWAITING" },
 831 #endif
 832 #ifdef SIGWINCH
 833   {  SIGWINCH,    "SIGWINCH" },
 834 #endif
 835 #ifdef SIGWINDOW
 836   {  SIGWINDOW,   "SIGWINDOW" },
 837 #endif
 838   {  SIGXCPU,     "SIGXCPU" },
 839   {  SIGXFSZ,     "SIGXFSZ" },
 840 #ifdef SIGXRES
 841   {  SIGXRES,     "SIGXRES" },
 842 #endif
 843   { -1, NULL }
 844 };
 845 
 846 // Returned string is a constant. For unknown signals "UNKNOWN" is returned.
 847 const char* os::Posix::get_signal_name(int sig, char* out, size_t outlen) {
 848 
 849   const char* ret = NULL;
 850 
 851 #ifdef SIGRTMIN
 852   if (sig >= SIGRTMIN && sig <= SIGRTMAX) {
 853     if (sig == SIGRTMIN) {
 854       ret = "SIGRTMIN";
 855     } else if (sig == SIGRTMAX) {
 856       ret = "SIGRTMAX";
 857     } else {
 858       jio_snprintf(out, outlen, "SIGRTMIN+%d", sig - SIGRTMIN);
 859       return out;
 860     }
 861   }
 862 #endif
 863 
 864   if (sig > 0) {
 865     for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
 866       if (g_signal_info[idx].sig == sig) {
 867         ret = g_signal_info[idx].name;
 868         break;
 869       }
 870     }
 871   }
 872 
 873   if (!ret) {
 874     if (!is_valid_signal(sig)) {
 875       ret = "INVALID";
 876     } else {
 877       ret = "UNKNOWN";
 878     }
 879   }
 880 
 881   if (out && outlen > 0) {
 882     strncpy(out, ret, outlen);
 883     out[outlen - 1] = '\0';
 884   }
 885   return out;
 886 }
 887 
 888 int os::Posix::get_signal_number(const char* signal_name) {
 889   char tmp[30];
 890   const char* s = signal_name;
 891   if (s[0] != 'S' || s[1] != 'I' || s[2] != 'G') {
 892     jio_snprintf(tmp, sizeof(tmp), "SIG%s", signal_name);
 893     s = tmp;
 894   }
 895   for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
 896     if (strcmp(g_signal_info[idx].name, s) == 0) {
 897       return g_signal_info[idx].sig;
 898     }
 899   }
 900   return -1;
 901 }
 902 
 903 int os::get_signal_number(const char* signal_name) {
 904   return os::Posix::get_signal_number(signal_name);
 905 }
 906 
 907 // Returns true if signal number is valid.
 908 bool os::Posix::is_valid_signal(int sig) {
 909   // MacOS not really POSIX compliant: sigaddset does not return
 910   // an error for invalid signal numbers. However, MacOS does not
 911   // support real time signals and simply seems to have just 33
 912   // signals with no holes in the signal range.
 913 #ifdef __APPLE__
 914   return sig >= 1 && sig < NSIG;
 915 #else
 916   // Use sigaddset to check for signal validity.
 917   sigset_t set;
 918   sigemptyset(&set);
 919   if (sigaddset(&set, sig) == -1 && errno == EINVAL) {
 920     return false;
 921   }
 922   return true;
 923 #endif
 924 }
 925 
 926 // Returns:
 927 // NULL for an invalid signal number
 928 // "SIG<num>" for a valid but unknown signal number
 929 // signal name otherwise.
 930 const char* os::exception_name(int sig, char* buf, size_t size) {
 931   if (!os::Posix::is_valid_signal(sig)) {
 932     return NULL;
 933   }
 934   const char* const name = os::Posix::get_signal_name(sig, buf, size);
 935   if (strcmp(name, "UNKNOWN") == 0) {
 936     jio_snprintf(buf, size, "SIG%d", sig);
 937   }
 938   return buf;
 939 }
 940 
 941 #define NUM_IMPORTANT_SIGS 32
 942 // Returns one-line short description of a signal set in a user provided buffer.
 943 const char* os::Posix::describe_signal_set_short(const sigset_t* set, char* buffer, size_t buf_size) {
 944   assert(buf_size == (NUM_IMPORTANT_SIGS + 1), "wrong buffer size");
 945   // Note: for shortness, just print out the first 32. That should
 946   // cover most of the useful ones, apart from realtime signals.
 947   for (int sig = 1; sig <= NUM_IMPORTANT_SIGS; sig++) {
 948     const int rc = sigismember(set, sig);
 949     if (rc == -1 && errno == EINVAL) {
 950       buffer[sig-1] = '?';
 951     } else {
 952       buffer[sig-1] = rc == 0 ? '0' : '1';
 953     }
 954   }
 955   buffer[NUM_IMPORTANT_SIGS] = 0;
 956   return buffer;
 957 }
 958 
 959 // Prints one-line description of a signal set.
 960 void os::Posix::print_signal_set_short(outputStream* st, const sigset_t* set) {
 961   char buf[NUM_IMPORTANT_SIGS + 1];
 962   os::Posix::describe_signal_set_short(set, buf, sizeof(buf));
 963   st->print("%s", buf);
 964 }
 965 
 966 // Writes one-line description of a combination of sigaction.sa_flags into a user
 967 // provided buffer. Returns that buffer.
 968 const char* os::Posix::describe_sa_flags(int flags, char* buffer, size_t size) {
 969   char* p = buffer;
 970   size_t remaining = size;
 971   bool first = true;
 972   int idx = 0;
 973 
 974   assert(buffer, "invalid argument");
 975 
 976   if (size == 0) {
 977     return buffer;
 978   }
 979 
 980   strncpy(buffer, "none", size);
 981 
 982   const struct {
 983     // NB: i is an unsigned int here because SA_RESETHAND is on some
 984     // systems 0x80000000, which is implicitly unsigned.  Assignining
 985     // it to an int field would be an overflow in unsigned-to-signed
 986     // conversion.
 987     unsigned int i;
 988     const char* s;
 989   } flaginfo [] = {
 990     { SA_NOCLDSTOP, "SA_NOCLDSTOP" },
 991     { SA_ONSTACK,   "SA_ONSTACK"   },
 992     { SA_RESETHAND, "SA_RESETHAND" },
 993     { SA_RESTART,   "SA_RESTART"   },
 994     { SA_SIGINFO,   "SA_SIGINFO"   },
 995     { SA_NOCLDWAIT, "SA_NOCLDWAIT" },
 996     { SA_NODEFER,   "SA_NODEFER"   },
 997 #ifdef AIX
 998     { SA_ONSTACK,   "SA_ONSTACK"   },
 999     { SA_OLDSTYLE,  "SA_OLDSTYLE"  },
1000 #endif
1001     { 0, NULL }
1002   };
1003 
1004   for (idx = 0; flaginfo[idx].s && remaining > 1; idx++) {
1005     if (flags & flaginfo[idx].i) {
1006       if (first) {
1007         jio_snprintf(p, remaining, "%s", flaginfo[idx].s);
1008         first = false;
1009       } else {
1010         jio_snprintf(p, remaining, "|%s", flaginfo[idx].s);
1011       }
1012       const size_t len = strlen(p);
1013       p += len;
1014       remaining -= len;
1015     }
1016   }
1017 
1018   buffer[size - 1] = '\0';
1019 
1020   return buffer;
1021 }
1022 
1023 // Prints one-line description of a combination of sigaction.sa_flags.
