os_solaris.cpp revision 691:956304450e80
1/*
2 * Copyright 1997-2009 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
24
25// do not include  precompiled  header file
26# include "incls/_os_solaris.cpp.incl"
27
28// put OS-includes here
29# include <dlfcn.h>
30# include <errno.h>
31# include <link.h>
32# include <poll.h>
33# include <pthread.h>
34# include <pwd.h>
35# include <schedctl.h>
36# include <setjmp.h>
37# include <signal.h>
38# include <stdio.h>
39# include <alloca.h>
40# include <sys/filio.h>
41# include <sys/ipc.h>
42# include <sys/lwp.h>
43# include <sys/machelf.h>     // for elf Sym structure used by dladdr1
44# include <sys/mman.h>
45# include <sys/processor.h>
46# include <sys/procset.h>
47# include <sys/pset.h>
48# include <sys/resource.h>
49# include <sys/shm.h>
50# include <sys/socket.h>
51# include <sys/stat.h>
52# include <sys/systeminfo.h>
53# include <sys/time.h>
54# include <sys/times.h>
55# include <sys/types.h>
56# include <sys/wait.h>
57# include <sys/utsname.h>
58# include <thread.h>
59# include <unistd.h>
60# include <sys/priocntl.h>
61# include <sys/rtpriocntl.h>
62# include <sys/tspriocntl.h>
63# include <sys/iapriocntl.h>
64# include <sys/loadavg.h>
65# include <string.h>
66
67# define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
68# include <sys/procfs.h>     //  see comment in <sys/procfs.h>
69
70#define MAX_PATH (2 * K)
71
72// for timer info max values which include all bits
73#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
74
75#ifdef _GNU_SOURCE
76// See bug #6514594
77extern "C" int madvise(caddr_t, size_t, int);
78extern "C"  int memcntl(caddr_t addr, size_t len, int cmd, caddr_t  arg,
79     int attr, int mask);
80#endif //_GNU_SOURCE
81
82/*
83  MPSS Changes Start.
84  The JVM binary needs to be built and run on pre-Solaris 9
85  systems, but the constants needed by MPSS are only in Solaris 9
86  header files.  They are textually replicated here to allow
87  building on earlier systems.  Once building on Solaris 8 is
88  no longer a requirement, these #defines can be replaced by ordinary
89  system .h inclusion.
90
91  In earlier versions of the  JDK and Solaris, we used ISM for large pages.
92  But ISM requires shared memory to achieve this and thus has many caveats.
93  MPSS is a fully transparent and is a cleaner way to get large pages.
94  Although we still require keeping ISM for backward compatiblitiy as well as
95  giving the opportunity to use large pages on older systems it is
96  recommended that MPSS be used for Solaris 9 and above.
97
98*/
99
100#ifndef MC_HAT_ADVISE
101
102struct memcntl_mha {
103  uint_t          mha_cmd;        /* command(s) */
104  uint_t          mha_flags;
105  size_t          mha_pagesize;
106};
107#define MC_HAT_ADVISE   7       /* advise hat map size */
108#define MHA_MAPSIZE_VA  0x1     /* set preferred page size */
109#define MAP_ALIGN       0x200   /* addr specifies alignment */
110
111#endif
112// MPSS Changes End.
113
114
115// Here are some liblgrp types from sys/lgrp_user.h to be able to
116// compile on older systems without this header file.
117
118#ifndef MADV_ACCESS_LWP
119# define  MADV_ACCESS_LWP         7       /* next LWP to access heavily */
120#endif
121#ifndef MADV_ACCESS_MANY
122# define  MADV_ACCESS_MANY        8       /* many processes to access heavily */
123#endif
124
125#ifndef LGRP_RSRC_CPU
126# define LGRP_RSRC_CPU           0       /* CPU resources */
127#endif
128#ifndef LGRP_RSRC_MEM
129# define LGRP_RSRC_MEM           1       /* memory resources */
130#endif
131
132// Some more macros from sys/mman.h that are not present in Solaris 8.
133
134#ifndef MAX_MEMINFO_CNT
135/*
136 * info_req request type definitions for meminfo
137 * request types starting with MEMINFO_V are used for Virtual addresses
138 * and should not be mixed with MEMINFO_PLGRP which is targeted for Physical
139 * addresses
140 */
141# define MEMINFO_SHIFT           16
142# define MEMINFO_MASK            (0xFF << MEMINFO_SHIFT)
143# define MEMINFO_VPHYSICAL       (0x01 << MEMINFO_SHIFT) /* get physical addr */
144# define MEMINFO_VLGRP           (0x02 << MEMINFO_SHIFT) /* get lgroup */
145# define MEMINFO_VPAGESIZE       (0x03 << MEMINFO_SHIFT) /* size of phys page */
146# define MEMINFO_VREPLCNT        (0x04 << MEMINFO_SHIFT) /* no. of replica */
147# define MEMINFO_VREPL           (0x05 << MEMINFO_SHIFT) /* physical replica */
148# define MEMINFO_VREPL_LGRP      (0x06 << MEMINFO_SHIFT) /* lgrp of replica */
149# define MEMINFO_PLGRP           (0x07 << MEMINFO_SHIFT) /* lgroup for paddr */
150
151/* maximum number of addresses meminfo() can process at a time */
152# define MAX_MEMINFO_CNT 256
153
154/* maximum number of request types */
155# define MAX_MEMINFO_REQ 31
156#endif
157
158// see thr_setprio(3T) for the basis of these numbers
159#define MinimumPriority 0
160#define NormalPriority  64
161#define MaximumPriority 127
162
163// Values for ThreadPriorityPolicy == 1
164int prio_policy1[MaxPriority+1] = { -99999, 0, 16, 32, 48, 64,
165                                        80, 96, 112, 124, 127 };
166
167// System parameters used internally
168static clock_t clock_tics_per_sec = 100;
169
170// For diagnostics to print a message once. see run_periodic_checks
171static bool check_addr0_done = false;
172static sigset_t check_signal_done;
173static bool check_signals = true;
174
175address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
176address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
177
178address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
179
180
181// "default" initializers for missing libc APIs
182extern "C" {
183  static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
184  static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
185
186  static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
187  static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
188}
189
190// "default" initializers for pthread-based synchronization
191extern "C" {
192  static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
193  static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
194}
195
196// Thread Local Storage
197// This is common to all Solaris platforms so it is defined here,
198// in this common file.
199// The declarations are in the os_cpu threadLS*.hpp files.
200//
201// Static member initialization for TLS
202Thread* ThreadLocalStorage::_get_thread_cache[ThreadLocalStorage::_pd_cache_size] = {NULL};
203
204#ifndef PRODUCT
205#define _PCT(n,d)       ((100.0*(double)(n))/(double)(d))
206
207int ThreadLocalStorage::_tcacheHit = 0;
208int ThreadLocalStorage::_tcacheMiss = 0;
209
210void ThreadLocalStorage::print_statistics() {
211  int total = _tcacheMiss+_tcacheHit;
212  tty->print_cr("Thread cache hits %d misses %d total %d percent %f\n",
213                _tcacheHit, _tcacheMiss, total, _PCT(_tcacheHit, total));
214}
215#undef _PCT
216#endif // PRODUCT
217
218Thread* ThreadLocalStorage::get_thread_via_cache_slowly(uintptr_t raw_id,
219                                                        int index) {
220  Thread *thread = get_thread_slow();
221  if (thread != NULL) {
222    address sp = os::current_stack_pointer();
223    guarantee(thread->_stack_base == NULL ||
224              (sp <= thread->_stack_base &&
225                 sp >= thread->_stack_base - thread->_stack_size) ||
226               is_error_reported(),
227              "sp must be inside of selected thread stack");
228
229    thread->_self_raw_id = raw_id;  // mark for quick retrieval
230    _get_thread_cache[ index ] = thread;
231  }
232  return thread;
233}
234
235
236static const double all_zero[ sizeof(Thread) / sizeof(double) + 1 ] = {0};
237#define NO_CACHED_THREAD ((Thread*)all_zero)
238
239void ThreadLocalStorage::pd_set_thread(Thread* thread) {
240
241  // Store the new value before updating the cache to prevent a race
242  // between get_thread_via_cache_slowly() and this store operation.
243  os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
244
245  // Update thread cache with new thread if setting on thread create,
246  // or NO_CACHED_THREAD (zeroed) thread if resetting thread on exit.
247  uintptr_t raw = pd_raw_thread_id();
248  int ix = pd_cache_index(raw);
249  _get_thread_cache[ix] = thread == NULL ? NO_CACHED_THREAD : thread;
250}
251
252void ThreadLocalStorage::pd_init() {
253  for (int i = 0; i < _pd_cache_size; i++) {
254    _get_thread_cache[i] = NO_CACHED_THREAD;
255  }
256}
257
258// Invalidate all the caches (happens to be the same as pd_init).
259void ThreadLocalStorage::pd_invalidate_all() { pd_init(); }
260
261#undef NO_CACHED_THREAD
262
263// END Thread Local Storage
264
265static inline size_t adjust_stack_size(address base, size_t size) {
266  if ((ssize_t)size < 0) {
267    // 4759953: Compensate for ridiculous stack size.
268    size = max_intx;
269  }
270  if (size > (size_t)base) {
271    // 4812466: Make sure size doesn't allow the stack to wrap the address space.
272    size = (size_t)base;
273  }
274  return size;
275}
276
277static inline stack_t get_stack_info() {
278  stack_t st;
279  int retval = thr_stksegment(&st);
280  st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
281  assert(retval == 0, "incorrect return value from thr_stksegment");
282  assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
283  assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
284  return st;
285}
286
287address os::current_stack_base() {
288  int r = thr_main() ;
289  guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
290  bool is_primordial_thread = r;
291
292  // Workaround 4352906, avoid calls to thr_stksegment by
293  // thr_main after the first one (it looks like we trash
294  // some data, causing the value for ss_sp to be incorrect).
295  if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
296    stack_t st = get_stack_info();
297    if (is_primordial_thread) {
298      // cache initial value of stack base
299      os::Solaris::_main_stack_base = (address)st.ss_sp;
300    }
301    return (address)st.ss_sp;
302  } else {
303    guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
304    return os::Solaris::_main_stack_base;
305  }
306}
307
308size_t os::current_stack_size() {
309  size_t size;
310
311  int r = thr_main() ;
312  guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
313  if(!r) {
314    size = get_stack_info().ss_size;
315  } else {
316    struct rlimit limits;
317    getrlimit(RLIMIT_STACK, &limits);
318    size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
319  }
320  // base may not be page aligned
321  address base = current_stack_base();
322  address bottom = (address)align_size_up((intptr_t)(base - size), os::vm_page_size());;
323  return (size_t)(base - bottom);
324}
325
326struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
327  return localtime_r(clock, res);
328}
329
330// interruptible infrastructure
331
332// setup_interruptible saves the thread state before going into an
333// interruptible system call.
334// The saved state is used to restore the thread to
335// its former state whether or not an interrupt is received.
336// Used by classloader os::read
337// hpi calls skip this layer and stay in _thread_in_native
338
339void os::Solaris::setup_interruptible(JavaThread* thread) {
340
341  JavaThreadState thread_state = thread->thread_state();
342
343  assert(thread_state != _thread_blocked, "Coming from the wrong thread");
344  assert(thread_state != _thread_in_native, "Native threads skip setup_interruptible");
345  OSThread* osthread = thread->osthread();
346  osthread->set_saved_interrupt_thread_state(thread_state);
347  thread->frame_anchor()->make_walkable(thread);
348  ThreadStateTransition::transition(thread, thread_state, _thread_blocked);
349}
350
351// Version of setup_interruptible() for threads that are already in
352// _thread_blocked. Used by os_sleep().
353void os::Solaris::setup_interruptible_already_blocked(JavaThread* thread) {
354  thread->frame_anchor()->make_walkable(thread);
355}
356
357JavaThread* os::Solaris::setup_interruptible() {
358  JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
359  setup_interruptible(thread);
360  return thread;
361}
362
363void os::Solaris::try_enable_extended_io() {
364  typedef int (*enable_extended_FILE_stdio_t)(int, int);
365
366  if (!UseExtendedFileIO) {
367    return;
368  }
369
370  enable_extended_FILE_stdio_t enabler =
371    (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
372                                         "enable_extended_FILE_stdio");
373  if (enabler) {
374    enabler(-1, -1);
375  }
376}
377
378
379#ifdef ASSERT
380
381JavaThread* os::Solaris::setup_interruptible_native() {
382  JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
383  JavaThreadState thread_state = thread->thread_state();
384  assert(thread_state == _thread_in_native, "Assumed thread_in_native");
385  return thread;
386}
387
388void os::Solaris::cleanup_interruptible_native(JavaThread* thread) {
389  JavaThreadState thread_state = thread->thread_state();
390  assert(thread_state == _thread_in_native, "Assumed thread_in_native");
391}
392#endif
393
394// cleanup_interruptible reverses the effects of setup_interruptible
395// setup_interruptible_already_blocked() does not need any cleanup.
396
397void os::Solaris::cleanup_interruptible(JavaThread* thread) {
398  OSThread* osthread = thread->osthread();
399
400  ThreadStateTransition::transition(thread, _thread_blocked, osthread->saved_interrupt_thread_state());
401}
402
403// I/O interruption related counters called in _INTERRUPTIBLE
404
405void os::Solaris::bump_interrupted_before_count() {
406  RuntimeService::record_interrupted_before_count();
407}
408
409void os::Solaris::bump_interrupted_during_count() {
410  RuntimeService::record_interrupted_during_count();
411}
412
413static int _processors_online = 0;
414
415         jint os::Solaris::_os_thread_limit = 0;
416volatile jint os::Solaris::_os_thread_count = 0;
417
418julong os::available_memory() {
419  return Solaris::available_memory();
420}
421
422julong os::Solaris::available_memory() {
423  return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
424}
425
426julong os::Solaris::_physical_memory = 0;
427
428julong os::physical_memory() {
429   return Solaris::physical_memory();
430}
431
432julong os::allocatable_physical_memory(julong size) {
433#ifdef _LP64
434   return size;
435#else
436   julong result = MIN2(size, (julong)3835*M);
437   if (!is_allocatable(result)) {
438     // Memory allocations will be aligned but the alignment
439     // is not known at this point.  Alignments will
440     // be at most to LargePageSizeInBytes.  Protect
441     // allocations from alignments up to illegal
442     // values. If at this point 2G is illegal.
443     julong reasonable_size = (julong)2*G - 2 * LargePageSizeInBytes;
444     result =  MIN2(size, reasonable_size);
445   }
446   return result;
447#endif
448}
449
450static hrtime_t first_hrtime = 0;
451static const hrtime_t hrtime_hz = 1000*1000*1000;
452const int LOCK_BUSY = 1;
453const int LOCK_FREE = 0;
454const int LOCK_INVALID = -1;
455static volatile hrtime_t max_hrtime = 0;
456static volatile int max_hrtime_lock = LOCK_FREE;     // Update counter with LSB as lock-in-progress
457
458
459void os::Solaris::initialize_system_info() {
460  _processor_count = sysconf(_SC_NPROCESSORS_CONF);
461  _processors_online = sysconf (_SC_NPROCESSORS_ONLN);
462  _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
463}
464
465int os::active_processor_count() {
466  int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
467  pid_t pid = getpid();
468  psetid_t pset = PS_NONE;
469  // Are we running in a processor set or is there any processor set around?
470  if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
471    uint_t pset_cpus;
472    // Query the number of cpus available to us.
473    if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
474      assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
475      _processors_online = pset_cpus;
476      return pset_cpus;
477    }
478  }
479  // Otherwise return number of online cpus
480  return online_cpus;
481}
482
483static bool find_processors_in_pset(psetid_t        pset,
484                                    processorid_t** id_array,
485                                    uint_t*         id_length) {
486  bool result = false;
487  // Find the number of processors in the processor set.
488  if (pset_info(pset, NULL, id_length, NULL) == 0) {
489    // Make up an array to hold their ids.
490    *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length);
491    // Fill in the array with their processor ids.
492    if (pset_info(pset, NULL, id_length, *id_array) == 0) {
493      result = true;
494    }
495  }
496  return result;
497}
498
499// Callers of find_processors_online() must tolerate imprecise results --
500// the system configuration can change asynchronously because of DR
501// or explicit psradm operations.
502//
503// We also need to take care that the loop (below) terminates as the
504// number of processors online can change between the _SC_NPROCESSORS_ONLN
505// request and the loop that builds the list of processor ids.   Unfortunately
506// there's no reliable way to determine the maximum valid processor id,
507// so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
508// man pages, which claim the processor id set is "sparse, but
509// not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
510// exit the loop.
511//
512// In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
513// not available on S8.0.
514
515static bool find_processors_online(processorid_t** id_array,
516                                   uint*           id_length) {
517  const processorid_t MAX_PROCESSOR_ID = 100000 ;
518  // Find the number of processors online.
519  *id_length = sysconf(_SC_NPROCESSORS_ONLN);
520  // Make up an array to hold their ids.
521  *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length);
522  // Processors need not be numbered consecutively.
523  long found = 0;
524  processorid_t next = 0;
525  while (found < *id_length && next < MAX_PROCESSOR_ID) {
526    processor_info_t info;
527    if (processor_info(next, &info) == 0) {
528      // NB, PI_NOINTR processors are effectively online ...
529      if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
530        (*id_array)[found] = next;
531        found += 1;
532      }
533    }
534    next += 1;
535  }
536  if (found < *id_length) {
537      // The loop above didn't identify the expected number of processors.
538      // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
539      // and re-running the loop, above, but there's no guarantee of progress
540      // if the system configuration is in flux.  Instead, we just return what
541      // we've got.  Note that in the worst case find_processors_online() could
542      // return an empty set.  (As a fall-back in the case of the empty set we
543      // could just return the ID of the current processor).
544      *id_length = found ;
545  }
546
547  return true;
548}
549
550static bool assign_distribution(processorid_t* id_array,
551                                uint           id_length,
552                                uint*          distribution,
553                                uint           distribution_length) {
554  // We assume we can assign processorid_t's to uint's.
555  assert(sizeof(processorid_t) == sizeof(uint),
556         "can't convert processorid_t to uint");
557  // Quick check to see if we won't succeed.
558  if (id_length < distribution_length) {
559    return false;
560  }
561  // Assign processor ids to the distribution.
562  // Try to shuffle processors to distribute work across boards,
563  // assuming 4 processors per board.
564  const uint processors_per_board = ProcessDistributionStride;
565  // Find the maximum processor id.
566  processorid_t max_id = 0;
567  for (uint m = 0; m < id_length; m += 1) {
568    max_id = MAX2(max_id, id_array[m]);
569  }
570  // The next id, to limit loops.
571  const processorid_t limit_id = max_id + 1;
572  // Make up markers for available processors.
573  bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id);
574  for (uint c = 0; c < limit_id; c += 1) {
575    available_id[c] = false;
576  }
577  for (uint a = 0; a < id_length; a += 1) {
578    available_id[id_array[a]] = true;
579  }
580  // Step by "boards", then by "slot", copying to "assigned".
581  // NEEDS_CLEANUP: The assignment of processors should be stateful,
582  //                remembering which processors have been assigned by
583  //                previous calls, etc., so as to distribute several
584  //                independent calls of this method.  What we'd like is
585  //                It would be nice to have an API that let us ask
586  //                how many processes are bound to a processor,
587  //                but we don't have that, either.
588  //                In the short term, "board" is static so that
589  //                subsequent distributions don't all start at board 0.
590  static uint board = 0;
591  uint assigned = 0;
592  // Until we've found enough processors ....
593  while (assigned < distribution_length) {
594    // ... find the next available processor in the board.
595    for (uint slot = 0; slot < processors_per_board; slot += 1) {
596      uint try_id = board * processors_per_board + slot;
597      if ((try_id < limit_id) && (available_id[try_id] == true)) {
598        distribution[assigned] = try_id;
599        available_id[try_id] = false;
600        assigned += 1;
601        break;
602      }
603    }
604    board += 1;
605    if (board * processors_per_board + 0 >= limit_id) {
606      board = 0;
607    }
608  }
609  if (available_id != NULL) {
610    FREE_C_HEAP_ARRAY(bool, available_id);
611  }
612  return true;
613}
614
615bool os::distribute_processes(uint length, uint* distribution) {
616  bool result = false;
617  // Find the processor id's of all the available CPUs.
618  processorid_t* id_array  = NULL;
619  uint           id_length = 0;
620  // There are some races between querying information and using it,
621  // since processor sets can change dynamically.
622  psetid_t pset = PS_NONE;
623  // Are we running in a processor set?
624  if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
625    result = find_processors_in_pset(pset, &id_array, &id_length);
626  } else {
627    result = find_processors_online(&id_array, &id_length);
628  }
629  if (result == true) {
630    if (id_length >= length) {
631      result = assign_distribution(id_array, id_length, distribution, length);
632    } else {
633      result = false;
634    }
635  }
636  if (id_array != NULL) {
637    FREE_C_HEAP_ARRAY(processorid_t, id_array);
638  }
639  return result;
640}
641
642bool os::bind_to_processor(uint processor_id) {
643  // We assume that a processorid_t can be stored in a uint.
644  assert(sizeof(uint) == sizeof(processorid_t),
645         "can't convert uint to processorid_t");
646  int bind_result =
647    processor_bind(P_LWPID,                       // bind LWP.
648                   P_MYID,                        // bind current LWP.
649                   (processorid_t) processor_id,  // id.
650                   NULL);                         // don't return old binding.
651  return (bind_result == 0);
652}
653
654bool os::getenv(const char* name, char* buffer, int len) {
655  char* val = ::getenv( name );
656  if ( val == NULL
657  ||   strlen(val) + 1  >  len ) {
658    if (len > 0)  buffer[0] = 0; // return a null string
659    return false;
660  }
661  strcpy( buffer, val );
662  return true;
663}
664
665
666// Return true if user is running as root.
667
668bool os::have_special_privileges() {
669  static bool init = false;
670  static bool privileges = false;
671  if (!init) {
672    privileges = (getuid() != geteuid()) || (getgid() != getegid());
673    init = true;
674  }
675  return privileges;
676}
677
678
679static char* get_property(char* name, char* buffer, int buffer_size) {
680  if (os::getenv(name, buffer, buffer_size)) {
681    return buffer;
682  }
683  static char empty[] = "";
684  return empty;
685}
686
687
688void os::init_system_properties_values() {
689  char arch[12];
690  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
691
692  // The next steps are taken in the product version:
693  //
694  // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
695  // This library should be located at:
696  // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
697  //
698  // If "/jre/lib/" appears at the right place in the path, then we
699  // assume libjvm[_g].so is installed in a JDK and we use this path.
700  //
701  // Otherwise exit with message: "Could not create the Java virtual machine."
702  //
703  // The following extra steps are taken in the debugging version:
704  //
705  // If "/jre/lib/" does NOT appear at the right place in the path
706  // instead of exit check for $JAVA_HOME environment variable.
707  //
708  // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
709  // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
710  // it looks like libjvm[_g].so is installed there
711  // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
712  //
713  // Otherwise exit.
714  //
715  // Important note: if the location of libjvm.so changes this
716  // code needs to be changed accordingly.
717
718  // The next few definitions allow the code to be verbatim:
719#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
720#define free(p) FREE_C_HEAP_ARRAY(char, p)
721#define getenv(n) ::getenv(n)
722
723#define EXTENSIONS_DIR  "/lib/ext"
724#define ENDORSED_DIR    "/lib/endorsed"
725#define COMMON_DIR      "/usr/jdk/packages"
726
727  {
728    /* sysclasspath, java_home, dll_dir */
729    {
730        char *home_path;
731        char *dll_path;
732        char *pslash;
733        char buf[MAXPATHLEN];
734        os::jvm_path(buf, sizeof(buf));
735
736        // Found the full path to libjvm.so.
737        // Now cut the path to <java_home>/jre if we can.
738        *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
739        pslash = strrchr(buf, '/');
740        if (pslash != NULL)
741            *pslash = '\0';           /* get rid of /{client|server|hotspot} */
742        dll_path = malloc(strlen(buf) + 1);
743        if (dll_path == NULL)
744            return;
745        strcpy(dll_path, buf);
746        Arguments::set_dll_dir(dll_path);
747
748        if (pslash != NULL) {
749            pslash = strrchr(buf, '/');
750            if (pslash != NULL) {
751                *pslash = '\0';       /* get rid of /<arch> */
752                pslash = strrchr(buf, '/');
753                if (pslash != NULL)
754                    *pslash = '\0';   /* get rid of /lib */
755            }
756        }
757
758        home_path = malloc(strlen(buf) + 1);
759        if (home_path == NULL)
760            return;
761        strcpy(home_path, buf);
762        Arguments::set_java_home(home_path);
763
764        if (!set_boot_path('/', ':'))
765            return;
766    }
767
768    /*
769     * Where to look for native libraries
770     */
771    {
772      // Use dlinfo() to determine the correct java.library.path.
773      //
774      // If we're launched by the Java launcher, and the user
775      // does not set java.library.path explicitly on the commandline,
776      // the Java launcher sets LD_LIBRARY_PATH for us and unsets
777      // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
778      // dlinfo returns LD_LIBRARY_PATH + crle settings (including
779      // /usr/lib), which is exactly what we want.
780      //
781      // If the user does set java.library.path, it completely
782      // overwrites this setting, and always has.
783      //
784      // If we're not launched by the Java launcher, we may
785      // get here with any/all of the LD_LIBRARY_PATH[_32|64]
786      // settings.  Again, dlinfo does exactly what we want.
