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 "jvm.h"
  26 #include "utilities/globalDefinitions.hpp"
  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 NOT_AIX( | 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   int ret = util_posix_fallocate(fd, 0, (off_t)size);
 247   if (ret != 0) {
 248     vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory. error(%d)", ret));
 249     return NULL;
 250   }
 251 
 252   int prot = PROT_READ | PROT_WRITE;
 253   int flags = MAP_SHARED;
 254   if (base != NULL) {
 255     flags |= MAP_FIXED;
 256   }
 257   char* addr = (char*)mmap(base, size, prot, flags, fd, 0);
 258 
 259   if (addr == MAP_FAILED) {
 260     warning("Failed mmap to file. (%s)", os::strerror(errno));
 261     return NULL;
 262   }
 263   if (base != NULL && addr != base) {
 264     if (!os::release_memory(addr, size)) {
 265       warning("Could not release memory on unsuccessful file mapping");
 266     }
 267     return NULL;
 268   }
 269   return addr;
 270 }
 271 
 272 char* os::replace_existing_mapping_with_file_mapping(char* base, size_t size, int fd) {
 273   assert(fd != -1, "File descriptor is not valid");
 274   assert(base != NULL, "Base cannot be NULL");
 275 
 276   return map_memory_to_file(base, size, fd);
 277 }
 278 
 279 // Multiple threads can race in this code, and can remap over each other with MAP_FIXED,
 280 // so on posix, unmap the section at the start and at the end of the chunk that we mapped
 281 // rather than unmapping and remapping the whole chunk to get requested alignment.
 282 char* os::reserve_memory_aligned(size_t size, size_t alignment, int file_desc) {
 283   assert((alignment & (os::vm_allocation_granularity() - 1)) == 0,
 284       "Alignment must be a multiple of allocation granularity (page size)");
 285   assert((size & (alignment -1)) == 0, "size must be 'alignment' aligned");
 286 
 287   size_t extra_size = size + alignment;
 288   assert(extra_size >= size, "overflow, size is too large to allow alignment");
 289 
 290   char* extra_base;
 291   if (file_desc != -1) {
 292     // For file mapping, we do not call os:reserve_memory(extra_size, NULL, alignment, file_desc) because
 293     // we need to deal with shrinking of the file space later when we release extra memory after alignment.
 294     // We also cannot called os:reserve_memory() with file_desc set to -1 because on aix we might get SHM memory.
 295     // So here to call a helper function while reserve memory for us. After we have a aligned base,
 296     // we will replace anonymous mapping with file mapping.
 297     extra_base = reserve_mmapped_memory(extra_size, NULL);
 298     if (extra_base != NULL) {
 299       MemTracker::record_virtual_memory_reserve((address)extra_base, extra_size, CALLER_PC);
 300     }
 301   } else {
 302     extra_base = os::reserve_memory(extra_size, NULL, alignment);
 303   }
 304 
 305   if (extra_base == NULL) {
 306     return NULL;
 307   }
 308 
 309   // Do manual alignment
 310   char* aligned_base = align_up(extra_base, alignment);
 311 
 312   // [  |                                       |  ]
 313   // ^ extra_base
 314   //    ^ extra_base + begin_offset == aligned_base
 315   //     extra_base + begin_offset + size       ^
 316   //                       extra_base + extra_size ^
 317   // |<>| == begin_offset
 318   //                              end_offset == |<>|
 319   size_t begin_offset = aligned_base - extra_base;
 320   size_t end_offset = (extra_base + extra_size) - (aligned_base + size);
 321 
 322   if (begin_offset > 0) {
 323       os::release_memory(extra_base, begin_offset);
 324   }
 325 
 326   if (end_offset > 0) {
 327       os::release_memory(extra_base + begin_offset + size, end_offset);
 328   }
 329 
 330   if (file_desc != -1) {
 331     // After we have an aligned address, we can replace anonymous mapping with file mapping
 332     if (replace_existing_mapping_with_file_mapping(aligned_base, size, file_desc) == NULL) {
 333       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
 334     }
 335     MemTracker::record_virtual_memory_commit((address)aligned_base, size, CALLER_PC);
 336   }
 337   return aligned_base;
 338 }
 339 
 340 int os::log_vsnprintf(char* buf, size_t len, const char* fmt, va_list args) {
 341     return vsnprintf(buf, len, fmt, args);
 342 }
 343 
 344 int os::get_fileno(FILE* fp) {
 345   return NOT_AIX(::)fileno(fp);
 346 }
 347 
 348 struct tm* os::gmtime_pd(const time_t* clock, struct tm*  res) {
 349   return gmtime_r(clock, res);
 350 }
 351 
 352 void os::Posix::print_load_average(outputStream* st) {
 353   st->print("load average:");
 354   double loadavg[3];
 355   os::loadavg(loadavg, 3);
 356   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
 357   st->cr();
 358 }
 359 
 360 void os::Posix::print_rlimit_info(outputStream* st) {
 361   st->print("rlimit:");
 362   struct rlimit rlim;
 363 
 364   st->print(" STACK ");
 365   getrlimit(RLIMIT_STACK, &rlim);
 366   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 367   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 368 
 369   st->print(", CORE ");
 370   getrlimit(RLIMIT_CORE, &rlim);
 371   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 372   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 373 
 374   // Isn't there on solaris
 375 #if defined(AIX)
 376   st->print(", NPROC ");
 377   st->print("%d", sysconf(_SC_CHILD_MAX));
 378 #elif !defined(SOLARIS)
 379   st->print(", NPROC ");
 380   getrlimit(RLIMIT_NPROC, &rlim);
 381   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 382   else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
 383 #endif
 384 
 385   st->print(", NOFILE ");
 386   getrlimit(RLIMIT_NOFILE, &rlim);
 387   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 388   else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
 389 
 390   st->print(", AS ");
 391   getrlimit(RLIMIT_AS, &rlim);
 392   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 393   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 394 
 395   st->print(", DATA ");
 396   getrlimit(RLIMIT_DATA, &rlim);
 397   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 398   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 399 
 400   st->print(", FSIZE ");
 401   getrlimit(RLIMIT_FSIZE, &rlim);
 402   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 403   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 404 
 405   st->cr();
 406 }
 407 
 408 void os::Posix::print_uname_info(outputStream* st) {
 409   // kernel
 410   st->print("uname:");
 411   struct utsname name;
 412   uname(&name);
 413   st->print("%s ", name.sysname);
 414 #ifdef ASSERT
 415   st->print("%s ", name.nodename);
 416 #endif
 417   st->print("%s ", name.release);
 418   st->print("%s ", name.version);
 419   st->print("%s", name.machine);
 420   st->cr();
 421 }
 422 
 423 bool os::get_host_name(char* buf, size_t buflen) {
 424   struct utsname name;
 425   uname(&name);
 426   jio_snprintf(buf, buflen, "%s", name.nodename);
 427   return true;
 428 }
 429 
 430 bool os::has_allocatable_memory_limit(julong* limit) {
 431   struct rlimit rlim;
 432   int getrlimit_res = getrlimit(RLIMIT_AS, &rlim);
 433   // if there was an error when calling getrlimit, assume that there is no limitation
 434   // on virtual memory.
 435   bool result;
 436   if ((getrlimit_res != 0) || (rlim.rlim_cur == RLIM_INFINITY)) {
 437     result = false;
 438   } else {
 439     *limit = (julong)rlim.rlim_cur;
 440     result = true;
 441   }
 442 #ifdef _LP64
 443   return result;
 444 #else
 445   // arbitrary virtual space limit for 32 bit Unices found by testing. If
 446   // getrlimit above returned a limit, bound it with this limit. Otherwise
 447   // directly use it.
 448   const julong max_virtual_limit = (julong)3800*M;
 449   if (result) {
 450     *limit = MIN2(*limit, max_virtual_limit);
 451   } else {
 452     *limit = max_virtual_limit;
 453   }
 454 
 455   // bound by actually allocatable memory. The algorithm uses two bounds, an
 456   // upper and a lower limit. The upper limit is the current highest amount of
 457   // memory that could not be allocated, the lower limit is the current highest
 458   // amount of memory that could be allocated.
 459   // The algorithm iteratively refines the result by halving the difference
 460   // between these limits, updating either the upper limit (if that value could
 461   // not be allocated) or the lower limit (if the that value could be allocated)
 462   // until the difference between these limits is "small".
 463 
 464   // the minimum amount of memory we care about allocating.
 465   const julong min_allocation_size = M;
 466 
 467   julong upper_limit = *limit;
 468 
 469   // first check a few trivial cases
 470   if (is_allocatable(upper_limit) || (upper_limit <= min_allocation_size)) {
 471     *limit = upper_limit;
 472   } else if (!is_allocatable(min_allocation_size)) {
 473     // we found that not even min_allocation_size is allocatable. Return it
 474     // anyway. There is no point to search for a better value any more.
 475     *limit = min_allocation_size;
 476   } else {
 477     // perform the binary search.
 478     julong lower_limit = min_allocation_size;
 479     while ((upper_limit - lower_limit) > min_allocation_size) {
 480       julong temp_limit = ((upper_limit - lower_limit) / 2) + lower_limit;
 481       temp_limit = align_down(temp_limit, min_allocation_size);
 482       if (is_allocatable(temp_limit)) {
 483         lower_limit = temp_limit;
 484       } else {
 485         upper_limit = temp_limit;
 486       }
 487     }
 488     *limit = lower_limit;
 489   }
 490   return true;
 491 #endif
 492 }
 493 
 494 const char* os::get_current_directory(char *buf, size_t buflen) {
 495   return getcwd(buf, buflen);
 496 }
 497 
 498 FILE* os::open(int fd, const char* mode) {
 499   return ::fdopen(fd, mode);
 500 }
 501 
 502 void os::flockfile(FILE* fp) {
 503   ::flockfile(fp);
 504 }
 505 
 506 void os::funlockfile(FILE* fp) {
 507   ::funlockfile(fp);
 508 }
 509 
 510 // Builds a platform dependent Agent_OnLoad_<lib_name> function name
 511 // which is used to find statically linked in agents.
