os_solaris.cpp revision 9737:e286c9ccd58d
1/*
2 * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25// no precompiled headers
26#include "classfile/classLoader.hpp"
27#include "classfile/systemDictionary.hpp"
28#include "classfile/vmSymbols.hpp"
29#include "code/icBuffer.hpp"
30#include "code/vtableStubs.hpp"
31#include "compiler/compileBroker.hpp"
32#include "compiler/disassembler.hpp"
33#include "interpreter/interpreter.hpp"
34#include "jvm_solaris.h"
35#include "memory/allocation.inline.hpp"
36#include "memory/filemap.hpp"
37#include "mutex_solaris.inline.hpp"
38#include "oops/oop.inline.hpp"
39#include "os_share_solaris.hpp"
40#include "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.inline.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/decoder.hpp"
68#include "utilities/defaultStream.hpp"
69#include "utilities/events.hpp"
70#include "utilities/growableArray.hpp"
71#include "utilities/vmError.hpp"
72
73// put OS-includes here
74# include <dlfcn.h>
75# include <errno.h>
76# include <exception>
77# include <link.h>
78# include <poll.h>
79# include <pthread.h>
80# include <pwd.h>
81# include <schedctl.h>
82# include <setjmp.h>
83# include <signal.h>
84# include <stdio.h>
85# include <alloca.h>
86# include <sys/filio.h>
87# include <sys/ipc.h>
88# include <sys/lwp.h>
89# include <sys/machelf.h>     // for elf Sym structure used by dladdr1
90# include <sys/mman.h>
91# include <sys/processor.h>
92# include <sys/procset.h>
93# include <sys/pset.h>
94# include <sys/resource.h>
95# include <sys/shm.h>
96# include <sys/socket.h>
97# include <sys/stat.h>
98# include <sys/systeminfo.h>
99# include <sys/time.h>
100# include <sys/times.h>
101# include <sys/types.h>
102# include <sys/wait.h>
103# include <sys/utsname.h>
104# include <thread.h>
105# include <unistd.h>
106# include <sys/priocntl.h>
107# include <sys/rtpriocntl.h>
108# include <sys/tspriocntl.h>
109# include <sys/iapriocntl.h>
110# include <sys/fxpriocntl.h>
111# include <sys/loadavg.h>
112# include <string.h>
113# include <stdio.h>
114
115# define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
116# include <sys/procfs.h>     //  see comment in <sys/procfs.h>
117
118#define MAX_PATH (2 * K)
119
120// for timer info max values which include all bits
121#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
122
123
124// Here are some liblgrp types from sys/lgrp_user.h to be able to
125// compile on older systems without this header file.
126
127#ifndef MADV_ACCESS_LWP
128  #define  MADV_ACCESS_LWP   7       /* next LWP to access heavily */
129#endif
130#ifndef MADV_ACCESS_MANY
131  #define  MADV_ACCESS_MANY  8       /* many processes to access heavily */
132#endif
133
134#ifndef LGRP_RSRC_CPU
135  #define LGRP_RSRC_CPU      0       /* CPU resources */
136#endif
137#ifndef LGRP_RSRC_MEM
138  #define LGRP_RSRC_MEM      1       /* memory resources */
139#endif
140
141// Values for ThreadPriorityPolicy == 1
142int prio_policy1[CriticalPriority+1] = {
143  -99999,  0, 16,  32,  48,  64,
144          80, 96, 112, 124, 127, 127 };
145
146// System parameters used internally
147static clock_t clock_tics_per_sec = 100;
148
149// Track if we have called enable_extended_FILE_stdio (on Solaris 10u4+)
150static bool enabled_extended_FILE_stdio = false;
151
152// For diagnostics to print a message once. see run_periodic_checks
153static bool check_addr0_done = false;
154static sigset_t check_signal_done;
155static bool check_signals = true;
156
157address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
158address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
159
160address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
161
162
163// "default" initializers for missing libc APIs
164extern "C" {
165  static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
166  static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
167
168  static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
169  static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
170}
171
172// "default" initializers for pthread-based synchronization
173extern "C" {
174  static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
175  static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
176}
177
178static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
179
180static inline size_t adjust_stack_size(address base, size_t size) {
181  if ((ssize_t)size < 0) {
182    // 4759953: Compensate for ridiculous stack size.
183    size = max_intx;
184  }
185  if (size > (size_t)base) {
186    // 4812466: Make sure size doesn't allow the stack to wrap the address space.