787
788      Dl_serinfo     _info, *info = &_info;
789      Dl_serpath     *path;
790      char*          library_path;
791      char           *common_path;
792      int            i;
793
794      // determine search path count and required buffer size
795      if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
796        vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
797      }
798
799      // allocate new buffer and initialize
800      info = (Dl_serinfo*)malloc(_info.dls_size);
801      if (info == NULL) {
802        vm_exit_out_of_memory(_info.dls_size,
803                              "init_system_properties_values info");
804      }
805      info->dls_size = _info.dls_size;
806      info->dls_cnt = _info.dls_cnt;
807
808      // obtain search path information
809      if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
810        free(info);
811        vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
812      }
813
814      path = &info->dls_serpath[0];
815
816      // Note: Due to a legacy implementation, most of the library path
817      // is set in the launcher.  This was to accomodate linking restrictions
818      // on legacy Solaris implementations (which are no longer supported).
819      // Eventually, all the library path setting will be done here.
820      //
821      // However, to prevent the proliferation of improperly built native
822      // libraries, the new path component /usr/jdk/packages is added here.
823
824      // Determine the actual CPU architecture.
825      char cpu_arch[12];
826      sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
827#ifdef _LP64
828      // If we are a 64-bit vm, perform the following translations:
829      //   sparc   -> sparcv9
830      //   i386    -> amd64
831      if (strcmp(cpu_arch, "sparc") == 0)
832        strcat(cpu_arch, "v9");
833      else if (strcmp(cpu_arch, "i386") == 0)
834        strcpy(cpu_arch, "amd64");
835#endif
836
837      // Construct the invariant part of ld_library_path. Note that the
838      // space for the colon and the trailing null are provided by the
839      // nulls included by the sizeof operator.
840      size_t bufsize = sizeof(COMMON_DIR) + sizeof("/lib/") + strlen(cpu_arch);
841      common_path = malloc(bufsize);
842      if (common_path == NULL) {
843        free(info);
844        vm_exit_out_of_memory(bufsize,
845                              "init_system_properties_values common_path");
846      }
847      sprintf(common_path, COMMON_DIR "/lib/%s", cpu_arch);
848
849      // struct size is more than sufficient for the path components obtained
850      // through the dlinfo() call, so only add additional space for the path
851      // components explicitly added here.
852      bufsize = info->dls_size + strlen(common_path);
853      library_path = malloc(bufsize);
854      if (library_path == NULL) {
855        free(info);
856        free(common_path);
857        vm_exit_out_of_memory(bufsize,
858                              "init_system_properties_values library_path");
859      }
860      library_path[0] = '\0';
861
862      // Construct the desired Java library path from the linker's library
863      // search path.
864      //
865      // For compatibility, it is optimal that we insert the additional path
866      // components specific to the Java VM after those components specified
867      // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
868      // infrastructure.
869      if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it
870        strcpy(library_path, common_path);
871      } else {
872        int inserted = 0;
873        for (i = 0; i < info->dls_cnt; i++, path++) {
874          uint_t flags = path->dls_flags & LA_SER_MASK;
875          if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
876            strcat(library_path, common_path);
877            strcat(library_path, os::path_separator());
878            inserted = 1;
879          }
880          strcat(library_path, path->dls_name);
881          strcat(library_path, os::path_separator());
882        }
883        // eliminate trailing path separator
884        library_path[strlen(library_path)-1] = '\0';
885      }
886
887      // happens before argument parsing - can't use a trace flag
888      // tty->print_raw("init_system_properties_values: native lib path: ");
889      // tty->print_raw_cr(library_path);
890
891      // callee copies into its own buffer
892      Arguments::set_library_path(library_path);
893
894      free(common_path);
895      free(library_path);
896      free(info);
897    }
898
899    /*
900     * Extensions directories.
901     *
902     * Note that the space for the colon and the trailing null are provided
903     * by the nulls included by the sizeof operator (so actually one byte more
904     * than necessary is allocated).
905     */
906    {
907        char *buf = (char *) malloc(strlen(Arguments::get_java_home()) +
908            sizeof(EXTENSIONS_DIR) + sizeof(COMMON_DIR) +
909            sizeof(EXTENSIONS_DIR));
910        sprintf(buf, "%s" EXTENSIONS_DIR ":" COMMON_DIR EXTENSIONS_DIR,
911            Arguments::get_java_home());
912        Arguments::set_ext_dirs(buf);
913    }
914
915    /* Endorsed standards default directory. */
916    {
917        char * buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
918        sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
919        Arguments::set_endorsed_dirs(buf);
920    }
921  }
922
923#undef malloc
924#undef free
925#undef getenv
926#undef EXTENSIONS_DIR
927#undef ENDORSED_DIR
928#undef COMMON_DIR
929
930}
931
932void os::breakpoint() {
933  BREAKPOINT;
934}
935
936bool os::obsolete_option(const JavaVMOption *option)
937{
938  if (!strncmp(option->optionString, "-Xt", 3)) {
939    return true;
940  } else if (!strncmp(option->optionString, "-Xtm", 4)) {
941    return true;
942  } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
943    return true;
944  } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
945    return true;
946  }
947  return false;
948}
949
950bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
951  address  stackStart  = (address)thread->stack_base();
952  address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
953  if (sp < stackStart && sp >= stackEnd ) return true;
954  return false;
955}
956
957extern "C" void breakpoint() {
958  // use debugger to set breakpoint here
959}
960
961// Returns an estimate of the current stack pointer. Result must be guaranteed to
962// point into the calling threads stack, and be no lower than the current stack
963// pointer.
964address os::current_stack_pointer() {
965  volatile int dummy;
966  address sp = (address)&dummy + 8;     // %%%% need to confirm if this is right
967  return sp;
968}
969
970static thread_t main_thread;
971
972// Thread start routine for all new Java threads
973extern "C" void* java_start(void* thread_addr) {
974  // Try to randomize the cache line index of hot stack frames.
975  // This helps when threads of the same stack traces evict each other's
976  // cache lines. The threads can be either from the same JVM instance, or
977  // from different JVM instances. The benefit is especially true for
978  // processors with hyperthreading technology.
979  static int counter = 0;
980  int pid = os::current_process_id();
981  alloca(((pid ^ counter++) & 7) * 128);
982
983  int prio;
984  Thread* thread = (Thread*)thread_addr;
985  OSThread* osthr = thread->osthread();
986
987  osthr->set_lwp_id( _lwp_self() );  // Store lwp in case we are bound
988  thread->_schedctl = (void *) schedctl_init () ;
989
990  if (UseNUMA) {
991    int lgrp_id = os::numa_get_group_id();
992    if (lgrp_id != -1) {
993      thread->set_lgrp_id(lgrp_id);
994    }
995  }
996
997  // If the creator called set priority before we started,
998  // we need to call set priority now that we have an lwp.
999  // Get the priority from libthread and set the priority
1000  // for the new Solaris lwp.
1001  if ( osthr->thread_id() != -1 ) {
1002    if ( UseThreadPriorities ) {
1003      thr_getprio(osthr->thread_id(), &prio);
1004      if (ThreadPriorityVerbose) {
1005        tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT ", setting priority: %d\n",
1006                      osthr->thread_id(), osthr->lwp_id(), prio );
1007      }
1008      os::set_native_priority(thread, prio);
1009    }
1010  } else if (ThreadPriorityVerbose) {
1011    warning("Can't set priority in _start routine, thread id hasn't been set\n");
1012  }
1013
1014  assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
1015
1016  // initialize signal mask for this thread
1017  os::Solaris::hotspot_sigmask(thread);
1018
1019  thread->run();
1020
1021  // One less thread is executing
1022  // When the VMThread gets here, the main thread may have already exited
1023  // which frees the CodeHeap containing the Atomic::dec code
1024  if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
1025    Atomic::dec(&os::Solaris::_os_thread_count);
1026  }
1027
1028  if (UseDetachedThreads) {
1029    thr_exit(NULL);
1030    ShouldNotReachHere();
1031  }
1032  return NULL;
1033}
1034
1035static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
1036  // Allocate the OSThread object
1037  OSThread* osthread = new OSThread(NULL, NULL);
1038  if (osthread == NULL) return NULL;
1039
1040  // Store info on the Solaris thread into the OSThread
1041  osthread->set_thread_id(thread_id);
1042  osthread->set_lwp_id(_lwp_self());
1043  thread->_schedctl = (void *) schedctl_init () ;
1044
1045  if (UseNUMA) {
1046    int lgrp_id = os::numa_get_group_id();
1047    if (lgrp_id != -1) {
1048      thread->set_lgrp_id(lgrp_id);
1049    }
1050  }
1051
1052  if ( ThreadPriorityVerbose ) {
1053    tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
1054                  osthread->thread_id(), osthread->lwp_id() );
1055  }
1056
1057  // Initial thread state is INITIALIZED, not SUSPENDED
1058  osthread->set_state(INITIALIZED);
1059
1060  return osthread;
1061}
1062
1063void os::Solaris::hotspot_sigmask(Thread* thread) {
1064
1065  //Save caller's signal mask
1066  sigset_t sigmask;
1067  thr_sigsetmask(SIG_SETMASK, NULL, &sigmask);
1068  OSThread *osthread = thread->osthread();
1069  osthread->set_caller_sigmask(sigmask);
1070
1071  thr_sigsetmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
1072  if (!ReduceSignalUsage) {
1073    if (thread->is_VM_thread()) {
1074      // Only the VM thread handles BREAK_SIGNAL ...
1075      thr_sigsetmask(SIG_UNBLOCK, vm_signals(), NULL);
1076    } else {
1077      // ... all other threads block BREAK_SIGNAL
1078      assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
1079      thr_sigsetmask(SIG_BLOCK, vm_signals(), NULL);
1080    }
1081  }
1082}
1083
1084bool os::create_attached_thread(JavaThread* thread) {
1085#ifdef ASSERT
1086  thread->verify_not_published();
1087#endif
1088  OSThread* osthread = create_os_thread(thread, thr_self());
1089  if (osthread == NULL) {
1090     return false;
1091  }
1092
1093  // Initial thread state is RUNNABLE
1094  osthread->set_state(RUNNABLE);
1095  thread->set_osthread(osthread);
1096
1097  // initialize signal mask for this thread
1098  // and save the caller's signal mask
1099  os::Solaris::hotspot_sigmask(thread);
1100
1101  return true;
1102}
1103
1104bool os::create_main_thread(JavaThread* thread) {
1105#ifdef ASSERT
1106  thread->verify_not_published();
1107#endif
1108  if (_starting_thread == NULL) {
1109    _starting_thread = create_os_thread(thread, main_thread);
1110     if (_starting_thread == NULL) {
1111        return false;
1112     }
1113  }
1114
1115  // The primodial thread is runnable from the start
1116  _starting_thread->set_state(RUNNABLE);
1117
1118  thread->set_osthread(_starting_thread);
1119
1120  // initialize signal mask for this thread
1121  // and save the caller's signal mask
1122  os::Solaris::hotspot_sigmask(thread);
1123
1124  return true;
1125}
1126
1127// _T2_libthread is true if we believe we are running with the newer
1128// SunSoft lwp/libthread.so (2.8 patch, 2.9 default)
1129bool os::Solaris::_T2_libthread = false;
1130
1131bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1132  // Allocate the OSThread object
1133  OSThread* osthread = new OSThread(NULL, NULL);
1134  if (osthread == NULL) {
1135    return false;
1136  }
1137
1138  if ( ThreadPriorityVerbose ) {
1139    char *thrtyp;
1140    switch ( thr_type ) {
1141      case vm_thread:
1142        thrtyp = (char *)"vm";
1143        break;
1144      case cgc_thread:
1145        thrtyp = (char *)"cgc";
1146        break;
1147      case pgc_thread:
1148        thrtyp = (char *)"pgc";
1149        break;
1150      case java_thread:
1151        thrtyp = (char *)"java";
1152        break;
1153      case compiler_thread:
1154        thrtyp = (char *)"compiler";
1155        break;
1156      case watcher_thread:
1157        thrtyp = (char *)"watcher";
1158        break;
1159      default:
1160        thrtyp = (char *)"unknown";
1161        break;
1162    }
1163    tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
1164  }
1165
1166  // Calculate stack size if it's not specified by caller.
1167  if (stack_size == 0) {
1168    // The default stack size 1M (2M for LP64).
1169    stack_size = (BytesPerWord >> 2) * K * K;
1170
1171    switch (thr_type) {
1172    case os::java_thread:
1173      // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
1174      if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
1175      break;
1176    case os::compiler_thread:
1177      if (CompilerThreadStackSize > 0) {
1178        stack_size = (size_t)(CompilerThreadStackSize * K);
1179        break;
1180      } // else fall through:
1181        // use VMThreadStackSize if CompilerThreadStackSize is not defined
1182    case os::vm_thread:
1183    case os::pgc_thread:
1184    case os::cgc_thread:
1185    case os::watcher_thread:
1186      if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1187      break;
1188    }
1189  }
1190  stack_size = MAX2(stack_size, os::Solaris::min_stack_allowed);
1191
1192  // Initial state is ALLOCATED but not INITIALIZED
1193  osthread->set_state(ALLOCATED);
1194
1195  if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
1196    // We got lots of threads. Check if we still have some address space left.
1197    // Need to be at least 5Mb of unreserved address space. We do check by
1198    // trying to reserve some.
1199    const size_t VirtualMemoryBangSize = 20*K*K;
1200    char* mem = os::reserve_memory(VirtualMemoryBangSize);
1201    if (mem == NULL) {
1202      delete osthread;
1203      return false;
1204    } else {
1205      // Release the memory again
1206      os::release_memory(mem, VirtualMemoryBangSize);
1207    }
1208  }
1209
1210  // Setup osthread because the child thread may need it.
1211  thread->set_osthread(osthread);
1212
1213  // Create the Solaris thread
1214  // explicit THR_BOUND for T2_libthread case in case
1215  // that assumption is not accurate, but our alternate signal stack
1216  // handling is based on it which must have bound threads
1217  thread_t tid = 0;
1218  long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED
1219                   | ((UseBoundThreads || os::Solaris::T2_libthread() ||
1220                       (thr_type == vm_thread) ||
1221                       (thr_type == cgc_thread) ||
1222                       (thr_type == pgc_thread) ||
1223                       (thr_type == compiler_thread && BackgroundCompilation)) ?
1224                      THR_BOUND : 0);
1225  int      status;
1226
1227  // 4376845 -- libthread/kernel don't provide enough LWPs to utilize all CPUs.
1228  //
1229  // On multiprocessors systems, libthread sometimes under-provisions our
1230  // process with LWPs.  On a 30-way systems, for instance, we could have
1231  // 50 user-level threads in ready state and only 2 or 3 LWPs assigned
1232  // to our process.  This can result in under utilization of PEs.
1233  // I suspect the problem is related to libthread's LWP
1234  // pool management and to the kernel's SIGBLOCKING "last LWP parked"
1235  // upcall policy.
1236  //
1237  // The following code is palliative -- it attempts to ensure that our
1238  // process has sufficient LWPs to take advantage of multiple PEs.
1239  // Proper long-term cures include using user-level threads bound to LWPs
1240  // (THR_BOUND) or using LWP-based synchronization.  Note that there is a
1241  // slight timing window with respect to sampling _os_thread_count, but
1242  // the race is benign.  Also, we should periodically recompute
1243  // _processors_online as the min of SC_NPROCESSORS_ONLN and the
1244  // the number of PEs in our partition.  You might be tempted to use
1245  // THR_NEW_LWP here, but I'd recommend against it as that could
1246  // result in undesirable growth of the libthread's LWP pool.
1247  // The fix below isn't sufficient; for instance, it doesn't take into count
1248  // LWPs parked on IO.  It does, however, help certain CPU-bound benchmarks.
1249  //
1250  // Some pathologies this scheme doesn't handle:
1251  // *  Threads can block, releasing the LWPs.  The LWPs can age out.
1252  //    When a large number of threads become ready again there aren't
1253  //    enough LWPs available to service them.  This can occur when the
1254  //    number of ready threads oscillates.
1255  // *  LWPs/Threads park on IO, thus taking the LWP out of circulation.
1256  //
1257  // Finally, we should call thr_setconcurrency() periodically to refresh
1258  // the LWP pool and thwart the LWP age-out mechanism.
1259  // The "+3" term provides a little slop -- we want to slightly overprovision.
1260
1261  if (AdjustConcurrency && os::Solaris::_os_thread_count < (_processors_online+3)) {
1262    if (!(flags & THR_BOUND)) {
1263      thr_setconcurrency (os::Solaris::_os_thread_count);       // avoid starvation
1264    }
1265  }
1266  // Although this doesn't hurt, we should warn of undefined behavior
1267  // when using unbound T1 threads with schedctl().  This should never
1268  // happen, as the compiler and VM threads are always created bound
1269  DEBUG_ONLY(
1270      if ((VMThreadHintNoPreempt || CompilerThreadHintNoPreempt) &&
1271          (!os::Solaris::T2_libthread() && (!(flags & THR_BOUND))) &&
1272          ((thr_type == vm_thread) || (thr_type == cgc_thread) ||
1273           (thr_type == pgc_thread) || (thr_type == compiler_thread && BackgroundCompilation))) {
1274         warning("schedctl behavior undefined when Compiler/VM/GC Threads are Unbound");
1275      }
1276  );
1277
1278
1279  // Mark that we don't have an lwp or thread id yet.
1280  // In case we attempt to set the priority before the thread starts.
1281  osthread->set_lwp_id(-1);
1282  osthread->set_thread_id(-1);
1283
1284  status = thr_create(NULL, stack_size, java_start, thread, flags, &tid);
1285  if (status != 0) {
1286    if (PrintMiscellaneous && (Verbose || WizardMode)) {
1287      perror("os::create_thread");
1288    }
1289    thread->set_osthread(NULL);
1290    // Need to clean up stuff we've allocated so far
1291    delete osthread;
1292    return false;
1293  }
1294
1295  Atomic::inc(&os::Solaris::_os_thread_count);
1296
1297  // Store info on the Solaris thread into the OSThread
1298  osthread->set_thread_id(tid);
1299
1300  // Remember that we created this thread so we can set priority on it
1301  osthread->set_vm_created();
1302
1303  // Set the default thread priority otherwise use NormalPriority
1304
1305  if ( UseThreadPriorities ) {
1306     thr_setprio(tid, (DefaultThreadPriority == -1) ?
1307                        java_to_os_priority[NormPriority] :
1308                        DefaultThreadPriority);
1309  }
1310
1311  // Initial thread state is INITIALIZED, not SUSPENDED
1312  osthread->set_state(INITIALIZED);
1313
1314  // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1315  return true;
1316}
1317
1318/* defined for >= Solaris 10. This allows builds on earlier versions
1319 *  of Solaris to take advantage of the newly reserved Solaris JVM signals
1320 *  With SIGJVM1, SIGJVM2, INTERRUPT_SIGNAL is SIGJVM1, ASYNC_SIGNAL is SIGJVM2
1321 *  and -XX:+UseAltSigs does nothing since these should have no conflict
1322 */
1323#if !defined(SIGJVM1)
1324#define SIGJVM1 39
1325#define SIGJVM2 40
1326#endif
1327
1328debug_only(static bool signal_sets_initialized = false);
1329static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
1330int os::Solaris::_SIGinterrupt = INTERRUPT_SIGNAL;
1331int os::Solaris::_SIGasync = ASYNC_SIGNAL;
1332
1333bool os::Solaris::is_sig_ignored(int sig) {
1334      struct sigaction oact;
1335      sigaction(sig, (struct sigaction*)NULL, &oact);
1336      void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1337                                     : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1338      if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
1339           return true;
1340      else
1341           return false;
1342}
1343
1344// Note: SIGRTMIN is a macro that calls sysconf() so it will
1345// dynamically detect SIGRTMIN value for the system at runtime, not buildtime
1346static bool isJVM1available() {
1347  return SIGJVM1 < SIGRTMIN;
1348}
1349
1350void os::Solaris::signal_sets_init() {
1351  // Should also have an assertion stating we are still single-threaded.
1352  assert(!signal_sets_initialized, "Already initialized");
1353  // Fill in signals that are necessarily unblocked for all threads in
1354  // the VM. Currently, we unblock the following signals:
1355  // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1356  //                         by -Xrs (=ReduceSignalUsage));
1357  // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1358  // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1359  // the dispositions or masks wrt these signals.
1360  // Programs embedding the VM that want to use the above signals for their
1361  // own purposes must, at this time, use the "-Xrs" option to prevent
1362  // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1363  // (See bug 4345157, and other related bugs).
1364  // In reality, though, unblocking these signals is really a nop, since
1365  // these signals are not blocked by default.
1366  sigemptyset(&unblocked_sigs);
1367  sigemptyset(&allowdebug_blocked_sigs);
1368  sigaddset(&unblocked_sigs, SIGILL);
1369  sigaddset(&unblocked_sigs, SIGSEGV);
1370  sigaddset(&unblocked_sigs, SIGBUS);
1371  sigaddset(&unblocked_sigs, SIGFPE);
1372
1373  if (isJVM1available) {
1374    os::Solaris::set_SIGinterrupt(SIGJVM1);
1375    os::Solaris::set_SIGasync(SIGJVM2);
1376  } else if (UseAltSigs) {
1377    os::Solaris::set_SIGinterrupt(ALT_INTERRUPT_SIGNAL);
1378    os::Solaris::set_SIGasync(ALT_ASYNC_SIGNAL);
1379  } else {
1380    os::Solaris::set_SIGinterrupt(INTERRUPT_SIGNAL);
1381    os::Solaris::set_SIGasync(ASYNC_SIGNAL);
1382  }
1383
1384  sigaddset(&unblocked_sigs, os::Solaris::SIGinterrupt());
1385  sigaddset(&unblocked_sigs, os::Solaris::SIGasync());
1386
1387  if (!ReduceSignalUsage) {
1388   if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1389      sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1390      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
1391   }
1392   if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1393      sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1394      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
1395   }
1396   if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1397      sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1398      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
1399   }
1400  }
1401  // Fill in signals that are blocked by all but the VM thread.
1402  sigemptyset(&vm_sigs);
1403  if (!ReduceSignalUsage)
1404    sigaddset(&vm_sigs, BREAK_SIGNAL);
1405  debug_only(signal_sets_initialized = true);
1406
1407  // For diagnostics only used in run_periodic_checks
1408  sigemptyset(&check_signal_done);
1409}
1410
1411// These are signals that are unblocked while a thread is running Java.
1412// (For some reason, they get blocked by default.)
1413sigset_t* os::Solaris::unblocked_signals() {
1414  assert(signal_sets_initialized, "Not initialized");
1415  return &unblocked_sigs;
1416}
1417
1418// These are the signals that are blocked while a (non-VM) thread is
1419// running Java. Only the VM thread handles these signals.
1420sigset_t* os::Solaris::vm_signals() {
1421  assert(signal_sets_initialized, "Not initialized");
1422  return &vm_sigs;
1423}
1424
1425// These are signals that are blocked during cond_wait to allow debugger in
1426sigset_t* os::Solaris::allowdebug_blocked_signals() {
1427  assert(signal_sets_initialized, "Not initialized");
1428  return &allowdebug_blocked_sigs;
1429}
1430
1431// First crack at OS-specific initialization, from inside the new thread.
1432void os::initialize_thread() {
1433  int r = thr_main() ;
1434  guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
1435  if (r) {
1436    JavaThread* jt = (JavaThread *)Thread::current();
1437    assert(jt != NULL,"Sanity check");
1438    size_t stack_size;
1439    address base = jt->stack_base();
1440    if (Arguments::created_by_java_launcher()) {
1441      // Use 2MB to allow for Solaris 7 64 bit mode.
1442      stack_size = JavaThread::stack_size_at_create() == 0
1443        ? 2048*K : JavaThread::stack_size_at_create();
1444
1445      // There are rare cases when we may have already used more than
1446      // the basic stack size allotment before this method is invoked.
1447      // Attempt to allow for a normally sized java_stack.
1448      size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1449      stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1450    } else {
1451      // 6269555: If we were not created by a Java launcher, i.e. if we are
1452      // running embedded in a native application, treat the primordial thread
1453      // as much like a native attached thread as possible.  This means using
1454      // the current stack size from thr_stksegment(), unless it is too large
1455      // to reliably setup guard pages.  A reasonable max size is 8MB.
1456      size_t current_size = current_stack_size();
1457      // This should never happen, but just in case....
1458      if (current_size == 0) current_size = 2 * K * K;
1459      stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1460    }
1461    address bottom = (address)align_size_up((intptr_t)(base - stack_size), os::vm_page_size());;
1462    stack_size = (size_t)(base - bottom);
1463
1464    assert(stack_size > 0, "Stack size calculation problem");
1465
1466    if (stack_size > jt->stack_size()) {
1467      NOT_PRODUCT(
1468        struct rlimit limits;
1469        getrlimit(RLIMIT_STACK, &limits);
1470        size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1471        assert(size >= jt->stack_size(), "Stack size problem in main thread");
1472      )
1473      tty->print_cr(
1474        "Stack size of %d Kb exceeds current limit of %d Kb.\n"
1475        "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1476        "See limit(1) to increase the stack size limit.",
1477        stack_size / K, jt->stack_size() / K);
1478      vm_exit(1);
1479    }
1480    assert(jt->stack_size() >= stack_size,
1481          "Attempt to map more stack than was allocated");
1482    jt->set_stack_size(stack_size);
1483  }
1484
1485   // 5/22/01: Right now alternate signal stacks do not handle
1486   // throwing stack overflow exceptions, see bug 4463178
1487   // Until a fix is found for this, T2 will NOT imply alternate signal
1488   // stacks.
1489   // If using T2 libthread threads, install an alternate signal stack.
1490   // Because alternate stacks associate with LWPs on Solaris,
1491   // see sigaltstack(2), if using UNBOUND threads, or if UseBoundThreads
1492   // we prefer to explicitly stack bang.