 512 // Parameters:
 513 //            sym_name: Symbol in library we are looking for
 514 //            lib_name: Name of library to look in, NULL for shared libs.
 515 //            is_absolute_path == true if lib_name is absolute path to agent
 516 //                                     such as "/a/b/libL.so"
 517 //            == false if only the base name of the library is passed in
 518 //               such as "L"
 519 char* os::build_agent_function_name(const char *sym_name, const char *lib_name,
 520                                     bool is_absolute_path) {
 521   char *agent_entry_name;
 522   size_t len;
 523   size_t name_len;
 524   size_t prefix_len = strlen(JNI_LIB_PREFIX);
 525   size_t suffix_len = strlen(JNI_LIB_SUFFIX);
 526   const char *start;
 527 
 528   if (lib_name != NULL) {
 529     name_len = strlen(lib_name);
 530     if (is_absolute_path) {
 531       // Need to strip path, prefix and suffix
 532       if ((start = strrchr(lib_name, *os::file_separator())) != NULL) {
 533         lib_name = ++start;
 534       }
 535       if (strlen(lib_name) <= (prefix_len + suffix_len)) {
 536         return NULL;
 537       }
 538       lib_name += prefix_len;
 539       name_len = strlen(lib_name) - suffix_len;
 540     }
 541   }
 542   len = (lib_name != NULL ? name_len : 0) + strlen(sym_name) + 2;
 543   agent_entry_name = NEW_C_HEAP_ARRAY_RETURN_NULL(char, len, mtThread);
 544   if (agent_entry_name == NULL) {
 545     return NULL;
 546   }
 547   strcpy(agent_entry_name, sym_name);
 548   if (lib_name != NULL) {
 549     strcat(agent_entry_name, "_");
 550     strncat(agent_entry_name, lib_name, name_len);
 551   }
 552   return agent_entry_name;
 553 }
 554 
 555 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
 556   assert(thread == Thread::current(),  "thread consistency check");
 557 
 558   ParkEvent * const slp = thread->_SleepEvent ;
 559   slp->reset() ;
 560   OrderAccess::fence() ;
 561 
 562   if (interruptible) {
 563     jlong prevtime = javaTimeNanos();
 564 
 565     for (;;) {
 566       if (os::is_interrupted(thread, true)) {
 567         return OS_INTRPT;
 568       }
 569 
 570       jlong newtime = javaTimeNanos();
 571 
 572       if (newtime - prevtime < 0) {
 573         // time moving backwards, should only happen if no monotonic clock
 574         // not a guarantee() because JVM should not abort on kernel/glibc bugs
 575         assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected in os::sleep(interruptible)");
 576       } else {
 577         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
 578       }
 579 
 580       if (millis <= 0) {
 581         return OS_OK;
 582       }
 583 
 584       prevtime = newtime;
 585 
 586       {
 587         assert(thread->is_Java_thread(), "sanity check");
 588         JavaThread *jt = (JavaThread *) thread;
 589         ThreadBlockInVM tbivm(jt);
 590         OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
 591 
 592         jt->set_suspend_equivalent();
 593         // cleared by handle_special_suspend_equivalent_condition() or
 594         // java_suspend_self() via check_and_wait_while_suspended()
 595 
 596         slp->park(millis);
 597 
 598         // were we externally suspended while we were waiting?
 599         jt->check_and_wait_while_suspended();
 600       }
 601     }
 602   } else {
 603     OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
 604     jlong prevtime = javaTimeNanos();
 605 
 606     for (;;) {
 607       // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
 608       // the 1st iteration ...
 609       jlong newtime = javaTimeNanos();
 610 
 611       if (newtime - prevtime < 0) {
 612         // time moving backwards, should only happen if no monotonic clock
 613         // not a guarantee() because JVM should not abort on kernel/glibc bugs
 614         assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected on os::sleep(!interruptible)");
 615       } else {
 616         millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
 617       }
 618 
 619       if (millis <= 0) break ;
 620 
 621       prevtime = newtime;
 622       slp->park(millis);
 623     }
 624     return OS_OK ;
 625   }
 626 }
 627 
 628 ////////////////////////////////////////////////////////////////////////////////
 629 // interrupt support
 630 
 631 void os::interrupt(Thread* thread) {
 632   debug_only(Thread::check_for_dangling_thread_pointer(thread);)
 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 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
 655   debug_only(Thread::check_for_dangling_thread_pointer(thread);)
 656 
 657   OSThread* osthread = thread->osthread();
 658 
 659   bool interrupted = osthread->interrupted();
 660 
 661   // NOTE that since there is no "lock" around the interrupt and
 662   // is_interrupted operations, there is the possibility that the
 663   // interrupted flag (in osThread) will be "false" but that the
 664   // low-level events will be in the signaled state. This is
 665   // intentional. The effect of this is that Object.wait() and
 666   // LockSupport.park() will appear to have a spurious wakeup, which
 667   // is allowed and not harmful, and the possibility is so rare that
 668   // it is not worth the added complexity to add yet another lock.
 669   // For the sleep event an explicit reset is performed on entry
 670   // to os::sleep, so there is no early return. It has also been
 671   // recommended not to put the interrupted flag into the "event"
 672   // structure because it hides the issue.
 673   if (interrupted && clear_interrupted) {
 674     osthread->set_interrupted(false);
 675     // consider thread->_SleepEvent->reset() ... optional optimization
 676   }
 677 
 678   return interrupted;
 679 }
 680 
 681 
 682 
 683 static const struct {
 684   int sig; const char* name;
 685 }
 686  g_signal_info[] =
 687   {
 688   {  SIGABRT,     "SIGABRT" },
 689 #ifdef SIGAIO
 690   {  SIGAIO,      "SIGAIO" },
 691 #endif
 692   {  SIGALRM,     "SIGALRM" },
 693 #ifdef SIGALRM1
 694   {  SIGALRM1,    "SIGALRM1" },
 695 #endif
 696   {  SIGBUS,      "SIGBUS" },
 697 #ifdef SIGCANCEL
 698   {  SIGCANCEL,   "SIGCANCEL" },
 699 #endif
 700   {  SIGCHLD,     "SIGCHLD" },
 701 #ifdef SIGCLD
 702   {  SIGCLD,      "SIGCLD" },
 703 #endif
 704   {  SIGCONT,     "SIGCONT" },
 705 #ifdef SIGCPUFAIL
 706   {  SIGCPUFAIL,  "SIGCPUFAIL" },
 707 #endif
 708 #ifdef SIGDANGER
 709   {  SIGDANGER,   "SIGDANGER" },
 710 #endif
 711 #ifdef SIGDIL
 712   {  SIGDIL,      "SIGDIL" },
 713 #endif
 714 #ifdef SIGEMT
 715   {  SIGEMT,      "SIGEMT" },
 716 #endif
 717   {  SIGFPE,      "SIGFPE" },
 718 #ifdef SIGFREEZE
 719   {  SIGFREEZE,   "SIGFREEZE" },
 720 #endif
 721 #ifdef SIGGFAULT
 722   {  SIGGFAULT,   "SIGGFAULT" },
 723 #endif
 724 #ifdef SIGGRANT
 725   {  SIGGRANT,    "SIGGRANT" },
 726 #endif
 727   {  SIGHUP,      "SIGHUP" },
 728   {  SIGILL,      "SIGILL" },
 729   {  SIGINT,      "SIGINT" },
 730 #ifdef SIGIO
 731   {  SIGIO,       "SIGIO" },
 732 #endif
 733 #ifdef SIGIOINT
 734   {  SIGIOINT,    "SIGIOINT" },
 735 #endif
 736 #ifdef SIGIOT
 737 // SIGIOT is there for BSD compatibility, but on most Unices just a
 738 // synonym for SIGABRT. The result should be "SIGABRT", not
 739 // "SIGIOT".
