os_solaris.cpp revision 13477:4d61110c6046
1/*
2 * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25// 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 "logging/log.hpp"
36#include "memory/allocation.inline.hpp"
37#include "memory/filemap.hpp"
38#include "oops/oop.inline.hpp"
39#include "os_share_solaris.hpp"
40#include "os_solaris.inline.hpp"
41#include "prims/jniFastGetField.hpp"
42#include "prims/jvm.h"
43#include "prims/jvm_misc.hpp"
44#include "runtime/arguments.hpp"
45#include "runtime/atomic.hpp"
46#include "runtime/extendedPC.hpp"
47#include "runtime/globals.hpp"
48#include "runtime/interfaceSupport.hpp"
49#include "runtime/java.hpp"
50#include "runtime/javaCalls.hpp"
51#include "runtime/mutexLocker.hpp"
52#include "runtime/objectMonitor.hpp"
53#include "runtime/orderAccess.inline.hpp"
54#include "runtime/osThread.hpp"
55#include "runtime/perfMemory.hpp"
56#include "runtime/sharedRuntime.hpp"
57#include "runtime/statSampler.hpp"
58#include "runtime/stubRoutines.hpp"
59#include "runtime/thread.inline.hpp"
60#include "runtime/threadCritical.hpp"
61#include "runtime/timer.hpp"
62#include "runtime/vm_version.hpp"
63#include "semaphore_posix.hpp"
64#include "services/attachListener.hpp"
65#include "services/memTracker.hpp"
66#include "services/runtimeService.hpp"
67#include "utilities/align.hpp"
68#include "utilities/decoder.hpp"
69#include "utilities/defaultStream.hpp"
70#include "utilities/events.hpp"
71#include "utilities/growableArray.hpp"
72#include "utilities/macros.hpp"
73#include "utilities/vmError.hpp"
74
75// put OS-includes here
76# include <dlfcn.h>
77# include <errno.h>
78# include <exception>
79# include <link.h>
80# include <poll.h>
81# include <pthread.h>
82# include <schedctl.h>
83# include <setjmp.h>
84# include <signal.h>
85# include <stdio.h>
86# include <alloca.h>
87# include <sys/filio.h>
88# include <sys/ipc.h>
89# include <sys/lwp.h>
90# include <sys/machelf.h>     // for elf Sym structure used by dladdr1
91# include <sys/mman.h>
92# include <sys/processor.h>
93# include <sys/procset.h>
94# include <sys/pset.h>
95# include <sys/resource.h>
96# include <sys/shm.h>
97# include <sys/socket.h>
98# include <sys/stat.h>
99# include <sys/systeminfo.h>
100# include <sys/time.h>
101# include <sys/times.h>
102# include <sys/types.h>
103# include <sys/wait.h>
104# include <sys/utsname.h>
105# include <thread.h>
106# include <unistd.h>
107# include <sys/priocntl.h>
108# include <sys/rtpriocntl.h>
109# include <sys/tspriocntl.h>
110# include <sys/iapriocntl.h>
111# include <sys/fxpriocntl.h>
112# include <sys/loadavg.h>
113# include <string.h>
114# include <stdio.h>
115
116# define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
117# include <sys/procfs.h>     //  see comment in <sys/procfs.h>
118
119#define MAX_PATH (2 * K)
120
121// for timer info max values which include all bits
122#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
123
124
125// Here are some liblgrp types from sys/lgrp_user.h to be able to
126// compile on older systems without this header file.
127
128#ifndef MADV_ACCESS_LWP
129  #define  MADV_ACCESS_LWP   7       /* next LWP to access heavily */
130#endif
131#ifndef MADV_ACCESS_MANY
132  #define  MADV_ACCESS_MANY  8       /* many processes to access heavily */
133#endif
134
135#ifndef LGRP_RSRC_CPU
136  #define LGRP_RSRC_CPU      0       /* CPU resources */
137#endif
138#ifndef LGRP_RSRC_MEM
139  #define LGRP_RSRC_MEM      1       /* memory resources */
140#endif
141
142// Values for ThreadPriorityPolicy == 1
143int prio_policy1[CriticalPriority+1] = {
144  -99999,  0, 16,  32,  48,  64,
145          80, 96, 112, 124, 127, 127 };
146
147// System parameters used internally
148static clock_t clock_tics_per_sec = 100;
149
150// Track if we have called enable_extended_FILE_stdio (on Solaris 10u4+)
151static bool enabled_extended_FILE_stdio = false;
152
153// For diagnostics to print a message once. see run_periodic_checks
154static bool check_addr0_done = false;
155static sigset_t check_signal_done;
156static bool check_signals = true;
157
158address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
159address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
160
161address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
162
163os::Solaris::pthread_setname_np_func_t os::Solaris::_pthread_setname_np = NULL;
164
165// "default" initializers for missing libc APIs
166extern "C" {
167  static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
168  static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
169
170  static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
171  static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
172}
173
174// "default" initializers for pthread-based synchronization
175extern "C" {
176  static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
177  static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
178}
179
180static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
181
182static inline size_t adjust_stack_size(address base, size_t size) {
183  if ((ssize_t)size < 0) {
184    // 4759953: Compensate for ridiculous stack size.
185    size = max_intx;
186  }
187  if (size > (size_t)base) {
188    // 4812466: Make sure size doesn't allow the stack to wrap the address space.
189    size = (size_t)base;
190  }
191  return size;
192}
193
194static inline stack_t get_stack_info() {
195  stack_t st;
196  int retval = thr_stksegment(&st);
197  st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
198  assert(retval == 0, "incorrect return value from thr_stksegment");
199  assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
200  assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
201  return st;
202}
203
204address os::current_stack_base() {
205  int r = thr_main();
206  guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
207  bool is_primordial_thread = r;
208
209  // Workaround 4352906, avoid calls to thr_stksegment by
210  // thr_main after the first one (it looks like we trash
211  // some data, causing the value for ss_sp to be incorrect).
212  if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
213    stack_t st = get_stack_info();
214    if (is_primordial_thread) {
215      // cache initial value of stack base
216      os::Solaris::_main_stack_base = (address)st.ss_sp;
217    }
218    return (address)st.ss_sp;
219  } else {
220    guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
221    return os::Solaris::_main_stack_base;
222  }
223}
224
225size_t os::current_stack_size() {
226  size_t size;
227
228  int r = thr_main();
229  guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
230  if (!r) {
231    size = get_stack_info().ss_size;
232  } else {
233    struct rlimit limits;
234    getrlimit(RLIMIT_STACK, &limits);
235    size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
236  }
237  // base may not be page aligned
238  address base = current_stack_base();
239  address bottom = align_up(base - size, os::vm_page_size());;
240  return (size_t)(base - bottom);
241}
242
243struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
244  return localtime_r(clock, res);
245}
246
247void os::Solaris::try_enable_extended_io() {
248  typedef int (*enable_extended_FILE_stdio_t)(int, int);
249
250  if (!UseExtendedFileIO) {
251    return;
252  }
253
254  enable_extended_FILE_stdio_t enabler =
255    (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
256                                         "enable_extended_FILE_stdio");
257  if (enabler) {
258    enabler(-1, -1);
259  }
260}
261
262static int _processors_online = 0;
263
264jint os::Solaris::_os_thread_limit = 0;
265volatile jint os::Solaris::_os_thread_count = 0;
266
267julong os::available_memory() {
268  return Solaris::available_memory();
269}
270
271julong os::Solaris::available_memory() {
272  return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
273}
274
275julong os::Solaris::_physical_memory = 0;
276
277julong os::physical_memory() {
278  return Solaris::physical_memory();
279}
280
281static hrtime_t first_hrtime = 0;
282static const hrtime_t hrtime_hz = 1000*1000*1000;
283static volatile hrtime_t max_hrtime = 0;
284
285
286void os::Solaris::initialize_system_info() {
287  set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
288  _processors_online = sysconf(_SC_NPROCESSORS_ONLN);
289  _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) *
290                                     (julong)sysconf(_SC_PAGESIZE);
291}
292
293int os::active_processor_count() {
294  int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
295  pid_t pid = getpid();
296  psetid_t pset = PS_NONE;
297  // Are we running in a processor set or is there any processor set around?
298  if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
299    uint_t pset_cpus;
300    // Query the number of cpus available to us.
301    if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
302      assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
303      _processors_online = pset_cpus;
304      return pset_cpus;
305    }
306  }
307  // Otherwise return number of online cpus
308  return online_cpus;
309}
310
311static bool find_processors_in_pset(psetid_t        pset,
312                                    processorid_t** id_array,
313                                    uint_t*         id_length) {
314  bool result = false;
315  // Find the number of processors in the processor set.
316  if (pset_info(pset, NULL, id_length, NULL) == 0) {
317    // Make up an array to hold their ids.
318    *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
319    // Fill in the array with their processor ids.
320    if (pset_info(pset, NULL, id_length, *id_array) == 0) {
321      result = true;
322    }
323  }
324  return result;
325}
326
327// Callers of find_processors_online() must tolerate imprecise results --
328// the system configuration can change asynchronously because of DR
329// or explicit psradm operations.
330//
331// We also need to take care that the loop (below) terminates as the
332// number of processors online can change between the _SC_NPROCESSORS_ONLN
333// request and the loop that builds the list of processor ids.   Unfortunately
334// there's no reliable way to determine the maximum valid processor id,
335// so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
336// man pages, which claim the processor id set is "sparse, but
337// not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
338// exit the loop.
339//
340// In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
341// not available on S8.0.
342
343static bool find_processors_online(processorid_t** id_array,
344                                   uint*           id_length) {
345  const processorid_t MAX_PROCESSOR_ID = 100000;
346  // Find the number of processors online.
347  *id_length = sysconf(_SC_NPROCESSORS_ONLN);
348  // Make up an array to hold their ids.
349  *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
350  // Processors need not be numbered consecutively.
351  long found = 0;
352  processorid_t next = 0;
353  while (found < *id_length && next < MAX_PROCESSOR_ID) {
354    processor_info_t info;
355    if (processor_info(next, &info) == 0) {
356      // NB, PI_NOINTR processors are effectively online ...
357      if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
358        (*id_array)[found] = next;
359        found += 1;
360      }
361    }
362    next += 1;
363  }
364  if (found < *id_length) {
365    // The loop above didn't identify the expected number of processors.
366    // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
367    // and re-running the loop, above, but there's no guarantee of progress
368    // if the system configuration is in flux.  Instead, we just return what
369    // we've got.  Note that in the worst case find_processors_online() could
370    // return an empty set.  (As a fall-back in the case of the empty set we
371    // could just return the ID of the current processor).
372    *id_length = found;
373  }
374
375  return true;
376}
377
378static bool assign_distribution(processorid_t* id_array,
379                                uint           id_length,
380                                uint*          distribution,
381                                uint           distribution_length) {
382  // We assume we can assign processorid_t's to uint's.
383  assert(sizeof(processorid_t) == sizeof(uint),
384         "can't convert processorid_t to uint");
385  // Quick check to see if we won't succeed.
386  if (id_length < distribution_length) {
387    return false;
388  }
389  // Assign processor ids to the distribution.
390  // Try to shuffle processors to distribute work across boards,
391  // assuming 4 processors per board.
392  const uint processors_per_board = ProcessDistributionStride;
393  // Find the maximum processor id.
394  processorid_t max_id = 0;
395  for (uint m = 0; m < id_length; m += 1) {
396    max_id = MAX2(max_id, id_array[m]);
397  }
398  // The next id, to limit loops.
399  const processorid_t limit_id = max_id + 1;
400  // Make up markers for available processors.
401  bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id, mtInternal);
402  for (uint c = 0; c < limit_id; c += 1) {
403    available_id[c] = false;
404  }
405  for (uint a = 0; a < id_length; a += 1) {
406    available_id[id_array[a]] = true;
407  }
408  // Step by "boards", then by "slot", copying to "assigned".
409  // NEEDS_CLEANUP: The assignment of processors should be stateful,
410  //                remembering which processors have been assigned by
411  //                previous calls, etc., so as to distribute several
412  //                independent calls of this method.  What we'd like is
413  //                It would be nice to have an API that let us ask
414  //                how many processes are bound to a processor,
415  //                but we don't have that, either.
416  //                In the short term, "board" is static so that
417  //                subsequent distributions don't all start at board 0.
418  static uint board = 0;
419  uint assigned = 0;
420  // Until we've found enough processors ....
421  while (assigned < distribution_length) {
422    // ... find the next available processor in the board.
423    for (uint slot = 0; slot < processors_per_board; slot += 1) {
424      uint try_id = board * processors_per_board + slot;
425      if ((try_id < limit_id) && (available_id[try_id] == true)) {
426        distribution[assigned] = try_id;
427        available_id[try_id] = false;
428        assigned += 1;
429        break;
430      }
431    }
432    board += 1;
433    if (board * processors_per_board + 0 >= limit_id) {
434      board = 0;
435    }
436  }
437  if (available_id != NULL) {
438    FREE_C_HEAP_ARRAY(bool, available_id);
439  }
440  return true;
441}
442
443void os::set_native_thread_name(const char *name) {
444  if (Solaris::_pthread_setname_np != NULL) {
445    // Only the first 31 bytes of 'name' are processed by pthread_setname_np
446    // but we explicitly copy into a size-limited buffer to avoid any
447    // possible overflow.
448    char buf[32];
449    snprintf(buf, sizeof(buf), "%s", name);
450    buf[sizeof(buf) - 1] = '\0';
451    Solaris::_pthread_setname_np(pthread_self(), buf);
452  }
453}
454
455bool os::distribute_processes(uint length, uint* distribution) {
456  bool result = false;
457  // Find the processor id's of all the available CPUs.
458  processorid_t* id_array  = NULL;
459  uint           id_length = 0;
460  // There are some races between querying information and using it,
461  // since processor sets can change dynamically.
462  psetid_t pset = PS_NONE;
463  // Are we running in a processor set?
464  if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
465    result = find_processors_in_pset(pset, &id_array, &id_length);
466  } else {
467    result = find_processors_online(&id_array, &id_length);
468  }
469  if (result == true) {
470    if (id_length >= length) {
471      result = assign_distribution(id_array, id_length, distribution, length);
472    } else {
473      result = false;
474    }
475  }
476  if (id_array != NULL) {
477    FREE_C_HEAP_ARRAY(processorid_t, id_array);
478  }
479  return result;
480}
481
482bool os::bind_to_processor(uint processor_id) {
483  // We assume that a processorid_t can be stored in a uint.
484  assert(sizeof(uint) == sizeof(processorid_t),
485         "can't convert uint to processorid_t");
486  int bind_result =
487    processor_bind(P_LWPID,                       // bind LWP.
488                   P_MYID,                        // bind current LWP.
489                   (processorid_t) processor_id,  // id.
490                   NULL);                         // don't return old binding.
491  return (bind_result == 0);
492}
493
494// Return true if user is running as root.
495
496bool os::have_special_privileges() {
497  static bool init = false;
498  static bool privileges = false;
499  if (!init) {
500    privileges = (getuid() != geteuid()) || (getgid() != getegid());
501    init = true;
502  }
503  return privileges;
504}
505
506
507void os::init_system_properties_values() {
508  // The next steps are taken in the product version:
509  //
510  // Obtain the JAVA_HOME value from the location of libjvm.so.
511  // This library should be located at:
512  // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
513  //
514  // If "/jre/lib/" appears at the right place in the path, then we
515  // assume libjvm.so is installed in a JDK and we use this path.
516  //
517  // Otherwise exit with message: "Could not create the Java virtual machine."
518  //
519  // The following extra steps are taken in the debugging version:
520  //
521  // If "/jre/lib/" does NOT appear at the right place in the path
522  // instead of exit check for $JAVA_HOME environment variable.
523  //
524  // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
525  // then we append a fake suffix "hotspot/libjvm.so" to this path so
526  // it looks like libjvm.so is installed there
527  // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
528  //
529  // Otherwise exit.
530  //
531  // Important note: if the location of libjvm.so changes this
532  // code needs to be changed accordingly.
533
534// Base path of extensions installed on the system.
535#define SYS_EXT_DIR     "/usr/jdk/packages"
536#define EXTENSIONS_DIR  "/lib/ext"
537
538  // Buffer that fits several sprintfs.
539  // Note that the space for the colon and the trailing null are provided
540  // by the nulls included by the sizeof operator.
541  const size_t bufsize =
542    MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
543         sizeof(SYS_EXT_DIR) + sizeof("/lib/"), // invariant ld_library_path
544         (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
545  char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
546
547  // sysclasspath, java_home, dll_dir
548  {
549    char *pslash;
550    os::jvm_path(buf, bufsize);
551
552    // Found the full path to libjvm.so.
553    // Now cut the path to <java_home>/jre if we can.
554    *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
555    pslash = strrchr(buf, '/');
556    if (pslash != NULL) {
557      *pslash = '\0';            // Get rid of /{client|server|hotspot}.
558    }
559    Arguments::set_dll_dir(buf);
560
561    if (pslash != NULL) {
562      pslash = strrchr(buf, '/');
563      if (pslash != NULL) {
564        *pslash = '\0';        // Get rid of /lib.
565      }
566    }
567    Arguments::set_java_home(buf);
568    set_boot_path('/', ':');
569  }
570
571  // Where to look for native libraries.
572  {
573    // Use dlinfo() to determine the correct java.library.path.
574    //
575    // If we're launched by the Java launcher, and the user
576    // does not set java.library.path explicitly on the commandline,
577    // the Java launcher sets LD_LIBRARY_PATH for us and unsets
578    // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
579    // dlinfo returns LD_LIBRARY_PATH + crle settings (including
580    // /usr/lib), which is exactly what we want.
581    //
582    // If the user does set java.library.path, it completely
583    // overwrites this setting, and always has.
584    //
585    // If we're not launched by the Java launcher, we may
586    // get here with any/all of the LD_LIBRARY_PATH[_32|64]
587    // settings.  Again, dlinfo does exactly what we want.
588
589    Dl_serinfo     info_sz, *info = &info_sz;
590    Dl_serpath     *path;
591    char           *library_path;
592    char           *common_path = buf;
593
594    // Determine search path count and required buffer size.
595    if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
596      FREE_C_HEAP_ARRAY(char, buf);
597      vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
598    }
599
600    // Allocate new buffer and initialize.
601    info = (Dl_serinfo*)NEW_C_HEAP_ARRAY(char, info_sz.dls_size, mtInternal);
602    info->dls_size = info_sz.dls_size;
603    info->dls_cnt = info_sz.dls_cnt;
604
605    // Obtain search path information.
606    if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
607      FREE_C_HEAP_ARRAY(char, buf);
608      FREE_C_HEAP_ARRAY(char, info);
609      vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
610    }
611
612    path = &info->dls_serpath[0];
613
614    // Note: Due to a legacy implementation, most of the library path
615    // is set in the launcher. This was to accomodate linking restrictions
616    // on legacy Solaris implementations (which are no longer supported).
617    // Eventually, all the library path setting will be done here.
618    //
619    // However, to prevent the proliferation of improperly built native
620    // libraries, the new path component /usr/jdk/packages is added here.
621
622    // Construct the invariant part of ld_library_path.
623    sprintf(common_path, SYS_EXT_DIR "/lib");
624
625    // Struct size is more than sufficient for the path components obtained
626    // through the dlinfo() call, so only add additional space for the path
627    // components explicitly added here.
628    size_t library_path_size = info->dls_size + strlen(common_path);
629    library_path = (char *)NEW_C_HEAP_ARRAY(char, library_path_size, mtInternal);
630    library_path[0] = '\0';
631
632    // Construct the desired Java library path from the linker's library
633    // search path.
634    //
635    // For compatibility, it is optimal that we insert the additional path
636    // components specific to the Java VM after those components specified
637    // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
638    // infrastructure.
639    if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it.
640      strcpy(library_path, common_path);
641    } else {
642      int inserted = 0;
643      int i;
644      for (i = 0; i < info->dls_cnt; i++, path++) {
645        uint_t flags = path->dls_flags & LA_SER_MASK;
646        if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
647          strcat(library_path, common_path);
648          strcat(library_path, os::path_separator());
649          inserted = 1;
650        }
651        strcat(library_path, path->dls_name);
652        strcat(library_path, os::path_separator());
653      }
654      // Eliminate trailing path separator.
655      library_path[strlen(library_path)-1] = '\0';
656    }
657
658    // happens before argument parsing - can't use a trace flag
659    // tty->print_raw("init_system_properties_values: native lib path: ");
660    // tty->print_raw_cr(library_path);
661
662    // Callee copies into its own buffer.
663    Arguments::set_library_path(library_path);
664
665    FREE_C_HEAP_ARRAY(char, library_path);
666    FREE_C_HEAP_ARRAY(char, info);
667  }
668
669  // Extensions directories.
670  sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
671  Arguments::set_ext_dirs(buf);
672
673  FREE_C_HEAP_ARRAY(char, buf);
674
675#undef SYS_EXT_DIR
676#undef EXTENSIONS_DIR
677}
678
679void os::breakpoint() {
680  BREAKPOINT;
681}
682
683bool os::obsolete_option(const JavaVMOption *option) {
684  if (!strncmp(option->optionString, "-Xt", 3)) {
685    return true;
686  } else if (!strncmp(option->optionString, "-Xtm", 4)) {
687    return true;
688  } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
689    return true;
690  } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
691    return true;
692  }
693  return false;
694}
695
696bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
697  address  stackStart  = (address)thread->stack_base();
698  address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
699  if (sp < stackStart && sp >= stackEnd) return true;
700  return false;
701}
702
703extern "C" void breakpoint() {
704  // use debugger to set breakpoint here
705}
706
707static thread_t main_thread;
708
709// Thread start routine for all newly created threads
710extern "C" void* thread_native_entry(void* thread_addr) {
711  // Try to randomize the cache line index of hot stack frames.
712  // This helps when threads of the same stack traces evict each other's
713  // cache lines. The threads can be either from the same JVM instance, or
714  // from different JVM instances. The benefit is especially true for
715  // processors with hyperthreading technology.
716  static int counter = 0;
717  int pid = os::current_process_id();
718  alloca(((pid ^ counter++) & 7) * 128);
719
720  int prio;
721  Thread* thread = (Thread*)thread_addr;
722
723  thread->initialize_thread_current();
724
725  OSThread* osthr = thread->osthread();
726
727  osthr->set_lwp_id(_lwp_self());  // Store lwp in case we are bound
728  thread->_schedctl = (void *) schedctl_init();
729
730  log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ").",
731    os::current_thread_id());
732
733  if (UseNUMA) {
734    int lgrp_id = os::numa_get_group_id();
735    if (lgrp_id != -1) {
736      thread->set_lgrp_id(lgrp_id);
737    }
738  }
739
740  // Our priority was set when we were created, and stored in the
741  // osthread, but couldn't be passed through to our LWP until now.
742  // So read back the priority and set it again.