1493   // If not using T2 libthread, but using UseBoundThreads any threads
1494   // (primordial thread, jni_attachCurrentThread) we do not create,
1495   // probably are not bound, therefore they can not have an alternate
1496   // signal stack. Since our stack banging code is generated and
1497   // is shared across threads, all threads must be bound to allow
1498   // using alternate signal stacks.  The alternative is to interpose
1499   // on _lwp_create to associate an alt sig stack with each LWP,
1500   // and this could be a problem when the JVM is embedded.
1501   // We would prefer to use alternate signal stacks with T2
1502   // Since there is currently no accurate way to detect T2
1503   // we do not. Assuming T2 when running T1 causes sig 11s or assertions
1504   // on installing alternate signal stacks
1505
1506
1507   // 05/09/03: removed alternate signal stack support for Solaris
1508   // The alternate signal stack mechanism is no longer needed to
1509   // handle stack overflow. This is now handled by allocating
1510   // guard pages (red zone) and stackbanging.
1511   // Initially the alternate signal stack mechanism was removed because
1512   // it did not work with T1 llibthread. Alternate
1513   // signal stacks MUST have all threads bound to lwps. Applications
1514   // can create their own threads and attach them without their being
1515   // bound under T1. This is frequently the case for the primordial thread.
1516   // If we were ever to reenable this mechanism we would need to
1517   // use the dynamic check for T2 libthread.
1518
1519  os::Solaris::init_thread_fpu_state();
1520}
1521
1522
1523
1524// Free Solaris resources related to the OSThread
1525void os::free_thread(OSThread* osthread) {
1526  assert(osthread != NULL, "os::free_thread but osthread not set");
1527
1528
1529  // We are told to free resources of the argument thread,
1530  // but we can only really operate on the current thread.
1531  // The main thread must take the VMThread down synchronously
1532  // before the main thread exits and frees up CodeHeap
1533  guarantee((Thread::current()->osthread() == osthread
1534     || (osthread == VMThread::vm_thread()->osthread())), "os::free_thread but not current thread");
1535  if (Thread::current()->osthread() == osthread) {
1536    // Restore caller's signal mask
1537    sigset_t sigmask = osthread->caller_sigmask();
1538    thr_sigsetmask(SIG_SETMASK, &sigmask, NULL);
1539  }
1540  delete osthread;
1541}
1542
1543void os::pd_start_thread(Thread* thread) {
1544  int status = thr_continue(thread->osthread()->thread_id());
1545  assert_status(status == 0, status, "thr_continue failed");
1546}
1547
1548
1549intx os::current_thread_id() {
1550  return (intx)thr_self();
1551}
1552
1553static pid_t _initial_pid = 0;
1554
1555int os::current_process_id() {
1556  return (int)(_initial_pid ? _initial_pid : getpid());
1557}
1558
1559int os::allocate_thread_local_storage() {
1560  // %%%       in Win32 this allocates a memory segment pointed to by a
1561  //           register.  Dan Stein can implement a similar feature in
1562  //           Solaris.  Alternatively, the VM can do the same thing
1563  //           explicitly: malloc some storage and keep the pointer in a
1564  //           register (which is part of the thread's context) (or keep it
1565  //           in TLS).
1566  // %%%       In current versions of Solaris, thr_self and TSD can
1567  //           be accessed via short sequences of displaced indirections.
1568  //           The value of thr_self is available as %g7(36).
1569  //           The value of thr_getspecific(k) is stored in %g7(12)(4)(k*4-4),
1570  //           assuming that the current thread already has a value bound to k.
1571  //           It may be worth experimenting with such access patterns,
1572  //           and later having the parameters formally exported from a Solaris
1573  //           interface.  I think, however, that it will be faster to
1574  //           maintain the invariant that %g2 always contains the
1575  //           JavaThread in Java code, and have stubs simply
1576  //           treat %g2 as a caller-save register, preserving it in a %lN.
1577  thread_key_t tk;
1578  if (thr_keycreate( &tk, NULL ) )
1579    fatal1("os::allocate_thread_local_storage: thr_keycreate failed (%s)", strerror(errno));
1580  return int(tk);
1581}
1582
1583void os::free_thread_local_storage(int index) {
1584  // %%% don't think we need anything here
1585  // if ( pthread_key_delete((pthread_key_t) tk) )
1586  //   fatal("os::free_thread_local_storage: pthread_key_delete failed");
1587}
1588
1589#define SMALLINT 32   // libthread allocate for tsd_common is a version specific
1590                      // small number - point is NO swap space available
1591void os::thread_local_storage_at_put(int index, void* value) {
1592  // %%% this is used only in threadLocalStorage.cpp
1593  if (thr_setspecific((thread_key_t)index, value)) {
1594    if (errno == ENOMEM) {
1595       vm_exit_out_of_memory(SMALLINT, "thr_setspecific: out of swap space");
1596    } else {
1597      fatal1("os::thread_local_storage_at_put: thr_setspecific failed (%s)", strerror(errno));
1598    }
1599  } else {
1600      ThreadLocalStorage::set_thread_in_slot ((Thread *) value) ;
1601  }
1602}
1603
1604// This function could be called before TLS is initialized, for example, when
1605// VM receives an async signal or when VM causes a fatal error during
1606// initialization. Return NULL if thr_getspecific() fails.
1607void* os::thread_local_storage_at(int index) {
1608  // %%% this is used only in threadLocalStorage.cpp
1609  void* r = NULL;
1610  return thr_getspecific((thread_key_t)index, &r) != 0 ? NULL : r;
1611}
1612
1613
1614const int NANOSECS_PER_MILLISECS = 1000000;
1615// gethrtime can move backwards if read from one cpu and then a different cpu
1616// getTimeNanos is guaranteed to not move backward on Solaris
1617// local spinloop created as faster for a CAS on an int than
1618// a CAS on a 64bit jlong. Also Atomic::cmpxchg for jlong is not
1619// supported on sparc v8 or pre supports_cx8 intel boxes.
1620// oldgetTimeNanos for systems which do not support CAS on 64bit jlong
1621// i.e. sparc v8 and pre supports_cx8 (i486) intel boxes
1622inline hrtime_t oldgetTimeNanos() {
1623  int gotlock = LOCK_INVALID;
1624  hrtime_t newtime = gethrtime();
1625
1626  for (;;) {
1627// grab lock for max_hrtime
1628    int curlock = max_hrtime_lock;
1629    if (curlock & LOCK_BUSY)  continue;
1630    if (gotlock = Atomic::cmpxchg(LOCK_BUSY, &max_hrtime_lock, LOCK_FREE) != LOCK_FREE) continue;
1631    if (newtime > max_hrtime) {
1632      max_hrtime = newtime;
1633    } else {
1634      newtime = max_hrtime;
1635    }
1636    // release lock
1637    max_hrtime_lock = LOCK_FREE;
1638    return newtime;
1639  }
1640}
1641// gethrtime can move backwards if read from one cpu and then a different cpu
1642// getTimeNanos is guaranteed to not move backward on Solaris
1643inline hrtime_t getTimeNanos() {
1644  if (VM_Version::supports_cx8()) {
1645    const hrtime_t now = gethrtime();
1646    const hrtime_t prev = max_hrtime;
1647    if (now <= prev)  return prev;   // same or retrograde time;
1648    const hrtime_t obsv = Atomic::cmpxchg(now, (volatile jlong*)&max_hrtime, prev);
1649    assert(obsv >= prev, "invariant");   // Monotonicity
1650    // If the CAS succeeded then we're done and return "now".
1651    // If the CAS failed and the observed value "obs" is >= now then
1652    // we should return "obs".  If the CAS failed and now > obs > prv then
1653    // some other thread raced this thread and installed a new value, in which case
1654    // we could either (a) retry the entire operation, (b) retry trying to install now
1655    // or (c) just return obs.  We use (c).   No loop is required although in some cases
1656    // we might discard a higher "now" value in deference to a slightly lower but freshly
1657    // installed obs value.   That's entirely benign -- it admits no new orderings compared
1658    // to (a) or (b) -- and greatly reduces coherence traffic.
1659    // We might also condition (c) on the magnitude of the delta between obs and now.
1660    // Avoiding excessive CAS operations to hot RW locations is critical.
1661    // See http://blogs.sun.com/dave/entry/cas_and_cache_trivia_invalidate
1662    return (prev == obsv) ? now : obsv ;
1663  } else {
1664    return oldgetTimeNanos();
1665  }
1666}
1667
1668// Time since start-up in seconds to a fine granularity.
1669// Used by VMSelfDestructTimer and the MemProfiler.
1670double os::elapsedTime() {
1671  return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1672}
1673
1674jlong os::elapsed_counter() {
1675  return (jlong)(getTimeNanos() - first_hrtime);
1676}
1677
1678jlong os::elapsed_frequency() {
1679   return hrtime_hz;
1680}
1681
1682// Return the real, user, and system times in seconds from an
1683// arbitrary fixed point in the past.
1684bool os::getTimesSecs(double* process_real_time,
1685                  double* process_user_time,
1686                  double* process_system_time) {
1687  struct tms ticks;
1688  clock_t real_ticks = times(&ticks);
1689
1690  if (real_ticks == (clock_t) (-1)) {
1691    return false;
1692  } else {
1693    double ticks_per_second = (double) clock_tics_per_sec;
1694    *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1695    *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1696    // For consistency return the real time from getTimeNanos()
1697    // converted to seconds.
1698    *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1699
1700    return true;
1701  }
1702}
1703
1704bool os::supports_vtime() { return true; }
1705
1706bool os::enable_vtime() {
1707  int fd = open("/proc/self/ctl", O_WRONLY);
1708  if (fd == -1)
1709    return false;
1710
1711  long cmd[] = { PCSET, PR_MSACCT };
1712  int res = write(fd, cmd, sizeof(long) * 2);
1713  close(fd);
1714  if (res != sizeof(long) * 2)
1715    return false;
1716
1717  return true;
1718}
1719
1720bool os::vtime_enabled() {
1721  int fd = open("/proc/self/status", O_RDONLY);
1722  if (fd == -1)
1723    return false;
1724
1725  pstatus_t status;
1726  int res = read(fd, (void*) &status, sizeof(pstatus_t));
1727  close(fd);
1728  if (res != sizeof(pstatus_t))
1729    return false;
1730
1731  return status.pr_flags & PR_MSACCT;
1732}
1733
1734double os::elapsedVTime() {
1735  return (double)gethrvtime() / (double)hrtime_hz;
1736}
1737
1738// Used internally for comparisons only
1739// getTimeMillis guaranteed to not move backwards on Solaris
1740jlong getTimeMillis() {
1741  jlong nanotime = getTimeNanos();
1742  return (jlong)(nanotime / NANOSECS_PER_MILLISECS);
1743}
1744
1745// Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1746jlong os::javaTimeMillis() {
1747  timeval t;
1748  if (gettimeofday( &t, NULL) == -1)
1749    fatal1("os::javaTimeMillis: gettimeofday (%s)", strerror(errno));
1750  return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1751}
1752
1753jlong os::javaTimeNanos() {
1754  return (jlong)getTimeNanos();
1755}
1756
1757void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1758  info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1759  info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1760  info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1761  info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1762}
1763
1764char * os::local_time_string(char *buf, size_t buflen) {
1765  struct tm t;
1766  time_t long_time;
1767  time(&long_time);
1768  localtime_r(&long_time, &t);
1769  jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1770               t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1771               t.tm_hour, t.tm_min, t.tm_sec);
1772  return buf;
1773}
1774
1775// Note: os::shutdown() might be called very early during initialization, or
1776// called from signal handler. Before adding something to os::shutdown(), make
1777// sure it is async-safe and can handle partially initialized VM.
1778void os::shutdown() {
1779
1780  // allow PerfMemory to attempt cleanup of any persistent resources
1781  perfMemory_exit();
1782
1783  // needs to remove object in file system
1784  AttachListener::abort();
1785
1786  // flush buffered output, finish log files
1787  ostream_abort();
1788
1789  // Check for abort hook
1790  abort_hook_t abort_hook = Arguments::abort_hook();
1791  if (abort_hook != NULL) {
1792    abort_hook();
1793  }
1794}
1795
1796// Note: os::abort() might be called very early during initialization, or
1797// called from signal handler. Before adding something to os::abort(), make
1798// sure it is async-safe and can handle partially initialized VM.
1799void os::abort(bool dump_core) {
1800  os::shutdown();
1801  if (dump_core) {
1802#ifndef PRODUCT
1803    fdStream out(defaultStream::output_fd());
1804    out.print_raw("Current thread is ");
1805    char buf[16];
1806    jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1807    out.print_raw_cr(buf);
1808    out.print_raw_cr("Dumping core ...");
1809#endif
1810    ::abort(); // dump core (for debugging)
1811  }
1812
1813  ::exit(1);
1814}
1815
1816// Die immediately, no exit hook, no abort hook, no cleanup.
1817void os::die() {
1818  _exit(-1);
1819}
1820
1821// unused
1822void os::set_error_file(const char *logfile) {}
1823
1824// DLL functions
1825
1826const char* os::dll_file_extension() { return ".so"; }
1827
1828const char* os::get_temp_directory() { return "/tmp/"; }
1829
1830static bool file_exists(const char* filename) {
1831  struct stat statbuf;
1832  if (filename == NULL || strlen(filename) == 0) {
1833    return false;
1834  }
1835  return os::stat(filename, &statbuf) == 0;
1836}
1837
1838void os::dll_build_name(char* buffer, size_t buflen,
1839                        const char* pname, const char* fname) {
1840  // Copied from libhpi
1841  const size_t pnamelen = pname ? strlen(pname) : 0;
1842
1843  // Quietly truncate on buffer overflow.  Should be an error.
1844  if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1845      *buffer = '\0';
1846      return;
1847  }
1848
1849  if (pnamelen == 0) {
1850    snprintf(buffer, buflen, "lib%s.so", fname);
1851  } else if (strchr(pname, *os::path_separator()) != NULL) {
1852    int n;
1853    char** pelements = split_path(pname, &n);
1854    for (int i = 0 ; i < n ; i++) {
1855      // really shouldn't be NULL but what the heck, check can't hurt
1856      if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1857        continue; // skip the empty path values
1858      }
1859      snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1860      if (file_exists(buffer)) {
1861        break;
1862      }
1863    }
1864    // release the storage
1865    for (int i = 0 ; i < n ; i++) {
1866      if (pelements[i] != NULL) {
1867        FREE_C_HEAP_ARRAY(char, pelements[i]);
1868      }
1869    }
1870    if (pelements != NULL) {
1871      FREE_C_HEAP_ARRAY(char*, pelements);
1872    }
1873  } else {
1874    snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1875  }
1876}
1877
1878const char* os::get_current_directory(char *buf, int buflen) {
1879  return getcwd(buf, buflen);
1880}
1881
1882// check if addr is inside libjvm[_g].so
1883bool os::address_is_in_vm(address addr) {
1884  static address libjvm_base_addr;
1885  Dl_info dlinfo;
1886
1887  if (libjvm_base_addr == NULL) {
1888    dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1889    libjvm_base_addr = (address)dlinfo.dli_fbase;
1890    assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1891  }
1892
1893  if (dladdr((void *)addr, &dlinfo)) {
1894    if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1895  }
1896
1897  return false;
1898}
1899
1900typedef int (*dladdr1_func_type) (void *, Dl_info *, void **, int);
1901static dladdr1_func_type dladdr1_func = NULL;
1902
1903bool os::dll_address_to_function_name(address addr, char *buf,
1904                                      int buflen, int * offset) {
1905  Dl_info dlinfo;
1906
1907  // dladdr1_func was initialized in os::init()
1908  if (dladdr1_func){
1909      // yes, we have dladdr1
1910
1911      // Support for dladdr1 is checked at runtime; it may be
1912      // available even if the vm is built on a machine that does
1913      // not have dladdr1 support.  Make sure there is a value for
1914      // RTLD_DL_SYMENT.
1915      #ifndef RTLD_DL_SYMENT
1916      #define RTLD_DL_SYMENT 1
1917      #endif
1918      Sym * info;
1919      if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1920                       RTLD_DL_SYMENT)) {
1921          if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1922          if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1923
1924          // check if the returned symbol really covers addr
1925          return ((char *)dlinfo.dli_saddr + info->st_size > (char *)addr);
1926      } else {
1927          if (buf) buf[0] = '\0';
1928          if (offset) *offset  = -1;
1929          return false;
1930      }
1931  } else {
1932      // no, only dladdr is available
1933      if(dladdr((void *)addr, &dlinfo)) {
1934          if (buf) jio_snprintf(buf, buflen, dlinfo.dli_sname);
1935          if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1936          return true;
1937      } else {
1938          if (buf) buf[0] = '\0';
1939          if (offset) *offset  = -1;
1940          return false;
1941      }
1942  }
1943}
1944
1945bool os::dll_address_to_library_name(address addr, char* buf,
1946                                     int buflen, int* offset) {
1947  Dl_info dlinfo;
1948
1949  if (dladdr((void*)addr, &dlinfo)){
1950     if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1951     if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1952     return true;
1953  } else {
1954     if (buf) buf[0] = '\0';
1955     if (offset) *offset = -1;
1956     return false;
1957  }
1958}
1959
1960// Prints the names and full paths of all opened dynamic libraries
1961// for current process
1962void os::print_dll_info(outputStream * st) {
1963    Dl_info dli;
1964    void *handle;
1965    Link_map *map;
1966    Link_map *p;
1967
1968    st->print_cr("Dynamic libraries:"); st->flush();
1969
1970    if (!dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli)) {
1971        st->print_cr("Error: Cannot print dynamic libraries.");
1972        return;
1973    }
1974    handle = dlopen(dli.dli_fname, RTLD_LAZY);
1975    if (handle == NULL) {
1976        st->print_cr("Error: Cannot print dynamic libraries.");
1977        return;
1978    }
1979    dlinfo(handle, RTLD_DI_LINKMAP, &map);
1980    if (map == NULL) {
1981        st->print_cr("Error: Cannot print dynamic libraries.");
1982        return;
1983    }
1984
1985    while (map->l_prev != NULL)
1986        map = map->l_prev;
1987
1988    while (map != NULL) {
1989        st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
1990        map = map->l_next;
1991    }
1992
1993    dlclose(handle);
1994}
1995
1996  // Loads .dll/.so and
1997  // in case of error it checks if .dll/.so was built for the
1998  // same architecture as Hotspot is running on
1999
2000void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
2001{
2002  void * result= ::dlopen(filename, RTLD_LAZY);
2003  if (result != NULL) {
2004    // Successful loading
2005    return result;
2006  }
2007
2008  Elf32_Ehdr elf_head;
2009
2010  // Read system error message into ebuf
2011  // It may or may not be overwritten below
2012  ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2013  ebuf[ebuflen-1]='\0';
2014  int diag_msg_max_length=ebuflen-strlen(ebuf);
2015  char* diag_msg_buf=ebuf+strlen(ebuf);
2016
2017  if (diag_msg_max_length==0) {
2018    // No more space in ebuf for additional diagnostics message
2019    return NULL;
2020  }
2021
2022
2023  int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2024
2025  if (file_descriptor < 0) {
2026    // Can't open library, report dlerror() message
2027    return NULL;
2028  }
2029
2030  bool failed_to_read_elf_head=
2031    (sizeof(elf_head)!=
2032        (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2033
2034  ::close(file_descriptor);
2035  if (failed_to_read_elf_head) {
2036    // file i/o error - report dlerror() msg
2037    return NULL;
2038  }
2039
2040  typedef struct {
2041    Elf32_Half  code;         // Actual value as defined in elf.h
2042    Elf32_Half  compat_class; // Compatibility of archs at VM's sense
2043    char        elf_class;    // 32 or 64 bit
2044    char        endianess;    // MSB or LSB
2045    char*       name;         // String representation
2046  } arch_t;
2047
2048  static const arch_t arch_array[]={
2049    {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2050    {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2051    {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2052    {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2053    {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2054    {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2055    {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2056    {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2057    {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}
2058  };
2059
2060  #if  (defined IA32)
2061    static  Elf32_Half running_arch_code=EM_386;
2062  #elif   (defined AMD64)
2063    static  Elf32_Half running_arch_code=EM_X86_64;
2064  #elif  (defined IA64)
2065    static  Elf32_Half running_arch_code=EM_IA_64;
2066  #elif  (defined __sparc) && (defined _LP64)
2067    static  Elf32_Half running_arch_code=EM_SPARCV9;
2068  #elif  (defined __sparc) && (!defined _LP64)
2069    static  Elf32_Half running_arch_code=EM_SPARC;
2070  #elif  (defined __powerpc64__)
2071    static  Elf32_Half running_arch_code=EM_PPC64;
2072  #elif  (defined __powerpc__)
2073    static  Elf32_Half running_arch_code=EM_PPC;
2074  #else
2075    #error Method os::dll_load requires that one of following is defined:\
2076         IA32, AMD64, IA64, __sparc, __powerpc__
2077  #endif
2078
2079  // Identify compatability class for VM's architecture and library's architecture
2080  // Obtain string descriptions for architectures
2081
2082  arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2083  int running_arch_index=-1;
2084
2085  for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2086    if (running_arch_code == arch_array[i].code) {
2087      running_arch_index    = i;
2088    }
2089    if (lib_arch.code == arch_array[i].code) {
2090      lib_arch.compat_class = arch_array[i].compat_class;
2091      lib_arch.name         = arch_array[i].name;
2092    }
2093  }
2094
2095  assert(running_arch_index != -1,
2096    "Didn't find running architecture code (running_arch_code) in arch_array");
2097  if (running_arch_index == -1) {
2098    // Even though running architecture detection failed
2099    // we may still continue with reporting dlerror() message
2100    return NULL;
2101  }
2102
2103  if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2104    ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2105    return NULL;
2106  }
2107
2108  if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2109    ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2110    return NULL;
2111  }
2112
2113  if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2114    if ( lib_arch.name!=NULL ) {
2115      ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2116        " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2117        lib_arch.name, arch_array[running_arch_index].name);
2118    } else {
2119      ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2120      " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2121        lib_arch.code,
2122        arch_array[running_arch_index].name);
2123    }
2124  }
2125
2126  return NULL;
2127}
2128
2129void* os::dll_lookup(void* handle, const char* name) {
2130  return dlsym(handle, name);
2131}
2132
2133
2134bool _print_ascii_file(const char* filename, outputStream* st) {
2135  int fd = open(filename, O_RDONLY);
2136  if (fd == -1) {
2137     return false;
2138  }
2139
2140  char buf[32];
2141  int bytes;
2142  while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
2143    st->print_raw(buf, bytes);
2144  }
2145
2146  close(fd);
2147
2148  return true;
2149}
2150
2151void os::print_os_info(outputStream* st) {
2152  st->print("OS:");
2153
2154  if (!_print_ascii_file("/etc/release", st)) {
2155    st->print("Solaris");
2156  }
2157  st->cr();
2158
2159  // kernel
2160  st->print("uname:");
2161  struct utsname name;
2162  uname(&name);
2163  st->print(name.sysname); st->print(" ");
2164  st->print(name.release); st->print(" ");
2165  st->print(name.version); st->print(" ");
2166  st->print(name.machine);
2167
2168  // libthread
2169  if (os::Solaris::T2_libthread()) st->print("  (T2 libthread)");
2170  else st->print("  (T1 libthread)");
2171  st->cr();
2172
2173  // rlimit
2174  st->print("rlimit:");
2175  struct rlimit rlim;
2176
2177  st->print(" STACK ");
2178  getrlimit(RLIMIT_STACK, &rlim);
2179  if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2180  else st->print("%uk", rlim.rlim_cur >> 10);
2181
2182  st->print(", CORE ");
2183  getrlimit(RLIMIT_CORE, &rlim);
2184  if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2185  else st->print("%uk", rlim.rlim_cur >> 10);
2186
2187  st->print(", NOFILE ");
2188  getrlimit(RLIMIT_NOFILE, &rlim);
2189  if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2190  else st->print("%d", rlim.rlim_cur);
2191
2192  st->print(", AS ");
2193  getrlimit(RLIMIT_AS, &rlim);
2194  if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2195  else st->print("%uk", rlim.rlim_cur >> 10);
2196  st->cr();
2197
2198  // load average
2199  st->print("load average:");
2200  double loadavg[3];
2201  os::loadavg(loadavg, 3);
2202  st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2203  st->cr();
2204}
2205
2206
2207static bool check_addr0(outputStream* st) {
2208  jboolean status = false;
2209  int fd = open("/proc/self/map",O_RDONLY);
2210  if (fd >= 0) {
2211    prmap_t p;
2212    while(read(fd, &p, sizeof(p)) > 0) {
2213      if (p.pr_vaddr == 0x0) {
2214        st->print("Warning: Address: 0x%x, Size: %dK, ",p.pr_vaddr, p.pr_size/1024, p.pr_mapname);
2215        st->print("Mapped file: %s, ", p.pr_mapname[0] == '\0' ? "None" : p.pr_mapname);
2216        st->print("Access:");
2217        st->print("%s",(p.pr_mflags & MA_READ)  ? "r" : "-");
2218        st->print("%s",(p.pr_mflags & MA_WRITE) ? "w" : "-");
2219        st->print("%s",(p.pr_mflags & MA_EXEC)  ? "x" : "-");
2220        st->cr();
2221        status = true;
2222      }
2223      close(fd);
2224    }
2225  }
2226  return status;
2227}
2228
2229void os::print_memory_info(outputStream* st) {
2230  st->print("Memory:");
2231  st->print(" %dk page", os::vm_page_size()>>10);
2232  st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
2233  st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
2234  st->cr();
2235  (void) check_addr0(st);
2236}
2237
2238// Taken from /usr/include/sys/machsig.h  Supposed to be architecture specific
2239// but they're the same for all the solaris architectures that we support.