 740 #if (SIGIOT != SIGABRT )
 741   {  SIGIOT,      "SIGIOT" },
 742 #endif
 743 #endif
 744 #ifdef SIGKAP
 745   {  SIGKAP,      "SIGKAP" },
 746 #endif
 747   {  SIGKILL,     "SIGKILL" },
 748 #ifdef SIGLOST
 749   {  SIGLOST,     "SIGLOST" },
 750 #endif
 751 #ifdef SIGLWP
 752   {  SIGLWP,      "SIGLWP" },
 753 #endif
 754 #ifdef SIGLWPTIMER
 755   {  SIGLWPTIMER, "SIGLWPTIMER" },
 756 #endif
 757 #ifdef SIGMIGRATE
 758   {  SIGMIGRATE,  "SIGMIGRATE" },
 759 #endif
 760 #ifdef SIGMSG
 761   {  SIGMSG,      "SIGMSG" },
 762 #endif
 763   {  SIGPIPE,     "SIGPIPE" },
 764 #ifdef SIGPOLL
 765   {  SIGPOLL,     "SIGPOLL" },
 766 #endif
 767 #ifdef SIGPRE
 768   {  SIGPRE,      "SIGPRE" },
 769 #endif
 770   {  SIGPROF,     "SIGPROF" },
 771 #ifdef SIGPTY
 772   {  SIGPTY,      "SIGPTY" },
 773 #endif
 774 #ifdef SIGPWR
 775   {  SIGPWR,      "SIGPWR" },
 776 #endif
 777   {  SIGQUIT,     "SIGQUIT" },
 778 #ifdef SIGRECONFIG
 779   {  SIGRECONFIG, "SIGRECONFIG" },
 780 #endif
 781 #ifdef SIGRECOVERY
 782   {  SIGRECOVERY, "SIGRECOVERY" },
 783 #endif
 784 #ifdef SIGRESERVE
 785   {  SIGRESERVE,  "SIGRESERVE" },
 786 #endif
 787 #ifdef SIGRETRACT
 788   {  SIGRETRACT,  "SIGRETRACT" },
 789 #endif
 790 #ifdef SIGSAK
 791   {  SIGSAK,      "SIGSAK" },
 792 #endif
 793   {  SIGSEGV,     "SIGSEGV" },
 794 #ifdef SIGSOUND
 795   {  SIGSOUND,    "SIGSOUND" },
 796 #endif
 797 #ifdef SIGSTKFLT
 798   {  SIGSTKFLT,    "SIGSTKFLT" },
 799 #endif
 800   {  SIGSTOP,     "SIGSTOP" },
 801   {  SIGSYS,      "SIGSYS" },
 802 #ifdef SIGSYSERROR
 803   {  SIGSYSERROR, "SIGSYSERROR" },
 804 #endif
 805 #ifdef SIGTALRM
 806   {  SIGTALRM,    "SIGTALRM" },
 807 #endif
 808   {  SIGTERM,     "SIGTERM" },
 809 #ifdef SIGTHAW
 810   {  SIGTHAW,     "SIGTHAW" },
 811 #endif
 812   {  SIGTRAP,     "SIGTRAP" },
 813 #ifdef SIGTSTP
 814   {  SIGTSTP,     "SIGTSTP" },
 815 #endif
 816   {  SIGTTIN,     "SIGTTIN" },
 817   {  SIGTTOU,     "SIGTTOU" },
 818 #ifdef SIGURG
 819   {  SIGURG,      "SIGURG" },
 820 #endif
 821   {  SIGUSR1,     "SIGUSR1" },
 822   {  SIGUSR2,     "SIGUSR2" },
 823 #ifdef SIGVIRT
 824   {  SIGVIRT,     "SIGVIRT" },
 825 #endif
 826   {  SIGVTALRM,   "SIGVTALRM" },
 827 #ifdef SIGWAITING
 828   {  SIGWAITING,  "SIGWAITING" },
 829 #endif
 830 #ifdef SIGWINCH
 831   {  SIGWINCH,    "SIGWINCH" },
 832 #endif
 833 #ifdef SIGWINDOW
 834   {  SIGWINDOW,   "SIGWINDOW" },
 835 #endif
 836   {  SIGXCPU,     "SIGXCPU" },
 837   {  SIGXFSZ,     "SIGXFSZ" },
 838 #ifdef SIGXRES
 839   {  SIGXRES,     "SIGXRES" },
 840 #endif
 841   { -1, NULL }
 842 };
 843 
 844 // Returned string is a constant. For unknown signals "UNKNOWN" is returned.
 845 const char* os::Posix::get_signal_name(int sig, char* out, size_t outlen) {
 846 
 847   const char* ret = NULL;
 848 
 849 #ifdef SIGRTMIN
 850   if (sig >= SIGRTMIN && sig <= SIGRTMAX) {
 851     if (sig == SIGRTMIN) {
 852       ret = "SIGRTMIN";
 853     } else if (sig == SIGRTMAX) {
 854       ret = "SIGRTMAX";
 855     } else {
 856       jio_snprintf(out, outlen, "SIGRTMIN+%d", sig - SIGRTMIN);
 857       return out;
 858     }
 859   }
 860 #endif
 861 
 862   if (sig > 0) {
 863     for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
 864       if (g_signal_info[idx].sig == sig) {
 865         ret = g_signal_info[idx].name;
 866         break;
 867       }
 868     }
 869   }
 870 
 871   if (!ret) {
 872     if (!is_valid_signal(sig)) {
 873       ret = "INVALID";
 874     } else {
 875       ret = "UNKNOWN";
 876     }
 877   }
 878 
 879   if (out && outlen > 0) {
 880     strncpy(out, ret, outlen);
 881     out[outlen - 1] = '\0';
 882   }
 883   return out;
 884 }
 885 
 886 int os::Posix::get_signal_number(const char* signal_name) {
 887   char tmp[30];
 888   const char* s = signal_name;
 889   if (s[0] != 'S' || s[1] != 'I' || s[2] != 'G') {
 890     jio_snprintf(tmp, sizeof(tmp), "SIG%s", signal_name);
 891     s = tmp;
 892   }
 893   for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
 894     if (strcmp(g_signal_info[idx].name, s) == 0) {
 895       return g_signal_info[idx].sig;
 896     }
 897   }
 898   return -1;
 899 }
 900 
 901 int os::get_signal_number(const char* signal_name) {
 902   return os::Posix::get_signal_number(signal_name);
 903 }
 904 
 905 // Returns true if signal number is valid.
 906 bool os::Posix::is_valid_signal(int sig) {
 907   // MacOS not really POSIX compliant: sigaddset does not return
 908   // an error for invalid signal numbers. However, MacOS does not
 909   // support real time signals and simply seems to have just 33
 910   // signals with no holes in the signal range.
 911 #ifdef __APPLE__
 912   return sig >= 1 && sig < NSIG;
 913 #else
 914   // Use sigaddset to check for signal validity.
 915   sigset_t set;
 916   sigemptyset(&set);
 917   if (sigaddset(&set, sig) == -1 && errno == EINVAL) {
 918     return false;
 919   }
 920   return true;
 921 #endif
 922 }
 923 
 924 // Returns:
 925 // NULL for an invalid signal number
 926 // "SIG<num>" for a valid but unknown signal number
 927 // signal name otherwise.
 928 const char* os::exception_name(int sig, char* buf, size_t size) {
 929   if (!os::Posix::is_valid_signal(sig)) {
 930     return NULL;
 931   }
 932   const char* const name = os::Posix::get_signal_name(sig, buf, size);
 933   if (strcmp(name, "UNKNOWN") == 0) {
 934     jio_snprintf(buf, size, "SIG%d", sig);
 935   }
 936   return buf;
 937 }
 938 
 939 #define NUM_IMPORTANT_SIGS 32
 940 // Returns one-line short description of a signal set in a user provided buffer.
 941 const char* os::Posix::describe_signal_set_short(const sigset_t* set, char* buffer, size_t buf_size) {
 942   assert(buf_size == (NUM_IMPORTANT_SIGS + 1), "wrong buffer size");
 943   // Note: for shortness, just print out the first 32. That should
 944   // cover most of the useful ones, apart from realtime signals.
 945   for (int sig = 1; sig <= NUM_IMPORTANT_SIGS; sig++) {
 946     const int rc = sigismember(set, sig);
 947     if (rc == -1 && errno == EINVAL) {
 948       buffer[sig-1] = '?';
 949     } else {
 950       buffer[sig-1] = rc == 0 ? '0' : '1';
 951     }
 952   }
 953   buffer[NUM_IMPORTANT_SIGS] = 0;
 954   return buffer;
 955 }
 956 
 957 // Prints one-line description of a signal set.
 958 void os::Posix::print_signal_set_short(outputStream* st, const sigset_t* set) {
 959   char buf[NUM_IMPORTANT_SIGS + 1];
 960   os::Posix::describe_signal_set_short(set, buf, sizeof(buf));
 961   st->print("%s", buf);
 962 }
 963 
 964 // Writes one-line description of a combination of sigaction.sa_flags into a user
 965 // provided buffer. Returns that buffer.
 966 const char* os::Posix::describe_sa_flags(int flags, char* buffer, size_t size) {
 967   char* p = buffer;
 968   size_t remaining = size;
 969   bool first = true;
 970   int idx = 0;
 971 
 972   assert(buffer, "invalid argument");
 973 
 974   if (size == 0) {
 975     return buffer;
 976   }
 977 
 978   strncpy(buffer, "none", size);
 979 
 980   const struct {
 981     // NB: i is an unsigned int here because SA_RESETHAND is on some
 982     // systems 0x80000000, which is implicitly unsigned.  Assignining
 983     // it to an int field would be an overflow in unsigned-to-signed
 984     // conversion.
 985     unsigned int i;
 986     const char* s;
 987   } flaginfo [] = {
 988     { SA_NOCLDSTOP, "SA_NOCLDSTOP" },
 989     { SA_ONSTACK,   "SA_ONSTACK"   },
 990     { SA_RESETHAND, "SA_RESETHAND" },
 991     { SA_RESTART,   "SA_RESTART"   },
 992     { SA_SIGINFO,   "SA_SIGINFO"   },
 993     { SA_NOCLDWAIT, "SA_NOCLDWAIT" },
 994     { SA_NODEFER,   "SA_NODEFER"   },
 995 #ifdef AIX
 996     { SA_ONSTACK,   "SA_ONSTACK"   },
 997     { SA_OLDSTYLE,  "SA_OLDSTYLE"  },
 998 #endif
 999     { 0, NULL }
1000   };
1001 
1002   for (idx = 0; flaginfo[idx].s && remaining > 1; idx++) {
1003     if (flags & flaginfo[idx].i) {
1004       if (first) {
1005         jio_snprintf(p, remaining, "%s", flaginfo[idx].s);
1006         first = false;
1007       } else {
1008         jio_snprintf(p, remaining, "|%s", flaginfo[idx].s);
1009       }
1010       const size_t len = strlen(p);
1011       p += len;
1012       remaining -= len;
1013     }
1014   }
1015 
1016   buffer[size - 1] = '\0';
1017 
1018   return buffer;
1019 }
1020 
1021 // Prints one-line description of a combination of sigaction.sa_flags.
1022 void os::Posix::print_sa_flags(outputStream* st, int flags) {
1023   char buffer[0x100];
1024   os::Posix::describe_sa_flags(flags, buffer, sizeof(buffer));
1025   st->print("%s", buffer);
1026 }
1027 
1028 // Helper function for os::Posix::print_siginfo_...():
1029 // return a textual description for signal code.
