os_solaris.cpp revision 4276:6b803ba47588
1/*
2 * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25// no precompiled headers
26#include "classfile/classLoader.hpp"
27#include "classfile/systemDictionary.hpp"
28#include "classfile/vmSymbols.hpp"
29#include "code/icBuffer.hpp"
30#include "code/vtableStubs.hpp"
31#include "compiler/compileBroker.hpp"
32#include "compiler/disassembler.hpp"
33#include "interpreter/interpreter.hpp"
34#include "jvm_solaris.h"
35#include "memory/allocation.inline.hpp"
36#include "memory/filemap.hpp"
37#include "mutex_solaris.inline.hpp"
38#include "oops/oop.inline.hpp"
39#include "os_share_solaris.hpp"
40#include "prims/jniFastGetField.hpp"
41#include "prims/jvm.h"
42#include "prims/jvm_misc.hpp"
43#include "runtime/arguments.hpp"
44#include "runtime/extendedPC.hpp"
45#include "runtime/globals.hpp"
46#include "runtime/interfaceSupport.hpp"
47#include "runtime/java.hpp"
48#include "runtime/javaCalls.hpp"
49#include "runtime/mutexLocker.hpp"
50#include "runtime/objectMonitor.hpp"
51#include "runtime/osThread.hpp"
52#include "runtime/perfMemory.hpp"
53#include "runtime/sharedRuntime.hpp"
54#include "runtime/statSampler.hpp"
55#include "runtime/stubRoutines.hpp"
56#include "runtime/thread.inline.hpp"
57#include "runtime/threadCritical.hpp"
58#include "runtime/timer.hpp"
59#include "services/attachListener.hpp"
60#include "services/memTracker.hpp"
61#include "services/runtimeService.hpp"
62#include "utilities/decoder.hpp"
63#include "utilities/defaultStream.hpp"
64#include "utilities/events.hpp"
65#include "utilities/growableArray.hpp"
66#include "utilities/vmError.hpp"
67
68// put OS-includes here
69# include <dlfcn.h>
70# include <errno.h>
71# include <exception>
72# include <link.h>
73# include <poll.h>
74# include <pthread.h>
75# include <pwd.h>
76# include <schedctl.h>
77# include <setjmp.h>
78# include <signal.h>
79# include <stdio.h>
80# include <alloca.h>
81# include <sys/filio.h>
82# include <sys/ipc.h>
83# include <sys/lwp.h>
84# include <sys/machelf.h>     // for elf Sym structure used by dladdr1
85# include <sys/mman.h>
86# include <sys/processor.h>
87# include <sys/procset.h>
88# include <sys/pset.h>
89# include <sys/resource.h>
90# include <sys/shm.h>
91# include <sys/socket.h>
92# include <sys/stat.h>
93# include <sys/systeminfo.h>
94# include <sys/time.h>
95# include <sys/times.h>
96# include <sys/types.h>
97# include <sys/wait.h>
98# include <sys/utsname.h>
99# include <thread.h>
100# include <unistd.h>
101# include <sys/priocntl.h>
102# include <sys/rtpriocntl.h>
103# include <sys/tspriocntl.h>
104# include <sys/iapriocntl.h>
105# include <sys/fxpriocntl.h>
106# include <sys/loadavg.h>
107# include <string.h>
108# include <stdio.h>
109
110# define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
111# include <sys/procfs.h>     //  see comment in <sys/procfs.h>
112
113#define MAX_PATH (2 * K)
114
115// for timer info max values which include all bits
116#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
117
118#ifdef _GNU_SOURCE
119// See bug #6514594
120extern "C" int madvise(caddr_t, size_t, int);
121extern "C" int memcntl(caddr_t addr, size_t len, int cmd, caddr_t arg,
122                       int attr, int mask);
123#endif //_GNU_SOURCE
124
125/*
126  MPSS Changes Start.
127  The JVM binary needs to be built and run on pre-Solaris 9
128  systems, but the constants needed by MPSS are only in Solaris 9
129  header files.  They are textually replicated here to allow
130  building on earlier systems.  Once building on Solaris 8 is
131  no longer a requirement, these #defines can be replaced by ordinary
132  system .h inclusion.
133
134  In earlier versions of the  JDK and Solaris, we used ISM for large pages.
135  But ISM requires shared memory to achieve this and thus has many caveats.
136  MPSS is a fully transparent and is a cleaner way to get large pages.
137  Although we still require keeping ISM for backward compatiblitiy as well as
138  giving the opportunity to use large pages on older systems it is
139  recommended that MPSS be used for Solaris 9 and above.
140
141*/
142
143#ifndef MC_HAT_ADVISE
144
145struct memcntl_mha {
146  uint_t          mha_cmd;        /* command(s) */
147  uint_t          mha_flags;
148  size_t          mha_pagesize;
149};
150#define MC_HAT_ADVISE   7       /* advise hat map size */
151#define MHA_MAPSIZE_VA  0x1     /* set preferred page size */
152#define MAP_ALIGN       0x200   /* addr specifies alignment */
153
154#endif
155// MPSS Changes End.
156
157
158// Here are some liblgrp types from sys/lgrp_user.h to be able to
159// compile on older systems without this header file.
160
161#ifndef MADV_ACCESS_LWP
162# define  MADV_ACCESS_LWP         7       /* next LWP to access heavily */
163#endif
164#ifndef MADV_ACCESS_MANY
165# define  MADV_ACCESS_MANY        8       /* many processes to access heavily */
166#endif
167
168#ifndef LGRP_RSRC_CPU
169# define LGRP_RSRC_CPU           0       /* CPU resources */
170#endif
171#ifndef LGRP_RSRC_MEM
172# define LGRP_RSRC_MEM           1       /* memory resources */
173#endif
174
175// Some more macros from sys/mman.h that are not present in Solaris 8.
176
177#ifndef MAX_MEMINFO_CNT
178/*
179 * info_req request type definitions for meminfo
180 * request types starting with MEMINFO_V are used for Virtual addresses
181 * and should not be mixed with MEMINFO_PLGRP which is targeted for Physical
182 * addresses
183 */
184# define MEMINFO_SHIFT           16
185# define MEMINFO_MASK            (0xFF << MEMINFO_SHIFT)
186# define MEMINFO_VPHYSICAL       (0x01 << MEMINFO_SHIFT) /* get physical addr */
187# define MEMINFO_VLGRP           (0x02 << MEMINFO_SHIFT) /* get lgroup */
188# define MEMINFO_VPAGESIZE       (0x03 << MEMINFO_SHIFT) /* size of phys page */
189# define MEMINFO_VREPLCNT        (0x04 << MEMINFO_SHIFT) /* no. of replica */
190# define MEMINFO_VREPL           (0x05 << MEMINFO_SHIFT) /* physical replica */
191# define MEMINFO_VREPL_LGRP      (0x06 << MEMINFO_SHIFT) /* lgrp of replica */
192# define MEMINFO_PLGRP           (0x07 << MEMINFO_SHIFT) /* lgroup for paddr */
193
194/* maximum number of addresses meminfo() can process at a time */
195# define MAX_MEMINFO_CNT 256
196
197/* maximum number of request types */
198# define MAX_MEMINFO_REQ 31
199#endif
200
201// see thr_setprio(3T) for the basis of these numbers
202#define MinimumPriority 0
203#define NormalPriority  64
204#define MaximumPriority 127
205
206// Values for ThreadPriorityPolicy == 1
207int prio_policy1[CriticalPriority+1] = {
208  -99999,  0, 16,  32,  48,  64,
209          80, 96, 112, 124, 127, 127 };
210
211// System parameters used internally
212static clock_t clock_tics_per_sec = 100;
213
214// Track if we have called enable_extended_FILE_stdio (on Solaris 10u4+)
215static bool enabled_extended_FILE_stdio = false;
216
217// For diagnostics to print a message once. see run_periodic_checks
218static bool check_addr0_done = false;
219static sigset_t check_signal_done;
220static bool check_signals = true;
221
222address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
223address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
224
225address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
226
227
228// "default" initializers for missing libc APIs
229extern "C" {
230  static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
231  static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
232
233  static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
234  static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
235}
236
237// "default" initializers for pthread-based synchronization
238extern "C" {
239  static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
240  static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
241}
242
243// Thread Local Storage
244// This is common to all Solaris platforms so it is defined here,
245// in this common file.
246// The declarations are in the os_cpu threadLS*.hpp files.
247//
248// Static member initialization for TLS
249Thread* ThreadLocalStorage::_get_thread_cache[ThreadLocalStorage::_pd_cache_size] = {NULL};
250
251#ifndef PRODUCT
252#define _PCT(n,d)       ((100.0*(double)(n))/(double)(d))
253
254int ThreadLocalStorage::_tcacheHit = 0;
255int ThreadLocalStorage::_tcacheMiss = 0;
256
257void ThreadLocalStorage::print_statistics() {
258  int total = _tcacheMiss+_tcacheHit;
259  tty->print_cr("Thread cache hits %d misses %d total %d percent %f\n",
260                _tcacheHit, _tcacheMiss, total, _PCT(_tcacheHit, total));
261}
262#undef _PCT
263#endif // PRODUCT
264
265Thread* ThreadLocalStorage::get_thread_via_cache_slowly(uintptr_t raw_id,
266                                                        int index) {
267  Thread *thread = get_thread_slow();
268  if (thread != NULL) {
269    address sp = os::current_stack_pointer();
270    guarantee(thread->_stack_base == NULL ||
271              (sp <= thread->_stack_base &&
272                 sp >= thread->_stack_base - thread->_stack_size) ||
273               is_error_reported(),
274              "sp must be inside of selected thread stack");
275
276    thread->set_self_raw_id(raw_id);  // mark for quick retrieval
277    _get_thread_cache[ index ] = thread;
278  }
279  return thread;
280}
281
282
283static const double all_zero[ sizeof(Thread) / sizeof(double) + 1 ] = {0};
284#define NO_CACHED_THREAD ((Thread*)all_zero)
285
286void ThreadLocalStorage::pd_set_thread(Thread* thread) {
287
288  // Store the new value before updating the cache to prevent a race
289  // between get_thread_via_cache_slowly() and this store operation.
290  os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
291
292  // Update thread cache with new thread if setting on thread create,
293  // or NO_CACHED_THREAD (zeroed) thread if resetting thread on exit.
294  uintptr_t raw = pd_raw_thread_id();
295  int ix = pd_cache_index(raw);
296  _get_thread_cache[ix] = thread == NULL ? NO_CACHED_THREAD : thread;
297}
298
299void ThreadLocalStorage::pd_init() {
300  for (int i = 0; i < _pd_cache_size; i++) {
301    _get_thread_cache[i] = NO_CACHED_THREAD;
302  }
303}
304
305// Invalidate all the caches (happens to be the same as pd_init).
306void ThreadLocalStorage::pd_invalidate_all() { pd_init(); }
307
308#undef NO_CACHED_THREAD
309
310// END Thread Local Storage
311
312static inline size_t adjust_stack_size(address base, size_t size) {
313  if ((ssize_t)size < 0) {
314    // 4759953: Compensate for ridiculous stack size.
315    size = max_intx;
316  }
317  if (size > (size_t)base) {
318    // 4812466: Make sure size doesn't allow the stack to wrap the address space.
319    size = (size_t)base;
320  }
321  return size;
322}
323
324static inline stack_t get_stack_info() {
325  stack_t st;
326  int retval = thr_stksegment(&st);
327  st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
328  assert(retval == 0, "incorrect return value from thr_stksegment");
329  assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
330  assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
331  return st;
332}
333
334address os::current_stack_base() {
335  int r = thr_main() ;
336  guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
337  bool is_primordial_thread = r;
338
339  // Workaround 4352906, avoid calls to thr_stksegment by
340  // thr_main after the first one (it looks like we trash
341  // some data, causing the value for ss_sp to be incorrect).
342  if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
343    stack_t st = get_stack_info();
344    if (is_primordial_thread) {
345      // cache initial value of stack base
346      os::Solaris::_main_stack_base = (address)st.ss_sp;
347    }
348    return (address)st.ss_sp;
349  } else {
350    guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
351    return os::Solaris::_main_stack_base;
352  }
353}
354
355size_t os::current_stack_size() {
356  size_t size;
357
358  int r = thr_main() ;
359  guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
360  if(!r) {
361    size = get_stack_info().ss_size;
362  } else {
363    struct rlimit limits;
364    getrlimit(RLIMIT_STACK, &limits);
365    size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
366  }
367  // base may not be page aligned
368  address base = current_stack_base();
369  address bottom = (address)align_size_up((intptr_t)(base - size), os::vm_page_size());;
370  return (size_t)(base - bottom);
371}
372
373struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
374  return localtime_r(clock, res);
375}
376
377// interruptible infrastructure
378
379// setup_interruptible saves the thread state before going into an
380// interruptible system call.
381// The saved state is used to restore the thread to
382// its former state whether or not an interrupt is received.
383// Used by classloader os::read
384// os::restartable_read calls skip this layer and stay in _thread_in_native
385
386void os::Solaris::setup_interruptible(JavaThread* thread) {
387
388  JavaThreadState thread_state = thread->thread_state();
389
390  assert(thread_state != _thread_blocked, "Coming from the wrong thread");
391  assert(thread_state != _thread_in_native, "Native threads skip setup_interruptible");
392  OSThread* osthread = thread->osthread();
393  osthread->set_saved_interrupt_thread_state(thread_state);
394  thread->frame_anchor()->make_walkable(thread);
395  ThreadStateTransition::transition(thread, thread_state, _thread_blocked);
396}
397
398// Version of setup_interruptible() for threads that are already in
399// _thread_blocked. Used by os_sleep().
400void os::Solaris::setup_interruptible_already_blocked(JavaThread* thread) {
401  thread->frame_anchor()->make_walkable(thread);
402}
403
404JavaThread* os::Solaris::setup_interruptible() {
405  JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
406  setup_interruptible(thread);
407  return thread;
408}
409
410void os::Solaris::try_enable_extended_io() {
411  typedef int (*enable_extended_FILE_stdio_t)(int, int);
412
413  if (!UseExtendedFileIO) {
414    return;
415  }
416
417  enable_extended_FILE_stdio_t enabler =
418    (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
419                                         "enable_extended_FILE_stdio");
420  if (enabler) {
421    enabler(-1, -1);
422  }
423}
424
425
426#ifdef ASSERT
427
428JavaThread* os::Solaris::setup_interruptible_native() {
429  JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
430  JavaThreadState thread_state = thread->thread_state();
431  assert(thread_state == _thread_in_native, "Assumed thread_in_native");
432  return thread;
433}
434
435void os::Solaris::cleanup_interruptible_native(JavaThread* thread) {
436  JavaThreadState thread_state = thread->thread_state();
437  assert(thread_state == _thread_in_native, "Assumed thread_in_native");
438}
439#endif
440
441// cleanup_interruptible reverses the effects of setup_interruptible
442// setup_interruptible_already_blocked() does not need any cleanup.
443
444void os::Solaris::cleanup_interruptible(JavaThread* thread) {
445  OSThread* osthread = thread->osthread();
446
447  ThreadStateTransition::transition(thread, _thread_blocked, osthread->saved_interrupt_thread_state());
448}
449
450// I/O interruption related counters called in _INTERRUPTIBLE
451
452void os::Solaris::bump_interrupted_before_count() {
453  RuntimeService::record_interrupted_before_count();
454}
455
456void os::Solaris::bump_interrupted_during_count() {
457  RuntimeService::record_interrupted_during_count();
458}
459
460static int _processors_online = 0;
461
462         jint os::Solaris::_os_thread_limit = 0;
463volatile jint os::Solaris::_os_thread_count = 0;
464
465julong os::available_memory() {
466  return Solaris::available_memory();
467}
468
469julong os::Solaris::available_memory() {
470  return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
471}
472
473julong os::Solaris::_physical_memory = 0;
474
475julong os::physical_memory() {
476   return Solaris::physical_memory();
477}
478
479julong os::allocatable_physical_memory(julong size) {
480#ifdef _LP64
481   return size;
482#else
483   julong result = MIN2(size, (julong)3835*M);
484   if (!is_allocatable(result)) {
485     // Memory allocations will be aligned but the alignment
486     // is not known at this point.  Alignments will
487     // be at most to LargePageSizeInBytes.  Protect
488     // allocations from alignments up to illegal
489     // values. If at this point 2G is illegal.
490     julong reasonable_size = (julong)2*G - 2 * LargePageSizeInBytes;
491     result =  MIN2(size, reasonable_size);
492   }
493   return result;
494#endif
495}
496
497static hrtime_t first_hrtime = 0;
498static const hrtime_t hrtime_hz = 1000*1000*1000;
499const int LOCK_BUSY = 1;
500const int LOCK_FREE = 0;
501const int LOCK_INVALID = -1;
502static volatile hrtime_t max_hrtime = 0;
503static volatile int max_hrtime_lock = LOCK_FREE;     // Update counter with LSB as lock-in-progress
504
505
506void os::Solaris::initialize_system_info() {
507  set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
508  _processors_online = sysconf (_SC_NPROCESSORS_ONLN);
509  _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
510}
511
512int os::active_processor_count() {
513  int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
514  pid_t pid = getpid();
515  psetid_t pset = PS_NONE;
516  // Are we running in a processor set or is there any processor set around?
517  if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
518    uint_t pset_cpus;
519    // Query the number of cpus available to us.
520    if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
521      assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
522      _processors_online = pset_cpus;
523      return pset_cpus;
524    }
525  }
526  // Otherwise return number of online cpus
527  return online_cpus;
528}
529
530static bool find_processors_in_pset(psetid_t        pset,
531                                    processorid_t** id_array,
532                                    uint_t*         id_length) {
533  bool result = false;
534  // Find the number of processors in the processor set.
535  if (pset_info(pset, NULL, id_length, NULL) == 0) {
536    // Make up an array to hold their ids.
537    *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
538    // Fill in the array with their processor ids.
539    if (pset_info(pset, NULL, id_length, *id_array) == 0) {
540      result = true;
541    }
542  }
543  return result;
544}
545
546// Callers of find_processors_online() must tolerate imprecise results --
547// the system configuration can change asynchronously because of DR
548// or explicit psradm operations.
549//
550// We also need to take care that the loop (below) terminates as the
551// number of processors online can change between the _SC_NPROCESSORS_ONLN
552// request and the loop that builds the list of processor ids.   Unfortunately
553// there's no reliable way to determine the maximum valid processor id,
554// so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
555// man pages, which claim the processor id set is "sparse, but
556// not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
557// exit the loop.
558//
559// In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
560// not available on S8.0.
561
562static bool find_processors_online(processorid_t** id_array,
563                                   uint*           id_length) {
564  const processorid_t MAX_PROCESSOR_ID = 100000 ;
565  // Find the number of processors online.
566  *id_length = sysconf(_SC_NPROCESSORS_ONLN);
567  // Make up an array to hold their ids.
568  *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
569  // Processors need not be numbered consecutively.
570  long found = 0;
571  processorid_t next = 0;
572  while (found < *id_length && next < MAX_PROCESSOR_ID) {
573    processor_info_t info;
574    if (processor_info(next, &info) == 0) {
575      // NB, PI_NOINTR processors are effectively online ...
576      if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
577        (*id_array)[found] = next;
578        found += 1;
579      }
580    }
581    next += 1;
582  }
583  if (found < *id_length) {
584      // The loop above didn't identify the expected number of processors.
585      // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
586      // and re-running the loop, above, but there's no guarantee of progress
587      // if the system configuration is in flux.  Instead, we just return what
588      // we've got.  Note that in the worst case find_processors_online() could
589      // return an empty set.  (As a fall-back in the case of the empty set we
590      // could just return the ID of the current processor).
591      *id_length = found ;
592  }
593
594  return true;
595}
596
597static bool assign_distribution(processorid_t* id_array,
598                                uint           id_length,
599                                uint*          distribution,
600                                uint           distribution_length) {
601  // We assume we can assign processorid_t's to uint's.
602  assert(sizeof(processorid_t) == sizeof(uint),
603         "can't convert processorid_t to uint");
604  // Quick check to see if we won't succeed.
605  if (id_length < distribution_length) {
606    return false;
607  }
608  // Assign processor ids to the distribution.
609  // Try to shuffle processors to distribute work across boards,
610  // assuming 4 processors per board.
611  const uint processors_per_board = ProcessDistributionStride;
612  // Find the maximum processor id.
613  processorid_t max_id = 0;
614  for (uint m = 0; m < id_length; m += 1) {
615    max_id = MAX2(max_id, id_array[m]);
616  }
617  // The next id, to limit loops.
618  const processorid_t limit_id = max_id + 1;
619  // Make up markers for available processors.
620  bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id, mtInternal);
621  for (uint c = 0; c < limit_id; c += 1) {
622    available_id[c] = false;
623  }
624  for (uint a = 0; a < id_length; a += 1) {
625    available_id[id_array[a]] = true;
626  }
627  // Step by "boards", then by "slot", copying to "assigned".
628  // NEEDS_CLEANUP: The assignment of processors should be stateful,
629  //                remembering which processors have been assigned by
630  //                previous calls, etc., so as to distribute several
631  //                independent calls of this method.  What we'd like is
632  //                It would be nice to have an API that let us ask
633  //                how many processes are bound to a processor,
634  //                but we don't have that, either.
635  //                In the short term, "board" is static so that
636  //                subsequent distributions don't all start at board 0.
637  static uint board = 0;
638  uint assigned = 0;
639  // Until we've found enough processors ....
640  while (assigned < distribution_length) {
641    // ... find the next available processor in the board.
642    for (uint slot = 0; slot < processors_per_board; slot += 1) {
643      uint try_id = board * processors_per_board + slot;
644      if ((try_id < limit_id) && (available_id[try_id] == true)) {
645        distribution[assigned] = try_id;
646        available_id[try_id] = false;
647        assigned += 1;
648        break;
649      }
650    }
651    board += 1;
652    if (board * processors_per_board + 0 >= limit_id) {
653      board = 0;
654    }
655  }
656  if (available_id != NULL) {
657    FREE_C_HEAP_ARRAY(bool, available_id, mtInternal);
658  }
659  return true;
660}
661
662void os::set_native_thread_name(const char *name) {
663  // Not yet implemented.
664  return;
665}
666
667bool os::distribute_processes(uint length, uint* distribution) {
668  bool result = false;
669  // Find the processor id's of all the available CPUs.
670  processorid_t* id_array  = NULL;
671  uint           id_length = 0;
672  // There are some races between querying information and using it,
673  // since processor sets can change dynamically.
674  psetid_t pset = PS_NONE;
675  // Are we running in a processor set?
676  if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
677    result = find_processors_in_pset(pset, &id_array, &id_length);
678  } else {
679    result = find_processors_online(&id_array, &id_length);
680  }
681  if (result == true) {
682    if (id_length >= length) {
683      result = assign_distribution(id_array, id_length, distribution, length);
684    } else {
685      result = false;
686    }
687  }
688  if (id_array != NULL) {
689    FREE_C_HEAP_ARRAY(processorid_t, id_array, mtInternal);
690  }
691  return result;
692}
693
694bool os::bind_to_processor(uint processor_id) {
695  // We assume that a processorid_t can be stored in a uint.
696  assert(sizeof(uint) == sizeof(processorid_t),
697         "can't convert uint to processorid_t");
698  int bind_result =
699    processor_bind(P_LWPID,                       // bind LWP.
700                   P_MYID,                        // bind current LWP.
701                   (processorid_t) processor_id,  // id.
702                   NULL);                         // don't return old binding.
703  return (bind_result == 0);
704}
705
706bool os::getenv(const char* name, char* buffer, int len) {
707  char* val = ::getenv( name );
708  if ( val == NULL
709  ||   strlen(val) + 1  >  len ) {
710    if (len > 0)  buffer[0] = 0; // return a null string
711    return false;
712  }
713  strcpy( buffer, val );
714  return true;
715}
716
717
718// Return true if user is running as root.
719
720bool os::have_special_privileges() {
721  static bool init = false;
722  static bool privileges = false;
723  if (!init) {
724    privileges = (getuid() != geteuid()) || (getgid() != getegid());
725    init = true;
726  }
727  return privileges;
728}
729
730
731void os::init_system_properties_values() {
732  char arch[12];
733  sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
734
735  // The next steps are taken in the product version:
736  //
737  // Obtain the JAVA_HOME value from the location of libjvm.so.
738  // This library should be located at:
739  // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
740  //
741  // If "/jre/lib/" appears at the right place in the path, then we
742  // assume libjvm.so is installed in a JDK and we use this path.
743  //
744  // Otherwise exit with message: "Could not create the Java virtual machine."
745  //
746  // The following extra steps are taken in the debugging version:
747  //
748  // If "/jre/lib/" does NOT appear at the right place in the path
749  // instead of exit check for $JAVA_HOME environment variable.
750  //
751  // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
752  // then we append a fake suffix "hotspot/libjvm.so" to this path so
753  // it looks like libjvm.so is installed there
754  // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
755  //
756  // Otherwise exit.
757  //
758  // Important note: if the location of libjvm.so changes this
759  // code needs to be changed accordingly.
760
761  // The next few definitions allow the code to be verbatim:
762#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
763#define free(p) FREE_C_HEAP_ARRAY(char, p, mtInternal)
764#define getenv(n) ::getenv(n)
765
766#define EXTENSIONS_DIR  "/lib/ext"
767#define ENDORSED_DIR    "/lib/endorsed"
768#define COMMON_DIR      "/usr/jdk/packages"
769
770  {
771    /* sysclasspath, java_home, dll_dir */
772    {
773        char *home_path;
774        char *dll_path;
775        char *pslash;
776        char buf[MAXPATHLEN];
777        os::jvm_path(buf, sizeof(buf));
778
779        // Found the full path to libjvm.so.
780        // Now cut the path to <java_home>/jre if we can.
781        *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
782        pslash = strrchr(buf, '/');
783        if (pslash != NULL)
784            *pslash = '\0';           /* get rid of /{client|server|hotspot} */
785        dll_path = malloc(strlen(buf) + 1);
786        if (dll_path == NULL)
787            return;
788        strcpy(dll_path, buf);
789        Arguments::set_dll_dir(dll_path);
790
791        if (pslash != NULL) {
792            pslash = strrchr(buf, '/');
793            if (pslash != NULL) {
794                *pslash = '\0';       /* get rid of /<arch> */
795                pslash = strrchr(buf, '/');
796                if (pslash != NULL)
797                    *pslash = '\0';   /* get rid of /lib */
798            }
799        }
800
801        home_path = malloc(strlen(buf) + 1);
802        if (home_path == NULL)
803            return;
804        strcpy(home_path, buf);
805        Arguments::set_java_home(home_path);
806
807        if (!set_boot_path('/', ':'))
808            return;
809    }
810
811    /*
812     * Where to look for native libraries
813     */
814    {
815      // Use dlinfo() to determine the correct java.library.path.
816      //
817      // If we're launched by the Java launcher, and the user
818      // does not set java.library.path explicitly on the commandline,
819      // the Java launcher sets LD_LIBRARY_PATH for us and unsets
820      // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
821      // dlinfo returns LD_LIBRARY_PATH + crle settings (including
822      // /usr/lib), which is exactly what we want.
823      //
824      // If the user does set java.library.path, it completely
825      // overwrites this setting, and always has.
826      //
827      // If we're not launched by the Java launcher, we may
828      // get here with any/all of the LD_LIBRARY_PATH[_32|64]
829      // settings.  Again, dlinfo does exactly what we want.
830
831      Dl_serinfo     _info, *info = &_info;
832      Dl_serpath     *path;
833      char*          library_path;
834      char           *common_path;
835      int            i;
836
837      // determine search path count and required buffer size
838      if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
839        vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
840      }
841
842      // allocate new buffer and initialize
843      info = (Dl_serinfo*)malloc(_info.dls_size);
844      if (info == NULL) {
845        vm_exit_out_of_memory(_info.dls_size,
846                              "init_system_properties_values info");
847      }
848      info->dls_size = _info.dls_size;
849      info->dls_cnt = _info.dls_cnt;
850
851      // obtain search path information
852      if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
853        free(info);
854        vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
855      }
856
857      path = &info->dls_serpath[0];
858
859      // Note: Due to a legacy implementation, most of the library path
860      // is set in the launcher.  This was to accomodate linking restrictions
861      // on legacy Solaris implementations (which are no longer supported).