2240const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2241                          "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2242                          "ILL_COPROC", "ILL_BADSTK" };
2243
2244const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2245                          "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2246                          "FPE_FLTINV", "FPE_FLTSUB" };
2247
2248const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2249
2250const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2251
2252void os::print_siginfo(outputStream* st, void* siginfo) {
2253  st->print("siginfo:");
2254
2255  const int buflen = 100;
2256  char buf[buflen];
2257  siginfo_t *si = (siginfo_t*)siginfo;
2258  st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2259  char *err = strerror(si->si_errno);
2260  if (si->si_errno != 0 && err != NULL) {
2261    st->print("si_errno=%s", err);
2262  } else {
2263    st->print("si_errno=%d", si->si_errno);
2264  }
2265  const int c = si->si_code;
2266  assert(c > 0, "unexpected si_code");
2267  switch (si->si_signo) {
2268  case SIGILL:
2269    st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2270    st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2271    break;
2272  case SIGFPE:
2273    st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2274    st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2275    break;
2276  case SIGSEGV:
2277    st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2278    st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2279    break;
2280  case SIGBUS:
2281    st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2282    st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2283    break;
2284  default:
2285    st->print(", si_code=%d", si->si_code);
2286    // no si_addr
2287  }
2288
2289  if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2290      UseSharedSpaces) {
2291    FileMapInfo* mapinfo = FileMapInfo::current_info();
2292    if (mapinfo->is_in_shared_space(si->si_addr)) {
2293      st->print("\n\nError accessing class data sharing archive."   \
2294                " Mapped file inaccessible during execution, "      \
2295                " possible disk/network problem.");
2296    }
2297  }
2298  st->cr();
2299}
2300
2301// Moved from whole group, because we need them here for diagnostic
2302// prints.
2303#define OLDMAXSIGNUM 32
2304static int Maxsignum = 0;
2305static int *ourSigFlags = NULL;
2306
2307extern "C" void sigINTRHandler(int, siginfo_t*, void*);
2308
2309int os::Solaris::get_our_sigflags(int sig) {
2310  assert(ourSigFlags!=NULL, "signal data structure not initialized");
2311  assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2312  return ourSigFlags[sig];
2313}
2314
2315void os::Solaris::set_our_sigflags(int sig, int flags) {
2316  assert(ourSigFlags!=NULL, "signal data structure not initialized");
2317  assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2318  ourSigFlags[sig] = flags;
2319}
2320
2321
2322static const char* get_signal_handler_name(address handler,
2323                                           char* buf, int buflen) {
2324  int offset;
2325  bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
2326  if (found) {
2327    // skip directory names
2328    const char *p1, *p2;
2329    p1 = buf;
2330    size_t len = strlen(os::file_separator());
2331    while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
2332    jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
2333  } else {
2334    jio_snprintf(buf, buflen, PTR_FORMAT, handler);
2335  }
2336  return buf;
2337}
2338
2339static void print_signal_handler(outputStream* st, int sig,
2340                                  char* buf, size_t buflen) {
2341  struct sigaction sa;
2342
2343  sigaction(sig, NULL, &sa);
2344
2345  st->print("%s: ", os::exception_name(sig, buf, buflen));
2346
2347  address handler = (sa.sa_flags & SA_SIGINFO)
2348                  ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
2349                  : CAST_FROM_FN_PTR(address, sa.sa_handler);
2350
2351  if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
2352    st->print("SIG_DFL");
2353  } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
2354    st->print("SIG_IGN");
2355  } else {
2356    st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
2357  }
2358
2359  st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
2360
2361  address rh = VMError::get_resetted_sighandler(sig);
2362  // May be, handler was resetted by VMError?
2363  if(rh != NULL) {
2364    handler = rh;
2365    sa.sa_flags = VMError::get_resetted_sigflags(sig);
2366  }
2367
2368  st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
2369
2370  // Check: is it our handler?
2371  if(handler == CAST_FROM_FN_PTR(address, signalHandler) ||
2372     handler == CAST_FROM_FN_PTR(address, sigINTRHandler)) {
2373    // It is our signal handler
2374    // check for flags
2375    if(sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2376      st->print(
2377        ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2378        os::Solaris::get_our_sigflags(sig));
2379    }
2380  }
2381  st->cr();
2382}
2383
2384void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2385  st->print_cr("Signal Handlers:");
2386  print_signal_handler(st, SIGSEGV, buf, buflen);
2387  print_signal_handler(st, SIGBUS , buf, buflen);
2388  print_signal_handler(st, SIGFPE , buf, buflen);
2389  print_signal_handler(st, SIGPIPE, buf, buflen);
2390  print_signal_handler(st, SIGXFSZ, buf, buflen);
2391  print_signal_handler(st, SIGILL , buf, buflen);
2392  print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2393  print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2394  print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2395  print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2396  print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2397  print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2398  print_signal_handler(st, os::Solaris::SIGinterrupt(), buf, buflen);
2399  print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2400}
2401
2402static char saved_jvm_path[MAXPATHLEN] = { 0 };
2403
2404// Find the full path to the current module, libjvm.so or libjvm_g.so
2405void os::jvm_path(char *buf, jint buflen) {
2406  // Error checking.
2407  if (buflen < MAXPATHLEN) {
2408    assert(false, "must use a large-enough buffer");
2409    buf[0] = '\0';
2410    return;
2411  }
2412  // Lazy resolve the path to current module.
2413  if (saved_jvm_path[0] != 0) {
2414    strcpy(buf, saved_jvm_path);
2415    return;
2416  }
2417
2418  Dl_info dlinfo;
2419  int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2420  assert(ret != 0, "cannot locate libjvm");
2421  realpath((char *)dlinfo.dli_fname, buf);
2422
2423  if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2424    // Support for the gamma launcher.  Typical value for buf is
2425    // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2426    // the right place in the string, then assume we are installed in a JDK and
2427    // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2428    // up the path so it looks like libjvm.so is installed there (append a
2429    // fake suffix hotspot/libjvm.so).
2430    const char *p = buf + strlen(buf) - 1;
2431    for (int count = 0; p > buf && count < 5; ++count) {
2432      for (--p; p > buf && *p != '/'; --p)
2433        /* empty */ ;
2434    }
2435
2436    if (strncmp(p, "/jre/lib/", 9) != 0) {
2437      // Look for JAVA_HOME in the environment.
2438      char* java_home_var = ::getenv("JAVA_HOME");
2439      if (java_home_var != NULL && java_home_var[0] != 0) {
2440        char cpu_arch[12];
2441        sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2442#ifdef _LP64
2443        // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2444        if (strcmp(cpu_arch, "sparc") == 0) {
2445          strcat(cpu_arch, "v9");
2446        } else if (strcmp(cpu_arch, "i386") == 0) {
2447          strcpy(cpu_arch, "amd64");
2448        }
2449#endif
2450        // Check the current module name "libjvm.so" or "libjvm_g.so".
2451        p = strrchr(buf, '/');
2452        assert(strstr(p, "/libjvm") == p, "invalid library name");
2453        p = strstr(p, "_g") ? "_g" : "";
2454
2455        realpath(java_home_var, buf);
2456        sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
2457        if (0 == access(buf, F_OK)) {
2458          // Use current module name "libjvm[_g].so" instead of
2459          // "libjvm"debug_only("_g")".so" since for fastdebug version
2460          // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2461          // It is used when we are choosing the HPI library's name
2462          // "libhpi[_g].so" in hpi::initialize_get_interface().
2463          sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
2464        } else {
2465          // Go back to path of .so
2466          realpath((char *)dlinfo.dli_fname, buf);
2467        }
2468      }
2469    }
2470  }
2471
2472  strcpy(saved_jvm_path, buf);
2473}
2474
2475
2476void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2477  // no prefix required, not even "_"
2478}
2479
2480
2481void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2482  // no suffix required
2483}
2484
2485
2486// sun.misc.Signal
2487
2488extern "C" {
2489  static void UserHandler(int sig, void *siginfo, void *context) {
2490    // Ctrl-C is pressed during error reporting, likely because the error
2491    // handler fails to abort. Let VM die immediately.
2492    if (sig == SIGINT && is_error_reported()) {
2493       os::die();
2494    }
2495
2496    os::signal_notify(sig);
2497    // We do not need to reinstate the signal handler each time...
2498  }
2499}
2500
2501void* os::user_handler() {
2502  return CAST_FROM_FN_PTR(void*, UserHandler);
2503}
2504
2505extern "C" {
2506  typedef void (*sa_handler_t)(int);
2507  typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2508}
2509
2510void* os::signal(int signal_number, void* handler) {
2511  struct sigaction sigAct, oldSigAct;
2512  sigfillset(&(sigAct.sa_mask));
2513  sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2514  sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2515
2516  if (sigaction(signal_number, &sigAct, &oldSigAct))
2517    // -1 means registration failed
2518    return (void *)-1;
2519
2520  return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2521}
2522
2523void os::signal_raise(int signal_number) {
2524  raise(signal_number);
2525}
2526
2527/*
2528 * The following code is moved from os.cpp for making this
2529 * code platform specific, which it is by its very nature.
2530 */
2531
2532// a counter for each possible signal value
2533static int Sigexit = 0;
2534static int Maxlibjsigsigs;
2535static jint *pending_signals = NULL;
2536static int *preinstalled_sigs = NULL;
2537static struct sigaction *chainedsigactions = NULL;
2538static sema_t sig_sem;
2539typedef int (*version_getting_t)();
2540version_getting_t os::Solaris::get_libjsig_version = NULL;
2541static int libjsigversion = NULL;
2542
2543int os::sigexitnum_pd() {
2544  assert(Sigexit > 0, "signal memory not yet initialized");
2545  return Sigexit;
2546}
2547
2548void os::Solaris::init_signal_mem() {
2549  // Initialize signal structures
2550  Maxsignum = SIGRTMAX;
2551  Sigexit = Maxsignum+1;
2552  assert(Maxsignum >0, "Unable to obtain max signal number");
2553
2554  Maxlibjsigsigs = Maxsignum;
2555
2556  // pending_signals has one int per signal
2557  // The additional signal is for SIGEXIT - exit signal to signal_thread
2558  pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1));
2559  memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2560
2561  if (UseSignalChaining) {
2562     chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2563       * (Maxsignum + 1));
2564     memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2565     preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1));
2566     memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2567  }
2568  ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1 ));
2569  memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2570}
2571
2572void os::signal_init_pd() {
2573  int ret;
2574
2575  ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2576  assert(ret == 0, "sema_init() failed");
2577}
2578
2579void os::signal_notify(int signal_number) {
2580  int ret;
2581
2582  Atomic::inc(&pending_signals[signal_number]);
2583  ret = ::sema_post(&sig_sem);
2584  assert(ret == 0, "sema_post() failed");
2585}
2586
2587static int check_pending_signals(bool wait_for_signal) {
2588  int ret;
2589  while (true) {
2590    for (int i = 0; i < Sigexit + 1; i++) {
2591      jint n = pending_signals[i];
2592      if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2593        return i;
2594      }
2595    }
2596    if (!wait_for_signal) {
2597      return -1;
2598    }
2599    JavaThread *thread = JavaThread::current();
2600    ThreadBlockInVM tbivm(thread);
2601
2602    bool threadIsSuspended;
2603    do {
2604      thread->set_suspend_equivalent();
2605      // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2606      while((ret = ::sema_wait(&sig_sem)) == EINTR)
2607          ;
2608      assert(ret == 0, "sema_wait() failed");
2609
2610      // were we externally suspended while we were waiting?
2611      threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2612      if (threadIsSuspended) {
2613        //
2614        // The semaphore has been incremented, but while we were waiting
2615        // another thread suspended us. We don't want to continue running
2616        // while suspended because that would surprise the thread that
2617        // suspended us.
2618        //
2619        ret = ::sema_post(&sig_sem);
2620        assert(ret == 0, "sema_post() failed");
2621
2622        thread->java_suspend_self();
2623      }
2624    } while (threadIsSuspended);
2625  }
2626}
2627
2628int os::signal_lookup() {
2629  return check_pending_signals(false);
2630}
2631
2632int os::signal_wait() {
2633  return check_pending_signals(true);
2634}
2635
2636////////////////////////////////////////////////////////////////////////////////
2637// Virtual Memory
2638
2639static int page_size = -1;
2640
2641// The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2642// clear this var if support is not available.
2643static bool has_map_align = true;
2644
2645int os::vm_page_size() {
2646  assert(page_size != -1, "must call os::init");
2647  return page_size;
2648}
2649
2650// Solaris allocates memory by pages.
2651int os::vm_allocation_granularity() {
2652  assert(page_size != -1, "must call os::init");
2653  return page_size;
2654}
2655
2656bool os::commit_memory(char* addr, size_t bytes, bool exec) {
2657  int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2658  size_t size = bytes;
2659  return
2660     NULL != Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2661}
2662
2663bool os::commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2664                       bool exec) {
2665  if (commit_memory(addr, bytes, exec)) {
2666    if (UseMPSS && alignment_hint > (size_t)vm_page_size()) {
2667      // If the large page size has been set and the VM
2668      // is using large pages, use the large page size
2669      // if it is smaller than the alignment hint. This is
2670      // a case where the VM wants to use a larger alignment size
2671      // for its own reasons but still want to use large pages
2672      // (which is what matters to setting the mpss range.
2673      size_t page_size = 0;
2674      if (large_page_size() < alignment_hint) {
2675        assert(UseLargePages, "Expected to be here for large page use only");
2676        page_size = large_page_size();
2677      } else {
2678        // If the alignment hint is less than the large page
2679        // size, the VM wants a particular alignment (thus the hint)
2680        // for internal reasons.  Try to set the mpss range using
2681        // the alignment_hint.
2682        page_size = alignment_hint;
2683      }
2684      // Since this is a hint, ignore any failures.
2685      (void)Solaris::set_mpss_range(addr, bytes, page_size);
2686    }
2687    return true;
2688  }
2689  return false;
2690}
2691
2692// Uncommit the pages in a specified region.
2693void os::free_memory(char* addr, size_t bytes) {
2694  if (madvise(addr, bytes, MADV_FREE) < 0) {
2695    debug_only(warning("MADV_FREE failed."));
2696    return;
2697  }
2698}
2699
2700// Change the page size in a given range.
2701void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2702  assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2703  assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2704  Solaris::set_mpss_range(addr, bytes, alignment_hint);
2705}
2706
2707// Tell the OS to make the range local to the first-touching LWP
2708void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2709  assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2710  if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2711    debug_only(warning("MADV_ACCESS_LWP failed."));
2712  }
2713}
2714
2715// Tell the OS that this range would be accessed from different LWPs.
2716void os::numa_make_global(char *addr, size_t bytes) {
2717  assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2718  if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2719    debug_only(warning("MADV_ACCESS_MANY failed."));
2720  }
2721}
2722
2723// Get the number of the locality groups.
2724size_t os::numa_get_groups_num() {
2725  size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2726  return n != -1 ? n : 1;
2727}
2728
2729// Get a list of leaf locality groups. A leaf lgroup is group that
2730// doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2731// board. An LWP is assigned to one of these groups upon creation.
2732size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2733   if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2734     ids[0] = 0;
2735     return 1;
2736   }
2737   int result_size = 0, top = 1, bottom = 0, cur = 0;
2738   for (int k = 0; k < size; k++) {
2739     int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2740                                    (Solaris::lgrp_id_t*)&ids[top], size - top);
2741     if (r == -1) {
2742       ids[0] = 0;
2743       return 1;
2744     }
2745     if (!r) {
2746       // That's a leaf node.
2747       assert (bottom <= cur, "Sanity check");
2748       // Check if the node has memory
2749       if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2750                                   NULL, 0, LGRP_RSRC_MEM) > 0) {
2751         ids[bottom++] = ids[cur];
2752       }
2753     }
2754     top += r;
2755     cur++;
2756   }
2757   if (bottom == 0) {
2758     // Handle a situation, when the OS reports no memory available.
2759     // Assume UMA architecture.
2760     ids[0] = 0;
2761     return 1;
2762   }
2763   return bottom;
2764}
2765
2766// Detect the topology change. Typically happens during CPU plugging-unplugging.
2767bool os::numa_topology_changed() {
2768  int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2769  if (is_stale != -1 && is_stale) {
2770    Solaris::lgrp_fini(Solaris::lgrp_cookie());
2771    Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2772    assert(c != 0, "Failure to initialize LGRP API");
2773    Solaris::set_lgrp_cookie(c);
2774    return true;
2775  }
2776  return false;
2777}
2778
2779// Get the group id of the current LWP.
2780int os::numa_get_group_id() {
2781  int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2782  if (lgrp_id == -1) {
2783    return 0;
2784  }
2785  const int size = os::numa_get_groups_num();
2786  int *ids = (int*)alloca(size * sizeof(int));
2787
2788  // Get the ids of all lgroups with memory; r is the count.
2789  int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2790                                  (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2791  if (r <= 0) {
2792    return 0;
2793  }
2794  return ids[os::random() % r];
2795}
2796
2797// Request information about the page.
2798bool os::get_page_info(char *start, page_info* info) {
2799  const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2800  uint64_t addr = (uintptr_t)start;
2801  uint64_t outdata[2];
2802  uint_t validity = 0;
2803
2804  if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2805    return false;
2806  }
2807
2808  info->size = 0;
2809  info->lgrp_id = -1;
2810
2811  if ((validity & 1) != 0) {
2812    if ((validity & 2) != 0) {
2813      info->lgrp_id = outdata[0];
2814    }
2815    if ((validity & 4) != 0) {
2816      info->size = outdata[1];
2817    }
2818    return true;
2819  }
2820  return false;
2821}
2822
2823// Scan the pages from start to end until a page different than
2824// the one described in the info parameter is encountered.
2825char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2826  const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2827  const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2828  uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT];
2829  uint_t validity[MAX_MEMINFO_CNT];
2830
2831  size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2832  uint64_t p = (uint64_t)start;
2833  while (p < (uint64_t)end) {
2834    addrs[0] = p;
2835    size_t addrs_count = 1;
2836    while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] < (uint64_t)end) {
2837      addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2838      addrs_count++;
2839    }
2840
2841    if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2842      return NULL;
2843    }
2844
2845    size_t i = 0;
2846    for (; i < addrs_count; i++) {
2847      if ((validity[i] & 1) != 0) {
2848        if ((validity[i] & 4) != 0) {
2849          if (outdata[types * i + 1] != page_expected->size) {
2850            break;
2851          }
2852        } else
2853          if (page_expected->size != 0) {
2854            break;
2855          }
2856
2857        if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2858          if (outdata[types * i] != page_expected->lgrp_id) {
2859            break;
2860          }
2861        }
2862      } else {
2863        return NULL;
2864      }
2865    }
2866
2867    if (i != addrs_count) {
2868      if ((validity[i] & 2) != 0) {
2869        page_found->lgrp_id = outdata[types * i];
2870      } else {
2871        page_found->lgrp_id = -1;
2872      }
2873      if ((validity[i] & 4) != 0) {
2874        page_found->size = outdata[types * i + 1];
2875      } else {
2876        page_found->size = 0;
2877      }
2878      return (char*)addrs[i];
2879    }
2880
2881    p = addrs[addrs_count - 1] + page_size;
2882  }
2883  return end;
2884}
2885
2886bool os::uncommit_memory(char* addr, size_t bytes) {
2887  size_t size = bytes;
2888  // Map uncommitted pages PROT_NONE so we fail early if we touch an
2889  // uncommitted page. Otherwise, the read/write might succeed if we
2890  // have enough swap space to back the physical page.
2891  return
2892    NULL != Solaris::mmap_chunk(addr, size,
2893                                MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2894                                PROT_NONE);
2895}
2896
2897char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2898  char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2899
2900  if (b == MAP_FAILED) {
2901    return NULL;
2902  }
2903  return b;
2904}
2905
2906char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes, size_t alignment_hint, bool fixed) {
2907  char* addr = requested_addr;
2908  int flags = MAP_PRIVATE | MAP_NORESERVE;
2909
2910  assert(!(fixed && (alignment_hint > 0)), "alignment hint meaningless with fixed mmap");
2911
2912  if (fixed) {
2913    flags |= MAP_FIXED;
2914  } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
2915    flags |= MAP_ALIGN;
2916    addr = (char*) alignment_hint;
2917  }
2918
2919  // Map uncommitted pages PROT_NONE so we fail early if we touch an
2920  // uncommitted page. Otherwise, the read/write might succeed if we
2921  // have enough swap space to back the physical page.
2922  return mmap_chunk(addr, bytes, flags, PROT_NONE);
2923}
2924
2925char* os::reserve_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
2926  char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint, (requested_addr != NULL));
2927
2928  guarantee(requested_addr == NULL || requested_addr == addr,
2929            "OS failed to return requested mmap address.");
2930  return addr;
2931}
2932
2933// Reserve memory at an arbitrary address, only if that area is
2934// available (and not reserved for something else).
2935
2936char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2937  const int max_tries = 10;
2938  char* base[max_tries];
2939  size_t size[max_tries];
2940
2941  // Solaris adds a gap between mmap'ed regions.  The size of the gap
2942  // is dependent on the requested size and the MMU.  Our initial gap
2943  // value here is just a guess and will be corrected later.
2944  bool had_top_overlap = false;
2945  bool have_adjusted_gap = false;
2946  size_t gap = 0x400000;
2947
2948  // Assert only that the size is a multiple of the page size, since
2949  // that's all that mmap requires, and since that's all we really know
2950  // about at this low abstraction level.  If we need higher alignment,
2951  // we can either pass an alignment to this method or verify alignment
2952  // in one of the methods further up the call chain.  See bug 5044738.
2953  assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2954
2955  // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2956  // Give it a try, if the kernel honors the hint we can return immediately.
2957  char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2958  volatile int err = errno;
2959  if (addr == requested_addr) {
2960    return addr;
2961  } else if (addr != NULL) {
2962    unmap_memory(addr, bytes);
2963  }
2964
2965  if (PrintMiscellaneous && Verbose) {
2966    char buf[256];
2967    buf[0] = '\0';
2968    if (addr == NULL) {
2969      jio_snprintf(buf, sizeof(buf), ": %s", strerror(err));
2970    }
2971    warning("attempt_reserve_memory_at: couldn't reserve %d bytes at "
2972            PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2973            "%s", bytes, requested_addr, addr, buf);
2974  }
2975
2976  // Address hint method didn't work.  Fall back to the old method.
2977  // In theory, once SNV becomes our oldest supported platform, this
2978  // code will no longer be needed.
2979  //
2980  // Repeatedly allocate blocks until the block is allocated at the
2981  // right spot. Give up after max_tries.
2982  int i;
2983  for (i = 0; i < max_tries; ++i) {
2984    base[i] = reserve_memory(bytes);
2985
2986    if (base[i] != NULL) {
2987      // Is this the block we wanted?
2988      if (base[i] == requested_addr) {
2989        size[i] = bytes;
2990        break;
2991      }
2992
2993      // check that the gap value is right
2994      if (had_top_overlap && !have_adjusted_gap) {
2995        size_t actual_gap = base[i-1] - base[i] - bytes;
2996        if (gap != actual_gap) {
2997          // adjust the gap value and retry the last 2 allocations
2998          assert(i > 0, "gap adjustment code problem");
2999          have_adjusted_gap = true;  // adjust the gap only once, just in case
3000          gap = actual_gap;
3001          if (PrintMiscellaneous && Verbose) {
3002            warning("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
3003          }
3004          unmap_memory(base[i], bytes);
3005          unmap_memory(base[i-1], size[i-1]);
3006          i-=2;
3007          continue;
3008        }
3009      }
3010
3011      // Does this overlap the block we wanted? Give back the overlapped
3012      // parts and try again.
3013      //
3014      // There is still a bug in this code: if top_overlap == bytes,
3015      // the overlap is offset from requested region by the value of gap.
3016      // In this case giving back the overlapped part will not work,
3017      // because we'll give back the entire block at base[i] and
3018      // therefore the subsequent allocation will not generate a new gap.
3019      // This could be fixed with a new algorithm that used larger
3020      // or variable size chunks to find the requested region -
3021      // but such a change would introduce additional complications.
3022      // It's rare enough that the planets align for this bug,
3023      // so we'll just wait for a fix for 6204603/5003415 which
3024      // will provide a mmap flag to allow us to avoid this business.
3025
3026      size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3027      if (top_overlap >= 0 && top_overlap < bytes) {
3028        had_top_overlap = true;
3029        unmap_memory(base[i], top_overlap);
3030        base[i] += top_overlap;
3031        size[i] = bytes - top_overlap;
3032      } else {
3033        size_t bottom_overlap = base[i] + bytes - requested_addr;
3034        if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3035          if (PrintMiscellaneous && Verbose && bottom_overlap == 0) {
3036            warning("attempt_reserve_memory_at: possible alignment bug");
3037          }
3038          unmap_memory(requested_addr, bottom_overlap);
3039          size[i] = bytes - bottom_overlap;
3040        } else {
3041          size[i] = bytes;
3042        }
3043      }
3044    }
3045  }
3046
3047  // Give back the unused reserved pieces.
3048
3049  for (int j = 0; j < i; ++j) {
3050    if (base[j] != NULL) {
3051      unmap_memory(base[j], size[j]);
3052    }
3053  }
3054
3055  return (i < max_tries) ? requested_addr : NULL;
3056}
3057
3058bool os::release_memory(char* addr, size_t bytes) {
3059  size_t size = bytes;
3060  return munmap(addr, size) == 0;
3061}
3062
3063static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
3064  assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
3065         "addr must be page aligned");
3066  int retVal = mprotect(addr, bytes, prot);
3067  return retVal == 0;
3068}
3069
3070// Protect memory (Used to pass readonly pages through
3071// JNI GetArray<type>Elements with empty arrays.)
3072// Also, used for serialization page and for compressed oops null pointer
3073// checking.
