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