743
744  if (osthr->thread_id() != -1) {
745    if (UseThreadPriorities) {
746      int prio = osthr->native_priority();
747      if (ThreadPriorityVerbose) {
748        tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
749                      INTPTR_FORMAT ", setting priority: %d\n",
750                      osthr->thread_id(), osthr->lwp_id(), prio);
751      }
752      os::set_native_priority(thread, prio);
753    }
754  } else if (ThreadPriorityVerbose) {
755    warning("Can't set priority in _start routine, thread id hasn't been set\n");
756  }
757
758  assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
759
760  // initialize signal mask for this thread
761  os::Solaris::hotspot_sigmask(thread);
762
763  thread->run();
764
765  // One less thread is executing
766  // When the VMThread gets here, the main thread may have already exited
767  // which frees the CodeHeap containing the Atomic::dec code
768  if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
769    Atomic::dec(&os::Solaris::_os_thread_count);
770  }
771
772  log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ").", os::current_thread_id());
773
774  // If a thread has not deleted itself ("delete this") as part of its
775  // termination sequence, we have to ensure thread-local-storage is
776  // cleared before we actually terminate. No threads should ever be
777  // deleted asynchronously with respect to their termination.
778  if (Thread::current_or_null_safe() != NULL) {
779    assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
780    thread->clear_thread_current();
781  }
782
783  if (UseDetachedThreads) {
784    thr_exit(NULL);
785    ShouldNotReachHere();
786  }
787  return NULL;
788}
789
790static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
791  // Allocate the OSThread object
792  OSThread* osthread = new OSThread(NULL, NULL);
793  if (osthread == NULL) return NULL;
794
795  // Store info on the Solaris thread into the OSThread
796  osthread->set_thread_id(thread_id);
797  osthread->set_lwp_id(_lwp_self());
798  thread->_schedctl = (void *) schedctl_init();
799
800  if (UseNUMA) {
801    int lgrp_id = os::numa_get_group_id();
802    if (lgrp_id != -1) {
803      thread->set_lgrp_id(lgrp_id);
804    }
805  }
806
807  if (ThreadPriorityVerbose) {
808    tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
809                  osthread->thread_id(), osthread->lwp_id());
810  }
811
812  // Initial thread state is INITIALIZED, not SUSPENDED
813  osthread->set_state(INITIALIZED);
814
815  return osthread;
816}
817
818void os::Solaris::hotspot_sigmask(Thread* thread) {
819  //Save caller's signal mask
820  sigset_t sigmask;
821  pthread_sigmask(SIG_SETMASK, NULL, &sigmask);
822  OSThread *osthread = thread->osthread();
823  osthread->set_caller_sigmask(sigmask);
824
825  pthread_sigmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
826  if (!ReduceSignalUsage) {
827    if (thread->is_VM_thread()) {
828      // Only the VM thread handles BREAK_SIGNAL ...
829      pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
830    } else {
831      // ... all other threads block BREAK_SIGNAL
832      assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
833      pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
834    }
835  }
836}
837
838bool os::create_attached_thread(JavaThread* thread) {
839#ifdef ASSERT
840  thread->verify_not_published();
841#endif
842  OSThread* osthread = create_os_thread(thread, thr_self());
843  if (osthread == NULL) {
844    return false;
845  }
846
847  // Initial thread state is RUNNABLE
848  osthread->set_state(RUNNABLE);
849  thread->set_osthread(osthread);
850
851  // initialize signal mask for this thread
852  // and save the caller's signal mask
853  os::Solaris::hotspot_sigmask(thread);
854
855  log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ").",
856    os::current_thread_id());
857
858  return true;
859}
860
861bool os::create_main_thread(JavaThread* thread) {
862#ifdef ASSERT
863  thread->verify_not_published();
864#endif
865  if (_starting_thread == NULL) {
866    _starting_thread = create_os_thread(thread, main_thread);
867    if (_starting_thread == NULL) {
868      return false;
869    }
870  }
871
872  // The primodial thread is runnable from the start
873  _starting_thread->set_state(RUNNABLE);
874
875  thread->set_osthread(_starting_thread);
876
877  // initialize signal mask for this thread
878  // and save the caller's signal mask
879  os::Solaris::hotspot_sigmask(thread);
880
881  return true;
882}
883
884// Helper function to trace thread attributes, similar to os::Posix::describe_pthread_attr()
885static char* describe_thr_create_attributes(char* buf, size_t buflen,
886                                            size_t stacksize, long flags) {
887  stringStream ss(buf, buflen);
888  ss.print("stacksize: " SIZE_FORMAT "k, ", stacksize / 1024);
889  ss.print("flags: ");
890  #define PRINT_FLAG(f) if (flags & f) ss.print( #f " ");
891  #define ALL(X) \
892    X(THR_SUSPENDED) \
893    X(THR_DETACHED) \
894    X(THR_BOUND) \
895    X(THR_NEW_LWP) \
896    X(THR_DAEMON)
897  ALL(PRINT_FLAG)
898  #undef ALL
899  #undef PRINT_FLAG
900  return buf;
901}
902
903// return default stack size for thr_type
904size_t os::Posix::default_stack_size(os::ThreadType thr_type) {
905  // default stack size when not specified by caller is 1M (2M for LP64)
906  size_t s = (BytesPerWord >> 2) * K * K;
907  return s;
908}
909
910bool os::create_thread(Thread* thread, ThreadType thr_type,
911                       size_t req_stack_size) {
912  // Allocate the OSThread object
913  OSThread* osthread = new OSThread(NULL, NULL);
914  if (osthread == NULL) {
915    return false;
916  }
917
918  if (ThreadPriorityVerbose) {
919    char *thrtyp;
920    switch (thr_type) {
921    case vm_thread:
922      thrtyp = (char *)"vm";
923      break;
924    case cgc_thread:
925      thrtyp = (char *)"cgc";
926      break;
927    case pgc_thread:
928      thrtyp = (char *)"pgc";
929      break;
930    case java_thread:
931      thrtyp = (char *)"java";
932      break;
933    case compiler_thread:
934      thrtyp = (char *)"compiler";
935      break;
936    case watcher_thread:
937      thrtyp = (char *)"watcher";
938      break;
939    default:
940      thrtyp = (char *)"unknown";
941      break;
942    }
943    tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
944  }
945
946  // calculate stack size if it's not specified by caller
947  size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
948
949  // Initial state is ALLOCATED but not INITIALIZED
950  osthread->set_state(ALLOCATED);
951
952  if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
953    // We got lots of threads. Check if we still have some address space left.
954    // Need to be at least 5Mb of unreserved address space. We do check by
955    // trying to reserve some.
956    const size_t VirtualMemoryBangSize = 20*K*K;
957    char* mem = os::reserve_memory(VirtualMemoryBangSize);
958    if (mem == NULL) {
959      delete osthread;
960      return false;
961    } else {
962      // Release the memory again
963      os::release_memory(mem, VirtualMemoryBangSize);
964    }
965  }
966
967  // Setup osthread because the child thread may need it.
968  thread->set_osthread(osthread);
969
970  // Create the Solaris thread
971  thread_t tid = 0;
972  long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED;
973  int      status;
974
975  // Mark that we don't have an lwp or thread id yet.
976  // In case we attempt to set the priority before the thread starts.
977  osthread->set_lwp_id(-1);
978  osthread->set_thread_id(-1);
979
980  status = thr_create(NULL, stack_size, thread_native_entry, thread, flags, &tid);
981
982  char buf[64];
983  if (status == 0) {
984    log_info(os, thread)("Thread started (tid: " UINTX_FORMAT ", attributes: %s). ",
985      (uintx) tid, describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
986  } else {
987    log_warning(os, thread)("Failed to start thread - thr_create failed (%s) for attributes: %s.",
988      os::errno_name(status), describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
989  }
990
991  if (status != 0) {
992    thread->set_osthread(NULL);
993    // Need to clean up stuff we've allocated so far
994    delete osthread;
995    return false;
996  }
997
998  Atomic::inc(&os::Solaris::_os_thread_count);
999
1000  // Store info on the Solaris thread into the OSThread
1001  osthread->set_thread_id(tid);
1002
1003  // Remember that we created this thread so we can set priority on it
1004  osthread->set_vm_created();
1005
1006  // Most thread types will set an explicit priority before starting the thread,
1007  // but for those that don't we need a valid value to read back in thread_native_entry.
1008  osthread->set_native_priority(NormPriority);
1009
1010  // Initial thread state is INITIALIZED, not SUSPENDED
1011  osthread->set_state(INITIALIZED);
1012
1013  // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1014  return true;
1015}
1016
1017debug_only(static bool signal_sets_initialized = false);
1018static sigset_t unblocked_sigs, vm_sigs;
1019
1020bool os::Solaris::is_sig_ignored(int sig) {
1021  struct sigaction oact;
1022  sigaction(sig, (struct sigaction*)NULL, &oact);
1023  void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1024                                 : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1025  if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
1026    return true;
1027  } else {
1028    return false;
1029  }
1030}
1031
1032void os::Solaris::signal_sets_init() {
1033  // Should also have an assertion stating we are still single-threaded.
1034  assert(!signal_sets_initialized, "Already initialized");
1035  // Fill in signals that are necessarily unblocked for all threads in
1036  // the VM. Currently, we unblock the following signals:
1037  // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1038  //                         by -Xrs (=ReduceSignalUsage));
1039  // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1040  // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1041  // the dispositions or masks wrt these signals.
1042  // Programs embedding the VM that want to use the above signals for their
1043  // own purposes must, at this time, use the "-Xrs" option to prevent
1044  // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1045  // (See bug 4345157, and other related bugs).
1046  // In reality, though, unblocking these signals is really a nop, since
1047  // these signals are not blocked by default.
1048  sigemptyset(&unblocked_sigs);
1049  sigaddset(&unblocked_sigs, SIGILL);
1050  sigaddset(&unblocked_sigs, SIGSEGV);
1051  sigaddset(&unblocked_sigs, SIGBUS);
1052  sigaddset(&unblocked_sigs, SIGFPE);
1053  sigaddset(&unblocked_sigs, ASYNC_SIGNAL);
1054
1055  if (!ReduceSignalUsage) {
1056    if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1057      sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1058    }
1059    if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1060      sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1061    }
1062    if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1063      sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1064    }
1065  }
1066  // Fill in signals that are blocked by all but the VM thread.
1067  sigemptyset(&vm_sigs);
1068  if (!ReduceSignalUsage) {
1069    sigaddset(&vm_sigs, BREAK_SIGNAL);
1070  }
1071  debug_only(signal_sets_initialized = true);
1072
1073  // For diagnostics only used in run_periodic_checks
1074  sigemptyset(&check_signal_done);
1075}
1076
1077// These are signals that are unblocked while a thread is running Java.
1078// (For some reason, they get blocked by default.)
1079sigset_t* os::Solaris::unblocked_signals() {
1080  assert(signal_sets_initialized, "Not initialized");
1081  return &unblocked_sigs;
1082}
1083
1084// These are the signals that are blocked while a (non-VM) thread is
1085// running Java. Only the VM thread handles these signals.
1086sigset_t* os::Solaris::vm_signals() {
1087  assert(signal_sets_initialized, "Not initialized");
1088  return &vm_sigs;
1089}
1090
1091void _handle_uncaught_cxx_exception() {
1092  VMError::report_and_die("An uncaught C++ exception");
1093}
1094
1095
1096// First crack at OS-specific initialization, from inside the new thread.
1097void os::initialize_thread(Thread* thr) {
1098  int r = thr_main();
1099  guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
1100  if (r) {
1101    JavaThread* jt = (JavaThread *)thr;
1102    assert(jt != NULL, "Sanity check");
1103    size_t stack_size;
1104    address base = jt->stack_base();
1105    if (Arguments::created_by_java_launcher()) {
1106      // Use 2MB to allow for Solaris 7 64 bit mode.
1107      stack_size = JavaThread::stack_size_at_create() == 0
1108        ? 2048*K : JavaThread::stack_size_at_create();
1109
1110      // There are rare cases when we may have already used more than
1111      // the basic stack size allotment before this method is invoked.
1112      // Attempt to allow for a normally sized java_stack.
1113      size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1114      stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1115    } else {
1116      // 6269555: If we were not created by a Java launcher, i.e. if we are
1117      // running embedded in a native application, treat the primordial thread
1118      // as much like a native attached thread as possible.  This means using
1119      // the current stack size from thr_stksegment(), unless it is too large
1120      // to reliably setup guard pages.  A reasonable max size is 8MB.
1121      size_t current_size = current_stack_size();
1122      // This should never happen, but just in case....
1123      if (current_size == 0) current_size = 2 * K * K;
1124      stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1125    }
1126    address bottom = align_up(base - stack_size, os::vm_page_size());;
1127    stack_size = (size_t)(base - bottom);
1128
1129    assert(stack_size > 0, "Stack size calculation problem");
1130
1131    if (stack_size > jt->stack_size()) {
1132#ifndef PRODUCT
1133      struct rlimit limits;
1134      getrlimit(RLIMIT_STACK, &limits);
1135      size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1136      assert(size >= jt->stack_size(), "Stack size problem in main thread");
1137#endif
1138      tty->print_cr("Stack size of %d Kb exceeds current limit of %d Kb.\n"
1139                    "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1140                    "See limit(1) to increase the stack size limit.",
1141                    stack_size / K, jt->stack_size() / K);
1142      vm_exit(1);
1143    }
1144    assert(jt->stack_size() >= stack_size,
1145           "Attempt to map more stack than was allocated");
1146    jt->set_stack_size(stack_size);
1147  }
1148
1149  // With the T2 libthread (T1 is no longer supported) threads are always bound
1150  // and we use stackbanging in all cases.
1151
1152  os::Solaris::init_thread_fpu_state();
1153  std::set_terminate(_handle_uncaught_cxx_exception);
1154}
1155
1156
1157
1158// Free Solaris resources related to the OSThread
1159void os::free_thread(OSThread* osthread) {
1160  assert(osthread != NULL, "os::free_thread but osthread not set");
1161
1162  // We are told to free resources of the argument thread,
1163  // but we can only really operate on the current thread.
1164  assert(Thread::current()->osthread() == osthread,
1165         "os::free_thread but not current thread");
1166
1167  // Restore caller's signal mask
1168  sigset_t sigmask = osthread->caller_sigmask();
1169  pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1170
1171  delete osthread;
1172}
1173
1174void os::pd_start_thread(Thread* thread) {
1175  int status = thr_continue(thread->osthread()->thread_id());
1176  assert_status(status == 0, status, "thr_continue failed");
1177}
1178
1179
1180intx os::current_thread_id() {
1181  return (intx)thr_self();
1182}
1183
1184static pid_t _initial_pid = 0;
1185
1186int os::current_process_id() {
1187  return (int)(_initial_pid ? _initial_pid : getpid());
1188}
1189
1190// gethrtime() should be monotonic according to the documentation,
1191// but some virtualized platforms are known to break this guarantee.
1192// getTimeNanos() must be guaranteed not to move backwards, so we
1193// are forced to add a check here.
1194inline hrtime_t getTimeNanos() {
1195  const hrtime_t now = gethrtime();
1196  const hrtime_t prev = max_hrtime;
1197  if (now <= prev) {
1198    return prev;   // same or retrograde time;
1199  }
1200  const hrtime_t obsv = Atomic::cmpxchg(now, &max_hrtime, prev);
1201  assert(obsv >= prev, "invariant");   // Monotonicity
1202  // If the CAS succeeded then we're done and return "now".
1203  // If the CAS failed and the observed value "obsv" is >= now then
1204  // we should return "obsv".  If the CAS failed and now > obsv > prv then
1205  // some other thread raced this thread and installed a new value, in which case
1206  // we could either (a) retry the entire operation, (b) retry trying to install now
1207  // or (c) just return obsv.  We use (c).   No loop is required although in some cases
1208  // we might discard a higher "now" value in deference to a slightly lower but freshly
1209  // installed obsv value.   That's entirely benign -- it admits no new orderings compared
1210  // to (a) or (b) -- and greatly reduces coherence traffic.
1211  // We might also condition (c) on the magnitude of the delta between obsv and now.
1212  // Avoiding excessive CAS operations to hot RW locations is critical.
1213  // See https://blogs.oracle.com/dave/entry/cas_and_cache_trivia_invalidate
1214  return (prev == obsv) ? now : obsv;
1215}
1216
1217// Time since start-up in seconds to a fine granularity.
1218// Used by VMSelfDestructTimer and the MemProfiler.
1219double os::elapsedTime() {
1220  return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1221}
1222
1223jlong os::elapsed_counter() {
1224  return (jlong)(getTimeNanos() - first_hrtime);
1225}
1226
1227jlong os::elapsed_frequency() {
1228  return hrtime_hz;
1229}
1230
1231// Return the real, user, and system times in seconds from an
1232// arbitrary fixed point in the past.
1233bool os::getTimesSecs(double* process_real_time,
1234                      double* process_user_time,
1235                      double* process_system_time) {
1236  struct tms ticks;
1237  clock_t real_ticks = times(&ticks);
1238
1239  if (real_ticks == (clock_t) (-1)) {
1240    return false;
1241  } else {
1242    double ticks_per_second = (double) clock_tics_per_sec;
1243    *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1244    *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1245    // For consistency return the real time from getTimeNanos()
1246    // converted to seconds.
1247    *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1248
1249    return true;
1250  }
1251}
1252
1253bool os::supports_vtime() { return true; }
1254bool os::enable_vtime() { return false; }
1255bool os::vtime_enabled() { return false; }
1256
1257double os::elapsedVTime() {
1258  return (double)gethrvtime() / (double)hrtime_hz;
1259}
1260
1261// Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1262jlong os::javaTimeMillis() {
1263  timeval t;
1264  if (gettimeofday(&t, NULL) == -1) {
1265    fatal("os::javaTimeMillis: gettimeofday (%s)", os::strerror(errno));
1266  }
1267  return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1268}
1269
1270// Must return seconds+nanos since Jan 1 1970. This must use the same
1271// time source as javaTimeMillis and can't use get_nsec_fromepoch as
1272// we need better than 1ms accuracy
1273void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1274  timeval t;
1275  if (gettimeofday(&t, NULL) == -1) {
1276    fatal("os::javaTimeSystemUTC: gettimeofday (%s)", os::strerror(errno));
1277  }
1278  seconds = jlong(t.tv_sec);
1279  nanos = jlong(t.tv_usec) * 1000;
1280}
1281
1282
1283jlong os::javaTimeNanos() {
1284  return (jlong)getTimeNanos();
1285}
1286
1287void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1288  info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1289  info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1290  info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1291  info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1292}
1293
1294char * os::local_time_string(char *buf, size_t buflen) {
1295  struct tm t;
1296  time_t long_time;
1297  time(&long_time);
1298  localtime_r(&long_time, &t);
1299  jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1300               t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1301               t.tm_hour, t.tm_min, t.tm_sec);
1302  return buf;
1303}
1304
1305// Note: os::shutdown() might be called very early during initialization, or
1306// called from signal handler. Before adding something to os::shutdown(), make
1307// sure it is async-safe and can handle partially initialized VM.
1308void os::shutdown() {
1309
1310  // allow PerfMemory to attempt cleanup of any persistent resources
1311  perfMemory_exit();
1312
1313  // needs to remove object in file system
1314  AttachListener::abort();
1315
1316  // flush buffered output, finish log files
1317  ostream_abort();
1318
1319  // Check for abort hook
1320  abort_hook_t abort_hook = Arguments::abort_hook();
1321  if (abort_hook != NULL) {
1322    abort_hook();
1323  }
1324}
1325
1326// Note: os::abort() might be called very early during initialization, or
1327// called from signal handler. Before adding something to os::abort(), make
1328// sure it is async-safe and can handle partially initialized VM.
1329void os::abort(bool dump_core, void* siginfo, const void* context) {
1330  os::shutdown();
1331  if (dump_core) {
1332#ifndef PRODUCT
1333    fdStream out(defaultStream::output_fd());
1334    out.print_raw("Current thread is ");
1335    char buf[16];
1336    jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1337    out.print_raw_cr(buf);
1338    out.print_raw_cr("Dumping core ...");
1339#endif
1340    ::abort(); // dump core (for debugging)
1341  }
1342
1343  ::exit(1);
1344}
1345
1346// Die immediately, no exit hook, no abort hook, no cleanup.
1347void os::die() {
1348  ::abort(); // dump core (for debugging)
1349}
1350
1351// DLL functions
1352
1353const char* os::dll_file_extension() { return ".so"; }
1354
1355// This must be hard coded because it's the system's temporary
1356// directory not the java application's temp directory, ala java.io.tmpdir.
1357const char* os::get_temp_directory() { return "/tmp"; }
1358
1359static bool file_exists(const char* filename) {
1360  struct stat statbuf;
1361  if (filename == NULL || strlen(filename) == 0) {
1362    return false;
1363  }
1364  return os::stat(filename, &statbuf) == 0;
1365}
1366
1367bool os::dll_build_name(char* buffer, size_t buflen,
1368                        const char* pname, const char* fname) {
1369  bool retval = false;
1370  const size_t pnamelen = pname ? strlen(pname) : 0;
1371
1372  // Return error on buffer overflow.
1373  if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1374    return retval;
1375  }
1376
1377  if (pnamelen == 0) {
1378    snprintf(buffer, buflen, "lib%s.so", fname);
1379    retval = true;
1380  } else if (strchr(pname, *os::path_separator()) != NULL) {
1381    int n;
1382    char** pelements = split_path(pname, &n);
1383    if (pelements == NULL) {
1384      return false;
1385    }
1386    for (int i = 0; i < n; i++) {
1387      // really shouldn't be NULL but what the heck, check can't hurt
1388      if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1389        continue; // skip the empty path values
1390      }
1391      snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1392      if (file_exists(buffer)) {
1393        retval = true;
1394        break;
1395      }
1396    }
1397    // release the storage
1398    for (int i = 0; i < n; i++) {
1399      if (pelements[i] != NULL) {
1400        FREE_C_HEAP_ARRAY(char, pelements[i]);
1401      }
1402    }
1403    if (pelements != NULL) {
1404      FREE_C_HEAP_ARRAY(char*, pelements);
1405    }
1406  } else {
1407    snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1408    retval = true;
1409  }
1410  return retval;
1411}
1412
1413// check if addr is inside libjvm.so
1414bool os::address_is_in_vm(address addr) {
1415  static address libjvm_base_addr;
1416  Dl_info dlinfo;
1417
1418  if (libjvm_base_addr == NULL) {
1419    if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1420      libjvm_base_addr = (address)dlinfo.dli_fbase;
1421    }
1422    assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1423  }
1424
1425  if (dladdr((void *)addr, &dlinfo) != 0) {
1426    if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1427  }
1428
1429  return false;
1430}
1431
1432typedef int (*dladdr1_func_type)(void *, Dl_info *, void **, int);
1433static dladdr1_func_type dladdr1_func = NULL;
1434
1435bool os::dll_address_to_function_name(address addr, char *buf,
1436                                      int buflen, int * offset,
1437                                      bool demangle) {
1438  // buf is not optional, but offset is optional
1439  assert(buf != NULL, "sanity check");
1440
1441  Dl_info dlinfo;
1442
1443  // dladdr1_func was initialized in os::init()
1444  if (dladdr1_func != NULL) {
1445    // yes, we have dladdr1
1446
1447    // Support for dladdr1 is checked at runtime; it may be
1448    // available even if the vm is built on a machine that does
1449    // not have dladdr1 support.  Make sure there is a value for
1450    // RTLD_DL_SYMENT.