1030 struct enum_sigcode_desc_t {
1031   const char* s_name;
1032   const char* s_desc;
1033 };
1034 
1035 static bool get_signal_code_description(const siginfo_t* si, enum_sigcode_desc_t* out) {
1036 
1037   const struct {
1038     int sig; int code; const char* s_code; const char* s_desc;
1039   } t1 [] = {
1040     { SIGILL,  ILL_ILLOPC,   "ILL_ILLOPC",   "Illegal opcode." },
1041     { SIGILL,  ILL_ILLOPN,   "ILL_ILLOPN",   "Illegal operand." },
1042     { SIGILL,  ILL_ILLADR,   "ILL_ILLADR",   "Illegal addressing mode." },
1043     { SIGILL,  ILL_ILLTRP,   "ILL_ILLTRP",   "Illegal trap." },
1044     { SIGILL,  ILL_PRVOPC,   "ILL_PRVOPC",   "Privileged opcode." },
1045     { SIGILL,  ILL_PRVREG,   "ILL_PRVREG",   "Privileged register." },
1046     { SIGILL,  ILL_COPROC,   "ILL_COPROC",   "Coprocessor error." },
1047     { SIGILL,  ILL_BADSTK,   "ILL_BADSTK",   "Internal stack error." },
1048 #if defined(IA64) && defined(LINUX)
1049     { SIGILL,  ILL_BADIADDR, "ILL_BADIADDR", "Unimplemented instruction address" },
1050     { SIGILL,  ILL_BREAK,    "ILL_BREAK",    "Application Break instruction" },
1051 #endif
1052     { SIGFPE,  FPE_INTDIV,   "FPE_INTDIV",   "Integer divide by zero." },
1053     { SIGFPE,  FPE_INTOVF,   "FPE_INTOVF",   "Integer overflow." },
1054     { SIGFPE,  FPE_FLTDIV,   "FPE_FLTDIV",   "Floating-point divide by zero." },
1055     { SIGFPE,  FPE_FLTOVF,   "FPE_FLTOVF",   "Floating-point overflow." },
1056     { SIGFPE,  FPE_FLTUND,   "FPE_FLTUND",   "Floating-point underflow." },
1057     { SIGFPE,  FPE_FLTRES,   "FPE_FLTRES",   "Floating-point inexact result." },
1058     { SIGFPE,  FPE_FLTINV,   "FPE_FLTINV",   "Invalid floating-point operation." },
1059     { SIGFPE,  FPE_FLTSUB,   "FPE_FLTSUB",   "Subscript out of range." },
1060     { SIGSEGV, SEGV_MAPERR,  "SEGV_MAPERR",  "Address not mapped to object." },
1061     { SIGSEGV, SEGV_ACCERR,  "SEGV_ACCERR",  "Invalid permissions for mapped object." },
1062 #ifdef AIX
1063     // no explanation found what keyerr would be
1064     { SIGSEGV, SEGV_KEYERR,  "SEGV_KEYERR",  "key error" },
1065 #endif
1066 #if defined(IA64) && !defined(AIX)
1067     { SIGSEGV, SEGV_PSTKOVF, "SEGV_PSTKOVF", "Paragraph stack overflow" },
1068 #endif
1069 #if defined(__sparc) && defined(SOLARIS)
1070 // define Solaris Sparc M7 ADI SEGV signals
1071 #if !defined(SEGV_ACCADI)
1072 #define SEGV_ACCADI 3
1073 #endif
1074     { SIGSEGV, SEGV_ACCADI,  "SEGV_ACCADI",  "ADI not enabled for mapped object." },
1075 #if !defined(SEGV_ACCDERR)
1076 #define SEGV_ACCDERR 4
1077 #endif
1078     { SIGSEGV, SEGV_ACCDERR, "SEGV_ACCDERR", "ADI disrupting exception." },
1079 #if !defined(SEGV_ACCPERR)
1080 #define SEGV_ACCPERR 5
1081 #endif
1082     { SIGSEGV, SEGV_ACCPERR, "SEGV_ACCPERR", "ADI precise exception." },
1083 #endif // defined(__sparc) && defined(SOLARIS)
1084     { SIGBUS,  BUS_ADRALN,   "BUS_ADRALN",   "Invalid address alignment." },
1085     { SIGBUS,  BUS_ADRERR,   "BUS_ADRERR",   "Nonexistent physical address." },
1086     { SIGBUS,  BUS_OBJERR,   "BUS_OBJERR",   "Object-specific hardware error." },
1087     { SIGTRAP, TRAP_BRKPT,   "TRAP_BRKPT",   "Process breakpoint." },
1088     { SIGTRAP, TRAP_TRACE,   "TRAP_TRACE",   "Process trace trap." },
1089     { SIGCHLD, CLD_EXITED,   "CLD_EXITED",   "Child has exited." },
1090     { SIGCHLD, CLD_KILLED,   "CLD_KILLED",   "Child has terminated abnormally and did not create a core file." },
1091     { SIGCHLD, CLD_DUMPED,   "CLD_DUMPED",   "Child has terminated abnormally and created a core file." },
1092     { SIGCHLD, CLD_TRAPPED,  "CLD_TRAPPED",  "Traced child has trapped." },
1093     { SIGCHLD, CLD_STOPPED,  "CLD_STOPPED",  "Child has stopped." },
1094     { SIGCHLD, CLD_CONTINUED,"CLD_CONTINUED","Stopped child has continued." },
1095 #ifdef SIGPOLL
1096     { SIGPOLL, POLL_OUT,     "POLL_OUT",     "Output buffers available." },
1097     { SIGPOLL, POLL_MSG,     "POLL_MSG",     "Input message available." },
1098     { SIGPOLL, POLL_ERR,     "POLL_ERR",     "I/O error." },
1099     { SIGPOLL, POLL_PRI,     "POLL_PRI",     "High priority input available." },
1100     { SIGPOLL, POLL_HUP,     "POLL_HUP",     "Device disconnected. [Option End]" },
1101 #endif
1102     { -1, -1, NULL, NULL }
1103   };
1104 
1105   // Codes valid in any signal context.
1106   const struct {
1107     int code; const char* s_code; const char* s_desc;
1108   } t2 [] = {
1109     { SI_USER,      "SI_USER",     "Signal sent by kill()." },
1110     { SI_QUEUE,     "SI_QUEUE",    "Signal sent by the sigqueue()." },
1111     { SI_TIMER,     "SI_TIMER",    "Signal generated by expiration of a timer set by timer_settime()." },
1112     { SI_ASYNCIO,   "SI_ASYNCIO",  "Signal generated by completion of an asynchronous I/O request." },
1113     { SI_MESGQ,     "SI_MESGQ",    "Signal generated by arrival of a message on an empty message queue." },
1114     // Linux specific
1115 #ifdef SI_TKILL
1116     { SI_TKILL,     "SI_TKILL",    "Signal sent by tkill (pthread_kill)" },
1117 #endif
1118 #ifdef SI_DETHREAD
1119     { SI_DETHREAD,  "SI_DETHREAD", "Signal sent by execve() killing subsidiary threads" },
1120 #endif
1121 #ifdef SI_KERNEL
1122     { SI_KERNEL,    "SI_KERNEL",   "Signal sent by kernel." },
1123 #endif
1124 #ifdef SI_SIGIO
1125     { SI_SIGIO,     "SI_SIGIO",    "Signal sent by queued SIGIO" },
1126 #endif
1127 
1128 #ifdef AIX
1129     { SI_UNDEFINED, "SI_UNDEFINED","siginfo contains partial information" },
1130     { SI_EMPTY,     "SI_EMPTY",    "siginfo contains no useful information" },
1131 #endif
1132 
1133 #ifdef __sun
1134     { SI_NOINFO,    "SI_NOINFO",   "No signal information" },
1135     { SI_RCTL,      "SI_RCTL",     "kernel generated signal via rctl action" },
1136     { SI_LWP,       "SI_LWP",      "Signal sent via lwp_kill" },
1137 #endif
1138 
1139     { -1, NULL, NULL }
1140   };
1141 
1142   const char* s_code = NULL;
1143   const char* s_desc = NULL;
1144 
1145   for (int i = 0; t1[i].sig != -1; i ++) {
1146     if (t1[i].sig == si->si_signo && t1[i].code == si->si_code) {
1147       s_code = t1[i].s_code;
1148       s_desc = t1[i].s_desc;
1149       break;
1150     }
1151   }
1152 
1153   if (s_code == NULL) {
1154     for (int i = 0; t2[i].s_code != NULL; i ++) {
1155       if (t2[i].code == si->si_code) {
1156         s_code = t2[i].s_code;
1157         s_desc = t2[i].s_desc;
1158       }
1159     }
1160   }
1161 
1162   if (s_code == NULL) {
1163     out->s_name = "unknown";
1164     out->s_desc = "unknown";
1165     return false;
1166   }
1167 
1168   out->s_name = s_code;
1169   out->s_desc = s_desc;
1170 
1171   return true;
1172 }
1173 
1174 void os::print_siginfo(outputStream* os, const void* si0) {
1175 
1176   const siginfo_t* const si = (const siginfo_t*) si0;
1177 
1178   char buf[20];
1179   os->print("siginfo:");
1180 
1181   if (!si) {
1182     os->print(" <null>");
1183     return;
1184   }
1185 
1186   const int sig = si->si_signo;
1187 
1188   os->print(" si_signo: %d (%s)", sig, os::Posix::get_signal_name(sig, buf, sizeof(buf)));
1189 
1190   enum_sigcode_desc_t ed;
1191   get_signal_code_description(si, &ed);
1192   os->print(", si_code: %d (%s)", si->si_code, ed.s_name);
1193 
1194   if (si->si_errno) {
1195     os->print(", si_errno: %d", si->si_errno);
1196   }
1197 
1198   // Output additional information depending on the signal code.