862      // Eventually, all the library path setting will be done here.
863      //
864      // However, to prevent the proliferation of improperly built native
865      // libraries, the new path component /usr/jdk/packages is added here.
866
867      // Determine the actual CPU architecture.
868      char cpu_arch[12];
869      sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
870#ifdef _LP64
871      // If we are a 64-bit vm, perform the following translations:
872      //   sparc   -> sparcv9
873      //   i386    -> amd64
874      if (strcmp(cpu_arch, "sparc") == 0)
875        strcat(cpu_arch, "v9");
876      else if (strcmp(cpu_arch, "i386") == 0)
877        strcpy(cpu_arch, "amd64");
878#endif
879
880      // Construct the invariant part of ld_library_path. Note that the
881      // space for the colon and the trailing null are provided by the
882      // nulls included by the sizeof operator.
883      size_t bufsize = sizeof(COMMON_DIR) + sizeof("/lib/") + strlen(cpu_arch);
884      common_path = malloc(bufsize);
885      if (common_path == NULL) {
886        free(info);
887        vm_exit_out_of_memory(bufsize,
888                              "init_system_properties_values common_path");
889      }
890      sprintf(common_path, COMMON_DIR "/lib/%s", cpu_arch);
891
892      // struct size is more than sufficient for the path components obtained
893      // through the dlinfo() call, so only add additional space for the path
894      // components explicitly added here.
895      bufsize = info->dls_size + strlen(common_path);
896      library_path = malloc(bufsize);
897      if (library_path == NULL) {
898        free(info);
899        free(common_path);
900        vm_exit_out_of_memory(bufsize,
901                              "init_system_properties_values library_path");
902      }
903      library_path[0] = '\0';
904
905      // Construct the desired Java library path from the linker's library
906      // search path.
907      //
908      // For compatibility, it is optimal that we insert the additional path
909      // components specific to the Java VM after those components specified
910      // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
911      // infrastructure.
912      if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it
913        strcpy(library_path, common_path);
914      } else {
915        int inserted = 0;
916        for (i = 0; i < info->dls_cnt; i++, path++) {
917          uint_t flags = path->dls_flags & LA_SER_MASK;
918          if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
919            strcat(library_path, common_path);
920            strcat(library_path, os::path_separator());
921            inserted = 1;
922          }
923          strcat(library_path, path->dls_name);
924          strcat(library_path, os::path_separator());
925        }
926        // eliminate trailing path separator
927        library_path[strlen(library_path)-1] = '\0';
928      }
929
930      // happens before argument parsing - can't use a trace flag
931      // tty->print_raw("init_system_properties_values: native lib path: ");
932      // tty->print_raw_cr(library_path);
933
934      // callee copies into its own buffer
935      Arguments::set_library_path(library_path);
936
937      free(common_path);
938      free(library_path);
939      free(info);
940    }
941
942    /*
943     * Extensions directories.
944     *
945     * Note that the space for the colon and the trailing null are provided
946     * by the nulls included by the sizeof operator (so actually one byte more
947     * than necessary is allocated).
948     */
949    {
950        char *buf = (char *) malloc(strlen(Arguments::get_java_home()) +
951            sizeof(EXTENSIONS_DIR) + sizeof(COMMON_DIR) +
952            sizeof(EXTENSIONS_DIR));
953        sprintf(buf, "%s" EXTENSIONS_DIR ":" COMMON_DIR EXTENSIONS_DIR,
954            Arguments::get_java_home());
955        Arguments::set_ext_dirs(buf);
956    }
957
958    /* Endorsed standards default directory. */
959    {
960        char * buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
961        sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
962        Arguments::set_endorsed_dirs(buf);
963    }
964  }
965
966#undef malloc
967#undef free
968#undef getenv
969#undef EXTENSIONS_DIR
970#undef ENDORSED_DIR
971#undef COMMON_DIR
972
973}
974
975void os::breakpoint() {
976  BREAKPOINT;
977}
978
979bool os::obsolete_option(const JavaVMOption *option)
980{
981  if (!strncmp(option->optionString, "-Xt", 3)) {
982    return true;
983  } else if (!strncmp(option->optionString, "-Xtm", 4)) {
984    return true;
985  } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
986    return true;
987  } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
988    return true;
989  }
990  return false;
991}
992
993bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
994  address  stackStart  = (address)thread->stack_base();
995  address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
996  if (sp < stackStart && sp >= stackEnd ) return true;
997  return false;
998}
999
1000extern "C" void breakpoint() {
1001  // use debugger to set breakpoint here
1002}
1003
1004static thread_t main_thread;
1005
1006// Thread start routine for all new Java threads
1007extern "C" void* java_start(void* thread_addr) {
1008  // Try to randomize the cache line index of hot stack frames.
1009  // This helps when threads of the same stack traces evict each other's
1010  // cache lines. The threads can be either from the same JVM instance, or
1011  // from different JVM instances. The benefit is especially true for
1012  // processors with hyperthreading technology.
1013  static int counter = 0;
1014  int pid = os::current_process_id();
1015  alloca(((pid ^ counter++) & 7) * 128);
1016
1017  int prio;
1018  Thread* thread = (Thread*)thread_addr;
1019  OSThread* osthr = thread->osthread();
1020
1021  osthr->set_lwp_id( _lwp_self() );  // Store lwp in case we are bound
1022  thread->_schedctl = (void *) schedctl_init () ;
1023
1024  if (UseNUMA) {
1025    int lgrp_id = os::numa_get_group_id();
1026    if (lgrp_id != -1) {
1027      thread->set_lgrp_id(lgrp_id);
1028    }
1029  }
1030
1031  // If the creator called set priority before we started,
1032  // we need to call set_native_priority now that we have an lwp.
1033  // We used to get the priority from thr_getprio (we called
1034  // thr_setprio way back in create_thread) and pass it to
1035  // set_native_priority, but Solaris scales the priority
1036  // in java_to_os_priority, so when we read it back here,
1037  // we pass trash to set_native_priority instead of what's
1038  // in java_to_os_priority. So we save the native priority
1039  // in the osThread and recall it here.
1040
1041  if ( osthr->thread_id() != -1 ) {
1042    if ( UseThreadPriorities ) {
1043      int prio = osthr->native_priority();
1044      if (ThreadPriorityVerbose) {
1045        tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
1046                      INTPTR_FORMAT ", setting priority: %d\n",
1047                      osthr->thread_id(), osthr->lwp_id(), prio);
1048      }
1049      os::set_native_priority(thread, prio);
1050    }
1051  } else if (ThreadPriorityVerbose) {
1052    warning("Can't set priority in _start routine, thread id hasn't been set\n");
1053  }
1054
1055  assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
1056
1057  // initialize signal mask for this thread
1058  os::Solaris::hotspot_sigmask(thread);
1059
1060  thread->run();
1061
1062  // One less thread is executing
1063  // When the VMThread gets here, the main thread may have already exited
1064  // which frees the CodeHeap containing the Atomic::dec code
1065  if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
1066    Atomic::dec(&os::Solaris::_os_thread_count);
1067  }
1068
1069  if (UseDetachedThreads) {
1070    thr_exit(NULL);
1071    ShouldNotReachHere();
1072  }
1073  return NULL;
1074}
1075
1076static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
1077  // Allocate the OSThread object
1078  OSThread* osthread = new OSThread(NULL, NULL);
1079  if (osthread == NULL) return NULL;
1080
1081  // Store info on the Solaris thread into the OSThread
1082  osthread->set_thread_id(thread_id);
1083  osthread->set_lwp_id(_lwp_self());
1084  thread->_schedctl = (void *) schedctl_init () ;
1085
1086  if (UseNUMA) {
1087    int lgrp_id = os::numa_get_group_id();
1088    if (lgrp_id != -1) {
1089      thread->set_lgrp_id(lgrp_id);
1090    }
1091  }
1092
1093  if ( ThreadPriorityVerbose ) {
1094    tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
1095                  osthread->thread_id(), osthread->lwp_id() );
1096  }
1097
1098  // Initial thread state is INITIALIZED, not SUSPENDED
1099  osthread->set_state(INITIALIZED);
1100
1101  return osthread;
1102}
1103
1104void os::Solaris::hotspot_sigmask(Thread* thread) {
1105
1106  //Save caller's signal mask
1107  sigset_t sigmask;
1108  thr_sigsetmask(SIG_SETMASK, NULL, &sigmask);
1109  OSThread *osthread = thread->osthread();
1110  osthread->set_caller_sigmask(sigmask);
1111
1112  thr_sigsetmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
1113  if (!ReduceSignalUsage) {
1114    if (thread->is_VM_thread()) {
1115      // Only the VM thread handles BREAK_SIGNAL ...
1116      thr_sigsetmask(SIG_UNBLOCK, vm_signals(), NULL);
1117    } else {
1118      // ... all other threads block BREAK_SIGNAL
1119      assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
1120      thr_sigsetmask(SIG_BLOCK, vm_signals(), NULL);
1121    }
1122  }
1123}
1124
1125bool os::create_attached_thread(JavaThread* thread) {
1126#ifdef ASSERT
1127  thread->verify_not_published();
1128#endif
1129  OSThread* osthread = create_os_thread(thread, thr_self());
1130  if (osthread == NULL) {
1131     return false;
1132  }
1133
1134  // Initial thread state is RUNNABLE
1135  osthread->set_state(RUNNABLE);
1136  thread->set_osthread(osthread);
1137
1138  // initialize signal mask for this thread
1139  // and save the caller's signal mask
1140  os::Solaris::hotspot_sigmask(thread);
1141
1142  return true;
1143}
1144
1145bool os::create_main_thread(JavaThread* thread) {
1146#ifdef ASSERT
1147  thread->verify_not_published();
1148#endif
1149  if (_starting_thread == NULL) {
1150    _starting_thread = create_os_thread(thread, main_thread);
1151     if (_starting_thread == NULL) {
1152        return false;
1153     }
1154  }
1155
1156  // The primodial thread is runnable from the start
1157  _starting_thread->set_state(RUNNABLE);
1158
1159  thread->set_osthread(_starting_thread);
1160
1161  // initialize signal mask for this thread
1162  // and save the caller's signal mask
1163  os::Solaris::hotspot_sigmask(thread);
1164
1165  return true;
1166}
1167
1168// _T2_libthread is true if we believe we are running with the newer
1169// SunSoft lwp/libthread.so (2.8 patch, 2.9 default)
1170bool os::Solaris::_T2_libthread = false;
1171
1172bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1173  // Allocate the OSThread object
1174  OSThread* osthread = new OSThread(NULL, NULL);
1175  if (osthread == NULL) {
1176    return false;
1177  }
1178
1179  if ( ThreadPriorityVerbose ) {
1180    char *thrtyp;
1181    switch ( thr_type ) {
1182      case vm_thread:
1183        thrtyp = (char *)"vm";
1184        break;
1185      case cgc_thread:
1186        thrtyp = (char *)"cgc";
1187        break;
1188      case pgc_thread:
1189        thrtyp = (char *)"pgc";
1190        break;
1191      case java_thread:
1192        thrtyp = (char *)"java";
1193        break;
1194      case compiler_thread:
1195        thrtyp = (char *)"compiler";
1196        break;
1197      case watcher_thread:
1198        thrtyp = (char *)"watcher";
1199        break;
1200      default:
1201        thrtyp = (char *)"unknown";
1202        break;
1203    }
1204    tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
1205  }
1206
1207  // Calculate stack size if it's not specified by caller.
1208  if (stack_size == 0) {
1209    // The default stack size 1M (2M for LP64).
1210    stack_size = (BytesPerWord >> 2) * K * K;
1211
1212    switch (thr_type) {
1213    case os::java_thread:
1214      // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
1215      if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
1216      break;
1217    case os::compiler_thread:
1218      if (CompilerThreadStackSize > 0) {
1219        stack_size = (size_t)(CompilerThreadStackSize * K);
1220        break;
1221      } // else fall through:
1222        // use VMThreadStackSize if CompilerThreadStackSize is not defined
1223    case os::vm_thread:
1224    case os::pgc_thread:
1225    case os::cgc_thread:
1226    case os::watcher_thread:
1227      if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1228      break;
1229    }
1230  }
1231  stack_size = MAX2(stack_size, os::Solaris::min_stack_allowed);
1232
1233  // Initial state is ALLOCATED but not INITIALIZED
1234  osthread->set_state(ALLOCATED);
1235
1236  if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
1237    // We got lots of threads. Check if we still have some address space left.
1238    // Need to be at least 5Mb of unreserved address space. We do check by
1239    // trying to reserve some.
1240    const size_t VirtualMemoryBangSize = 20*K*K;
1241    char* mem = os::reserve_memory(VirtualMemoryBangSize);
1242    if (mem == NULL) {
1243      delete osthread;
1244      return false;
1245    } else {
1246      // Release the memory again
1247      os::release_memory(mem, VirtualMemoryBangSize);
1248    }
1249  }
1250
1251  // Setup osthread because the child thread may need it.
1252  thread->set_osthread(osthread);
1253
1254  // Create the Solaris thread
1255  // explicit THR_BOUND for T2_libthread case in case
1256  // that assumption is not accurate, but our alternate signal stack
1257  // handling is based on it which must have bound threads
1258  thread_t tid = 0;
1259  long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED
1260                   | ((UseBoundThreads || os::Solaris::T2_libthread() ||
1261                       (thr_type == vm_thread) ||
1262                       (thr_type == cgc_thread) ||
1263                       (thr_type == pgc_thread) ||
1264                       (thr_type == compiler_thread && BackgroundCompilation)) ?
1265                      THR_BOUND : 0);
1266  int      status;
1267
1268  // 4376845 -- libthread/kernel don't provide enough LWPs to utilize all CPUs.
1269  //
1270  // On multiprocessors systems, libthread sometimes under-provisions our
1271  // process with LWPs.  On a 30-way systems, for instance, we could have
1272  // 50 user-level threads in ready state and only 2 or 3 LWPs assigned
1273  // to our process.  This can result in under utilization of PEs.
1274  // I suspect the problem is related to libthread's LWP
1275  // pool management and to the kernel's SIGBLOCKING "last LWP parked"
1276  // upcall policy.
1277  //
1278  // The following code is palliative -- it attempts to ensure that our
1279  // process has sufficient LWPs to take advantage of multiple PEs.
1280  // Proper long-term cures include using user-level threads bound to LWPs
1281  // (THR_BOUND) or using LWP-based synchronization.  Note that there is a
1282  // slight timing window with respect to sampling _os_thread_count, but
1283  // the race is benign.  Also, we should periodically recompute
1284  // _processors_online as the min of SC_NPROCESSORS_ONLN and the
1285  // the number of PEs in our partition.  You might be tempted to use
1286  // THR_NEW_LWP here, but I'd recommend against it as that could
1287  // result in undesirable growth of the libthread's LWP pool.
1288  // The fix below isn't sufficient; for instance, it doesn't take into count
1289  // LWPs parked on IO.  It does, however, help certain CPU-bound benchmarks.
1290  //
1291  // Some pathologies this scheme doesn't handle:
1292  // *  Threads can block, releasing the LWPs.  The LWPs can age out.
1293  //    When a large number of threads become ready again there aren't
1294  //    enough LWPs available to service them.  This can occur when the
1295  //    number of ready threads oscillates.
1296  // *  LWPs/Threads park on IO, thus taking the LWP out of circulation.
1297  //
1298  // Finally, we should call thr_setconcurrency() periodically to refresh
1299  // the LWP pool and thwart the LWP age-out mechanism.
1300  // The "+3" term provides a little slop -- we want to slightly overprovision.
1301
1302  if (AdjustConcurrency && os::Solaris::_os_thread_count < (_processors_online+3)) {
1303    if (!(flags & THR_BOUND)) {
1304      thr_setconcurrency (os::Solaris::_os_thread_count);       // avoid starvation
1305    }
1306  }
1307  // Although this doesn't hurt, we should warn of undefined behavior
1308  // when using unbound T1 threads with schedctl().  This should never
1309  // happen, as the compiler and VM threads are always created bound
1310  DEBUG_ONLY(
1311      if ((VMThreadHintNoPreempt || CompilerThreadHintNoPreempt) &&
1312          (!os::Solaris::T2_libthread() && (!(flags & THR_BOUND))) &&
1313          ((thr_type == vm_thread) || (thr_type == cgc_thread) ||
1314           (thr_type == pgc_thread) || (thr_type == compiler_thread && BackgroundCompilation))) {
1315         warning("schedctl behavior undefined when Compiler/VM/GC Threads are Unbound");
1316      }
1317  );
1318
1319
1320  // Mark that we don't have an lwp or thread id yet.
1321  // In case we attempt to set the priority before the thread starts.
1322  osthread->set_lwp_id(-1);
1323  osthread->set_thread_id(-1);
1324
1325  status = thr_create(NULL, stack_size, java_start, thread, flags, &tid);
1326  if (status != 0) {
1327    if (PrintMiscellaneous && (Verbose || WizardMode)) {
1328      perror("os::create_thread");
1329    }
1330    thread->set_osthread(NULL);
1331    // Need to clean up stuff we've allocated so far
1332    delete osthread;
1333    return false;
1334  }
1335
1336  Atomic::inc(&os::Solaris::_os_thread_count);
1337
1338  // Store info on the Solaris thread into the OSThread
1339  osthread->set_thread_id(tid);
1340
1341  // Remember that we created this thread so we can set priority on it
1342  osthread->set_vm_created();
1343
1344  // Set the default thread priority.  If using bound threads, setting
1345  // lwp priority will be delayed until thread start.
1346  set_native_priority(thread,
1347                      DefaultThreadPriority == -1 ?
1348                        java_to_os_priority[NormPriority] :
1349                        DefaultThreadPriority);
1350
1351  // Initial thread state is INITIALIZED, not SUSPENDED
1352  osthread->set_state(INITIALIZED);
1353
1354  // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1355  return true;
1356}
1357
1358/* defined for >= Solaris 10. This allows builds on earlier versions
1359 *  of Solaris to take advantage of the newly reserved Solaris JVM signals
1360 *  With SIGJVM1, SIGJVM2, INTERRUPT_SIGNAL is SIGJVM1, ASYNC_SIGNAL is SIGJVM2
1361 *  and -XX:+UseAltSigs does nothing since these should have no conflict
1362 */
1363#if !defined(SIGJVM1)
1364#define SIGJVM1 39
1365#define SIGJVM2 40
1366#endif
1367
1368debug_only(static bool signal_sets_initialized = false);
1369static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
1370int os::Solaris::_SIGinterrupt = INTERRUPT_SIGNAL;
1371int os::Solaris::_SIGasync = ASYNC_SIGNAL;
1372
1373bool os::Solaris::is_sig_ignored(int sig) {
1374      struct sigaction oact;
1375      sigaction(sig, (struct sigaction*)NULL, &oact);
1376      void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1377                                     : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1378      if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
1379           return true;
1380      else
1381           return false;
1382}
1383
1384// Note: SIGRTMIN is a macro that calls sysconf() so it will
1385// dynamically detect SIGRTMIN value for the system at runtime, not buildtime
1386static bool isJVM1available() {
1387  return SIGJVM1 < SIGRTMIN;
1388}
1389
1390void os::Solaris::signal_sets_init() {
1391  // Should also have an assertion stating we are still single-threaded.
1392  assert(!signal_sets_initialized, "Already initialized");
1393  // Fill in signals that are necessarily unblocked for all threads in
1394  // the VM. Currently, we unblock the following signals:
1395  // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1396  //                         by -Xrs (=ReduceSignalUsage));
1397  // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1398  // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1399  // the dispositions or masks wrt these signals.
1400  // Programs embedding the VM that want to use the above signals for their
1401  // own purposes must, at this time, use the "-Xrs" option to prevent
1402  // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1403  // (See bug 4345157, and other related bugs).
1404  // In reality, though, unblocking these signals is really a nop, since
1405  // these signals are not blocked by default.
1406  sigemptyset(&unblocked_sigs);
1407  sigemptyset(&allowdebug_blocked_sigs);
1408  sigaddset(&unblocked_sigs, SIGILL);
1409  sigaddset(&unblocked_sigs, SIGSEGV);
1410  sigaddset(&unblocked_sigs, SIGBUS);
1411  sigaddset(&unblocked_sigs, SIGFPE);
1412
1413  if (isJVM1available) {
1414    os::Solaris::set_SIGinterrupt(SIGJVM1);
1415    os::Solaris::set_SIGasync(SIGJVM2);
1416  } else if (UseAltSigs) {
1417    os::Solaris::set_SIGinterrupt(ALT_INTERRUPT_SIGNAL);
1418    os::Solaris::set_SIGasync(ALT_ASYNC_SIGNAL);
1419  } else {
1420    os::Solaris::set_SIGinterrupt(INTERRUPT_SIGNAL);
1421    os::Solaris::set_SIGasync(ASYNC_SIGNAL);
1422  }
1423
1424  sigaddset(&unblocked_sigs, os::Solaris::SIGinterrupt());
1425  sigaddset(&unblocked_sigs, os::Solaris::SIGasync());
1426
1427  if (!ReduceSignalUsage) {
1428   if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1429      sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1430      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
1431   }
1432   if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1433      sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1434      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
1435   }
1436   if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1437      sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1438      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
1439   }
1440  }
1441  // Fill in signals that are blocked by all but the VM thread.
1442  sigemptyset(&vm_sigs);
1443  if (!ReduceSignalUsage)
1444    sigaddset(&vm_sigs, BREAK_SIGNAL);
1445  debug_only(signal_sets_initialized = true);
1446
1447  // For diagnostics only used in run_periodic_checks
1448  sigemptyset(&check_signal_done);
1449}
1450
1451// These are signals that are unblocked while a thread is running Java.
1452// (For some reason, they get blocked by default.)
1453sigset_t* os::Solaris::unblocked_signals() {
1454  assert(signal_sets_initialized, "Not initialized");
1455  return &unblocked_sigs;
1456}
1457
1458// These are the signals that are blocked while a (non-VM) thread is
1459// running Java. Only the VM thread handles these signals.
1460sigset_t* os::Solaris::vm_signals() {
1461  assert(signal_sets_initialized, "Not initialized");
1462  return &vm_sigs;
1463}
1464
1465// These are signals that are blocked during cond_wait to allow debugger in
1466sigset_t* os::Solaris::allowdebug_blocked_signals() {
1467  assert(signal_sets_initialized, "Not initialized");
1468  return &allowdebug_blocked_sigs;
1469}
1470
1471
1472void _handle_uncaught_cxx_exception() {
1473  VMError err("An uncaught C++ exception");
1474  err.report_and_die();
1475}
1476
1477
1478// First crack at OS-specific initialization, from inside the new thread.
1479void os::initialize_thread(Thread* thr) {
1480  int r = thr_main() ;
1481  guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
1482  if (r) {
1483    JavaThread* jt = (JavaThread *)thr;
1484    assert(jt != NULL,"Sanity check");
1485    size_t stack_size;
1486    address base = jt->stack_base();
1487    if (Arguments::created_by_java_launcher()) {
1488      // Use 2MB to allow for Solaris 7 64 bit mode.
1489      stack_size = JavaThread::stack_size_at_create() == 0
1490        ? 2048*K : JavaThread::stack_size_at_create();
1491
1492      // There are rare cases when we may have already used more than
1493      // the basic stack size allotment before this method is invoked.
1494      // Attempt to allow for a normally sized java_stack.
1495      size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1496      stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1497    } else {
1498      // 6269555: If we were not created by a Java launcher, i.e. if we are
1499      // running embedded in a native application, treat the primordial thread
1500      // as much like a native attached thread as possible.  This means using
1501      // the current stack size from thr_stksegment(), unless it is too large
1502      // to reliably setup guard pages.  A reasonable max size is 8MB.
1503      size_t current_size = current_stack_size();
1504      // This should never happen, but just in case....
1505      if (current_size == 0) current_size = 2 * K * K;
1506      stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1507    }
1508    address bottom = (address)align_size_up((intptr_t)(base - stack_size), os::vm_page_size());;
1509    stack_size = (size_t)(base - bottom);
1510
1511    assert(stack_size > 0, "Stack size calculation problem");
1512
1513    if (stack_size > jt->stack_size()) {
1514      NOT_PRODUCT(
1515        struct rlimit limits;
1516        getrlimit(RLIMIT_STACK, &limits);
1517        size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1518        assert(size >= jt->stack_size(), "Stack size problem in main thread");
1519      )
1520      tty->print_cr(
1521        "Stack size of %d Kb exceeds current limit of %d Kb.\n"
1522        "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1523        "See limit(1) to increase the stack size limit.",
1524        stack_size / K, jt->stack_size() / K);
1525      vm_exit(1);
1526    }
1527    assert(jt->stack_size() >= stack_size,
1528          "Attempt to map more stack than was allocated");
1529    jt->set_stack_size(stack_size);
1530  }
1531
1532   // 5/22/01: Right now alternate signal stacks do not handle
1533   // throwing stack overflow exceptions, see bug 4463178
1534   // Until a fix is found for this, T2 will NOT imply alternate signal
1535   // stacks.
1536   // If using T2 libthread threads, install an alternate signal stack.
1537   // Because alternate stacks associate with LWPs on Solaris,
1538   // see sigaltstack(2), if using UNBOUND threads, or if UseBoundThreads
1539   // we prefer to explicitly stack bang.
1540   // If not using T2 libthread, but using UseBoundThreads any threads
1541   // (primordial thread, jni_attachCurrentThread) we do not create,
1542   // probably are not bound, therefore they can not have an alternate
1543   // signal stack. Since our stack banging code is generated and
1544   // is shared across threads, all threads must be bound to allow
1545   // using alternate signal stacks.  The alternative is to interpose
1546   // on _lwp_create to associate an alt sig stack with each LWP,
1547   // and this could be a problem when the JVM is embedded.
1548   // We would prefer to use alternate signal stacks with T2
1549   // Since there is currently no accurate way to detect T2
1550   // we do not. Assuming T2 when running T1 causes sig 11s or assertions
1551   // on installing alternate signal stacks
1552
1553
1554   // 05/09/03: removed alternate signal stack support for Solaris
1555   // The alternate signal stack mechanism is no longer needed to
1556   // handle stack overflow. This is now handled by allocating
1557   // guard pages (red zone) and stackbanging.
1558   // Initially the alternate signal stack mechanism was removed because
1559   // it did not work with T1 llibthread. Alternate
1560   // signal stacks MUST have all threads bound to lwps. Applications
1561   // can create their own threads and attach them without their being
1562   // bound under T1. This is frequently the case for the primordial thread.
1563   // If we were ever to reenable this mechanism we would need to
1564   // use the dynamic check for T2 libthread.
1565
1566  os::Solaris::init_thread_fpu_state();
1567  std::set_terminate(_handle_uncaught_cxx_exception);
1568}
1569
1570
1571
1572// Free Solaris resources related to the OSThread
1573void os::free_thread(OSThread* osthread) {
1574  assert(osthread != NULL, "os::free_thread but osthread not set");
1575
1576
1577  // We are told to free resources of the argument thread,
1578  // but we can only really operate on the current thread.
1579  // The main thread must take the VMThread down synchronously
1580  // before the main thread exits and frees up CodeHeap
1581  guarantee((Thread::current()->osthread() == osthread
1582     || (osthread == VMThread::vm_thread()->osthread())), "os::free_thread but not current thread");
1583  if (Thread::current()->osthread() == osthread) {
1584    // Restore caller's signal mask
1585    sigset_t sigmask = osthread->caller_sigmask();
1586    thr_sigsetmask(SIG_SETMASK, &sigmask, NULL);
1587  }
1588  delete osthread;
1589}
1590
1591void os::pd_start_thread(Thread* thread) {
1592  int status = thr_continue(thread->osthread()->thread_id());
1593  assert_status(status == 0, status, "thr_continue failed");
1594}
1595
1596
1597intx os::current_thread_id() {
1598  return (intx)thr_self();
1599}
1600
1601static pid_t _initial_pid = 0;
1602
1603int os::current_process_id() {
1604  return (int)(_initial_pid ? _initial_pid : getpid());
1605}
1606
1607int os::allocate_thread_local_storage() {
1608  // %%%       in Win32 this allocates a memory segment pointed to by a
1609  //           register.  Dan Stein can implement a similar feature in
1610  //           Solaris.  Alternatively, the VM can do the same thing
1611  //           explicitly: malloc some storage and keep the pointer in a
1612  //           register (which is part of the thread's context) (or keep it
1613  //           in TLS).