3074bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3075                        bool is_committed) {
3076  unsigned int p = 0;
3077  switch (prot) {
3078  case MEM_PROT_NONE: p = PROT_NONE; break;
3079  case MEM_PROT_READ: p = PROT_READ; break;
3080  case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3081  case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3082  default:
3083    ShouldNotReachHere();
3084  }
3085  // is_committed is unused.
3086  return solaris_mprotect(addr, bytes, p);
3087}
3088
3089// guard_memory and unguard_memory only happens within stack guard pages.
3090// Since ISM pertains only to the heap, guard and unguard memory should not
3091/// happen with an ISM region.
3092bool os::guard_memory(char* addr, size_t bytes) {
3093  return solaris_mprotect(addr, bytes, PROT_NONE);
3094}
3095
3096bool os::unguard_memory(char* addr, size_t bytes) {
3097  return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
3098}
3099
3100// Large page support
3101
3102// UseLargePages is the master flag to enable/disable large page memory.
3103// UseMPSS and UseISM are supported for compatibility reasons. Their combined
3104// effects can be described in the following table:
3105//
3106// UseLargePages UseMPSS UseISM
3107//    false         *       *   => UseLargePages is the master switch, turning
3108//                                 it off will turn off both UseMPSS and
3109//                                 UseISM. VM will not use large page memory
3110//                                 regardless the settings of UseMPSS/UseISM.
3111//     true      false    false => Unless future Solaris provides other
3112//                                 mechanism to use large page memory, this
3113//                                 combination is equivalent to -UseLargePages,
3114//                                 VM will not use large page memory
3115//     true      true     false => JVM will use MPSS for large page memory.
3116//                                 This is the default behavior.
3117//     true      false    true  => JVM will use ISM for large page memory.
3118//     true      true     true  => JVM will use ISM if it is available.
3119//                                 Otherwise, JVM will fall back to MPSS.
3120//                                 Becaues ISM is now available on all
3121//                                 supported Solaris versions, this combination
3122//                                 is equivalent to +UseISM -UseMPSS.
3123
3124typedef int (*getpagesizes_func_type) (size_t[], int);
3125static size_t _large_page_size = 0;
3126
3127bool os::Solaris::ism_sanity_check(bool warn, size_t * page_size) {
3128  // x86 uses either 2M or 4M page, depending on whether PAE (Physical Address
3129  // Extensions) mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. Sparc
3130  // can support multiple page sizes.
3131
3132  // Don't bother to probe page size because getpagesizes() comes with MPSS.
3133  // ISM is only recommended on old Solaris where there is no MPSS support.
3134  // Simply choose a conservative value as default.
3135  *page_size = LargePageSizeInBytes ? LargePageSizeInBytes :
3136               SPARC_ONLY(4 * M) IA32_ONLY(4 * M) AMD64_ONLY(2 * M);
3137
3138  // ISM is available on all supported Solaris versions
3139  return true;
3140}
3141
3142// Insertion sort for small arrays (descending order).
3143static void insertion_sort_descending(size_t* array, int len) {
3144  for (int i = 0; i < len; i++) {
3145    size_t val = array[i];
3146    for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
3147      size_t tmp = array[key];
3148      array[key] = array[key - 1];
3149      array[key - 1] = tmp;
3150    }
3151  }
3152}
3153
3154bool os::Solaris::mpss_sanity_check(bool warn, size_t * page_size) {
3155  getpagesizes_func_type getpagesizes_func =
3156    CAST_TO_FN_PTR(getpagesizes_func_type, dlsym(RTLD_DEFAULT, "getpagesizes"));
3157  if (getpagesizes_func == NULL) {
3158    if (warn) {
3159      warning("MPSS is not supported by the operating system.");
3160    }
3161    return false;
3162  }
3163
3164  const unsigned int usable_count = VM_Version::page_size_count();
3165  if (usable_count == 1) {
3166    return false;
3167  }
3168
3169  // Fill the array of page sizes.
3170  int n = getpagesizes_func(_page_sizes, page_sizes_max);
3171  assert(n > 0, "Solaris bug?");
3172  if (n == page_sizes_max) {
3173    // Add a sentinel value (necessary only if the array was completely filled
3174    // since it is static (zeroed at initialization)).
3175    _page_sizes[--n] = 0;
3176    DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
3177  }
3178  assert(_page_sizes[n] == 0, "missing sentinel");
3179
3180  if (n == 1) return false;     // Only one page size available.
3181
3182  // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
3183  // select up to usable_count elements.  First sort the array, find the first
3184  // acceptable value, then copy the usable sizes to the top of the array and
3185  // trim the rest.  Make sure to include the default page size :-).
3186  //
3187  // A better policy could get rid of the 4M limit by taking the sizes of the
3188  // important VM memory regions (java heap and possibly the code cache) into
3189  // account.
3190  insertion_sort_descending(_page_sizes, n);
3191  const size_t size_limit =
3192    FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
3193  int beg;
3194  for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */ ;
3195  const int end = MIN2((int)usable_count, n) - 1;
3196  for (int cur = 0; cur < end; ++cur, ++beg) {
3197    _page_sizes[cur] = _page_sizes[beg];
3198  }
3199  _page_sizes[end] = vm_page_size();
3200  _page_sizes[end + 1] = 0;
3201
3202  if (_page_sizes[end] > _page_sizes[end - 1]) {
3203    // Default page size is not the smallest; sort again.
3204    insertion_sort_descending(_page_sizes, end + 1);
3205  }
3206  *page_size = _page_sizes[0];
3207
3208  return true;
3209}
3210
3211bool os::large_page_init() {
3212  if (!UseLargePages) {
3213    UseISM = false;
3214    UseMPSS = false;
3215    return false;
3216  }
3217
3218  // print a warning if any large page related flag is specified on command line
3219  bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
3220                         !FLAG_IS_DEFAULT(UseISM)               ||
3221                         !FLAG_IS_DEFAULT(UseMPSS)              ||
3222                         !FLAG_IS_DEFAULT(LargePageSizeInBytes);
3223  UseISM = UseISM &&
3224           Solaris::ism_sanity_check(warn_on_failure, &_large_page_size);
3225  if (UseISM) {
3226    // ISM disables MPSS to be compatible with old JDK behavior
3227    UseMPSS = false;
3228    _page_sizes[0] = _large_page_size;
3229    _page_sizes[1] = vm_page_size();
3230  }
3231
3232  UseMPSS = UseMPSS &&
3233            Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
3234
3235  UseLargePages = UseISM || UseMPSS;
3236  return UseLargePages;
3237}
3238
3239bool os::Solaris::set_mpss_range(caddr_t start, size_t bytes, size_t align) {
3240  // Signal to OS that we want large pages for addresses
3241  // from addr, addr + bytes
3242  struct memcntl_mha mpss_struct;
3243  mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
3244  mpss_struct.mha_pagesize = align;
3245  mpss_struct.mha_flags = 0;
3246  if (memcntl(start, bytes, MC_HAT_ADVISE,
3247              (caddr_t) &mpss_struct, 0, 0) < 0) {
3248    debug_only(warning("Attempt to use MPSS failed."));
3249    return false;
3250  }
3251  return true;
3252}
3253
3254char* os::reserve_memory_special(size_t bytes, char* addr, bool exec) {
3255  // "exec" is passed in but not used.  Creating the shared image for
3256  // the code cache doesn't have an SHM_X executable permission to check.
3257  assert(UseLargePages && UseISM, "only for ISM large pages");
3258
3259  size_t size = bytes;
3260  char* retAddr = NULL;
3261  int shmid;
3262  key_t ismKey;
3263
3264  bool warn_on_failure = UseISM &&
3265                        (!FLAG_IS_DEFAULT(UseLargePages)         ||
3266                         !FLAG_IS_DEFAULT(UseISM)                ||
3267                         !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3268                        );
3269  char msg[128];
3270
3271  ismKey = IPC_PRIVATE;
3272
3273  // Create a large shared memory region to attach to based on size.
3274  // Currently, size is the total size of the heap
3275  shmid = shmget(ismKey, size, SHM_R | SHM_W | IPC_CREAT);
3276  if (shmid == -1){
3277     if (warn_on_failure) {
3278       jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3279       warning(msg);
3280     }
3281     return NULL;
3282  }
3283
3284  // Attach to the region
3285  retAddr = (char *) shmat(shmid, 0, SHM_SHARE_MMU | SHM_R | SHM_W);
3286  int err = errno;
3287
3288  // Remove shmid. If shmat() is successful, the actual shared memory segment
3289  // will be deleted when it's detached by shmdt() or when the process
3290  // terminates. If shmat() is not successful this will remove the shared
3291  // segment immediately.
3292  shmctl(shmid, IPC_RMID, NULL);
3293
3294  if (retAddr == (char *) -1) {
3295    if (warn_on_failure) {
3296      jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3297      warning(msg);
3298    }
3299    return NULL;
3300  }
3301
3302  return retAddr;
3303}
3304
3305bool os::release_memory_special(char* base, size_t bytes) {
3306  // detaching the SHM segment will also delete it, see reserve_memory_special()
3307  int rslt = shmdt(base);
3308  return rslt == 0;
3309}
3310
3311size_t os::large_page_size() {
3312  return _large_page_size;
3313}
3314
3315// MPSS allows application to commit large page memory on demand; with ISM
3316// the entire memory region must be allocated as shared memory.
3317bool os::can_commit_large_page_memory() {
3318  return UseISM ? false : true;
3319}
3320
3321bool os::can_execute_large_page_memory() {
3322  return UseISM ? false : true;
3323}
3324
3325static int os_sleep(jlong millis, bool interruptible) {
3326  const jlong limit = INT_MAX;
3327  jlong prevtime;
3328  int res;
3329
3330  while (millis > limit) {
3331    if ((res = os_sleep(limit, interruptible)) != OS_OK)
3332      return res;
3333    millis -= limit;
3334  }
3335
3336  // Restart interrupted polls with new parameters until the proper delay
3337  // has been completed.
3338
3339  prevtime = getTimeMillis();
3340
3341  while (millis > 0) {
3342    jlong newtime;
3343
3344    if (!interruptible) {
3345      // Following assert fails for os::yield_all:
3346      // assert(!thread->is_Java_thread(), "must not be java thread");
3347      res = poll(NULL, 0, millis);
3348    } else {
3349      JavaThread *jt = JavaThread::current();
3350
3351      INTERRUPTIBLE_NORESTART_VM_ALWAYS(poll(NULL, 0, millis), res, jt,
3352        os::Solaris::clear_interrupted);
3353    }
3354
3355    // INTERRUPTIBLE_NORESTART_VM_ALWAYS returns res == OS_INTRPT for
3356    // thread.Interrupt.
3357
3358    if((res == OS_ERR) && (errno == EINTR)) {
3359      newtime = getTimeMillis();
3360      assert(newtime >= prevtime, "time moving backwards");
3361    /* Doing prevtime and newtime in microseconds doesn't help precision,
3362       and trying to round up to avoid lost milliseconds can result in a
3363       too-short delay. */
3364      millis -= newtime - prevtime;
3365      if(millis <= 0)
3366        return OS_OK;
3367      prevtime = newtime;
3368    } else
3369      return res;
3370  }
3371
3372  return OS_OK;
3373}
3374
3375// Read calls from inside the vm need to perform state transitions
3376size_t os::read(int fd, void *buf, unsigned int nBytes) {
3377  INTERRUPTIBLE_RETURN_INT_VM(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3378}
3379
3380int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3381  assert(thread == Thread::current(),  "thread consistency check");
3382
3383  // TODO-FIXME: this should be removed.
3384  // On Solaris machines (especially 2.5.1) we found that sometimes the VM gets into a live lock
3385  // situation with a JavaThread being starved out of a lwp. The kernel doesn't seem to generate
3386  // a SIGWAITING signal which would enable the threads library to create a new lwp for the starving
3387  // thread. We suspect that because the Watcher thread keeps waking up at periodic intervals the kernel
3388  // is fooled into believing that the system is making progress. In the code below we block the
3389  // the watcher thread while safepoint is in progress so that it would not appear as though the
3390  // system is making progress.
3391  if (!Solaris::T2_libthread() &&
3392      thread->is_Watcher_thread() && SafepointSynchronize::is_synchronizing() && !Arguments::has_profile()) {
3393    // We now try to acquire the threads lock. Since this lock is held by the VM thread during
3394    // the entire safepoint, the watcher thread will  line up here during the safepoint.
3395    Threads_lock->lock_without_safepoint_check();
3396    Threads_lock->unlock();
3397  }
3398
3399  if (thread->is_Java_thread()) {
3400    // This is a JavaThread so we honor the _thread_blocked protocol
3401    // even for sleeps of 0 milliseconds. This was originally done
3402    // as a workaround for bug 4338139. However, now we also do it
3403    // to honor the suspend-equivalent protocol.
3404
3405    JavaThread *jt = (JavaThread *) thread;
3406    ThreadBlockInVM tbivm(jt);
3407
3408    jt->set_suspend_equivalent();
3409    // cleared by handle_special_suspend_equivalent_condition() or
3410    // java_suspend_self() via check_and_wait_while_suspended()
3411
3412    int ret_code;
3413    if (millis <= 0) {
3414      thr_yield();
3415      ret_code = 0;
3416    } else {
3417      // The original sleep() implementation did not create an
3418      // OSThreadWaitState helper for sleeps of 0 milliseconds.
3419      // I'm preserving that decision for now.
3420      OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3421
3422      ret_code = os_sleep(millis, interruptible);
3423    }
3424
3425    // were we externally suspended while we were waiting?
3426    jt->check_and_wait_while_suspended();
3427
3428    return ret_code;
3429  }
3430
3431  // non-JavaThread from this point on:
3432
3433  if (millis <= 0) {
3434    thr_yield();
3435    return 0;
3436  }
3437
3438  OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3439
3440  return os_sleep(millis, interruptible);
3441}
3442
3443int os::naked_sleep() {
3444  // %% make the sleep time an integer flag. for now use 1 millisec.
3445  return os_sleep(1, false);
3446}
3447
3448// Sleep forever; naked call to OS-specific sleep; use with CAUTION
3449void os::infinite_sleep() {
3450  while (true) {    // sleep forever ...
3451    ::sleep(100);   // ... 100 seconds at a time
3452  }
3453}
3454
3455// Used to convert frequent JVM_Yield() to nops
3456bool os::dont_yield() {
3457  if (DontYieldALot) {
3458    static hrtime_t last_time = 0;
3459    hrtime_t diff = getTimeNanos() - last_time;
3460
3461    if (diff < DontYieldALotInterval * 1000000)
3462      return true;
3463
3464    last_time += diff;
3465
3466    return false;
3467  }
3468  else {
3469    return false;
3470  }
3471}
3472
3473// Caveat: Solaris os::yield() causes a thread-state transition whereas
3474// the linux and win32 implementations do not.  This should be checked.
3475
3476void os::yield() {
3477  // Yields to all threads with same or greater priority
3478  os::sleep(Thread::current(), 0, false);
3479}
3480
3481// Note that yield semantics are defined by the scheduling class to which
3482// the thread currently belongs.  Typically, yield will _not yield to
3483// other equal or higher priority threads that reside on the dispatch queues
3484// of other CPUs.
3485
3486os::YieldResult os::NakedYield() { thr_yield(); return os::YIELD_UNKNOWN; }
3487
3488
3489// On Solaris we found that yield_all doesn't always yield to all other threads.
3490// There have been cases where there is a thread ready to execute but it doesn't
3491// get an lwp as the VM thread continues to spin with sleeps of 1 millisecond.
3492// The 1 millisecond wait doesn't seem long enough for the kernel to issue a
3493// SIGWAITING signal which will cause a new lwp to be created. So we count the
3494// number of times yield_all is called in the one loop and increase the sleep
3495// time after 8 attempts. If this fails too we increase the concurrency level
3496// so that the starving thread would get an lwp
3497
3498void os::yield_all(int attempts) {
3499  // Yields to all threads, including threads with lower priorities
3500  if (attempts == 0) {
3501    os::sleep(Thread::current(), 1, false);
3502  } else {
3503    int iterations = attempts % 30;
3504    if (iterations == 0 && !os::Solaris::T2_libthread()) {
3505      // thr_setconcurrency and _getconcurrency make sense only under T1.
3506      int noofLWPS = thr_getconcurrency();
3507      if (noofLWPS < (Threads::number_of_threads() + 2)) {
3508        thr_setconcurrency(thr_getconcurrency() + 1);
3509      }
3510    } else if (iterations < 25) {
3511      os::sleep(Thread::current(), 1, false);
3512    } else {
3513      os::sleep(Thread::current(), 10, false);
3514    }
3515  }
3516}
3517
3518// Called from the tight loops to possibly influence time-sharing heuristics
3519void os::loop_breaker(int attempts) {
3520  os::yield_all(attempts);
3521}
3522
3523
3524// Interface for setting lwp priorities.  If we are using T2 libthread,
3525// which forces the use of BoundThreads or we manually set UseBoundThreads,
3526// all of our threads will be assigned to real lwp's.  Using the thr_setprio
3527// function is meaningless in this mode so we must adjust the real lwp's priority
3528// The routines below implement the getting and setting of lwp priorities.
3529//
3530// Note: There are three priority scales used on Solaris.  Java priotities
3531//       which range from 1 to 10, libthread "thr_setprio" scale which range
3532//       from 0 to 127, and the current scheduling class of the process we
3533//       are running in.  This is typically from -60 to +60.
3534//       The setting of the lwp priorities in done after a call to thr_setprio
3535//       so Java priorities are mapped to libthread priorities and we map from
3536//       the latter to lwp priorities.  We don't keep priorities stored in
3537//       Java priorities since some of our worker threads want to set priorities
3538//       higher than all Java threads.
3539//
3540// For related information:
3541// (1)  man -s 2 priocntl
3542// (2)  man -s 4 priocntl
3543// (3)  man dispadmin
3544// =    librt.so
3545// =    libthread/common/rtsched.c - thrp_setlwpprio().
3546// =    ps -cL <pid> ... to validate priority.
3547// =    sched_get_priority_min and _max
3548//              pthread_create
3549//              sched_setparam
3550//              pthread_setschedparam
3551//
3552// Assumptions:
3553// +    We assume that all threads in the process belong to the same
3554//              scheduling class.   IE. an homogenous process.
3555// +    Must be root or in IA group to change change "interactive" attribute.
3556//              Priocntl() will fail silently.  The only indication of failure is when
3557//              we read-back the value and notice that it hasn't changed.
3558// +    Interactive threads enter the runq at the head, non-interactive at the tail.
3559// +    For RT, change timeslice as well.  Invariant:
3560//              constant "priority integral"
3561//              Konst == TimeSlice * (60-Priority)
3562//              Given a priority, compute appropriate timeslice.
3563// +    Higher numerical values have higher priority.
3564
3565// sched class attributes
3566typedef struct {
3567        int   schedPolicy;              // classID
3568        int   maxPrio;
3569        int   minPrio;
3570} SchedInfo;
3571
3572
3573static SchedInfo tsLimits, iaLimits, rtLimits;
3574
3575#ifdef ASSERT
3576static int  ReadBackValidate = 1;
3577#endif
3578static int  myClass     = 0;
3579static int  myMin       = 0;
3580static int  myMax       = 0;
3581static int  myCur       = 0;
3582static bool priocntl_enable = false;
3583
3584
3585// Call the version of priocntl suitable for all supported versions
3586// of Solaris. We need to call through this wrapper so that we can
3587// build on Solaris 9 and run on Solaris 8, 9 and 10.
3588//
3589// This code should be removed if we ever stop supporting Solaris 8
3590// and earlier releases.
3591
3592static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3593typedef long (*priocntl_type)(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3594static priocntl_type priocntl_ptr = priocntl_stub;
3595
3596// Stub to set the value of the real pointer, and then call the real
3597// function.
3598
3599static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg) {
3600  // Try Solaris 8- name only.
3601  priocntl_type tmp = (priocntl_type)dlsym(RTLD_DEFAULT, "__priocntl");
3602  guarantee(tmp != NULL, "priocntl function not found.");
3603  priocntl_ptr = tmp;
3604  return (*priocntl_ptr)(PC_VERSION, idtype, id, cmd, arg);
3605}
3606
3607
3608// lwp_priocntl_init
3609//
3610// Try to determine the priority scale for our process.
3611//
3612// Return errno or 0 if OK.
3613//
3614static
3615int     lwp_priocntl_init ()
3616{
3617  int rslt;
3618  pcinfo_t ClassInfo;
3619  pcparms_t ParmInfo;
3620  int i;
3621
3622  if (!UseThreadPriorities) return 0;
3623
3624  // We are using Bound threads, we need to determine our priority ranges
3625  if (os::Solaris::T2_libthread() || UseBoundThreads) {
3626    // If ThreadPriorityPolicy is 1, switch tables
3627    if (ThreadPriorityPolicy == 1) {
3628      for (i = 0 ; i < MaxPriority+1; i++)
3629        os::java_to_os_priority[i] = prio_policy1[i];
3630    }
3631  }
3632  // Not using Bound Threads, set to ThreadPolicy 1
3633  else {
3634    for ( i = 0 ; i < MaxPriority+1; i++ ) {
3635      os::java_to_os_priority[i] = prio_policy1[i];
3636    }
3637    return 0;
3638  }
3639
3640
3641  // Get IDs for a set of well-known scheduling classes.
3642  // TODO-FIXME: GETCLINFO returns the current # of classes in the
3643  // the system.  We should have a loop that iterates over the
3644  // classID values, which are known to be "small" integers.
3645
3646  strcpy(ClassInfo.pc_clname, "TS");
3647  ClassInfo.pc_cid = -1;
3648  rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3649  if (rslt < 0) return errno;
3650  assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3651  tsLimits.schedPolicy = ClassInfo.pc_cid;
3652  tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3653  tsLimits.minPrio = -tsLimits.maxPrio;
3654
3655  strcpy(ClassInfo.pc_clname, "IA");
3656  ClassInfo.pc_cid = -1;
3657  rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3658  if (rslt < 0) return errno;
3659  assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3660  iaLimits.schedPolicy = ClassInfo.pc_cid;
3661  iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3662  iaLimits.minPrio = -iaLimits.maxPrio;
3663
3664  strcpy(ClassInfo.pc_clname, "RT");
3665  ClassInfo.pc_cid = -1;
3666  rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3667  if (rslt < 0) return errno;
3668  assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3669  rtLimits.schedPolicy = ClassInfo.pc_cid;
3670  rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3671  rtLimits.minPrio = 0;
3672
3673
3674  // Query our "current" scheduling class.
3675  // This will normally be IA,TS or, rarely, RT.
3676  memset (&ParmInfo, 0, sizeof(ParmInfo));
3677  ParmInfo.pc_cid = PC_CLNULL;
3678  rslt = (*priocntl_ptr) (PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo );
3679  if ( rslt < 0 ) return errno;
3680  myClass = ParmInfo.pc_cid;
3681
3682  // We now know our scheduling classId, get specific information
3683  // the class.
3684  ClassInfo.pc_cid = myClass;
3685  ClassInfo.pc_clname[0] = 0;
3686  rslt = (*priocntl_ptr) (PC_VERSION, (idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo );
3687  if ( rslt < 0 ) return errno;
3688
3689  if (ThreadPriorityVerbose)
3690    tty->print_cr ("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3691
3692  memset(&ParmInfo, 0, sizeof(pcparms_t));
3693  ParmInfo.pc_cid = PC_CLNULL;
3694  rslt = (*priocntl_ptr)(PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3695  if (rslt < 0) return errno;
3696
3697  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3698    myMin = rtLimits.minPrio;
3699    myMax = rtLimits.maxPrio;
3700  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3701    iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3702    myMin = iaLimits.minPrio;
3703    myMax = iaLimits.maxPrio;
3704    myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3705  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3706    tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3707    myMin = tsLimits.minPrio;
3708    myMax = tsLimits.maxPrio;
3709    myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3710  } else {
3711    // No clue - punt
3712    if (ThreadPriorityVerbose)
3713      tty->print_cr ("Unknown scheduling class: %s ... \n", ClassInfo.pc_clname);
3714    return EINVAL;      // no clue, punt
3715  }
3716
3717  if (ThreadPriorityVerbose)
3718        tty->print_cr ("Thread priority Range: [%d..%d]\n", myMin, myMax);
3719
3720  priocntl_enable = true;  // Enable changing priorities
3721  return 0;
3722}
3723
3724#define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3725#define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3726#define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3727
3728
3729// scale_to_lwp_priority
3730//
3731// Convert from the libthread "thr_setprio" scale to our current
3732// lwp scheduling class scale.
3733//
3734static
3735int     scale_to_lwp_priority (int rMin, int rMax, int x)
3736{
3737  int v;
3738
3739  if (x == 127) return rMax;            // avoid round-down
3740    v = (((x*(rMax-rMin)))/128)+rMin;
3741  return v;
3742}
3743
3744
3745// set_lwp_priority
3746//
3747// Set the priority of the lwp.  This call should only be made
3748// when using bound threads (T2 threads are bound by default).
3749//
3750int     set_lwp_priority (int ThreadID, int lwpid, int newPrio )
3751{
3752  int rslt;
3753  int Actual, Expected, prv;
3754  pcparms_t ParmInfo;                   // for GET-SET
3755#ifdef ASSERT
3756  pcparms_t ReadBack;                   // for readback
3757#endif
3758
3759  // Set priority via PC_GETPARMS, update, PC_SETPARMS
3760  // Query current values.
3761  // TODO: accelerate this by eliminating the PC_GETPARMS call.
3762  // Cache "pcparms_t" in global ParmCache.
3763  // TODO: elide set-to-same-value
3764
3765  // If something went wrong on init, don't change priorities.