1451#ifndef RTLD_DL_SYMENT
1452  #define RTLD_DL_SYMENT 1
1453#endif
1454#ifdef _LP64
1455    Elf64_Sym * info;
1456#else
1457    Elf32_Sym * info;
1458#endif
1459    if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1460                     RTLD_DL_SYMENT) != 0) {
1461      // see if we have a matching symbol that covers our address
1462      if (dlinfo.dli_saddr != NULL &&
1463          (char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1464        if (dlinfo.dli_sname != NULL) {
1465          if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1466            jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1467          }
1468          if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1469          return true;
1470        }
1471      }
1472      // no matching symbol so try for just file info
1473      if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1474        if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1475                            buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1476          return true;
1477        }
1478      }
1479    }
1480    buf[0] = '\0';
1481    if (offset != NULL) *offset  = -1;
1482    return false;
1483  }
1484
1485  // no, only dladdr is available
1486  if (dladdr((void *)addr, &dlinfo) != 0) {
1487    // see if we have a matching symbol
1488    if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1489      if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1490        jio_snprintf(buf, buflen, dlinfo.dli_sname);
1491      }
1492      if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1493      return true;
1494    }
1495    // no matching symbol so try for just file info
1496    if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1497      if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1498                          buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1499        return true;
1500      }
1501    }
1502  }
1503  buf[0] = '\0';
1504  if (offset != NULL) *offset  = -1;
1505  return false;
1506}
1507
1508bool os::dll_address_to_library_name(address addr, char* buf,
1509                                     int buflen, int* offset) {
1510  // buf is not optional, but offset is optional
1511  assert(buf != NULL, "sanity check");
1512
1513  Dl_info dlinfo;
1514
1515  if (dladdr((void*)addr, &dlinfo) != 0) {
1516    if (dlinfo.dli_fname != NULL) {
1517      jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1518    }
1519    if (dlinfo.dli_fbase != NULL && offset != NULL) {
1520      *offset = addr - (address)dlinfo.dli_fbase;
1521    }
1522    return true;
1523  }
1524
1525  buf[0] = '\0';
1526  if (offset) *offset = -1;
1527  return false;
1528}
1529
1530int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1531  Dl_info dli;
1532  // Sanity check?
1533  if (dladdr(CAST_FROM_FN_PTR(void *, os::get_loaded_modules_info), &dli) == 0 ||
1534      dli.dli_fname == NULL) {
1535    return 1;
1536  }
1537
1538  void * handle = dlopen(dli.dli_fname, RTLD_LAZY);
1539  if (handle == NULL) {
1540    return 1;
1541  }
1542
1543  Link_map *map;
1544  dlinfo(handle, RTLD_DI_LINKMAP, &map);
1545  if (map == NULL) {
1546    dlclose(handle);
1547    return 1;
1548  }
1549
1550  while (map->l_prev != NULL) {
1551    map = map->l_prev;
1552  }
1553
1554  while (map != NULL) {
1555    // Iterate through all map entries and call callback with fields of interest
1556    if(callback(map->l_name, (address)map->l_addr, (address)0, param)) {
1557      dlclose(handle);
1558      return 1;
1559    }
1560    map = map->l_next;
1561  }
1562
1563  dlclose(handle);
1564  return 0;
1565}
1566
1567int _print_dll_info_cb(const char * name, address base_address, address top_address, void * param) {
1568  outputStream * out = (outputStream *) param;
1569  out->print_cr(PTR_FORMAT " \t%s", base_address, name);
1570  return 0;
1571}
1572
1573void os::print_dll_info(outputStream * st) {
1574  st->print_cr("Dynamic libraries:"); st->flush();
1575  if (get_loaded_modules_info(_print_dll_info_cb, (void *)st)) {
1576    st->print_cr("Error: Cannot print dynamic libraries.");
1577  }
1578}
1579
1580// Loads .dll/.so and
1581// in case of error it checks if .dll/.so was built for the
1582// same architecture as Hotspot is running on
1583
1584void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1585  void * result= ::dlopen(filename, RTLD_LAZY);
1586  if (result != NULL) {
1587    // Successful loading
1588    return result;
1589  }
1590
1591  Elf32_Ehdr elf_head;
1592
1593  // Read system error message into ebuf
1594  // It may or may not be overwritten below
1595  ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1596  ebuf[ebuflen-1]='\0';
1597  int diag_msg_max_length=ebuflen-strlen(ebuf);
1598  char* diag_msg_buf=ebuf+strlen(ebuf);
1599
1600  if (diag_msg_max_length==0) {
1601    // No more space in ebuf for additional diagnostics message
1602    return NULL;
1603  }
1604
1605
1606  int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1607
1608  if (file_descriptor < 0) {
1609    // Can't open library, report dlerror() message
1610    return NULL;
1611  }
1612
1613  bool failed_to_read_elf_head=
1614    (sizeof(elf_head)!=
1615     (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1616
1617  ::close(file_descriptor);
1618  if (failed_to_read_elf_head) {
1619    // file i/o error - report dlerror() msg
1620    return NULL;
1621  }
1622
1623  typedef struct {
1624    Elf32_Half  code;         // Actual value as defined in elf.h
1625    Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1626    char        elf_class;    // 32 or 64 bit
1627    char        endianess;    // MSB or LSB
1628    char*       name;         // String representation
1629  } arch_t;
1630
1631  static const arch_t arch_array[]={
1632    {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1633    {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1634    {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1635    {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1636    {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1637    {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1638    {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1639    {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1640    {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1641    {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
1642  };
1643
1644#if  (defined IA32)
1645  static  Elf32_Half running_arch_code=EM_386;
1646#elif   (defined AMD64)
1647  static  Elf32_Half running_arch_code=EM_X86_64;
1648#elif  (defined IA64)
1649  static  Elf32_Half running_arch_code=EM_IA_64;
1650#elif  (defined __sparc) && (defined _LP64)
1651  static  Elf32_Half running_arch_code=EM_SPARCV9;
1652#elif  (defined __sparc) && (!defined _LP64)
1653  static  Elf32_Half running_arch_code=EM_SPARC;
1654#elif  (defined __powerpc64__)
1655  static  Elf32_Half running_arch_code=EM_PPC64;
1656#elif  (defined __powerpc__)
1657  static  Elf32_Half running_arch_code=EM_PPC;
1658#elif (defined ARM)
1659  static  Elf32_Half running_arch_code=EM_ARM;
1660#else
1661  #error Method os::dll_load requires that one of following is defined:\
1662       IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
1663#endif
1664
1665  // Identify compatability class for VM's architecture and library's architecture
1666  // Obtain string descriptions for architectures
1667
1668  arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1669  int running_arch_index=-1;
1670
1671  for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1672    if (running_arch_code == arch_array[i].code) {
1673      running_arch_index    = i;
1674    }
1675    if (lib_arch.code == arch_array[i].code) {
1676      lib_arch.compat_class = arch_array[i].compat_class;
1677      lib_arch.name         = arch_array[i].name;
1678    }
1679  }
1680
1681  assert(running_arch_index != -1,
1682         "Didn't find running architecture code (running_arch_code) in arch_array");
1683  if (running_arch_index == -1) {
1684    // Even though running architecture detection failed
1685    // we may still continue with reporting dlerror() message
1686    return NULL;
1687  }
1688
1689  if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1690    ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1691    return NULL;
1692  }
1693
1694  if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1695    ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1696    return NULL;
1697  }
1698
1699  if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1700    if (lib_arch.name!=NULL) {
1701      ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1702                 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1703                 lib_arch.name, arch_array[running_arch_index].name);
1704    } else {
1705      ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1706                 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1707                 lib_arch.code,
1708                 arch_array[running_arch_index].name);
1709    }
1710  }
1711
1712  return NULL;
1713}
1714
1715void* os::dll_lookup(void* handle, const char* name) {
1716  return dlsym(handle, name);
1717}
1718
1719void* os::get_default_process_handle() {
1720  return (void*)::dlopen(NULL, RTLD_LAZY);
1721}
1722
1723int os::stat(const char *path, struct stat *sbuf) {
1724  char pathbuf[MAX_PATH];
1725  if (strlen(path) > MAX_PATH - 1) {
1726    errno = ENAMETOOLONG;
1727    return -1;
1728  }
1729  os::native_path(strcpy(pathbuf, path));
1730  return ::stat(pathbuf, sbuf);
1731}
1732
1733static inline time_t get_mtime(const char* filename) {
1734  struct stat st;
1735  int ret = os::stat(filename, &st);
1736  assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
1737  return st.st_mtime;
1738}
1739
1740int os::compare_file_modified_times(const char* file1, const char* file2) {
1741  time_t t1 = get_mtime(file1);
1742  time_t t2 = get_mtime(file2);
1743  return t1 - t2;
1744}
1745
1746static bool _print_ascii_file(const char* filename, outputStream* st) {
1747  int fd = ::open(filename, O_RDONLY);
1748  if (fd == -1) {
1749    return false;
1750  }
1751
1752  char buf[32];
1753  int bytes;
1754  while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1755    st->print_raw(buf, bytes);
1756  }
1757
1758  ::close(fd);
1759
1760  return true;
1761}
1762
1763void os::print_os_info_brief(outputStream* st) {
1764  os::Solaris::print_distro_info(st);
1765
1766  os::Posix::print_uname_info(st);
1767
1768  os::Solaris::print_libversion_info(st);
1769}
1770
1771void os::print_os_info(outputStream* st) {
1772  st->print("OS:");
1773
1774  os::Solaris::print_distro_info(st);
1775
1776  os::Posix::print_uname_info(st);
1777
1778  os::Solaris::print_libversion_info(st);
1779
1780  os::Posix::print_rlimit_info(st);
1781
1782  os::Posix::print_load_average(st);
1783}
1784
1785void os::Solaris::print_distro_info(outputStream* st) {
1786  if (!_print_ascii_file("/etc/release", st)) {
1787    st->print("Solaris");
1788  }
1789  st->cr();
1790}
1791
1792void os::get_summary_os_info(char* buf, size_t buflen) {
1793  strncpy(buf, "Solaris", buflen);  // default to plain solaris
1794  FILE* fp = fopen("/etc/release", "r");
1795  if (fp != NULL) {
1796    char tmp[256];
1797    // Only get the first line and chop out everything but the os name.
1798    if (fgets(tmp, sizeof(tmp), fp)) {
1799      char* ptr = tmp;
1800      // skip past whitespace characters
1801      while (*ptr != '\0' && (*ptr == ' ' || *ptr == '\t' || *ptr == '\n')) ptr++;
1802      if (*ptr != '\0') {
1803        char* nl = strchr(ptr, '\n');
1804        if (nl != NULL) *nl = '\0';
1805        strncpy(buf, ptr, buflen);
1806      }
1807    }
1808    fclose(fp);
1809  }
1810}
1811
1812void os::Solaris::print_libversion_info(outputStream* st) {
1813  st->print("  (T2 libthread)");
1814  st->cr();
1815}
1816
1817static bool check_addr0(outputStream* st) {
1818  jboolean status = false;
1819  const int read_chunk = 200;
1820  int ret = 0;
1821  int nmap = 0;
1822  int fd = ::open("/proc/self/map",O_RDONLY);
1823  if (fd >= 0) {
1824    prmap_t *p = NULL;
1825    char *mbuff = (char *) calloc(read_chunk, sizeof(prmap_t));
1826    if (NULL == mbuff) {
1827      ::close(fd);
1828      return status;
1829    }
1830    while ((ret = ::read(fd, mbuff, read_chunk*sizeof(prmap_t))) > 0) {
1831      //check if read() has not read partial data
1832      if( 0 != ret % sizeof(prmap_t)){
1833        break;
1834      }
1835      nmap = ret / sizeof(prmap_t);
1836      p = (prmap_t *)mbuff;
1837      for(int i = 0; i < nmap; i++){
1838        if (p->pr_vaddr == 0x0) {
1839          st->print("Warning: Address: " PTR_FORMAT ", Size: " SIZE_FORMAT "K, ",p->pr_vaddr, p->pr_size/1024);
1840          st->print("Mapped file: %s, ", p->pr_mapname[0] == '\0' ? "None" : p->pr_mapname);
1841          st->print("Access: ");
1842          st->print("%s",(p->pr_mflags & MA_READ)  ? "r" : "-");
1843          st->print("%s",(p->pr_mflags & MA_WRITE) ? "w" : "-");
1844          st->print("%s",(p->pr_mflags & MA_EXEC)  ? "x" : "-");
1845          st->cr();
1846          status = true;
1847        }
1848        p++;
1849      }
1850    }
1851    free(mbuff);
1852    ::close(fd);
1853  }
1854  return status;
1855}
1856
1857void os::get_summary_cpu_info(char* buf, size_t buflen) {
1858  // Get MHz with system call. We don't seem to already have this.
1859  processor_info_t stats;
1860  processorid_t id = getcpuid();
1861  int clock = 0;
1862  if (processor_info(id, &stats) != -1) {
1863    clock = stats.pi_clock;  // pi_processor_type isn't more informative than below
1864  }
1865#ifdef AMD64
1866  snprintf(buf, buflen, "x86 64 bit %d MHz", clock);
1867#else
1868  // must be sparc
1869  snprintf(buf, buflen, "Sparcv9 64 bit %d MHz", clock);
1870#endif
1871}
1872
1873void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
1874  // Nothing to do for now.
1875}
1876
1877void os::print_memory_info(outputStream* st) {
1878  st->print("Memory:");
1879  st->print(" %dk page", os::vm_page_size()>>10);
1880  st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
1881  st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
1882  st->cr();
1883  (void) check_addr0(st);
1884}
1885
1886// Moved from whole group, because we need them here for diagnostic
1887// prints.
1888static int Maxsignum = 0;
1889static int *ourSigFlags = NULL;
1890
1891int os::Solaris::get_our_sigflags(int sig) {
1892  assert(ourSigFlags!=NULL, "signal data structure not initialized");
1893  assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1894  return ourSigFlags[sig];
1895}
1896
1897void os::Solaris::set_our_sigflags(int sig, int flags) {
1898  assert(ourSigFlags!=NULL, "signal data structure not initialized");
1899  assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1900  ourSigFlags[sig] = flags;
1901}
1902
1903
1904static const char* get_signal_handler_name(address handler,
1905                                           char* buf, int buflen) {
1906  int offset;
1907  bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
1908  if (found) {
1909    // skip directory names
1910    const char *p1, *p2;
1911    p1 = buf;
1912    size_t len = strlen(os::file_separator());
1913    while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
1914    jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
1915  } else {
1916    jio_snprintf(buf, buflen, PTR_FORMAT, handler);
1917  }
1918  return buf;
1919}
1920
1921static void print_signal_handler(outputStream* st, int sig,
1922                                 char* buf, size_t buflen) {
1923  struct sigaction sa;
1924
1925  sigaction(sig, NULL, &sa);
1926
1927  st->print("%s: ", os::exception_name(sig, buf, buflen));
1928
1929  address handler = (sa.sa_flags & SA_SIGINFO)
1930                  ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
1931                  : CAST_FROM_FN_PTR(address, sa.sa_handler);
1932
1933  if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
1934    st->print("SIG_DFL");
1935  } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
1936    st->print("SIG_IGN");
1937  } else {
1938    st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
1939  }
1940
1941  st->print(", sa_mask[0]=");
1942  os::Posix::print_signal_set_short(st, &sa.sa_mask);
1943
1944  address rh = VMError::get_resetted_sighandler(sig);
1945  // May be, handler was resetted by VMError?
1946  if (rh != NULL) {
1947    handler = rh;
1948    sa.sa_flags = VMError::get_resetted_sigflags(sig);
1949  }
1950
1951  st->print(", sa_flags=");
1952  os::Posix::print_sa_flags(st, sa.sa_flags);
1953
1954  // Check: is it our handler?
1955  if (handler == CAST_FROM_FN_PTR(address, signalHandler)) {
1956    // It is our signal handler
1957    // check for flags
1958    if (sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
1959      st->print(
1960                ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
1961                os::Solaris::get_our_sigflags(sig));
1962    }
1963  }
1964  st->cr();
1965}
1966
1967void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1968  st->print_cr("Signal Handlers:");
1969  print_signal_handler(st, SIGSEGV, buf, buflen);
1970  print_signal_handler(st, SIGBUS , buf, buflen);
1971  print_signal_handler(st, SIGFPE , buf, buflen);
1972  print_signal_handler(st, SIGPIPE, buf, buflen);
1973  print_signal_handler(st, SIGXFSZ, buf, buflen);
1974  print_signal_handler(st, SIGILL , buf, buflen);
1975  print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
1976  print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
1977  print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
1978  print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
1979  print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
1980}
1981
1982static char saved_jvm_path[MAXPATHLEN] = { 0 };
1983
1984// Find the full path to the current module, libjvm.so
1985void os::jvm_path(char *buf, jint buflen) {
1986  // Error checking.
1987  if (buflen < MAXPATHLEN) {
1988    assert(false, "must use a large-enough buffer");
1989    buf[0] = '\0';
1990    return;
1991  }
1992  // Lazy resolve the path to current module.
1993  if (saved_jvm_path[0] != 0) {
1994    strcpy(buf, saved_jvm_path);
1995    return;
1996  }
1997
1998  Dl_info dlinfo;
1999  int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2000  assert(ret != 0, "cannot locate libjvm");
2001  if (ret != 0 && dlinfo.dli_fname != NULL) {
2002    if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
2003      return;
2004    }
2005  } else {
2006    buf[0] = '\0';
2007    return;
2008  }
2009
2010  if (Arguments::sun_java_launcher_is_altjvm()) {
2011    // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2012    // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2013    // If "/jre/lib/" appears at the right place in the string, then
2014    // assume we are installed in a JDK and we're done.  Otherwise, check
2015    // for a JAVA_HOME environment variable and fix up the path so it
2016    // looks like libjvm.so is installed there (append a fake suffix
2017    // hotspot/libjvm.so).
2018    const char *p = buf + strlen(buf) - 1;
2019    for (int count = 0; p > buf && count < 5; ++count) {
2020      for (--p; p > buf && *p != '/'; --p)
2021        /* empty */ ;
2022    }
2023
2024    if (strncmp(p, "/jre/lib/", 9) != 0) {
2025      // Look for JAVA_HOME in the environment.
2026      char* java_home_var = ::getenv("JAVA_HOME");
2027      if (java_home_var != NULL && java_home_var[0] != 0) {
2028        char* jrelib_p;
2029        int   len;
2030
2031        // Check the current module name "libjvm.so".
2032        p = strrchr(buf, '/');
2033        assert(strstr(p, "/libjvm") == p, "invalid library name");
2034
2035        if (os::Posix::realpath(java_home_var, buf, buflen) == NULL) {
2036          return;
2037        }
2038        // determine if this is a legacy image or modules image
2039        // modules image doesn't have "jre" subdirectory
2040        len = strlen(buf);
2041        assert(len < buflen, "Ran out of buffer space");
2042        jrelib_p = buf + len;
2043        snprintf(jrelib_p, buflen-len, "/jre/lib");
2044        if (0 != access(buf, F_OK)) {
2045          snprintf(jrelib_p, buflen-len, "/lib");
2046        }
2047
2048        if (0 == access(buf, F_OK)) {
2049          // Use current module name "libjvm.so"
2050          len = strlen(buf);
2051          snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2052        } else {
2053          // Go back to path of .so
2054          if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
2055            return;
2056          }
2057        }
2058      }
2059    }
2060  }
2061
2062  strncpy(saved_jvm_path, buf, MAXPATHLEN);
2063  saved_jvm_path[MAXPATHLEN - 1] = '\0';
2064}
2065
2066
2067void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2068  // no prefix required, not even "_"
2069}
2070
2071
2072void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2073  // no suffix required
2074}
2075
2076// This method is a copy of JDK's sysGetLastErrorString
2077// from src/solaris/hpi/src/system_md.c
2078
2079size_t os::lasterror(char *buf, size_t len) {
2080  if (errno == 0)  return 0;
2081
2082  const char *s = os::strerror(errno);
2083  size_t n = ::strlen(s);
2084  if (n >= len) {
2085    n = len - 1;
2086  }
2087  ::strncpy(buf, s, n);
2088  buf[n] = '\0';
2089  return n;
2090}
2091
2092
2093// sun.misc.Signal
2094
2095extern "C" {
2096  static void UserHandler(int sig, void *siginfo, void *context) {
2097    // Ctrl-C is pressed during error reporting, likely because the error
2098    // handler fails to abort. Let VM die immediately.
2099    if (sig == SIGINT && VMError::is_error_reported()) {
2100      os::die();
2101    }
2102
2103    os::signal_notify(sig);
2104    // We do not need to reinstate the signal handler each time...
2105  }
2106}
2107
2108void* os::user_handler() {
2109  return CAST_FROM_FN_PTR(void*, UserHandler);
2110}
2111
2112struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2113  struct timespec ts;
2114  unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2115
2116  return ts;
2117}
2118
2119extern "C" {
2120  typedef void (*sa_handler_t)(int);
2121  typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2122}
2123
2124void* os::signal(int signal_number, void* handler) {
2125  struct sigaction sigAct, oldSigAct;
2126  sigfillset(&(sigAct.sa_mask));
2127  sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2128  sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2129
2130  if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2131    // -1 means registration failed
2132    return (void *)-1;
2133  }
2134
2135  return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2136}
2137
2138void os::signal_raise(int signal_number) {
2139  raise(signal_number);
2140}
2141
2142// The following code is moved from os.cpp for making this
2143// code platform specific, which it is by its very nature.