1024 void os::Posix::print_sa_flags(outputStream* st, int flags) {
1025   char buffer[0x100];
1026   os::Posix::describe_sa_flags(flags, buffer, sizeof(buffer));
1027   st->print("%s", buffer);
1028 }
1029 
1030 // Helper function for os::Posix::print_siginfo_...():
1031 // return a textual description for signal code.
1032 struct enum_sigcode_desc_t {
1033   const char* s_name;
1034   const char* s_desc;
1035 };
1036 
1037 static bool get_signal_code_description(const siginfo_t* si, enum_sigcode_desc_t* out) {
1038 
1039   const struct {
1040     int sig; int code; const char* s_code; const char* s_desc;
1041   } t1 [] = {
1042     { SIGILL,  ILL_ILLOPC,   "ILL_ILLOPC",   "Illegal opcode." },
1043     { SIGILL,  ILL_ILLOPN,   "ILL_ILLOPN",   "Illegal operand." },
1044     { SIGILL,  ILL_ILLADR,   "ILL_ILLADR",   "Illegal addressing mode." },
1045     { SIGILL,  ILL_ILLTRP,   "ILL_ILLTRP",   "Illegal trap." },
1046     { SIGILL,  ILL_PRVOPC,   "ILL_PRVOPC",   "Privileged opcode." },
1047     { SIGILL,  ILL_PRVREG,   "ILL_PRVREG",   "Privileged register." },
1048     { SIGILL,  ILL_COPROC,   "ILL_COPROC",   "Coprocessor error." },
1049     { SIGILL,  ILL_BADSTK,   "ILL_BADSTK",   "Internal stack error." },
1050 #if defined(IA64) && defined(LINUX)
1051     { SIGILL,  ILL_BADIADDR, "ILL_BADIADDR", "Unimplemented instruction address" },
1052     { SIGILL,  ILL_BREAK,    "ILL_BREAK",    "Application Break instruction" },
1053 #endif
1054     { SIGFPE,  FPE_INTDIV,   "FPE_INTDIV",   "Integer divide by zero." },
1055     { SIGFPE,  FPE_INTOVF,   "FPE_INTOVF",   "Integer overflow." },
1056     { SIGFPE,  FPE_FLTDIV,   "FPE_FLTDIV",   "Floating-point divide by zero." },
1057     { SIGFPE,  FPE_FLTOVF,   "FPE_FLTOVF",   "Floating-point overflow." },
1058     { SIGFPE,  FPE_FLTUND,   "FPE_FLTUND",   "Floating-point underflow." },
1059     { SIGFPE,  FPE_FLTRES,   "FPE_FLTRES",   "Floating-point inexact result." },
1060     { SIGFPE,  FPE_FLTINV,   "FPE_FLTINV",   "Invalid floating-point operation." },
1061     { SIGFPE,  FPE_FLTSUB,   "FPE_FLTSUB",   "Subscript out of range." },
1062     { SIGSEGV, SEGV_MAPERR,  "SEGV_MAPERR",  "Address not mapped to object." },
1063     { SIGSEGV, SEGV_ACCERR,  "SEGV_ACCERR",  "Invalid permissions for mapped object." },
1064 #ifdef AIX
1065     // no explanation found what keyerr would be
1066     { SIGSEGV, SEGV_KEYERR,  "SEGV_KEYERR",  "key error" },
1067 #endif
1068 #if defined(IA64) && !defined(AIX)
1069     { SIGSEGV, SEGV_PSTKOVF, "SEGV_PSTKOVF", "Paragraph stack overflow" },
1070 #endif
1071 #if defined(__sparc) && defined(SOLARIS)
1072 // define Solaris Sparc M7 ADI SEGV signals
1073 #if !defined(SEGV_ACCADI)
1074 #define SEGV_ACCADI 3
1075 #endif
1076     { SIGSEGV, SEGV_ACCADI,  "SEGV_ACCADI",  "ADI not enabled for mapped object." },
1077 #if !defined(SEGV_ACCDERR)
1078 #define SEGV_ACCDERR 4
1079 #endif
1080     { SIGSEGV, SEGV_ACCDERR, "SEGV_ACCDERR", "ADI disrupting exception." },
1081 #if !defined(SEGV_ACCPERR)
1082 #define SEGV_ACCPERR 5
1083 #endif
1084     { SIGSEGV, SEGV_ACCPERR, "SEGV_ACCPERR", "ADI precise exception." },
1085 #endif // defined(__sparc) && defined(SOLARIS)
1086     { SIGBUS,  BUS_ADRALN,   "BUS_ADRALN",   "Invalid address alignment." },
1087     { SIGBUS,  BUS_ADRERR,   "BUS_ADRERR",   "Nonexistent physical address." },
1088     { SIGBUS,  BUS_OBJERR,   "BUS_OBJERR",   "Object-specific hardware error." },
1089     { SIGTRAP, TRAP_BRKPT,   "TRAP_BRKPT",   "Process breakpoint." },
1090     { SIGTRAP, TRAP_TRACE,   "TRAP_TRACE",   "Process trace trap." },
1091     { SIGCHLD, CLD_EXITED,   "CLD_EXITED",   "Child has exited." },
1092     { SIGCHLD, CLD_KILLED,   "CLD_KILLED",   "Child has terminated abnormally and did not create a core file." },
1093     { SIGCHLD, CLD_DUMPED,   "CLD_DUMPED",   "Child has terminated abnormally and created a core file." },
1094     { SIGCHLD, CLD_TRAPPED,  "CLD_TRAPPED",  "Traced child has trapped." },
1095     { SIGCHLD, CLD_STOPPED,  "CLD_STOPPED",  "Child has stopped." },
1096     { SIGCHLD, CLD_CONTINUED,"CLD_CONTINUED","Stopped child has continued." },
1097 #ifdef SIGPOLL
1098     { SIGPOLL, POLL_OUT,     "POLL_OUT",     "Output buffers available." },
1099     { SIGPOLL, POLL_MSG,     "POLL_MSG",     "Input message available." },
1100     { SIGPOLL, POLL_ERR,     "POLL_ERR",     "I/O error." },
1101     { SIGPOLL, POLL_PRI,     "POLL_PRI",     "High priority input available." },
1102     { SIGPOLL, POLL_HUP,     "POLL_HUP",     "Device disconnected. [Option End]" },
1103 #endif
1104     { -1, -1, NULL, NULL }
1105   };
1106 
1107   // Codes valid in any signal context.