3766  if ( !priocntl_enable ) {
3767    if (ThreadPriorityVerbose)
3768      tty->print_cr("Trying to set priority but init failed, ignoring");
3769    return EINVAL;
3770  }
3771
3772
3773  // If lwp hasn't started yet, just return
3774  // the _start routine will call us again.
3775  if ( lwpid <= 0 ) {
3776    if (ThreadPriorityVerbose) {
3777      tty->print_cr ("deferring the set_lwp_priority of thread " INTPTR_FORMAT " to %d, lwpid not set",
3778                     ThreadID, newPrio);
3779    }
3780    return 0;
3781  }
3782
3783  if (ThreadPriorityVerbose) {
3784    tty->print_cr ("set_lwp_priority(" INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3785                   ThreadID, lwpid, newPrio);
3786  }
3787
3788  memset(&ParmInfo, 0, sizeof(pcparms_t));
3789  ParmInfo.pc_cid = PC_CLNULL;
3790  rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3791  if (rslt < 0) return errno;
3792
3793  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3794    rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3795    rtInfo->rt_pri     = scale_to_lwp_priority (rtLimits.minPrio, rtLimits.maxPrio, newPrio);
3796    rtInfo->rt_tqsecs  = RT_NOCHANGE;
3797    rtInfo->rt_tqnsecs = RT_NOCHANGE;
3798    if (ThreadPriorityVerbose) {
3799      tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3800    }
3801  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3802    iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3803    int maxClamped     = MIN2(iaLimits.maxPrio, (int)iaInfo->ia_uprilim);
3804    iaInfo->ia_upri    = scale_to_lwp_priority(iaLimits.minPrio, maxClamped, newPrio);
3805    iaInfo->ia_uprilim = IA_NOCHANGE;
3806    iaInfo->ia_mode    = IA_NOCHANGE;
3807    if (ThreadPriorityVerbose) {
3808      tty->print_cr ("IA: [%d...%d] %d->%d\n",
3809               iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3810    }
3811  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3812    tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3813    int maxClamped     = MIN2(tsLimits.maxPrio, (int)tsInfo->ts_uprilim);
3814    prv                = tsInfo->ts_upri;
3815    tsInfo->ts_upri    = scale_to_lwp_priority(tsLimits.minPrio, maxClamped, newPrio);
3816    tsInfo->ts_uprilim = IA_NOCHANGE;
3817    if (ThreadPriorityVerbose) {
3818      tty->print_cr ("TS: %d [%d...%d] %d->%d\n",
3819               prv, tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3820    }
3821    if (prv == tsInfo->ts_upri) return 0;
3822  } else {
3823    if ( ThreadPriorityVerbose ) {
3824      tty->print_cr ("Unknown scheduling class\n");
3825    }
3826      return EINVAL;    // no clue, punt
3827  }
3828
3829  rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3830  if (ThreadPriorityVerbose && rslt) {
3831    tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3832  }
3833  if (rslt < 0) return errno;
3834
3835#ifdef ASSERT
3836  // Sanity check: read back what we just attempted to set.
3837  // In theory it could have changed in the interim ...
3838  //
3839  // The priocntl system call is tricky.
3840  // Sometimes it'll validate the priority value argument and
3841  // return EINVAL if unhappy.  At other times it fails silently.
3842  // Readbacks are prudent.
3843
3844  if (!ReadBackValidate) return 0;
3845
3846  memset(&ReadBack, 0, sizeof(pcparms_t));
3847  ReadBack.pc_cid = PC_CLNULL;
3848  rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3849  assert(rslt >= 0, "priocntl failed");
3850  Actual = Expected = 0xBAD;
3851  assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3852  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3853    Actual   = RTPRI(ReadBack)->rt_pri;
3854    Expected = RTPRI(ParmInfo)->rt_pri;
3855  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3856    Actual   = IAPRI(ReadBack)->ia_upri;
3857    Expected = IAPRI(ParmInfo)->ia_upri;
3858  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3859    Actual   = TSPRI(ReadBack)->ts_upri;
3860    Expected = TSPRI(ParmInfo)->ts_upri;
3861  } else {
3862    if ( ThreadPriorityVerbose ) {
3863      tty->print_cr("set_lwp_priority: unexpected class in readback: %d\n", ParmInfo.pc_cid);
3864    }
3865  }
3866
3867  if (Actual != Expected) {
3868    if ( ThreadPriorityVerbose ) {
3869      tty->print_cr ("set_lwp_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3870             lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3871    }
3872  }
3873#endif
3874
3875  return 0;
3876}
3877
3878
3879
3880// Solaris only gives access to 128 real priorities at a time,
3881// so we expand Java's ten to fill this range.  This would be better
3882// if we dynamically adjusted relative priorities.
3883//
3884// The ThreadPriorityPolicy option allows us to select 2 different
3885// priority scales.
3886//
3887// ThreadPriorityPolicy=0
3888// Since the Solaris' default priority is MaximumPriority, we do not
3889// set a priority lower than Max unless a priority lower than
3890// NormPriority is requested.
3891//
3892// ThreadPriorityPolicy=1
3893// This mode causes the priority table to get filled with
3894// linear values.  NormPriority get's mapped to 50% of the
3895// Maximum priority an so on.  This will cause VM threads
3896// to get unfair treatment against other Solaris processes
3897// which do not explicitly alter their thread priorities.
3898//
3899
3900
3901int os::java_to_os_priority[MaxPriority + 1] = {
3902  -99999,         // 0 Entry should never be used
3903
3904  0,              // 1 MinPriority
3905  32,             // 2
3906  64,             // 3
3907
3908  96,             // 4
3909  127,            // 5 NormPriority
3910  127,            // 6
3911
3912  127,            // 7
3913  127,            // 8
3914  127,            // 9 NearMaxPriority
3915
3916  127             // 10 MaxPriority
3917};
3918
3919
3920OSReturn os::set_native_priority(Thread* thread, int newpri) {
3921  assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3922  if ( !UseThreadPriorities ) return OS_OK;
3923  int status = thr_setprio(thread->osthread()->thread_id(), newpri);
3924  if ( os::Solaris::T2_libthread() || (UseBoundThreads && thread->osthread()->is_vm_created()) )
3925    status |= (set_lwp_priority (thread->osthread()->thread_id(),
3926                    thread->osthread()->lwp_id(), newpri ));
3927  return (status == 0) ? OS_OK : OS_ERR;
3928}
3929
3930
3931OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3932  int p;
3933  if ( !UseThreadPriorities ) {
3934    *priority_ptr = NormalPriority;
3935    return OS_OK;
3936  }
3937  int status = thr_getprio(thread->osthread()->thread_id(), &p);
3938  if (status != 0) {
3939    return OS_ERR;
3940  }
3941  *priority_ptr = p;
3942  return OS_OK;
3943}
3944
3945
3946// Hint to the underlying OS that a task switch would not be good.
3947// Void return because it's a hint and can fail.
3948void os::hint_no_preempt() {
3949  schedctl_start(schedctl_init());
3950}
3951
3952void os::interrupt(Thread* thread) {
3953  assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
3954
3955  OSThread* osthread = thread->osthread();
3956
3957  int isInterrupted = osthread->interrupted();
3958  if (!isInterrupted) {
3959      osthread->set_interrupted(true);
3960      OrderAccess::fence();
3961      // os::sleep() is implemented with either poll (NULL,0,timeout) or
3962      // by parking on _SleepEvent.  If the former, thr_kill will unwedge
3963      // the sleeper by SIGINTR, otherwise the unpark() will wake the sleeper.
3964      ParkEvent * const slp = thread->_SleepEvent ;
3965      if (slp != NULL) slp->unpark() ;
3966  }
3967
3968  // For JSR166:  unpark after setting status but before thr_kill -dl
3969  if (thread->is_Java_thread()) {
3970    ((JavaThread*)thread)->parker()->unpark();
3971  }
3972
3973  // Handle interruptible wait() ...
3974  ParkEvent * const ev = thread->_ParkEvent ;
3975  if (ev != NULL) ev->unpark() ;
3976
3977  // When events are used everywhere for os::sleep, then this thr_kill
3978  // will only be needed if UseVMInterruptibleIO is true.
3979
3980  if (!isInterrupted) {
3981    int status = thr_kill(osthread->thread_id(), os::Solaris::SIGinterrupt());
3982    assert_status(status == 0, status, "thr_kill");
3983
3984    // Bump thread interruption counter
3985    RuntimeService::record_thread_interrupt_signaled_count();
3986  }
3987}
3988
3989
3990bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3991  assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
3992
3993  OSThread* osthread = thread->osthread();
3994
3995  bool res = osthread->interrupted();
3996
3997  // NOTE that since there is no "lock" around these two operations,
3998  // there is the possibility that the interrupted flag will be
3999  // "false" but that the interrupt event will be set. This is
4000  // intentional. The effect of this is that Object.wait() will appear
4001  // to have a spurious wakeup, which is not harmful, and the
4002  // possibility is so rare that it is not worth the added complexity
4003  // to add yet another lock. It has also been recommended not to put
4004  // the interrupted flag into the os::Solaris::Event structure,
4005  // because it hides the issue.
4006  if (res && clear_interrupted) {
4007    osthread->set_interrupted(false);
4008  }
4009  return res;
4010}
4011
4012
4013void os::print_statistics() {
4014}
4015
4016int os::message_box(const char* title, const char* message) {
4017  int i;
4018  fdStream err(defaultStream::error_fd());
4019  for (i = 0; i < 78; i++) err.print_raw("=");
4020  err.cr();
4021  err.print_raw_cr(title);
4022  for (i = 0; i < 78; i++) err.print_raw("-");
4023  err.cr();
4024  err.print_raw_cr(message);
4025  for (i = 0; i < 78; i++) err.print_raw("=");
4026  err.cr();
4027
4028  char buf[16];
4029  // Prevent process from exiting upon "read error" without consuming all CPU
4030  while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4031
4032  return buf[0] == 'y' || buf[0] == 'Y';
4033}
4034
4035// A lightweight implementation that does not suspend the target thread and
4036// thus returns only a hint. Used for profiling only!
4037ExtendedPC os::get_thread_pc(Thread* thread) {
4038  // Make sure that it is called by the watcher and the Threads lock is owned.
4039  assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
4040  // For now, is only used to profile the VM Thread
4041  assert(thread->is_VM_thread(), "Can only be called for VMThread");
4042  ExtendedPC epc;
4043
4044  GetThreadPC_Callback  cb(ProfileVM_lock);
4045  OSThread *osthread = thread->osthread();
4046  const int time_to_wait = 400; // 400ms wait for initial response
4047  int status = cb.interrupt(thread, time_to_wait);
4048
4049  if (cb.is_done() ) {
4050    epc = cb.addr();
4051  } else {
4052    DEBUG_ONLY(tty->print_cr("Failed to get pc for thread: %d got %d status",
4053                              osthread->thread_id(), status););
4054    // epc is already NULL
4055  }
4056  return epc;
4057}
4058
4059
4060// This does not do anything on Solaris. This is basically a hook for being
4061// able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
4062void os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread) {
4063  f(value, method, args, thread);
4064}
4065
4066// This routine may be used by user applications as a "hook" to catch signals.
4067// The user-defined signal handler must pass unrecognized signals to this
4068// routine, and if it returns true (non-zero), then the signal handler must
4069// return immediately.  If the flag "abort_if_unrecognized" is true, then this
4070// routine will never retun false (zero), but instead will execute a VM panic
4071// routine kill the process.
4072//
4073// If this routine returns false, it is OK to call it again.  This allows
4074// the user-defined signal handler to perform checks either before or after
4075// the VM performs its own checks.  Naturally, the user code would be making
4076// a serious error if it tried to handle an exception (such as a null check
4077// or breakpoint) that the VM was generating for its own correct operation.
4078//
4079// This routine may recognize any of the following kinds of signals:
4080// SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
4081// os::Solaris::SIGasync
4082// It should be consulted by handlers for any of those signals.
4083// It explicitly does not recognize os::Solaris::SIGinterrupt
4084//
4085// The caller of this routine must pass in the three arguments supplied
4086// to the function referred to in the "sa_sigaction" (not the "sa_handler")
4087// field of the structure passed to sigaction().  This routine assumes that
4088// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4089//
4090// Note that the VM will print warnings if it detects conflicting signal
4091// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4092//
4093extern "C" int JVM_handle_solaris_signal(int signo, siginfo_t* siginfo, void* ucontext, int abort_if_unrecognized);
4094
4095
4096void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
4097  JVM_handle_solaris_signal(sig, info, ucVoid, true);
4098}
4099
4100/* Do not delete - if guarantee is ever removed,  a signal handler (even empty)
4101   is needed to provoke threads blocked on IO to return an EINTR
4102   Note: this explicitly does NOT call JVM_handle_solaris_signal and
4103   does NOT participate in signal chaining due to requirement for
4104   NOT setting SA_RESTART to make EINTR work. */
4105extern "C" void sigINTRHandler(int sig, siginfo_t* info, void* ucVoid) {
4106   if (UseSignalChaining) {
4107      struct sigaction *actp = os::Solaris::get_chained_signal_action(sig);
4108      if (actp && actp->sa_handler) {
4109        vm_exit_during_initialization("Signal chaining detected for VM interrupt signal, try -XX:+UseAltSigs");
4110      }
4111   }
4112}
4113
4114// This boolean allows users to forward their own non-matching signals
4115// to JVM_handle_solaris_signal, harmlessly.
4116bool os::Solaris::signal_handlers_are_installed = false;
4117
4118// For signal-chaining
4119bool os::Solaris::libjsig_is_loaded = false;
4120typedef struct sigaction *(*get_signal_t)(int);
4121get_signal_t os::Solaris::get_signal_action = NULL;
4122
4123struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
4124  struct sigaction *actp = NULL;
4125
4126  if ((libjsig_is_loaded)  && (sig <= Maxlibjsigsigs)) {
4127    // Retrieve the old signal handler from libjsig
4128    actp = (*get_signal_action)(sig);
4129  }
4130  if (actp == NULL) {
4131    // Retrieve the preinstalled signal handler from jvm
4132    actp = get_preinstalled_handler(sig);
4133  }
4134
4135  return actp;
4136}
4137
4138static bool call_chained_handler(struct sigaction *actp, int sig,
4139                                 siginfo_t *siginfo, void *context) {
4140  // Call the old signal handler
4141  if (actp->sa_handler == SIG_DFL) {
4142    // It's more reasonable to let jvm treat it as an unexpected exception
4143    // instead of taking the default action.
4144    return false;
4145  } else if (actp->sa_handler != SIG_IGN) {
4146    if ((actp->sa_flags & SA_NODEFER) == 0) {
4147      // automaticlly block the signal
4148      sigaddset(&(actp->sa_mask), sig);
4149    }
4150
4151    sa_handler_t hand;
4152    sa_sigaction_t sa;
4153    bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4154    // retrieve the chained handler
4155    if (siginfo_flag_set) {
4156      sa = actp->sa_sigaction;
4157    } else {
4158      hand = actp->sa_handler;
4159    }
4160
4161    if ((actp->sa_flags & SA_RESETHAND) != 0) {
4162      actp->sa_handler = SIG_DFL;
4163    }
4164
4165    // try to honor the signal mask
4166    sigset_t oset;
4167    thr_sigsetmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4168
4169    // call into the chained handler
4170    if (siginfo_flag_set) {
4171      (*sa)(sig, siginfo, context);
4172    } else {
4173      (*hand)(sig);
4174    }
4175
4176    // restore the signal mask
4177    thr_sigsetmask(SIG_SETMASK, &oset, 0);
4178  }
4179  // Tell jvm's signal handler the signal is taken care of.
4180  return true;
4181}
4182
4183bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4184  bool chained = false;
4185  // signal-chaining
4186  if (UseSignalChaining) {
4187    struct sigaction *actp = get_chained_signal_action(sig);
4188    if (actp != NULL) {
4189      chained = call_chained_handler(actp, sig, siginfo, context);
4190    }
4191  }
4192  return chained;
4193}
4194
4195struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
4196  assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4197  if (preinstalled_sigs[sig] != 0) {
4198    return &chainedsigactions[sig];
4199  }
4200  return NULL;
4201}
4202
4203void os::Solaris::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4204
4205  assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
4206  assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4207  chainedsigactions[sig] = oldAct;
4208  preinstalled_sigs[sig] = 1;
4209}
4210
4211void os::Solaris::set_signal_handler(int sig, bool set_installed, bool oktochain) {
4212  // Check for overwrite.
4213  struct sigaction oldAct;
4214  sigaction(sig, (struct sigaction*)NULL, &oldAct);
4215  void* oldhand = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4216                                      : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4217  if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4218      oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4219      oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
4220    if (AllowUserSignalHandlers || !set_installed) {
4221      // Do not overwrite; user takes responsibility to forward to us.
4222      return;
4223    } else if (UseSignalChaining) {
4224      if (oktochain) {
4225        // save the old handler in jvm
4226        save_preinstalled_handler(sig, oldAct);
4227      } else {
4228        vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal, try -XX:+UseAltSigs.");
4229      }
4230      // libjsig also interposes the sigaction() call below and saves the
4231      // old sigaction on it own.
4232    } else {
4233      fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig);
4234    }
4235  }
4236
4237  struct sigaction sigAct;
4238  sigfillset(&(sigAct.sa_mask));
4239  sigAct.sa_handler = SIG_DFL;
4240
4241  sigAct.sa_sigaction = signalHandler;
4242  // Handle SIGSEGV on alternate signal stack if
4243  // not using stack banging
4244  if (!UseStackBanging && sig == SIGSEGV) {
4245    sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
4246  // Interruptible i/o requires SA_RESTART cleared so EINTR
4247  // is returned instead of restarting system calls
4248  } else if (sig == os::Solaris::SIGinterrupt()) {
4249    sigemptyset(&sigAct.sa_mask);
4250    sigAct.sa_handler = NULL;
4251    sigAct.sa_flags = SA_SIGINFO;
4252    sigAct.sa_sigaction = sigINTRHandler;
4253  } else {
4254    sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
4255  }
4256  os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
4257
4258  sigaction(sig, &sigAct, &oldAct);
4259
4260  void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4261                                       : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4262  assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4263}
4264
4265
4266#define DO_SIGNAL_CHECK(sig) \
4267  if (!sigismember(&check_signal_done, sig)) \
4268    os::Solaris::check_signal_handler(sig)
4269
4270// This method is a periodic task to check for misbehaving JNI applications
4271// under CheckJNI, we can add any periodic checks here
4272
4273void os::run_periodic_checks() {
4274  // A big source of grief is hijacking virt. addr 0x0 on Solaris,
4275  // thereby preventing a NULL checks.
4276  if(!check_addr0_done) check_addr0_done = check_addr0(tty);
4277
4278  if (check_signals == false) return;
4279
4280  // SEGV and BUS if overridden could potentially prevent
4281  // generation of hs*.log in the event of a crash, debugging
4282  // such a case can be very challenging, so we absolutely
4283  // check for the following for a good measure:
4284  DO_SIGNAL_CHECK(SIGSEGV);
4285  DO_SIGNAL_CHECK(SIGILL);
4286  DO_SIGNAL_CHECK(SIGFPE);
4287  DO_SIGNAL_CHECK(SIGBUS);
4288  DO_SIGNAL_CHECK(SIGPIPE);
4289  DO_SIGNAL_CHECK(SIGXFSZ);
4290
4291  // ReduceSignalUsage allows the user to override these handlers
4292  // see comments at the very top and jvm_solaris.h
4293  if (!ReduceSignalUsage) {
4294    DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4295    DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4296    DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4297    DO_SIGNAL_CHECK(BREAK_SIGNAL);
4298  }
4299
4300  // See comments above for using JVM1/JVM2 and UseAltSigs
4301  DO_SIGNAL_CHECK(os::Solaris::SIGinterrupt());
4302  DO_SIGNAL_CHECK(os::Solaris::SIGasync());
4303
4304}
4305
4306typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4307
4308static os_sigaction_t os_sigaction = NULL;
4309
4310void os::Solaris::check_signal_handler(int sig) {
4311  char buf[O_BUFLEN];
4312  address jvmHandler = NULL;
4313
4314  struct sigaction act;
4315  if (os_sigaction == NULL) {
4316    // only trust the default sigaction, in case it has been interposed
4317    os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4318    if (os_sigaction == NULL) return;
4319  }
4320
4321  os_sigaction(sig, (struct sigaction*)NULL, &act);
4322
4323  address thisHandler = (act.sa_flags & SA_SIGINFO)
4324    ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4325    : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4326
4327
4328  switch(sig) {
4329    case SIGSEGV:
4330    case SIGBUS:
4331    case SIGFPE:
4332    case SIGPIPE:
4333    case SIGXFSZ:
4334    case SIGILL:
4335      jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4336      break;
4337
4338    case SHUTDOWN1_SIGNAL:
4339    case SHUTDOWN2_SIGNAL:
4340    case SHUTDOWN3_SIGNAL:
4341    case BREAK_SIGNAL:
4342      jvmHandler = (address)user_handler();
4343      break;
4344
4345    default:
4346      int intrsig = os::Solaris::SIGinterrupt();
4347      int asynsig = os::Solaris::SIGasync();
4348
4349      if (sig == intrsig) {
4350        jvmHandler = CAST_FROM_FN_PTR(address, sigINTRHandler);
4351      } else if (sig == asynsig) {
4352        jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4353      } else {
4354        return;
4355      }
4356      break;
4357  }
4358
4359
4360  if (thisHandler != jvmHandler) {
4361    tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4362    tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4363    tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4364    // No need to check this sig any longer
4365    sigaddset(&check_signal_done, sig);
4366  } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4367    tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4368    tty->print("expected:" PTR32_FORMAT, os::Solaris::get_our_sigflags(sig));
4369    tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4370    // No need to check this sig any longer
4371    sigaddset(&check_signal_done, sig);
4372  }
4373
4374  // Print all the signal handler state
4375  if (sigismember(&check_signal_done, sig)) {
4376    print_signal_handlers(tty, buf, O_BUFLEN);
4377  }
4378
4379}
4380
4381void os::Solaris::install_signal_handlers() {
4382  bool libjsigdone = false;
4383  signal_handlers_are_installed = true;
4384
4385  // signal-chaining
4386  typedef void (*signal_setting_t)();
4387  signal_setting_t begin_signal_setting = NULL;
4388  signal_setting_t end_signal_setting = NULL;
4389  begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4390                                        dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4391  if (begin_signal_setting != NULL) {
4392    end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4393                                        dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4394    get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4395                                       dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4396    get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4397                                         dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4398    libjsig_is_loaded = true;
4399    if (os::Solaris::get_libjsig_version != NULL) {
4400      libjsigversion =  (*os::Solaris::get_libjsig_version)();
4401    }
4402    assert(UseSignalChaining, "should enable signal-chaining");
4403  }
4404  if (libjsig_is_loaded) {
4405    // Tell libjsig jvm is setting signal handlers
4406    (*begin_signal_setting)();
4407  }
4408
4409  set_signal_handler(SIGSEGV, true, true);
4410  set_signal_handler(SIGPIPE, true, true);
4411  set_signal_handler(SIGXFSZ, true, true);
4412  set_signal_handler(SIGBUS, true, true);
4413  set_signal_handler(SIGILL, true, true);
4414  set_signal_handler(SIGFPE, true, true);
4415
4416
4417  if (os::Solaris::SIGinterrupt() > OLDMAXSIGNUM || os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4418
4419    // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4420    // can not register overridable signals which might be > 32
4421    if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4422    // Tell libjsig jvm has finished setting signal handlers
4423      (*end_signal_setting)();
4424      libjsigdone = true;
4425    }
4426  }
4427
4428  // Never ok to chain our SIGinterrupt
4429  set_signal_handler(os::Solaris::SIGinterrupt(), true, false);
4430  set_signal_handler(os::Solaris::SIGasync(), true, true);
4431
4432  if (libjsig_is_loaded && !libjsigdone) {
4433    // Tell libjsig jvm finishes setting signal handlers
4434    (*end_signal_setting)();
4435  }
4436
4437  // We don't activate signal checker if libjsig is in place, we trust ourselves
4438  // and if UserSignalHandler is installed all bets are off
4439  if (CheckJNICalls) {
4440    if (libjsig_is_loaded) {
4441      tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4442      check_signals = false;
4443    }
4444    if (AllowUserSignalHandlers) {
4445      tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4446      check_signals = false;
4447    }
4448  }
4449}
4450
4451
4452void report_error(const char* file_name, int line_no, const char* title, const char* format, ...);
4453
4454const char * signames[] = {
4455  "SIG0",
4456  "SIGHUP", "SIGINT", "SIGQUIT", "SIGILL", "SIGTRAP",
4457  "SIGABRT", "SIGEMT", "SIGFPE", "SIGKILL", "SIGBUS",
4458  "SIGSEGV", "SIGSYS", "SIGPIPE", "SIGALRM", "SIGTERM",
4459  "SIGUSR1", "SIGUSR2", "SIGCLD", "SIGPWR", "SIGWINCH",
4460  "SIGURG", "SIGPOLL", "SIGSTOP", "SIGTSTP", "SIGCONT",
4461  "SIGTTIN", "SIGTTOU", "SIGVTALRM", "SIGPROF", "SIGXCPU",
4462  "SIGXFSZ", "SIGWAITING", "SIGLWP", "SIGFREEZE", "SIGTHAW",
4463  "SIGCANCEL", "SIGLOST"
4464};
4465
4466const char* os::exception_name(int exception_code, char* buf, size_t size) {
4467  if (0 < exception_code && exception_code <= SIGRTMAX) {
4468    // signal
4469    if (exception_code < sizeof(signames)/sizeof(const char*)) {
4470       jio_snprintf(buf, size, "%s", signames[exception_code]);
4471    } else {
4472       jio_snprintf(buf, size, "SIG%d", exception_code);
4473    }
4474    return buf;
4475  } else {
4476    return NULL;
4477  }
4478}
4479
4480// (Static) wrappers for the new libthread API
4481int_fnP_thread_t_iP_uP_stack_tP_gregset_t os::Solaris::_thr_getstate;
4482int_fnP_thread_t_i_gregset_t os::Solaris::_thr_setstate;
4483int_fnP_thread_t_i os::Solaris::_thr_setmutator;
4484int_fnP_thread_t os::Solaris::_thr_suspend_mutator;
4485int_fnP_thread_t os::Solaris::_thr_continue_mutator;
4486
4487// (Static) wrapper for getisax(2) call.