2144
2145// a counter for each possible signal value
2146static int Sigexit = 0;
2147static jint *pending_signals = NULL;
2148static int *preinstalled_sigs = NULL;
2149static struct sigaction *chainedsigactions = NULL;
2150static sema_t sig_sem;
2151typedef int (*version_getting_t)();
2152version_getting_t os::Solaris::get_libjsig_version = NULL;
2153
2154int os::sigexitnum_pd() {
2155  assert(Sigexit > 0, "signal memory not yet initialized");
2156  return Sigexit;
2157}
2158
2159void os::Solaris::init_signal_mem() {
2160  // Initialize signal structures
2161  Maxsignum = SIGRTMAX;
2162  Sigexit = Maxsignum+1;
2163  assert(Maxsignum >0, "Unable to obtain max signal number");
2164
2165  // pending_signals has one int per signal
2166  // The additional signal is for SIGEXIT - exit signal to signal_thread
2167  pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2168  memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2169
2170  if (UseSignalChaining) {
2171    chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2172                                                   * (Maxsignum + 1), mtInternal);
2173    memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2174    preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2175    memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2176  }
2177  ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2178  memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2179}
2180
2181void os::signal_init_pd() {
2182  int ret;
2183
2184  ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2185  assert(ret == 0, "sema_init() failed");
2186}
2187
2188void os::signal_notify(int signal_number) {
2189  int ret;
2190
2191  Atomic::inc(&pending_signals[signal_number]);
2192  ret = ::sema_post(&sig_sem);
2193  assert(ret == 0, "sema_post() failed");
2194}
2195
2196static int check_pending_signals(bool wait_for_signal) {
2197  int ret;
2198  while (true) {
2199    for (int i = 0; i < Sigexit + 1; i++) {
2200      jint n = pending_signals[i];
2201      if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2202        return i;
2203      }
2204    }
2205    if (!wait_for_signal) {
2206      return -1;
2207    }
2208    JavaThread *thread = JavaThread::current();
2209    ThreadBlockInVM tbivm(thread);
2210
2211    bool threadIsSuspended;
2212    do {
2213      thread->set_suspend_equivalent();
2214      // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2215      while ((ret = ::sema_wait(&sig_sem)) == EINTR)
2216        ;
2217      assert(ret == 0, "sema_wait() failed");
2218
2219      // were we externally suspended while we were waiting?
2220      threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2221      if (threadIsSuspended) {
2222        // The semaphore has been incremented, but while we were waiting
2223        // another thread suspended us. We don't want to continue running
2224        // while suspended because that would surprise the thread that
2225        // suspended us.
2226        ret = ::sema_post(&sig_sem);
2227        assert(ret == 0, "sema_post() failed");
2228
2229        thread->java_suspend_self();
2230      }
2231    } while (threadIsSuspended);
2232  }
2233}
2234
2235int os::signal_lookup() {
2236  return check_pending_signals(false);
2237}
2238
2239int os::signal_wait() {
2240  return check_pending_signals(true);
2241}
2242
2243////////////////////////////////////////////////////////////////////////////////
2244// Virtual Memory
2245
2246static int page_size = -1;
2247
2248// The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2249// clear this var if support is not available.
2250static bool has_map_align = true;
2251
2252int os::vm_page_size() {
2253  assert(page_size != -1, "must call os::init");
2254  return page_size;
2255}
2256
2257// Solaris allocates memory by pages.
2258int os::vm_allocation_granularity() {
2259  assert(page_size != -1, "must call os::init");
2260  return page_size;
2261}
2262
2263static bool recoverable_mmap_error(int err) {
2264  // See if the error is one we can let the caller handle. This
2265  // list of errno values comes from the Solaris mmap(2) man page.
2266  switch (err) {
2267  case EBADF:
2268  case EINVAL:
2269  case ENOTSUP:
2270    // let the caller deal with these errors
2271    return true;
2272
2273  default:
2274    // Any remaining errors on this OS can cause our reserved mapping
2275    // to be lost. That can cause confusion where different data
2276    // structures think they have the same memory mapped. The worst
2277    // scenario is if both the VM and a library think they have the
2278    // same memory mapped.
2279    return false;
2280  }
2281}
2282
2283static void warn_fail_commit_memory(char* addr, size_t bytes, bool exec,
2284                                    int err) {
2285  warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2286          ", %d) failed; error='%s' (errno=%d)", addr, bytes, exec,
2287          os::strerror(err), err);
2288}
2289
2290static void warn_fail_commit_memory(char* addr, size_t bytes,
2291                                    size_t alignment_hint, bool exec,
2292                                    int err) {
2293  warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2294          ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, bytes,
2295          alignment_hint, exec, os::strerror(err), err);
2296}
2297
2298int os::Solaris::commit_memory_impl(char* addr, size_t bytes, bool exec) {
2299  int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2300  size_t size = bytes;
2301  char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2302  if (res != NULL) {
2303    if (UseNUMAInterleaving) {
2304      numa_make_global(addr, bytes);
2305    }
2306    return 0;
2307  }
2308
2309  int err = errno;  // save errno from mmap() call in mmap_chunk()
2310
2311  if (!recoverable_mmap_error(err)) {
2312    warn_fail_commit_memory(addr, bytes, exec, err);
2313    vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "committing reserved memory.");
2314  }
2315
2316  return err;
2317}
2318
2319bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2320  return Solaris::commit_memory_impl(addr, bytes, exec) == 0;
2321}
2322
2323void os::pd_commit_memory_or_exit(char* addr, size_t bytes, bool exec,
2324                                  const char* mesg) {
2325  assert(mesg != NULL, "mesg must be specified");
2326  int err = os::Solaris::commit_memory_impl(addr, bytes, exec);
2327  if (err != 0) {
2328    // the caller wants all commit errors to exit with the specified mesg:
2329    warn_fail_commit_memory(addr, bytes, exec, err);
2330    vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2331  }
2332}
2333
2334size_t os::Solaris::page_size_for_alignment(size_t alignment) {
2335  assert(is_aligned(alignment, (size_t) vm_page_size()),
2336         SIZE_FORMAT " is not aligned to " SIZE_FORMAT,
2337         alignment, (size_t) vm_page_size());
2338
2339  for (int i = 0; _page_sizes[i] != 0; i++) {
2340    if (is_aligned(alignment, _page_sizes[i])) {
2341      return _page_sizes[i];
2342    }
2343  }
2344
2345  return (size_t) vm_page_size();
2346}
2347
2348int os::Solaris::commit_memory_impl(char* addr, size_t bytes,
2349                                    size_t alignment_hint, bool exec) {
2350  int err = Solaris::commit_memory_impl(addr, bytes, exec);
2351  if (err == 0 && UseLargePages && alignment_hint > 0) {
2352    assert(is_aligned(bytes, alignment_hint),
2353           SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, alignment_hint);
2354
2355    // The syscall memcntl requires an exact page size (see man memcntl for details).
2356    size_t page_size = page_size_for_alignment(alignment_hint);
2357    if (page_size > (size_t) vm_page_size()) {
2358      (void)Solaris::setup_large_pages(addr, bytes, page_size);
2359    }
2360  }
2361  return err;
2362}
2363
2364bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2365                          bool exec) {
2366  return Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec) == 0;
2367}
2368
2369void os::pd_commit_memory_or_exit(char* addr, size_t bytes,
2370                                  size_t alignment_hint, bool exec,
2371                                  const char* mesg) {
2372  assert(mesg != NULL, "mesg must be specified");
2373  int err = os::Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec);
2374  if (err != 0) {
2375    // the caller wants all commit errors to exit with the specified mesg:
2376    warn_fail_commit_memory(addr, bytes, alignment_hint, exec, err);
2377    vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2378  }
2379}
2380
2381// Uncommit the pages in a specified region.
2382void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2383  if (madvise(addr, bytes, MADV_FREE) < 0) {
2384    debug_only(warning("MADV_FREE failed."));
2385    return;
2386  }
2387}
2388
2389bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2390  return os::commit_memory(addr, size, !ExecMem);
2391}
2392
2393bool os::remove_stack_guard_pages(char* addr, size_t size) {
2394  return os::uncommit_memory(addr, size);
2395}
2396
2397// Change the page size in a given range.
2398void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2399  assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2400  assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2401  if (UseLargePages) {
2402    size_t page_size = Solaris::page_size_for_alignment(alignment_hint);
2403    if (page_size > (size_t) vm_page_size()) {
2404      Solaris::setup_large_pages(addr, bytes, page_size);
2405    }
2406  }
2407}
2408
2409// Tell the OS to make the range local to the first-touching LWP
2410void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2411  assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2412  if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2413    debug_only(warning("MADV_ACCESS_LWP failed."));
2414  }
2415}
2416
2417// Tell the OS that this range would be accessed from different LWPs.
2418void os::numa_make_global(char *addr, size_t bytes) {
2419  assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2420  if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2421    debug_only(warning("MADV_ACCESS_MANY failed."));
2422  }
2423}
2424
2425// Get the number of the locality groups.
2426size_t os::numa_get_groups_num() {
2427  size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2428  return n != -1 ? n : 1;
2429}
2430
2431// Get a list of leaf locality groups. A leaf lgroup is group that
2432// doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2433// board. An LWP is assigned to one of these groups upon creation.
2434size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2435  if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2436    ids[0] = 0;
2437    return 1;
2438  }
2439  int result_size = 0, top = 1, bottom = 0, cur = 0;
2440  for (int k = 0; k < size; k++) {
2441    int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2442                                   (Solaris::lgrp_id_t*)&ids[top], size - top);
2443    if (r == -1) {
2444      ids[0] = 0;
2445      return 1;
2446    }
2447    if (!r) {
2448      // That's a leaf node.
2449      assert(bottom <= cur, "Sanity check");
2450      // Check if the node has memory
2451      if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2452                                  NULL, 0, LGRP_RSRC_MEM) > 0) {
2453        ids[bottom++] = ids[cur];
2454      }
2455    }
2456    top += r;
2457    cur++;
2458  }
2459  if (bottom == 0) {
2460    // Handle a situation, when the OS reports no memory available.
2461    // Assume UMA architecture.
2462    ids[0] = 0;
2463    return 1;
2464  }
2465  return bottom;
2466}
2467
2468// Detect the topology change. Typically happens during CPU plugging-unplugging.
2469bool os::numa_topology_changed() {
2470  int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2471  if (is_stale != -1 && is_stale) {
2472    Solaris::lgrp_fini(Solaris::lgrp_cookie());
2473    Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2474    assert(c != 0, "Failure to initialize LGRP API");
2475    Solaris::set_lgrp_cookie(c);
2476    return true;
2477  }
2478  return false;
2479}
2480
2481// Get the group id of the current LWP.
2482int os::numa_get_group_id() {
2483  int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2484  if (lgrp_id == -1) {
2485    return 0;
2486  }
2487  const int size = os::numa_get_groups_num();
2488  int *ids = (int*)alloca(size * sizeof(int));
2489
2490  // Get the ids of all lgroups with memory; r is the count.
2491  int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2492                                  (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2493  if (r <= 0) {
2494    return 0;
2495  }
2496  return ids[os::random() % r];
2497}
2498
2499// Request information about the page.
2500bool os::get_page_info(char *start, page_info* info) {
2501  const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2502  uint64_t addr = (uintptr_t)start;
2503  uint64_t outdata[2];
2504  uint_t validity = 0;
2505
2506  if (meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2507    return false;
2508  }
2509
2510  info->size = 0;
2511  info->lgrp_id = -1;
2512
2513  if ((validity & 1) != 0) {
2514    if ((validity & 2) != 0) {
2515      info->lgrp_id = outdata[0];
2516    }
2517    if ((validity & 4) != 0) {
2518      info->size = outdata[1];
2519    }
2520    return true;
2521  }
2522  return false;
2523}
2524
2525// Scan the pages from start to end until a page different than
2526// the one described in the info parameter is encountered.
2527char *os::scan_pages(char *start, char* end, page_info* page_expected,
2528                     page_info* page_found) {
2529  const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2530  const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2531  uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT + 1];
2532  uint_t validity[MAX_MEMINFO_CNT];
2533
2534  size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2535  uint64_t p = (uint64_t)start;
2536  while (p < (uint64_t)end) {
2537    addrs[0] = p;
2538    size_t addrs_count = 1;
2539    while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] + page_size < (uint64_t)end) {
2540      addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2541      addrs_count++;
2542    }
2543
2544    if (meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2545      return NULL;
2546    }
2547
2548    size_t i = 0;
2549    for (; i < addrs_count; i++) {
2550      if ((validity[i] & 1) != 0) {
2551        if ((validity[i] & 4) != 0) {
2552          if (outdata[types * i + 1] != page_expected->size) {
2553            break;
2554          }
2555        } else if (page_expected->size != 0) {
2556          break;
2557        }
2558
2559        if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2560          if (outdata[types * i] != page_expected->lgrp_id) {
2561            break;
2562          }
2563        }
2564      } else {
2565        return NULL;
2566      }
2567    }
2568
2569    if (i < addrs_count) {
2570      if ((validity[i] & 2) != 0) {
2571        page_found->lgrp_id = outdata[types * i];
2572      } else {
2573        page_found->lgrp_id = -1;
2574      }
2575      if ((validity[i] & 4) != 0) {
2576        page_found->size = outdata[types * i + 1];
2577      } else {
2578        page_found->size = 0;
2579      }
2580      return (char*)addrs[i];
2581    }
2582
2583    p = addrs[addrs_count - 1] + page_size;
2584  }
2585  return end;
2586}
2587
2588bool os::pd_uncommit_memory(char* addr, size_t bytes) {
2589  size_t size = bytes;
2590  // Map uncommitted pages PROT_NONE so we fail early if we touch an
2591  // uncommitted page. Otherwise, the read/write might succeed if we
2592  // have enough swap space to back the physical page.
2593  return
2594    NULL != Solaris::mmap_chunk(addr, size,
2595                                MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2596                                PROT_NONE);
2597}
2598
2599char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2600  char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2601
2602  if (b == MAP_FAILED) {
2603    return NULL;
2604  }
2605  return b;
2606}
2607
2608char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes,
2609                             size_t alignment_hint, bool fixed) {
2610  char* addr = requested_addr;
2611  int flags = MAP_PRIVATE | MAP_NORESERVE;
2612
2613  assert(!(fixed && (alignment_hint > 0)),
2614         "alignment hint meaningless with fixed mmap");
2615
2616  if (fixed) {
2617    flags |= MAP_FIXED;
2618  } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
2619    flags |= MAP_ALIGN;
2620    addr = (char*) alignment_hint;
2621  }
2622
2623  // Map uncommitted pages PROT_NONE so we fail early if we touch an
2624  // uncommitted page. Otherwise, the read/write might succeed if we
2625  // have enough swap space to back the physical page.
2626  return mmap_chunk(addr, bytes, flags, PROT_NONE);
2627}
2628
2629char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2630                            size_t alignment_hint) {
2631  char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint,
2632                                  (requested_addr != NULL));
2633
2634  guarantee(requested_addr == NULL || requested_addr == addr,
2635            "OS failed to return requested mmap address.");
2636  return addr;
2637}
2638
2639// Reserve memory at an arbitrary address, only if that area is
2640// available (and not reserved for something else).
2641
2642char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2643  const int max_tries = 10;
2644  char* base[max_tries];
2645  size_t size[max_tries];
2646
2647  // Solaris adds a gap between mmap'ed regions.  The size of the gap
2648  // is dependent on the requested size and the MMU.  Our initial gap
2649  // value here is just a guess and will be corrected later.
2650  bool had_top_overlap = false;
2651  bool have_adjusted_gap = false;
2652  size_t gap = 0x400000;
2653
2654  // Assert only that the size is a multiple of the page size, since
2655  // that's all that mmap requires, and since that's all we really know
2656  // about at this low abstraction level.  If we need higher alignment,
2657  // we can either pass an alignment to this method or verify alignment
2658  // in one of the methods further up the call chain.  See bug 5044738.
2659  assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2660
2661  // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2662  // Give it a try, if the kernel honors the hint we can return immediately.
2663  char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2664
2665  volatile int err = errno;
2666  if (addr == requested_addr) {
2667    return addr;
2668  } else if (addr != NULL) {
2669    pd_unmap_memory(addr, bytes);
2670  }
2671
2672  if (log_is_enabled(Warning, os)) {
2673    char buf[256];
2674    buf[0] = '\0';
2675    if (addr == NULL) {
2676      jio_snprintf(buf, sizeof(buf), ": %s", os::strerror(err));
2677    }
2678    log_info(os)("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
2679            PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2680            "%s", bytes, requested_addr, addr, buf);
2681  }
2682
2683  // Address hint method didn't work.  Fall back to the old method.
2684  // In theory, once SNV becomes our oldest supported platform, this
2685  // code will no longer be needed.
2686  //
2687  // Repeatedly allocate blocks until the block is allocated at the
2688  // right spot. Give up after max_tries.
2689  int i;
2690  for (i = 0; i < max_tries; ++i) {
2691    base[i] = reserve_memory(bytes);
2692
2693    if (base[i] != NULL) {
2694      // Is this the block we wanted?
2695      if (base[i] == requested_addr) {
2696        size[i] = bytes;
2697        break;
2698      }
2699
2700      // check that the gap value is right
2701      if (had_top_overlap && !have_adjusted_gap) {
2702        size_t actual_gap = base[i-1] - base[i] - bytes;
2703        if (gap != actual_gap) {
2704          // adjust the gap value and retry the last 2 allocations
2705          assert(i > 0, "gap adjustment code problem");
2706          have_adjusted_gap = true;  // adjust the gap only once, just in case
2707          gap = actual_gap;
2708          log_info(os)("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
2709          unmap_memory(base[i], bytes);
2710          unmap_memory(base[i-1], size[i-1]);
2711          i-=2;
2712          continue;
2713        }
2714      }
2715
2716      // Does this overlap the block we wanted? Give back the overlapped
2717      // parts and try again.
2718      //
2719      // There is still a bug in this code: if top_overlap == bytes,
2720      // the overlap is offset from requested region by the value of gap.
2721      // In this case giving back the overlapped part will not work,
2722      // because we'll give back the entire block at base[i] and
2723      // therefore the subsequent allocation will not generate a new gap.
2724      // This could be fixed with a new algorithm that used larger
2725      // or variable size chunks to find the requested region -
2726      // but such a change would introduce additional complications.
2727      // It's rare enough that the planets align for this bug,
2728      // so we'll just wait for a fix for 6204603/5003415 which
2729      // will provide a mmap flag to allow us to avoid this business.
2730
2731      size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2732      if (top_overlap >= 0 && top_overlap < bytes) {
2733        had_top_overlap = true;
2734        unmap_memory(base[i], top_overlap);
2735        base[i] += top_overlap;
2736        size[i] = bytes - top_overlap;
2737      } else {
2738        size_t bottom_overlap = base[i] + bytes - requested_addr;
2739        if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2740          if (bottom_overlap == 0) {
2741            log_info(os)("attempt_reserve_memory_at: possible alignment bug");
2742          }
2743          unmap_memory(requested_addr, bottom_overlap);
2744          size[i] = bytes - bottom_overlap;
2745        } else {
2746          size[i] = bytes;
2747        }
2748      }
2749    }
2750  }
2751
2752  // Give back the unused reserved pieces.
2753
2754  for (int j = 0; j < i; ++j) {
2755    if (base[j] != NULL) {
2756      unmap_memory(base[j], size[j]);
2757    }
2758  }
2759
2760  return (i < max_tries) ? requested_addr : NULL;
2761}
2762
2763bool os::pd_release_memory(char* addr, size_t bytes) {
2764  size_t size = bytes;
2765  return munmap(addr, size) == 0;
2766}
2767
2768static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
2769  assert(addr == (char*)align_down((uintptr_t)addr, os::vm_page_size()),
2770         "addr must be page aligned");
2771  int retVal = mprotect(addr, bytes, prot);
2772  return retVal == 0;
2773}
2774
2775// Protect memory (Used to pass readonly pages through
2776// JNI GetArray<type>Elements with empty arrays.)
2777// Also, used for serialization page and for compressed oops null pointer
2778// checking.
2779bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2780                        bool is_committed) {
2781  unsigned int p = 0;
2782  switch (prot) {
2783  case MEM_PROT_NONE: p = PROT_NONE; break;
2784  case MEM_PROT_READ: p = PROT_READ; break;
2785  case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2786  case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2787  default:
2788    ShouldNotReachHere();
2789  }
2790  // is_committed is unused.
2791  return solaris_mprotect(addr, bytes, p);
2792}
2793
2794// guard_memory and unguard_memory only happens within stack guard pages.
2795// Since ISM pertains only to the heap, guard and unguard memory should not
2796/// happen with an ISM region.
2797bool os::guard_memory(char* addr, size_t bytes) {
2798  return solaris_mprotect(addr, bytes, PROT_NONE);
2799}
2800
2801bool os::unguard_memory(char* addr, size_t bytes) {
2802  return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
2803}
2804
2805// Large page support
2806static size_t _large_page_size = 0;
2807
2808// Insertion sort for small arrays (descending order).
2809static void insertion_sort_descending(size_t* array, int len) {
2810  for (int i = 0; i < len; i++) {
2811    size_t val = array[i];
2812    for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
2813      size_t tmp = array[key];
2814      array[key] = array[key - 1];
2815      array[key - 1] = tmp;
2816    }
2817  }
2818}
2819
2820bool os::Solaris::mpss_sanity_check(bool warn, size_t* page_size) {
2821  const unsigned int usable_count = VM_Version::page_size_count();
2822  if (usable_count == 1) {
2823    return false;
2824  }
2825
2826  // Find the right getpagesizes interface.  When solaris 11 is the minimum
2827  // build platform, getpagesizes() (without the '2') can be called directly.
2828  typedef int (*gps_t)(size_t[], int);
2829  gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
2830  if (gps_func == NULL) {
2831    gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
2832    if (gps_func == NULL) {
2833      if (warn) {
2834        warning("MPSS is not supported by the operating system.");
2835      }
2836      return false;
2837    }
2838  }
2839
2840  // Fill the array of page sizes.
2841  int n = (*gps_func)(_page_sizes, page_sizes_max);
2842  assert(n > 0, "Solaris bug?");
2843
2844  if (n == page_sizes_max) {
2845    // Add a sentinel value (necessary only if the array was completely filled
2846    // since it is static (zeroed at initialization)).
2847    _page_sizes[--n] = 0;
2848    DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
2849  }
2850  assert(_page_sizes[n] == 0, "missing sentinel");
2851  trace_page_sizes("available page sizes", _page_sizes, n);
2852
2853  if (n == 1) return false;     // Only one page size available.
2854
2855  // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
2856  // select up to usable_count elements.  First sort the array, find the first
2857  // acceptable value, then copy the usable sizes to the top of the array and
2858  // trim the rest.  Make sure to include the default page size :-).
2859  //
2860  // A better policy could get rid of the 4M limit by taking the sizes of the
2861  // important VM memory regions (java heap and possibly the code cache) into
2862  // account.
2863  insertion_sort_descending(_page_sizes, n);
2864  const size_t size_limit =
2865    FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
2866  int beg;
2867  for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */;
2868  const int end = MIN2((int)usable_count, n) - 1;
2869  for (int cur = 0; cur < end; ++cur, ++beg) {
2870    _page_sizes[cur] = _page_sizes[beg];
2871  }
2872  _page_sizes[end] = vm_page_size();
2873  _page_sizes[end + 1] = 0;
2874
2875  if (_page_sizes[end] > _page_sizes[end - 1]) {
2876    // Default page size is not the smallest; sort again.