1108   const struct {
1109     int code; const char* s_code; const char* s_desc;
1110   } t2 [] = {
1111     { SI_USER,      "SI_USER",     "Signal sent by kill()." },
1112     { SI_QUEUE,     "SI_QUEUE",    "Signal sent by the sigqueue()." },
1113     { SI_TIMER,     "SI_TIMER",    "Signal generated by expiration of a timer set by timer_settime()." },
1114     { SI_ASYNCIO,   "SI_ASYNCIO",  "Signal generated by completion of an asynchronous I/O request." },
1115     { SI_MESGQ,     "SI_MESGQ",    "Signal generated by arrival of a message on an empty message queue." },
1116     // Linux specific
1117 #ifdef SI_TKILL
1118     { SI_TKILL,     "SI_TKILL",    "Signal sent by tkill (pthread_kill)" },
1119 #endif
1120 #ifdef SI_DETHREAD
1121     { SI_DETHREAD,  "SI_DETHREAD", "Signal sent by execve() killing subsidiary threads" },
1122 #endif
1123 #ifdef SI_KERNEL
1124     { SI_KERNEL,    "SI_KERNEL",   "Signal sent by kernel." },
1125 #endif
1126 #ifdef SI_SIGIO
1127     { SI_SIGIO,     "SI_SIGIO",    "Signal sent by queued SIGIO" },
1128 #endif
1129 
1130 #ifdef AIX
1131     { SI_UNDEFINED, "SI_UNDEFINED","siginfo contains partial information" },
1132     { SI_EMPTY,     "SI_EMPTY",    "siginfo contains no useful information" },
1133 #endif
1134 
1135 #ifdef __sun
1136     { SI_NOINFO,    "SI_NOINFO",   "No signal information" },
1137     { SI_RCTL,      "SI_RCTL",     "kernel generated signal via rctl action" },
1138     { SI_LWP,       "SI_LWP",      "Signal sent via lwp_kill" },
1139 #endif
1140 
1141     { -1, NULL, NULL }
1142   };
1143 
1144   const char* s_code = NULL;
1145   const char* s_desc = NULL;
1146 
1147   for (int i = 0; t1[i].sig != -1; i ++) {
1148     if (t1[i].sig == si->si_signo && t1[i].code == si->si_code) {
1149       s_code = t1[i].s_code;
1150       s_desc = t1[i].s_desc;
1151       break;
1152     }
1153   }
1154 
1155   if (s_code == NULL) {
1156     for (int i = 0; t2[i].s_code != NULL; i ++) {
1157       if (t2[i].code == si->si_code) {
1158         s_code = t2[i].s_code;
1159         s_desc = t2[i].s_desc;
1160       }
1161     }
1162   }
1163 
1164   if (s_code == NULL) {
1165     out->s_name = "unknown";
1166     out->s_desc = "unknown";
1167     return false;
1168   }
1169 
1170   out->s_name = s_code;
1171   out->s_desc = s_desc;
1172 
1173   return true;
1174 }
1175 
1176 void os::print_siginfo(outputStream* os, const void* si0) {
1177 
1178   const siginfo_t* const si = (const siginfo_t*) si0;
1179 
1180   char buf[20];
1181   os->print("siginfo:");
1182 
1183   if (!si) {
1184     os->print(" <null>");
1185     return;
1186   }
1187 
1188   const int sig = si->si_signo;
1189 
1190   os->print(" si_signo: %d (%s)", sig, os::Posix::get_signal_name(sig, buf, sizeof(buf)));
1191 
1192   enum_sigcode_desc_t ed;
1193   get_signal_code_description(si, &ed);
1194   os->print(", si_code: %d (%s)", si->si_code, ed.s_name);
1195 
1196   if (si->si_errno) {
1197     os->print(", si_errno: %d", si->si_errno);
1198   }
1199 
1200   // Output additional information depending on the signal code.
1201 
1202   // Note: Many implementations lump si_addr, si_pid, si_uid etc. together as unions,
1203   // so it depends on the context which member to use. For synchronous error signals,
1204   // we print si_addr, unless the signal was sent by another process or thread, in
1205   // which case we print out pid or tid of the sender.
1206   if (si->si_code == SI_USER || si->si_code == SI_QUEUE) {
1207     const pid_t pid = si->si_pid;
1208     os->print(", si_pid: %ld", (long) pid);
1209     if (IS_VALID_PID(pid)) {
1210       const pid_t me = getpid();
1211       if (me == pid) {
1212         os->print(" (current process)");
1213       }
1214     } else {
1215       os->print(" (invalid)");
1216     }
1217     os->print(", si_uid: %ld", (long) si->si_uid);
1218     if (sig == SIGCHLD) {
1219       os->print(", si_status: %d", si->si_status);
1220     }
1221   } else if (sig == SIGSEGV || sig == SIGBUS || sig == SIGILL ||
1222              sig == SIGTRAP || sig == SIGFPE) {
1223     os->print(", si_addr: " PTR_FORMAT, p2i(si->si_addr));
1224 #ifdef SIGPOLL
1225   } else if (sig == SIGPOLL) {
1226     os->print(", si_band: %ld", si->si_band);
1227 #endif
1228   }
1229 
1230 }
1231 
1232 int os::Posix::unblock_thread_signal_mask(const sigset_t *set) {
1233   return pthread_sigmask(SIG_UNBLOCK, set, NULL);
1234 }
1235 
1236 address os::Posix::ucontext_get_pc(const ucontext_t* ctx) {
1237 #if defined(AIX)
1238    return Aix::ucontext_get_pc(ctx);
1239 #elif defined(BSD)
1240    return Bsd::ucontext_get_pc(ctx);
1241 #elif defined(LINUX)
1242    return Linux::ucontext_get_pc(ctx);
1243 #elif defined(SOLARIS)
1244    return Solaris::ucontext_get_pc(ctx);
1245 #else
1246    VMError::report_and_die("unimplemented ucontext_get_pc");
1247 #endif
1248 }
1249 
1250 void os::Posix::ucontext_set_pc(ucontext_t* ctx, address pc) {
1251 #if defined(AIX)
1252    Aix::ucontext_set_pc(ctx, pc);
1253 #elif defined(BSD)
1254    Bsd::ucontext_set_pc(ctx, pc);
1255 #elif defined(LINUX)
1256    Linux::ucontext_set_pc(ctx, pc);
1257 #elif defined(SOLARIS)
1258    Solaris::ucontext_set_pc(ctx, pc);
1259 #else
1260    VMError::report_and_die("unimplemented ucontext_get_pc");
1261 #endif
1262 }
1263 
1264 char* os::Posix::describe_pthread_attr(char* buf, size_t buflen, const pthread_attr_t* attr) {
1265   size_t stack_size = 0;
1266   size_t guard_size = 0;
1267   int detachstate = 0;
1268   pthread_attr_getstacksize(attr, &stack_size);
1269   pthread_attr_getguardsize(attr, &guard_size);
1270   // Work around linux NPTL implementation error, see also os::create_thread() in os_linux.cpp.
1271   LINUX_ONLY(stack_size -= guard_size);
1272   pthread_attr_getdetachstate(attr, &detachstate);
1273   jio_snprintf(buf, buflen, "stacksize: " SIZE_FORMAT "k, guardsize: " SIZE_FORMAT "k, %s",
1274     stack_size / 1024, guard_size / 1024,
1275     (detachstate == PTHREAD_CREATE_DETACHED ? "detached" : "joinable"));
1276   return buf;
1277 }
1278 
1279 char* os::Posix::realpath(const char* filename, char* outbuf, size_t outbuflen) {
1280 
1281   if (filename == NULL || outbuf == NULL || outbuflen < 1) {
1282     assert(false, "os::Posix::realpath: invalid arguments.");
1283     errno = EINVAL;
1284     return NULL;
1285   }
1286 
1287   char* result = NULL;
1288 
1289   // This assumes platform realpath() is implemented according to POSIX.1-2008.