1199 
1200   // Note: Many implementations lump si_addr, si_pid, si_uid etc. together as unions,
1201   // so it depends on the context which member to use. For synchronous error signals,
1202   // we print si_addr, unless the signal was sent by another process or thread, in
1203   // which case we print out pid or tid of the sender.
1204   if (si->si_code == SI_USER || si->si_code == SI_QUEUE) {
1205     const pid_t pid = si->si_pid;
1206     os->print(", si_pid: %ld", (long) pid);
1207     if (IS_VALID_PID(pid)) {
1208       const pid_t me = getpid();
1209       if (me == pid) {
1210         os->print(" (current process)");
1211       }
1212     } else {
1213       os->print(" (invalid)");
1214     }
1215     os->print(", si_uid: %ld", (long) si->si_uid);
1216     if (sig == SIGCHLD) {
1217       os->print(", si_status: %d", si->si_status);
1218     }
1219   } else if (sig == SIGSEGV || sig == SIGBUS || sig == SIGILL ||
1220              sig == SIGTRAP || sig == SIGFPE) {
1221     os->print(", si_addr: " PTR_FORMAT, p2i(si->si_addr));
1222 #ifdef SIGPOLL
1223   } else if (sig == SIGPOLL) {
1224     os->print(", si_band: %ld", si->si_band);
1225 #endif
1226   }
1227 
1228 }
1229 
1230 int os::Posix::unblock_thread_signal_mask(const sigset_t *set) {
1231   return pthread_sigmask(SIG_UNBLOCK, set, NULL);
1232 }
1233 
1234 address os::Posix::ucontext_get_pc(const ucontext_t* ctx) {
1235 #if defined(AIX)
1236    return Aix::ucontext_get_pc(ctx);
1237 #elif defined(BSD)
1238    return Bsd::ucontext_get_pc(ctx);
1239 #elif defined(LINUX)
1240    return Linux::ucontext_get_pc(ctx);
1241 #elif defined(SOLARIS)
1242    return Solaris::ucontext_get_pc(ctx);
1243 #else
1244    VMError::report_and_die("unimplemented ucontext_get_pc");
1245 #endif
1246 }
1247 
1248 void os::Posix::ucontext_set_pc(ucontext_t* ctx, address pc) {
1249 #if defined(AIX)
1250    Aix::ucontext_set_pc(ctx, pc);
1251 #elif defined(BSD)
1252    Bsd::ucontext_set_pc(ctx, pc);
1253 #elif defined(LINUX)
1254    Linux::ucontext_set_pc(ctx, pc);
1255 #elif defined(SOLARIS)
1256    Solaris::ucontext_set_pc(ctx, pc);
1257 #else
1258    VMError::report_and_die("unimplemented ucontext_get_pc");
1259 #endif
1260 }
1261 
1262 char* os::Posix::describe_pthread_attr(char* buf, size_t buflen, const pthread_attr_t* attr) {
1263   size_t stack_size = 0;
1264   size_t guard_size = 0;
1265   int detachstate = 0;
1266   pthread_attr_getstacksize(attr, &stack_size);
1267   pthread_attr_getguardsize(attr, &guard_size);
1268   // Work around linux NPTL implementation error, see also os::create_thread() in os_linux.cpp.
1269   LINUX_ONLY(stack_size -= guard_size);
1270   pthread_attr_getdetachstate(attr, &detachstate);
1271   jio_snprintf(buf, buflen, "stacksize: " SIZE_FORMAT "k, guardsize: " SIZE_FORMAT "k, %s",
1272     stack_size / 1024, guard_size / 1024,
1273     (detachstate == PTHREAD_CREATE_DETACHED ? "detached" : "joinable"));
1274   return buf;
1275 }
1276 
1277 char* os::Posix::realpath(const char* filename, char* outbuf, size_t outbuflen) {
1278 
1279   if (filename == NULL || outbuf == NULL || outbuflen < 1) {
1280     assert(false, "os::Posix::realpath: invalid arguments.");
1281     errno = EINVAL;
1282     return NULL;
1283   }
1284 
1285   char* result = NULL;
1286 
1287   // This assumes platform realpath() is implemented according to POSIX.1-2008.
1288   // POSIX.1-2008 allows to specify NULL for the output buffer, in which case
1289   // output buffer is dynamically allocated and must be ::free()'d by the caller.
1290   char* p = ::realpath(filename, NULL);
1291   if (p != NULL) {
1292     if (strlen(p) < outbuflen) {
1293       strcpy(outbuf, p);
1294       result = outbuf;
1295     } else {
1296       errno = ENAMETOOLONG;
1297     }
1298     ::free(p); // *not* os::free
1299   } else {
1300     // Fallback for platforms struggling with modern Posix standards (AIX 5.3, 6.1). If realpath
1301     // returns EINVAL, this may indicate that realpath is not POSIX.1-2008 compatible and
1302     // that it complains about the NULL we handed down as user buffer.
1303     // In this case, use the user provided buffer but at least check whether realpath caused
1304     // a memory overwrite.
1305     if (errno == EINVAL) {
1306       outbuf[outbuflen - 1] = '\0';
1307       p = ::realpath(filename, outbuf);
1308       if (p != NULL) {
1309         guarantee(outbuf[outbuflen - 1] == '\0', "realpath buffer overwrite detected.");
1310         result = p;
1311       }
1312     }
1313   }
1314   return result;
1315 
1316 }
1317 
1318 
1319 // Check minimum allowable stack sizes for thread creation and to initialize
1320 // the java system classes, including StackOverflowError - depends on page
1321 // size.
1322 // The space needed for frames during startup is platform dependent. It
1323 // depends on word size, platform calling conventions, C frame layout and
1324 // interpreter/C1/C2 design decisions. Therefore this is given in a
1325 // platform (os/cpu) dependent constant.
1326 // To this, space for guard mechanisms is added, which depends on the
1327 // page size which again depends on the concrete system the VM is running
1328 // on. Space for libc guard pages is not included in this size.
1329 jint os::Posix::set_minimum_stack_sizes() {
1330   size_t os_min_stack_allowed = SOLARIS_ONLY(thr_min_stack()) NOT_SOLARIS(PTHREAD_STACK_MIN);
1331 
1332   _java_thread_min_stack_allowed = _java_thread_min_stack_allowed +
1333                                    JavaThread::stack_guard_zone_size() +
1334                                    JavaThread::stack_shadow_zone_size();
1335 
1336   _java_thread_min_stack_allowed = align_up(_java_thread_min_stack_allowed, vm_page_size());
1337   _java_thread_min_stack_allowed = MAX2(_java_thread_min_stack_allowed, os_min_stack_allowed);
1338 
1339   size_t stack_size_in_bytes = ThreadStackSize * K;
1340   if (stack_size_in_bytes != 0 &&
1341       stack_size_in_bytes < _java_thread_min_stack_allowed) {
1342     // The '-Xss' and '-XX:ThreadStackSize=N' options both set
1343     // ThreadStackSize so we go with "Java thread stack size" instead
1344     // of "ThreadStackSize" to be more friendly.
1345     tty->print_cr("\nThe Java thread stack size specified is too small. "
1346                   "Specify at least " SIZE_FORMAT "k",
1347                   _java_thread_min_stack_allowed / K);
1348     return JNI_ERR;
1349   }
1350 
1351   // Make the stack size a multiple of the page size so that
1352   // the yellow/red zones can be guarded.
1353   JavaThread::set_stack_size_at_create(align_up(stack_size_in_bytes, vm_page_size()));
1354 
1355   // Reminder: a compiler thread is a Java thread.
1356   _compiler_thread_min_stack_allowed = _compiler_thread_min_stack_allowed +
1357                                        JavaThread::stack_guard_zone_size() +
1358                                        JavaThread::stack_shadow_zone_size();
1359 
1360   _compiler_thread_min_stack_allowed = align_up(_compiler_thread_min_stack_allowed, vm_page_size());
1361   _compiler_thread_min_stack_allowed = MAX2(_compiler_thread_min_stack_allowed, os_min_stack_allowed);
1362 
1363   stack_size_in_bytes = CompilerThreadStackSize * K;
1364   if (stack_size_in_bytes != 0 &&
1365       stack_size_in_bytes < _compiler_thread_min_stack_allowed) {
1366     tty->print_cr("\nThe CompilerThreadStackSize specified is too small. "
1367                   "Specify at least " SIZE_FORMAT "k",
1368                   _compiler_thread_min_stack_allowed / K);
1369     return JNI_ERR;
1370   }
1371 
1372   _vm_internal_thread_min_stack_allowed = align_up(_vm_internal_thread_min_stack_allowed, vm_page_size());
1373   _vm_internal_thread_min_stack_allowed = MAX2(_vm_internal_thread_min_stack_allowed, os_min_stack_allowed);
1374 
1375   stack_size_in_bytes = VMThreadStackSize * K;
1376   if (stack_size_in_bytes != 0 &&
1377       stack_size_in_bytes < _vm_internal_thread_min_stack_allowed) {
1378     tty->print_cr("\nThe VMThreadStackSize specified is too small. "
1379                   "Specify at least " SIZE_FORMAT "k",
1380                   _vm_internal_thread_min_stack_allowed / K);
1381     return JNI_ERR;
1382   }
1383   return JNI_OK;
1384 }
1385 
1386 // Called when creating the thread.  The minimum stack sizes have already been calculated
1387 size_t os::Posix::get_initial_stack_size(ThreadType thr_type, size_t req_stack_size) {
1388   size_t stack_size;
1389   if (req_stack_size == 0) {
1390     stack_size = default_stack_size(thr_type);
1391   } else {
1392     stack_size = req_stack_size;
1393   }
1394 
1395   switch (thr_type) {
1396   case os::java_thread:
1397     // Java threads use ThreadStackSize which default value can be
1398     // changed with the flag -Xss
1399     if (req_stack_size == 0 && JavaThread::stack_size_at_create() > 0) {
1400       // no requested size and we have a more specific default value
1401       stack_size = JavaThread::stack_size_at_create();
1402     }
1403     stack_size = MAX2(stack_size,
1404                       _java_thread_min_stack_allowed);
1405     break;
1406   case os::compiler_thread:
1407     if (req_stack_size == 0 && CompilerThreadStackSize > 0) {
1408       // no requested size and we have a more specific default value
1409       stack_size = (size_t)(CompilerThreadStackSize * K);
1410     }
1411     stack_size = MAX2(stack_size,
1412                       _compiler_thread_min_stack_allowed);
1413     break;
1414   case os::vm_thread:
1415   case os::pgc_thread:
1416   case os::cgc_thread:
1417   case os::watcher_thread:
1418   default:  // presume the unknown thr_type is a VM internal
1419     if (req_stack_size == 0 && VMThreadStackSize > 0) {
1420       // no requested size and we have a more specific default value
1421       stack_size = (size_t)(VMThreadStackSize * K);
1422     }
1423 
1424     stack_size = MAX2(stack_size,
1425                       _vm_internal_thread_min_stack_allowed);
1426     break;
1427   }
1428 
1429   // pthread_attr_setstacksize() may require that the size be rounded up to the OS page size.