2877    insertion_sort_descending(_page_sizes, end + 1);
2878  }
2879  *page_size = _page_sizes[0];
2880
2881  trace_page_sizes("usable page sizes", _page_sizes, end + 1);
2882  return true;
2883}
2884
2885void os::large_page_init() {
2886  if (UseLargePages) {
2887    // print a warning if any large page related flag is specified on command line
2888    bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
2889                           !FLAG_IS_DEFAULT(LargePageSizeInBytes);
2890
2891    UseLargePages = Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
2892  }
2893}
2894
2895bool os::Solaris::is_valid_page_size(size_t bytes) {
2896  for (int i = 0; _page_sizes[i] != 0; i++) {
2897    if (_page_sizes[i] == bytes) {
2898      return true;
2899    }
2900  }
2901  return false;
2902}
2903
2904bool os::Solaris::setup_large_pages(caddr_t start, size_t bytes, size_t align) {
2905  assert(is_valid_page_size(align), SIZE_FORMAT " is not a valid page size", align);
2906  assert(is_aligned((void*) start, align),
2907         PTR_FORMAT " is not aligned to " SIZE_FORMAT, p2i((void*) start), align);
2908  assert(is_aligned(bytes, align),
2909         SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, align);
2910
2911  // Signal to OS that we want large pages for addresses
2912  // from addr, addr + bytes
2913  struct memcntl_mha mpss_struct;
2914  mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
2915  mpss_struct.mha_pagesize = align;
2916  mpss_struct.mha_flags = 0;
2917  // Upon successful completion, memcntl() returns 0
2918  if (memcntl(start, bytes, MC_HAT_ADVISE, (caddr_t) &mpss_struct, 0, 0)) {
2919    debug_only(warning("Attempt to use MPSS failed."));
2920    return false;
2921  }
2922  return true;
2923}
2924
2925char* os::reserve_memory_special(size_t size, size_t alignment, char* addr, bool exec) {
2926  fatal("os::reserve_memory_special should not be called on Solaris.");
2927  return NULL;
2928}
2929
2930bool os::release_memory_special(char* base, size_t bytes) {
2931  fatal("os::release_memory_special should not be called on Solaris.");
2932  return false;
2933}
2934
2935size_t os::large_page_size() {
2936  return _large_page_size;
2937}
2938
2939// MPSS allows application to commit large page memory on demand; with ISM
2940// the entire memory region must be allocated as shared memory.
2941bool os::can_commit_large_page_memory() {
2942  return true;
2943}
2944
2945bool os::can_execute_large_page_memory() {
2946  return true;
2947}
2948
2949// Read calls from inside the vm need to perform state transitions
2950size_t os::read(int fd, void *buf, unsigned int nBytes) {
2951  size_t res;
2952  JavaThread* thread = (JavaThread*)Thread::current();
2953  assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2954  ThreadBlockInVM tbiv(thread);
2955  RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2956  return res;
2957}
2958
2959size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
2960  size_t res;
2961  JavaThread* thread = (JavaThread*)Thread::current();
2962  assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2963  ThreadBlockInVM tbiv(thread);
2964  RESTARTABLE(::pread(fd, buf, (size_t) nBytes, offset), res);
2965  return res;
2966}
2967
2968size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
2969  size_t res;
2970  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
2971         "Assumed _thread_in_native");
2972  RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2973  return res;
2974}
2975
2976void os::naked_short_sleep(jlong ms) {
2977  assert(ms < 1000, "Un-interruptable sleep, short time use only");
2978
2979  // usleep is deprecated and removed from POSIX, in favour of nanosleep, but
2980  // Solaris requires -lrt for this.
2981  usleep((ms * 1000));
2982
2983  return;
2984}
2985
2986// Sleep forever; naked call to OS-specific sleep; use with CAUTION
2987void os::infinite_sleep() {
2988  while (true) {    // sleep forever ...
2989    ::sleep(100);   // ... 100 seconds at a time
2990  }
2991}
2992
2993// Used to convert frequent JVM_Yield() to nops
2994bool os::dont_yield() {
2995  if (DontYieldALot) {
2996    static hrtime_t last_time = 0;
2997    hrtime_t diff = getTimeNanos() - last_time;
2998
2999    if (diff < DontYieldALotInterval * 1000000) {
3000      return true;
3001    }
3002
3003    last_time += diff;
3004
3005    return false;
3006  } else {
3007    return false;
3008  }
3009}
3010
3011// Note that yield semantics are defined by the scheduling class to which
3012// the thread currently belongs.  Typically, yield will _not yield to
3013// other equal or higher priority threads that reside on the dispatch queues
3014// of other CPUs.
3015
3016void os::naked_yield() {
3017  thr_yield();
3018}
3019
3020// Interface for setting lwp priorities.  We are using T2 libthread,
3021// which forces the use of bound threads, so all of our threads will
3022// be assigned to real lwp's.  Using the thr_setprio function is
3023// meaningless in this mode so we must adjust the real lwp's priority.
3024// The routines below implement the getting and setting of lwp priorities.
3025//
3026// Note: There are three priority scales used on Solaris.  Java priotities
3027//       which range from 1 to 10, libthread "thr_setprio" scale which range
3028//       from 0 to 127, and the current scheduling class of the process we
3029//       are running in.  This is typically from -60 to +60.
3030//       The setting of the lwp priorities in done after a call to thr_setprio
3031//       so Java priorities are mapped to libthread priorities and we map from
3032//       the latter to lwp priorities.  We don't keep priorities stored in
3033//       Java priorities since some of our worker threads want to set priorities
3034//       higher than all Java threads.
3035//
3036// For related information:
3037// (1)  man -s 2 priocntl
3038// (2)  man -s 4 priocntl
3039// (3)  man dispadmin
3040// =    librt.so
3041// =    libthread/common/rtsched.c - thrp_setlwpprio().
3042// =    ps -cL <pid> ... to validate priority.
3043// =    sched_get_priority_min and _max
3044//              pthread_create
3045//              sched_setparam
3046//              pthread_setschedparam
3047//
3048// Assumptions:
3049// +    We assume that all threads in the process belong to the same
3050//              scheduling class.   IE. an homogenous process.
3051// +    Must be root or in IA group to change change "interactive" attribute.
3052//              Priocntl() will fail silently.  The only indication of failure is when
3053//              we read-back the value and notice that it hasn't changed.
3054// +    Interactive threads enter the runq at the head, non-interactive at the tail.
3055// +    For RT, change timeslice as well.  Invariant:
3056//              constant "priority integral"
3057//              Konst == TimeSlice * (60-Priority)
3058//              Given a priority, compute appropriate timeslice.
3059// +    Higher numerical values have higher priority.
3060
3061// sched class attributes
3062typedef struct {
3063  int   schedPolicy;              // classID
3064  int   maxPrio;
3065  int   minPrio;
3066} SchedInfo;
3067
3068
3069static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3070
3071#ifdef ASSERT
3072static int  ReadBackValidate = 1;
3073#endif
3074static int  myClass     = 0;
3075static int  myMin       = 0;
3076static int  myMax       = 0;
3077static int  myCur       = 0;
3078static bool priocntl_enable = false;
3079
3080static const int criticalPrio = FXCriticalPriority;
3081static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3082
3083
3084// lwp_priocntl_init
3085//
3086// Try to determine the priority scale for our process.
3087//
3088// Return errno or 0 if OK.
3089//
3090static int lwp_priocntl_init() {
3091  int rslt;
3092  pcinfo_t ClassInfo;
3093  pcparms_t ParmInfo;
3094  int i;
3095
3096  if (!UseThreadPriorities) return 0;
3097
3098  // If ThreadPriorityPolicy is 1, switch tables
3099  if (ThreadPriorityPolicy == 1) {
3100    for (i = 0; i < CriticalPriority+1; i++)
3101      os::java_to_os_priority[i] = prio_policy1[i];
3102  }
3103  if (UseCriticalJavaThreadPriority) {
3104    // MaxPriority always maps to the FX scheduling class and criticalPrio.
3105    // See set_native_priority() and set_lwp_class_and_priority().
3106    // Save original MaxPriority mapping in case attempt to
3107    // use critical priority fails.
3108    java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3109    // Set negative to distinguish from other priorities
3110    os::java_to_os_priority[MaxPriority] = -criticalPrio;
3111  }
3112
3113  // Get IDs for a set of well-known scheduling classes.
3114  // TODO-FIXME: GETCLINFO returns the current # of classes in the
3115  // the system.  We should have a loop that iterates over the
3116  // classID values, which are known to be "small" integers.
3117
3118  strcpy(ClassInfo.pc_clname, "TS");
3119  ClassInfo.pc_cid = -1;
3120  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3121  if (rslt < 0) return errno;
3122  assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3123  tsLimits.schedPolicy = ClassInfo.pc_cid;
3124  tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3125  tsLimits.minPrio = -tsLimits.maxPrio;
3126
3127  strcpy(ClassInfo.pc_clname, "IA");
3128  ClassInfo.pc_cid = -1;
3129  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3130  if (rslt < 0) return errno;
3131  assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3132  iaLimits.schedPolicy = ClassInfo.pc_cid;
3133  iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3134  iaLimits.minPrio = -iaLimits.maxPrio;
3135
3136  strcpy(ClassInfo.pc_clname, "RT");
3137  ClassInfo.pc_cid = -1;
3138  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3139  if (rslt < 0) return errno;
3140  assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3141  rtLimits.schedPolicy = ClassInfo.pc_cid;
3142  rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3143  rtLimits.minPrio = 0;
3144
3145  strcpy(ClassInfo.pc_clname, "FX");
3146  ClassInfo.pc_cid = -1;
3147  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3148  if (rslt < 0) return errno;
3149  assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3150  fxLimits.schedPolicy = ClassInfo.pc_cid;
3151  fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3152  fxLimits.minPrio = 0;
3153
3154  // Query our "current" scheduling class.
3155  // This will normally be IA, TS or, rarely, FX or RT.
3156  memset(&ParmInfo, 0, sizeof(ParmInfo));
3157  ParmInfo.pc_cid = PC_CLNULL;
3158  rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3159  if (rslt < 0) return errno;
3160  myClass = ParmInfo.pc_cid;
3161
3162  // We now know our scheduling classId, get specific information
3163  // about the class.
3164  ClassInfo.pc_cid = myClass;
3165  ClassInfo.pc_clname[0] = 0;
3166  rslt = priocntl((idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3167  if (rslt < 0) return errno;
3168
3169  if (ThreadPriorityVerbose) {
3170    tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3171  }
3172
3173  memset(&ParmInfo, 0, sizeof(pcparms_t));
3174  ParmInfo.pc_cid = PC_CLNULL;
3175  rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3176  if (rslt < 0) return errno;
3177
3178  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3179    myMin = rtLimits.minPrio;
3180    myMax = rtLimits.maxPrio;
3181  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3182    iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3183    myMin = iaLimits.minPrio;
3184    myMax = iaLimits.maxPrio;
3185    myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3186  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3187    tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3188    myMin = tsLimits.minPrio;
3189    myMax = tsLimits.maxPrio;
3190    myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3191  } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3192    fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3193    myMin = fxLimits.minPrio;
3194    myMax = fxLimits.maxPrio;
3195    myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3196  } else {
3197    // No clue - punt
3198    if (ThreadPriorityVerbose) {
3199      tty->print_cr("Unknown scheduling class: %s ... \n",
3200                    ClassInfo.pc_clname);
3201    }
3202    return EINVAL;      // no clue, punt
3203  }
3204
3205  if (ThreadPriorityVerbose) {
3206    tty->print_cr("Thread priority Range: [%d..%d]\n", myMin, myMax);
3207  }
3208
3209  priocntl_enable = true;  // Enable changing priorities
3210  return 0;
3211}
3212
3213#define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3214#define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3215#define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3216#define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3217
3218
3219// scale_to_lwp_priority
3220//
3221// Convert from the libthread "thr_setprio" scale to our current
3222// lwp scheduling class scale.
3223//
3224static int scale_to_lwp_priority(int rMin, int rMax, int x) {
3225  int v;
3226
3227  if (x == 127) return rMax;            // avoid round-down
3228  v = (((x*(rMax-rMin)))/128)+rMin;
3229  return v;
3230}
3231
3232
3233// set_lwp_class_and_priority
3234int set_lwp_class_and_priority(int ThreadID, int lwpid,
3235                               int newPrio, int new_class, bool scale) {
3236  int rslt;
3237  int Actual, Expected, prv;
3238  pcparms_t ParmInfo;                   // for GET-SET
3239#ifdef ASSERT
3240  pcparms_t ReadBack;                   // for readback
3241#endif
3242
3243  // Set priority via PC_GETPARMS, update, PC_SETPARMS
3244  // Query current values.
3245  // TODO: accelerate this by eliminating the PC_GETPARMS call.
3246  // Cache "pcparms_t" in global ParmCache.
3247  // TODO: elide set-to-same-value
3248
3249  // If something went wrong on init, don't change priorities.
3250  if (!priocntl_enable) {
3251    if (ThreadPriorityVerbose) {
3252      tty->print_cr("Trying to set priority but init failed, ignoring");
3253    }
3254    return EINVAL;
3255  }
3256
3257  // If lwp hasn't started yet, just return
3258  // the _start routine will call us again.
3259  if (lwpid <= 0) {
3260    if (ThreadPriorityVerbose) {
3261      tty->print_cr("deferring the set_lwp_class_and_priority of thread "
3262                    INTPTR_FORMAT " to %d, lwpid not set",
3263                    ThreadID, newPrio);
3264    }
3265    return 0;
3266  }
3267
3268  if (ThreadPriorityVerbose) {
3269    tty->print_cr ("set_lwp_class_and_priority("
3270                   INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3271                   ThreadID, lwpid, newPrio);
3272  }
3273
3274  memset(&ParmInfo, 0, sizeof(pcparms_t));
3275  ParmInfo.pc_cid = PC_CLNULL;
3276  rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3277  if (rslt < 0) return errno;
3278
3279  int cur_class = ParmInfo.pc_cid;
3280  ParmInfo.pc_cid = (id_t)new_class;
3281
3282  if (new_class == rtLimits.schedPolicy) {
3283    rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3284    rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3285                                                       rtLimits.maxPrio, newPrio)
3286                               : newPrio;
3287    rtInfo->rt_tqsecs  = RT_NOCHANGE;
3288    rtInfo->rt_tqnsecs = RT_NOCHANGE;
3289    if (ThreadPriorityVerbose) {
3290      tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3291    }
3292  } else if (new_class == iaLimits.schedPolicy) {
3293    iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3294    int maxClamped     = MIN2(iaLimits.maxPrio,
3295                              cur_class == new_class
3296                              ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3297    iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3298                                                       maxClamped, newPrio)
3299                               : newPrio;
3300    iaInfo->ia_uprilim = cur_class == new_class
3301                           ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3302    iaInfo->ia_mode    = IA_NOCHANGE;
3303    if (ThreadPriorityVerbose) {
3304      tty->print_cr("IA: [%d...%d] %d->%d\n",
3305                    iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3306    }
3307  } else if (new_class == tsLimits.schedPolicy) {
3308    tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3309    int maxClamped     = MIN2(tsLimits.maxPrio,
3310                              cur_class == new_class
3311                              ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3312    tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3313                                                       maxClamped, newPrio)
3314                               : newPrio;
3315    tsInfo->ts_uprilim = cur_class == new_class
3316                           ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
3317    if (ThreadPriorityVerbose) {
3318      tty->print_cr("TS: [%d...%d] %d->%d\n",
3319                    tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3320    }
3321  } else if (new_class == fxLimits.schedPolicy) {
3322    fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
3323    int maxClamped     = MIN2(fxLimits.maxPrio,
3324                              cur_class == new_class
3325                              ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
3326    fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
3327                                                       maxClamped, newPrio)
3328                               : newPrio;
3329    fxInfo->fx_uprilim = cur_class == new_class
3330                           ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
3331    fxInfo->fx_tqsecs  = FX_NOCHANGE;
3332    fxInfo->fx_tqnsecs = FX_NOCHANGE;
3333    if (ThreadPriorityVerbose) {
3334      tty->print_cr("FX: [%d...%d] %d->%d\n",
3335                    fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
3336    }
3337  } else {
3338    if (ThreadPriorityVerbose) {
3339      tty->print_cr("Unknown new scheduling class %d\n", new_class);
3340    }
3341    return EINVAL;    // no clue, punt
3342  }
3343
3344  rslt = priocntl(P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3345  if (ThreadPriorityVerbose && rslt) {
3346    tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3347  }
3348  if (rslt < 0) return errno;
3349
3350#ifdef ASSERT
3351  // Sanity check: read back what we just attempted to set.
3352  // In theory it could have changed in the interim ...
3353  //
3354  // The priocntl system call is tricky.
3355  // Sometimes it'll validate the priority value argument and
3356  // return EINVAL if unhappy.  At other times it fails silently.
3357  // Readbacks are prudent.
3358
3359  if (!ReadBackValidate) return 0;
3360
3361  memset(&ReadBack, 0, sizeof(pcparms_t));
3362  ReadBack.pc_cid = PC_CLNULL;
3363  rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3364  assert(rslt >= 0, "priocntl failed");
3365  Actual = Expected = 0xBAD;
3366  assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3367  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3368    Actual   = RTPRI(ReadBack)->rt_pri;
3369    Expected = RTPRI(ParmInfo)->rt_pri;
3370  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3371    Actual   = IAPRI(ReadBack)->ia_upri;
3372    Expected = IAPRI(ParmInfo)->ia_upri;
3373  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3374    Actual   = TSPRI(ReadBack)->ts_upri;
3375    Expected = TSPRI(ParmInfo)->ts_upri;
3376  } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3377    Actual   = FXPRI(ReadBack)->fx_upri;
3378    Expected = FXPRI(ParmInfo)->fx_upri;
3379  } else {
3380    if (ThreadPriorityVerbose) {
3381      tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
3382                    ParmInfo.pc_cid);
3383    }
3384  }
3385
3386  if (Actual != Expected) {
3387    if (ThreadPriorityVerbose) {
3388      tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3389                     lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3390    }
3391  }
3392#endif
3393
3394  return 0;
3395}
3396
3397// Solaris only gives access to 128 real priorities at a time,
3398// so we expand Java's ten to fill this range.  This would be better
3399// if we dynamically adjusted relative priorities.
3400//
3401// The ThreadPriorityPolicy option allows us to select 2 different
3402// priority scales.
3403//
3404// ThreadPriorityPolicy=0
3405// Since the Solaris' default priority is MaximumPriority, we do not
3406// set a priority lower than Max unless a priority lower than
3407// NormPriority is requested.
3408//
3409// ThreadPriorityPolicy=1
3410// This mode causes the priority table to get filled with
3411// linear values.  NormPriority get's mapped to 50% of the
3412// Maximum priority an so on.  This will cause VM threads
3413// to get unfair treatment against other Solaris processes
3414// which do not explicitly alter their thread priorities.
3415
3416int os::java_to_os_priority[CriticalPriority + 1] = {
3417  -99999,         // 0 Entry should never be used
3418
3419  0,              // 1 MinPriority
3420  32,             // 2
3421  64,             // 3
3422
3423  96,             // 4
3424  127,            // 5 NormPriority
3425  127,            // 6
3426
3427  127,            // 7
3428  127,            // 8
3429  127,            // 9 NearMaxPriority
3430
3431  127,            // 10 MaxPriority
3432
3433  -criticalPrio   // 11 CriticalPriority
3434};
3435
3436OSReturn os::set_native_priority(Thread* thread, int newpri) {
3437  OSThread* osthread = thread->osthread();
3438
3439  // Save requested priority in case the thread hasn't been started
3440  osthread->set_native_priority(newpri);
3441
3442  // Check for critical priority request
3443  bool fxcritical = false;
3444  if (newpri == -criticalPrio) {
3445    fxcritical = true;
3446    newpri = criticalPrio;
3447  }
3448
3449  assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3450  if (!UseThreadPriorities) return OS_OK;
3451
3452  int status = 0;
3453
3454  if (!fxcritical) {
3455    // Use thr_setprio only if we have a priority that thr_setprio understands
3456    status = thr_setprio(thread->osthread()->thread_id(), newpri);
3457  }
3458
3459  int lwp_status =
3460          set_lwp_class_and_priority(osthread->thread_id(),
3461                                     osthread->lwp_id(),
3462                                     newpri,
3463                                     fxcritical ? fxLimits.schedPolicy : myClass,
3464                                     !fxcritical);
3465  if (lwp_status != 0 && fxcritical) {
3466    // Try again, this time without changing the scheduling class
3467    newpri = java_MaxPriority_to_os_priority;
3468    lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
3469                                            osthread->lwp_id(),
3470                                            newpri, myClass, false);
3471  }
3472  status |= lwp_status;
3473  return (status == 0) ? OS_OK : OS_ERR;
3474}
3475
3476
3477OSReturn os::get_native_priority(const Thread* const thread,
3478                                 int *priority_ptr) {
3479  int p;
3480  if (!UseThreadPriorities) {
3481    *priority_ptr = NormalPriority;
3482    return OS_OK;
3483  }
3484  int status = thr_getprio(thread->osthread()->thread_id(), &p);
3485  if (status != 0) {
3486    return OS_ERR;
3487  }
3488  *priority_ptr = p;
3489  return OS_OK;
3490}
3491
3492
3493// Hint to the underlying OS that a task switch would not be good.
3494// Void return because it's a hint and can fail.
3495void os::hint_no_preempt() {
3496  schedctl_start(schedctl_init());
3497}
3498
3499static void resume_clear_context(OSThread *osthread) {
3500  osthread->set_ucontext(NULL);
3501}
3502
3503static void suspend_save_context(OSThread *osthread, ucontext_t* context) {
3504  osthread->set_ucontext(context);
3505}
3506
3507static PosixSemaphore sr_semaphore;
3508
3509void os::Solaris::SR_handler(Thread* thread, ucontext_t* uc) {
3510  // Save and restore errno to avoid confusing native code with EINTR
3511  // after sigsuspend.