1290   // POSIX.1-2008 allows to specify NULL for the output buffer, in which case
1291   // output buffer is dynamically allocated and must be ::free()'d by the caller.
1292   char* p = ::realpath(filename, NULL);
1293   if (p != NULL) {
1294     if (strlen(p) < outbuflen) {
1295       strcpy(outbuf, p);
1296       result = outbuf;
1297     } else {
1298       errno = ENAMETOOLONG;
1299     }
1300     ::free(p); // *not* os::free
1301   } else {
1302     // Fallback for platforms struggling with modern Posix standards (AIX 5.3, 6.1). If realpath
1303     // returns EINVAL, this may indicate that realpath is not POSIX.1-2008 compatible and
1304     // that it complains about the NULL we handed down as user buffer.
1305     // In this case, use the user provided buffer but at least check whether realpath caused
1306     // a memory overwrite.
1307     if (errno == EINVAL) {
1308       outbuf[outbuflen - 1] = '\0';
1309       p = ::realpath(filename, outbuf);
1310       if (p != NULL) {
1311         guarantee(outbuf[outbuflen - 1] == '\0', "realpath buffer overwrite detected.");
1312         result = p;
1313       }
1314     }
1315   }
1316   return result;
1317 
1318 }
1319 
1320 
1321 // Check minimum allowable stack sizes for thread creation and to initialize
1322 // the java system classes, including StackOverflowError - depends on page
1323 // size.
1324 // The space needed for frames during startup is platform dependent. It
1325 // depends on word size, platform calling conventions, C frame layout and
1326 // interpreter/C1/C2 design decisions. Therefore this is given in a
1327 // platform (os/cpu) dependent constant.
1328 // To this, space for guard mechanisms is added, which depends on the
1329 // page size which again depends on the concrete system the VM is running
1330 // on. Space for libc guard pages is not included in this size.
1331 jint os::Posix::set_minimum_stack_sizes() {
1332   size_t os_min_stack_allowed = SOLARIS_ONLY(thr_min_stack()) NOT_SOLARIS(PTHREAD_STACK_MIN);
1333 
1334   _java_thread_min_stack_allowed = _java_thread_min_stack_allowed +
1335                                    JavaThread::stack_guard_zone_size() +
1336                                    JavaThread::stack_shadow_zone_size();
1337 
1338   _java_thread_min_stack_allowed = align_up(_java_thread_min_stack_allowed, vm_page_size());
1339   _java_thread_min_stack_allowed = MAX2(_java_thread_min_stack_allowed, os_min_stack_allowed);
1340 
1341   size_t stack_size_in_bytes = ThreadStackSize * K;
1342   if (stack_size_in_bytes != 0 &&
1343       stack_size_in_bytes < _java_thread_min_stack_allowed) {
1344     // The '-Xss' and '-XX:ThreadStackSize=N' options both set
1345     // ThreadStackSize so we go with "Java thread stack size" instead
1346     // of "ThreadStackSize" to be more friendly.
1347     tty->print_cr("\nThe Java thread stack size specified is too small. "
1348                   "Specify at least " SIZE_FORMAT "k",
1349                   _java_thread_min_stack_allowed / K);
1350     return JNI_ERR;
1351   }
1352 
1353   // Make the stack size a multiple of the page size so that
1354   // the yellow/red zones can be guarded.
1355   JavaThread::set_stack_size_at_create(align_up(stack_size_in_bytes, vm_page_size()));
1356 
1357   // Reminder: a compiler thread is a Java thread.
1358   _compiler_thread_min_stack_allowed = _compiler_thread_min_stack_allowed +
1359                                        JavaThread::stack_guard_zone_size() +
1360                                        JavaThread::stack_shadow_zone_size();
1361 
1362   _compiler_thread_min_stack_allowed = align_up(_compiler_thread_min_stack_allowed, vm_page_size());
1363   _compiler_thread_min_stack_allowed = MAX2(_compiler_thread_min_stack_allowed, os_min_stack_allowed);
1364 
1365   stack_size_in_bytes = CompilerThreadStackSize * K;
1366   if (stack_size_in_bytes != 0 &&
1367       stack_size_in_bytes < _compiler_thread_min_stack_allowed) {
1368     tty->print_cr("\nThe CompilerThreadStackSize specified is too small. "
1369                   "Specify at least " SIZE_FORMAT "k",
1370                   _compiler_thread_min_stack_allowed / K);
1371     return JNI_ERR;
1372   }
1373 
1374   _vm_internal_thread_min_stack_allowed = align_up(_vm_internal_thread_min_stack_allowed, vm_page_size());
1375   _vm_internal_thread_min_stack_allowed = MAX2(_vm_internal_thread_min_stack_allowed, os_min_stack_allowed);
1376 
1377   stack_size_in_bytes = VMThreadStackSize * K;
1378   if (stack_size_in_bytes != 0 &&
1379       stack_size_in_bytes < _vm_internal_thread_min_stack_allowed) {
1380     tty->print_cr("\nThe VMThreadStackSize specified is too small. "
1381                   "Specify at least " SIZE_FORMAT "k",
1382                   _vm_internal_thread_min_stack_allowed / K);
1383     return JNI_ERR;
1384   }
1385   return JNI_OK;
1386 }
1387 
1388 // Called when creating the thread.  The minimum stack sizes have already been calculated
1389 size_t os::Posix::get_initial_stack_size(ThreadType thr_type, size_t req_stack_size) {
1390   size_t stack_size;
1391   if (req_stack_size == 0) {
1392     stack_size = default_stack_size(thr_type);
1393   } else {
1394     stack_size = req_stack_size;
1395   }
1396 
1397   switch (thr_type) {
1398   case os::java_thread:
1399     // Java threads use ThreadStackSize which default value can be
1400     // changed with the flag -Xss
1401     if (req_stack_size == 0 && JavaThread::stack_size_at_create() > 0) {
1402       // no requested size and we have a more specific default value
1403       stack_size = JavaThread::stack_size_at_create();
1404     }
1405     stack_size = MAX2(stack_size,
1406                       _java_thread_min_stack_allowed);
1407     break;
1408   case os::compiler_thread:
1409     if (req_stack_size == 0 && CompilerThreadStackSize > 0) {
1410       // no requested size and we have a more specific default value
1411       stack_size = (size_t)(CompilerThreadStackSize * K);
1412     }
1413     stack_size = MAX2(stack_size,
1414                       _compiler_thread_min_stack_allowed);
1415     break;
1416   case os::vm_thread:
1417   case os::pgc_thread:
1418   case os::cgc_thread:
1419   case os::watcher_thread:
1420   default:  // presume the unknown thr_type is a VM internal
1421     if (req_stack_size == 0 && VMThreadStackSize > 0) {
1422       // no requested size and we have a more specific default value
1423       stack_size = (size_t)(VMThreadStackSize * K);
1424     }
1425 
1426     stack_size = MAX2(stack_size,
1427                       _vm_internal_thread_min_stack_allowed);
1428     break;
1429   }
1430 
1431   // pthread_attr_setstacksize() may require that the size be rounded up to the OS page size.