1614  // %%%       In current versions of Solaris, thr_self and TSD can
1615  //           be accessed via short sequences of displaced indirections.
1616  //           The value of thr_self is available as %g7(36).
1617  //           The value of thr_getspecific(k) is stored in %g7(12)(4)(k*4-4),
1618  //           assuming that the current thread already has a value bound to k.
1619  //           It may be worth experimenting with such access patterns,
1620  //           and later having the parameters formally exported from a Solaris
1621  //           interface.  I think, however, that it will be faster to
1622  //           maintain the invariant that %g2 always contains the
1623  //           JavaThread in Java code, and have stubs simply
1624  //           treat %g2 as a caller-save register, preserving it in a %lN.
1625  thread_key_t tk;
1626  if (thr_keycreate( &tk, NULL ) )
1627    fatal(err_msg("os::allocate_thread_local_storage: thr_keycreate failed "
1628                  "(%s)", strerror(errno)));
1629  return int(tk);
1630}
1631
1632void os::free_thread_local_storage(int index) {
1633  // %%% don't think we need anything here
1634  // if ( pthread_key_delete((pthread_key_t) tk) )
1635  //   fatal("os::free_thread_local_storage: pthread_key_delete failed");
1636}
1637
1638#define SMALLINT 32   // libthread allocate for tsd_common is a version specific
1639                      // small number - point is NO swap space available
1640void os::thread_local_storage_at_put(int index, void* value) {
1641  // %%% this is used only in threadLocalStorage.cpp
1642  if (thr_setspecific((thread_key_t)index, value)) {
1643    if (errno == ENOMEM) {
1644       vm_exit_out_of_memory(SMALLINT, "thr_setspecific: out of swap space");
1645    } else {
1646      fatal(err_msg("os::thread_local_storage_at_put: thr_setspecific failed "
1647                    "(%s)", strerror(errno)));
1648    }
1649  } else {
1650      ThreadLocalStorage::set_thread_in_slot ((Thread *) value) ;
1651  }
1652}
1653
1654// This function could be called before TLS is initialized, for example, when
1655// VM receives an async signal or when VM causes a fatal error during
1656// initialization. Return NULL if thr_getspecific() fails.
1657void* os::thread_local_storage_at(int index) {
1658  // %%% this is used only in threadLocalStorage.cpp
1659  void* r = NULL;
1660  return thr_getspecific((thread_key_t)index, &r) != 0 ? NULL : r;
1661}
1662
1663
1664// gethrtime can move backwards if read from one cpu and then a different cpu
1665// getTimeNanos is guaranteed to not move backward on Solaris
1666// local spinloop created as faster for a CAS on an int than
1667// a CAS on a 64bit jlong. Also Atomic::cmpxchg for jlong is not
1668// supported on sparc v8 or pre supports_cx8 intel boxes.
1669// oldgetTimeNanos for systems which do not support CAS on 64bit jlong
1670// i.e. sparc v8 and pre supports_cx8 (i486) intel boxes
1671inline hrtime_t oldgetTimeNanos() {
1672  int gotlock = LOCK_INVALID;
1673  hrtime_t newtime = gethrtime();
1674
1675  for (;;) {
1676// grab lock for max_hrtime
1677    int curlock = max_hrtime_lock;
1678    if (curlock & LOCK_BUSY)  continue;
1679    if (gotlock = Atomic::cmpxchg(LOCK_BUSY, &max_hrtime_lock, LOCK_FREE) != LOCK_FREE) continue;
1680    if (newtime > max_hrtime) {
1681      max_hrtime = newtime;
1682    } else {
1683      newtime = max_hrtime;
1684    }
1685    // release lock
1686    max_hrtime_lock = LOCK_FREE;
1687    return newtime;
1688  }
1689}
1690// gethrtime can move backwards if read from one cpu and then a different cpu
1691// getTimeNanos is guaranteed to not move backward on Solaris
1692inline hrtime_t getTimeNanos() {
1693  if (VM_Version::supports_cx8()) {
1694    const hrtime_t now = gethrtime();
1695    // Use atomic long load since 32-bit x86 uses 2 registers to keep long.
1696    const hrtime_t prev = Atomic::load((volatile jlong*)&max_hrtime);
1697    if (now <= prev)  return prev;   // same or retrograde time;
1698    const hrtime_t obsv = Atomic::cmpxchg(now, (volatile jlong*)&max_hrtime, prev);
1699    assert(obsv >= prev, "invariant");   // Monotonicity
1700    // If the CAS succeeded then we're done and return "now".
1701    // If the CAS failed and the observed value "obs" is >= now then
1702    // we should return "obs".  If the CAS failed and now > obs > prv then
1703    // some other thread raced this thread and installed a new value, in which case
1704    // we could either (a) retry the entire operation, (b) retry trying to install now
1705    // or (c) just return obs.  We use (c).   No loop is required although in some cases
1706    // we might discard a higher "now" value in deference to a slightly lower but freshly
1707    // installed obs value.   That's entirely benign -- it admits no new orderings compared
1708    // to (a) or (b) -- and greatly reduces coherence traffic.
1709    // We might also condition (c) on the magnitude of the delta between obs and now.
1710    // Avoiding excessive CAS operations to hot RW locations is critical.
1711    // See http://blogs.sun.com/dave/entry/cas_and_cache_trivia_invalidate
1712    return (prev == obsv) ? now : obsv ;
1713  } else {
1714    return oldgetTimeNanos();
1715  }
1716}
1717
1718// Time since start-up in seconds to a fine granularity.
1719// Used by VMSelfDestructTimer and the MemProfiler.
1720double os::elapsedTime() {
1721  return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1722}
1723
1724jlong os::elapsed_counter() {
1725  return (jlong)(getTimeNanos() - first_hrtime);
1726}
1727
1728jlong os::elapsed_frequency() {
1729   return hrtime_hz;
1730}
1731
1732// Return the real, user, and system times in seconds from an
1733// arbitrary fixed point in the past.
1734bool os::getTimesSecs(double* process_real_time,
1735                  double* process_user_time,
1736                  double* process_system_time) {
1737  struct tms ticks;
1738  clock_t real_ticks = times(&ticks);
1739
1740  if (real_ticks == (clock_t) (-1)) {
1741    return false;
1742  } else {
1743    double ticks_per_second = (double) clock_tics_per_sec;
1744    *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1745    *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1746    // For consistency return the real time from getTimeNanos()
1747    // converted to seconds.
1748    *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1749
1750    return true;
1751  }
1752}
1753
1754bool os::supports_vtime() { return true; }
1755
1756bool os::enable_vtime() {
1757  int fd = ::open("/proc/self/ctl", O_WRONLY);
1758  if (fd == -1)
1759    return false;
1760
1761  long cmd[] = { PCSET, PR_MSACCT };
1762  int res = ::write(fd, cmd, sizeof(long) * 2);
1763  ::close(fd);
1764  if (res != sizeof(long) * 2)
1765    return false;
1766
1767  return true;
1768}
1769
1770bool os::vtime_enabled() {
1771  int fd = ::open("/proc/self/status", O_RDONLY);
1772  if (fd == -1)
1773    return false;
1774
1775  pstatus_t status;
1776  int res = os::read(fd, (void*) &status, sizeof(pstatus_t));
1777  ::close(fd);
1778  if (res != sizeof(pstatus_t))
1779    return false;
1780
1781  return status.pr_flags & PR_MSACCT;
1782}
1783
1784double os::elapsedVTime() {
1785  return (double)gethrvtime() / (double)hrtime_hz;
1786}
1787
1788// Used internally for comparisons only
1789// getTimeMillis guaranteed to not move backwards on Solaris
1790jlong getTimeMillis() {
1791  jlong nanotime = getTimeNanos();
1792  return (jlong)(nanotime / NANOSECS_PER_MILLISEC);
1793}
1794
1795// Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1796jlong os::javaTimeMillis() {
1797  timeval t;
1798  if (gettimeofday( &t, NULL) == -1)
1799    fatal(err_msg("os::javaTimeMillis: gettimeofday (%s)", strerror(errno)));
1800  return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1801}
1802
1803jlong os::javaTimeNanos() {
1804  return (jlong)getTimeNanos();
1805}
1806
1807void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1808  info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1809  info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1810  info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1811  info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1812}
1813
1814char * os::local_time_string(char *buf, size_t buflen) {
1815  struct tm t;
1816  time_t long_time;
1817  time(&long_time);
1818  localtime_r(&long_time, &t);
1819  jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1820               t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1821               t.tm_hour, t.tm_min, t.tm_sec);
1822  return buf;
1823}
1824
1825// Note: os::shutdown() might be called very early during initialization, or
1826// called from signal handler. Before adding something to os::shutdown(), make
1827// sure it is async-safe and can handle partially initialized VM.
1828void os::shutdown() {
1829
1830  // allow PerfMemory to attempt cleanup of any persistent resources
1831  perfMemory_exit();
1832
1833  // needs to remove object in file system
1834  AttachListener::abort();
1835
1836  // flush buffered output, finish log files
1837  ostream_abort();
1838
1839  // Check for abort hook
1840  abort_hook_t abort_hook = Arguments::abort_hook();
1841  if (abort_hook != NULL) {
1842    abort_hook();
1843  }
1844}
1845
1846// Note: os::abort() might be called very early during initialization, or
1847// called from signal handler. Before adding something to os::abort(), make
1848// sure it is async-safe and can handle partially initialized VM.
1849void os::abort(bool dump_core) {
1850  os::shutdown();
1851  if (dump_core) {
1852#ifndef PRODUCT
1853    fdStream out(defaultStream::output_fd());
1854    out.print_raw("Current thread is ");
1855    char buf[16];
1856    jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1857    out.print_raw_cr(buf);
1858    out.print_raw_cr("Dumping core ...");
1859#endif
1860    ::abort(); // dump core (for debugging)
1861  }
1862
1863  ::exit(1);
1864}
1865
1866// Die immediately, no exit hook, no abort hook, no cleanup.
1867void os::die() {
1868  ::abort(); // dump core (for debugging)
1869}
1870
1871// unused
1872void os::set_error_file(const char *logfile) {}
1873
1874// DLL functions
1875
1876const char* os::dll_file_extension() { return ".so"; }
1877
1878// This must be hard coded because it's the system's temporary
1879// directory not the java application's temp directory, ala java.io.tmpdir.
1880const char* os::get_temp_directory() { return "/tmp"; }
1881
1882static bool file_exists(const char* filename) {
1883  struct stat statbuf;
1884  if (filename == NULL || strlen(filename) == 0) {
1885    return false;
1886  }
1887  return os::stat(filename, &statbuf) == 0;
1888}
1889
1890bool os::dll_build_name(char* buffer, size_t buflen,
1891                        const char* pname, const char* fname) {
1892  bool retval = false;
1893  const size_t pnamelen = pname ? strlen(pname) : 0;
1894
1895  // Return error on buffer overflow.
1896  if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1897    return retval;
1898  }
1899
1900  if (pnamelen == 0) {
1901    snprintf(buffer, buflen, "lib%s.so", fname);
1902    retval = true;
1903  } else if (strchr(pname, *os::path_separator()) != NULL) {
1904    int n;
1905    char** pelements = split_path(pname, &n);
1906    for (int i = 0 ; i < n ; i++) {
1907      // really shouldn't be NULL but what the heck, check can't hurt
1908      if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1909        continue; // skip the empty path values
1910      }
1911      snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1912      if (file_exists(buffer)) {
1913        retval = true;
1914        break;
1915      }
1916    }
1917    // release the storage
1918    for (int i = 0 ; i < n ; i++) {
1919      if (pelements[i] != NULL) {
1920        FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1921      }
1922    }
1923    if (pelements != NULL) {
1924      FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1925    }
1926  } else {
1927    snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1928    retval = true;
1929  }
1930  return retval;
1931}
1932
1933const char* os::get_current_directory(char *buf, int buflen) {
1934  return getcwd(buf, buflen);
1935}
1936
1937// check if addr is inside libjvm.so
1938bool os::address_is_in_vm(address addr) {
1939  static address libjvm_base_addr;
1940  Dl_info dlinfo;
1941
1942  if (libjvm_base_addr == NULL) {
1943    dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1944    libjvm_base_addr = (address)dlinfo.dli_fbase;
1945    assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1946  }
1947
1948  if (dladdr((void *)addr, &dlinfo)) {
1949    if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1950  }
1951
1952  return false;
1953}
1954
1955typedef int (*dladdr1_func_type) (void *, Dl_info *, void **, int);
1956static dladdr1_func_type dladdr1_func = NULL;
1957
1958bool os::dll_address_to_function_name(address addr, char *buf,
1959                                      int buflen, int * offset) {
1960  Dl_info dlinfo;
1961
1962  // dladdr1_func was initialized in os::init()
1963  if (dladdr1_func){
1964      // yes, we have dladdr1
1965
1966      // Support for dladdr1 is checked at runtime; it may be
1967      // available even if the vm is built on a machine that does
1968      // not have dladdr1 support.  Make sure there is a value for
1969      // RTLD_DL_SYMENT.
1970      #ifndef RTLD_DL_SYMENT
1971      #define RTLD_DL_SYMENT 1
1972      #endif
1973#ifdef _LP64
1974      Elf64_Sym * info;
1975#else
1976      Elf32_Sym * info;
1977#endif
1978      if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1979                       RTLD_DL_SYMENT)) {
1980        if ((char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1981          if (buf != NULL) {
1982            if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen))
1983              jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1984            }
1985            if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1986            return true;
1987        }
1988      }
1989      if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1990        if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1991           buf, buflen, offset, dlinfo.dli_fname)) {
1992          return true;
1993        }
1994      }
1995      if (buf != NULL) buf[0] = '\0';
1996      if (offset != NULL) *offset  = -1;
1997      return false;
1998  } else {
1999      // no, only dladdr is available
2000      if (dladdr((void *)addr, &dlinfo)) {
2001        if (buf != NULL) {
2002          if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen))
2003            jio_snprintf(buf, buflen, dlinfo.dli_sname);
2004        }
2005        if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
2006        return true;
2007      } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
2008        if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
2009          buf, buflen, offset, dlinfo.dli_fname)) {
2010          return true;
2011        }
2012      }
2013      if (buf != NULL) buf[0] = '\0';
2014      if (offset != NULL) *offset  = -1;
2015      return false;
2016  }
2017}
2018
2019bool os::dll_address_to_library_name(address addr, char* buf,
2020                                     int buflen, int* offset) {
2021  Dl_info dlinfo;
2022
2023  if (dladdr((void*)addr, &dlinfo)){
2024     if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
2025     if (offset) *offset = addr - (address)dlinfo.dli_fbase;
2026     return true;
2027  } else {
2028     if (buf) buf[0] = '\0';
2029     if (offset) *offset = -1;
2030     return false;
2031  }
2032}
2033
2034// Prints the names and full paths of all opened dynamic libraries
2035// for current process
2036void os::print_dll_info(outputStream * st) {
2037    Dl_info dli;
2038    void *handle;
2039    Link_map *map;
2040    Link_map *p;
2041
2042    st->print_cr("Dynamic libraries:"); st->flush();
2043
2044    if (!dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli)) {
2045        st->print_cr("Error: Cannot print dynamic libraries.");
2046        return;
2047    }
2048    handle = dlopen(dli.dli_fname, RTLD_LAZY);
2049    if (handle == NULL) {
2050        st->print_cr("Error: Cannot print dynamic libraries.");
2051        return;
2052    }
2053    dlinfo(handle, RTLD_DI_LINKMAP, &map);
2054    if (map == NULL) {
2055        st->print_cr("Error: Cannot print dynamic libraries.");
2056        return;
2057    }
2058
2059    while (map->l_prev != NULL)
2060        map = map->l_prev;
2061
2062    while (map != NULL) {
2063        st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
2064        map = map->l_next;
2065    }
2066
2067    dlclose(handle);
2068}
2069
2070  // Loads .dll/.so and
2071  // in case of error it checks if .dll/.so was built for the
2072  // same architecture as Hotspot is running on
2073
2074void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
2075{
2076  void * result= ::dlopen(filename, RTLD_LAZY);
2077  if (result != NULL) {
2078    // Successful loading
2079    return result;
2080  }
2081
2082  Elf32_Ehdr elf_head;
2083
2084  // Read system error message into ebuf
2085  // It may or may not be overwritten below
2086  ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2087  ebuf[ebuflen-1]='\0';
2088  int diag_msg_max_length=ebuflen-strlen(ebuf);
2089  char* diag_msg_buf=ebuf+strlen(ebuf);
2090
2091  if (diag_msg_max_length==0) {
2092    // No more space in ebuf for additional diagnostics message
2093    return NULL;
2094  }
2095
2096
2097  int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2098
2099  if (file_descriptor < 0) {
2100    // Can't open library, report dlerror() message
2101    return NULL;
2102  }
2103
2104  bool failed_to_read_elf_head=
2105    (sizeof(elf_head)!=
2106        (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2107
2108  ::close(file_descriptor);
2109  if (failed_to_read_elf_head) {
2110    // file i/o error - report dlerror() msg
2111    return NULL;
2112  }
2113
2114  typedef struct {
2115    Elf32_Half  code;         // Actual value as defined in elf.h
2116    Elf32_Half  compat_class; // Compatibility of archs at VM's sense
2117    char        elf_class;    // 32 or 64 bit
2118    char        endianess;    // MSB or LSB
2119    char*       name;         // String representation
2120  } arch_t;
2121
2122  static const arch_t arch_array[]={
2123    {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2124    {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2125    {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2126    {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2127    {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2128    {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2129    {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2130    {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2131    {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
2132    {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
2133  };
2134
2135  #if  (defined IA32)
2136    static  Elf32_Half running_arch_code=EM_386;
2137  #elif   (defined AMD64)
2138    static  Elf32_Half running_arch_code=EM_X86_64;
2139  #elif  (defined IA64)
2140    static  Elf32_Half running_arch_code=EM_IA_64;
2141  #elif  (defined __sparc) && (defined _LP64)
2142    static  Elf32_Half running_arch_code=EM_SPARCV9;
2143  #elif  (defined __sparc) && (!defined _LP64)
2144    static  Elf32_Half running_arch_code=EM_SPARC;
2145  #elif  (defined __powerpc64__)
2146    static  Elf32_Half running_arch_code=EM_PPC64;
2147  #elif  (defined __powerpc__)
2148    static  Elf32_Half running_arch_code=EM_PPC;
2149  #elif (defined ARM)
2150    static  Elf32_Half running_arch_code=EM_ARM;
2151  #else
2152    #error Method os::dll_load requires that one of following is defined:\
2153         IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
2154  #endif
2155
2156  // Identify compatability class for VM's architecture and library's architecture
2157  // Obtain string descriptions for architectures
2158
2159  arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2160  int running_arch_index=-1;
2161
2162  for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2163    if (running_arch_code == arch_array[i].code) {
2164      running_arch_index    = i;
2165    }
2166    if (lib_arch.code == arch_array[i].code) {
2167      lib_arch.compat_class = arch_array[i].compat_class;
2168      lib_arch.name         = arch_array[i].name;
2169    }
2170  }
2171
2172  assert(running_arch_index != -1,
2173    "Didn't find running architecture code (running_arch_code) in arch_array");
2174  if (running_arch_index == -1) {
2175    // Even though running architecture detection failed
2176    // we may still continue with reporting dlerror() message
2177    return NULL;
2178  }
2179
2180  if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2181    ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2182    return NULL;
2183  }
2184
2185  if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2186    ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2187    return NULL;
2188  }
2189
2190  if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2191    if ( lib_arch.name!=NULL ) {
2192      ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2193        " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2194        lib_arch.name, arch_array[running_arch_index].name);
2195    } else {
2196      ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2197      " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2198        lib_arch.code,
2199        arch_array[running_arch_index].name);
2200    }
2201  }
2202
2203  return NULL;
2204}
2205
2206void* os::dll_lookup(void* handle, const char* name) {
2207  return dlsym(handle, name);
2208}
2209
2210int os::stat(const char *path, struct stat *sbuf) {
2211  char pathbuf[MAX_PATH];
2212  if (strlen(path) > MAX_PATH - 1) {
2213    errno = ENAMETOOLONG;
2214    return -1;
2215  }
2216  os::native_path(strcpy(pathbuf, path));
2217  return ::stat(pathbuf, sbuf);
2218}
2219
2220static bool _print_ascii_file(const char* filename, outputStream* st) {
2221  int fd = ::open(filename, O_RDONLY);
2222  if (fd == -1) {
2223     return false;
2224  }
2225
2226  char buf[32];
2227  int bytes;
2228  while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2229    st->print_raw(buf, bytes);
2230  }
2231
2232  ::close(fd);
2233
2234  return true;
2235}
2236
2237void os::print_os_info_brief(outputStream* st) {
2238  os::Solaris::print_distro_info(st);
2239
2240  os::Posix::print_uname_info(st);
2241
2242  os::Solaris::print_libversion_info(st);
2243}
2244
2245void os::print_os_info(outputStream* st) {
2246  st->print("OS:");
2247
2248  os::Solaris::print_distro_info(st);
2249
2250  os::Posix::print_uname_info(st);
2251
2252  os::Solaris::print_libversion_info(st);
2253
2254  os::Posix::print_rlimit_info(st);
2255
2256  os::Posix::print_load_average(st);
2257}
2258
2259void os::Solaris::print_distro_info(outputStream* st) {
2260  if (!_print_ascii_file("/etc/release", st)) {
2261      st->print("Solaris");
2262    }
2263    st->cr();
2264}
2265
2266void os::Solaris::print_libversion_info(outputStream* st) {
2267  if (os::Solaris::T2_libthread()) {
2268    st->print("  (T2 libthread)");
2269  }
2270  else {
2271    st->print("  (T1 libthread)");
2272  }
2273  st->cr();
2274}
2275
2276static bool check_addr0(outputStream* st) {
2277  jboolean status = false;
2278  int fd = ::open("/proc/self/map",O_RDONLY);
2279  if (fd >= 0) {
2280    prmap_t p;
2281    while(::read(fd, &p, sizeof(p)) > 0) {
2282      if (p.pr_vaddr == 0x0) {
2283        st->print("Warning: Address: 0x%x, Size: %dK, ",p.pr_vaddr, p.pr_size/1024, p.pr_mapname);
2284        st->print("Mapped file: %s, ", p.pr_mapname[0] == '\0' ? "None" : p.pr_mapname);
2285        st->print("Access:");
2286        st->print("%s",(p.pr_mflags & MA_READ)  ? "r" : "-");
2287        st->print("%s",(p.pr_mflags & MA_WRITE) ? "w" : "-");
2288        st->print("%s",(p.pr_mflags & MA_EXEC)  ? "x" : "-");
2289        st->cr();
2290        status = true;
2291      }
2292      ::close(fd);
2293    }
2294  }
2295  return status;
2296}
2297
2298void os::pd_print_cpu_info(outputStream* st) {
2299  // Nothing to do for now.
2300}
2301
2302void os::print_memory_info(outputStream* st) {
2303  st->print("Memory:");
2304  st->print(" %dk page", os::vm_page_size()>>10);
2305  st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
2306  st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
2307  st->cr();
2308  (void) check_addr0(st);
2309}
2310
2311// Taken from /usr/include/sys/machsig.h  Supposed to be architecture specific
2312// but they're the same for all the solaris architectures that we support.
2313const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2314                          "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2315                          "ILL_COPROC", "ILL_BADSTK" };
2316
2317const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2318                          "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2319                          "FPE_FLTINV", "FPE_FLTSUB" };
2320
2321const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2322
2323const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2324
2325void os::print_siginfo(outputStream* st, void* siginfo) {
2326  st->print("siginfo:");
2327
2328  const int buflen = 100;
2329  char buf[buflen];
2330  siginfo_t *si = (siginfo_t*)siginfo;
2331  st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2332  char *err = strerror(si->si_errno);
2333  if (si->si_errno != 0 && err != NULL) {
2334    st->print("si_errno=%s", err);
2335  } else {
2336    st->print("si_errno=%d", si->si_errno);
2337  }
2338  const int c = si->si_code;
2339  assert(c > 0, "unexpected si_code");
2340  switch (si->si_signo) {
2341  case SIGILL:
2342    st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2343    st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2344    break;
2345  case SIGFPE:
2346    st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2347    st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2348    break;
2349  case SIGSEGV:
2350    st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2351    st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2352    break;
2353  case SIGBUS:
2354    st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2355    st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2356    break;
2357  default:
2358    st->print(", si_code=%d", si->si_code);
2359    // no si_addr
2360  }
2361
2362  if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2363      UseSharedSpaces) {
2364    FileMapInfo* mapinfo = FileMapInfo::current_info();
2365    if (mapinfo->is_in_shared_space(si->si_addr)) {
2366      st->print("\n\nError accessing class data sharing archive."   \
2367                " Mapped file inaccessible during execution, "      \
2368                " possible disk/network problem.");
2369    }
2370  }
2371  st->cr();
2372}
2373
2374// Moved from whole group, because we need them here for diagnostic
2375// prints.
2376#define OLDMAXSIGNUM 32
2377static int Maxsignum = 0;
2378static int *ourSigFlags = NULL;
2379
2380extern "C" void sigINTRHandler(int, siginfo_t*, void*);
2381
2382int os::Solaris::get_our_sigflags(int sig) {
2383  assert(ourSigFlags!=NULL, "signal data structure not initialized");
2384  assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2385  return ourSigFlags[sig];
2386}
2387
2388void os::Solaris::set_our_sigflags(int sig, int flags) {
2389  assert(ourSigFlags!=NULL, "signal data structure not initialized");
2390  assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2391  ourSigFlags[sig] = flags;
2392}
2393
2394
2395static const char* get_signal_handler_name(address handler,
2396                                           char* buf, int buflen) {
2397  int offset;
2398  bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
2399  if (found) {
2400    // skip directory names
2401    const char *p1, *p2;
2402    p1 = buf;
2403    size_t len = strlen(os::file_separator());
2404    while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
2405    jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
2406  } else {
2407    jio_snprintf(buf, buflen, PTR_FORMAT, handler);
2408  }
2409  return buf;
2410}
2411
2412static void print_signal_handler(outputStream* st, int sig,
2413                                  char* buf, size_t buflen) {
2414  struct sigaction sa;
2415
2416  sigaction(sig, NULL, &sa);
2417
2418  st->print("%s: ", os::exception_name(sig, buf, buflen));
2419
2420  address handler = (sa.sa_flags & SA_SIGINFO)
2421                  ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
2422                  : CAST_FROM_FN_PTR(address, sa.sa_handler);
2423
2424  if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
2425    st->print("SIG_DFL");
2426  } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
2427    st->print("SIG_IGN");
2428  } else {
2429    st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
2430  }
2431
2432  st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
2433
2434  address rh = VMError::get_resetted_sighandler(sig);
2435  // May be, handler was resetted by VMError?
2436  if(rh != NULL) {
2437    handler = rh;
2438    sa.sa_flags = VMError::get_resetted_sigflags(sig);
2439  }
2440
2441  st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
2442
2443  // Check: is it our handler?
2444  if(handler == CAST_FROM_FN_PTR(address, signalHandler) ||
2445     handler == CAST_FROM_FN_PTR(address, sigINTRHandler)) {
2446    // It is our signal handler
2447    // check for flags
2448    if(sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2449      st->print(
2450        ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2451        os::Solaris::get_our_sigflags(sig));
2452    }
2453  }
2454  st->cr();
2455}
2456
2457void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2458  st->print_cr("Signal Handlers:");
2459  print_signal_handler(st, SIGSEGV, buf, buflen);
2460  print_signal_handler(st, SIGBUS , buf, buflen);
2461  print_signal_handler(st, SIGFPE , buf, buflen);
2462  print_signal_handler(st, SIGPIPE, buf, buflen);
2463  print_signal_handler(st, SIGXFSZ, buf, buflen);
2464  print_signal_handler(st, SIGILL , buf, buflen);
2465  print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2466  print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2467  print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2468  print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2469  print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2470  print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2471  print_signal_handler(st, os::Solaris::SIGinterrupt(), buf, buflen);
2472  print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2473}
2474
2475static char saved_jvm_path[MAXPATHLEN] = { 0 };
2476
2477// Find the full path to the current module, libjvm.so
2478void os::jvm_path(char *buf, jint buflen) {
2479  // Error checking.