3512  int old_errno = errno;
3513
3514  OSThread* osthread = thread->osthread();
3515  assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3516
3517  os::SuspendResume::State current = osthread->sr.state();
3518  if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3519    suspend_save_context(osthread, uc);
3520
3521    // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3522    os::SuspendResume::State state = osthread->sr.suspended();
3523    if (state == os::SuspendResume::SR_SUSPENDED) {
3524      sigset_t suspend_set;  // signals for sigsuspend()
3525
3526      // get current set of blocked signals and unblock resume signal
3527      pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3528      sigdelset(&suspend_set, ASYNC_SIGNAL);
3529
3530      sr_semaphore.signal();
3531      // wait here until we are resumed
3532      while (1) {
3533        sigsuspend(&suspend_set);
3534
3535        os::SuspendResume::State result = osthread->sr.running();
3536        if (result == os::SuspendResume::SR_RUNNING) {
3537          sr_semaphore.signal();
3538          break;
3539        }
3540      }
3541
3542    } else if (state == os::SuspendResume::SR_RUNNING) {
3543      // request was cancelled, continue
3544    } else {
3545      ShouldNotReachHere();
3546    }
3547
3548    resume_clear_context(osthread);
3549  } else if (current == os::SuspendResume::SR_RUNNING) {
3550    // request was cancelled, continue
3551  } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3552    // ignore
3553  } else {
3554    // ignore
3555  }
3556
3557  errno = old_errno;
3558}
3559
3560void os::print_statistics() {
3561}
3562
3563bool os::message_box(const char* title, const char* message) {
3564  int i;
3565  fdStream err(defaultStream::error_fd());
3566  for (i = 0; i < 78; i++) err.print_raw("=");
3567  err.cr();
3568  err.print_raw_cr(title);
3569  for (i = 0; i < 78; i++) err.print_raw("-");
3570  err.cr();
3571  err.print_raw_cr(message);
3572  for (i = 0; i < 78; i++) err.print_raw("=");
3573  err.cr();
3574
3575  char buf[16];
3576  // Prevent process from exiting upon "read error" without consuming all CPU
3577  while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3578
3579  return buf[0] == 'y' || buf[0] == 'Y';
3580}
3581
3582static int sr_notify(OSThread* osthread) {
3583  int status = thr_kill(osthread->thread_id(), ASYNC_SIGNAL);
3584  assert_status(status == 0, status, "thr_kill");
3585  return status;
3586}
3587
3588// "Randomly" selected value for how long we want to spin
3589// before bailing out on suspending a thread, also how often
3590// we send a signal to a thread we want to resume
3591static const int RANDOMLY_LARGE_INTEGER = 1000000;
3592static const int RANDOMLY_LARGE_INTEGER2 = 100;
3593
3594static bool do_suspend(OSThread* osthread) {
3595  assert(osthread->sr.is_running(), "thread should be running");
3596  assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3597
3598  // mark as suspended and send signal
3599  if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3600    // failed to switch, state wasn't running?
3601    ShouldNotReachHere();
3602    return false;
3603  }
3604
3605  if (sr_notify(osthread) != 0) {
3606    ShouldNotReachHere();
3607  }
3608
3609  // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3610  while (true) {
3611    if (sr_semaphore.timedwait(0, 2000 * NANOSECS_PER_MILLISEC)) {
3612      break;
3613    } else {
3614      // timeout
3615      os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3616      if (cancelled == os::SuspendResume::SR_RUNNING) {
3617        return false;
3618      } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3619        // make sure that we consume the signal on the semaphore as well
3620        sr_semaphore.wait();
3621        break;
3622      } else {
3623        ShouldNotReachHere();
3624        return false;
3625      }
3626    }
3627  }
3628
3629  guarantee(osthread->sr.is_suspended(), "Must be suspended");
3630  return true;
3631}
3632
3633static void do_resume(OSThread* osthread) {
3634  assert(osthread->sr.is_suspended(), "thread should be suspended");
3635  assert(!sr_semaphore.trywait(), "invalid semaphore state");
3636
3637  if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3638    // failed to switch to WAKEUP_REQUEST
3639    ShouldNotReachHere();
3640    return;
3641  }
3642
3643  while (true) {
3644    if (sr_notify(osthread) == 0) {
3645      if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3646        if (osthread->sr.is_running()) {
3647          return;
3648        }
3649      }
3650    } else {
3651      ShouldNotReachHere();
3652    }
3653  }
3654
3655  guarantee(osthread->sr.is_running(), "Must be running!");
3656}
3657
3658void os::SuspendedThreadTask::internal_do_task() {
3659  if (do_suspend(_thread->osthread())) {
3660    SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3661    do_task(context);
3662    do_resume(_thread->osthread());
3663  }
3664}
3665
3666class PcFetcher : public os::SuspendedThreadTask {
3667 public:
3668  PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
3669  ExtendedPC result();
3670 protected:
3671  void do_task(const os::SuspendedThreadTaskContext& context);
3672 private:
3673  ExtendedPC _epc;
3674};
3675
3676ExtendedPC PcFetcher::result() {
3677  guarantee(is_done(), "task is not done yet.");
3678  return _epc;
3679}
3680
3681void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
3682  Thread* thread = context.thread();
3683  OSThread* osthread = thread->osthread();
3684  if (osthread->ucontext() != NULL) {
3685    _epc = os::Solaris::ucontext_get_pc((const ucontext_t *) context.ucontext());
3686  } else {
3687    // NULL context is unexpected, double-check this is the VMThread
3688    guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3689  }
3690}
3691
3692// A lightweight implementation that does not suspend the target thread and
3693// thus returns only a hint. Used for profiling only!
3694ExtendedPC os::get_thread_pc(Thread* thread) {
3695  // Make sure that it is called by the watcher and the Threads lock is owned.
3696  assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
3697  // For now, is only used to profile the VM Thread
3698  assert(thread->is_VM_thread(), "Can only be called for VMThread");
3699  PcFetcher fetcher(thread);
3700  fetcher.run();
3701  return fetcher.result();
3702}
3703
3704
3705// This does not do anything on Solaris. This is basically a hook for being
3706// able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
3707void os::os_exception_wrapper(java_call_t f, JavaValue* value,
3708                              const methodHandle& method, JavaCallArguments* args,
3709                              Thread* thread) {
3710  f(value, method, args, thread);
3711}
3712
3713// This routine may be used by user applications as a "hook" to catch signals.
3714// The user-defined signal handler must pass unrecognized signals to this
3715// routine, and if it returns true (non-zero), then the signal handler must
3716// return immediately.  If the flag "abort_if_unrecognized" is true, then this
3717// routine will never retun false (zero), but instead will execute a VM panic
3718// routine kill the process.
3719//
3720// If this routine returns false, it is OK to call it again.  This allows
3721// the user-defined signal handler to perform checks either before or after
3722// the VM performs its own checks.  Naturally, the user code would be making
3723// a serious error if it tried to handle an exception (such as a null check
3724// or breakpoint) that the VM was generating for its own correct operation.
3725//
3726// This routine may recognize any of the following kinds of signals:
3727// SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
3728// ASYNC_SIGNAL.
3729// It should be consulted by handlers for any of those signals.
3730//
3731// The caller of this routine must pass in the three arguments supplied
3732// to the function referred to in the "sa_sigaction" (not the "sa_handler")
3733// field of the structure passed to sigaction().  This routine assumes that
3734// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3735//
3736// Note that the VM will print warnings if it detects conflicting signal
3737// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3738//
3739extern "C" JNIEXPORT int JVM_handle_solaris_signal(int signo,
3740                                                   siginfo_t* siginfo,
3741                                                   void* ucontext,
3742                                                   int abort_if_unrecognized);
3743
3744
3745void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
3746  int orig_errno = errno;  // Preserve errno value over signal handler.
3747  JVM_handle_solaris_signal(sig, info, ucVoid, true);
3748  errno = orig_errno;
3749}
3750
3751// This boolean allows users to forward their own non-matching signals
3752// to JVM_handle_solaris_signal, harmlessly.
3753bool os::Solaris::signal_handlers_are_installed = false;
3754
3755// For signal-chaining
3756bool os::Solaris::libjsig_is_loaded = false;
3757typedef struct sigaction *(*get_signal_t)(int);
3758get_signal_t os::Solaris::get_signal_action = NULL;
3759
3760struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
3761  struct sigaction *actp = NULL;
3762
3763  if ((libjsig_is_loaded)  && (sig <= Maxsignum)) {
3764    // Retrieve the old signal handler from libjsig
3765    actp = (*get_signal_action)(sig);
3766  }
3767  if (actp == NULL) {
3768    // Retrieve the preinstalled signal handler from jvm
3769    actp = get_preinstalled_handler(sig);
3770  }
3771
3772  return actp;
3773}
3774
3775static bool call_chained_handler(struct sigaction *actp, int sig,
3776                                 siginfo_t *siginfo, void *context) {
3777  // Call the old signal handler
3778  if (actp->sa_handler == SIG_DFL) {
3779    // It's more reasonable to let jvm treat it as an unexpected exception
3780    // instead of taking the default action.
3781    return false;
3782  } else if (actp->sa_handler != SIG_IGN) {
3783    if ((actp->sa_flags & SA_NODEFER) == 0) {
3784      // automaticlly block the signal
3785      sigaddset(&(actp->sa_mask), sig);
3786    }
3787
3788    sa_handler_t hand;
3789    sa_sigaction_t sa;
3790    bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3791    // retrieve the chained handler
3792    if (siginfo_flag_set) {
3793      sa = actp->sa_sigaction;
3794    } else {
3795      hand = actp->sa_handler;
3796    }
3797
3798    if ((actp->sa_flags & SA_RESETHAND) != 0) {
3799      actp->sa_handler = SIG_DFL;
3800    }
3801
3802    // try to honor the signal mask
3803    sigset_t oset;
3804    pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3805
3806    // call into the chained handler
3807    if (siginfo_flag_set) {
3808      (*sa)(sig, siginfo, context);
3809    } else {
3810      (*hand)(sig);
3811    }
3812
3813    // restore the signal mask
3814    pthread_sigmask(SIG_SETMASK, &oset, 0);
3815  }
3816  // Tell jvm's signal handler the signal is taken care of.
3817  return true;
3818}
3819
3820bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3821  bool chained = false;
3822  // signal-chaining
3823  if (UseSignalChaining) {
3824    struct sigaction *actp = get_chained_signal_action(sig);
3825    if (actp != NULL) {
3826      chained = call_chained_handler(actp, sig, siginfo, context);
3827    }
3828  }
3829  return chained;
3830}
3831
3832struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
3833  assert((chainedsigactions != (struct sigaction *)NULL) &&
3834         (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3835  if (preinstalled_sigs[sig] != 0) {
3836    return &chainedsigactions[sig];
3837  }
3838  return NULL;
3839}
3840
3841void os::Solaris::save_preinstalled_handler(int sig,
3842                                            struct sigaction& oldAct) {
3843  assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
3844  assert((chainedsigactions != (struct sigaction *)NULL) &&
3845         (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3846  chainedsigactions[sig] = oldAct;
3847  preinstalled_sigs[sig] = 1;
3848}
3849
3850void os::Solaris::set_signal_handler(int sig, bool set_installed,
3851                                     bool oktochain) {
3852  // Check for overwrite.
3853  struct sigaction oldAct;
3854  sigaction(sig, (struct sigaction*)NULL, &oldAct);
3855  void* oldhand =
3856      oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3857                          : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3858  if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3859      oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3860      oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
3861    if (AllowUserSignalHandlers || !set_installed) {
3862      // Do not overwrite; user takes responsibility to forward to us.
3863      return;
3864    } else if (UseSignalChaining) {
3865      if (oktochain) {
3866        // save the old handler in jvm
3867        save_preinstalled_handler(sig, oldAct);
3868      } else {
3869        vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal.");
3870      }
3871      // libjsig also interposes the sigaction() call below and saves the
3872      // old sigaction on it own.
3873    } else {
3874      fatal("Encountered unexpected pre-existing sigaction handler "
3875            "%#lx for signal %d.", (long)oldhand, sig);
3876    }
3877  }
3878
3879  struct sigaction sigAct;
3880  sigfillset(&(sigAct.sa_mask));
3881  sigAct.sa_handler = SIG_DFL;
3882
3883  sigAct.sa_sigaction = signalHandler;
3884  // Handle SIGSEGV on alternate signal stack if
3885  // not using stack banging
3886  if (!UseStackBanging && sig == SIGSEGV) {
3887    sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
3888  } else {
3889    sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
3890  }
3891  os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
3892
3893  sigaction(sig, &sigAct, &oldAct);
3894
3895  void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3896                                       : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3897  assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3898}
3899
3900
3901#define DO_SIGNAL_CHECK(sig)                      \
3902  do {                                            \
3903    if (!sigismember(&check_signal_done, sig)) {  \
3904      os::Solaris::check_signal_handler(sig);     \
3905    }                                             \
3906  } while (0)
3907
3908// This method is a periodic task to check for misbehaving JNI applications
3909// under CheckJNI, we can add any periodic checks here
3910
3911void os::run_periodic_checks() {
3912  // A big source of grief is hijacking virt. addr 0x0 on Solaris,
3913  // thereby preventing a NULL checks.
3914  if (!check_addr0_done) check_addr0_done = check_addr0(tty);
3915
3916  if (check_signals == false) return;
3917
3918  // SEGV and BUS if overridden could potentially prevent
3919  // generation of hs*.log in the event of a crash, debugging
3920  // such a case can be very challenging, so we absolutely
3921  // check for the following for a good measure:
3922  DO_SIGNAL_CHECK(SIGSEGV);
3923  DO_SIGNAL_CHECK(SIGILL);
3924  DO_SIGNAL_CHECK(SIGFPE);
3925  DO_SIGNAL_CHECK(SIGBUS);
3926  DO_SIGNAL_CHECK(SIGPIPE);
3927  DO_SIGNAL_CHECK(SIGXFSZ);
3928  DO_SIGNAL_CHECK(ASYNC_SIGNAL);
3929
3930  // ReduceSignalUsage allows the user to override these handlers
3931  // see comments at the very top and jvm_solaris.h
3932  if (!ReduceSignalUsage) {
3933    DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3934    DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3935    DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3936    DO_SIGNAL_CHECK(BREAK_SIGNAL);
3937  }
3938}
3939
3940typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3941
3942static os_sigaction_t os_sigaction = NULL;
3943
3944void os::Solaris::check_signal_handler(int sig) {
3945  char buf[O_BUFLEN];
3946  address jvmHandler = NULL;
3947
3948  struct sigaction act;
3949  if (os_sigaction == NULL) {
3950    // only trust the default sigaction, in case it has been interposed
3951    os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3952    if (os_sigaction == NULL) return;
3953  }
3954
3955  os_sigaction(sig, (struct sigaction*)NULL, &act);
3956
3957  address thisHandler = (act.sa_flags & SA_SIGINFO)
3958    ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3959    : CAST_FROM_FN_PTR(address, act.sa_handler);
3960
3961
3962  switch (sig) {
3963  case SIGSEGV:
3964  case SIGBUS:
3965  case SIGFPE:
3966  case SIGPIPE:
3967  case SIGXFSZ:
3968  case SIGILL:
3969  case ASYNC_SIGNAL:
3970    jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
3971    break;
3972
3973  case SHUTDOWN1_SIGNAL:
3974  case SHUTDOWN2_SIGNAL:
3975  case SHUTDOWN3_SIGNAL:
3976  case BREAK_SIGNAL:
3977    jvmHandler = (address)user_handler();
3978    break;
3979
3980  default:
3981      return;
3982  }
3983
3984  if (thisHandler != jvmHandler) {
3985    tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3986    tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3987    tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3988    // No need to check this sig any longer
3989    sigaddset(&check_signal_done, sig);
3990    // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
3991    if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
3992      tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
3993                    exception_name(sig, buf, O_BUFLEN));
3994    }
3995  } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
3996    tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3997    tty->print("expected:");
3998    os::Posix::print_sa_flags(tty, os::Solaris::get_our_sigflags(sig));
3999    tty->cr();
4000    tty->print("  found:");
4001    os::Posix::print_sa_flags(tty, act.sa_flags);
4002    tty->cr();
4003    // No need to check this sig any longer
4004    sigaddset(&check_signal_done, sig);
4005  }
4006
4007  // Print all the signal handler state
4008  if (sigismember(&check_signal_done, sig)) {
4009    print_signal_handlers(tty, buf, O_BUFLEN);
4010  }
4011
4012}
4013
4014void os::Solaris::install_signal_handlers() {
4015  signal_handlers_are_installed = true;
4016
4017  // signal-chaining
4018  typedef void (*signal_setting_t)();
4019  signal_setting_t begin_signal_setting = NULL;
4020  signal_setting_t end_signal_setting = NULL;
4021  begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4022                                        dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4023  if (begin_signal_setting != NULL) {
4024    end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4025                                        dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4026    get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4027                                       dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4028    get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4029                                         dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4030    libjsig_is_loaded = true;
4031    if (os::Solaris::get_libjsig_version != NULL) {
4032      int libjsigversion =  (*os::Solaris::get_libjsig_version)();
4033      assert(libjsigversion == JSIG_VERSION_1_4_1, "libjsig version mismatch");
4034    }
4035    assert(UseSignalChaining, "should enable signal-chaining");
4036  }
4037  if (libjsig_is_loaded) {
4038    // Tell libjsig jvm is setting signal handlers
4039    (*begin_signal_setting)();
4040  }
4041
4042  set_signal_handler(SIGSEGV, true, true);
4043  set_signal_handler(SIGPIPE, true, true);
4044  set_signal_handler(SIGXFSZ, true, true);
4045  set_signal_handler(SIGBUS, true, true);
4046  set_signal_handler(SIGILL, true, true);
4047  set_signal_handler(SIGFPE, true, true);
4048  set_signal_handler(ASYNC_SIGNAL, true, true);
4049
4050  if (libjsig_is_loaded) {
4051    // Tell libjsig jvm finishes setting signal handlers
4052    (*end_signal_setting)();
4053  }
4054
4055  // We don't activate signal checker if libjsig is in place, we trust ourselves
4056  // and if UserSignalHandler is installed all bets are off.
4057  // Log that signal checking is off only if -verbose:jni is specified.
4058  if (CheckJNICalls) {
4059    if (libjsig_is_loaded) {
4060      if (PrintJNIResolving) {
4061        tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4062      }
4063      check_signals = false;
4064    }
4065    if (AllowUserSignalHandlers) {
4066      if (PrintJNIResolving) {
4067        tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4068      }
4069      check_signals = false;
4070    }
4071  }
4072}
4073
4074
4075void report_error(const char* file_name, int line_no, const char* title,
4076                  const char* format, ...);
4077
4078// (Static) wrappers for the liblgrp API
4079os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4080os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4081os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4082os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4083os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4084os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4085os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4086os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4087os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4088
4089static address resolve_symbol_lazy(const char* name) {
4090  address addr = (address) dlsym(RTLD_DEFAULT, name);
4091  if (addr == NULL) {
4092    // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4093    addr = (address) dlsym(RTLD_NEXT, name);
4094  }
4095  return addr;
4096}
4097
4098static address resolve_symbol(const char* name) {
4099  address addr = resolve_symbol_lazy(name);
4100  if (addr == NULL) {
4101    fatal(dlerror());
4102  }
4103  return addr;
4104}
4105
4106void os::Solaris::libthread_init() {
4107  address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4108
4109  lwp_priocntl_init();
4110
4111  // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4112  if (func == NULL) {
4113    func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4114    // Guarantee that this VM is running on an new enough OS (5.6 or
4115    // later) that it will have a new enough libthread.so.
4116    guarantee(func != NULL, "libthread.so is too old.");
4117  }
4118
4119  int size;
4120  void (*handler_info_func)(address *, int *);
4121  handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4122  handler_info_func(&handler_start, &size);
4123  handler_end = handler_start + size;
4124}
4125
4126
4127int_fnP_mutex_tP os::Solaris::_mutex_lock;
4128int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4129int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4130int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4131int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4132int os::Solaris::_mutex_scope = USYNC_THREAD;
4133
4134int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4135int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4136int_fnP_cond_tP os::Solaris::_cond_signal;
4137int_fnP_cond_tP os::Solaris::_cond_broadcast;
4138int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4139int_fnP_cond_tP os::Solaris::_cond_destroy;
4140int os::Solaris::_cond_scope = USYNC_THREAD;
4141
4142void os::Solaris::synchronization_init() {
4143  if (UseLWPSynchronization) {
4144    os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4145    os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4146    os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4147    os::Solaris::set_mutex_init(lwp_mutex_init);
4148    os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4149    os::Solaris::set_mutex_scope(USYNC_THREAD);
4150
4151    os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4152    os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4153    os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4154    os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4155    os::Solaris::set_cond_init(lwp_cond_init);
4156    os::Solaris::set_cond_destroy(lwp_cond_destroy);
4157    os::Solaris::set_cond_scope(USYNC_THREAD);
4158  } else {
4159    os::Solaris::set_mutex_scope(USYNC_THREAD);
4160    os::Solaris::set_cond_scope(USYNC_THREAD);
4161
4162    if (UsePthreads) {
4163      os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4164      os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4165      os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4166      os::Solaris::set_mutex_init(pthread_mutex_default_init);
4167      os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4168
4169      os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4170      os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4171      os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4172      os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4173      os::Solaris::set_cond_init(pthread_cond_default_init);
4174      os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4175    } else {
4176      os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4177      os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4178      os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4179      os::Solaris::set_mutex_init(::mutex_init);
4180      os::Solaris::set_mutex_destroy(::mutex_destroy);
4181
4182      os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4183      os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4184      os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4185      os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4186      os::Solaris::set_cond_init(::cond_init);
4187      os::Solaris::set_cond_destroy(::cond_destroy);
4188    }
4189  }
4190}
4191
4192bool os::Solaris::liblgrp_init() {
4193  void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4194  if (handle != NULL) {
4195    os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4196    os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4197    os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4198    os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4199    os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4200    os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4201    os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4202    os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4203                                                      dlsym(handle, "lgrp_cookie_stale")));
4204
4205    lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4206    set_lgrp_cookie(c);
4207    return true;
4208  }
4209  return false;
4210}
4211
4212// int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4213typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4214static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4215
4216void init_pset_getloadavg_ptr(void) {
4217  pset_getloadavg_ptr =
4218    (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4219  if (pset_getloadavg_ptr == NULL) {
4220    log_warning(os)("pset_getloadavg function not found");
4221  }
4222}
4223
4224int os::Solaris::_dev_zero_fd = -1;
4225
4226// this is called _before_ the global arguments have been parsed
4227void os::init(void) {
4228  _initial_pid = getpid();
4229
4230  max_hrtime = first_hrtime = gethrtime();
4231
4232  init_random(1234567);
4233
4234  page_size = sysconf(_SC_PAGESIZE);
4235  if (page_size == -1) {
4236    fatal("os_solaris.cpp: os::init: sysconf failed (%s)", os::strerror(errno));
4237  }
4238  init_page_sizes((size_t) page_size);
4239
4240  Solaris::initialize_system_info();
4241
4242  int fd = ::open("/dev/zero", O_RDWR);
4243  if (fd < 0) {
4244    fatal("os::init: cannot open /dev/zero (%s)", os::strerror(errno));
4245  } else {
4246    Solaris::set_dev_zero_fd(fd);
4247
4248    // Close on exec, child won't inherit.
4249    fcntl(fd, F_SETFD, FD_CLOEXEC);
4250  }
4251
4252  clock_tics_per_sec = CLK_TCK;
4253
4254  // check if dladdr1() exists; dladdr1 can provide more information than
4255  // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4256  // and is available on linker patches for 5.7 and 5.8.
4257  // libdl.so must have been loaded, this call is just an entry lookup
4258  void * hdl = dlopen("libdl.so", RTLD_NOW);
4259  if (hdl) {
4260    dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4261  }
4262
4263  // (Solaris only) this switches to calls that actually do locking.