1432   // Be careful not to round up to 0. Align down in that case.
1433   if (stack_size <= SIZE_MAX - vm_page_size()) {
1434     stack_size = align_up(stack_size, vm_page_size());
1435   } else {
1436     stack_size = align_down(stack_size, vm_page_size());
1437   }
1438 
1439   return stack_size;
1440 }
1441 
1442 Thread* os::ThreadCrashProtection::_protected_thread = NULL;
1443 os::ThreadCrashProtection* os::ThreadCrashProtection::_crash_protection = NULL;
1444 volatile intptr_t os::ThreadCrashProtection::_crash_mux = 0;
1445 
1446 os::ThreadCrashProtection::ThreadCrashProtection() {
1447 }
1448 
1449 /*
1450  * See the caveats for this class in os_posix.hpp
1451  * Protects the callback call so that SIGSEGV / SIGBUS jumps back into this
1452  * method and returns false. If none of the signals are raised, returns true.
1453  * The callback is supposed to provide the method that should be protected.
1454  */
1455 bool os::ThreadCrashProtection::call(os::CrashProtectionCallback& cb) {
1456   sigset_t saved_sig_mask;
1457 
1458   Thread::muxAcquire(&_crash_mux, "CrashProtection");
1459 
1460   _protected_thread = Thread::current_or_null();
1461   assert(_protected_thread != NULL, "Cannot crash protect a NULL thread");
1462 
1463   // we cannot rely on sigsetjmp/siglongjmp to save/restore the signal mask
1464   // since on at least some systems (OS X) siglongjmp will restore the mask
1465   // for the process, not the thread
1466   pthread_sigmask(0, NULL, &saved_sig_mask);
1467   if (sigsetjmp(_jmpbuf, 0) == 0) {
1468     // make sure we can see in the signal handler that we have crash protection
1469     // installed
1470     _crash_protection = this;
1471     cb.call();
1472     // and clear the crash protection
1473     _crash_protection = NULL;
1474     _protected_thread = NULL;
1475     Thread::muxRelease(&_crash_mux);
1476     return true;
1477   }
1478   // this happens when we siglongjmp() back
1479   pthread_sigmask(SIG_SETMASK, &saved_sig_mask, NULL);
1480   _crash_protection = NULL;
1481   _protected_thread = NULL;
1482   Thread::muxRelease(&_crash_mux);
1483   return false;
1484 }
1485 
1486 void os::ThreadCrashProtection::restore() {
1487   assert(_crash_protection != NULL, "must have crash protection");
1488   siglongjmp(_jmpbuf, 1);
1489 }
1490 
1491 void os::ThreadCrashProtection::check_crash_protection(int sig,
1492     Thread* thread) {
1493 
1494   if (thread != NULL &&
1495       thread == _protected_thread &&
1496       _crash_protection != NULL) {
1497 
1498     if (sig == SIGSEGV || sig == SIGBUS) {
1499       _crash_protection->restore();
1500     }
1501   }
1502 }
1503 
1504 // POSIX unamed semaphores are not supported on OS X.
1505 #ifndef __APPLE__
1506 
1507 PosixSemaphore::PosixSemaphore(uint value) {
1508   int ret = sem_init(&_semaphore, 0, value);
1509 
1510   guarantee_with_errno(ret == 0, "Failed to initialize semaphore");
1511 }
1512 
1513 PosixSemaphore::~PosixSemaphore() {
1514   sem_destroy(&_semaphore);
1515 }
1516 
1517 void PosixSemaphore::signal(uint count) {
1518   for (uint i = 0; i < count; i++) {
1519     int ret = sem_post(&_semaphore);
1520 
1521     assert_with_errno(ret == 0, "sem_post failed");
1522   }
1523 }
1524 
1525 void PosixSemaphore::wait() {
1526   int ret;
1527 
1528   do {
1529     ret = sem_wait(&_semaphore);
1530   } while (ret != 0 && errno == EINTR);
1531 
1532   assert_with_errno(ret == 0, "sem_wait failed");
1533 }
1534 
1535 bool PosixSemaphore::trywait() {
1536   int ret;
1537 
1538   do {
1539     ret = sem_trywait(&_semaphore);
1540   } while (ret != 0 && errno == EINTR);
1541 
1542   assert_with_errno(ret == 0 || errno == EAGAIN, "trywait failed");
1543 
1544   return ret == 0;
1545 }
1546 
1547 bool PosixSemaphore::timedwait(struct timespec ts) {
1548   while (true) {
1549     int result = sem_timedwait(&_semaphore, &ts);
1550     if (result == 0) {
1551       return true;
1552     } else if (errno == EINTR) {
1553       continue;
1554     } else if (errno == ETIMEDOUT) {
1555       return false;
1556     } else {
1557       assert_with_errno(false, "timedwait failed");
1558       return false;
1559     }
1560   }
1561 }
1562 
1563 #endif // __APPLE__
1564 
1565 
1566 // Shared pthread_mutex/cond based PlatformEvent implementation.
1567 // Not currently usable by Solaris.
1568 
1569 #ifndef SOLARIS
1570 
1571 // Shared condattr object for use with relative timed-waits. Will be associated
1572 // with CLOCK_MONOTONIC if available to avoid issues with time-of-day changes,
1573 // but otherwise whatever default is used by the platform - generally the
1574 // time-of-day clock.
1575 static pthread_condattr_t _condAttr[1];
1576 
1577 // Shared mutexattr to explicitly set the type to PTHREAD_MUTEX_NORMAL as not
1578 // all systems (e.g. FreeBSD) map the default to "normal".
1579 static pthread_mutexattr_t _mutexAttr[1];
1580 
1581 // common basic initialization that is always supported
1582 static void pthread_init_common(void) {
1583   int status;
1584   if ((status = pthread_condattr_init(_condAttr)) != 0) {
1585     fatal("pthread_condattr_init: %s", os::strerror(status));
1586   }
1587   if ((status = pthread_mutexattr_init(_mutexAttr)) != 0) {
1588     fatal("pthread_mutexattr_init: %s", os::strerror(status));
1589   }
1590   if ((status = pthread_mutexattr_settype(_mutexAttr, PTHREAD_MUTEX_NORMAL)) != 0) {
1591     fatal("pthread_mutexattr_settype: %s", os::strerror(status));
1592   }
1593 }
1594 
1595 // Not all POSIX types and API's are available on all notionally "posix"
1596 // platforms. If we have build-time support then we will check for actual
1597 // runtime support via dlopen/dlsym lookup. This allows for running on an
1598 // older OS version compared to the build platform. But if there is no
1599 // build time support then there cannot be any runtime support as we do not
1600 // know what the runtime types would be (for example clockid_t might be an
1601 // int or int64_t).