2480  if (buflen < MAXPATHLEN) {
2481    assert(false, "must use a large-enough buffer");
2482    buf[0] = '\0';
2483    return;
2484  }
2485  // Lazy resolve the path to current module.
2486  if (saved_jvm_path[0] != 0) {
2487    strcpy(buf, saved_jvm_path);
2488    return;
2489  }
2490
2491  Dl_info dlinfo;
2492  int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2493  assert(ret != 0, "cannot locate libjvm");
2494  realpath((char *)dlinfo.dli_fname, buf);
2495
2496  if (Arguments::created_by_gamma_launcher()) {
2497    // Support for the gamma launcher.  Typical value for buf is
2498    // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2499    // the right place in the string, then assume we are installed in a JDK and
2500    // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2501    // up the path so it looks like libjvm.so is installed there (append a
2502    // fake suffix hotspot/libjvm.so).
2503    const char *p = buf + strlen(buf) - 1;
2504    for (int count = 0; p > buf && count < 5; ++count) {
2505      for (--p; p > buf && *p != '/'; --p)
2506        /* empty */ ;
2507    }
2508
2509    if (strncmp(p, "/jre/lib/", 9) != 0) {
2510      // Look for JAVA_HOME in the environment.
2511      char* java_home_var = ::getenv("JAVA_HOME");
2512      if (java_home_var != NULL && java_home_var[0] != 0) {
2513        char cpu_arch[12];
2514        char* jrelib_p;
2515        int   len;
2516        sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2517#ifdef _LP64
2518        // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2519        if (strcmp(cpu_arch, "sparc") == 0) {
2520          strcat(cpu_arch, "v9");
2521        } else if (strcmp(cpu_arch, "i386") == 0) {
2522          strcpy(cpu_arch, "amd64");
2523        }
2524#endif
2525        // Check the current module name "libjvm.so".
2526        p = strrchr(buf, '/');
2527        assert(strstr(p, "/libjvm") == p, "invalid library name");
2528
2529        realpath(java_home_var, buf);
2530        // determine if this is a legacy image or modules image
2531        // modules image doesn't have "jre" subdirectory
2532        len = strlen(buf);
2533        jrelib_p = buf + len;
2534        snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2535        if (0 != access(buf, F_OK)) {
2536          snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2537        }
2538
2539        if (0 == access(buf, F_OK)) {
2540          // Use current module name "libjvm.so"
2541          len = strlen(buf);
2542          snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2543        } else {
2544          // Go back to path of .so
2545          realpath((char *)dlinfo.dli_fname, buf);
2546        }
2547      }
2548    }
2549  }
2550
2551  strcpy(saved_jvm_path, buf);
2552}
2553
2554
2555void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2556  // no prefix required, not even "_"
2557}
2558
2559
2560void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2561  // no suffix required
2562}
2563
2564// This method is a copy of JDK's sysGetLastErrorString
2565// from src/solaris/hpi/src/system_md.c
2566
2567size_t os::lasterror(char *buf, size_t len) {
2568
2569  if (errno == 0)  return 0;
2570
2571  const char *s = ::strerror(errno);
2572  size_t n = ::strlen(s);
2573  if (n >= len) {
2574    n = len - 1;
2575  }
2576  ::strncpy(buf, s, n);
2577  buf[n] = '\0';
2578  return n;
2579}
2580
2581
2582// sun.misc.Signal
2583
2584extern "C" {
2585  static void UserHandler(int sig, void *siginfo, void *context) {
2586    // Ctrl-C is pressed during error reporting, likely because the error
2587    // handler fails to abort. Let VM die immediately.
2588    if (sig == SIGINT && is_error_reported()) {
2589       os::die();
2590    }
2591
2592    os::signal_notify(sig);
2593    // We do not need to reinstate the signal handler each time...
2594  }
2595}
2596
2597void* os::user_handler() {
2598  return CAST_FROM_FN_PTR(void*, UserHandler);
2599}
2600
2601extern "C" {
2602  typedef void (*sa_handler_t)(int);
2603  typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2604}
2605
2606void* os::signal(int signal_number, void* handler) {
2607  struct sigaction sigAct, oldSigAct;
2608  sigfillset(&(sigAct.sa_mask));
2609  sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2610  sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2611
2612  if (sigaction(signal_number, &sigAct, &oldSigAct))
2613    // -1 means registration failed
2614    return (void *)-1;
2615
2616  return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2617}
2618
2619void os::signal_raise(int signal_number) {
2620  raise(signal_number);
2621}
2622
2623/*
2624 * The following code is moved from os.cpp for making this
2625 * code platform specific, which it is by its very nature.
2626 */
2627
2628// a counter for each possible signal value
2629static int Sigexit = 0;
2630static int Maxlibjsigsigs;
2631static jint *pending_signals = NULL;
2632static int *preinstalled_sigs = NULL;
2633static struct sigaction *chainedsigactions = NULL;
2634static sema_t sig_sem;
2635typedef int (*version_getting_t)();
2636version_getting_t os::Solaris::get_libjsig_version = NULL;
2637static int libjsigversion = NULL;
2638
2639int os::sigexitnum_pd() {
2640  assert(Sigexit > 0, "signal memory not yet initialized");
2641  return Sigexit;
2642}
2643
2644void os::Solaris::init_signal_mem() {
2645  // Initialize signal structures
2646  Maxsignum = SIGRTMAX;
2647  Sigexit = Maxsignum+1;
2648  assert(Maxsignum >0, "Unable to obtain max signal number");
2649
2650  Maxlibjsigsigs = Maxsignum;
2651
2652  // pending_signals has one int per signal
2653  // The additional signal is for SIGEXIT - exit signal to signal_thread
2654  pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2655  memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2656
2657  if (UseSignalChaining) {
2658     chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2659       * (Maxsignum + 1), mtInternal);
2660     memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2661     preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2662     memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2663  }
2664  ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1 ), mtInternal);
2665  memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2666}
2667
2668void os::signal_init_pd() {
2669  int ret;
2670
2671  ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2672  assert(ret == 0, "sema_init() failed");
2673}
2674
2675void os::signal_notify(int signal_number) {
2676  int ret;
2677
2678  Atomic::inc(&pending_signals[signal_number]);
2679  ret = ::sema_post(&sig_sem);
2680  assert(ret == 0, "sema_post() failed");
2681}
2682
2683static int check_pending_signals(bool wait_for_signal) {
2684  int ret;
2685  while (true) {
2686    for (int i = 0; i < Sigexit + 1; i++) {
2687      jint n = pending_signals[i];
2688      if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2689        return i;
2690      }
2691    }
2692    if (!wait_for_signal) {
2693      return -1;
2694    }
2695    JavaThread *thread = JavaThread::current();
2696    ThreadBlockInVM tbivm(thread);
2697
2698    bool threadIsSuspended;
2699    do {
2700      thread->set_suspend_equivalent();
2701      // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2702      while((ret = ::sema_wait(&sig_sem)) == EINTR)
2703          ;
2704      assert(ret == 0, "sema_wait() failed");
2705
2706      // were we externally suspended while we were waiting?
2707      threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2708      if (threadIsSuspended) {
2709        //
2710        // The semaphore has been incremented, but while we were waiting
2711        // another thread suspended us. We don't want to continue running
2712        // while suspended because that would surprise the thread that
2713        // suspended us.
2714        //
2715        ret = ::sema_post(&sig_sem);
2716        assert(ret == 0, "sema_post() failed");
2717
2718        thread->java_suspend_self();
2719      }
2720    } while (threadIsSuspended);
2721  }
2722}
2723
2724int os::signal_lookup() {
2725  return check_pending_signals(false);
2726}
2727
2728int os::signal_wait() {
2729  return check_pending_signals(true);
2730}
2731
2732////////////////////////////////////////////////////////////////////////////////
2733// Virtual Memory
2734
2735static int page_size = -1;
2736
2737// The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2738// clear this var if support is not available.
2739static bool has_map_align = true;
2740
2741int os::vm_page_size() {
2742  assert(page_size != -1, "must call os::init");
2743  return page_size;
2744}
2745
2746// Solaris allocates memory by pages.
2747int os::vm_allocation_granularity() {
2748  assert(page_size != -1, "must call os::init");
2749  return page_size;
2750}
2751
2752bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2753  int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2754  size_t size = bytes;
2755  char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2756  if (res != NULL) {
2757    if (UseNUMAInterleaving) {
2758      numa_make_global(addr, bytes);
2759    }
2760    return true;
2761  }
2762  return false;
2763}
2764
2765bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2766                       bool exec) {
2767  if (commit_memory(addr, bytes, exec)) {
2768    if (UseMPSS && alignment_hint > (size_t)vm_page_size()) {
2769      // If the large page size has been set and the VM
2770      // is using large pages, use the large page size
2771      // if it is smaller than the alignment hint. This is
2772      // a case where the VM wants to use a larger alignment size
2773      // for its own reasons but still want to use large pages
2774      // (which is what matters to setting the mpss range.
2775      size_t page_size = 0;
2776      if (large_page_size() < alignment_hint) {
2777        assert(UseLargePages, "Expected to be here for large page use only");
2778        page_size = large_page_size();
2779      } else {
2780        // If the alignment hint is less than the large page
2781        // size, the VM wants a particular alignment (thus the hint)
2782        // for internal reasons.  Try to set the mpss range using
2783        // the alignment_hint.
2784        page_size = alignment_hint;
2785      }
2786      // Since this is a hint, ignore any failures.
2787      (void)Solaris::set_mpss_range(addr, bytes, page_size);
2788    }
2789    return true;
2790  }
2791  return false;
2792}
2793
2794// Uncommit the pages in a specified region.
2795void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2796  if (madvise(addr, bytes, MADV_FREE) < 0) {
2797    debug_only(warning("MADV_FREE failed."));
2798    return;
2799  }
2800}
2801
2802bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2803  return os::commit_memory(addr, size);
2804}
2805
2806bool os::remove_stack_guard_pages(char* addr, size_t size) {
2807  return os::uncommit_memory(addr, size);
2808}
2809
2810// Change the page size in a given range.
2811void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2812  assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2813  assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2814  if (UseLargePages && UseMPSS) {
2815    Solaris::set_mpss_range(addr, bytes, alignment_hint);
2816  }
2817}
2818
2819// Tell the OS to make the range local to the first-touching LWP
2820void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2821  assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2822  if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2823    debug_only(warning("MADV_ACCESS_LWP failed."));
2824  }
2825}
2826
2827// Tell the OS that this range would be accessed from different LWPs.
2828void os::numa_make_global(char *addr, size_t bytes) {
2829  assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2830  if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2831    debug_only(warning("MADV_ACCESS_MANY failed."));
2832  }
2833}
2834
2835// Get the number of the locality groups.
2836size_t os::numa_get_groups_num() {
2837  size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2838  return n != -1 ? n : 1;
2839}
2840
2841// Get a list of leaf locality groups. A leaf lgroup is group that
2842// doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2843// board. An LWP is assigned to one of these groups upon creation.
2844size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2845   if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2846     ids[0] = 0;
2847     return 1;
2848   }
2849   int result_size = 0, top = 1, bottom = 0, cur = 0;
2850   for (int k = 0; k < size; k++) {
2851     int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2852                                    (Solaris::lgrp_id_t*)&ids[top], size - top);
2853     if (r == -1) {
2854       ids[0] = 0;
2855       return 1;
2856     }
2857     if (!r) {
2858       // That's a leaf node.
2859       assert (bottom <= cur, "Sanity check");
2860       // Check if the node has memory
2861       if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2862                                   NULL, 0, LGRP_RSRC_MEM) > 0) {
2863         ids[bottom++] = ids[cur];
2864       }
2865     }
2866     top += r;
2867     cur++;
2868   }
2869   if (bottom == 0) {
2870     // Handle a situation, when the OS reports no memory available.
2871     // Assume UMA architecture.
2872     ids[0] = 0;
2873     return 1;
2874   }
2875   return bottom;
2876}
2877
2878// Detect the topology change. Typically happens during CPU plugging-unplugging.
2879bool os::numa_topology_changed() {
2880  int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2881  if (is_stale != -1 && is_stale) {
2882    Solaris::lgrp_fini(Solaris::lgrp_cookie());
2883    Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2884    assert(c != 0, "Failure to initialize LGRP API");
2885    Solaris::set_lgrp_cookie(c);
2886    return true;
2887  }
2888  return false;
2889}
2890
2891// Get the group id of the current LWP.
2892int os::numa_get_group_id() {
2893  int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2894  if (lgrp_id == -1) {
2895    return 0;
2896  }
2897  const int size = os::numa_get_groups_num();
2898  int *ids = (int*)alloca(size * sizeof(int));
2899
2900  // Get the ids of all lgroups with memory; r is the count.
2901  int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2902                                  (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2903  if (r <= 0) {
2904    return 0;
2905  }
2906  return ids[os::random() % r];
2907}
2908
2909// Request information about the page.
2910bool os::get_page_info(char *start, page_info* info) {
2911  const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2912  uint64_t addr = (uintptr_t)start;
2913  uint64_t outdata[2];
2914  uint_t validity = 0;
2915
2916  if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2917    return false;
2918  }
2919
2920  info->size = 0;
2921  info->lgrp_id = -1;
2922
2923  if ((validity & 1) != 0) {
2924    if ((validity & 2) != 0) {
2925      info->lgrp_id = outdata[0];
2926    }
2927    if ((validity & 4) != 0) {
2928      info->size = outdata[1];
2929    }
2930    return true;
2931  }
2932  return false;
2933}
2934
2935// Scan the pages from start to end until a page different than
2936// the one described in the info parameter is encountered.
2937char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2938  const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2939  const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2940  uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT];
2941  uint_t validity[MAX_MEMINFO_CNT];
2942
2943  size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2944  uint64_t p = (uint64_t)start;
2945  while (p < (uint64_t)end) {
2946    addrs[0] = p;
2947    size_t addrs_count = 1;
2948    while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] < (uint64_t)end) {
2949      addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2950      addrs_count++;
2951    }
2952
2953    if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2954      return NULL;
2955    }
2956
2957    size_t i = 0;
2958    for (; i < addrs_count; i++) {
2959      if ((validity[i] & 1) != 0) {
2960        if ((validity[i] & 4) != 0) {
2961          if (outdata[types * i + 1] != page_expected->size) {
2962            break;
2963          }
2964        } else
2965          if (page_expected->size != 0) {
2966            break;
2967          }
2968
2969        if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2970          if (outdata[types * i] != page_expected->lgrp_id) {
2971            break;
2972          }
2973        }
2974      } else {
2975        return NULL;
2976      }
2977    }
2978
2979    if (i != addrs_count) {
2980      if ((validity[i] & 2) != 0) {
2981        page_found->lgrp_id = outdata[types * i];
2982      } else {
2983        page_found->lgrp_id = -1;
2984      }
2985      if ((validity[i] & 4) != 0) {
2986        page_found->size = outdata[types * i + 1];
2987      } else {
2988        page_found->size = 0;
2989      }
2990      return (char*)addrs[i];
2991    }
2992
2993    p = addrs[addrs_count - 1] + page_size;
2994  }
2995  return end;
2996}
2997
2998bool os::pd_uncommit_memory(char* addr, size_t bytes) {
2999  size_t size = bytes;
3000  // Map uncommitted pages PROT_NONE so we fail early if we touch an
3001  // uncommitted page. Otherwise, the read/write might succeed if we
3002  // have enough swap space to back the physical page.
3003  return
3004    NULL != Solaris::mmap_chunk(addr, size,
3005                                MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
3006                                PROT_NONE);
3007}
3008
3009char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
3010  char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
3011
3012  if (b == MAP_FAILED) {
3013    return NULL;
3014  }
3015  return b;
3016}
3017
3018char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes, size_t alignment_hint, bool fixed) {
3019  char* addr = requested_addr;
3020  int flags = MAP_PRIVATE | MAP_NORESERVE;
3021
3022  assert(!(fixed && (alignment_hint > 0)), "alignment hint meaningless with fixed mmap");
3023
3024  if (fixed) {
3025    flags |= MAP_FIXED;
3026  } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
3027    flags |= MAP_ALIGN;
3028    addr = (char*) alignment_hint;
3029  }
3030
3031  // Map uncommitted pages PROT_NONE so we fail early if we touch an
3032  // uncommitted page. Otherwise, the read/write might succeed if we
3033  // have enough swap space to back the physical page.
3034  return mmap_chunk(addr, bytes, flags, PROT_NONE);
3035}
3036
3037char* os::pd_reserve_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
3038  char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint, (requested_addr != NULL));
3039
3040  guarantee(requested_addr == NULL || requested_addr == addr,
3041            "OS failed to return requested mmap address.");
3042  return addr;
3043}
3044
3045// Reserve memory at an arbitrary address, only if that area is
3046// available (and not reserved for something else).
3047
3048char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3049  const int max_tries = 10;
3050  char* base[max_tries];
3051  size_t size[max_tries];
3052
3053  // Solaris adds a gap between mmap'ed regions.  The size of the gap
3054  // is dependent on the requested size and the MMU.  Our initial gap
3055  // value here is just a guess and will be corrected later.
3056  bool had_top_overlap = false;
3057  bool have_adjusted_gap = false;
3058  size_t gap = 0x400000;
3059
3060  // Assert only that the size is a multiple of the page size, since
3061  // that's all that mmap requires, and since that's all we really know
3062  // about at this low abstraction level.  If we need higher alignment,
3063  // we can either pass an alignment to this method or verify alignment
3064  // in one of the methods further up the call chain.  See bug 5044738.
3065  assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3066
3067  // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
3068  // Give it a try, if the kernel honors the hint we can return immediately.
3069  char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
3070
3071  volatile int err = errno;
3072  if (addr == requested_addr) {
3073    return addr;
3074  } else if (addr != NULL) {
3075    pd_unmap_memory(addr, bytes);
3076  }
3077
3078  if (PrintMiscellaneous && Verbose) {
3079    char buf[256];
3080    buf[0] = '\0';
3081    if (addr == NULL) {
3082      jio_snprintf(buf, sizeof(buf), ": %s", strerror(err));
3083    }
3084    warning("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
3085            PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
3086            "%s", bytes, requested_addr, addr, buf);
3087  }
3088
3089  // Address hint method didn't work.  Fall back to the old method.
3090  // In theory, once SNV becomes our oldest supported platform, this
3091  // code will no longer be needed.
3092  //
3093  // Repeatedly allocate blocks until the block is allocated at the
3094  // right spot. Give up after max_tries.
3095  int i;
3096  for (i = 0; i < max_tries; ++i) {
3097    base[i] = reserve_memory(bytes);
3098
3099    if (base[i] != NULL) {
3100      // Is this the block we wanted?
3101      if (base[i] == requested_addr) {
3102        size[i] = bytes;
3103        break;
3104      }
3105
3106      // check that the gap value is right
3107      if (had_top_overlap && !have_adjusted_gap) {
3108        size_t actual_gap = base[i-1] - base[i] - bytes;
3109        if (gap != actual_gap) {
3110          // adjust the gap value and retry the last 2 allocations
3111          assert(i > 0, "gap adjustment code problem");
3112          have_adjusted_gap = true;  // adjust the gap only once, just in case
3113          gap = actual_gap;
3114          if (PrintMiscellaneous && Verbose) {
3115            warning("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
3116          }
3117          unmap_memory(base[i], bytes);
3118          unmap_memory(base[i-1], size[i-1]);
3119          i-=2;
3120          continue;
3121        }
3122      }
3123
3124      // Does this overlap the block we wanted? Give back the overlapped
3125      // parts and try again.
3126      //
3127      // There is still a bug in this code: if top_overlap == bytes,
3128      // the overlap is offset from requested region by the value of gap.
3129      // In this case giving back the overlapped part will not work,
3130      // because we'll give back the entire block at base[i] and
3131      // therefore the subsequent allocation will not generate a new gap.
3132      // This could be fixed with a new algorithm that used larger
3133      // or variable size chunks to find the requested region -
3134      // but such a change would introduce additional complications.
3135      // It's rare enough that the planets align for this bug,
3136      // so we'll just wait for a fix for 6204603/5003415 which
3137      // will provide a mmap flag to allow us to avoid this business.
3138
3139      size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3140      if (top_overlap >= 0 && top_overlap < bytes) {
3141        had_top_overlap = true;
3142        unmap_memory(base[i], top_overlap);
3143        base[i] += top_overlap;
3144        size[i] = bytes - top_overlap;
3145      } else {
3146        size_t bottom_overlap = base[i] + bytes - requested_addr;
3147        if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3148          if (PrintMiscellaneous && Verbose && bottom_overlap == 0) {
3149            warning("attempt_reserve_memory_at: possible alignment bug");
3150          }
3151          unmap_memory(requested_addr, bottom_overlap);
3152          size[i] = bytes - bottom_overlap;
3153        } else {
3154          size[i] = bytes;
3155        }
3156      }
3157    }
3158  }
3159
3160  // Give back the unused reserved pieces.
3161
3162  for (int j = 0; j < i; ++j) {
3163    if (base[j] != NULL) {
3164      unmap_memory(base[j], size[j]);
3165    }
3166  }
3167
3168  return (i < max_tries) ? requested_addr : NULL;
3169}
3170
3171bool os::pd_release_memory(char* addr, size_t bytes) {
3172  size_t size = bytes;
3173  return munmap(addr, size) == 0;
3174}
3175
3176static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
3177  assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
3178         "addr must be page aligned");
3179  int retVal = mprotect(addr, bytes, prot);
3180  return retVal == 0;
3181}
3182
3183// Protect memory (Used to pass readonly pages through
3184// JNI GetArray<type>Elements with empty arrays.)
3185// Also, used for serialization page and for compressed oops null pointer
3186// checking.
3187bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3188                        bool is_committed) {
3189  unsigned int p = 0;
3190  switch (prot) {
3191  case MEM_PROT_NONE: p = PROT_NONE; break;
3192  case MEM_PROT_READ: p = PROT_READ; break;
3193  case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3194  case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3195  default:
3196    ShouldNotReachHere();
3197  }
3198  // is_committed is unused.
3199  return solaris_mprotect(addr, bytes, p);
3200}
3201
3202// guard_memory and unguard_memory only happens within stack guard pages.
3203// Since ISM pertains only to the heap, guard and unguard memory should not
3204/// happen with an ISM region.
3205bool os::guard_memory(char* addr, size_t bytes) {
3206  return solaris_mprotect(addr, bytes, PROT_NONE);
3207}
3208
3209bool os::unguard_memory(char* addr, size_t bytes) {
3210  return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
3211}
3212
3213// Large page support
3214
3215// UseLargePages is the master flag to enable/disable large page memory.
3216// UseMPSS and UseISM are supported for compatibility reasons. Their combined
3217// effects can be described in the following table:
3218//
3219// UseLargePages UseMPSS UseISM
3220//    false         *       *   => UseLargePages is the master switch, turning
3221//                                 it off will turn off both UseMPSS and
3222//                                 UseISM. VM will not use large page memory
3223//                                 regardless the settings of UseMPSS/UseISM.
3224//     true      false    false => Unless future Solaris provides other
3225//                                 mechanism to use large page memory, this
3226//                                 combination is equivalent to -UseLargePages,
3227//                                 VM will not use large page memory
3228//     true      true     false => JVM will use MPSS for large page memory.
3229//                                 This is the default behavior.
3230//     true      false    true  => JVM will use ISM for large page memory.
3231//     true      true     true  => JVM will use ISM if it is available.
3232//                                 Otherwise, JVM will fall back to MPSS.
3233//                                 Becaues ISM is now available on all
3234//                                 supported Solaris versions, this combination
3235//                                 is equivalent to +UseISM -UseMPSS.
3236
3237static size_t _large_page_size = 0;
3238
3239bool os::Solaris::ism_sanity_check(bool warn, size_t * page_size) {
3240  // x86 uses either 2M or 4M page, depending on whether PAE (Physical Address
3241  // Extensions) mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. Sparc
3242  // can support multiple page sizes.
3243
3244  // Don't bother to probe page size because getpagesizes() comes with MPSS.
3245  // ISM is only recommended on old Solaris where there is no MPSS support.
3246  // Simply choose a conservative value as default.
3247  *page_size = LargePageSizeInBytes ? LargePageSizeInBytes :
3248               SPARC_ONLY(4 * M) IA32_ONLY(4 * M) AMD64_ONLY(2 * M)
3249               ARM_ONLY(2 * M);
3250
3251  // ISM is available on all supported Solaris versions
3252  return true;
3253}
3254
3255// Insertion sort for small arrays (descending order).
3256static void insertion_sort_descending(size_t* array, int len) {
3257  for (int i = 0; i < len; i++) {
3258    size_t val = array[i];
3259    for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
3260      size_t tmp = array[key];
3261      array[key] = array[key - 1];
3262      array[key - 1] = tmp;
3263    }
3264  }
3265}
3266
3267bool os::Solaris::mpss_sanity_check(bool warn, size_t * page_size) {
3268  const unsigned int usable_count = VM_Version::page_size_count();
3269  if (usable_count == 1) {
3270    return false;
3271  }
3272
3273  // Find the right getpagesizes interface.  When solaris 11 is the minimum
3274  // build platform, getpagesizes() (without the '2') can be called directly.
3275  typedef int (*gps_t)(size_t[], int);
3276  gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
3277  if (gps_func == NULL) {
3278    gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
3279    if (gps_func == NULL) {
3280      if (warn) {
3281        warning("MPSS is not supported by the operating system.");
3282      }
3283      return false;
3284    }
3285  }
3286
3287  // Fill the array of page sizes.
3288  int n = (*gps_func)(_page_sizes, page_sizes_max);
3289  assert(n > 0, "Solaris bug?");
3290
3291  if (n == page_sizes_max) {
3292    // Add a sentinel value (necessary only if the array was completely filled
3293    // since it is static (zeroed at initialization)).
3294    _page_sizes[--n] = 0;
3295    DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
3296  }
3297  assert(_page_sizes[n] == 0, "missing sentinel");
3298  trace_page_sizes("available page sizes", _page_sizes, n);
3299
3300  if (n == 1) return false;     // Only one page size available.
3301
3302  // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
3303  // select up to usable_count elements.  First sort the array, find the first
3304  // acceptable value, then copy the usable sizes to the top of the array and
3305  // trim the rest.  Make sure to include the default page size :-).
3306  //
3307  // A better policy could get rid of the 4M limit by taking the sizes of the
3308  // important VM memory regions (java heap and possibly the code cache) into
3309  // account.
3310  insertion_sort_descending(_page_sizes, n);
3311  const size_t size_limit =
3312    FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
3313  int beg;
3314  for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */ ;
3315  const int end = MIN2((int)usable_count, n) - 1;
3316  for (int cur = 0; cur < end; ++cur, ++beg) {
3317    _page_sizes[cur] = _page_sizes[beg];
3318  }
3319  _page_sizes[end] = vm_page_size();
3320  _page_sizes[end + 1] = 0;
3321
3322  if (_page_sizes[end] > _page_sizes[end - 1]) {
3323    // Default page size is not the smallest; sort again.