4264  ThreadCritical::initialize();
4265
4266  main_thread = thr_self();
4267
4268  // dynamic lookup of functions that may not be available in our lowest
4269  // supported Solaris release
4270  void * handle = dlopen("libc.so.1", RTLD_LAZY);
4271  if (handle != NULL) {
4272    Solaris::_pthread_setname_np =  // from 11.3
4273        (Solaris::pthread_setname_np_func_t)dlsym(handle, "pthread_setname_np");
4274  }
4275}
4276
4277// To install functions for atexit system call
4278extern "C" {
4279  static void perfMemory_exit_helper() {
4280    perfMemory_exit();
4281  }
4282}
4283
4284// this is called _after_ the global arguments have been parsed
4285jint os::init_2(void) {
4286  // try to enable extended file IO ASAP, see 6431278
4287  os::Solaris::try_enable_extended_io();
4288
4289  // Allocate a single page and mark it as readable for safepoint polling.  Also
4290  // use this first mmap call to check support for MAP_ALIGN.
4291  address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
4292                                                      page_size,
4293                                                      MAP_PRIVATE | MAP_ALIGN,
4294                                                      PROT_READ);
4295  if (polling_page == NULL) {
4296    has_map_align = false;
4297    polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
4298                                                PROT_READ);
4299  }
4300
4301  os::set_polling_page(polling_page);
4302  log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4303
4304  if (!UseMembar) {
4305    address mem_serialize_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE);
4306    guarantee(mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4307    os::set_memory_serialize_page(mem_serialize_page);
4308    log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4309  }
4310
4311  // Check and sets minimum stack sizes against command line options
4312  if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4313    return JNI_ERR;
4314  }
4315
4316  Solaris::libthread_init();
4317
4318  if (UseNUMA) {
4319    if (!Solaris::liblgrp_init()) {
4320      UseNUMA = false;
4321    } else {
4322      size_t lgrp_limit = os::numa_get_groups_num();
4323      int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
4324      size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4325      FREE_C_HEAP_ARRAY(int, lgrp_ids);
4326      if (lgrp_num < 2) {
4327        // There's only one locality group, disable NUMA.
4328        UseNUMA = false;
4329      }
4330    }
4331    if (!UseNUMA && ForceNUMA) {
4332      UseNUMA = true;
4333    }
4334  }
4335
4336  Solaris::signal_sets_init();
4337  Solaris::init_signal_mem();
4338  Solaris::install_signal_handlers();
4339
4340  // initialize synchronization primitives to use either thread or
4341  // lwp synchronization (controlled by UseLWPSynchronization)
4342  Solaris::synchronization_init();
4343
4344  if (MaxFDLimit) {
4345    // set the number of file descriptors to max. print out error
4346    // if getrlimit/setrlimit fails but continue regardless.
4347    struct rlimit nbr_files;
4348    int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4349    if (status != 0) {
4350      log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4351    } else {
4352      nbr_files.rlim_cur = nbr_files.rlim_max;
4353      status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4354      if (status != 0) {
4355        log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4356      }
4357    }
4358  }
4359
4360  // Calculate theoretical max. size of Threads to guard gainst
4361  // artifical out-of-memory situations, where all available address-
4362  // space has been reserved by thread stacks. Default stack size is 1Mb.
4363  size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4364    JavaThread::stack_size_at_create() : (1*K*K);
4365  assert(pre_thread_stack_size != 0, "Must have a stack");
4366  // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4367  // we should start doing Virtual Memory banging. Currently when the threads will
4368  // have used all but 200Mb of space.
4369  size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4370  Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4371
4372  // at-exit methods are called in the reverse order of their registration.
4373  // In Solaris 7 and earlier, atexit functions are called on return from
4374  // main or as a result of a call to exit(3C). There can be only 32 of
4375  // these functions registered and atexit() does not set errno. In Solaris
4376  // 8 and later, there is no limit to the number of functions registered
4377  // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4378  // functions are called upon dlclose(3DL) in addition to return from main
4379  // and exit(3C).
4380
4381  if (PerfAllowAtExitRegistration) {
4382    // only register atexit functions if PerfAllowAtExitRegistration is set.
4383    // atexit functions can be delayed until process exit time, which
4384    // can be problematic for embedded VM situations. Embedded VMs should
4385    // call DestroyJavaVM() to assure that VM resources are released.
4386
4387    // note: perfMemory_exit_helper atexit function may be removed in
4388    // the future if the appropriate cleanup code can be added to the
4389    // VM_Exit VMOperation's doit method.
4390    if (atexit(perfMemory_exit_helper) != 0) {
4391      warning("os::init2 atexit(perfMemory_exit_helper) failed");
4392    }
4393  }
4394
4395  // Init pset_loadavg function pointer
4396  init_pset_getloadavg_ptr();
4397
4398  return JNI_OK;
4399}
4400
4401// Mark the polling page as unreadable
4402void os::make_polling_page_unreadable(void) {
4403  if (mprotect((char *)_polling_page, page_size, PROT_NONE) != 0) {
4404    fatal("Could not disable polling page");
4405  }
4406}
4407
4408// Mark the polling page as readable
4409void os::make_polling_page_readable(void) {
4410  if (mprotect((char *)_polling_page, page_size, PROT_READ) != 0) {
4411    fatal("Could not enable polling page");
4412  }
4413}
4414
4415// Is a (classpath) directory empty?
4416bool os::dir_is_empty(const char* path) {
4417  DIR *dir = NULL;
4418  struct dirent *ptr;
4419
4420  dir = opendir(path);
4421  if (dir == NULL) return true;
4422
4423  // Scan the directory
4424  bool result = true;
4425  char buf[sizeof(struct dirent) + MAX_PATH];
4426  struct dirent *dbuf = (struct dirent *) buf;
4427  while (result && (ptr = readdir(dir, dbuf)) != NULL) {
4428    if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4429      result = false;
4430    }
4431  }
4432  closedir(dir);
4433  return result;
4434}
4435
4436// This code originates from JDK's sysOpen and open64_w
4437// from src/solaris/hpi/src/system_md.c
4438
4439int os::open(const char *path, int oflag, int mode) {
4440  if (strlen(path) > MAX_PATH - 1) {
4441    errno = ENAMETOOLONG;
4442    return -1;
4443  }
4444  int fd;
4445
4446  fd = ::open64(path, oflag, mode);
4447  if (fd == -1) return -1;
4448
4449  // If the open succeeded, the file might still be a directory
4450  {
4451    struct stat64 buf64;
4452    int ret = ::fstat64(fd, &buf64);
4453    int st_mode = buf64.st_mode;
4454
4455    if (ret != -1) {
4456      if ((st_mode & S_IFMT) == S_IFDIR) {
4457        errno = EISDIR;
4458        ::close(fd);
4459        return -1;
4460      }
4461    } else {
4462      ::close(fd);
4463      return -1;
4464    }
4465  }
4466
4467  // 32-bit Solaris systems suffer from:
4468  //
4469  // - an historical default soft limit of 256 per-process file
4470  //   descriptors that is too low for many Java programs.
4471  //
4472  // - a design flaw where file descriptors created using stdio
4473  //   fopen must be less than 256, _even_ when the first limit above
4474  //   has been raised.  This can cause calls to fopen (but not calls to
4475  //   open, for example) to fail mysteriously, perhaps in 3rd party
4476  //   native code (although the JDK itself uses fopen).  One can hardly
4477  //   criticize them for using this most standard of all functions.
4478  //
4479  // We attempt to make everything work anyways by:
4480  //
4481  // - raising the soft limit on per-process file descriptors beyond
4482  //   256
4483  //
4484  // - As of Solaris 10u4, we can request that Solaris raise the 256
4485  //   stdio fopen limit by calling function enable_extended_FILE_stdio.
4486  //   This is done in init_2 and recorded in enabled_extended_FILE_stdio
4487  //
4488  // - If we are stuck on an old (pre 10u4) Solaris system, we can
4489  //   workaround the bug by remapping non-stdio file descriptors below
4490  //   256 to ones beyond 256, which is done below.
4491  //
4492  // See:
4493  // 1085341: 32-bit stdio routines should support file descriptors >255
4494  // 6533291: Work around 32-bit Solaris stdio limit of 256 open files
4495  // 6431278: Netbeans crash on 32 bit Solaris: need to call
4496  //          enable_extended_FILE_stdio() in VM initialisation
4497  // Giri Mandalika's blog
4498  // http://technopark02.blogspot.com/2005_05_01_archive.html
4499  //
4500#ifndef  _LP64
4501  if ((!enabled_extended_FILE_stdio) && fd < 256) {
4502    int newfd = ::fcntl(fd, F_DUPFD, 256);
4503    if (newfd != -1) {
4504      ::close(fd);
4505      fd = newfd;
4506    }
4507  }
4508#endif // 32-bit Solaris
4509
4510  // All file descriptors that are opened in the JVM and not
4511  // specifically destined for a subprocess should have the
4512  // close-on-exec flag set.  If we don't set it, then careless 3rd
4513  // party native code might fork and exec without closing all
4514  // appropriate file descriptors (e.g. as we do in closeDescriptors in
4515  // UNIXProcess.c), and this in turn might:
4516  //
4517  // - cause end-of-file to fail to be detected on some file
4518  //   descriptors, resulting in mysterious hangs, or
4519  //
4520  // - might cause an fopen in the subprocess to fail on a system
4521  //   suffering from bug 1085341.
4522  //
4523  // (Yes, the default setting of the close-on-exec flag is a Unix
4524  // design flaw)
4525  //
4526  // See:
4527  // 1085341: 32-bit stdio routines should support file descriptors >255
4528  // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4529  // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4530  //
4531#ifdef FD_CLOEXEC
4532  {
4533    int flags = ::fcntl(fd, F_GETFD);
4534    if (flags != -1) {
4535      ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4536    }
4537  }
4538#endif
4539
4540  return fd;
4541}
4542
4543// create binary file, rewriting existing file if required
4544int os::create_binary_file(const char* path, bool rewrite_existing) {
4545  int oflags = O_WRONLY | O_CREAT;
4546  if (!rewrite_existing) {
4547    oflags |= O_EXCL;
4548  }
4549  return ::open64(path, oflags, S_IREAD | S_IWRITE);
4550}
4551
4552// return current position of file pointer
4553jlong os::current_file_offset(int fd) {
4554  return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4555}
4556
4557// move file pointer to the specified offset
4558jlong os::seek_to_file_offset(int fd, jlong offset) {
4559  return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4560}
4561
4562jlong os::lseek(int fd, jlong offset, int whence) {
4563  return (jlong) ::lseek64(fd, offset, whence);
4564}
4565
4566char * os::native_path(char *path) {
4567  return path;
4568}
4569
4570int os::ftruncate(int fd, jlong length) {
4571  return ::ftruncate64(fd, length);
4572}
4573
4574int os::fsync(int fd)  {
4575  RESTARTABLE_RETURN_INT(::fsync(fd));
4576}
4577
4578int os::available(int fd, jlong *bytes) {
4579  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
4580         "Assumed _thread_in_native");
4581  jlong cur, end;
4582  int mode;
4583  struct stat64 buf64;
4584
4585  if (::fstat64(fd, &buf64) >= 0) {
4586    mode = buf64.st_mode;
4587    if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4588      int n,ioctl_return;
4589
4590      RESTARTABLE(::ioctl(fd, FIONREAD, &n), ioctl_return);
4591      if (ioctl_return>= 0) {
4592        *bytes = n;
4593        return 1;
4594      }
4595    }
4596  }
4597  if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4598    return 0;
4599  } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4600    return 0;
4601  } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4602    return 0;
4603  }
4604  *bytes = end - cur;
4605  return 1;
4606}
4607
4608// Map a block of memory.
4609char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4610                        char *addr, size_t bytes, bool read_only,
4611                        bool allow_exec) {
4612  int prot;
4613  int flags;
4614
4615  if (read_only) {
4616    prot = PROT_READ;
4617    flags = MAP_SHARED;
4618  } else {
4619    prot = PROT_READ | PROT_WRITE;
4620    flags = MAP_PRIVATE;
4621  }
4622
4623  if (allow_exec) {
4624    prot |= PROT_EXEC;
4625  }
4626
4627  if (addr != NULL) {
4628    flags |= MAP_FIXED;
4629  }
4630
4631  char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4632                                     fd, file_offset);
4633  if (mapped_address == MAP_FAILED) {
4634    return NULL;
4635  }
4636  return mapped_address;
4637}
4638
4639
4640// Remap a block of memory.
4641char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4642                          char *addr, size_t bytes, bool read_only,
4643                          bool allow_exec) {
4644  // same as map_memory() on this OS
4645  return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4646                        allow_exec);
4647}
4648
4649
4650// Unmap a block of memory.
4651bool os::pd_unmap_memory(char* addr, size_t bytes) {
4652  return munmap(addr, bytes) == 0;
4653}
4654
4655void os::pause() {
4656  char filename[MAX_PATH];
4657  if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4658    jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4659  } else {
4660    jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4661  }
4662
4663  int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4664  if (fd != -1) {
4665    struct stat buf;
4666    ::close(fd);
4667    while (::stat(filename, &buf) == 0) {
4668      (void)::poll(NULL, 0, 100);
4669    }
4670  } else {
4671    jio_fprintf(stderr,
4672                "Could not open pause file '%s', continuing immediately.\n", filename);
4673  }
4674}
4675
4676#ifndef PRODUCT
4677#ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4678// Turn this on if you need to trace synch operations.
4679// Set RECORD_SYNCH_LIMIT to a large-enough value,
4680// and call record_synch_enable and record_synch_disable
4681// around the computation of interest.
4682
4683void record_synch(char* name, bool returning);  // defined below
4684
4685class RecordSynch {
4686  char* _name;
4687 public:
4688  RecordSynch(char* name) :_name(name) { record_synch(_name, false); }
4689  ~RecordSynch()                       { record_synch(_name, true); }
4690};
4691
4692#define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
4693extern "C" ret name params {                                    \
4694  typedef ret name##_t params;                                  \
4695  static name##_t* implem = NULL;                               \
4696  static int callcount = 0;                                     \
4697  if (implem == NULL) {                                         \
4698    implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
4699    if (implem == NULL)  fatal(dlerror());                      \
4700  }                                                             \
4701  ++callcount;                                                  \
4702  RecordSynch _rs(#name);                                       \
4703  inner;                                                        \
4704  return implem args;                                           \
4705}
4706// in dbx, examine callcounts this way:
4707// for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
4708
4709#define CHECK_POINTER_OK(p) \
4710  (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
4711#define CHECK_MU \
4712  if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
4713#define CHECK_CV \
4714  if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
4715#define CHECK_P(p) \
4716  if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
4717
4718#define CHECK_MUTEX(mutex_op) \
4719  CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
4720
4721CHECK_MUTEX(   mutex_lock)
4722CHECK_MUTEX(  _mutex_lock)
4723CHECK_MUTEX( mutex_unlock)
4724CHECK_MUTEX(_mutex_unlock)
4725CHECK_MUTEX( mutex_trylock)
4726CHECK_MUTEX(_mutex_trylock)
4727
4728#define CHECK_COND(cond_op) \
4729  CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU; CHECK_CV);
4730
4731CHECK_COND( cond_wait);
4732CHECK_COND(_cond_wait);
4733CHECK_COND(_cond_wait_cancel);
4734
4735#define CHECK_COND2(cond_op) \
4736  CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU; CHECK_CV);
4737
4738CHECK_COND2( cond_timedwait);
4739CHECK_COND2(_cond_timedwait);
4740CHECK_COND2(_cond_timedwait_cancel);
4741
4742// do the _lwp_* versions too
4743#define mutex_t lwp_mutex_t
4744#define cond_t  lwp_cond_t
4745CHECK_MUTEX(  _lwp_mutex_lock)
4746CHECK_MUTEX(  _lwp_mutex_unlock)
4747CHECK_MUTEX(  _lwp_mutex_trylock)
4748CHECK_MUTEX( __lwp_mutex_lock)
4749CHECK_MUTEX( __lwp_mutex_unlock)
4750CHECK_MUTEX( __lwp_mutex_trylock)
4751CHECK_MUTEX(___lwp_mutex_lock)
4752CHECK_MUTEX(___lwp_mutex_unlock)
4753
4754CHECK_COND(  _lwp_cond_wait);
4755CHECK_COND( __lwp_cond_wait);
4756CHECK_COND(___lwp_cond_wait);
4757
4758CHECK_COND2(  _lwp_cond_timedwait);
4759CHECK_COND2( __lwp_cond_timedwait);
4760#undef mutex_t
4761#undef cond_t
4762
4763CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4764CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4765CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4766CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4767CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4768CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4769CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4770CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4771
4772
4773// recording machinery:
4774
4775enum { RECORD_SYNCH_LIMIT = 200 };
4776char* record_synch_name[RECORD_SYNCH_LIMIT];
4777void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
4778bool record_synch_returning[RECORD_SYNCH_LIMIT];
4779thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
4780int record_synch_count = 0;
4781bool record_synch_enabled = false;
4782
4783// in dbx, examine recorded data this way:
4784// for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
4785
4786void record_synch(char* name, bool returning) {
4787  if (record_synch_enabled) {
4788    if (record_synch_count < RECORD_SYNCH_LIMIT) {
4789      record_synch_name[record_synch_count] = name;
4790      record_synch_returning[record_synch_count] = returning;
4791      record_synch_thread[record_synch_count] = thr_self();
4792      record_synch_arg0ptr[record_synch_count] = &name;
4793      record_synch_count++;
4794    }
4795    // put more checking code here:
4796    // ...
4797  }
4798}
4799
4800void record_synch_enable() {
4801  // start collecting trace data, if not already doing so
4802  if (!record_synch_enabled)  record_synch_count = 0;
4803  record_synch_enabled = true;
4804}
4805
4806void record_synch_disable() {
4807  // stop collecting trace data
4808  record_synch_enabled = false;
4809}
4810
4811#endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4812#endif // PRODUCT
4813
4814const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4815const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
4816                               (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4817
4818
4819// JVMTI & JVM monitoring and management support
4820// The thread_cpu_time() and current_thread_cpu_time() are only
4821// supported if is_thread_cpu_time_supported() returns true.
4822// They are not supported on Solaris T1.
4823
4824// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4825// are used by JVM M&M and JVMTI to get user+sys or user CPU time
4826// of a thread.
4827//
4828// current_thread_cpu_time() and thread_cpu_time(Thread *)
4829// returns the fast estimate available on the platform.
4830
4831// hrtime_t gethrvtime() return value includes
4832// user time but does not include system time
4833jlong os::current_thread_cpu_time() {
4834  return (jlong) gethrvtime();
4835}
4836
4837jlong os::thread_cpu_time(Thread *thread) {
4838  // return user level CPU time only to be consistent with
4839  // what current_thread_cpu_time returns.
4840  // thread_cpu_time_info() must be changed if this changes
4841  return os::thread_cpu_time(thread, false /* user time only */);
4842}
4843
4844jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4845  if (user_sys_cpu_time) {
4846    return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
4847  } else {
4848    return os::current_thread_cpu_time();
4849  }
4850}
4851
4852jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4853  char proc_name[64];
4854  int count;
4855  prusage_t prusage;
4856  jlong lwp_time;
4857  int fd;
4858
4859  sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
4860          getpid(),
4861          thread->osthread()->lwp_id());
4862  fd = ::open(proc_name, O_RDONLY);
4863  if (fd == -1) return -1;
4864
4865  do {
4866    count = ::pread(fd,
4867                    (void *)&prusage.pr_utime,
4868                    thr_time_size,
4869                    thr_time_off);
4870  } while (count < 0 && errno == EINTR);
4871  ::close(fd);
4872  if (count < 0) return -1;
4873
4874  if (user_sys_cpu_time) {
4875    // user + system CPU time
4876    lwp_time = (((jlong)prusage.pr_stime.tv_sec +
4877                 (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
4878                 (jlong)prusage.pr_stime.tv_nsec +
4879                 (jlong)prusage.pr_utime.tv_nsec;
4880  } else {
4881    // user level CPU time only
4882    lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
4883                (jlong)prusage.pr_utime.tv_nsec;
4884  }
4885
4886  return (lwp_time);
4887}
4888
4889void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4890  info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4891  info_ptr->may_skip_backward = false;    // elapsed time not wall time
4892  info_ptr->may_skip_forward = false;     // elapsed time not wall time
4893  info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4894}
4895
4896void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4897  info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4898  info_ptr->may_skip_backward = false;    // elapsed time not wall time
4899  info_ptr->may_skip_forward = false;     // elapsed time not wall time
4900  info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4901}
4902
4903bool os::is_thread_cpu_time_supported() {
4904  return true;
4905}
4906
4907// System loadavg support.  Returns -1 if load average cannot be obtained.
4908// Return the load average for our processor set if the primitive exists
4909// (Solaris 9 and later).  Otherwise just return system wide loadavg.
4910int os::loadavg(double loadavg[], int nelem) {
4911  if (pset_getloadavg_ptr != NULL) {
4912    return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
4913  } else {
4914    return ::getloadavg(loadavg, nelem);
4915  }
4916}
4917
4918//---------------------------------------------------------------------------------
4919
4920bool os::find(address addr, outputStream* st) {
4921  Dl_info dlinfo;
4922  memset(&dlinfo, 0, sizeof(dlinfo));
4923  if (dladdr(addr, &dlinfo) != 0) {
4924    st->print(PTR_FORMAT ": ", addr);
4925    if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4926      st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
4927    } else if (dlinfo.dli_fbase != NULL) {
4928      st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
4929    } else {
4930      st->print("<absolute address>");
4931    }
4932    if (dlinfo.dli_fname != NULL) {
4933      st->print(" in %s", dlinfo.dli_fname);
4934    }
4935    if (dlinfo.dli_fbase != NULL) {
4936      st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4937    }
4938    st->cr();
4939
4940    if (Verbose) {
4941      // decode some bytes around the PC
4942      address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4943      address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4944      address       lowest = (address) dlinfo.dli_sname;
4945      if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4946      if (begin < lowest)  begin = lowest;
4947      Dl_info dlinfo2;
4948      if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4949          && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4950        end = (address) dlinfo2.dli_saddr;
4951      }
4952      Disassembler::decode(begin, end, st);
4953    }
4954    return true;
4955  }
4956  return false;
4957}
4958
4959// Following function has been added to support HotSparc's libjvm.so running
4960// under Solaris production JDK 1.2.2 / 1.3.0.  These came from
4961// src/solaris/hpi/native_threads in the EVM codebase.
4962//
4963// NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
4964// libraries and should thus be removed. We will leave it behind for a while
4965// until we no longer want to able to run on top of 1.3.0 Solaris production
4966// JDK. See 4341971.
4967
4968#define STACK_SLACK 0x800
4969
4970extern "C" {
4971  intptr_t sysThreadAvailableStackWithSlack() {
4972    stack_t st;
4973    intptr_t retval, stack_top;
4974    retval = thr_stksegment(&st);
4975    assert(retval == 0, "incorrect return value from thr_stksegment");
4976    assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
4977    assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
4978    stack_top=(intptr_t)st.ss_sp-st.ss_size;
4979    return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
4980  }
4981}
4982
4983// ObjectMonitor park-unpark infrastructure ...