1602 //
1603 #ifdef SUPPORTS_CLOCK_MONOTONIC
1604 
1605 // This means we have clockid_t, clock_gettime et al and CLOCK_MONOTONIC
1606 
1607 static int (*_clock_gettime)(clockid_t, struct timespec *);
1608 static int (*_pthread_condattr_setclock)(pthread_condattr_t *, clockid_t);
1609 
1610 static bool _use_clock_monotonic_condattr;
1611 
1612 // Determine what POSIX API's are present and do appropriate
1613 // configuration.
1614 void os::Posix::init(void) {
1615 
1616   // NOTE: no logging available when this is called. Put logging
1617   // statements in init_2().
1618 
1619   // Copied from os::Linux::clock_init(). The duplication is temporary.
1620 
1621   // 1. Check for CLOCK_MONOTONIC support.
1622 
1623   void* handle = NULL;
1624 
1625   // For linux we need librt, for other OS we can find
1626   // this function in regular libc.
1627 #ifdef NEEDS_LIBRT
1628   // We do dlopen's in this particular order due to bug in linux
1629   // dynamic loader (see 6348968) leading to crash on exit.
1630   handle = dlopen("librt.so.1", RTLD_LAZY);
1631   if (handle == NULL) {
1632     handle = dlopen("librt.so", RTLD_LAZY);
1633   }
1634 #endif
1635 
1636   if (handle == NULL) {
1637     handle = RTLD_DEFAULT;
1638   }
1639 
1640   _clock_gettime = NULL;
1641 
1642   int (*clock_getres_func)(clockid_t, struct timespec*) =
1643     (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1644   int (*clock_gettime_func)(clockid_t, struct timespec*) =
1645     (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1646   if (clock_getres_func != NULL && clock_gettime_func != NULL) {
1647     // We assume that if both clock_gettime and clock_getres support
1648     // CLOCK_MONOTONIC then the OS provides true high-res monotonic clock.
1649     struct timespec res;
1650     struct timespec tp;
1651     if (clock_getres_func(CLOCK_MONOTONIC, &res) == 0 &&
1652         clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1653       // Yes, monotonic clock is supported.
1654       _clock_gettime = clock_gettime_func;
1655     } else {
1656 #ifdef NEEDS_LIBRT
1657       // Close librt if there is no monotonic clock.
1658       if (handle != RTLD_DEFAULT) {
1659         dlclose(handle);
1660       }
1661 #endif
1662     }
1663   }
1664 
1665   // 2. Check for pthread_condattr_setclock support.
1666 
1667   _pthread_condattr_setclock = NULL;
1668 
1669   // libpthread is already loaded.
1670   int (*condattr_setclock_func)(pthread_condattr_t*, clockid_t) =
1671     (int (*)(pthread_condattr_t*, clockid_t))dlsym(RTLD_DEFAULT,
1672                                                    "pthread_condattr_setclock");
1673   if (condattr_setclock_func != NULL) {
1674     _pthread_condattr_setclock = condattr_setclock_func;
1675   }
1676 
1677   // Now do general initialization.
1678 
1679   pthread_init_common();
1680 
1681   int status;
1682   if (_pthread_condattr_setclock != NULL && _clock_gettime != NULL) {
1683     if ((status = _pthread_condattr_setclock(_condAttr, CLOCK_MONOTONIC)) != 0) {
1684       if (status == EINVAL) {
1685         _use_clock_monotonic_condattr = false;
1686         warning("Unable to use monotonic clock with relative timed-waits" \
1687                 " - changes to the time-of-day clock may have adverse affects");
1688       } else {
1689         fatal("pthread_condattr_setclock: %s", os::strerror(status));
1690       }
1691     } else {
1692       _use_clock_monotonic_condattr = true;
1693     }
1694   } else {
1695     _use_clock_monotonic_condattr = false;
1696   }
1697 }
1698 
1699 void os::Posix::init_2(void) {
1700   log_info(os)("Use of CLOCK_MONOTONIC is%s supported",
1701                (_clock_gettime != NULL ? "" : " not"));
1702   log_info(os)("Use of pthread_condattr_setclock is%s supported",
1703                (_pthread_condattr_setclock != NULL ? "" : " not"));
1704   log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with %s",
1705                _use_clock_monotonic_condattr ? "CLOCK_MONOTONIC" : "the default clock");
1706 }
1707 
1708 #else // !SUPPORTS_CLOCK_MONOTONIC
1709 
1710 void os::Posix::init(void) {
1711   pthread_init_common();
1712 }
1713 
1714 void os::Posix::init_2(void) {
1715   log_info(os)("Use of CLOCK_MONOTONIC is not supported");
1716   log_info(os)("Use of pthread_condattr_setclock is not supported");
1717   log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with the default clock");
1718 }
1719 
1720 #endif // SUPPORTS_CLOCK_MONOTONIC
1721 
1722 os::PlatformEvent::PlatformEvent() {
1723   int status = pthread_cond_init(_cond, _condAttr);
1724   assert_status(status == 0, status, "cond_init");
1725   status = pthread_mutex_init(_mutex, _mutexAttr);
1726   assert_status(status == 0, status, "mutex_init");
1727   _event   = 0;
1728   _nParked = 0;
1729 }
1730 
1731 // Utility to convert the given timeout to an absolute timespec
1732 // (based on the appropriate clock) to use with pthread_cond_timewait.
1733 // The clock queried here must be the clock used to manage the
1734 // timeout of the condition variable.
1735 //
1736 // The passed in timeout value is either a relative time in nanoseconds
1737 // or an absolute time in milliseconds. A relative timeout will be
1738 // associated with CLOCK_MONOTONIC if available; otherwise, or if absolute,
1739 // the default time-of-day clock will be used.
1740 
1741 // Given time is a 64-bit value and the time_t used in the timespec is
1742 // sometimes a signed-32-bit value we have to watch for overflow if times
1743 // way in the future are given. Further on Solaris versions
1744 // prior to 10 there is a restriction (see cond_timedwait) that the specified
1745 // number of seconds, in abstime, is less than current_time + 100000000.
1746 // As it will be over 20 years before "now + 100000000" will overflow we can
1747 // ignore overflow and just impose a hard-limit on seconds using the value
1748 // of "now + 100000000". This places a limit on the timeout of about 3.17
1749 // years from "now".
1750 //
1751 #define MAX_SECS 100000000
1752 
1753 // Calculate a new absolute time that is "timeout" nanoseconds from "now".
1754 // "unit" indicates the unit of "now_part_sec" (may be nanos or micros depending
1755 // on which clock is being used).
1756 static void calc_rel_time(timespec* abstime, jlong timeout, jlong now_sec,
1757                           jlong now_part_sec, jlong unit) {
1758   time_t max_secs = now_sec + MAX_SECS;
1759 
1760   jlong seconds = timeout / NANOUNITS;
1761   timeout %= NANOUNITS; // remaining nanos
1762 
1763   if (seconds >= MAX_SECS) {
1764     // More seconds than we can add, so pin to max_secs.
1765     abstime->tv_sec = max_secs;
1766     abstime->tv_nsec = 0;
1767   } else {
1768     abstime->tv_sec = now_sec  + seconds;
1769     long nanos = (now_part_sec * (NANOUNITS / unit)) + timeout;
1770     if (nanos >= NANOUNITS) { // overflow
1771       abstime->tv_sec += 1;
1772       nanos -= NANOUNITS;
1773     }
1774     abstime->tv_nsec = nanos;
1775   }
1776 }
1777 
1778 // Unpack the given deadline in milliseconds since the epoch, into the given timespec.