1430   // Be careful not to round up to 0. Align down in that case.
1431   if (stack_size <= SIZE_MAX - vm_page_size()) {
1432     stack_size = align_up(stack_size, vm_page_size());
1433   } else {
1434     stack_size = align_down(stack_size, vm_page_size());
1435   }
1436 
1437   return stack_size;
1438 }
1439 
1440 Thread* os::ThreadCrashProtection::_protected_thread = NULL;
1441 os::ThreadCrashProtection* os::ThreadCrashProtection::_crash_protection = NULL;
1442 volatile intptr_t os::ThreadCrashProtection::_crash_mux = 0;
1443 
1444 os::ThreadCrashProtection::ThreadCrashProtection() {
1445 }
1446 
1447 /*
1448  * See the caveats for this class in os_posix.hpp
1449  * Protects the callback call so that SIGSEGV / SIGBUS jumps back into this
1450  * method and returns false. If none of the signals are raised, returns true.
1451  * The callback is supposed to provide the method that should be protected.
1452  */
1453 bool os::ThreadCrashProtection::call(os::CrashProtectionCallback& cb) {
1454   sigset_t saved_sig_mask;
1455 
1456   Thread::muxAcquire(&_crash_mux, "CrashProtection");
1457 
1458   _protected_thread = Thread::current_or_null();
1459   assert(_protected_thread != NULL, "Cannot crash protect a NULL thread");
1460 
1461   // we cannot rely on sigsetjmp/siglongjmp to save/restore the signal mask
1462   // since on at least some systems (OS X) siglongjmp will restore the mask
1463   // for the process, not the thread
1464   pthread_sigmask(0, NULL, &saved_sig_mask);
1465   if (sigsetjmp(_jmpbuf, 0) == 0) {
1466     // make sure we can see in the signal handler that we have crash protection
1467     // installed
1468     _crash_protection = this;
1469     cb.call();
1470     // and clear the crash protection
1471     _crash_protection = NULL;
1472     _protected_thread = NULL;
1473     Thread::muxRelease(&_crash_mux);
1474     return true;
1475   }
1476   // this happens when we siglongjmp() back
1477   pthread_sigmask(SIG_SETMASK, &saved_sig_mask, NULL);
1478   _crash_protection = NULL;
1479   _protected_thread = NULL;
1480   Thread::muxRelease(&_crash_mux);
1481   return false;
1482 }
1483 
1484 void os::ThreadCrashProtection::restore() {
1485   assert(_crash_protection != NULL, "must have crash protection");
1486   siglongjmp(_jmpbuf, 1);
1487 }
1488 
1489 void os::ThreadCrashProtection::check_crash_protection(int sig,
1490     Thread* thread) {
1491 
1492   if (thread != NULL &&
1493       thread == _protected_thread &&
1494       _crash_protection != NULL) {
1495 
1496     if (sig == SIGSEGV || sig == SIGBUS) {
1497       _crash_protection->restore();
1498     }
1499   }
1500 }
1501 
1502 // POSIX unamed semaphores are not supported on OS X.
1503 #ifndef __APPLE__
1504 
1505 PosixSemaphore::PosixSemaphore(uint value) {
1506   int ret = sem_init(&_semaphore, 0, value);
1507 
1508   guarantee_with_errno(ret == 0, "Failed to initialize semaphore");
1509 }
1510 
1511 PosixSemaphore::~PosixSemaphore() {
1512   sem_destroy(&_semaphore);
1513 }
1514 
1515 void PosixSemaphore::signal(uint count) {
1516   for (uint i = 0; i < count; i++) {
1517     int ret = sem_post(&_semaphore);
1518 
1519     assert_with_errno(ret == 0, "sem_post failed");
1520   }
1521 }
1522 
1523 void PosixSemaphore::wait() {
1524   int ret;
1525 
1526   do {
1527     ret = sem_wait(&_semaphore);
1528   } while (ret != 0 && errno == EINTR);
1529 
1530   assert_with_errno(ret == 0, "sem_wait failed");
1531 }
1532 
1533 bool PosixSemaphore::trywait() {
1534   int ret;
1535 
1536   do {
1537     ret = sem_trywait(&_semaphore);
1538   } while (ret != 0 && errno == EINTR);
1539 
1540   assert_with_errno(ret == 0 || errno == EAGAIN, "trywait failed");
1541 
1542   return ret == 0;
1543 }
1544 
1545 bool PosixSemaphore::timedwait(struct timespec ts) {
1546   while (true) {
1547     int result = sem_timedwait(&_semaphore, &ts);
1548     if (result == 0) {
1549       return true;
1550     } else if (errno == EINTR) {
1551       continue;
1552     } else if (errno == ETIMEDOUT) {
1553       return false;
1554     } else {
1555       assert_with_errno(false, "timedwait failed");
1556       return false;
1557     }
1558   }
1559 }
1560 
1561 #endif // __APPLE__
1562 
1563 
1564 // Shared pthread_mutex/cond based PlatformEvent implementation.
1565 // Not currently usable by Solaris.
1566 
1567 #ifndef SOLARIS
1568 
1569 // Shared condattr object for use with relative timed-waits. Will be associated
1570 // with CLOCK_MONOTONIC if available to avoid issues with time-of-day changes,
1571 // but otherwise whatever default is used by the platform - generally the
1572 // time-of-day clock.
1573 static pthread_condattr_t _condAttr[1];
1574 
1575 // Shared mutexattr to explicitly set the type to PTHREAD_MUTEX_NORMAL as not
1576 // all systems (e.g. FreeBSD) map the default to "normal".
1577 static pthread_mutexattr_t _mutexAttr[1];
1578 
1579 // common basic initialization that is always supported
1580 static void pthread_init_common(void) {
1581   int status;
1582   if ((status = pthread_condattr_init(_condAttr)) != 0) {
1583     fatal("pthread_condattr_init: %s", os::strerror(status));
1584   }
1585   if ((status = pthread_mutexattr_init(_mutexAttr)) != 0) {
1586     fatal("pthread_mutexattr_init: %s", os::strerror(status));
1587   }
1588   if ((status = pthread_mutexattr_settype(_mutexAttr, PTHREAD_MUTEX_NORMAL)) != 0) {
1589     fatal("pthread_mutexattr_settype: %s", os::strerror(status));
1590   }
1591 }
1592 
1593 // Not all POSIX types and API's are available on all notionally "posix"
1594 // platforms. If we have build-time support then we will check for actual
1595 // runtime support via dlopen/dlsym lookup. This allows for running on an
1596 // older OS version compared to the build platform. But if there is no
1597 // build time support then there cannot be any runtime support as we do not
1598 // know what the runtime types would be (for example clockid_t might be an
1599 // int or int64_t).
1600 //
1601 #ifdef SUPPORTS_CLOCK_MONOTONIC
1602 
1603 // This means we have clockid_t, clock_gettime et al and CLOCK_MONOTONIC
1604 
1605 static int (*_clock_gettime)(clockid_t, struct timespec *);
1606 static int (*_pthread_condattr_setclock)(pthread_condattr_t *, clockid_t);
1607 
1608 static bool _use_clock_monotonic_condattr;
1609 
1610 // Determine what POSIX API's are present and do appropriate
1611 // configuration.
1612 void os::Posix::init(void) {
1613 
1614   // NOTE: no logging available when this is called. Put logging
1615   // statements in init_2().
1616 
1617   // Copied from os::Linux::clock_init(). The duplication is temporary.
1618 
1619   // 1. Check for CLOCK_MONOTONIC support.
1620 
1621   void* handle = NULL;
1622 
1623   // For linux we need librt, for other OS we can find
1624   // this function in regular libc.
1625 #ifdef NEEDS_LIBRT
1626   // We do dlopen's in this particular order due to bug in linux
1627   // dynamic loader (see 6348968) leading to crash on exit.
1628   handle = dlopen("librt.so.1", RTLD_LAZY);
1629   if (handle == NULL) {
1630     handle = dlopen("librt.so", RTLD_LAZY);
1631   }
1632 #endif
1633 
1634   if (handle == NULL) {
1635     handle = RTLD_DEFAULT;
1636   }
1637 
1638   _clock_gettime = NULL;
1639 
1640   int (*clock_getres_func)(clockid_t, struct timespec*) =
1641     (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1642   int (*clock_gettime_func)(clockid_t, struct timespec*) =
1643     (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1644   if (clock_getres_func != NULL && clock_gettime_func != NULL) {
1645     // We assume that if both clock_gettime and clock_getres support
1646     // CLOCK_MONOTONIC then the OS provides true high-res monotonic clock.