3324    insertion_sort_descending(_page_sizes, end + 1);
3325  }
3326  *page_size = _page_sizes[0];
3327
3328  trace_page_sizes("usable page sizes", _page_sizes, end + 1);
3329  return true;
3330}
3331
3332void os::large_page_init() {
3333  if (!UseLargePages) {
3334    UseISM = false;
3335    UseMPSS = false;
3336    return;
3337  }
3338
3339  // print a warning if any large page related flag is specified on command line
3340  bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
3341                         !FLAG_IS_DEFAULT(UseISM)               ||
3342                         !FLAG_IS_DEFAULT(UseMPSS)              ||
3343                         !FLAG_IS_DEFAULT(LargePageSizeInBytes);
3344  UseISM = UseISM &&
3345           Solaris::ism_sanity_check(warn_on_failure, &_large_page_size);
3346  if (UseISM) {
3347    // ISM disables MPSS to be compatible with old JDK behavior
3348    UseMPSS = false;
3349    _page_sizes[0] = _large_page_size;
3350    _page_sizes[1] = vm_page_size();
3351  }
3352
3353  UseMPSS = UseMPSS &&
3354            Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
3355
3356  UseLargePages = UseISM || UseMPSS;
3357}
3358
3359bool os::Solaris::set_mpss_range(caddr_t start, size_t bytes, size_t align) {
3360  // Signal to OS that we want large pages for addresses
3361  // from addr, addr + bytes
3362  struct memcntl_mha mpss_struct;
3363  mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
3364  mpss_struct.mha_pagesize = align;
3365  mpss_struct.mha_flags = 0;
3366  if (memcntl(start, bytes, MC_HAT_ADVISE,
3367              (caddr_t) &mpss_struct, 0, 0) < 0) {
3368    debug_only(warning("Attempt to use MPSS failed."));
3369    return false;
3370  }
3371  return true;
3372}
3373
3374char* os::reserve_memory_special(size_t size, char* addr, bool exec) {
3375  // "exec" is passed in but not used.  Creating the shared image for
3376  // the code cache doesn't have an SHM_X executable permission to check.
3377  assert(UseLargePages && UseISM, "only for ISM large pages");
3378
3379  char* retAddr = NULL;
3380  int shmid;
3381  key_t ismKey;
3382
3383  bool warn_on_failure = UseISM &&
3384                        (!FLAG_IS_DEFAULT(UseLargePages)         ||
3385                         !FLAG_IS_DEFAULT(UseISM)                ||
3386                         !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3387                        );
3388  char msg[128];
3389
3390  ismKey = IPC_PRIVATE;
3391
3392  // Create a large shared memory region to attach to based on size.
3393  // Currently, size is the total size of the heap
3394  shmid = shmget(ismKey, size, SHM_R | SHM_W | IPC_CREAT);
3395  if (shmid == -1){
3396     if (warn_on_failure) {
3397       jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3398       warning(msg);
3399     }
3400     return NULL;
3401  }
3402
3403  // Attach to the region
3404  retAddr = (char *) shmat(shmid, 0, SHM_SHARE_MMU | SHM_R | SHM_W);
3405  int err = errno;
3406
3407  // Remove shmid. If shmat() is successful, the actual shared memory segment
3408  // will be deleted when it's detached by shmdt() or when the process
3409  // terminates. If shmat() is not successful this will remove the shared
3410  // segment immediately.
3411  shmctl(shmid, IPC_RMID, NULL);
3412
3413  if (retAddr == (char *) -1) {
3414    if (warn_on_failure) {
3415      jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3416      warning(msg);
3417    }
3418    return NULL;
3419  }
3420  if ((retAddr != NULL) && UseNUMAInterleaving) {
3421    numa_make_global(retAddr, size);
3422  }
3423
3424  // The memory is committed
3425  address pc = CALLER_PC;
3426  MemTracker::record_virtual_memory_reserve((address)retAddr, size, pc);
3427  MemTracker::record_virtual_memory_commit((address)retAddr, size, pc);
3428
3429  return retAddr;
3430}
3431
3432bool os::release_memory_special(char* base, size_t bytes) {
3433  // detaching the SHM segment will also delete it, see reserve_memory_special()
3434  int rslt = shmdt(base);
3435  if (rslt == 0) {
3436    MemTracker::record_virtual_memory_uncommit((address)base, bytes);
3437    MemTracker::record_virtual_memory_release((address)base, bytes);
3438    return true;
3439  } else {
3440   return false;
3441  }
3442}
3443
3444size_t os::large_page_size() {
3445  return _large_page_size;
3446}
3447
3448// MPSS allows application to commit large page memory on demand; with ISM
3449// the entire memory region must be allocated as shared memory.
3450bool os::can_commit_large_page_memory() {
3451  return UseISM ? false : true;
3452}
3453
3454bool os::can_execute_large_page_memory() {
3455  return UseISM ? false : true;
3456}
3457
3458static int os_sleep(jlong millis, bool interruptible) {
3459  const jlong limit = INT_MAX;
3460  jlong prevtime;
3461  int res;
3462
3463  while (millis > limit) {
3464    if ((res = os_sleep(limit, interruptible)) != OS_OK)
3465      return res;
3466    millis -= limit;
3467  }
3468
3469  // Restart interrupted polls with new parameters until the proper delay
3470  // has been completed.
3471
3472  prevtime = getTimeMillis();
3473
3474  while (millis > 0) {
3475    jlong newtime;
3476
3477    if (!interruptible) {
3478      // Following assert fails for os::yield_all:
3479      // assert(!thread->is_Java_thread(), "must not be java thread");
3480      res = poll(NULL, 0, millis);
3481    } else {
3482      JavaThread *jt = JavaThread::current();
3483
3484      INTERRUPTIBLE_NORESTART_VM_ALWAYS(poll(NULL, 0, millis), res, jt,
3485        os::Solaris::clear_interrupted);
3486    }
3487
3488    // INTERRUPTIBLE_NORESTART_VM_ALWAYS returns res == OS_INTRPT for
3489    // thread.Interrupt.
3490
3491    // See c/r 6751923. Poll can return 0 before time
3492    // has elapsed if time is set via clock_settime (as NTP does).
3493    // res == 0 if poll timed out (see man poll RETURN VALUES)
3494    // using the logic below checks that we really did
3495    // sleep at least "millis" if not we'll sleep again.
3496    if( ( res == 0 ) || ((res == OS_ERR) && (errno == EINTR))) {
3497      newtime = getTimeMillis();
3498      assert(newtime >= prevtime, "time moving backwards");
3499    /* Doing prevtime and newtime in microseconds doesn't help precision,
3500       and trying to round up to avoid lost milliseconds can result in a
3501       too-short delay. */
3502      millis -= newtime - prevtime;
3503      if(millis <= 0)
3504        return OS_OK;
3505      prevtime = newtime;
3506    } else
3507      return res;
3508  }
3509
3510  return OS_OK;
3511}
3512
3513// Read calls from inside the vm need to perform state transitions
3514size_t os::read(int fd, void *buf, unsigned int nBytes) {
3515  INTERRUPTIBLE_RETURN_INT_VM(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3516}
3517
3518size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
3519  INTERRUPTIBLE_RETURN_INT(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3520}
3521
3522int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3523  assert(thread == Thread::current(),  "thread consistency check");
3524
3525  // TODO-FIXME: this should be removed.
3526  // On Solaris machines (especially 2.5.1) we found that sometimes the VM gets into a live lock
3527  // situation with a JavaThread being starved out of a lwp. The kernel doesn't seem to generate
3528  // a SIGWAITING signal which would enable the threads library to create a new lwp for the starving
3529  // thread. We suspect that because the Watcher thread keeps waking up at periodic intervals the kernel
3530  // is fooled into believing that the system is making progress. In the code below we block the
3531  // the watcher thread while safepoint is in progress so that it would not appear as though the
3532  // system is making progress.
3533  if (!Solaris::T2_libthread() &&
3534      thread->is_Watcher_thread() && SafepointSynchronize::is_synchronizing() && !Arguments::has_profile()) {
3535    // We now try to acquire the threads lock. Since this lock is held by the VM thread during
3536    // the entire safepoint, the watcher thread will  line up here during the safepoint.
3537    Threads_lock->lock_without_safepoint_check();
3538    Threads_lock->unlock();
3539  }
3540
3541  if (thread->is_Java_thread()) {
3542    // This is a JavaThread so we honor the _thread_blocked protocol
3543    // even for sleeps of 0 milliseconds. This was originally done
3544    // as a workaround for bug 4338139. However, now we also do it
3545    // to honor the suspend-equivalent protocol.
3546
3547    JavaThread *jt = (JavaThread *) thread;
3548    ThreadBlockInVM tbivm(jt);
3549
3550    jt->set_suspend_equivalent();
3551    // cleared by handle_special_suspend_equivalent_condition() or
3552    // java_suspend_self() via check_and_wait_while_suspended()
3553
3554    int ret_code;
3555    if (millis <= 0) {
3556      thr_yield();
3557      ret_code = 0;
3558    } else {
3559      // The original sleep() implementation did not create an
3560      // OSThreadWaitState helper for sleeps of 0 milliseconds.
3561      // I'm preserving that decision for now.
3562      OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3563
3564      ret_code = os_sleep(millis, interruptible);
3565    }
3566
3567    // were we externally suspended while we were waiting?
3568    jt->check_and_wait_while_suspended();
3569
3570    return ret_code;
3571  }
3572
3573  // non-JavaThread from this point on:
3574
3575  if (millis <= 0) {
3576    thr_yield();
3577    return 0;
3578  }
3579
3580  OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3581
3582  return os_sleep(millis, interruptible);
3583}
3584
3585int os::naked_sleep() {
3586  // %% make the sleep time an integer flag. for now use 1 millisec.
3587  return os_sleep(1, false);
3588}
3589
3590// Sleep forever; naked call to OS-specific sleep; use with CAUTION
3591void os::infinite_sleep() {
3592  while (true) {    // sleep forever ...
3593    ::sleep(100);   // ... 100 seconds at a time
3594  }
3595}
3596
3597// Used to convert frequent JVM_Yield() to nops
3598bool os::dont_yield() {
3599  if (DontYieldALot) {
3600    static hrtime_t last_time = 0;
3601    hrtime_t diff = getTimeNanos() - last_time;
3602
3603    if (diff < DontYieldALotInterval * 1000000)
3604      return true;
3605
3606    last_time += diff;
3607
3608    return false;
3609  }
3610  else {
3611    return false;
3612  }
3613}
3614
3615// Caveat: Solaris os::yield() causes a thread-state transition whereas
3616// the linux and win32 implementations do not.  This should be checked.
3617
3618void os::yield() {
3619  // Yields to all threads with same or greater priority
3620  os::sleep(Thread::current(), 0, false);
3621}
3622
3623// Note that yield semantics are defined by the scheduling class to which
3624// the thread currently belongs.  Typically, yield will _not yield to
3625// other equal or higher priority threads that reside on the dispatch queues
3626// of other CPUs.
3627
3628os::YieldResult os::NakedYield() { thr_yield(); return os::YIELD_UNKNOWN; }
3629
3630
3631// On Solaris we found that yield_all doesn't always yield to all other threads.
3632// There have been cases where there is a thread ready to execute but it doesn't
3633// get an lwp as the VM thread continues to spin with sleeps of 1 millisecond.
3634// The 1 millisecond wait doesn't seem long enough for the kernel to issue a
3635// SIGWAITING signal which will cause a new lwp to be created. So we count the
3636// number of times yield_all is called in the one loop and increase the sleep
3637// time after 8 attempts. If this fails too we increase the concurrency level
3638// so that the starving thread would get an lwp
3639
3640void os::yield_all(int attempts) {
3641  // Yields to all threads, including threads with lower priorities
3642  if (attempts == 0) {
3643    os::sleep(Thread::current(), 1, false);
3644  } else {
3645    int iterations = attempts % 30;
3646    if (iterations == 0 && !os::Solaris::T2_libthread()) {
3647      // thr_setconcurrency and _getconcurrency make sense only under T1.
3648      int noofLWPS = thr_getconcurrency();
3649      if (noofLWPS < (Threads::number_of_threads() + 2)) {
3650        thr_setconcurrency(thr_getconcurrency() + 1);
3651      }
3652    } else if (iterations < 25) {
3653      os::sleep(Thread::current(), 1, false);
3654    } else {
3655      os::sleep(Thread::current(), 10, false);
3656    }
3657  }
3658}
3659
3660// Called from the tight loops to possibly influence time-sharing heuristics
3661void os::loop_breaker(int attempts) {
3662  os::yield_all(attempts);
3663}
3664
3665
3666// Interface for setting lwp priorities.  If we are using T2 libthread,
3667// which forces the use of BoundThreads or we manually set UseBoundThreads,
3668// all of our threads will be assigned to real lwp's.  Using the thr_setprio
3669// function is meaningless in this mode so we must adjust the real lwp's priority
3670// The routines below implement the getting and setting of lwp priorities.
3671//
3672// Note: There are three priority scales used on Solaris.  Java priotities
3673//       which range from 1 to 10, libthread "thr_setprio" scale which range
3674//       from 0 to 127, and the current scheduling class of the process we
3675//       are running in.  This is typically from -60 to +60.
3676//       The setting of the lwp priorities in done after a call to thr_setprio
3677//       so Java priorities are mapped to libthread priorities and we map from
3678//       the latter to lwp priorities.  We don't keep priorities stored in
3679//       Java priorities since some of our worker threads want to set priorities
3680//       higher than all Java threads.
3681//
3682// For related information:
3683// (1)  man -s 2 priocntl
3684// (2)  man -s 4 priocntl
3685// (3)  man dispadmin
3686// =    librt.so
3687// =    libthread/common/rtsched.c - thrp_setlwpprio().
3688// =    ps -cL <pid> ... to validate priority.
3689// =    sched_get_priority_min and _max
3690//              pthread_create
3691//              sched_setparam
3692//              pthread_setschedparam
3693//
3694// Assumptions:
3695// +    We assume that all threads in the process belong to the same
3696//              scheduling class.   IE. an homogenous process.
3697// +    Must be root or in IA group to change change "interactive" attribute.
3698//              Priocntl() will fail silently.  The only indication of failure is when
3699//              we read-back the value and notice that it hasn't changed.
3700// +    Interactive threads enter the runq at the head, non-interactive at the tail.
3701// +    For RT, change timeslice as well.  Invariant:
3702//              constant "priority integral"
3703//              Konst == TimeSlice * (60-Priority)
3704//              Given a priority, compute appropriate timeslice.
3705// +    Higher numerical values have higher priority.
3706
3707// sched class attributes
3708typedef struct {
3709        int   schedPolicy;              // classID
3710        int   maxPrio;
3711        int   minPrio;
3712} SchedInfo;
3713
3714
3715static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3716
3717#ifdef ASSERT
3718static int  ReadBackValidate = 1;
3719#endif
3720static int  myClass     = 0;
3721static int  myMin       = 0;
3722static int  myMax       = 0;
3723static int  myCur       = 0;
3724static bool priocntl_enable = false;
3725
3726static const int criticalPrio = 60; // FX/60 is critical thread class/priority on T4
3727static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3728
3729// Call the version of priocntl suitable for all supported versions
3730// of Solaris. We need to call through this wrapper so that we can
3731// build on Solaris 9 and run on Solaris 8, 9 and 10.
3732//
3733// This code should be removed if we ever stop supporting Solaris 8
3734// and earlier releases.
3735
3736static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3737typedef long (*priocntl_type)(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3738static priocntl_type priocntl_ptr = priocntl_stub;
3739
3740// Stub to set the value of the real pointer, and then call the real
3741// function.
3742
3743static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg) {
3744  // Try Solaris 8- name only.
3745  priocntl_type tmp = (priocntl_type)dlsym(RTLD_DEFAULT, "__priocntl");
3746  guarantee(tmp != NULL, "priocntl function not found.");
3747  priocntl_ptr = tmp;
3748  return (*priocntl_ptr)(PC_VERSION, idtype, id, cmd, arg);
3749}
3750
3751
3752// lwp_priocntl_init
3753//
3754// Try to determine the priority scale for our process.
3755//
3756// Return errno or 0 if OK.
3757//
3758static
3759int     lwp_priocntl_init ()
3760{
3761  int rslt;
3762  pcinfo_t ClassInfo;
3763  pcparms_t ParmInfo;
3764  int i;
3765
3766  if (!UseThreadPriorities) return 0;
3767
3768  // We are using Bound threads, we need to determine our priority ranges
3769  if (os::Solaris::T2_libthread() || UseBoundThreads) {
3770    // If ThreadPriorityPolicy is 1, switch tables
3771    if (ThreadPriorityPolicy == 1) {
3772      for (i = 0 ; i < CriticalPriority+1; i++)
3773        os::java_to_os_priority[i] = prio_policy1[i];
3774    }
3775    if (UseCriticalJavaThreadPriority) {
3776      // MaxPriority always maps to the FX scheduling class and criticalPrio.
3777      // See set_native_priority() and set_lwp_class_and_priority().
3778      // Save original MaxPriority mapping in case attempt to
3779      // use critical priority fails.
3780      java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3781      // Set negative to distinguish from other priorities
3782      os::java_to_os_priority[MaxPriority] = -criticalPrio;
3783    }
3784  }
3785  // Not using Bound Threads, set to ThreadPolicy 1
3786  else {
3787    for ( i = 0 ; i < CriticalPriority+1; i++ ) {
3788      os::java_to_os_priority[i] = prio_policy1[i];
3789    }
3790    return 0;
3791  }
3792
3793  // Get IDs for a set of well-known scheduling classes.
3794  // TODO-FIXME: GETCLINFO returns the current # of classes in the
3795  // the system.  We should have a loop that iterates over the
3796  // classID values, which are known to be "small" integers.
3797
3798  strcpy(ClassInfo.pc_clname, "TS");
3799  ClassInfo.pc_cid = -1;
3800  rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3801  if (rslt < 0) return errno;
3802  assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3803  tsLimits.schedPolicy = ClassInfo.pc_cid;
3804  tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3805  tsLimits.minPrio = -tsLimits.maxPrio;
3806
3807  strcpy(ClassInfo.pc_clname, "IA");
3808  ClassInfo.pc_cid = -1;
3809  rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3810  if (rslt < 0) return errno;
3811  assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3812  iaLimits.schedPolicy = ClassInfo.pc_cid;
3813  iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3814  iaLimits.minPrio = -iaLimits.maxPrio;
3815
3816  strcpy(ClassInfo.pc_clname, "RT");
3817  ClassInfo.pc_cid = -1;
3818  rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3819  if (rslt < 0) return errno;
3820  assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3821  rtLimits.schedPolicy = ClassInfo.pc_cid;
3822  rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3823  rtLimits.minPrio = 0;
3824
3825  strcpy(ClassInfo.pc_clname, "FX");
3826  ClassInfo.pc_cid = -1;
3827  rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3828  if (rslt < 0) return errno;
3829  assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3830  fxLimits.schedPolicy = ClassInfo.pc_cid;
3831  fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3832  fxLimits.minPrio = 0;
3833
3834  // Query our "current" scheduling class.
3835  // This will normally be IA, TS or, rarely, FX or RT.
3836  memset(&ParmInfo, 0, sizeof(ParmInfo));
3837  ParmInfo.pc_cid = PC_CLNULL;
3838  rslt = (*priocntl_ptr) (PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3839  if (rslt < 0) return errno;
3840  myClass = ParmInfo.pc_cid;
3841
3842  // We now know our scheduling classId, get specific information
3843  // about the class.
3844  ClassInfo.pc_cid = myClass;
3845  ClassInfo.pc_clname[0] = 0;
3846  rslt = (*priocntl_ptr) (PC_VERSION, (idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3847  if (rslt < 0) return errno;
3848
3849  if (ThreadPriorityVerbose) {
3850    tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3851  }
3852
3853  memset(&ParmInfo, 0, sizeof(pcparms_t));
3854  ParmInfo.pc_cid = PC_CLNULL;
3855  rslt = (*priocntl_ptr)(PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3856  if (rslt < 0) return errno;
3857
3858  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3859    myMin = rtLimits.minPrio;
3860    myMax = rtLimits.maxPrio;
3861  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3862    iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3863    myMin = iaLimits.minPrio;
3864    myMax = iaLimits.maxPrio;
3865    myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3866  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3867    tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3868    myMin = tsLimits.minPrio;
3869    myMax = tsLimits.maxPrio;
3870    myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3871  } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3872    fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3873    myMin = fxLimits.minPrio;
3874    myMax = fxLimits.maxPrio;
3875    myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3876  } else {
3877    // No clue - punt
3878    if (ThreadPriorityVerbose)
3879      tty->print_cr ("Unknown scheduling class: %s ... \n", ClassInfo.pc_clname);
3880    return EINVAL;      // no clue, punt
3881  }
3882
3883  if (ThreadPriorityVerbose) {
3884    tty->print_cr ("Thread priority Range: [%d..%d]\n", myMin, myMax);
3885  }
3886
3887  priocntl_enable = true;  // Enable changing priorities
3888  return 0;
3889}
3890
3891#define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3892#define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3893#define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3894#define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3895
3896
3897// scale_to_lwp_priority
3898//
3899// Convert from the libthread "thr_setprio" scale to our current
3900// lwp scheduling class scale.
3901//
3902static
3903int     scale_to_lwp_priority (int rMin, int rMax, int x)
3904{
3905  int v;
3906
3907  if (x == 127) return rMax;            // avoid round-down
3908    v = (((x*(rMax-rMin)))/128)+rMin;
3909  return v;
3910}
3911
3912
3913// set_lwp_class_and_priority
3914//
3915// Set the class and priority of the lwp.  This call should only
3916// be made when using bound threads (T2 threads are bound by default).
3917//
3918int set_lwp_class_and_priority(int ThreadID, int lwpid,
3919                               int newPrio, int new_class, bool scale) {
3920  int rslt;
3921  int Actual, Expected, prv;
3922  pcparms_t ParmInfo;                   // for GET-SET
3923#ifdef ASSERT
3924  pcparms_t ReadBack;                   // for readback
3925#endif
3926
3927  // Set priority via PC_GETPARMS, update, PC_SETPARMS
3928  // Query current values.
3929  // TODO: accelerate this by eliminating the PC_GETPARMS call.
3930  // Cache "pcparms_t" in global ParmCache.
3931  // TODO: elide set-to-same-value
3932
3933  // If something went wrong on init, don't change priorities.
3934  if ( !priocntl_enable ) {
3935    if (ThreadPriorityVerbose)
3936      tty->print_cr("Trying to set priority but init failed, ignoring");
3937    return EINVAL;
3938  }
3939
3940  // If lwp hasn't started yet, just return
3941  // the _start routine will call us again.
3942  if ( lwpid <= 0 ) {
3943    if (ThreadPriorityVerbose) {
3944      tty->print_cr ("deferring the set_lwp_class_and_priority of thread "
3945                     INTPTR_FORMAT " to %d, lwpid not set",
3946                     ThreadID, newPrio);
3947    }
3948    return 0;
3949  }
3950
3951  if (ThreadPriorityVerbose) {
3952    tty->print_cr ("set_lwp_class_and_priority("
3953                   INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3954                   ThreadID, lwpid, newPrio);
3955  }
3956
3957  memset(&ParmInfo, 0, sizeof(pcparms_t));
3958  ParmInfo.pc_cid = PC_CLNULL;
3959  rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3960  if (rslt < 0) return errno;
3961
3962  int cur_class = ParmInfo.pc_cid;
3963  ParmInfo.pc_cid = (id_t)new_class;
3964
3965  if (new_class == rtLimits.schedPolicy) {
3966    rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3967    rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3968                                                       rtLimits.maxPrio, newPrio)
3969                               : newPrio;
3970    rtInfo->rt_tqsecs  = RT_NOCHANGE;
3971    rtInfo->rt_tqnsecs = RT_NOCHANGE;
3972    if (ThreadPriorityVerbose) {
3973      tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3974    }
3975  } else if (new_class == iaLimits.schedPolicy) {
3976    iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3977    int maxClamped     = MIN2(iaLimits.maxPrio,
3978                              cur_class == new_class
3979                                ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3980    iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3981                                                       maxClamped, newPrio)
3982                               : newPrio;
3983    iaInfo->ia_uprilim = cur_class == new_class
3984                           ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3985    iaInfo->ia_mode    = IA_NOCHANGE;
3986    if (ThreadPriorityVerbose) {
3987      tty->print_cr("IA: [%d...%d] %d->%d\n",
3988                    iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3989    }
3990  } else if (new_class == tsLimits.schedPolicy) {
3991    tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3992    int maxClamped     = MIN2(tsLimits.maxPrio,
3993                              cur_class == new_class
3994                                ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3995    tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3996                                                       maxClamped, newPrio)
3997                               : newPrio;
3998    tsInfo->ts_uprilim = cur_class == new_class
3999                           ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
4000    if (ThreadPriorityVerbose) {
4001      tty->print_cr("TS: [%d...%d] %d->%d\n",
4002                    tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
4003    }
4004  } else if (new_class == fxLimits.schedPolicy) {
4005    fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
4006    int maxClamped     = MIN2(fxLimits.maxPrio,
4007                              cur_class == new_class
4008                                ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
4009    fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
4010                                                       maxClamped, newPrio)
4011                               : newPrio;
4012    fxInfo->fx_uprilim = cur_class == new_class
4013                           ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
4014    fxInfo->fx_tqsecs  = FX_NOCHANGE;
4015    fxInfo->fx_tqnsecs = FX_NOCHANGE;
4016    if (ThreadPriorityVerbose) {
4017      tty->print_cr("FX: [%d...%d] %d->%d\n",
4018                    fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
4019    }
4020  } else {
4021    if (ThreadPriorityVerbose) {
4022      tty->print_cr("Unknown new scheduling class %d\n", new_class);
4023    }
4024    return EINVAL;    // no clue, punt
4025  }
4026
4027  rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
4028  if (ThreadPriorityVerbose && rslt) {
4029    tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
4030  }
4031  if (rslt < 0) return errno;
4032
4033#ifdef ASSERT
4034  // Sanity check: read back what we just attempted to set.
4035  // In theory it could have changed in the interim ...
4036  //
4037  // The priocntl system call is tricky.
4038  // Sometimes it'll validate the priority value argument and
4039  // return EINVAL if unhappy.  At other times it fails silently.
4040  // Readbacks are prudent.
4041
4042  if (!ReadBackValidate) return 0;
4043
4044  memset(&ReadBack, 0, sizeof(pcparms_t));
4045  ReadBack.pc_cid = PC_CLNULL;
4046  rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
4047  assert(rslt >= 0, "priocntl failed");
4048  Actual = Expected = 0xBAD;
4049  assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
4050  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
4051    Actual   = RTPRI(ReadBack)->rt_pri;
4052    Expected = RTPRI(ParmInfo)->rt_pri;
4053  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
4054    Actual   = IAPRI(ReadBack)->ia_upri;
4055    Expected = IAPRI(ParmInfo)->ia_upri;
4056  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
4057    Actual   = TSPRI(ReadBack)->ts_upri;
4058    Expected = TSPRI(ParmInfo)->ts_upri;
4059  } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
4060    Actual   = FXPRI(ReadBack)->fx_upri;
4061    Expected = FXPRI(ParmInfo)->fx_upri;
4062  } else {
4063    if (ThreadPriorityVerbose) {
4064      tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
4065                    ParmInfo.pc_cid);
4066    }
4067  }
4068
4069  if (Actual != Expected) {
4070    if (ThreadPriorityVerbose) {
4071      tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
4072                     lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
4073    }
4074  }
4075#endif
4076
4077  return 0;
4078}
4079
4080// Solaris only gives access to 128 real priorities at a time,
4081// so we expand Java's ten to fill this range.  This would be better
4082// if we dynamically adjusted relative priorities.
4083//
4084// The ThreadPriorityPolicy option allows us to select 2 different
4085// priority scales.
4086//
4087// ThreadPriorityPolicy=0
4088// Since the Solaris' default priority is MaximumPriority, we do not
4089// set a priority lower than Max unless a priority lower than
4090// NormPriority is requested.
4091//
4092// ThreadPriorityPolicy=1
4093// This mode causes the priority table to get filled with
4094// linear values.  NormPriority get's mapped to 50% of the
4095// Maximum priority an so on.  This will cause VM threads
4096// to get unfair treatment against other Solaris processes
4097// which do not explicitly alter their thread priorities.