4488os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4489
4490// (Static) wrappers for the liblgrp API
4491os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4492os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4493os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4494os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4495os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4496os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4497os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4498os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4499os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4500
4501// (Static) wrapper for meminfo() call.
4502os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4503
4504static address resolve_symbol_lazy(const char* name) {
4505  address addr = (address) dlsym(RTLD_DEFAULT, name);
4506  if(addr == NULL) {
4507    // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4508    addr = (address) dlsym(RTLD_NEXT, name);
4509  }
4510  return addr;
4511}
4512
4513static address resolve_symbol(const char* name) {
4514  address addr = resolve_symbol_lazy(name);
4515  if(addr == NULL) {
4516    fatal(dlerror());
4517  }
4518  return addr;
4519}
4520
4521
4522
4523// isT2_libthread()
4524//
4525// Routine to determine if we are currently using the new T2 libthread.
4526//
4527// We determine if we are using T2 by reading /proc/self/lstatus and
4528// looking for a thread with the ASLWP bit set.  If we find this status
4529// bit set, we must assume that we are NOT using T2.  The T2 team
4530// has approved this algorithm.
4531//
4532// We need to determine if we are running with the new T2 libthread
4533// since setting native thread priorities is handled differently
4534// when using this library.  All threads created using T2 are bound
4535// threads. Calling thr_setprio is meaningless in this case.
4536//
4537bool isT2_libthread() {
4538  static prheader_t * lwpArray = NULL;
4539  static int lwpSize = 0;
4540  static int lwpFile = -1;
4541  lwpstatus_t * that;
4542  char lwpName [128];
4543  bool isT2 = false;
4544
4545#define ADR(x)  ((uintptr_t)(x))
4546#define LWPINDEX(ary,ix)   ((lwpstatus_t *)(((ary)->pr_entsize * (ix)) + (ADR((ary) + 1))))
4547
4548  lwpFile = open("/proc/self/lstatus", O_RDONLY, 0);
4549  if (lwpFile < 0) {
4550      if (ThreadPriorityVerbose) warning ("Couldn't open /proc/self/lstatus\n");
4551      return false;
4552  }
4553  lwpSize = 16*1024;
4554  for (;;) {
4555    lseek (lwpFile, 0, SEEK_SET);
4556    lwpArray = (prheader_t *)NEW_C_HEAP_ARRAY(char, lwpSize);
4557    if (read(lwpFile, lwpArray, lwpSize) < 0) {
4558      if (ThreadPriorityVerbose) warning("Error reading /proc/self/lstatus\n");
4559      break;
4560    }
4561    if ((lwpArray->pr_nent * lwpArray->pr_entsize) <= lwpSize) {
4562       // We got a good snapshot - now iterate over the list.
4563      int aslwpcount = 0;
4564      for (int i = 0; i < lwpArray->pr_nent; i++ ) {
4565        that = LWPINDEX(lwpArray,i);
4566        if (that->pr_flags & PR_ASLWP) {
4567          aslwpcount++;
4568        }
4569      }
4570      if (aslwpcount == 0) isT2 = true;
4571      break;
4572    }
4573    lwpSize = lwpArray->pr_nent * lwpArray->pr_entsize;
4574    FREE_C_HEAP_ARRAY(char, lwpArray);  // retry.
4575  }
4576
4577  FREE_C_HEAP_ARRAY(char, lwpArray);
4578  close (lwpFile);
4579  if (ThreadPriorityVerbose) {
4580    if (isT2) tty->print_cr("We are running with a T2 libthread\n");
4581    else tty->print_cr("We are not running with a T2 libthread\n");
4582  }
4583  return isT2;
4584}
4585
4586
4587void os::Solaris::libthread_init() {
4588  address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4589
4590  // Determine if we are running with the new T2 libthread
4591  os::Solaris::set_T2_libthread(isT2_libthread());
4592
4593  lwp_priocntl_init();
4594
4595  // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4596  if(func == NULL) {
4597    func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4598    // Guarantee that this VM is running on an new enough OS (5.6 or
4599    // later) that it will have a new enough libthread.so.
4600    guarantee(func != NULL, "libthread.so is too old.");
4601  }
4602
4603  // Initialize the new libthread getstate API wrappers
4604  func = resolve_symbol("thr_getstate");
4605  os::Solaris::set_thr_getstate(CAST_TO_FN_PTR(int_fnP_thread_t_iP_uP_stack_tP_gregset_t, func));
4606
4607  func = resolve_symbol("thr_setstate");
4608  os::Solaris::set_thr_setstate(CAST_TO_FN_PTR(int_fnP_thread_t_i_gregset_t, func));
4609
4610  func = resolve_symbol("thr_setmutator");
4611  os::Solaris::set_thr_setmutator(CAST_TO_FN_PTR(int_fnP_thread_t_i, func));
4612
4613  func = resolve_symbol("thr_suspend_mutator");
4614  os::Solaris::set_thr_suspend_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4615
4616  func = resolve_symbol("thr_continue_mutator");
4617  os::Solaris::set_thr_continue_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4618
4619  int size;
4620  void (*handler_info_func)(address *, int *);
4621  handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4622  handler_info_func(&handler_start, &size);
4623  handler_end = handler_start + size;
4624}
4625
4626
4627int_fnP_mutex_tP os::Solaris::_mutex_lock;
4628int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4629int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4630int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4631int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4632int os::Solaris::_mutex_scope = USYNC_THREAD;
4633
4634int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4635int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4636int_fnP_cond_tP os::Solaris::_cond_signal;
4637int_fnP_cond_tP os::Solaris::_cond_broadcast;
4638int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4639int_fnP_cond_tP os::Solaris::_cond_destroy;
4640int os::Solaris::_cond_scope = USYNC_THREAD;
4641
4642void os::Solaris::synchronization_init() {
4643  if(UseLWPSynchronization) {
4644    os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4645    os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4646    os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4647    os::Solaris::set_mutex_init(lwp_mutex_init);
4648    os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4649    os::Solaris::set_mutex_scope(USYNC_THREAD);
4650
4651    os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4652    os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4653    os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4654    os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4655    os::Solaris::set_cond_init(lwp_cond_init);
4656    os::Solaris::set_cond_destroy(lwp_cond_destroy);
4657    os::Solaris::set_cond_scope(USYNC_THREAD);
4658  }
4659  else {
4660    os::Solaris::set_mutex_scope(USYNC_THREAD);
4661    os::Solaris::set_cond_scope(USYNC_THREAD);
4662
4663    if(UsePthreads) {
4664      os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4665      os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4666      os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4667      os::Solaris::set_mutex_init(pthread_mutex_default_init);
4668      os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4669
4670      os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4671      os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4672      os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4673      os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4674      os::Solaris::set_cond_init(pthread_cond_default_init);
4675      os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4676    }
4677    else {
4678      os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4679      os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4680      os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4681      os::Solaris::set_mutex_init(::mutex_init);
4682      os::Solaris::set_mutex_destroy(::mutex_destroy);
4683
4684      os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4685      os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4686      os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4687      os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4688      os::Solaris::set_cond_init(::cond_init);
4689      os::Solaris::set_cond_destroy(::cond_destroy);
4690    }
4691  }
4692}
4693
4694bool os::Solaris::liblgrp_init() {
4695  void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4696  if (handle != NULL) {
4697    os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4698    os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4699    os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4700    os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4701    os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4702    os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4703    os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4704    os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4705                                       dlsym(handle, "lgrp_cookie_stale")));
4706
4707    lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4708    set_lgrp_cookie(c);
4709    return true;
4710  }
4711  return false;
4712}
4713
4714void os::Solaris::misc_sym_init() {
4715  address func;
4716
4717  // getisax
4718  func = resolve_symbol_lazy("getisax");
4719  if (func != NULL) {
4720    os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
4721  }
4722
4723  // meminfo
4724  func = resolve_symbol_lazy("meminfo");
4725  if (func != NULL) {
4726    os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
4727  }
4728}
4729
4730uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
4731  assert(_getisax != NULL, "_getisax not set");
4732  return _getisax(array, n);
4733}
4734
4735// Symbol doesn't exist in Solaris 8 pset.h
4736#ifndef PS_MYID
4737#define PS_MYID -3
4738#endif
4739
4740// int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4741typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4742static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4743
4744void init_pset_getloadavg_ptr(void) {
4745  pset_getloadavg_ptr =
4746    (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4747  if (PrintMiscellaneous && Verbose && pset_getloadavg_ptr == NULL) {
4748    warning("pset_getloadavg function not found");
4749  }
4750}
4751
4752int os::Solaris::_dev_zero_fd = -1;
4753
4754// this is called _before_ the global arguments have been parsed
4755void os::init(void) {
4756  _initial_pid = getpid();
4757
4758  max_hrtime = first_hrtime = gethrtime();
4759
4760  init_random(1234567);
4761
4762  page_size = sysconf(_SC_PAGESIZE);
4763  if (page_size == -1)
4764    fatal1("os_solaris.cpp: os::init: sysconf failed (%s)", strerror(errno));
4765  init_page_sizes((size_t) page_size);
4766
4767  Solaris::initialize_system_info();
4768
4769  // Initialize misc. symbols as soon as possible, so we can use them
4770  // if we need them.
4771  Solaris::misc_sym_init();
4772
4773  int fd = open("/dev/zero", O_RDWR);
4774  if (fd < 0) {
4775    fatal1("os::init: cannot open /dev/zero (%s)", strerror(errno));
4776  } else {
4777    Solaris::set_dev_zero_fd(fd);
4778
4779    // Close on exec, child won't inherit.
4780    fcntl(fd, F_SETFD, FD_CLOEXEC);
4781  }
4782
4783  clock_tics_per_sec = CLK_TCK;
4784
4785  // check if dladdr1() exists; dladdr1 can provide more information than
4786  // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4787  // and is available on linker patches for 5.7 and 5.8.
4788  // libdl.so must have been loaded, this call is just an entry lookup
4789  void * hdl = dlopen("libdl.so", RTLD_NOW);
4790  if (hdl)
4791    dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4792
4793  // (Solaris only) this switches to calls that actually do locking.
4794  ThreadCritical::initialize();
4795
4796  main_thread = thr_self();
4797
4798  // Constant minimum stack size allowed. It must be at least
4799  // the minimum of what the OS supports (thr_min_stack()), and
4800  // enough to allow the thread to get to user bytecode execution.
4801  Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
4802  // If the pagesize of the VM is greater than 8K determine the appropriate
4803  // number of initial guard pages.  The user can change this with the
4804  // command line arguments, if needed.
4805  if (vm_page_size() > 8*K) {
4806    StackYellowPages = 1;
4807    StackRedPages = 1;
4808    StackShadowPages = round_to((StackShadowPages*8*K), vm_page_size()) / vm_page_size();
4809  }
4810}
4811
4812// To install functions for atexit system call
4813extern "C" {
4814  static void perfMemory_exit_helper() {
4815    perfMemory_exit();
4816  }
4817}
4818
4819// this is called _after_ the global arguments have been parsed
4820jint os::init_2(void) {
4821  // try to enable extended file IO ASAP, see 6431278
4822  os::Solaris::try_enable_extended_io();
4823
4824  // Allocate a single page and mark it as readable for safepoint polling.  Also
4825  // use this first mmap call to check support for MAP_ALIGN.
4826  address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
4827                                                      page_size,
4828                                                      MAP_PRIVATE | MAP_ALIGN,
4829                                                      PROT_READ);
4830  if (polling_page == NULL) {
4831    has_map_align = false;
4832    polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
4833                                                PROT_READ);
4834  }
4835
4836  os::set_polling_page(polling_page);
4837
4838#ifndef PRODUCT
4839  if( Verbose && PrintMiscellaneous )
4840    tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4841#endif
4842
4843  if (!UseMembar) {
4844    address mem_serialize_page = (address)Solaris::mmap_chunk( NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE );
4845    guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4846    os::set_memory_serialize_page( mem_serialize_page );
4847
4848#ifndef PRODUCT
4849    if(Verbose && PrintMiscellaneous)
4850      tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4851#endif
4852}
4853
4854  FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
4855
4856  // Check minimum allowable stack size for thread creation and to initialize
4857  // the java system classes, including StackOverflowError - depends on page
4858  // size.  Add a page for compiler2 recursion in main thread.
4859  // Add in BytesPerWord times page size to account for VM stack during
4860  // class initialization depending on 32 or 64 bit VM.
4861  guarantee((Solaris::min_stack_allowed >=
4862    (StackYellowPages+StackRedPages+StackShadowPages+BytesPerWord
4863     COMPILER2_PRESENT(+1)) * page_size),
4864    "need to increase Solaris::min_stack_allowed on this platform");
4865
4866  size_t threadStackSizeInBytes = ThreadStackSize * K;
4867  if (threadStackSizeInBytes != 0 &&
4868    threadStackSizeInBytes < Solaris::min_stack_allowed) {
4869    tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
4870                  Solaris::min_stack_allowed/K);
4871    return JNI_ERR;
4872  }
4873
4874  // For 64kbps there will be a 64kb page size, which makes
4875  // the usable default stack size quite a bit less.  Increase the
4876  // stack for 64kb (or any > than 8kb) pages, this increases
4877  // virtual memory fragmentation (since we're not creating the
4878  // stack on a power of 2 boundary.  The real fix for this
4879  // should be to fix the guard page mechanism.
4880
4881  if (vm_page_size() > 8*K) {
4882      threadStackSizeInBytes = (threadStackSizeInBytes != 0)
4883         ? threadStackSizeInBytes +
4884           ((StackYellowPages + StackRedPages) * vm_page_size())
4885         : 0;
4886      ThreadStackSize = threadStackSizeInBytes/K;
4887  }
4888
4889  // Make the stack size a multiple of the page size so that
4890  // the yellow/red zones can be guarded.
4891  JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4892        vm_page_size()));
4893
4894  Solaris::libthread_init();
4895
4896  if (UseNUMA) {
4897    if (!Solaris::liblgrp_init()) {
4898      UseNUMA = false;
4899    } else {
4900      size_t lgrp_limit = os::numa_get_groups_num();
4901      int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit);
4902      size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4903      FREE_C_HEAP_ARRAY(int, lgrp_ids);
4904      if (lgrp_num < 2) {
4905        // There's only one locality group, disable NUMA.
4906        UseNUMA = false;
4907      }
4908    }
4909    if (!UseNUMA && ForceNUMA) {
4910      UseNUMA = true;
4911    }
4912  }
4913
4914  Solaris::signal_sets_init();
4915  Solaris::init_signal_mem();
4916  Solaris::install_signal_handlers();
4917
4918  if (libjsigversion < JSIG_VERSION_1_4_1) {
4919    Maxlibjsigsigs = OLDMAXSIGNUM;
4920  }
4921
4922  // initialize synchronization primitives to use either thread or
4923  // lwp synchronization (controlled by UseLWPSynchronization)
4924  Solaris::synchronization_init();
4925
4926  if (MaxFDLimit) {
4927    // set the number of file descriptors to max. print out error
4928    // if getrlimit/setrlimit fails but continue regardless.
4929    struct rlimit nbr_files;
4930    int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4931    if (status != 0) {
4932      if (PrintMiscellaneous && (Verbose || WizardMode))
4933        perror("os::init_2 getrlimit failed");
4934    } else {
4935      nbr_files.rlim_cur = nbr_files.rlim_max;
4936      status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4937      if (status != 0) {
4938        if (PrintMiscellaneous && (Verbose || WizardMode))
4939          perror("os::init_2 setrlimit failed");
4940      }
4941    }
4942  }
4943
4944  // Initialize HPI.
4945  jint hpi_result = hpi::initialize();
4946  if (hpi_result != JNI_OK) {
4947    tty->print_cr("There was an error trying to initialize the HPI library.");
4948    return hpi_result;
4949  }
4950
4951  // Calculate theoretical max. size of Threads to guard gainst
4952  // artifical out-of-memory situations, where all available address-
4953  // space has been reserved by thread stacks. Default stack size is 1Mb.
4954  size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4955    JavaThread::stack_size_at_create() : (1*K*K);
4956  assert(pre_thread_stack_size != 0, "Must have a stack");
4957  // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4958  // we should start doing Virtual Memory banging. Currently when the threads will
4959  // have used all but 200Mb of space.
4960  size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4961  Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4962
4963  // at-exit methods are called in the reverse order of their registration.
4964  // In Solaris 7 and earlier, atexit functions are called on return from
4965  // main or as a result of a call to exit(3C). There can be only 32 of
4966  // these functions registered and atexit() does not set errno. In Solaris
4967  // 8 and later, there is no limit to the number of functions registered
4968  // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4969  // functions are called upon dlclose(3DL) in addition to return from main
4970  // and exit(3C).
4971
4972  if (PerfAllowAtExitRegistration) {
4973    // only register atexit functions if PerfAllowAtExitRegistration is set.
4974    // atexit functions can be delayed until process exit time, which
4975    // can be problematic for embedded VM situations. Embedded VMs should
4976    // call DestroyJavaVM() to assure that VM resources are released.
4977
4978    // note: perfMemory_exit_helper atexit function may be removed in
4979    // the future if the appropriate cleanup code can be added to the
4980    // VM_Exit VMOperation's doit method.
4981    if (atexit(perfMemory_exit_helper) != 0) {
4982      warning("os::init2 atexit(perfMemory_exit_helper) failed");
4983    }
4984  }
4985
4986  // Init pset_loadavg function pointer
4987  init_pset_getloadavg_ptr();
4988
4989  return JNI_OK;
4990}
4991
4992
4993// Mark the polling page as unreadable
4994void os::make_polling_page_unreadable(void) {
4995  if( mprotect((char *)_polling_page, page_size, PROT_NONE) != 0 )
4996    fatal("Could not disable polling page");
4997};
4998
4999// Mark the polling page as readable
5000void os::make_polling_page_readable(void) {
5001  if( mprotect((char *)_polling_page, page_size, PROT_READ) != 0 )
5002    fatal("Could not enable polling page");
5003};
5004
5005// OS interface.
5006
5007int os::stat(const char *path, struct stat *sbuf) {
5008  char pathbuf[MAX_PATH];
5009  if (strlen(path) > MAX_PATH - 1) {
5010    errno = ENAMETOOLONG;
5011    return -1;
5012  }
5013  hpi::native_path(strcpy(pathbuf, path));
5014  return ::stat(pathbuf, sbuf);
5015}
5016
5017
5018bool os::check_heap(bool force) { return true; }
5019
5020typedef int (*vsnprintf_t)(char* buf, size_t count, const char* fmt, va_list argptr);
5021static vsnprintf_t sol_vsnprintf = NULL;
5022
5023int local_vsnprintf(char* buf, size_t count, const char* fmt, va_list argptr) {
5024  if (!sol_vsnprintf) {
5025    //search  for the named symbol in the objects that were loaded after libjvm
5026    void* where = RTLD_NEXT;
5027    if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5028        sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5029    if (!sol_vsnprintf){
5030      //search  for the named symbol in the objects that were loaded before libjvm
5031      where = RTLD_DEFAULT;
5032      if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5033        sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5034      assert(sol_vsnprintf != NULL, "vsnprintf not found");
5035    }
5036  }
5037  return (*sol_vsnprintf)(buf, count, fmt, argptr);
5038}
5039
5040
5041// Is a (classpath) directory empty?
5042bool os::dir_is_empty(const char* path) {
5043  DIR *dir = NULL;
5044  struct dirent *ptr;
5045
5046  dir = opendir(path);
5047  if (dir == NULL) return true;
5048
5049  /* Scan the directory */
5050  bool result = true;
5051  char buf[sizeof(struct dirent) + MAX_PATH];
5052  struct dirent *dbuf = (struct dirent *) buf;
5053  while (result && (ptr = readdir(dir, dbuf)) != NULL) {
5054    if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5055      result = false;
5056    }
5057  }
5058  closedir(dir);
5059  return result;
5060}
5061
5062// create binary file, rewriting existing file if required
5063int os::create_binary_file(const char* path, bool rewrite_existing) {
5064  int oflags = O_WRONLY | O_CREAT;
5065  if (!rewrite_existing) {
5066    oflags |= O_EXCL;
5067  }
5068  return ::open64(path, oflags, S_IREAD | S_IWRITE);
5069}
5070
5071// return current position of file pointer
5072jlong os::current_file_offset(int fd) {
5073  return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5074}
5075
5076// move file pointer to the specified offset
5077jlong os::seek_to_file_offset(int fd, jlong offset) {
5078  return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5079}
5080
5081// Map a block of memory.
5082char* os::map_memory(int fd, const char* file_name, size_t file_offset,
5083                     char *addr, size_t bytes, bool read_only,
5084                     bool allow_exec) {
5085  int prot;
5086  int flags;
5087
5088  if (read_only) {
5089    prot = PROT_READ;
5090    flags = MAP_SHARED;
5091  } else {
5092    prot = PROT_READ | PROT_WRITE;
5093    flags = MAP_PRIVATE;
5094  }
5095
5096  if (allow_exec) {
5097    prot |= PROT_EXEC;
5098  }
5099
5100  if (addr != NULL) {
5101    flags |= MAP_FIXED;
5102  }
5103
5104  char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5105                                     fd, file_offset);
5106  if (mapped_address == MAP_FAILED) {
5107    return NULL;
5108  }
5109  return mapped_address;
5110}
5111
5112
5113// Remap a block of memory.
5114char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
5115                       char *addr, size_t bytes, bool read_only,
5116                       bool allow_exec) {
5117  // same as map_memory() on this OS
5118  return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5119                        allow_exec);
5120}
5121
5122
5123// Unmap a block of memory.
5124bool os::unmap_memory(char* addr, size_t bytes) {
5125  return munmap(addr, bytes) == 0;
5126}
5127
5128void os::pause() {
5129  char filename[MAX_PATH];
5130  if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5131    jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5132  } else {
5133    jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5134  }
5135
5136  int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5137  if (fd != -1) {
5138    struct stat buf;
5139    close(fd);
5140    while (::stat(filename, &buf) == 0) {
5141      (void)::poll(NULL, 0, 100);
5142    }
5143  } else {
5144    jio_fprintf(stderr,
5145      "Could not open pause file '%s', continuing immediately.\n", filename);
5146  }
5147}
5148
5149#ifndef PRODUCT
5150#ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5151// Turn this on if you need to trace synch operations.
5152// Set RECORD_SYNCH_LIMIT to a large-enough value,
5153// and call record_synch_enable and record_synch_disable
5154// around the computation of interest.
5155
5156void record_synch(char* name, bool returning);  // defined below
5157
5158class RecordSynch {
5159  char* _name;
5160 public:
5161  RecordSynch(char* name) :_name(name)
5162                 { record_synch(_name, false); }
5163  ~RecordSynch() { record_synch(_name,   true);  }
5164};
5165
5166#define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
5167extern "C" ret name params {                                    \
5168  typedef ret name##_t params;                                  \
5169  static name##_t* implem = NULL;                               \
5170  static int callcount = 0;                                     \
5171  if (implem == NULL) {                                         \
5172    implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
5173    if (implem == NULL)  fatal(dlerror());                      \
5174  }                                                             \
5175  ++callcount;                                                  \
5176  RecordSynch _rs(#name);                                       \
5177  inner;                                                        \
5178  return implem args;                                           \
5179}
5180// in dbx, examine callcounts this way:
5181// for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
5182
5183#define CHECK_POINTER_OK(p) \
5184  (Universe::perm_gen() == NULL || !Universe::is_reserved_heap((oop)(p)))
5185#define CHECK_MU \
5186  if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
5187#define CHECK_CV \
5188  if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
5189#define CHECK_P(p) \
5190  if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
5191
5192#define CHECK_MUTEX(mutex_op) \
5193CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
5194
5195CHECK_MUTEX(   mutex_lock)
5196CHECK_MUTEX(  _mutex_lock)
5197CHECK_MUTEX( mutex_unlock)
5198CHECK_MUTEX(_mutex_unlock)
5199CHECK_MUTEX( mutex_trylock)
5200CHECK_MUTEX(_mutex_trylock)
5201
5202#define CHECK_COND(cond_op) \
5203CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU;CHECK_CV);
5204
5205CHECK_COND( cond_wait);
5206CHECK_COND(_cond_wait);
5207CHECK_COND(_cond_wait_cancel);
5208
5209#define CHECK_COND2(cond_op) \
5210CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU;CHECK_CV);
5211
5212CHECK_COND2( cond_timedwait);
5213CHECK_COND2(_cond_timedwait);
5214CHECK_COND2(_cond_timedwait_cancel);
5215
5216// do the _lwp_* versions too
5217#define mutex_t lwp_mutex_t
5218#define cond_t  lwp_cond_t
5219CHECK_MUTEX(  _lwp_mutex_lock)
5220CHECK_MUTEX(  _lwp_mutex_unlock)
5221CHECK_MUTEX(  _lwp_mutex_trylock)
5222CHECK_MUTEX( __lwp_mutex_lock)
5223CHECK_MUTEX( __lwp_mutex_unlock)
5224CHECK_MUTEX( __lwp_mutex_trylock)
5225CHECK_MUTEX(___lwp_mutex_lock)
5226CHECK_MUTEX(___lwp_mutex_unlock)
5227
5228CHECK_COND(  _lwp_cond_wait);
5229CHECK_COND( __lwp_cond_wait);
5230CHECK_COND(___lwp_cond_wait);
5231
5232CHECK_COND2(  _lwp_cond_timedwait);
5233CHECK_COND2( __lwp_cond_timedwait);
5234#undef mutex_t
5235#undef cond_t
5236
5237CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5238CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5239CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5240CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5241CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5242CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5243CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5244CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5245
5246
5247// recording machinery:
5248
5249enum { RECORD_SYNCH_LIMIT = 200 };
5250char* record_synch_name[RECORD_SYNCH_LIMIT];
5251void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
5252bool record_synch_returning[RECORD_SYNCH_LIMIT];
5253thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
5254int record_synch_count = 0;
5255bool record_synch_enabled = false;
5256
5257// in dbx, examine recorded data this way:
5258// for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
5259
5260void record_synch(char* name, bool returning) {
5261  if (record_synch_enabled) {
5262    if (record_synch_count < RECORD_SYNCH_LIMIT) {
5263      record_synch_name[record_synch_count] = name;
5264      record_synch_returning[record_synch_count] = returning;
5265      record_synch_thread[record_synch_count] = thr_self();
5266      record_synch_arg0ptr[record_synch_count] = &name;
5267      record_synch_count++;
5268    }
5269    // put more checking code here:
5270    // ...