4984//
4985// We implement Solaris and Linux PlatformEvents with the
4986// obvious condvar-mutex-flag triple.
4987// Another alternative that works quite well is pipes:
4988// Each PlatformEvent consists of a pipe-pair.
4989// The thread associated with the PlatformEvent
4990// calls park(), which reads from the input end of the pipe.
4991// Unpark() writes into the other end of the pipe.
4992// The write-side of the pipe must be set NDELAY.
4993// Unfortunately pipes consume a large # of handles.
4994// Native solaris lwp_park() and lwp_unpark() work nicely, too.
4995// Using pipes for the 1st few threads might be workable, however.
4996//
4997// park() is permitted to return spuriously.
4998// Callers of park() should wrap the call to park() in
4999// an appropriate loop.  A litmus test for the correct
5000// usage of park is the following: if park() were modified
5001// to immediately return 0 your code should still work,
5002// albeit degenerating to a spin loop.
5003//
5004// In a sense, park()-unpark() just provides more polite spinning
5005// and polling with the key difference over naive spinning being
5006// that a parked thread needs to be explicitly unparked() in order
5007// to wake up and to poll the underlying condition.
5008//
5009// Assumption:
5010//    Only one parker can exist on an event, which is why we allocate
5011//    them per-thread. Multiple unparkers can coexist.
5012//
5013// _Event transitions in park()
5014//   -1 => -1 : illegal
5015//    1 =>  0 : pass - return immediately
5016//    0 => -1 : block; then set _Event to 0 before returning
5017//
5018// _Event transitions in unpark()
5019//    0 => 1 : just return
5020//    1 => 1 : just return
5021//   -1 => either 0 or 1; must signal target thread
5022//         That is, we can safely transition _Event from -1 to either
5023//         0 or 1.
5024//
5025// _Event serves as a restricted-range semaphore.
5026//   -1 : thread is blocked, i.e. there is a waiter
5027//    0 : neutral: thread is running or ready,
5028//        could have been signaled after a wait started
5029//    1 : signaled - thread is running or ready
5030//
5031// Another possible encoding of _Event would be with
5032// explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5033//
5034// TODO-FIXME: add DTRACE probes for:
5035// 1.   Tx parks
5036// 2.   Ty unparks Tx
5037// 3.   Tx resumes from park
5038
5039
5040// value determined through experimentation
5041#define ROUNDINGFIX 11
5042
5043// utility to compute the abstime argument to timedwait.
5044// TODO-FIXME: switch from compute_abstime() to unpackTime().
5045
5046static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5047  // millis is the relative timeout time
5048  // abstime will be the absolute timeout time
5049  if (millis < 0)  millis = 0;
5050  struct timeval now;
5051  int status = gettimeofday(&now, NULL);
5052  assert(status == 0, "gettimeofday");
5053  jlong seconds = millis / 1000;
5054  jlong max_wait_period;
5055
5056  if (UseLWPSynchronization) {
5057    // forward port of fix for 4275818 (not sleeping long enough)
5058    // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5059    // _lwp_cond_timedwait() used a round_down algorithm rather
5060    // than a round_up. For millis less than our roundfactor
5061    // it rounded down to 0 which doesn't meet the spec.
5062    // For millis > roundfactor we may return a bit sooner, but
5063    // since we can not accurately identify the patch level and
5064    // this has already been fixed in Solaris 9 and 8 we will
5065    // leave it alone rather than always rounding down.
5066
5067    if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5068    // It appears that when we go directly through Solaris _lwp_cond_timedwait()
5069    // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
5070    max_wait_period = 21000000;
5071  } else {
5072    max_wait_period = 50000000;
5073  }
5074  millis %= 1000;
5075  if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
5076    seconds = max_wait_period;
5077  }
5078  abstime->tv_sec = now.tv_sec  + seconds;
5079  long       usec = now.tv_usec + millis * 1000;
5080  if (usec >= 1000000) {
5081    abstime->tv_sec += 1;
5082    usec -= 1000000;
5083  }
5084  abstime->tv_nsec = usec * 1000;
5085  return abstime;
5086}
5087
5088void os::PlatformEvent::park() {           // AKA: down()
5089  // Transitions for _Event:
5090  //   -1 => -1 : illegal
5091  //    1 =>  0 : pass - return immediately
5092  //    0 => -1 : block; then set _Event to 0 before returning
5093
5094  // Invariant: Only the thread associated with the Event/PlatformEvent
5095  // may call park().
5096  assert(_nParked == 0, "invariant");
5097
5098  int v;
5099  for (;;) {
5100    v = _Event;
5101    if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5102  }
5103  guarantee(v >= 0, "invariant");
5104  if (v == 0) {
5105    // Do this the hard way by blocking ...
5106    // See http://monaco.sfbay/detail.jsf?cr=5094058.
5107    int status = os::Solaris::mutex_lock(_mutex);
5108    assert_status(status == 0, status, "mutex_lock");
5109    guarantee(_nParked == 0, "invariant");
5110    ++_nParked;
5111    while (_Event < 0) {
5112      // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5113      // Treat this the same as if the wait was interrupted
5114      // With usr/lib/lwp going to kernel, always handle ETIME
5115      status = os::Solaris::cond_wait(_cond, _mutex);
5116      if (status == ETIME) status = EINTR;
5117      assert_status(status == 0 || status == EINTR, status, "cond_wait");
5118    }
5119    --_nParked;
5120    _Event = 0;
5121    status = os::Solaris::mutex_unlock(_mutex);
5122    assert_status(status == 0, status, "mutex_unlock");
5123    // Paranoia to ensure our locked and lock-free paths interact
5124    // correctly with each other.
5125    OrderAccess::fence();
5126  }
5127}
5128
5129int os::PlatformEvent::park(jlong millis) {
5130  // Transitions for _Event:
5131  //   -1 => -1 : illegal
5132  //    1 =>  0 : pass - return immediately
5133  //    0 => -1 : block; then set _Event to 0 before returning
5134
5135  guarantee(_nParked == 0, "invariant");
5136  int v;
5137  for (;;) {
5138    v = _Event;
5139    if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5140  }
5141  guarantee(v >= 0, "invariant");
5142  if (v != 0) return OS_OK;
5143
5144  int ret = OS_TIMEOUT;
5145  timestruc_t abst;
5146  compute_abstime(&abst, millis);
5147
5148  // See http://monaco.sfbay/detail.jsf?cr=5094058.
5149  int status = os::Solaris::mutex_lock(_mutex);
5150  assert_status(status == 0, status, "mutex_lock");
5151  guarantee(_nParked == 0, "invariant");
5152  ++_nParked;
5153  while (_Event < 0) {
5154    int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5155    assert_status(status == 0 || status == EINTR ||
5156                  status == ETIME || status == ETIMEDOUT,
5157                  status, "cond_timedwait");
5158    if (!FilterSpuriousWakeups) break;                // previous semantics
5159    if (status == ETIME || status == ETIMEDOUT) break;
5160    // We consume and ignore EINTR and spurious wakeups.
5161  }
5162  --_nParked;
5163  if (_Event >= 0) ret = OS_OK;
5164  _Event = 0;
5165  status = os::Solaris::mutex_unlock(_mutex);
5166  assert_status(status == 0, status, "mutex_unlock");
5167  // Paranoia to ensure our locked and lock-free paths interact
5168  // correctly with each other.
5169  OrderAccess::fence();
5170  return ret;
5171}
5172
5173void os::PlatformEvent::unpark() {
5174  // Transitions for _Event:
5175  //    0 => 1 : just return
5176  //    1 => 1 : just return
5177  //   -1 => either 0 or 1; must signal target thread
5178  //         That is, we can safely transition _Event from -1 to either
5179  //         0 or 1.
5180  // See also: "Semaphores in Plan 9" by Mullender & Cox
5181  //
5182  // Note: Forcing a transition from "-1" to "1" on an unpark() means
5183  // that it will take two back-to-back park() calls for the owning
5184  // thread to block. This has the benefit of forcing a spurious return
5185  // from the first park() call after an unpark() call which will help
5186  // shake out uses of park() and unpark() without condition variables.
5187
5188  if (Atomic::xchg(1, &_Event) >= 0) return;
5189
5190  // If the thread associated with the event was parked, wake it.
5191  // Wait for the thread assoc with the PlatformEvent to vacate.
5192  int status = os::Solaris::mutex_lock(_mutex);
5193  assert_status(status == 0, status, "mutex_lock");
5194  int AnyWaiters = _nParked;
5195  status = os::Solaris::mutex_unlock(_mutex);
5196  assert_status(status == 0, status, "mutex_unlock");
5197  guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5198  if (AnyWaiters != 0) {
5199    // Note that we signal() *after* dropping the lock for "immortal" Events.
5200    // This is safe and avoids a common class of  futile wakeups.  In rare
5201    // circumstances this can cause a thread to return prematurely from
5202    // cond_{timed}wait() but the spurious wakeup is benign and the victim
5203    // will simply re-test the condition and re-park itself.
5204    // This provides particular benefit if the underlying platform does not
5205    // provide wait morphing.
5206    status = os::Solaris::cond_signal(_cond);
5207    assert_status(status == 0, status, "cond_signal");
5208  }
5209}
5210
5211// JSR166
5212// -------------------------------------------------------
5213
5214// The solaris and linux implementations of park/unpark are fairly
5215// conservative for now, but can be improved. They currently use a
5216// mutex/condvar pair, plus _counter.
5217// Park decrements _counter if > 0, else does a condvar wait.  Unpark
5218// sets count to 1 and signals condvar.  Only one thread ever waits
5219// on the condvar. Contention seen when trying to park implies that someone
5220// is unparking you, so don't wait. And spurious returns are fine, so there
5221// is no need to track notifications.
5222
5223#define MAX_SECS 100000000
5224
5225// This code is common to linux and solaris and will be moved to a
5226// common place in dolphin.
5227//
5228// The passed in time value is either a relative time in nanoseconds
5229// or an absolute time in milliseconds. Either way it has to be unpacked
5230// into suitable seconds and nanoseconds components and stored in the
5231// given timespec structure.
5232// Given time is a 64-bit value and the time_t used in the timespec is only
5233// a signed-32-bit value (except on 64-bit Linux) we have to watch for
5234// overflow if times way in the future are given. Further on Solaris versions
5235// prior to 10 there is a restriction (see cond_timedwait) that the specified
5236// number of seconds, in abstime, is less than current_time  + 100,000,000.
5237// As it will be 28 years before "now + 100000000" will overflow we can
5238// ignore overflow and just impose a hard-limit on seconds using the value
5239// of "now + 100,000,000". This places a limit on the timeout of about 3.17
5240// years from "now".
5241//
5242static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5243  assert(time > 0, "convertTime");
5244
5245  struct timeval now;
5246  int status = gettimeofday(&now, NULL);
5247  assert(status == 0, "gettimeofday");
5248
5249  time_t max_secs = now.tv_sec + MAX_SECS;
5250
5251  if (isAbsolute) {
5252    jlong secs = time / 1000;
5253    if (secs > max_secs) {
5254      absTime->tv_sec = max_secs;
5255    } else {
5256      absTime->tv_sec = secs;
5257    }
5258    absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5259  } else {
5260    jlong secs = time / NANOSECS_PER_SEC;
5261    if (secs >= MAX_SECS) {
5262      absTime->tv_sec = max_secs;
5263      absTime->tv_nsec = 0;
5264    } else {
5265      absTime->tv_sec = now.tv_sec + secs;
5266      absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5267      if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5268        absTime->tv_nsec -= NANOSECS_PER_SEC;
5269        ++absTime->tv_sec; // note: this must be <= max_secs
5270      }
5271    }
5272  }
5273  assert(absTime->tv_sec >= 0, "tv_sec < 0");
5274  assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5275  assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5276  assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5277}
5278
5279void Parker::park(bool isAbsolute, jlong time) {
5280  // Ideally we'd do something useful while spinning, such
5281  // as calling unpackTime().
5282
5283  // Optional fast-path check:
5284  // Return immediately if a permit is available.
5285  // We depend on Atomic::xchg() having full barrier semantics
5286  // since we are doing a lock-free update to _counter.
5287  if (Atomic::xchg(0, &_counter) > 0) return;
5288
5289  // Optional fast-exit: Check interrupt before trying to wait
5290  Thread* thread = Thread::current();
5291  assert(thread->is_Java_thread(), "Must be JavaThread");
5292  JavaThread *jt = (JavaThread *)thread;
5293  if (Thread::is_interrupted(thread, false)) {
5294    return;
5295  }
5296
5297  // First, demultiplex/decode time arguments
5298  timespec absTime;
5299  if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5300    return;
5301  }
5302  if (time > 0) {
5303    // Warning: this code might be exposed to the old Solaris time
5304    // round-down bugs.  Grep "roundingFix" for details.
5305    unpackTime(&absTime, isAbsolute, time);
5306  }
5307
5308  // Enter safepoint region
5309  // Beware of deadlocks such as 6317397.
5310  // The per-thread Parker:: _mutex is a classic leaf-lock.
5311  // In particular a thread must never block on the Threads_lock while
5312  // holding the Parker:: mutex.  If safepoints are pending both the
5313  // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5314  ThreadBlockInVM tbivm(jt);
5315
5316  // Don't wait if cannot get lock since interference arises from
5317  // unblocking.  Also. check interrupt before trying wait
5318  if (Thread::is_interrupted(thread, false) ||
5319      os::Solaris::mutex_trylock(_mutex) != 0) {
5320    return;
5321  }
5322
5323  int status;
5324
5325  if (_counter > 0)  { // no wait needed
5326    _counter = 0;
5327    status = os::Solaris::mutex_unlock(_mutex);
5328    assert(status == 0, "invariant");
5329    // Paranoia to ensure our locked and lock-free paths interact
5330    // correctly with each other and Java-level accesses.
5331    OrderAccess::fence();
5332    return;
5333  }
5334
5335  OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5336  jt->set_suspend_equivalent();
5337  // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5338
5339  // Do this the hard way by blocking ...
5340  // See http://monaco.sfbay/detail.jsf?cr=5094058.
5341  if (time == 0) {
5342    status = os::Solaris::cond_wait(_cond, _mutex);
5343  } else {
5344    status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5345  }
5346  // Note that an untimed cond_wait() can sometimes return ETIME on older
5347  // versions of the Solaris.
5348  assert_status(status == 0 || status == EINTR ||
5349                status == ETIME || status == ETIMEDOUT,
5350                status, "cond_timedwait");
5351
5352  _counter = 0;
5353  status = os::Solaris::mutex_unlock(_mutex);
5354  assert_status(status == 0, status, "mutex_unlock");
5355  // Paranoia to ensure our locked and lock-free paths interact
5356  // correctly with each other and Java-level accesses.
5357  OrderAccess::fence();
5358
5359  // If externally suspended while waiting, re-suspend
5360  if (jt->handle_special_suspend_equivalent_condition()) {
5361    jt->java_suspend_self();
5362  }
5363}
5364
5365void Parker::unpark() {
5366  int status = os::Solaris::mutex_lock(_mutex);
5367  assert(status == 0, "invariant");
5368  const int s = _counter;
5369  _counter = 1;
5370  status = os::Solaris::mutex_unlock(_mutex);
5371  assert(status == 0, "invariant");
5372
5373  if (s < 1) {
5374    status = os::Solaris::cond_signal(_cond);
5375    assert(status == 0, "invariant");
5376  }
5377}
5378
5379extern char** environ;
5380
5381// Run the specified command in a separate process. Return its exit value,
5382// or -1 on failure (e.g. can't fork a new process).
5383// Unlike system(), this function can be called from signal handler. It
5384// doesn't block SIGINT et al.
5385int os::fork_and_exec(char* cmd) {
5386  char * argv[4];
5387  argv[0] = (char *)"sh";
5388  argv[1] = (char *)"-c";
5389  argv[2] = cmd;
5390  argv[3] = NULL;
5391
5392  // fork is async-safe, fork1 is not so can't use in signal handler
5393  pid_t pid;
5394  Thread* t = Thread::current_or_null_safe();
5395  if (t != NULL && t->is_inside_signal_handler()) {
5396    pid = fork();
5397  } else {
5398    pid = fork1();
5399  }
5400
5401  if (pid < 0) {
5402    // fork failed
5403    warning("fork failed: %s", os::strerror(errno));
5404    return -1;
5405
5406  } else if (pid == 0) {
5407    // child process
5408
5409    // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5410    execve("/usr/bin/sh", argv, environ);
5411
5412    // execve failed
5413    _exit(-1);
5414
5415  } else  {
5416    // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5417    // care about the actual exit code, for now.
5418
5419    int status;
5420
5421    // Wait for the child process to exit.  This returns immediately if
5422    // the child has already exited. */
5423    while (waitpid(pid, &status, 0) < 0) {
5424      switch (errno) {
5425      case ECHILD: return 0;
5426      case EINTR: break;
5427      default: return -1;
5428      }
5429    }
5430
5431    if (WIFEXITED(status)) {
5432      // The child exited normally; get its exit code.
5433      return WEXITSTATUS(status);
5434    } else if (WIFSIGNALED(status)) {
5435      // The child exited because of a signal
5436      // The best value to return is 0x80 + signal number,
5437      // because that is what all Unix shells do, and because
5438      // it allows callers to distinguish between process exit and
5439      // process death by signal.
5440      return 0x80 + WTERMSIG(status);
5441    } else {
5442      // Unknown exit code; pass it through
5443      return status;
5444    }
5445  }
5446}
5447
5448// is_headless_jre()
5449//
5450// Test for the existence of xawt/libmawt.so or libawt_xawt.so
5451// in order to report if we are running in a headless jre
5452//
5453// Since JDK8 xawt/libmawt.so was moved into the same directory
5454// as libawt.so, and renamed libawt_xawt.so
5455//
5456bool os::is_headless_jre() {
5457  struct stat statbuf;
5458  char buf[MAXPATHLEN];
5459  char libmawtpath[MAXPATHLEN];
5460  const char *xawtstr  = "/xawt/libmawt.so";
5461  const char *new_xawtstr = "/libawt_xawt.so";
5462  char *p;
5463
5464  // Get path to libjvm.so
5465  os::jvm_path(buf, sizeof(buf));
5466
5467  // Get rid of libjvm.so
5468  p = strrchr(buf, '/');
5469  if (p == NULL) {
5470    return false;
5471  } else {
5472    *p = '\0';
5473  }
5474
5475  // Get rid of client or server
5476  p = strrchr(buf, '/');
5477  if (p == NULL) {
5478    return false;
5479  } else {
5480    *p = '\0';
5481  }
5482
5483  // check xawt/libmawt.so
5484  strcpy(libmawtpath, buf);
5485  strcat(libmawtpath, xawtstr);
5486  if (::stat(libmawtpath, &statbuf) == 0) return false;
5487
5488  // check libawt_xawt.so
5489  strcpy(libmawtpath, buf);
5490  strcat(libmawtpath, new_xawtstr);
5491  if (::stat(libmawtpath, &statbuf) == 0) return false;
5492
5493  return true;
5494}
5495
5496size_t os::write(int fd, const void *buf, unsigned int nBytes) {
5497  size_t res;
5498  RESTARTABLE((size_t) ::write(fd, buf, (size_t) nBytes), res);
5499  return res;
5500}
5501
5502int os::close(int fd) {
5503  return ::close(fd);
5504}
5505
5506int os::socket_close(int fd) {
5507  return ::close(fd);
5508}
5509
5510int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
5511  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5512         "Assumed _thread_in_native");
5513  RESTARTABLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags));
5514}
5515
5516int os::send(int fd, char* buf, size_t nBytes, uint flags) {
5517  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5518         "Assumed _thread_in_native");
5519  RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5520}
5521
5522int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
5523  RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5524}
5525
5526// As both poll and select can be interrupted by signals, we have to be
5527// prepared to restart the system call after updating the timeout, unless
5528// a poll() is done with timeout == -1, in which case we repeat with this
5529// "wait forever" value.
5530
5531int os::connect(int fd, struct sockaddr *him, socklen_t len) {
5532  int _result;
5533  _result = ::connect(fd, him, len);
5534
5535  // On Solaris, when a connect() call is interrupted, the connection
5536  // can be established asynchronously (see 6343810). Subsequent calls
5537  // to connect() must check the errno value which has the semantic
5538  // described below (copied from the connect() man page). Handling
5539  // of asynchronously established connections is required for both
5540  // blocking and non-blocking sockets.
5541  //     EINTR            The  connection  attempt  was   interrupted
5542  //                      before  any data arrived by the delivery of
5543  //                      a signal. The connection, however, will  be
5544  //                      established asynchronously.
5545  //
5546  //     EINPROGRESS      The socket is non-blocking, and the connec-
5547  //                      tion  cannot  be completed immediately.
5548  //
5549  //     EALREADY         The socket is non-blocking,  and a previous
5550  //                      connection  attempt  has  not yet been com-
5551  //                      pleted.
5552  //
5553  //     EISCONN          The socket is already connected.
5554  if (_result == OS_ERR && errno == EINTR) {
5555    // restarting a connect() changes its errno semantics
5556    RESTARTABLE(::connect(fd, him, len), _result);
5557    // undo these changes
5558    if (_result == OS_ERR) {
5559      if (errno == EALREADY) {
5560        errno = EINPROGRESS; // fall through
5561      } else if (errno == EISCONN) {
5562        errno = 0;
5563        return OS_OK;
5564      }
5565    }
5566  }
5567  return _result;
5568}
5569
5570// Get the default path to the core file
5571// Returns the length of the string
5572int os::get_core_path(char* buffer, size_t bufferSize) {
5573  const char* p = get_current_directory(buffer, bufferSize);
5574
5575  if (p == NULL) {
5576    assert(p != NULL, "failed to get current directory");
5577    return 0;
5578  }
5579
5580  jio_snprintf(buffer, bufferSize, "%s/core or core.%d",
5581                                              p, current_process_id());
5582
5583  return strlen(buffer);
5584}
5585
5586#ifndef PRODUCT
5587void TestReserveMemorySpecial_test() {
5588  // No tests available for this platform
5589}
5590#endif
5591
5592bool os::start_debugging(char *buf, int buflen) {
5593  int len = (int)strlen(buf);
5594  char *p = &buf[len];
5595
5596  jio_snprintf(p, buflen-len,
5597               "\n\n"
5598               "Do you want to debug the problem?\n\n"
5599               "To debug, run 'dbx - %d'; then switch to thread " INTX_FORMAT "\n"
5600               "Enter 'yes' to launch dbx automatically (PATH must include dbx)\n"
5601               "Otherwise, press RETURN to abort...",
5602               os::current_process_id(), os::current_thread_id());
5603
5604  bool yes = os::message_box("Unexpected Error", buf);
5605
5606  if (yes) {
5607    // yes, user asked VM to launch debugger
5608    jio_snprintf(buf, sizeof(buf), "dbx - %d", os::current_process_id());
5609
5610    os::fork_and_exec(buf);
5611    yes = false;
5612  }
5613  return yes;
5614}
5615