4098//
4099
4100int os::java_to_os_priority[CriticalPriority + 1] = {
4101  -99999,         // 0 Entry should never be used
4102
4103  0,              // 1 MinPriority
4104  32,             // 2
4105  64,             // 3
4106
4107  96,             // 4
4108  127,            // 5 NormPriority
4109  127,            // 6
4110
4111  127,            // 7
4112  127,            // 8
4113  127,            // 9 NearMaxPriority
4114
4115  127,            // 10 MaxPriority
4116
4117  -criticalPrio   // 11 CriticalPriority
4118};
4119
4120OSReturn os::set_native_priority(Thread* thread, int newpri) {
4121  OSThread* osthread = thread->osthread();
4122
4123  // Save requested priority in case the thread hasn't been started
4124  osthread->set_native_priority(newpri);
4125
4126  // Check for critical priority request
4127  bool fxcritical = false;
4128  if (newpri == -criticalPrio) {
4129    fxcritical = true;
4130    newpri = criticalPrio;
4131  }
4132
4133  assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
4134  if (!UseThreadPriorities) return OS_OK;
4135
4136  int status = 0;
4137
4138  if (!fxcritical) {
4139    // Use thr_setprio only if we have a priority that thr_setprio understands
4140    status = thr_setprio(thread->osthread()->thread_id(), newpri);
4141  }
4142
4143  if (os::Solaris::T2_libthread() ||
4144      (UseBoundThreads && osthread->is_vm_created())) {
4145    int lwp_status =
4146      set_lwp_class_and_priority(osthread->thread_id(),
4147                                 osthread->lwp_id(),
4148                                 newpri,
4149                                 fxcritical ? fxLimits.schedPolicy : myClass,
4150                                 !fxcritical);
4151    if (lwp_status != 0 && fxcritical) {
4152      // Try again, this time without changing the scheduling class
4153      newpri = java_MaxPriority_to_os_priority;
4154      lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
4155                                              osthread->lwp_id(),
4156                                              newpri, myClass, false);
4157    }
4158    status |= lwp_status;
4159  }
4160  return (status == 0) ? OS_OK : OS_ERR;
4161}
4162
4163
4164OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4165  int p;
4166  if ( !UseThreadPriorities ) {
4167    *priority_ptr = NormalPriority;
4168    return OS_OK;
4169  }
4170  int status = thr_getprio(thread->osthread()->thread_id(), &p);
4171  if (status != 0) {
4172    return OS_ERR;
4173  }
4174  *priority_ptr = p;
4175  return OS_OK;
4176}
4177
4178
4179// Hint to the underlying OS that a task switch would not be good.
4180// Void return because it's a hint and can fail.
4181void os::hint_no_preempt() {
4182  schedctl_start(schedctl_init());
4183}
4184
4185void os::interrupt(Thread* thread) {
4186  assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4187
4188  OSThread* osthread = thread->osthread();
4189
4190  int isInterrupted = osthread->interrupted();
4191  if (!isInterrupted) {
4192      osthread->set_interrupted(true);
4193      OrderAccess::fence();
4194      // os::sleep() is implemented with either poll (NULL,0,timeout) or
4195      // by parking on _SleepEvent.  If the former, thr_kill will unwedge
4196      // the sleeper by SIGINTR, otherwise the unpark() will wake the sleeper.
4197      ParkEvent * const slp = thread->_SleepEvent ;
4198      if (slp != NULL) slp->unpark() ;
4199  }
4200
4201  // For JSR166:  unpark after setting status but before thr_kill -dl
4202  if (thread->is_Java_thread()) {
4203    ((JavaThread*)thread)->parker()->unpark();
4204  }
4205
4206  // Handle interruptible wait() ...
4207  ParkEvent * const ev = thread->_ParkEvent ;
4208  if (ev != NULL) ev->unpark() ;
4209
4210  // When events are used everywhere for os::sleep, then this thr_kill
4211  // will only be needed if UseVMInterruptibleIO is true.
4212
4213  if (!isInterrupted) {
4214    int status = thr_kill(osthread->thread_id(), os::Solaris::SIGinterrupt());
4215    assert_status(status == 0, status, "thr_kill");
4216
4217    // Bump thread interruption counter
4218    RuntimeService::record_thread_interrupt_signaled_count();
4219  }
4220}
4221
4222
4223bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4224  assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4225
4226  OSThread* osthread = thread->osthread();
4227
4228  bool res = osthread->interrupted();
4229
4230  // NOTE that since there is no "lock" around these two operations,
4231  // there is the possibility that the interrupted flag will be
4232  // "false" but that the interrupt event will be set. This is
4233  // intentional. The effect of this is that Object.wait() will appear
4234  // to have a spurious wakeup, which is not harmful, and the
4235  // possibility is so rare that it is not worth the added complexity
4236  // to add yet another lock. It has also been recommended not to put
4237  // the interrupted flag into the os::Solaris::Event structure,
4238  // because it hides the issue.
4239  if (res && clear_interrupted) {
4240    osthread->set_interrupted(false);
4241  }
4242  return res;
4243}
4244
4245
4246void os::print_statistics() {
4247}
4248
4249int os::message_box(const char* title, const char* message) {
4250  int i;
4251  fdStream err(defaultStream::error_fd());
4252  for (i = 0; i < 78; i++) err.print_raw("=");
4253  err.cr();
4254  err.print_raw_cr(title);
4255  for (i = 0; i < 78; i++) err.print_raw("-");
4256  err.cr();
4257  err.print_raw_cr(message);
4258  for (i = 0; i < 78; i++) err.print_raw("=");
4259  err.cr();
4260
4261  char buf[16];
4262  // Prevent process from exiting upon "read error" without consuming all CPU
4263  while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4264
4265  return buf[0] == 'y' || buf[0] == 'Y';
4266}
4267
4268// A lightweight implementation that does not suspend the target thread and
4269// thus returns only a hint. Used for profiling only!
4270ExtendedPC os::get_thread_pc(Thread* thread) {
4271  // Make sure that it is called by the watcher and the Threads lock is owned.
4272  assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
4273  // For now, is only used to profile the VM Thread
4274  assert(thread->is_VM_thread(), "Can only be called for VMThread");
4275  ExtendedPC epc;
4276
4277  GetThreadPC_Callback  cb(ProfileVM_lock);
4278  OSThread *osthread = thread->osthread();
4279  const int time_to_wait = 400; // 400ms wait for initial response
4280  int status = cb.interrupt(thread, time_to_wait);
4281
4282  if (cb.is_done() ) {
4283    epc = cb.addr();
4284  } else {
4285    DEBUG_ONLY(tty->print_cr("Failed to get pc for thread: %d got %d status",
4286                              osthread->thread_id(), status););
4287    // epc is already NULL
4288  }
4289  return epc;
4290}
4291
4292
4293// This does not do anything on Solaris. This is basically a hook for being
4294// able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
4295void os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread) {
4296  f(value, method, args, thread);
4297}
4298
4299// This routine may be used by user applications as a "hook" to catch signals.
4300// The user-defined signal handler must pass unrecognized signals to this
4301// routine, and if it returns true (non-zero), then the signal handler must
4302// return immediately.  If the flag "abort_if_unrecognized" is true, then this
4303// routine will never retun false (zero), but instead will execute a VM panic
4304// routine kill the process.
4305//
4306// If this routine returns false, it is OK to call it again.  This allows
4307// the user-defined signal handler to perform checks either before or after
4308// the VM performs its own checks.  Naturally, the user code would be making
4309// a serious error if it tried to handle an exception (such as a null check
4310// or breakpoint) that the VM was generating for its own correct operation.
4311//
4312// This routine may recognize any of the following kinds of signals:
4313// SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
4314// os::Solaris::SIGasync
4315// It should be consulted by handlers for any of those signals.
4316// It explicitly does not recognize os::Solaris::SIGinterrupt
4317//
4318// The caller of this routine must pass in the three arguments supplied
4319// to the function referred to in the "sa_sigaction" (not the "sa_handler")
4320// field of the structure passed to sigaction().  This routine assumes that
4321// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4322//
4323// Note that the VM will print warnings if it detects conflicting signal
4324// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4325//
4326extern "C" JNIEXPORT int
4327JVM_handle_solaris_signal(int signo, siginfo_t* siginfo, void* ucontext,
4328                          int abort_if_unrecognized);
4329
4330
4331void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
4332  int orig_errno = errno;  // Preserve errno value over signal handler.
4333  JVM_handle_solaris_signal(sig, info, ucVoid, true);
4334  errno = orig_errno;
4335}
4336
4337/* Do not delete - if guarantee is ever removed,  a signal handler (even empty)
4338   is needed to provoke threads blocked on IO to return an EINTR
4339   Note: this explicitly does NOT call JVM_handle_solaris_signal and
4340   does NOT participate in signal chaining due to requirement for
4341   NOT setting SA_RESTART to make EINTR work. */
4342extern "C" void sigINTRHandler(int sig, siginfo_t* info, void* ucVoid) {
4343   if (UseSignalChaining) {
4344      struct sigaction *actp = os::Solaris::get_chained_signal_action(sig);
4345      if (actp && actp->sa_handler) {
4346        vm_exit_during_initialization("Signal chaining detected for VM interrupt signal, try -XX:+UseAltSigs");
4347      }
4348   }
4349}
4350
4351// This boolean allows users to forward their own non-matching signals
4352// to JVM_handle_solaris_signal, harmlessly.
4353bool os::Solaris::signal_handlers_are_installed = false;
4354
4355// For signal-chaining
4356bool os::Solaris::libjsig_is_loaded = false;
4357typedef struct sigaction *(*get_signal_t)(int);
4358get_signal_t os::Solaris::get_signal_action = NULL;
4359
4360struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
4361  struct sigaction *actp = NULL;
4362
4363  if ((libjsig_is_loaded)  && (sig <= Maxlibjsigsigs)) {
4364    // Retrieve the old signal handler from libjsig
4365    actp = (*get_signal_action)(sig);
4366  }
4367  if (actp == NULL) {
4368    // Retrieve the preinstalled signal handler from jvm
4369    actp = get_preinstalled_handler(sig);
4370  }
4371
4372  return actp;
4373}
4374
4375static bool call_chained_handler(struct sigaction *actp, int sig,
4376                                 siginfo_t *siginfo, void *context) {
4377  // Call the old signal handler
4378  if (actp->sa_handler == SIG_DFL) {
4379    // It's more reasonable to let jvm treat it as an unexpected exception
4380    // instead of taking the default action.
4381    return false;
4382  } else if (actp->sa_handler != SIG_IGN) {
4383    if ((actp->sa_flags & SA_NODEFER) == 0) {
4384      // automaticlly block the signal
4385      sigaddset(&(actp->sa_mask), sig);
4386    }
4387
4388    sa_handler_t hand;
4389    sa_sigaction_t sa;
4390    bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4391    // retrieve the chained handler
4392    if (siginfo_flag_set) {
4393      sa = actp->sa_sigaction;
4394    } else {
4395      hand = actp->sa_handler;
4396    }
4397
4398    if ((actp->sa_flags & SA_RESETHAND) != 0) {
4399      actp->sa_handler = SIG_DFL;
4400    }
4401
4402    // try to honor the signal mask
4403    sigset_t oset;
4404    thr_sigsetmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4405
4406    // call into the chained handler
4407    if (siginfo_flag_set) {
4408      (*sa)(sig, siginfo, context);
4409    } else {
4410      (*hand)(sig);
4411    }
4412
4413    // restore the signal mask
4414    thr_sigsetmask(SIG_SETMASK, &oset, 0);
4415  }
4416  // Tell jvm's signal handler the signal is taken care of.
4417  return true;
4418}
4419
4420bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4421  bool chained = false;
4422  // signal-chaining
4423  if (UseSignalChaining) {
4424    struct sigaction *actp = get_chained_signal_action(sig);
4425    if (actp != NULL) {
4426      chained = call_chained_handler(actp, sig, siginfo, context);
4427    }
4428  }
4429  return chained;
4430}
4431
4432struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
4433  assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4434  if (preinstalled_sigs[sig] != 0) {
4435    return &chainedsigactions[sig];
4436  }
4437  return NULL;
4438}
4439
4440void os::Solaris::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4441
4442  assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
4443  assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4444  chainedsigactions[sig] = oldAct;
4445  preinstalled_sigs[sig] = 1;
4446}
4447
4448void os::Solaris::set_signal_handler(int sig, bool set_installed, bool oktochain) {
4449  // Check for overwrite.
4450  struct sigaction oldAct;
4451  sigaction(sig, (struct sigaction*)NULL, &oldAct);
4452  void* oldhand = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4453                                      : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4454  if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4455      oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4456      oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
4457    if (AllowUserSignalHandlers || !set_installed) {
4458      // Do not overwrite; user takes responsibility to forward to us.
4459      return;
4460    } else if (UseSignalChaining) {
4461      if (oktochain) {
4462        // save the old handler in jvm
4463        save_preinstalled_handler(sig, oldAct);
4464      } else {
4465        vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal, try -XX:+UseAltSigs.");
4466      }
4467      // libjsig also interposes the sigaction() call below and saves the
4468      // old sigaction on it own.
4469    } else {
4470      fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4471                    "%#lx for signal %d.", (long)oldhand, sig));
4472    }
4473  }
4474
4475  struct sigaction sigAct;
4476  sigfillset(&(sigAct.sa_mask));
4477  sigAct.sa_handler = SIG_DFL;
4478
4479  sigAct.sa_sigaction = signalHandler;
4480  // Handle SIGSEGV on alternate signal stack if
4481  // not using stack banging
4482  if (!UseStackBanging && sig == SIGSEGV) {
4483    sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
4484  // Interruptible i/o requires SA_RESTART cleared so EINTR
4485  // is returned instead of restarting system calls
4486  } else if (sig == os::Solaris::SIGinterrupt()) {
4487    sigemptyset(&sigAct.sa_mask);
4488    sigAct.sa_handler = NULL;
4489    sigAct.sa_flags = SA_SIGINFO;
4490    sigAct.sa_sigaction = sigINTRHandler;
4491  } else {
4492    sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
4493  }
4494  os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
4495
4496  sigaction(sig, &sigAct, &oldAct);
4497
4498  void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4499                                       : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4500  assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4501}
4502
4503
4504#define DO_SIGNAL_CHECK(sig) \
4505  if (!sigismember(&check_signal_done, sig)) \
4506    os::Solaris::check_signal_handler(sig)
4507
4508// This method is a periodic task to check for misbehaving JNI applications
4509// under CheckJNI, we can add any periodic checks here
4510
4511void os::run_periodic_checks() {
4512  // A big source of grief is hijacking virt. addr 0x0 on Solaris,
4513  // thereby preventing a NULL checks.
4514  if(!check_addr0_done) check_addr0_done = check_addr0(tty);
4515
4516  if (check_signals == false) return;
4517
4518  // SEGV and BUS if overridden could potentially prevent
4519  // generation of hs*.log in the event of a crash, debugging
4520  // such a case can be very challenging, so we absolutely
4521  // check for the following for a good measure:
4522  DO_SIGNAL_CHECK(SIGSEGV);
4523  DO_SIGNAL_CHECK(SIGILL);
4524  DO_SIGNAL_CHECK(SIGFPE);
4525  DO_SIGNAL_CHECK(SIGBUS);
4526  DO_SIGNAL_CHECK(SIGPIPE);
4527  DO_SIGNAL_CHECK(SIGXFSZ);
4528
4529  // ReduceSignalUsage allows the user to override these handlers
4530  // see comments at the very top and jvm_solaris.h
4531  if (!ReduceSignalUsage) {
4532    DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4533    DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4534    DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4535    DO_SIGNAL_CHECK(BREAK_SIGNAL);
4536  }
4537
4538  // See comments above for using JVM1/JVM2 and UseAltSigs
4539  DO_SIGNAL_CHECK(os::Solaris::SIGinterrupt());
4540  DO_SIGNAL_CHECK(os::Solaris::SIGasync());
4541
4542}
4543
4544typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4545
4546static os_sigaction_t os_sigaction = NULL;
4547
4548void os::Solaris::check_signal_handler(int sig) {
4549  char buf[O_BUFLEN];
4550  address jvmHandler = NULL;
4551
4552  struct sigaction act;
4553  if (os_sigaction == NULL) {
4554    // only trust the default sigaction, in case it has been interposed
4555    os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4556    if (os_sigaction == NULL) return;
4557  }
4558
4559  os_sigaction(sig, (struct sigaction*)NULL, &act);
4560
4561  address thisHandler = (act.sa_flags & SA_SIGINFO)
4562    ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4563    : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4564
4565
4566  switch(sig) {
4567    case SIGSEGV:
4568    case SIGBUS:
4569    case SIGFPE:
4570    case SIGPIPE:
4571    case SIGXFSZ:
4572    case SIGILL:
4573      jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4574      break;
4575
4576    case SHUTDOWN1_SIGNAL:
4577    case SHUTDOWN2_SIGNAL:
4578    case SHUTDOWN3_SIGNAL:
4579    case BREAK_SIGNAL:
4580      jvmHandler = (address)user_handler();
4581      break;
4582
4583    default:
4584      int intrsig = os::Solaris::SIGinterrupt();
4585      int asynsig = os::Solaris::SIGasync();
4586
4587      if (sig == intrsig) {
4588        jvmHandler = CAST_FROM_FN_PTR(address, sigINTRHandler);
4589      } else if (sig == asynsig) {
4590        jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4591      } else {
4592        return;
4593      }
4594      break;
4595  }
4596
4597
4598  if (thisHandler != jvmHandler) {
4599    tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4600    tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4601    tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4602    // No need to check this sig any longer
4603    sigaddset(&check_signal_done, sig);
4604  } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4605    tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4606    tty->print("expected:" PTR32_FORMAT, os::Solaris::get_our_sigflags(sig));
4607    tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4608    // No need to check this sig any longer
4609    sigaddset(&check_signal_done, sig);
4610  }
4611
4612  // Print all the signal handler state
4613  if (sigismember(&check_signal_done, sig)) {
4614    print_signal_handlers(tty, buf, O_BUFLEN);
4615  }
4616
4617}
4618
4619void os::Solaris::install_signal_handlers() {
4620  bool libjsigdone = false;
4621  signal_handlers_are_installed = true;
4622
4623  // signal-chaining
4624  typedef void (*signal_setting_t)();
4625  signal_setting_t begin_signal_setting = NULL;
4626  signal_setting_t end_signal_setting = NULL;
4627  begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4628                                        dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4629  if (begin_signal_setting != NULL) {
4630    end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4631                                        dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4632    get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4633                                       dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4634    get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4635                                         dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4636    libjsig_is_loaded = true;
4637    if (os::Solaris::get_libjsig_version != NULL) {
4638      libjsigversion =  (*os::Solaris::get_libjsig_version)();
4639    }
4640    assert(UseSignalChaining, "should enable signal-chaining");
4641  }
4642  if (libjsig_is_loaded) {
4643    // Tell libjsig jvm is setting signal handlers
4644    (*begin_signal_setting)();
4645  }
4646
4647  set_signal_handler(SIGSEGV, true, true);
4648  set_signal_handler(SIGPIPE, true, true);
4649  set_signal_handler(SIGXFSZ, true, true);
4650  set_signal_handler(SIGBUS, true, true);
4651  set_signal_handler(SIGILL, true, true);
4652  set_signal_handler(SIGFPE, true, true);
4653
4654
4655  if (os::Solaris::SIGinterrupt() > OLDMAXSIGNUM || os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4656
4657    // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4658    // can not register overridable signals which might be > 32
4659    if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4660    // Tell libjsig jvm has finished setting signal handlers
4661      (*end_signal_setting)();
4662      libjsigdone = true;
4663    }
4664  }
4665
4666  // Never ok to chain our SIGinterrupt
4667  set_signal_handler(os::Solaris::SIGinterrupt(), true, false);
4668  set_signal_handler(os::Solaris::SIGasync(), true, true);
4669
4670  if (libjsig_is_loaded && !libjsigdone) {
4671    // Tell libjsig jvm finishes setting signal handlers
4672    (*end_signal_setting)();
4673  }
4674
4675  // We don't activate signal checker if libjsig is in place, we trust ourselves
4676  // and if UserSignalHandler is installed all bets are off.
4677  // Log that signal checking is off only if -verbose:jni is specified.
4678  if (CheckJNICalls) {
4679    if (libjsig_is_loaded) {
4680      if (PrintJNIResolving) {
4681        tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4682      }
4683      check_signals = false;
4684    }
4685    if (AllowUserSignalHandlers) {
4686      if (PrintJNIResolving) {
4687        tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4688      }
4689      check_signals = false;
4690    }
4691  }
4692}
4693
4694
4695void report_error(const char* file_name, int line_no, const char* title, const char* format, ...);
4696
4697const char * signames[] = {
4698  "SIG0",
4699  "SIGHUP", "SIGINT", "SIGQUIT", "SIGILL", "SIGTRAP",
4700  "SIGABRT", "SIGEMT", "SIGFPE", "SIGKILL", "SIGBUS",
4701  "SIGSEGV", "SIGSYS", "SIGPIPE", "SIGALRM", "SIGTERM",
4702  "SIGUSR1", "SIGUSR2", "SIGCLD", "SIGPWR", "SIGWINCH",
4703  "SIGURG", "SIGPOLL", "SIGSTOP", "SIGTSTP", "SIGCONT",
4704  "SIGTTIN", "SIGTTOU", "SIGVTALRM", "SIGPROF", "SIGXCPU",
4705  "SIGXFSZ", "SIGWAITING", "SIGLWP", "SIGFREEZE", "SIGTHAW",
4706  "SIGCANCEL", "SIGLOST"
4707};
4708
4709const char* os::exception_name(int exception_code, char* buf, size_t size) {
4710  if (0 < exception_code && exception_code <= SIGRTMAX) {
4711    // signal
4712    if (exception_code < sizeof(signames)/sizeof(const char*)) {
4713       jio_snprintf(buf, size, "%s", signames[exception_code]);
4714    } else {
4715       jio_snprintf(buf, size, "SIG%d", exception_code);
4716    }
4717    return buf;
4718  } else {
4719    return NULL;
4720  }
4721}
4722
4723// (Static) wrappers for the new libthread API
4724int_fnP_thread_t_iP_uP_stack_tP_gregset_t os::Solaris::_thr_getstate;
4725int_fnP_thread_t_i_gregset_t os::Solaris::_thr_setstate;
4726int_fnP_thread_t_i os::Solaris::_thr_setmutator;
4727int_fnP_thread_t os::Solaris::_thr_suspend_mutator;
4728int_fnP_thread_t os::Solaris::_thr_continue_mutator;
4729
4730// (Static) wrapper for getisax(2) call.
4731os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4732
4733// (Static) wrappers for the liblgrp API
4734os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4735os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4736os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4737os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4738os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4739os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4740os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4741os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4742os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4743
4744// (Static) wrapper for meminfo() call.
4745os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4746
4747static address resolve_symbol_lazy(const char* name) {
4748  address addr = (address) dlsym(RTLD_DEFAULT, name);
4749  if(addr == NULL) {
4750    // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4751    addr = (address) dlsym(RTLD_NEXT, name);
4752  }
4753  return addr;
4754}
4755
4756static address resolve_symbol(const char* name) {
4757  address addr = resolve_symbol_lazy(name);
4758  if(addr == NULL) {
4759    fatal(dlerror());
4760  }
4761  return addr;
4762}
4763
4764
4765
4766// isT2_libthread()
4767//
4768// Routine to determine if we are currently using the new T2 libthread.
4769//
4770// We determine if we are using T2 by reading /proc/self/lstatus and
4771// looking for a thread with the ASLWP bit set.  If we find this status
4772// bit set, we must assume that we are NOT using T2.  The T2 team
4773// has approved this algorithm.
4774//
4775// We need to determine if we are running with the new T2 libthread
4776// since setting native thread priorities is handled differently
4777// when using this library.  All threads created using T2 are bound
4778// threads. Calling thr_setprio is meaningless in this case.
4779//
4780bool isT2_libthread() {
4781  static prheader_t * lwpArray = NULL;
4782  static int lwpSize = 0;
4783  static int lwpFile = -1;
4784  lwpstatus_t * that;
4785  char lwpName [128];
4786  bool isT2 = false;
4787
4788#define ADR(x)  ((uintptr_t)(x))
4789#define LWPINDEX(ary,ix)   ((lwpstatus_t *)(((ary)->pr_entsize * (ix)) + (ADR((ary) + 1))))
4790
4791  lwpFile = ::open("/proc/self/lstatus", O_RDONLY, 0);
4792  if (lwpFile < 0) {
4793      if (ThreadPriorityVerbose) warning ("Couldn't open /proc/self/lstatus\n");
4794      return false;
4795  }
4796  lwpSize = 16*1024;
4797  for (;;) {
4798    ::lseek64 (lwpFile, 0, SEEK_SET);
4799    lwpArray = (prheader_t *)NEW_C_HEAP_ARRAY(char, lwpSize, mtInternal);
4800    if (::read(lwpFile, lwpArray, lwpSize) < 0) {
4801      if (ThreadPriorityVerbose) warning("Error reading /proc/self/lstatus\n");
4802      break;
4803    }
4804    if ((lwpArray->pr_nent * lwpArray->pr_entsize) <= lwpSize) {
4805       // We got a good snapshot - now iterate over the list.
4806      int aslwpcount = 0;
4807      for (int i = 0; i < lwpArray->pr_nent; i++ ) {
4808        that = LWPINDEX(lwpArray,i);
4809        if (that->pr_flags & PR_ASLWP) {
4810          aslwpcount++;
4811        }
4812      }
4813      if (aslwpcount == 0) isT2 = true;
4814      break;
4815    }
4816    lwpSize = lwpArray->pr_nent * lwpArray->pr_entsize;
4817    FREE_C_HEAP_ARRAY(char, lwpArray, mtInternal);  // retry.
4818  }
4819
4820  FREE_C_HEAP_ARRAY(char, lwpArray, mtInternal);
4821  ::close (lwpFile);
4822  if (ThreadPriorityVerbose) {
4823    if (isT2) tty->print_cr("We are running with a T2 libthread\n");
4824    else tty->print_cr("We are not running with a T2 libthread\n");
4825  }
4826  return isT2;
4827}
4828
4829
4830void os::Solaris::libthread_init() {
4831  address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4832
4833  // Determine if we are running with the new T2 libthread
4834  os::Solaris::set_T2_libthread(isT2_libthread());
4835
4836  lwp_priocntl_init();
4837
4838  // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4839  if(func == NULL) {
4840    func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4841    // Guarantee that this VM is running on an new enough OS (5.6 or
4842    // later) that it will have a new enough libthread.so.