1779 // The current time in seconds is also passed in to enforce an upper bound as discussed above.
1780 static void unpack_abs_time(timespec* abstime, jlong deadline, jlong now_sec) {
1781   time_t max_secs = now_sec + MAX_SECS;
1782 
1783   jlong seconds = deadline / MILLIUNITS;
1784   jlong millis = deadline % MILLIUNITS;
1785 
1786   if (seconds >= max_secs) {
1787     // Absolute seconds exceeds allowed max, so pin to max_secs.
1788     abstime->tv_sec = max_secs;
1789     abstime->tv_nsec = 0;
1790   } else {
1791     abstime->tv_sec = seconds;
1792     abstime->tv_nsec = millis * (NANOUNITS / MILLIUNITS);
1793   }
1794 }
1795 
1796 static void to_abstime(timespec* abstime, jlong timeout, bool isAbsolute) {
1797   DEBUG_ONLY(int max_secs = MAX_SECS;)
1798 
1799   if (timeout < 0) {
1800     timeout = 0;
1801   }
1802 
1803 #ifdef SUPPORTS_CLOCK_MONOTONIC
1804 
1805   if (_use_clock_monotonic_condattr && !isAbsolute) {
1806     struct timespec now;
1807     int status = _clock_gettime(CLOCK_MONOTONIC, &now);
1808     assert_status(status == 0, status, "clock_gettime");
1809     calc_rel_time(abstime, timeout, now.tv_sec, now.tv_nsec, NANOUNITS);
1810     DEBUG_ONLY(max_secs += now.tv_sec;)
1811   } else {
1812 
1813 #else
1814 
1815   { // Match the block scope.
1816 
1817 #endif // SUPPORTS_CLOCK_MONOTONIC
1818 
1819     // Time-of-day clock is all we can reliably use.
1820     struct timeval now;
1821     int status = gettimeofday(&now, NULL);
1822     assert_status(status == 0, errno, "gettimeofday");
1823     if (isAbsolute) {
1824       unpack_abs_time(abstime, timeout, now.tv_sec);
1825     } else {
1826       calc_rel_time(abstime, timeout, now.tv_sec, now.tv_usec, MICROUNITS);
1827     }
1828     DEBUG_ONLY(max_secs += now.tv_sec;)
1829   }
1830 
1831   assert(abstime->tv_sec >= 0, "tv_sec < 0");
1832   assert(abstime->tv_sec <= max_secs, "tv_sec > max_secs");
1833   assert(abstime->tv_nsec >= 0, "tv_nsec < 0");
1834   assert(abstime->tv_nsec < NANOUNITS, "tv_nsec >= NANOUNITS");
1835 }
1836 
1837 // PlatformEvent
1838 //
1839 // Assumption:
1840 //    Only one parker can exist on an event, which is why we allocate
1841 //    them per-thread. Multiple unparkers can coexist.
1842 //
1843 // _event serves as a restricted-range semaphore.
1844 //   -1 : thread is blocked, i.e. there is a waiter
1845 //    0 : neutral: thread is running or ready,
1846 //        could have been signaled after a wait started
1847 //    1 : signaled - thread is running or ready
1848 //
1849 //    Having three states allows for some detection of bad usage - see
1850 //    comments on unpark().
1851 
1852 void os::PlatformEvent::park() {       // AKA "down()"
1853   // Transitions for _event:
1854   //   -1 => -1 : illegal
1855   //    1 =>  0 : pass - return immediately
1856   //    0 => -1 : block; then set _event to 0 before returning
1857 
1858   // Invariant: Only the thread associated with the PlatformEvent
1859   // may call park().
1860   assert(_nParked == 0, "invariant");
1861 
1862   int v;
1863 
1864   // atomically decrement _event
1865   for (;;) {
1866     v = _event;
1867     if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1868   }
1869   guarantee(v >= 0, "invariant");
1870 
1871   if (v == 0) { // Do this the hard way by blocking ...
1872     int status = pthread_mutex_lock(_mutex);
1873     assert_status(status == 0, status, "mutex_lock");
1874     guarantee(_nParked == 0, "invariant");
1875     ++_nParked;
1876     while (_event < 0) {
1877       // OS-level "spurious wakeups" are ignored
1878       status = pthread_cond_wait(_cond, _mutex);
1879       assert_status(status == 0, status, "cond_wait");
1880     }
1881     --_nParked;
1882 
1883     _event = 0;
1884     status = pthread_mutex_unlock(_mutex);
1885     assert_status(status == 0, status, "mutex_unlock");
1886     // Paranoia to ensure our locked and lock-free paths interact
1887     // correctly with each other.
1888     OrderAccess::fence();
1889   }
1890   guarantee(_event >= 0, "invariant");
1891 }
1892 
1893 int os::PlatformEvent::park(jlong millis) {
1894   // Transitions for _event:
1895   //   -1 => -1 : illegal
1896   //    1 =>  0 : pass - return immediately
1897   //    0 => -1 : block; then set _event to 0 before returning
1898 
1899   // Invariant: Only the thread associated with the Event/PlatformEvent
1900   // may call park().
1901   assert(_nParked == 0, "invariant");
1902 
1903   int v;
1904   // atomically decrement _event
1905   for (;;) {
1906     v = _event;
1907     if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1908   }
1909   guarantee(v >= 0, "invariant");
1910 
1911   if (v == 0) { // Do this the hard way by blocking ...
1912     struct timespec abst;
1913     // We have to watch for overflow when converting millis to nanos,
1914     // but if millis is that large then we will end up limiting to
1915     // MAX_SECS anyway, so just do that here.
1916     if (millis / MILLIUNITS > MAX_SECS) {
1917       millis = jlong(MAX_SECS) * MILLIUNITS;
1918     }
1919     to_abstime(&abst, millis * (NANOUNITS / MILLIUNITS), false);
1920 
1921     int ret = OS_TIMEOUT;
1922     int status = pthread_mutex_lock(_mutex);
1923     assert_status(status == 0, status, "mutex_lock");
1924     guarantee(_nParked == 0, "invariant");
1925     ++_nParked;
1926 
1927     while (_event < 0) {
1928       status = pthread_cond_timedwait(_cond, _mutex, &abst);
1929       assert_status(status == 0 || status == ETIMEDOUT,
1930                     status, "cond_timedwait");
1931       // OS-level "spurious wakeups" are ignored unless the archaic
1932       // FilterSpuriousWakeups is set false. That flag should be obsoleted.
1933       if (!FilterSpuriousWakeups) break;
1934       if (status == ETIMEDOUT) break;
1935     }
1936     --_nParked;
1937 
1938     if (_event >= 0) {
1939       ret = OS_OK;
1940     }
1941 
1942     _event = 0;
1943     status = pthread_mutex_unlock(_mutex);
1944     assert_status(status == 0, status, "mutex_unlock");
1945     // Paranoia to ensure our locked and lock-free paths interact
1946     // correctly with each other.