1647     struct timespec res;
1648     struct timespec tp;
1649     if (clock_getres_func(CLOCK_MONOTONIC, &res) == 0 &&
1650         clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1651       // Yes, monotonic clock is supported.
1652       _clock_gettime = clock_gettime_func;
1653     } else {
1654 #ifdef NEEDS_LIBRT
1655       // Close librt if there is no monotonic clock.
1656       if (handle != RTLD_DEFAULT) {
1657         dlclose(handle);
1658       }
1659 #endif
1660     }
1661   }
1662 
1663   // 2. Check for pthread_condattr_setclock support.
1664 
1665   _pthread_condattr_setclock = NULL;
1666 
1667   // libpthread is already loaded.
1668   int (*condattr_setclock_func)(pthread_condattr_t*, clockid_t) =
1669     (int (*)(pthread_condattr_t*, clockid_t))dlsym(RTLD_DEFAULT,
1670                                                    "pthread_condattr_setclock");
1671   if (condattr_setclock_func != NULL) {
1672     _pthread_condattr_setclock = condattr_setclock_func;
1673   }
1674 
1675   // Now do general initialization.
1676 
1677   pthread_init_common();
1678 
1679   int status;
1680   if (_pthread_condattr_setclock != NULL && _clock_gettime != NULL) {
1681     if ((status = _pthread_condattr_setclock(_condAttr, CLOCK_MONOTONIC)) != 0) {
1682       if (status == EINVAL) {
1683         _use_clock_monotonic_condattr = false;
1684         warning("Unable to use monotonic clock with relative timed-waits" \
1685                 " - changes to the time-of-day clock may have adverse affects");
1686       } else {
1687         fatal("pthread_condattr_setclock: %s", os::strerror(status));
1688       }
1689     } else {
1690       _use_clock_monotonic_condattr = true;
1691     }
1692   } else {
1693     _use_clock_monotonic_condattr = false;
1694   }
1695 }
1696 
1697 void os::Posix::init_2(void) {
1698   log_info(os)("Use of CLOCK_MONOTONIC is%s supported",
1699                (_clock_gettime != NULL ? "" : " not"));
1700   log_info(os)("Use of pthread_condattr_setclock is%s supported",
1701                (_pthread_condattr_setclock != NULL ? "" : " not"));
1702   log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with %s",
1703                _use_clock_monotonic_condattr ? "CLOCK_MONOTONIC" : "the default clock");
1704 }
1705 
1706 #else // !SUPPORTS_CLOCK_MONOTONIC
1707 
1708 void os::Posix::init(void) {
1709   pthread_init_common();
1710 }
1711 
1712 void os::Posix::init_2(void) {
1713   log_info(os)("Use of CLOCK_MONOTONIC is not supported");
1714   log_info(os)("Use of pthread_condattr_setclock is not supported");
1715   log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with the default clock");
1716 }
1717 
1718 #endif // SUPPORTS_CLOCK_MONOTONIC
1719 
1720 os::PlatformEvent::PlatformEvent() {
1721   int status = pthread_cond_init(_cond, _condAttr);
1722   assert_status(status == 0, status, "cond_init");
1723   status = pthread_mutex_init(_mutex, _mutexAttr);
1724   assert_status(status == 0, status, "mutex_init");
1725   _event   = 0;
1726   _nParked = 0;
1727 }
1728 
1729 // Utility to convert the given timeout to an absolute timespec
1730 // (based on the appropriate clock) to use with pthread_cond_timewait.
1731 // The clock queried here must be the clock used to manage the
1732 // timeout of the condition variable.
1733 //
1734 // The passed in timeout value is either a relative time in nanoseconds
1735 // or an absolute time in milliseconds. A relative timeout will be
1736 // associated with CLOCK_MONOTONIC if available; otherwise, or if absolute,
1737 // the default time-of-day clock will be used.
1738 
1739 // Given time is a 64-bit value and the time_t used in the timespec is
1740 // sometimes a signed-32-bit value we have to watch for overflow if times
1741 // way in the future are given. Further on Solaris versions
1742 // prior to 10 there is a restriction (see cond_timedwait) that the specified
1743 // number of seconds, in abstime, is less than current_time + 100000000.
1744 // As it will be over 20 years before "now + 100000000" will overflow we can
1745 // ignore overflow and just impose a hard-limit on seconds using the value
1746 // of "now + 100000000". This places a limit on the timeout of about 3.17
1747 // years from "now".
1748 //
1749 #define MAX_SECS 100000000
1750 
1751 // Calculate a new absolute time that is "timeout" nanoseconds from "now".
1752 // "unit" indicates the unit of "now_part_sec" (may be nanos or micros depending
1753 // on which clock is being used).
1754 static void calc_rel_time(timespec* abstime, jlong timeout, jlong now_sec,
1755                           jlong now_part_sec, jlong unit) {
1756   time_t max_secs = now_sec + MAX_SECS;
1757 
1758   jlong seconds = timeout / NANOUNITS;
1759   timeout %= NANOUNITS; // remaining nanos
1760 
1761   if (seconds >= MAX_SECS) {
1762     // More seconds than we can add, so pin to max_secs.
1763     abstime->tv_sec = max_secs;
1764     abstime->tv_nsec = 0;
1765   } else {
1766     abstime->tv_sec = now_sec  + seconds;
1767     long nanos = (now_part_sec * (NANOUNITS / unit)) + timeout;
1768     if (nanos >= NANOUNITS) { // overflow
1769       abstime->tv_sec += 1;
1770       nanos -= NANOUNITS;
1771     }
1772     abstime->tv_nsec = nanos;
1773   }
1774 }
1775 
1776 // Unpack the given deadline in milliseconds since the epoch, into the given timespec.
1777 // The current time in seconds is also passed in to enforce an upper bound as discussed above.
1778 static void unpack_abs_time(timespec* abstime, jlong deadline, jlong now_sec) {
1779   time_t max_secs = now_sec + MAX_SECS;
1780 
1781   jlong seconds = deadline / MILLIUNITS;
1782   jlong millis = deadline % MILLIUNITS;
1783 
1784   if (seconds >= max_secs) {
1785     // Absolute seconds exceeds allowed max, so pin to max_secs.
1786     abstime->tv_sec = max_secs;
1787     abstime->tv_nsec = 0;
1788   } else {
1789     abstime->tv_sec = seconds;
1790     abstime->tv_nsec = millis * (NANOUNITS / MILLIUNITS);
1791   }
1792 }
1793 
1794 static void to_abstime(timespec* abstime, jlong timeout, bool isAbsolute) {
1795   DEBUG_ONLY(int max_secs = MAX_SECS;)
1796 
1797   if (timeout < 0) {
1798     timeout = 0;
1799   }
1800 
1801 #ifdef SUPPORTS_CLOCK_MONOTONIC
1802 
1803   if (_use_clock_monotonic_condattr && !isAbsolute) {
1804     struct timespec now;
1805     int status = _clock_gettime(CLOCK_MONOTONIC, &now);
1806     assert_status(status == 0, status, "clock_gettime");
1807     calc_rel_time(abstime, timeout, now.tv_sec, now.tv_nsec, NANOUNITS);
1808     DEBUG_ONLY(max_secs += now.tv_sec;)
1809   } else {
1810 
1811 #else
1812 
1813   { // Match the block scope.
1814 
1815 #endif // SUPPORTS_CLOCK_MONOTONIC
1816 
1817     // Time-of-day clock is all we can reliably use.
1818     struct timeval now;
1819     int status = gettimeofday(&now, NULL);
1820     assert_status(status == 0, errno, "gettimeofday");
1821     if (isAbsolute) {
1822       unpack_abs_time(abstime, timeout, now.tv_sec);
1823     } else {
1824       calc_rel_time(abstime, timeout, now.tv_sec, now.tv_usec, MICROUNITS);
1825     }
1826     DEBUG_ONLY(max_secs += now.tv_sec;)
1827   }
1828 
1829   assert(abstime->tv_sec >= 0, "tv_sec < 0");
1830   assert(abstime->tv_sec <= max_secs, "tv_sec > max_secs");
1831   assert(abstime->tv_nsec >= 0, "tv_nsec < 0");
1832   assert(abstime->tv_nsec < NANOUNITS, "tv_nsec >= NANOUNITS");
1833 }
1834 
1835 // PlatformEvent
1836 //
1837 // Assumption:
1838 //    Only one parker can exist on an event, which is why we allocate
1839 //    them per-thread. Multiple unparkers can coexist.
1840 //
1841 // _event serves as a restricted-range semaphore.
1842 //   -1 : thread is blocked, i.e. there is a waiter
1843 //    0 : neutral: thread is running or ready,
1844 //        could have been signaled after a wait started
1845 //    1 : signaled - thread is running or ready
1846 //
1847 //    Having three states allows for some detection of bad usage - see
1848 //    comments on unpark().
1849 
1850 void os::PlatformEvent::park() {       // AKA "down()"
1851   // Transitions for _event:
1852   //   -1 => -1 : illegal
1853   //    1 =>  0 : pass - return immediately
1854   //    0 => -1 : block; then set _event to 0 before returning
1855 
1856   // Invariant: Only the thread associated with the PlatformEvent
1857   // may call park().
1858   assert(_nParked == 0, "invariant");
1859 
1860   int v;
1861 
1862   // atomically decrement _event
1863   for (;;) {
1864     v = _event;
1865     if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1866   }
1867   guarantee(v >= 0, "invariant");
1868 
1869   if (v == 0) { // Do this the hard way by blocking ...