4843    guarantee(func != NULL, "libthread.so is too old.");
4844  }
4845
4846  // Initialize the new libthread getstate API wrappers
4847  func = resolve_symbol("thr_getstate");
4848  os::Solaris::set_thr_getstate(CAST_TO_FN_PTR(int_fnP_thread_t_iP_uP_stack_tP_gregset_t, func));
4849
4850  func = resolve_symbol("thr_setstate");
4851  os::Solaris::set_thr_setstate(CAST_TO_FN_PTR(int_fnP_thread_t_i_gregset_t, func));
4852
4853  func = resolve_symbol("thr_setmutator");
4854  os::Solaris::set_thr_setmutator(CAST_TO_FN_PTR(int_fnP_thread_t_i, func));
4855
4856  func = resolve_symbol("thr_suspend_mutator");
4857  os::Solaris::set_thr_suspend_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4858
4859  func = resolve_symbol("thr_continue_mutator");
4860  os::Solaris::set_thr_continue_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4861
4862  int size;
4863  void (*handler_info_func)(address *, int *);
4864  handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4865  handler_info_func(&handler_start, &size);
4866  handler_end = handler_start + size;
4867}
4868
4869
4870int_fnP_mutex_tP os::Solaris::_mutex_lock;
4871int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4872int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4873int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4874int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4875int os::Solaris::_mutex_scope = USYNC_THREAD;
4876
4877int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4878int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4879int_fnP_cond_tP os::Solaris::_cond_signal;
4880int_fnP_cond_tP os::Solaris::_cond_broadcast;
4881int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4882int_fnP_cond_tP os::Solaris::_cond_destroy;
4883int os::Solaris::_cond_scope = USYNC_THREAD;
4884
4885void os::Solaris::synchronization_init() {
4886  if(UseLWPSynchronization) {
4887    os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4888    os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4889    os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4890    os::Solaris::set_mutex_init(lwp_mutex_init);
4891    os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4892    os::Solaris::set_mutex_scope(USYNC_THREAD);
4893
4894    os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4895    os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4896    os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4897    os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4898    os::Solaris::set_cond_init(lwp_cond_init);
4899    os::Solaris::set_cond_destroy(lwp_cond_destroy);
4900    os::Solaris::set_cond_scope(USYNC_THREAD);
4901  }
4902  else {
4903    os::Solaris::set_mutex_scope(USYNC_THREAD);
4904    os::Solaris::set_cond_scope(USYNC_THREAD);
4905
4906    if(UsePthreads) {
4907      os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4908      os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4909      os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4910      os::Solaris::set_mutex_init(pthread_mutex_default_init);
4911      os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4912
4913      os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4914      os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4915      os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4916      os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4917      os::Solaris::set_cond_init(pthread_cond_default_init);
4918      os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4919    }
4920    else {
4921      os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4922      os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4923      os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4924      os::Solaris::set_mutex_init(::mutex_init);
4925      os::Solaris::set_mutex_destroy(::mutex_destroy);
4926
4927      os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4928      os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4929      os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4930      os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4931      os::Solaris::set_cond_init(::cond_init);
4932      os::Solaris::set_cond_destroy(::cond_destroy);
4933    }
4934  }
4935}
4936
4937bool os::Solaris::liblgrp_init() {
4938  void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4939  if (handle != NULL) {
4940    os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4941    os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4942    os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4943    os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4944    os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4945    os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4946    os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4947    os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4948                                       dlsym(handle, "lgrp_cookie_stale")));
4949
4950    lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4951    set_lgrp_cookie(c);
4952    return true;
4953  }
4954  return false;
4955}
4956
4957void os::Solaris::misc_sym_init() {
4958  address func;
4959
4960  // getisax
4961  func = resolve_symbol_lazy("getisax");
4962  if (func != NULL) {
4963    os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
4964  }
4965
4966  // meminfo
4967  func = resolve_symbol_lazy("meminfo");
4968  if (func != NULL) {
4969    os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
4970  }
4971}
4972
4973uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
4974  assert(_getisax != NULL, "_getisax not set");
4975  return _getisax(array, n);
4976}
4977
4978// Symbol doesn't exist in Solaris 8 pset.h
4979#ifndef PS_MYID
4980#define PS_MYID -3
4981#endif
4982
4983// int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4984typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4985static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4986
4987void init_pset_getloadavg_ptr(void) {
4988  pset_getloadavg_ptr =
4989    (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4990  if (PrintMiscellaneous && Verbose && pset_getloadavg_ptr == NULL) {
4991    warning("pset_getloadavg function not found");
4992  }
4993}
4994
4995int os::Solaris::_dev_zero_fd = -1;
4996
4997// this is called _before_ the global arguments have been parsed
4998void os::init(void) {
4999  _initial_pid = getpid();
5000
5001  max_hrtime = first_hrtime = gethrtime();
5002
5003  init_random(1234567);
5004
5005  page_size = sysconf(_SC_PAGESIZE);
5006  if (page_size == -1)
5007    fatal(err_msg("os_solaris.cpp: os::init: sysconf failed (%s)",
5008                  strerror(errno)));
5009  init_page_sizes((size_t) page_size);
5010
5011  Solaris::initialize_system_info();
5012
5013  // Initialize misc. symbols as soon as possible, so we can use them
5014  // if we need them.
5015  Solaris::misc_sym_init();
5016
5017  int fd = ::open("/dev/zero", O_RDWR);
5018  if (fd < 0) {
5019    fatal(err_msg("os::init: cannot open /dev/zero (%s)", strerror(errno)));
5020  } else {
5021    Solaris::set_dev_zero_fd(fd);
5022
5023    // Close on exec, child won't inherit.
5024    fcntl(fd, F_SETFD, FD_CLOEXEC);
5025  }
5026
5027  clock_tics_per_sec = CLK_TCK;
5028
5029  // check if dladdr1() exists; dladdr1 can provide more information than
5030  // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
5031  // and is available on linker patches for 5.7 and 5.8.
5032  // libdl.so must have been loaded, this call is just an entry lookup
5033  void * hdl = dlopen("libdl.so", RTLD_NOW);
5034  if (hdl)
5035    dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
5036
5037  // (Solaris only) this switches to calls that actually do locking.
5038  ThreadCritical::initialize();
5039
5040  main_thread = thr_self();
5041
5042  // Constant minimum stack size allowed. It must be at least
5043  // the minimum of what the OS supports (thr_min_stack()), and
5044  // enough to allow the thread to get to user bytecode execution.
5045  Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
5046  // If the pagesize of the VM is greater than 8K determine the appropriate
5047  // number of initial guard pages.  The user can change this with the
5048  // command line arguments, if needed.
5049  if (vm_page_size() > 8*K) {
5050    StackYellowPages = 1;
5051    StackRedPages = 1;
5052    StackShadowPages = round_to((StackShadowPages*8*K), vm_page_size()) / vm_page_size();
5053  }
5054}
5055
5056// To install functions for atexit system call
5057extern "C" {
5058  static void perfMemory_exit_helper() {
5059    perfMemory_exit();
5060  }
5061}
5062
5063// this is called _after_ the global arguments have been parsed
5064jint os::init_2(void) {
5065  // try to enable extended file IO ASAP, see 6431278
5066  os::Solaris::try_enable_extended_io();
5067
5068  // Allocate a single page and mark it as readable for safepoint polling.  Also
5069  // use this first mmap call to check support for MAP_ALIGN.
5070  address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
5071                                                      page_size,
5072                                                      MAP_PRIVATE | MAP_ALIGN,
5073                                                      PROT_READ);
5074  if (polling_page == NULL) {
5075    has_map_align = false;
5076    polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
5077                                                PROT_READ);
5078  }
5079
5080  os::set_polling_page(polling_page);
5081
5082#ifndef PRODUCT
5083  if( Verbose && PrintMiscellaneous )
5084    tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5085#endif
5086
5087  if (!UseMembar) {
5088    address mem_serialize_page = (address)Solaris::mmap_chunk( NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE );
5089    guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
5090    os::set_memory_serialize_page( mem_serialize_page );
5091
5092#ifndef PRODUCT
5093    if(Verbose && PrintMiscellaneous)
5094      tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5095#endif
5096}
5097
5098  os::large_page_init();
5099
5100  // Check minimum allowable stack size for thread creation and to initialize
5101  // the java system classes, including StackOverflowError - depends on page
5102  // size.  Add a page for compiler2 recursion in main thread.
5103  // Add in 2*BytesPerWord times page size to account for VM stack during
5104  // class initialization depending on 32 or 64 bit VM.
5105  os::Solaris::min_stack_allowed = MAX2(os::Solaris::min_stack_allowed,
5106            (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
5107                    2*BytesPerWord COMPILER2_PRESENT(+1)) * page_size);
5108
5109  size_t threadStackSizeInBytes = ThreadStackSize * K;
5110  if (threadStackSizeInBytes != 0 &&
5111    threadStackSizeInBytes < os::Solaris::min_stack_allowed) {
5112    tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
5113                  os::Solaris::min_stack_allowed/K);
5114    return JNI_ERR;
5115  }
5116
5117  // For 64kbps there will be a 64kb page size, which makes
5118  // the usable default stack size quite a bit less.  Increase the
5119  // stack for 64kb (or any > than 8kb) pages, this increases
5120  // virtual memory fragmentation (since we're not creating the
5121  // stack on a power of 2 boundary.  The real fix for this
5122  // should be to fix the guard page mechanism.
5123
5124  if (vm_page_size() > 8*K) {
5125      threadStackSizeInBytes = (threadStackSizeInBytes != 0)
5126         ? threadStackSizeInBytes +
5127           ((StackYellowPages + StackRedPages) * vm_page_size())
5128         : 0;
5129      ThreadStackSize = threadStackSizeInBytes/K;
5130  }
5131
5132  // Make the stack size a multiple of the page size so that
5133  // the yellow/red zones can be guarded.
5134  JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5135        vm_page_size()));
5136
5137  Solaris::libthread_init();
5138
5139  if (UseNUMA) {
5140    if (!Solaris::liblgrp_init()) {
5141      UseNUMA = false;
5142    } else {
5143      size_t lgrp_limit = os::numa_get_groups_num();
5144      int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
5145      size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
5146      FREE_C_HEAP_ARRAY(int, lgrp_ids, mtInternal);
5147      if (lgrp_num < 2) {
5148        // There's only one locality group, disable NUMA.
5149        UseNUMA = false;
5150      }
5151    }
5152    // ISM is not compatible with the NUMA allocator - it always allocates
5153    // pages round-robin across the lgroups.
5154    if (UseNUMA && UseLargePages && UseISM) {
5155      if (!FLAG_IS_DEFAULT(UseNUMA)) {
5156        if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseISM)) {
5157          UseLargePages = false;
5158        } else {
5159          warning("UseNUMA is not compatible with ISM large pages, disabling NUMA allocator");
5160          UseNUMA = false;
5161        }
5162      } else {
5163        UseNUMA = false;
5164      }
5165    }
5166    if (!UseNUMA && ForceNUMA) {
5167      UseNUMA = true;
5168    }
5169  }
5170
5171  Solaris::signal_sets_init();
5172  Solaris::init_signal_mem();
5173  Solaris::install_signal_handlers();
5174
5175  if (libjsigversion < JSIG_VERSION_1_4_1) {
5176    Maxlibjsigsigs = OLDMAXSIGNUM;
5177  }
5178
5179  // initialize synchronization primitives to use either thread or
5180  // lwp synchronization (controlled by UseLWPSynchronization)
5181  Solaris::synchronization_init();
5182
5183  if (MaxFDLimit) {
5184    // set the number of file descriptors to max. print out error
5185    // if getrlimit/setrlimit fails but continue regardless.
5186    struct rlimit nbr_files;
5187    int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5188    if (status != 0) {
5189      if (PrintMiscellaneous && (Verbose || WizardMode))
5190        perror("os::init_2 getrlimit failed");
5191    } else {
5192      nbr_files.rlim_cur = nbr_files.rlim_max;
5193      status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5194      if (status != 0) {
5195        if (PrintMiscellaneous && (Verbose || WizardMode))
5196          perror("os::init_2 setrlimit failed");
5197      }
5198    }
5199  }
5200
5201  // Calculate theoretical max. size of Threads to guard gainst
5202  // artifical out-of-memory situations, where all available address-
5203  // space has been reserved by thread stacks. Default stack size is 1Mb.
5204  size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
5205    JavaThread::stack_size_at_create() : (1*K*K);
5206  assert(pre_thread_stack_size != 0, "Must have a stack");
5207  // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
5208  // we should start doing Virtual Memory banging. Currently when the threads will
5209  // have used all but 200Mb of space.
5210  size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
5211  Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
5212
5213  // at-exit methods are called in the reverse order of their registration.
5214  // In Solaris 7 and earlier, atexit functions are called on return from
5215  // main or as a result of a call to exit(3C). There can be only 32 of
5216  // these functions registered and atexit() does not set errno. In Solaris
5217  // 8 and later, there is no limit to the number of functions registered
5218  // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
5219  // functions are called upon dlclose(3DL) in addition to return from main
5220  // and exit(3C).
5221
5222  if (PerfAllowAtExitRegistration) {
5223    // only register atexit functions if PerfAllowAtExitRegistration is set.
5224    // atexit functions can be delayed until process exit time, which
5225    // can be problematic for embedded VM situations. Embedded VMs should
5226    // call DestroyJavaVM() to assure that VM resources are released.
5227
5228    // note: perfMemory_exit_helper atexit function may be removed in
5229    // the future if the appropriate cleanup code can be added to the
5230    // VM_Exit VMOperation's doit method.
5231    if (atexit(perfMemory_exit_helper) != 0) {
5232      warning("os::init2 atexit(perfMemory_exit_helper) failed");
5233    }
5234  }
5235
5236  // Init pset_loadavg function pointer
5237  init_pset_getloadavg_ptr();
5238
5239  return JNI_OK;
5240}
5241
5242void os::init_3(void) {
5243  return;
5244}
5245
5246// Mark the polling page as unreadable
5247void os::make_polling_page_unreadable(void) {
5248  if( mprotect((char *)_polling_page, page_size, PROT_NONE) != 0 )
5249    fatal("Could not disable polling page");
5250};
5251
5252// Mark the polling page as readable
5253void os::make_polling_page_readable(void) {
5254  if( mprotect((char *)_polling_page, page_size, PROT_READ) != 0 )
5255    fatal("Could not enable polling page");
5256};
5257
5258// OS interface.
5259
5260bool os::check_heap(bool force) { return true; }
5261
5262typedef int (*vsnprintf_t)(char* buf, size_t count, const char* fmt, va_list argptr);
5263static vsnprintf_t sol_vsnprintf = NULL;
5264
5265int local_vsnprintf(char* buf, size_t count, const char* fmt, va_list argptr) {
5266  if (!sol_vsnprintf) {
5267    //search  for the named symbol in the objects that were loaded after libjvm
5268    void* where = RTLD_NEXT;
5269    if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5270        sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5271    if (!sol_vsnprintf){
5272      //search  for the named symbol in the objects that were loaded before libjvm
5273      where = RTLD_DEFAULT;
5274      if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5275        sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5276      assert(sol_vsnprintf != NULL, "vsnprintf not found");
5277    }
5278  }
5279  return (*sol_vsnprintf)(buf, count, fmt, argptr);
5280}
5281
5282
5283// Is a (classpath) directory empty?
5284bool os::dir_is_empty(const char* path) {
5285  DIR *dir = NULL;
5286  struct dirent *ptr;
5287
5288  dir = opendir(path);
5289  if (dir == NULL) return true;
5290
5291  /* Scan the directory */
5292  bool result = true;
5293  char buf[sizeof(struct dirent) + MAX_PATH];
5294  struct dirent *dbuf = (struct dirent *) buf;
5295  while (result && (ptr = readdir(dir, dbuf)) != NULL) {
5296    if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5297      result = false;
5298    }
5299  }
5300  closedir(dir);
5301  return result;
5302}
5303
5304// This code originates from JDK's sysOpen and open64_w
5305// from src/solaris/hpi/src/system_md.c
5306
5307#ifndef O_DELETE
5308#define O_DELETE 0x10000
5309#endif
5310
5311// Open a file. Unlink the file immediately after open returns
5312// if the specified oflag has the O_DELETE flag set.
5313// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5314
5315int os::open(const char *path, int oflag, int mode) {
5316  if (strlen(path) > MAX_PATH - 1) {
5317    errno = ENAMETOOLONG;
5318    return -1;
5319  }
5320  int fd;
5321  int o_delete = (oflag & O_DELETE);
5322  oflag = oflag & ~O_DELETE;
5323
5324  fd = ::open64(path, oflag, mode);
5325  if (fd == -1) return -1;
5326
5327  //If the open succeeded, the file might still be a directory
5328  {
5329    struct stat64 buf64;
5330    int ret = ::fstat64(fd, &buf64);
5331    int st_mode = buf64.st_mode;
5332
5333    if (ret != -1) {
5334      if ((st_mode & S_IFMT) == S_IFDIR) {
5335        errno = EISDIR;
5336        ::close(fd);
5337        return -1;
5338      }
5339    } else {
5340      ::close(fd);
5341      return -1;
5342    }
5343  }
5344    /*
5345     * 32-bit Solaris systems suffer from:
5346     *
5347     * - an historical default soft limit of 256 per-process file
5348     *   descriptors that is too low for many Java programs.
5349     *
5350     * - a design flaw where file descriptors created using stdio
5351     *   fopen must be less than 256, _even_ when the first limit above
5352     *   has been raised.  This can cause calls to fopen (but not calls to
5353     *   open, for example) to fail mysteriously, perhaps in 3rd party
5354     *   native code (although the JDK itself uses fopen).  One can hardly
5355     *   criticize them for using this most standard of all functions.
5356     *
5357     * We attempt to make everything work anyways by:
5358     *
5359     * - raising the soft limit on per-process file descriptors beyond
5360     *   256
5361     *
5362     * - As of Solaris 10u4, we can request that Solaris raise the 256
5363     *   stdio fopen limit by calling function enable_extended_FILE_stdio.
5364     *   This is done in init_2 and recorded in enabled_extended_FILE_stdio
5365     *
5366     * - If we are stuck on an old (pre 10u4) Solaris system, we can
5367     *   workaround the bug by remapping non-stdio file descriptors below
5368     *   256 to ones beyond 256, which is done below.
5369     *
5370     * See:
5371     * 1085341: 32-bit stdio routines should support file descriptors >255
5372     * 6533291: Work around 32-bit Solaris stdio limit of 256 open files
5373     * 6431278: Netbeans crash on 32 bit Solaris: need to call
5374     *          enable_extended_FILE_stdio() in VM initialisation
5375     * Giri Mandalika's blog
5376     * http://technopark02.blogspot.com/2005_05_01_archive.html
5377     */
5378#ifndef  _LP64
5379     if ((!enabled_extended_FILE_stdio) && fd < 256) {
5380         int newfd = ::fcntl(fd, F_DUPFD, 256);
5381         if (newfd != -1) {
5382             ::close(fd);
5383             fd = newfd;
5384         }
5385     }
5386#endif // 32-bit Solaris
5387    /*
5388     * All file descriptors that are opened in the JVM and not
5389     * specifically destined for a subprocess should have the
5390     * close-on-exec flag set.  If we don't set it, then careless 3rd
5391     * party native code might fork and exec without closing all
5392     * appropriate file descriptors (e.g. as we do in closeDescriptors in
5393     * UNIXProcess.c), and this in turn might:
5394     *
5395     * - cause end-of-file to fail to be detected on some file
5396     *   descriptors, resulting in mysterious hangs, or
5397     *
5398     * - might cause an fopen in the subprocess to fail on a system
5399     *   suffering from bug 1085341.
5400     *
5401     * (Yes, the default setting of the close-on-exec flag is a Unix
5402     * design flaw)
5403     *
5404     * See:
5405     * 1085341: 32-bit stdio routines should support file descriptors >255
5406     * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5407     * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5408     */
5409#ifdef FD_CLOEXEC
5410    {
5411        int flags = ::fcntl(fd, F_GETFD);
5412        if (flags != -1)
5413            ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5414    }
5415#endif
5416
5417  if (o_delete != 0) {
5418    ::unlink(path);
5419  }
5420  return fd;
5421}
5422
5423// create binary file, rewriting existing file if required
5424int os::create_binary_file(const char* path, bool rewrite_existing) {
5425  int oflags = O_WRONLY | O_CREAT;
5426  if (!rewrite_existing) {
5427    oflags |= O_EXCL;
5428  }
5429  return ::open64(path, oflags, S_IREAD | S_IWRITE);
5430}
5431
5432// return current position of file pointer
5433jlong os::current_file_offset(int fd) {
5434  return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5435}
5436
5437// move file pointer to the specified offset
5438jlong os::seek_to_file_offset(int fd, jlong offset) {
5439  return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5440}
5441
5442jlong os::lseek(int fd, jlong offset, int whence) {
5443  return (jlong) ::lseek64(fd, offset, whence);
5444}
5445
5446char * os::native_path(char *path) {
5447  return path;
5448}
5449
5450int os::ftruncate(int fd, jlong length) {
5451  return ::ftruncate64(fd, length);
5452}
5453
5454int os::fsync(int fd)  {
5455  RESTARTABLE_RETURN_INT(::fsync(fd));
5456}
5457
5458int os::available(int fd, jlong *bytes) {
5459  jlong cur, end;
5460  int mode;
5461  struct stat64 buf64;
5462
5463  if (::fstat64(fd, &buf64) >= 0) {
5464    mode = buf64.st_mode;
5465    if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5466      /*
5467      * XXX: is the following call interruptible? If so, this might
5468      * need to go through the INTERRUPT_IO() wrapper as for other
5469      * blocking, interruptible calls in this file.
5470      */
5471      int n,ioctl_return;
5472
5473      INTERRUPTIBLE(::ioctl(fd, FIONREAD, &n),ioctl_return,os::Solaris::clear_interrupted);
5474      if (ioctl_return>= 0) {
5475          *bytes = n;
5476        return 1;
5477      }
5478    }
5479  }
5480  if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5481    return 0;
5482  } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5483    return 0;
5484  } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5485    return 0;
5486  }
5487  *bytes = end - cur;
5488  return 1;
5489}
5490
5491// Map a block of memory.
5492char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5493                     char *addr, size_t bytes, bool read_only,
5494                     bool allow_exec) {
5495  int prot;
5496  int flags;
5497
5498  if (read_only) {
5499    prot = PROT_READ;
5500    flags = MAP_SHARED;
5501  } else {
5502    prot = PROT_READ | PROT_WRITE;
5503    flags = MAP_PRIVATE;
5504  }
5505
5506  if (allow_exec) {
5507    prot |= PROT_EXEC;
5508  }
5509
5510  if (addr != NULL) {
5511    flags |= MAP_FIXED;
5512  }
5513
5514  char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5515                                     fd, file_offset);
5516  if (mapped_address == MAP_FAILED) {
5517    return NULL;
5518  }
5519  return mapped_address;
5520}
5521
5522
5523// Remap a block of memory.
5524char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5525                       char *addr, size_t bytes, bool read_only,
5526                       bool allow_exec) {
5527  // same as map_memory() on this OS
5528  return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5529                        allow_exec);
5530}
5531
5532
5533// Unmap a block of memory.
5534bool os::pd_unmap_memory(char* addr, size_t bytes) {
5535  return munmap(addr, bytes) == 0;
5536}
5537
5538void os::pause() {
5539  char filename[MAX_PATH];
5540  if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5541    jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5542  } else {
5543    jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5544  }
5545
5546  int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5547  if (fd != -1) {
5548    struct stat buf;
5549    ::close(fd);
5550    while (::stat(filename, &buf) == 0) {
5551      (void)::poll(NULL, 0, 100);
5552    }
5553  } else {
5554    jio_fprintf(stderr,
5555      "Could not open pause file '%s', continuing immediately.\n", filename);
5556  }
5557}
5558
5559#ifndef PRODUCT
5560#ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5561// Turn this on if you need to trace synch operations.
5562// Set RECORD_SYNCH_LIMIT to a large-enough value,
5563// and call record_synch_enable and record_synch_disable
5564// around the computation of interest.
5565
5566void record_synch(char* name, bool returning);  // defined below
5567
5568class RecordSynch {
5569  char* _name;
5570 public:
5571  RecordSynch(char* name) :_name(name)
5572                 { record_synch(_name, false); }
5573  ~RecordSynch() { record_synch(_name,   true);  }
5574};
5575
5576#define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
5577extern "C" ret name params {                                    \
5578  typedef ret name##_t params;                                  \
5579  static name##_t* implem = NULL;                               \
5580  static int callcount = 0;                                     \
5581  if (implem == NULL) {                                         \
5582    implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
5583    if (implem == NULL)  fatal(dlerror());                      \
5584  }                                                             \
5585  ++callcount;                                                  \
5586  RecordSynch _rs(#name);                                       \
5587  inner;                                                        \
5588  return implem args;                                           \
5589}
5590// in dbx, examine callcounts this way:
5591// for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
5592
5593#define CHECK_POINTER_OK(p) \
5594  (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
5595#define CHECK_MU \
5596  if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
5597#define CHECK_CV \
5598  if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
5599#define CHECK_P(p) \
5600  if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
5601
5602#define CHECK_MUTEX(mutex_op) \
5603CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
5604
5605CHECK_MUTEX(   mutex_lock)
5606CHECK_MUTEX(  _mutex_lock)
5607CHECK_MUTEX( mutex_unlock)
5608CHECK_MUTEX(_mutex_unlock)
5609CHECK_MUTEX( mutex_trylock)
5610CHECK_MUTEX(_mutex_trylock)
5611
5612#define CHECK_COND(cond_op) \
5613CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU;CHECK_CV);
5614
5615CHECK_COND( cond_wait);
5616CHECK_COND(_cond_wait);
5617CHECK_COND(_cond_wait_cancel);
5618
5619#define CHECK_COND2(cond_op) \
5620CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU;CHECK_CV);
5621
5622CHECK_COND2( cond_timedwait);
5623CHECK_COND2(_cond_timedwait);
5624CHECK_COND2(_cond_timedwait_cancel);
5625
5626// do the _lwp_* versions too
5627#define mutex_t lwp_mutex_t
5628#define cond_t  lwp_cond_t
5629CHECK_MUTEX(  _lwp_mutex_lock)
5630CHECK_MUTEX(  _lwp_mutex_unlock)
5631CHECK_MUTEX(  _lwp_mutex_trylock)
5632CHECK_MUTEX( __lwp_mutex_lock)
5633CHECK_MUTEX( __lwp_mutex_unlock)
5634CHECK_MUTEX( __lwp_mutex_trylock)
5635CHECK_MUTEX(___lwp_mutex_lock)
5636CHECK_MUTEX(___lwp_mutex_unlock)
5637
5638CHECK_COND(  _lwp_cond_wait);
5639CHECK_COND( __lwp_cond_wait);
5640CHECK_COND(___lwp_cond_wait);
5641
5642CHECK_COND2(  _lwp_cond_timedwait);
5643CHECK_COND2( __lwp_cond_timedwait);
5644#undef mutex_t
5645#undef cond_t
5646
5647CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5648CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5649CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5650CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5651CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5652CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5653CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5654CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5655
5656
5657// recording machinery:
5658
5659enum { RECORD_SYNCH_LIMIT = 200 };
5660char* record_synch_name[RECORD_SYNCH_LIMIT];
5661void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
5662bool record_synch_returning[RECORD_SYNCH_LIMIT];
5663thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
5664int record_synch_count = 0;
5665bool record_synch_enabled = false;
5666
5667// in dbx, examine recorded data this way:
5668// for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
5669
5670void record_synch(char* name, bool returning) {
5671  if (record_synch_enabled) {
5672    if (record_synch_count < RECORD_SYNCH_LIMIT) {
5673      record_synch_name[record_synch_count] = name;
5674      record_synch_returning[record_synch_count] = returning;
5675      record_synch_thread[record_synch_count] = thr_self();
5676      record_synch_arg0ptr[record_synch_count] = &name;
5677      record_synch_count++;
5678    }
5679    // put more checking code here:
5680    // ...
5681  }
5682}
5683
5684void record_synch_enable() {
5685  // start collecting trace data, if not already doing so
5686  if (!record_synch_enabled)  record_synch_count = 0;
5687  record_synch_enabled = true;
5688}
5689
5690void record_synch_disable() {
5691  // stop collecting trace data
5692  record_synch_enabled = false;
5693}
5694
5695#endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5696#endif // PRODUCT
5697
5698const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5699const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
5700                               (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5701
5702
5703// JVMTI & JVM monitoring and management support
5704// The thread_cpu_time() and current_thread_cpu_time() are only
5705// supported if is_thread_cpu_time_supported() returns true.
5706// They are not supported on Solaris T1.
5707
5708// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5709// are used by JVM M&M and JVMTI to get user+sys or user CPU time
5710// of a thread.
5711//
5712// current_thread_cpu_time() and thread_cpu_time(Thread *)
5713// returns the fast estimate available on the platform.
5714
5715// hrtime_t gethrvtime() return value includes
5716// user time but does not include system time
5717jlong os::current_thread_cpu_time() {
5718  return (jlong) gethrvtime();
5719}
5720
5721jlong os::thread_cpu_time(Thread *thread) {
5722  // return user level CPU time only to be consistent with
5723  // what current_thread_cpu_time returns.
5724  // thread_cpu_time_info() must be changed if this changes
5725  return os::thread_cpu_time(thread, false /* user time only */);
5726}
5727
5728jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5729  if (user_sys_cpu_time) {
5730    return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5731  } else {
5732    return os::current_thread_cpu_time();
5733  }
5734}
5735
5736jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5737  char proc_name[64];
5738  int count;
5739  prusage_t prusage;
5740  jlong lwp_time;
5741  int fd;
5742
5743  sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5744                     getpid(),
5745                     thread->osthread()->lwp_id());
5746  fd = ::open(proc_name, O_RDONLY);
5747  if ( fd == -1 ) return -1;
5748
5749  do {
5750    count = ::pread(fd,
5751                  (void *)&prusage.pr_utime,
5752                  thr_time_size,
5753                  thr_time_off);
5754  } while (count < 0 && errno == EINTR);
5755  ::close(fd);
5756  if ( count < 0 ) return -1;
5757
5758  if (user_sys_cpu_time) {
5759    // user + system CPU time
5760    lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5761                 (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5762                 (jlong)prusage.pr_stime.tv_nsec +
5763                 (jlong)prusage.pr_utime.tv_nsec;
5764  } else {
5765    // user level CPU time only
5766    lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5767                (jlong)prusage.pr_utime.tv_nsec;
5768  }
5769
5770  return(lwp_time);
5771}
5772
5773void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5774  info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5775  info_ptr->may_skip_backward = false;    // elapsed time not wall time
5776  info_ptr->may_skip_forward = false;     // elapsed time not wall time
5777  info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5778}
5779
5780void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5781  info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5782  info_ptr->may_skip_backward = false;    // elapsed time not wall time
5783  info_ptr->may_skip_forward = false;     // elapsed time not wall time
5784  info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5785}
5786
5787bool os::is_thread_cpu_time_supported() {
5788  if ( os::Solaris::T2_libthread() || UseBoundThreads ) {
5789    return true;
5790  } else {
5791    return false;
5792  }
5793}
5794
5795// System loadavg support.  Returns -1 if load average cannot be obtained.