1947     OrderAccess::fence();
1948     return ret;
1949   }
1950   return OS_OK;
1951 }
1952 
1953 void os::PlatformEvent::unpark() {
1954   // Transitions for _event:
1955   //    0 => 1 : just return
1956   //    1 => 1 : just return
1957   //   -1 => either 0 or 1; must signal target thread
1958   //         That is, we can safely transition _event from -1 to either
1959   //         0 or 1.
1960   // See also: "Semaphores in Plan 9" by Mullender & Cox
1961   //
1962   // Note: Forcing a transition from "-1" to "1" on an unpark() means
1963   // that it will take two back-to-back park() calls for the owning
1964   // thread to block. This has the benefit of forcing a spurious return
1965   // from the first park() call after an unpark() call which will help
1966   // shake out uses of park() and unpark() without checking state conditions
1967   // properly. This spurious return doesn't manifest itself in any user code
1968   // but only in the correctly written condition checking loops of ObjectMonitor,
1969   // Mutex/Monitor, Thread::muxAcquire and os::sleep
1970 
1971   if (Atomic::xchg(1, &_event) >= 0) return;
1972 
1973   int status = pthread_mutex_lock(_mutex);
1974   assert_status(status == 0, status, "mutex_lock");
1975   int anyWaiters = _nParked;
1976   assert(anyWaiters == 0 || anyWaiters == 1, "invariant");
1977   status = pthread_mutex_unlock(_mutex);
1978   assert_status(status == 0, status, "mutex_unlock");
1979 
1980   // Note that we signal() *after* dropping the lock for "immortal" Events.
1981   // This is safe and avoids a common class of futile wakeups.  In rare
1982   // circumstances this can cause a thread to return prematurely from
1983   // cond_{timed}wait() but the spurious wakeup is benign and the victim
1984   // will simply re-test the condition and re-park itself.
1985   // This provides particular benefit if the underlying platform does not
1986   // provide wait morphing.
1987 
1988   if (anyWaiters != 0) {
1989     status = pthread_cond_signal(_cond);
1990     assert_status(status == 0, status, "cond_signal");
1991   }
1992 }
1993 
1994 // JSR166 support
1995 
1996  os::PlatformParker::PlatformParker() {
1997   int status;
1998   status = pthread_cond_init(&_cond[REL_INDEX], _condAttr);
1999   assert_status(status == 0, status, "cond_init rel");
2000   status = pthread_cond_init(&_cond[ABS_INDEX], NULL);
2001   assert_status(status == 0, status, "cond_init abs");
2002   status = pthread_mutex_init(_mutex, _mutexAttr);
2003   assert_status(status == 0, status, "mutex_init");
2004   _cur_index = -1; // mark as unused
2005 }
2006 
2007 // Parker::park decrements count if > 0, else does a condvar wait.  Unpark
2008 // sets count to 1 and signals condvar.  Only one thread ever waits
2009 // on the condvar. Contention seen when trying to park implies that someone
2010 // is unparking you, so don't wait. And spurious returns are fine, so there
2011 // is no need to track notifications.
2012 
2013 void Parker::park(bool isAbsolute, jlong time) {
2014 
2015   // Optional fast-path check:
2016   // Return immediately if a permit is available.
2017   // We depend on Atomic::xchg() having full barrier semantics
2018   // since we are doing a lock-free update to _counter.
2019   if (Atomic::xchg(0, &_counter) > 0) return;
2020 
2021   Thread* thread = Thread::current();
2022   assert(thread->is_Java_thread(), "Must be JavaThread");
2023   JavaThread *jt = (JavaThread *)thread;
2024 
2025   // Optional optimization -- avoid state transitions if there's
2026   // an interrupt pending.
2027   if (Thread::is_interrupted(thread, false)) {
2028     return;
2029   }
2030 
2031   // Next, demultiplex/decode time arguments
2032   struct timespec absTime;
2033   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
2034     return;
2035   }
2036   if (time > 0) {
2037     to_abstime(&absTime, time, isAbsolute);
2038   }
2039 
2040   // Enter safepoint region
2041   // Beware of deadlocks such as 6317397.
2042   // The per-thread Parker:: mutex is a classic leaf-lock.
2043   // In particular a thread must never block on the Threads_lock while
2044   // holding the Parker:: mutex.  If safepoints are pending both the
2045   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
2046   ThreadBlockInVM tbivm(jt);
2047 
2048   // Don't wait if cannot get lock since interference arises from
2049   // unparking. Also re-check interrupt before trying wait.
2050   if (Thread::is_interrupted(thread, false) ||
2051       pthread_mutex_trylock(_mutex) != 0) {
2052     return;
2053   }
2054 
2055   int status;
2056   if (_counter > 0)  { // no wait needed
2057     _counter = 0;
2058     status = pthread_mutex_unlock(_mutex);
2059     assert_status(status == 0, status, "invariant");
2060     // Paranoia to ensure our locked and lock-free paths interact
2061     // correctly with each other and Java-level accesses.
2062     OrderAccess::fence();
2063     return;
2064   }
2065 
2066   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2067   jt->set_suspend_equivalent();
2068   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2069 
2070   assert(_cur_index == -1, "invariant");
2071   if (time == 0) {
2072     _cur_index = REL_INDEX; // arbitrary choice when not timed
2073     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
2074     assert_status(status == 0, status, "cond_timedwait");
2075   }
2076   else {
2077     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
2078     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
2079     assert_status(status == 0 || status == ETIMEDOUT,
2080                   status, "cond_timedwait");
2081   }
2082   _cur_index = -1;
2083 
2084   _counter = 0;
2085   status = pthread_mutex_unlock(_mutex);
2086   assert_status(status == 0, status, "invariant");
2087   // Paranoia to ensure our locked and lock-free paths interact
2088   // correctly with each other and Java-level accesses.
2089   OrderAccess::fence();
2090 
2091   // If externally suspended while waiting, re-suspend
2092   if (jt->handle_special_suspend_equivalent_condition()) {
2093     jt->java_suspend_self();
2094   }
2095 }
2096 
2097 void Parker::unpark() {
2098   int status = pthread_mutex_lock(_mutex);
2099   assert_status(status == 0, status, "invariant");
2100   const int s = _counter;
2101   _counter = 1;
2102   // must capture correct index before unlocking
2103   int index = _cur_index;
2104   status = pthread_mutex_unlock(_mutex);
2105   assert_status(status == 0, status, "invariant");
2106 
2107   // Note that we signal() *after* dropping the lock for "immortal" Events.
2108   // This is safe and avoids a common class of futile wakeups.  In rare
2109   // circumstances this can cause a thread to return prematurely from
2110   // cond_{timed}wait() but the spurious wakeup is benign and the victim
2111   // will simply re-test the condition and re-park itself.
2112   // This provides particular benefit if the underlying platform does not
2113   // provide wait morphing.
2114 
2115   if (s < 1 && index != -1) {
2116     // thread is definitely parked
2117     status = pthread_cond_signal(&_cond[index]);
2118     assert_status(status == 0, status, "invariant");
2119   }
2120 }
2121 
2122 
2123 #endif // !SOLARIS