1870     int status = pthread_mutex_lock(_mutex);
1871     assert_status(status == 0, status, "mutex_lock");
1872     guarantee(_nParked == 0, "invariant");
1873     ++_nParked;
1874     while (_event < 0) {
1875       // OS-level "spurious wakeups" are ignored
1876       status = pthread_cond_wait(_cond, _mutex);
1877       assert_status(status == 0, status, "cond_wait");
1878     }
1879     --_nParked;
1880 
1881     _event = 0;
1882     status = pthread_mutex_unlock(_mutex);
1883     assert_status(status == 0, status, "mutex_unlock");
1884     // Paranoia to ensure our locked and lock-free paths interact
1885     // correctly with each other.
1886     OrderAccess::fence();
1887   }
1888   guarantee(_event >= 0, "invariant");
1889 }
1890 
1891 int os::PlatformEvent::park(jlong millis) {
1892   // Transitions for _event:
1893   //   -1 => -1 : illegal
1894   //    1 =>  0 : pass - return immediately
1895   //    0 => -1 : block; then set _event to 0 before returning
1896 
1897   // Invariant: Only the thread associated with the Event/PlatformEvent
1898   // may call park().
1899   assert(_nParked == 0, "invariant");
1900 
1901   int v;
1902   // atomically decrement _event
1903   for (;;) {
1904     v = _event;
1905     if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1906   }
1907   guarantee(v >= 0, "invariant");
1908 
1909   if (v == 0) { // Do this the hard way by blocking ...
1910     struct timespec abst;
1911     // We have to watch for overflow when converting millis to nanos,
1912     // but if millis is that large then we will end up limiting to
1913     // MAX_SECS anyway, so just do that here.
1914     if (millis / MILLIUNITS > MAX_SECS) {
1915       millis = jlong(MAX_SECS) * MILLIUNITS;
1916     }
1917     to_abstime(&abst, millis * (NANOUNITS / MILLIUNITS), false);
1918 
1919     int ret = OS_TIMEOUT;
1920     int status = pthread_mutex_lock(_mutex);
1921     assert_status(status == 0, status, "mutex_lock");
1922     guarantee(_nParked == 0, "invariant");
1923     ++_nParked;
1924 
1925     while (_event < 0) {
1926       status = pthread_cond_timedwait(_cond, _mutex, &abst);
1927       assert_status(status == 0 || status == ETIMEDOUT,
1928                     status, "cond_timedwait");
1929       // OS-level "spurious wakeups" are ignored unless the archaic
1930       // FilterSpuriousWakeups is set false. That flag should be obsoleted.
1931       if (!FilterSpuriousWakeups) break;
1932       if (status == ETIMEDOUT) break;
1933     }
1934     --_nParked;
1935 
1936     if (_event >= 0) {
1937       ret = OS_OK;
1938     }
1939 
1940     _event = 0;
1941     status = pthread_mutex_unlock(_mutex);
1942     assert_status(status == 0, status, "mutex_unlock");
1943     // Paranoia to ensure our locked and lock-free paths interact
1944     // correctly with each other.
1945     OrderAccess::fence();
1946     return ret;
1947   }
1948   return OS_OK;
1949 }
1950 
1951 void os::PlatformEvent::unpark() {
1952   // Transitions for _event:
1953   //    0 => 1 : just return
1954   //    1 => 1 : just return
1955   //   -1 => either 0 or 1; must signal target thread
1956   //         That is, we can safely transition _event from -1 to either
1957   //         0 or 1.
1958   // See also: "Semaphores in Plan 9" by Mullender & Cox
1959   //
1960   // Note: Forcing a transition from "-1" to "1" on an unpark() means
1961   // that it will take two back-to-back park() calls for the owning
1962   // thread to block. This has the benefit of forcing a spurious return
1963   // from the first park() call after an unpark() call which will help
1964   // shake out uses of park() and unpark() without checking state conditions
1965   // properly. This spurious return doesn't manifest itself in any user code
1966   // but only in the correctly written condition checking loops of ObjectMonitor,
1967   // Mutex/Monitor, Thread::muxAcquire and os::sleep
1968 
1969   if (Atomic::xchg(1, &_event) >= 0) return;
1970 
1971   int status = pthread_mutex_lock(_mutex);
1972   assert_status(status == 0, status, "mutex_lock");
1973   int anyWaiters = _nParked;
1974   assert(anyWaiters == 0 || anyWaiters == 1, "invariant");
1975   status = pthread_mutex_unlock(_mutex);
1976   assert_status(status == 0, status, "mutex_unlock");
1977 
1978   // Note that we signal() *after* dropping the lock for "immortal" Events.
1979   // This is safe and avoids a common class of futile wakeups.  In rare
1980   // circumstances this can cause a thread to return prematurely from
1981   // cond_{timed}wait() but the spurious wakeup is benign and the victim
1982   // will simply re-test the condition and re-park itself.
1983   // This provides particular benefit if the underlying platform does not
1984   // provide wait morphing.
1985 
1986   if (anyWaiters != 0) {
1987     status = pthread_cond_signal(_cond);
1988     assert_status(status == 0, status, "cond_signal");
1989   }
1990 }
1991 
1992 // JSR166 support
1993 
1994  os::PlatformParker::PlatformParker() {
1995   int status;
1996   status = pthread_cond_init(&_cond[REL_INDEX], _condAttr);
1997   assert_status(status == 0, status, "cond_init rel");
1998   status = pthread_cond_init(&_cond[ABS_INDEX], NULL);
1999   assert_status(status == 0, status, "cond_init abs");
2000   status = pthread_mutex_init(_mutex, _mutexAttr);
2001   assert_status(status == 0, status, "mutex_init");
2002   _cur_index = -1; // mark as unused
2003 }
2004 
2005 // Parker::park decrements count if > 0, else does a condvar wait.  Unpark
2006 // sets count to 1 and signals condvar.  Only one thread ever waits
2007 // on the condvar. Contention seen when trying to park implies that someone
2008 // is unparking you, so don't wait. And spurious returns are fine, so there
2009 // is no need to track notifications.
2010 
2011 void Parker::park(bool isAbsolute, jlong time) {
2012 
2013   // Optional fast-path check:
2014   // Return immediately if a permit is available.
2015   // We depend on Atomic::xchg() having full barrier semantics
2016   // since we are doing a lock-free update to _counter.
2017   if (Atomic::xchg(0, &_counter) > 0) return;
2018 
2019   Thread* thread = Thread::current();
2020   assert(thread->is_Java_thread(), "Must be JavaThread");
2021   JavaThread *jt = (JavaThread *)thread;
2022 
2023   // Optional optimization -- avoid state transitions if there's
2024   // an interrupt pending.
2025   if (Thread::is_interrupted(thread, false)) {
2026     return;
2027   }
2028 
2029   // Next, demultiplex/decode time arguments
2030   struct timespec absTime;
2031   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
2032     return;
2033   }
2034   if (time > 0) {
2035     to_abstime(&absTime, time, isAbsolute);
2036   }
2037 
2038   // Enter safepoint region
2039   // Beware of deadlocks such as 6317397.
2040   // The per-thread Parker:: mutex is a classic leaf-lock.
2041   // In particular a thread must never block on the Threads_lock while
2042   // holding the Parker:: mutex.  If safepoints are pending both the
2043   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
2044   ThreadBlockInVM tbivm(jt);
2045 
2046   // Don't wait if cannot get lock since interference arises from
2047   // unparking. Also re-check interrupt before trying wait.
2048   if (Thread::is_interrupted(thread, false) ||
2049       pthread_mutex_trylock(_mutex) != 0) {
2050     return;
2051   }
2052 
2053   int status;
2054   if (_counter > 0)  { // no wait needed
2055     _counter = 0;
2056     status = pthread_mutex_unlock(_mutex);
2057     assert_status(status == 0, status, "invariant");
2058     // Paranoia to ensure our locked and lock-free paths interact
2059     // correctly with each other and Java-level accesses.
2060     OrderAccess::fence();
2061     return;
2062   }
2063 
2064   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2065   jt->set_suspend_equivalent();
2066   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2067 
2068   assert(_cur_index == -1, "invariant");
2069   if (time == 0) {
2070     _cur_index = REL_INDEX; // arbitrary choice when not timed
2071     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
2072     assert_status(status == 0, status, "cond_timedwait");
2073   }
2074   else {
2075     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
2076     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
2077     assert_status(status == 0 || status == ETIMEDOUT,
2078                   status, "cond_timedwait");
2079   }
2080   _cur_index = -1;
2081 
2082   _counter = 0;
2083   status = pthread_mutex_unlock(_mutex);
2084   assert_status(status == 0, status, "invariant");
2085   // Paranoia to ensure our locked and lock-free paths interact
2086   // correctly with each other and Java-level accesses.
2087   OrderAccess::fence();
2088 
2089   // If externally suspended while waiting, re-suspend
2090   if (jt->handle_special_suspend_equivalent_condition()) {
2091     jt->java_suspend_self();
2092   }
2093 }
2094 
2095 void Parker::unpark() {
2096   int status = pthread_mutex_lock(_mutex);
2097   assert_status(status == 0, status, "invariant");
2098   const int s = _counter;
2099   _counter = 1;
2100   // must capture correct index before unlocking
2101   int index = _cur_index;
2102   status = pthread_mutex_unlock(_mutex);
2103   assert_status(status == 0, status, "invariant");
2104 
2105   // Note that we signal() *after* dropping the lock for "immortal" Events.
2106   // This is safe and avoids a common class of futile wakeups.  In rare
2107   // circumstances this can cause a thread to return prematurely from
2108   // cond_{timed}wait() but the spurious wakeup is benign and the victim
2109   // will simply re-test the condition and re-park itself.
2110   // This provides particular benefit if the underlying platform does not
2111   // provide wait morphing.
2112 
2113   if (s < 1 && index != -1) {
2114     // thread is definitely parked
2115     status = pthread_cond_signal(&_cond[index]);
2116     assert_status(status == 0, status, "invariant");
2117   }
2118 }
2119 
2120 
2121 #endif // !SOLARIS