5796// Return the load average for our processor set if the primitive exists
5797// (Solaris 9 and later).  Otherwise just return system wide loadavg.
5798int os::loadavg(double loadavg[], int nelem) {
5799  if (pset_getloadavg_ptr != NULL) {
5800    return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5801  } else {
5802    return ::getloadavg(loadavg, nelem);
5803  }
5804}
5805
5806//---------------------------------------------------------------------------------
5807
5808static address same_page(address x, address y) {
5809  intptr_t page_bits = -os::vm_page_size();
5810  if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
5811    return x;
5812  else if (x > y)
5813    return (address)(intptr_t(y) | ~page_bits) + 1;
5814  else
5815    return (address)(intptr_t(y) & page_bits);
5816}
5817
5818bool os::find(address addr, outputStream* st) {
5819  Dl_info dlinfo;
5820  memset(&dlinfo, 0, sizeof(dlinfo));
5821  if (dladdr(addr, &dlinfo)) {
5822#ifdef _LP64
5823    st->print("0x%016lx: ", addr);
5824#else
5825    st->print("0x%08x: ", addr);
5826#endif
5827    if (dlinfo.dli_sname != NULL)
5828      st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5829    else if (dlinfo.dli_fname)
5830      st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5831    else
5832      st->print("<absolute address>");
5833    if (dlinfo.dli_fname)  st->print(" in %s", dlinfo.dli_fname);
5834#ifdef _LP64
5835    if (dlinfo.dli_fbase)  st->print(" at 0x%016lx", dlinfo.dli_fbase);
5836#else
5837    if (dlinfo.dli_fbase)  st->print(" at 0x%08x", dlinfo.dli_fbase);
5838#endif
5839    st->cr();
5840
5841    if (Verbose) {
5842      // decode some bytes around the PC
5843      address begin = same_page(addr-40, addr);
5844      address end   = same_page(addr+40, addr);
5845      address       lowest = (address) dlinfo.dli_sname;
5846      if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5847      if (begin < lowest)  begin = lowest;
5848      Dl_info dlinfo2;
5849      if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
5850          && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5851        end = (address) dlinfo2.dli_saddr;
5852      Disassembler::decode(begin, end, st);
5853    }
5854    return true;
5855  }
5856  return false;
5857}
5858
5859// Following function has been added to support HotSparc's libjvm.so running
5860// under Solaris production JDK 1.2.2 / 1.3.0.  These came from
5861// src/solaris/hpi/native_threads in the EVM codebase.
5862//
5863// NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5864// libraries and should thus be removed. We will leave it behind for a while
5865// until we no longer want to able to run on top of 1.3.0 Solaris production
5866// JDK. See 4341971.
5867
5868#define STACK_SLACK 0x800
5869
5870extern "C" {
5871  intptr_t sysThreadAvailableStackWithSlack() {
5872    stack_t st;
5873    intptr_t retval, stack_top;
5874    retval = thr_stksegment(&st);
5875    assert(retval == 0, "incorrect return value from thr_stksegment");
5876    assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5877    assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5878    stack_top=(intptr_t)st.ss_sp-st.ss_size;
5879    return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5880  }
5881}
5882
5883// ObjectMonitor park-unpark infrastructure ...
5884//
5885// We implement Solaris and Linux PlatformEvents with the
5886// obvious condvar-mutex-flag triple.
5887// Another alternative that works quite well is pipes:
5888// Each PlatformEvent consists of a pipe-pair.
5889// The thread associated with the PlatformEvent
5890// calls park(), which reads from the input end of the pipe.
5891// Unpark() writes into the other end of the pipe.
5892// The write-side of the pipe must be set NDELAY.
5893// Unfortunately pipes consume a large # of handles.
5894// Native solaris lwp_park() and lwp_unpark() work nicely, too.
5895// Using pipes for the 1st few threads might be workable, however.
5896//
5897// park() is permitted to return spuriously.
5898// Callers of park() should wrap the call to park() in
5899// an appropriate loop.  A litmus test for the correct
5900// usage of park is the following: if park() were modified
5901// to immediately return 0 your code should still work,
5902// albeit degenerating to a spin loop.
5903//
5904// An interesting optimization for park() is to use a trylock()
5905// to attempt to acquire the mutex.  If the trylock() fails
5906// then we know that a concurrent unpark() operation is in-progress.
5907// in that case the park() code could simply set _count to 0
5908// and return immediately.  The subsequent park() operation *might*
5909// return immediately.  That's harmless as the caller of park() is
5910// expected to loop.  By using trylock() we will have avoided a
5911// avoided a context switch caused by contention on the per-thread mutex.
5912//
5913// TODO-FIXME:
5914// 1.  Reconcile Doug's JSR166 j.u.c park-unpark with the
5915//     objectmonitor implementation.
5916// 2.  Collapse the JSR166 parker event, and the
5917//     objectmonitor ParkEvent into a single "Event" construct.
5918// 3.  In park() and unpark() add:
5919//     assert (Thread::current() == AssociatedWith).
5920// 4.  add spurious wakeup injection on a -XX:EarlyParkReturn=N switch.
5921//     1-out-of-N park() operations will return immediately.
5922//
5923// _Event transitions in park()
5924//   -1 => -1 : illegal
5925//    1 =>  0 : pass - return immediately
5926//    0 => -1 : block
5927//
5928// _Event serves as a restricted-range semaphore.
5929//
5930// Another possible encoding of _Event would be with
5931// explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5932//
5933// TODO-FIXME: add DTRACE probes for:
5934// 1.   Tx parks
5935// 2.   Ty unparks Tx
5936// 3.   Tx resumes from park
5937
5938
5939// value determined through experimentation
5940#define ROUNDINGFIX 11
5941
5942// utility to compute the abstime argument to timedwait.
5943// TODO-FIXME: switch from compute_abstime() to unpackTime().
5944
5945static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5946  // millis is the relative timeout time
5947  // abstime will be the absolute timeout time
5948  if (millis < 0)  millis = 0;
5949  struct timeval now;
5950  int status = gettimeofday(&now, NULL);
5951  assert(status == 0, "gettimeofday");
5952  jlong seconds = millis / 1000;
5953  jlong max_wait_period;
5954
5955  if (UseLWPSynchronization) {
5956    // forward port of fix for 4275818 (not sleeping long enough)
5957    // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5958    // _lwp_cond_timedwait() used a round_down algorithm rather
5959    // than a round_up. For millis less than our roundfactor
5960    // it rounded down to 0 which doesn't meet the spec.
5961    // For millis > roundfactor we may return a bit sooner, but
5962    // since we can not accurately identify the patch level and
5963    // this has already been fixed in Solaris 9 and 8 we will
5964    // leave it alone rather than always rounding down.
5965
5966    if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5967       // It appears that when we go directly through Solaris _lwp_cond_timedwait()
5968           // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
5969           max_wait_period = 21000000;
5970  } else {
5971    max_wait_period = 50000000;
5972  }
5973  millis %= 1000;
5974  if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
5975     seconds = max_wait_period;
5976  }
5977  abstime->tv_sec = now.tv_sec  + seconds;
5978  long       usec = now.tv_usec + millis * 1000;
5979  if (usec >= 1000000) {
5980    abstime->tv_sec += 1;
5981    usec -= 1000000;
5982  }
5983  abstime->tv_nsec = usec * 1000;
5984  return abstime;
5985}
5986
5987// Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5988// Conceptually TryPark() should be equivalent to park(0).
5989
5990int os::PlatformEvent::TryPark() {
5991  for (;;) {
5992    const int v = _Event ;
5993    guarantee ((v == 0) || (v == 1), "invariant") ;
5994    if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5995  }
5996}
5997
5998void os::PlatformEvent::park() {           // AKA: down()
5999  // Invariant: Only the thread associated with the Event/PlatformEvent
6000  // may call park().
6001  int v ;
6002  for (;;) {
6003      v = _Event ;
6004      if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6005  }
6006  guarantee (v >= 0, "invariant") ;
6007  if (v == 0) {
6008     // Do this the hard way by blocking ...
6009     // See http://monaco.sfbay/detail.jsf?cr=5094058.
6010     // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
6011     // Only for SPARC >= V8PlusA
6012#if defined(__sparc) && defined(COMPILER2)
6013     if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6014#endif
6015     int status = os::Solaris::mutex_lock(_mutex);
6016     assert_status(status == 0, status,  "mutex_lock");
6017     guarantee (_nParked == 0, "invariant") ;
6018     ++ _nParked ;
6019     while (_Event < 0) {
6020        // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
6021        // Treat this the same as if the wait was interrupted
6022        // With usr/lib/lwp going to kernel, always handle ETIME
6023        status = os::Solaris::cond_wait(_cond, _mutex);
6024        if (status == ETIME) status = EINTR ;
6025        assert_status(status == 0 || status == EINTR, status, "cond_wait");
6026     }
6027     -- _nParked ;
6028     _Event = 0 ;
6029     status = os::Solaris::mutex_unlock(_mutex);
6030     assert_status(status == 0, status, "mutex_unlock");
6031    // Paranoia to ensure our locked and lock-free paths interact
6032    // correctly with each other.
6033    OrderAccess::fence();
6034  }
6035}
6036
6037int os::PlatformEvent::park(jlong millis) {
6038  guarantee (_nParked == 0, "invariant") ;
6039  int v ;
6040  for (;;) {
6041      v = _Event ;
6042      if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6043  }
6044  guarantee (v >= 0, "invariant") ;
6045  if (v != 0) return OS_OK ;
6046
6047  int ret = OS_TIMEOUT;
6048  timestruc_t abst;
6049  compute_abstime (&abst, millis);
6050
6051  // See http://monaco.sfbay/detail.jsf?cr=5094058.
6052  // For Solaris SPARC set fprs.FEF=0 prior to parking.
6053  // Only for SPARC >= V8PlusA
6054#if defined(__sparc) && defined(COMPILER2)
6055 if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6056#endif
6057  int status = os::Solaris::mutex_lock(_mutex);
6058  assert_status(status == 0, status, "mutex_lock");
6059  guarantee (_nParked == 0, "invariant") ;
6060  ++ _nParked ;
6061  while (_Event < 0) {
6062     int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
6063     assert_status(status == 0 || status == EINTR ||
6064                   status == ETIME || status == ETIMEDOUT,
6065                   status, "cond_timedwait");
6066     if (!FilterSpuriousWakeups) break ;                // previous semantics
6067     if (status == ETIME || status == ETIMEDOUT) break ;
6068     // We consume and ignore EINTR and spurious wakeups.
6069  }
6070  -- _nParked ;
6071  if (_Event >= 0) ret = OS_OK ;
6072  _Event = 0 ;
6073  status = os::Solaris::mutex_unlock(_mutex);
6074  assert_status(status == 0, status, "mutex_unlock");
6075  // Paranoia to ensure our locked and lock-free paths interact
6076  // correctly with each other.
6077  OrderAccess::fence();
6078  return ret;
6079}
6080
6081void os::PlatformEvent::unpark() {
6082  // Transitions for _Event:
6083  //    0 :=> 1
6084  //    1 :=> 1
6085  //   -1 :=> either 0 or 1; must signal target thread
6086  //          That is, we can safely transition _Event from -1 to either
6087  //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
6088  //          unpark() calls.
6089  // See also: "Semaphores in Plan 9" by Mullender & Cox
6090  //
6091  // Note: Forcing a transition from "-1" to "1" on an unpark() means
6092  // that it will take two back-to-back park() calls for the owning
6093  // thread to block. This has the benefit of forcing a spurious return
6094  // from the first park() call after an unpark() call which will help
6095  // shake out uses of park() and unpark() without condition variables.
6096
6097  if (Atomic::xchg(1, &_Event) >= 0) return;
6098
6099  // If the thread associated with the event was parked, wake it.
6100  // Wait for the thread assoc with the PlatformEvent to vacate.
6101  int status = os::Solaris::mutex_lock(_mutex);
6102  assert_status(status == 0, status, "mutex_lock");
6103  int AnyWaiters = _nParked;
6104  status = os::Solaris::mutex_unlock(_mutex);
6105  assert_status(status == 0, status, "mutex_unlock");
6106  guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
6107  if (AnyWaiters != 0) {
6108    // We intentional signal *after* dropping the lock
6109    // to avoid a common class of futile wakeups.
6110    status = os::Solaris::cond_signal(_cond);
6111    assert_status(status == 0, status, "cond_signal");
6112  }
6113}
6114
6115// JSR166
6116// -------------------------------------------------------
6117
6118/*
6119 * The solaris and linux implementations of park/unpark are fairly
6120 * conservative for now, but can be improved. They currently use a
6121 * mutex/condvar pair, plus _counter.
6122 * Park decrements _counter if > 0, else does a condvar wait.  Unpark
6123 * sets count to 1 and signals condvar.  Only one thread ever waits
6124 * on the condvar. Contention seen when trying to park implies that someone
6125 * is unparking you, so don't wait. And spurious returns are fine, so there
6126 * is no need to track notifications.
6127 */
6128
6129#define MAX_SECS 100000000
6130/*
6131 * This code is common to linux and solaris and will be moved to a
6132 * common place in dolphin.
6133 *
6134 * The passed in time value is either a relative time in nanoseconds
6135 * or an absolute time in milliseconds. Either way it has to be unpacked
6136 * into suitable seconds and nanoseconds components and stored in the
6137 * given timespec structure.
6138 * Given time is a 64-bit value and the time_t used in the timespec is only
6139 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6140 * overflow if times way in the future are given. Further on Solaris versions
6141 * prior to 10 there is a restriction (see cond_timedwait) that the specified
6142 * number of seconds, in abstime, is less than current_time  + 100,000,000.
6143 * As it will be 28 years before "now + 100000000" will overflow we can
6144 * ignore overflow and just impose a hard-limit on seconds using the value
6145 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6146 * years from "now".
6147 */
6148static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6149  assert (time > 0, "convertTime");
6150
6151  struct timeval now;
6152  int status = gettimeofday(&now, NULL);
6153  assert(status == 0, "gettimeofday");
6154
6155  time_t max_secs = now.tv_sec + MAX_SECS;
6156
6157  if (isAbsolute) {
6158    jlong secs = time / 1000;
6159    if (secs > max_secs) {
6160      absTime->tv_sec = max_secs;
6161    }
6162    else {
6163      absTime->tv_sec = secs;
6164    }
6165    absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6166  }
6167  else {
6168    jlong secs = time / NANOSECS_PER_SEC;
6169    if (secs >= MAX_SECS) {
6170      absTime->tv_sec = max_secs;
6171      absTime->tv_nsec = 0;
6172    }
6173    else {
6174      absTime->tv_sec = now.tv_sec + secs;
6175      absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6176      if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6177        absTime->tv_nsec -= NANOSECS_PER_SEC;
6178        ++absTime->tv_sec; // note: this must be <= max_secs
6179      }
6180    }
6181  }
6182  assert(absTime->tv_sec >= 0, "tv_sec < 0");
6183  assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6184  assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6185  assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6186}
6187
6188void Parker::park(bool isAbsolute, jlong time) {
6189  // Ideally we'd do something useful while spinning, such
6190  // as calling unpackTime().
6191
6192  // Optional fast-path check:
6193  // Return immediately if a permit is available.
6194  // We depend on Atomic::xchg() having full barrier semantics
6195  // since we are doing a lock-free update to _counter.
6196  if (Atomic::xchg(0, &_counter) > 0) return;
6197
6198  // Optional fast-exit: Check interrupt before trying to wait
6199  Thread* thread = Thread::current();
6200  assert(thread->is_Java_thread(), "Must be JavaThread");
6201  JavaThread *jt = (JavaThread *)thread;
6202  if (Thread::is_interrupted(thread, false)) {
6203    return;
6204  }
6205
6206  // First, demultiplex/decode time arguments
6207  timespec absTime;
6208  if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6209    return;
6210  }
6211  if (time > 0) {
6212    // Warning: this code might be exposed to the old Solaris time
6213    // round-down bugs.  Grep "roundingFix" for details.
6214    unpackTime(&absTime, isAbsolute, time);
6215  }
6216
6217  // Enter safepoint region
6218  // Beware of deadlocks such as 6317397.
6219  // The per-thread Parker:: _mutex is a classic leaf-lock.
6220  // In particular a thread must never block on the Threads_lock while
6221  // holding the Parker:: mutex.  If safepoints are pending both the
6222  // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6223  ThreadBlockInVM tbivm(jt);
6224
6225  // Don't wait if cannot get lock since interference arises from
6226  // unblocking.  Also. check interrupt before trying wait
6227  if (Thread::is_interrupted(thread, false) ||
6228      os::Solaris::mutex_trylock(_mutex) != 0) {
6229    return;
6230  }
6231
6232  int status ;
6233
6234  if (_counter > 0)  { // no wait needed
6235    _counter = 0;
6236    status = os::Solaris::mutex_unlock(_mutex);
6237    assert (status == 0, "invariant") ;
6238    // Paranoia to ensure our locked and lock-free paths interact
6239    // correctly with each other and Java-level accesses.
6240    OrderAccess::fence();
6241    return;
6242  }
6243
6244#ifdef ASSERT
6245  // Don't catch signals while blocked; let the running threads have the signals.
6246  // (This allows a debugger to break into the running thread.)
6247  sigset_t oldsigs;
6248  sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
6249  thr_sigsetmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6250#endif
6251
6252  OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6253  jt->set_suspend_equivalent();
6254  // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6255
6256  // Do this the hard way by blocking ...
6257  // See http://monaco.sfbay/detail.jsf?cr=5094058.
6258  // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
6259  // Only for SPARC >= V8PlusA
6260#if defined(__sparc) && defined(COMPILER2)
6261  if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6262#endif
6263
6264  if (time == 0) {
6265    status = os::Solaris::cond_wait (_cond, _mutex) ;
6266  } else {
6267    status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
6268  }
6269  // Note that an untimed cond_wait() can sometimes return ETIME on older
6270  // versions of the Solaris.
6271  assert_status(status == 0 || status == EINTR ||
6272                status == ETIME || status == ETIMEDOUT,
6273                status, "cond_timedwait");
6274
6275#ifdef ASSERT
6276  thr_sigsetmask(SIG_SETMASK, &oldsigs, NULL);
6277#endif
6278  _counter = 0 ;
6279  status = os::Solaris::mutex_unlock(_mutex);
6280  assert_status(status == 0, status, "mutex_unlock") ;
6281  // Paranoia to ensure our locked and lock-free paths interact
6282  // correctly with each other and Java-level accesses.
6283  OrderAccess::fence();
6284
6285  // If externally suspended while waiting, re-suspend
6286  if (jt->handle_special_suspend_equivalent_condition()) {
6287    jt->java_suspend_self();
6288  }
6289}
6290
6291void Parker::unpark() {
6292  int s, status ;
6293  status = os::Solaris::mutex_lock (_mutex) ;
6294  assert (status == 0, "invariant") ;
6295  s = _counter;
6296  _counter = 1;
6297  status = os::Solaris::mutex_unlock (_mutex) ;
6298  assert (status == 0, "invariant") ;
6299
6300  if (s < 1) {
6301    status = os::Solaris::cond_signal (_cond) ;
6302    assert (status == 0, "invariant") ;
6303  }
6304}
6305
6306extern char** environ;
6307
6308// Run the specified command in a separate process. Return its exit value,
6309// or -1 on failure (e.g. can't fork a new process).
6310// Unlike system(), this function can be called from signal handler. It
6311// doesn't block SIGINT et al.
6312int os::fork_and_exec(char* cmd) {
6313  char * argv[4];
6314  argv[0] = (char *)"sh";
6315  argv[1] = (char *)"-c";
6316  argv[2] = cmd;
6317  argv[3] = NULL;
6318
6319  // fork is async-safe, fork1 is not so can't use in signal handler
6320  pid_t pid;
6321  Thread* t = ThreadLocalStorage::get_thread_slow();
6322  if (t != NULL && t->is_inside_signal_handler()) {
6323    pid = fork();
6324  } else {
6325    pid = fork1();
6326  }
6327
6328  if (pid < 0) {
6329    // fork failed
6330    warning("fork failed: %s", strerror(errno));
6331    return -1;
6332
6333  } else if (pid == 0) {
6334    // child process
6335
6336    // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
6337    execve("/usr/bin/sh", argv, environ);
6338
6339    // execve failed
6340    _exit(-1);
6341
6342  } else  {
6343    // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6344    // care about the actual exit code, for now.
6345
6346    int status;
6347
6348    // Wait for the child process to exit.  This returns immediately if
6349    // the child has already exited. */
6350    while (waitpid(pid, &status, 0) < 0) {
6351        switch (errno) {
6352        case ECHILD: return 0;
6353        case EINTR: break;
6354        default: return -1;
6355        }
6356    }
6357
6358    if (WIFEXITED(status)) {
6359       // The child exited normally; get its exit code.
6360       return WEXITSTATUS(status);
6361    } else if (WIFSIGNALED(status)) {
6362       // The child exited because of a signal
6363       // The best value to return is 0x80 + signal number,
6364       // because that is what all Unix shells do, and because
6365       // it allows callers to distinguish between process exit and
6366       // process death by signal.
6367       return 0x80 + WTERMSIG(status);
6368    } else {
6369       // Unknown exit code; pass it through
6370       return status;
6371    }
6372  }
6373}
6374
6375// is_headless_jre()
6376//
6377// Test for the existence of xawt/libmawt.so or libawt_xawt.so
6378// in order to report if we are running in a headless jre
6379//
6380// Since JDK8 xawt/libmawt.so was moved into the same directory
6381// as libawt.so, and renamed libawt_xawt.so
6382//
6383bool os::is_headless_jre() {
6384    struct stat statbuf;
6385    char buf[MAXPATHLEN];
6386    char libmawtpath[MAXPATHLEN];
6387    const char *xawtstr  = "/xawt/libmawt.so";
6388    const char *new_xawtstr = "/libawt_xawt.so";
6389    char *p;
6390
6391    // Get path to libjvm.so
6392    os::jvm_path(buf, sizeof(buf));
6393
6394    // Get rid of libjvm.so
6395    p = strrchr(buf, '/');
6396    if (p == NULL) return false;
6397    else *p = '\0';
6398
6399    // Get rid of client or server
6400    p = strrchr(buf, '/');
6401    if (p == NULL) return false;
6402    else *p = '\0';
6403
6404    // check xawt/libmawt.so
6405    strcpy(libmawtpath, buf);
6406    strcat(libmawtpath, xawtstr);
6407    if (::stat(libmawtpath, &statbuf) == 0) return false;
6408
6409    // check libawt_xawt.so
6410    strcpy(libmawtpath, buf);
6411    strcat(libmawtpath, new_xawtstr);
6412    if (::stat(libmawtpath, &statbuf) == 0) return false;
6413
6414    return true;
6415}
6416
6417size_t os::write(int fd, const void *buf, unsigned int nBytes) {
6418  INTERRUPTIBLE_RETURN_INT(::write(fd, buf, nBytes), os::Solaris::clear_interrupted);
6419}
6420
6421int os::close(int fd) {
6422  RESTARTABLE_RETURN_INT(::close(fd));
6423}
6424
6425int os::socket_close(int fd) {
6426  RESTARTABLE_RETURN_INT(::close(fd));
6427}
6428
6429int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
6430  INTERRUPTIBLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6431}
6432
6433int os::send(int fd, char* buf, size_t nBytes, uint flags) {
6434  INTERRUPTIBLE_RETURN_INT((int)::send(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6435}
6436
6437int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
6438  RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
6439}
6440
6441// As both poll and select can be interrupted by signals, we have to be
6442// prepared to restart the system call after updating the timeout, unless
6443// a poll() is done with timeout == -1, in which case we repeat with this
6444// "wait forever" value.
6445
6446int os::timeout(int fd, long timeout) {
6447  int res;
6448  struct timeval t;
6449  julong prevtime, newtime;
6450  static const char* aNull = 0;
6451  struct pollfd pfd;
6452  pfd.fd = fd;
6453  pfd.events = POLLIN;
6454
6455  gettimeofday(&t, &aNull);
6456  prevtime = ((julong)t.tv_sec * 1000)  +  t.tv_usec / 1000;
6457
6458  for(;;) {
6459    INTERRUPTIBLE_NORESTART(::poll(&pfd, 1, timeout), res, os::Solaris::clear_interrupted);
6460    if(res == OS_ERR && errno == EINTR) {
6461        if(timeout != -1) {
6462          gettimeofday(&t, &aNull);
6463          newtime = ((julong)t.tv_sec * 1000)  +  t.tv_usec /1000;
6464          timeout -= newtime - prevtime;
6465          if(timeout <= 0)
6466            return OS_OK;
6467          prevtime = newtime;
6468        }
6469    } else return res;
6470  }
6471}
6472
6473int os::connect(int fd, struct sockaddr *him, socklen_t len) {
6474  int _result;
6475  INTERRUPTIBLE_NORESTART(::connect(fd, him, len), _result,\
6476                          os::Solaris::clear_interrupted);
6477
6478  // Depending on when thread interruption is reset, _result could be
6479  // one of two values when errno == EINTR
6480
6481  if (((_result == OS_INTRPT) || (_result == OS_ERR))
6482      && (errno == EINTR)) {
6483     /* restarting a connect() changes its errno semantics */
6484     INTERRUPTIBLE(::connect(fd, him, len), _result,\
6485                   os::Solaris::clear_interrupted);
6486     /* undo these changes */
6487     if (_result == OS_ERR) {
6488       if (errno == EALREADY) {
6489         errno = EINPROGRESS; /* fall through */
6490       } else if (errno == EISCONN) {
6491         errno = 0;
6492         return OS_OK;
6493       }
6494     }
6495   }
6496   return _result;
6497 }
6498
6499int os::accept(int fd, struct sockaddr* him, socklen_t* len) {
6500  if (fd < 0) {
6501    return OS_ERR;
6502  }
6503  INTERRUPTIBLE_RETURN_INT((int)::accept(fd, him, len),\
6504                           os::Solaris::clear_interrupted);
6505}
6506
6507int os::recvfrom(int fd, char* buf, size_t nBytes, uint flags,
6508                 sockaddr* from, socklen_t* fromlen) {
6509  INTERRUPTIBLE_RETURN_INT((int)::recvfrom(fd, buf, nBytes, flags, from, fromlen),\
6510                           os::Solaris::clear_interrupted);
6511}
6512
6513int os::sendto(int fd, char* buf, size_t len, uint flags,
6514               struct sockaddr* to, socklen_t tolen) {
6515  INTERRUPTIBLE_RETURN_INT((int)::sendto(fd, buf, len, flags, to, tolen),\
6516                           os::Solaris::clear_interrupted);
6517}
6518
6519int os::socket_available(int fd, jint *pbytes) {
6520  if (fd < 0) {
6521    return OS_OK;
6522  }
6523  int ret;
6524  RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
6525  // note: ioctl can return 0 when successful, JVM_SocketAvailable
6526  // is expected to return 0 on failure and 1 on success to the jdk.
6527  return (ret == OS_ERR) ? 0 : 1;
6528}
6529
6530int os::bind(int fd, struct sockaddr* him, socklen_t len) {
6531   INTERRUPTIBLE_RETURN_INT_NORESTART(::bind(fd, him, len),\
6532                                      os::Solaris::clear_interrupted);
6533}
6534
6535// Get the default path to the core file
6536// Returns the length of the string
6537int os::get_core_path(char* buffer, size_t bufferSize) {
6538  const char* p = get_current_directory(buffer, bufferSize);
6539
6540  if (p == NULL) {
6541    assert(p != NULL, "failed to get current directory");
6542    return 0;
6543  }
6544
6545  return strlen(buffer);
6546}
6547