5271  }
5272}
5273
5274void record_synch_enable() {
5275  // start collecting trace data, if not already doing so
5276  if (!record_synch_enabled)  record_synch_count = 0;
5277  record_synch_enabled = true;
5278}
5279
5280void record_synch_disable() {
5281  // stop collecting trace data
5282  record_synch_enabled = false;
5283}
5284
5285#endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5286#endif // PRODUCT
5287
5288const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5289const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
5290                               (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5291
5292
5293// JVMTI & JVM monitoring and management support
5294// The thread_cpu_time() and current_thread_cpu_time() are only
5295// supported if is_thread_cpu_time_supported() returns true.
5296// They are not supported on Solaris T1.
5297
5298// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5299// are used by JVM M&M and JVMTI to get user+sys or user CPU time
5300// of a thread.
5301//
5302// current_thread_cpu_time() and thread_cpu_time(Thread *)
5303// returns the fast estimate available on the platform.
5304
5305// hrtime_t gethrvtime() return value includes
5306// user time but does not include system time
5307jlong os::current_thread_cpu_time() {
5308  return (jlong) gethrvtime();
5309}
5310
5311jlong os::thread_cpu_time(Thread *thread) {
5312  // return user level CPU time only to be consistent with
5313  // what current_thread_cpu_time returns.
5314  // thread_cpu_time_info() must be changed if this changes
5315  return os::thread_cpu_time(thread, false /* user time only */);
5316}
5317
5318jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5319  if (user_sys_cpu_time) {
5320    return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5321  } else {
5322    return os::current_thread_cpu_time();
5323  }
5324}
5325
5326jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5327  char proc_name[64];
5328  int count;
5329  prusage_t prusage;
5330  jlong lwp_time;
5331  int fd;
5332
5333  sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5334                     getpid(),
5335                     thread->osthread()->lwp_id());
5336  fd = open(proc_name, O_RDONLY);
5337  if ( fd == -1 ) return -1;
5338
5339  do {
5340    count = pread(fd,
5341                  (void *)&prusage.pr_utime,
5342                  thr_time_size,
5343                  thr_time_off);
5344  } while (count < 0 && errno == EINTR);
5345  close(fd);
5346  if ( count < 0 ) return -1;
5347
5348  if (user_sys_cpu_time) {
5349    // user + system CPU time
5350    lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5351                 (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5352                 (jlong)prusage.pr_stime.tv_nsec +
5353                 (jlong)prusage.pr_utime.tv_nsec;
5354  } else {
5355    // user level CPU time only
5356    lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5357                (jlong)prusage.pr_utime.tv_nsec;
5358  }
5359
5360  return(lwp_time);
5361}
5362
5363void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5364  info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5365  info_ptr->may_skip_backward = false;    // elapsed time not wall time
5366  info_ptr->may_skip_forward = false;     // elapsed time not wall time
5367  info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5368}
5369
5370void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5371  info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5372  info_ptr->may_skip_backward = false;    // elapsed time not wall time
5373  info_ptr->may_skip_forward = false;     // elapsed time not wall time
5374  info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5375}
5376
5377bool os::is_thread_cpu_time_supported() {
5378  if ( os::Solaris::T2_libthread() || UseBoundThreads ) {
5379    return true;
5380  } else {
5381    return false;
5382  }
5383}
5384
5385// System loadavg support.  Returns -1 if load average cannot be obtained.
5386// Return the load average for our processor set if the primitive exists
5387// (Solaris 9 and later).  Otherwise just return system wide loadavg.
5388int os::loadavg(double loadavg[], int nelem) {
5389  if (pset_getloadavg_ptr != NULL) {
5390    return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5391  } else {
5392    return ::getloadavg(loadavg, nelem);
5393  }
5394}
5395
5396//---------------------------------------------------------------------------------
5397#ifndef PRODUCT
5398
5399static address same_page(address x, address y) {
5400  intptr_t page_bits = -os::vm_page_size();
5401  if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
5402    return x;
5403  else if (x > y)
5404    return (address)(intptr_t(y) | ~page_bits) + 1;
5405  else
5406    return (address)(intptr_t(y) & page_bits);
5407}
5408
5409bool os::find(address addr) {
5410  Dl_info dlinfo;
5411  memset(&dlinfo, 0, sizeof(dlinfo));
5412  if (dladdr(addr, &dlinfo)) {
5413#ifdef _LP64
5414    tty->print("0x%016lx: ", addr);
5415#else
5416    tty->print("0x%08x: ", addr);
5417#endif
5418    if (dlinfo.dli_sname != NULL)
5419      tty->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5420    else if (dlinfo.dli_fname)
5421      tty->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5422    else
5423      tty->print("<absolute address>");
5424    if (dlinfo.dli_fname)  tty->print(" in %s", dlinfo.dli_fname);
5425#ifdef _LP64
5426    if (dlinfo.dli_fbase)  tty->print(" at 0x%016lx", dlinfo.dli_fbase);
5427#else
5428    if (dlinfo.dli_fbase)  tty->print(" at 0x%08x", dlinfo.dli_fbase);
5429#endif
5430    tty->cr();
5431
5432    if (Verbose) {
5433      // decode some bytes around the PC
5434      address begin = same_page(addr-40, addr);
5435      address end   = same_page(addr+40, addr);
5436      address       lowest = (address) dlinfo.dli_sname;
5437      if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5438      if (begin < lowest)  begin = lowest;
5439      Dl_info dlinfo2;
5440      if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
5441          && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5442        end = (address) dlinfo2.dli_saddr;
5443      Disassembler::decode(begin, end);
5444    }
5445    return true;
5446  }
5447  return false;
5448}
5449
5450#endif
5451
5452
5453// Following function has been added to support HotSparc's libjvm.so running
5454// under Solaris production JDK 1.2.2 / 1.3.0.  These came from
5455// src/solaris/hpi/native_threads in the EVM codebase.
5456//
5457// NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5458// libraries and should thus be removed. We will leave it behind for a while
5459// until we no longer want to able to run on top of 1.3.0 Solaris production
5460// JDK. See 4341971.
5461
5462#define STACK_SLACK 0x800
5463
5464extern "C" {
5465  intptr_t sysThreadAvailableStackWithSlack() {
5466    stack_t st;
5467    intptr_t retval, stack_top;
5468    retval = thr_stksegment(&st);
5469    assert(retval == 0, "incorrect return value from thr_stksegment");
5470    assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5471    assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5472    stack_top=(intptr_t)st.ss_sp-st.ss_size;
5473    return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5474  }
5475}
5476
5477// Just to get the Kernel build to link on solaris for testing.
5478
5479extern "C" {
5480class ASGCT_CallTrace;
5481void AsyncGetCallTrace(ASGCT_CallTrace *trace, jint depth, void* ucontext)
5482  KERNEL_RETURN;
5483}
5484
5485
5486// ObjectMonitor park-unpark infrastructure ...
5487//
5488// We implement Solaris and Linux PlatformEvents with the
5489// obvious condvar-mutex-flag triple.
5490// Another alternative that works quite well is pipes:
5491// Each PlatformEvent consists of a pipe-pair.
5492// The thread associated with the PlatformEvent
5493// calls park(), which reads from the input end of the pipe.
5494// Unpark() writes into the other end of the pipe.
5495// The write-side of the pipe must be set NDELAY.
5496// Unfortunately pipes consume a large # of handles.
5497// Native solaris lwp_park() and lwp_unpark() work nicely, too.
5498// Using pipes for the 1st few threads might be workable, however.
5499//
5500// park() is permitted to return spuriously.
5501// Callers of park() should wrap the call to park() in
5502// an appropriate loop.  A litmus test for the correct
5503// usage of park is the following: if park() were modified
5504// to immediately return 0 your code should still work,
5505// albeit degenerating to a spin loop.
5506//
5507// An interesting optimization for park() is to use a trylock()
5508// to attempt to acquire the mutex.  If the trylock() fails
5509// then we know that a concurrent unpark() operation is in-progress.
5510// in that case the park() code could simply set _count to 0
5511// and return immediately.  The subsequent park() operation *might*
5512// return immediately.  That's harmless as the caller of park() is
5513// expected to loop.  By using trylock() we will have avoided a
5514// avoided a context switch caused by contention on the per-thread mutex.
5515//
5516// TODO-FIXME:
5517// 1.  Reconcile Doug's JSR166 j.u.c park-unpark with the
5518//     objectmonitor implementation.
5519// 2.  Collapse the JSR166 parker event, and the
5520//     objectmonitor ParkEvent into a single "Event" construct.
5521// 3.  In park() and unpark() add:
5522//     assert (Thread::current() == AssociatedWith).
5523// 4.  add spurious wakeup injection on a -XX:EarlyParkReturn=N switch.
5524//     1-out-of-N park() operations will return immediately.
5525//
5526// _Event transitions in park()
5527//   -1 => -1 : illegal
5528//    1 =>  0 : pass - return immediately
5529//    0 => -1 : block
5530//
5531// _Event serves as a restricted-range semaphore.
5532//
5533// Another possible encoding of _Event would be with
5534// explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5535//
5536// TODO-FIXME: add DTRACE probes for:
5537// 1.   Tx parks
5538// 2.   Ty unparks Tx
5539// 3.   Tx resumes from park
5540
5541
5542// value determined through experimentation
5543#define ROUNDINGFIX 11
5544
5545// utility to compute the abstime argument to timedwait.
5546// TODO-FIXME: switch from compute_abstime() to unpackTime().
5547
5548static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5549  // millis is the relative timeout time
5550  // abstime will be the absolute timeout time
5551  if (millis < 0)  millis = 0;
5552  struct timeval now;
5553  int status = gettimeofday(&now, NULL);
5554  assert(status == 0, "gettimeofday");
5555  jlong seconds = millis / 1000;
5556  jlong max_wait_period;
5557
5558  if (UseLWPSynchronization) {
5559    // forward port of fix for 4275818 (not sleeping long enough)
5560    // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5561    // _lwp_cond_timedwait() used a round_down algorithm rather
5562    // than a round_up. For millis less than our roundfactor
5563    // it rounded down to 0 which doesn't meet the spec.
5564    // For millis > roundfactor we may return a bit sooner, but
5565    // since we can not accurately identify the patch level and
5566    // this has already been fixed in Solaris 9 and 8 we will
5567    // leave it alone rather than always rounding down.
5568
5569    if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5570       // It appears that when we go directly through Solaris _lwp_cond_timedwait()
5571           // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
5572           max_wait_period = 21000000;
5573  } else {
5574    max_wait_period = 50000000;
5575  }
5576  millis %= 1000;
5577  if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
5578     seconds = max_wait_period;
5579  }
5580  abstime->tv_sec = now.tv_sec  + seconds;
5581  long       usec = now.tv_usec + millis * 1000;
5582  if (usec >= 1000000) {
5583    abstime->tv_sec += 1;
5584    usec -= 1000000;
5585  }
5586  abstime->tv_nsec = usec * 1000;
5587  return abstime;
5588}
5589
5590// Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5591// Conceptually TryPark() should be equivalent to park(0).
5592
5593int os::PlatformEvent::TryPark() {
5594  for (;;) {
5595    const int v = _Event ;
5596    guarantee ((v == 0) || (v == 1), "invariant") ;
5597    if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5598  }
5599}
5600
5601void os::PlatformEvent::park() {           // AKA: down()
5602  // Invariant: Only the thread associated with the Event/PlatformEvent
5603  // may call park().
5604  int v ;
5605  for (;;) {
5606      v = _Event ;
5607      if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5608  }
5609  guarantee (v >= 0, "invariant") ;
5610  if (v == 0) {
5611     // Do this the hard way by blocking ...
5612     // See http://monaco.sfbay/detail.jsf?cr=5094058.
5613     // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5614     // Only for SPARC >= V8PlusA
5615#if defined(__sparc) && defined(COMPILER2)
5616     if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
5617#endif
5618     int status = os::Solaris::mutex_lock(_mutex);
5619     assert_status(status == 0, status,  "mutex_lock");
5620     guarantee (_nParked == 0, "invariant") ;
5621     ++ _nParked ;
5622     while (_Event < 0) {
5623        // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5624        // Treat this the same as if the wait was interrupted
5625        // With usr/lib/lwp going to kernel, always handle ETIME
5626        status = os::Solaris::cond_wait(_cond, _mutex);
5627        if (status == ETIME) status = EINTR ;
5628        assert_status(status == 0 || status == EINTR, status, "cond_wait");
5629     }
5630     -- _nParked ;
5631     _Event = 0 ;
5632     status = os::Solaris::mutex_unlock(_mutex);
5633     assert_status(status == 0, status, "mutex_unlock");
5634  }
5635}
5636
5637int os::PlatformEvent::park(jlong millis) {
5638  guarantee (_nParked == 0, "invariant") ;
5639  int v ;
5640  for (;;) {
5641      v = _Event ;
5642      if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5643  }
5644  guarantee (v >= 0, "invariant") ;
5645  if (v != 0) return OS_OK ;
5646
5647  int ret = OS_TIMEOUT;
5648  timestruc_t abst;
5649  compute_abstime (&abst, millis);
5650
5651  // See http://monaco.sfbay/detail.jsf?cr=5094058.
5652  // For Solaris SPARC set fprs.FEF=0 prior to parking.
5653  // Only for SPARC >= V8PlusA
5654#if defined(__sparc) && defined(COMPILER2)
5655 if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
5656#endif
5657  int status = os::Solaris::mutex_lock(_mutex);
5658  assert_status(status == 0, status, "mutex_lock");
5659  guarantee (_nParked == 0, "invariant") ;
5660  ++ _nParked ;
5661  while (_Event < 0) {
5662     int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5663     assert_status(status == 0 || status == EINTR ||
5664                   status == ETIME || status == ETIMEDOUT,
5665                   status, "cond_timedwait");
5666     if (!FilterSpuriousWakeups) break ;                // previous semantics
5667     if (status == ETIME || status == ETIMEDOUT) break ;
5668     // We consume and ignore EINTR and spurious wakeups.
5669  }
5670  -- _nParked ;
5671  if (_Event >= 0) ret = OS_OK ;
5672  _Event = 0 ;
5673  status = os::Solaris::mutex_unlock(_mutex);
5674  assert_status(status == 0, status, "mutex_unlock");
5675  return ret;
5676}
5677
5678void os::PlatformEvent::unpark() {
5679  int v, AnyWaiters;
5680
5681  // Increment _Event.
5682  // Another acceptable implementation would be to simply swap 1
5683  // into _Event:
5684  //   if (Swap (&_Event, 1) < 0) {
5685  //      mutex_lock (_mutex) ; AnyWaiters = nParked; mutex_unlock (_mutex) ;
5686  //      if (AnyWaiters) cond_signal (_cond) ;
5687  //   }
5688
5689  for (;;) {
5690    v = _Event ;
5691    if (v > 0) {
5692       // The LD of _Event could have reordered or be satisfied
5693       // by a read-aside from this processor's write buffer.
5694       // To avoid problems execute a barrier and then
5695       // ratify the value.  A degenerate CAS() would also work.
5696       // Viz., CAS (v+0, &_Event, v) == v).
5697       OrderAccess::fence() ;
5698       if (_Event == v) return ;
5699       continue ;
5700    }
5701    if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5702  }
5703
5704  // If the thread associated with the event was parked, wake it.
5705  if (v < 0) {
5706     int status ;
5707     // Wait for the thread assoc with the PlatformEvent to vacate.
5708     status = os::Solaris::mutex_lock(_mutex);
5709     assert_status(status == 0, status, "mutex_lock");
5710     AnyWaiters = _nParked ;
5711     status = os::Solaris::mutex_unlock(_mutex);
5712     assert_status(status == 0, status, "mutex_unlock");
5713     guarantee (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5714     if (AnyWaiters != 0) {
5715       // We intentional signal *after* dropping the lock
5716       // to avoid a common class of futile wakeups.
5717       status = os::Solaris::cond_signal(_cond);
5718       assert_status(status == 0, status, "cond_signal");
5719     }
5720  }
5721}
5722
5723// JSR166
5724// -------------------------------------------------------
5725
5726/*
5727 * The solaris and linux implementations of park/unpark are fairly
5728 * conservative for now, but can be improved. They currently use a
5729 * mutex/condvar pair, plus _counter.
5730 * Park decrements _counter if > 0, else does a condvar wait.  Unpark
5731 * sets count to 1 and signals condvar.  Only one thread ever waits
5732 * on the condvar. Contention seen when trying to park implies that someone
5733 * is unparking you, so don't wait. And spurious returns are fine, so there
5734 * is no need to track notifications.
5735 */
5736
5737#define NANOSECS_PER_SEC 1000000000
5738#define NANOSECS_PER_MILLISEC 1000000
5739#define MAX_SECS 100000000
5740
5741/*
5742 * This code is common to linux and solaris and will be moved to a
5743 * common place in dolphin.
5744 *
5745 * The passed in time value is either a relative time in nanoseconds
5746 * or an absolute time in milliseconds. Either way it has to be unpacked
5747 * into suitable seconds and nanoseconds components and stored in the
5748 * given timespec structure.
5749 * Given time is a 64-bit value and the time_t used in the timespec is only
5750 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5751 * overflow if times way in the future are given. Further on Solaris versions
5752 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5753 * number of seconds, in abstime, is less than current_time  + 100,000,000.
5754 * As it will be 28 years before "now + 100000000" will overflow we can
5755 * ignore overflow and just impose a hard-limit on seconds using the value
5756 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5757 * years from "now".
5758 */
5759static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5760  assert (time > 0, "convertTime");
5761
5762  struct timeval now;
5763  int status = gettimeofday(&now, NULL);
5764  assert(status == 0, "gettimeofday");
5765
5766  time_t max_secs = now.tv_sec + MAX_SECS;
5767
5768  if (isAbsolute) {
5769    jlong secs = time / 1000;
5770    if (secs > max_secs) {
5771      absTime->tv_sec = max_secs;
5772    }
5773    else {
5774      absTime->tv_sec = secs;
5775    }
5776    absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5777  }
5778  else {
5779    jlong secs = time / NANOSECS_PER_SEC;
5780    if (secs >= MAX_SECS) {
5781      absTime->tv_sec = max_secs;
5782      absTime->tv_nsec = 0;
5783    }
5784    else {
5785      absTime->tv_sec = now.tv_sec + secs;
5786      absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5787      if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5788        absTime->tv_nsec -= NANOSECS_PER_SEC;
5789        ++absTime->tv_sec; // note: this must be <= max_secs
5790      }
5791    }
5792  }
5793  assert(absTime->tv_sec >= 0, "tv_sec < 0");
5794  assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5795  assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5796  assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5797}
5798
5799void Parker::park(bool isAbsolute, jlong time) {
5800
5801  // Optional fast-path check:
5802  // Return immediately if a permit is available.
5803  if (_counter > 0) {
5804      _counter = 0 ;
5805      return ;
5806  }
5807
5808  // Optional fast-exit: Check interrupt before trying to wait
5809  Thread* thread = Thread::current();
5810  assert(thread->is_Java_thread(), "Must be JavaThread");
5811  JavaThread *jt = (JavaThread *)thread;
5812  if (Thread::is_interrupted(thread, false)) {
5813    return;
5814  }
5815
5816  // First, demultiplex/decode time arguments
5817  timespec absTime;
5818  if (time < 0) { // don't wait at all
5819    return;
5820  }
5821  if (time > 0) {
5822    // Warning: this code might be exposed to the old Solaris time
5823    // round-down bugs.  Grep "roundingFix" for details.
5824    unpackTime(&absTime, isAbsolute, time);
5825  }
5826
5827  // Enter safepoint region
5828  // Beware of deadlocks such as 6317397.
5829  // The per-thread Parker:: _mutex is a classic leaf-lock.
5830  // In particular a thread must never block on the Threads_lock while
5831  // holding the Parker:: mutex.  If safepoints are pending both the
5832  // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5833  ThreadBlockInVM tbivm(jt);
5834
5835  // Don't wait if cannot get lock since interference arises from
5836  // unblocking.  Also. check interrupt before trying wait
5837  if (Thread::is_interrupted(thread, false) ||
5838      os::Solaris::mutex_trylock(_mutex) != 0) {
5839    return;
5840  }
5841
5842  int status ;
5843
5844  if (_counter > 0)  { // no wait needed
5845    _counter = 0;
5846    status = os::Solaris::mutex_unlock(_mutex);
5847    assert (status == 0, "invariant") ;
5848    return;
5849  }
5850
5851#ifdef ASSERT
5852  // Don't catch signals while blocked; let the running threads have the signals.
5853  // (This allows a debugger to break into the running thread.)
5854  sigset_t oldsigs;
5855  sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
5856  thr_sigsetmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5857#endif
5858
5859  OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5860  jt->set_suspend_equivalent();
5861  // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5862
5863  // Do this the hard way by blocking ...
5864  // See http://monaco.sfbay/detail.jsf?cr=5094058.
5865  // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5866  // Only for SPARC >= V8PlusA
5867#if defined(__sparc) && defined(COMPILER2)
5868  if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
5869#endif
5870
5871  if (time == 0) {
5872    status = os::Solaris::cond_wait (_cond, _mutex) ;
5873  } else {
5874    status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5875  }
5876  // Note that an untimed cond_wait() can sometimes return ETIME on older
5877  // versions of the Solaris.
5878  assert_status(status == 0 || status == EINTR ||
5879                status == ETIME || status == ETIMEDOUT,
5880                status, "cond_timedwait");
5881
5882#ifdef ASSERT
5883  thr_sigsetmask(SIG_SETMASK, &oldsigs, NULL);
5884#endif
5885  _counter = 0 ;
5886  status = os::Solaris::mutex_unlock(_mutex);
5887  assert_status(status == 0, status, "mutex_unlock") ;
5888
5889  // If externally suspended while waiting, re-suspend
5890  if (jt->handle_special_suspend_equivalent_condition()) {
5891    jt->java_suspend_self();
5892  }
5893
5894}
5895
5896void Parker::unpark() {
5897  int s, status ;
5898  status = os::Solaris::mutex_lock (_mutex) ;
5899  assert (status == 0, "invariant") ;
5900  s = _counter;
5901  _counter = 1;
5902  status = os::Solaris::mutex_unlock (_mutex) ;
5903  assert (status == 0, "invariant") ;
5904
5905  if (s < 1) {
5906    status = os::Solaris::cond_signal (_cond) ;
5907    assert (status == 0, "invariant") ;
5908  }
5909}
5910
5911extern char** environ;
5912
5913// Run the specified command in a separate process. Return its exit value,
5914// or -1 on failure (e.g. can't fork a new process).
5915// Unlike system(), this function can be called from signal handler. It
5916// doesn't block SIGINT et al.
5917int os::fork_and_exec(char* cmd) {
5918  char * argv[4];
5919  argv[0] = (char *)"sh";
5920  argv[1] = (char *)"-c";
5921  argv[2] = cmd;
5922  argv[3] = NULL;
5923
5924  // fork is async-safe, fork1 is not so can't use in signal handler
5925  pid_t pid;
5926  Thread* t = ThreadLocalStorage::get_thread_slow();
5927  if (t != NULL && t->is_inside_signal_handler()) {
5928    pid = fork();
5929  } else {
5930    pid = fork1();
5931  }
5932
5933  if (pid < 0) {
5934    // fork failed
5935    warning("fork failed: %s", strerror(errno));
5936    return -1;
5937
5938  } else if (pid == 0) {
5939    // child process
5940
5941    // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5942    execve("/usr/bin/sh", argv, environ);
5943
5944    // execve failed
5945    _exit(-1);
5946
5947  } else  {
5948    // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5949    // care about the actual exit code, for now.
5950
5951    int status;
5952
5953    // Wait for the child process to exit.  This returns immediately if
5954    // the child has already exited. */
5955    while (waitpid(pid, &status, 0) < 0) {
5956        switch (errno) {
5957        case ECHILD: return 0;
5958        case EINTR: break;
5959        default: return -1;
5960        }
5961    }
5962
5963    if (WIFEXITED(status)) {
5964       // The child exited normally; get its exit code.
5965       return WEXITSTATUS(status);
5966    } else if (WIFSIGNALED(status)) {
5967       // The child exited because of a signal
5968       // The best value to return is 0x80 + signal number,
5969       // because that is what all Unix shells do, and because
5970       // it allows callers to distinguish between process exit and
5971       // process death by signal.
5972       return 0x80 + WTERMSIG(status);
5973    } else {
5974       // Unknown exit code; pass it through
5975       return status;
5976    }
5977  }
5978}
5979