187    size = (size_t)base;
188  }
189  return size;
190}
191
192static inline stack_t get_stack_info() {
193  stack_t st;
194  int retval = thr_stksegment(&st);
195  st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
196  assert(retval == 0, "incorrect return value from thr_stksegment");
197  assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
198  assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
199  return st;
200}
201
202address os::current_stack_base() {
203  int r = thr_main();
204  guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
205  bool is_primordial_thread = r;
206
207  // Workaround 4352906, avoid calls to thr_stksegment by
208  // thr_main after the first one (it looks like we trash
209  // some data, causing the value for ss_sp to be incorrect).
210  if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
211    stack_t st = get_stack_info();
212    if (is_primordial_thread) {
213      // cache initial value of stack base
214      os::Solaris::_main_stack_base = (address)st.ss_sp;
215    }
216    return (address)st.ss_sp;
217  } else {
218    guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
219    return os::Solaris::_main_stack_base;
220  }
221}
222
223size_t os::current_stack_size() {
224  size_t size;
225
226  int r = thr_main();
227  guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
228  if (!r) {
229    size = get_stack_info().ss_size;
230  } else {
231    struct rlimit limits;
232    getrlimit(RLIMIT_STACK, &limits);
233    size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
234  }
235  // base may not be page aligned
236  address base = current_stack_base();
237  address bottom = (address)align_size_up((intptr_t)(base - size), os::vm_page_size());;
238  return (size_t)(base - bottom);
239}
240
241struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
242  return localtime_r(clock, res);
243}
244
245void os::Solaris::try_enable_extended_io() {
246  typedef int (*enable_extended_FILE_stdio_t)(int, int);
247
248  if (!UseExtendedFileIO) {
249    return;
250  }
251
252  enable_extended_FILE_stdio_t enabler =
253    (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
254                                         "enable_extended_FILE_stdio");
255  if (enabler) {
256    enabler(-1, -1);
257  }
258}
259
260static int _processors_online = 0;
261
262jint os::Solaris::_os_thread_limit = 0;
263volatile jint os::Solaris::_os_thread_count = 0;
264
265julong os::available_memory() {
266  return Solaris::available_memory();
267}
268
269julong os::Solaris::available_memory() {
270  return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
271}
272
273julong os::Solaris::_physical_memory = 0;
274
275julong os::physical_memory() {
276  return Solaris::physical_memory();
277}
278
279static hrtime_t first_hrtime = 0;
280static const hrtime_t hrtime_hz = 1000*1000*1000;
281static volatile hrtime_t max_hrtime = 0;
282
283
284void os::Solaris::initialize_system_info() {
285  set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
286  _processors_online = sysconf(_SC_NPROCESSORS_ONLN);
287  _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) *
288                                     (julong)sysconf(_SC_PAGESIZE);
289}
290
291int os::active_processor_count() {
292  int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
293  pid_t pid = getpid();
294  psetid_t pset = PS_NONE;
295  // Are we running in a processor set or is there any processor set around?
296  if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
297    uint_t pset_cpus;
298    // Query the number of cpus available to us.
299    if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
300      assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
301      _processors_online = pset_cpus;
302      return pset_cpus;
303    }
304  }
305  // Otherwise return number of online cpus
306  return online_cpus;
307}
308
309static bool find_processors_in_pset(psetid_t        pset,
310                                    processorid_t** id_array,
311                                    uint_t*         id_length) {
312  bool result = false;
313  // Find the number of processors in the processor set.
314  if (pset_info(pset, NULL, id_length, NULL) == 0) {
315    // Make up an array to hold their ids.
316    *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
317    // Fill in the array with their processor ids.
318    if (pset_info(pset, NULL, id_length, *id_array) == 0) {
319      result = true;
320    }
321  }
322  return result;
323}
324
325// Callers of find_processors_online() must tolerate imprecise results --
326// the system configuration can change asynchronously because of DR
327// or explicit psradm operations.
328//
329// We also need to take care that the loop (below) terminates as the
330// number of processors online can change between the _SC_NPROCESSORS_ONLN
331// request and the loop that builds the list of processor ids.   Unfortunately
332// there's no reliable way to determine the maximum valid processor id,
333// so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
334// man pages, which claim the processor id set is "sparse, but
335// not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
336// exit the loop.
337//
338// In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
339// not available on S8.0.
340
341static bool find_processors_online(processorid_t** id_array,
342                                   uint*           id_length) {
343  const processorid_t MAX_PROCESSOR_ID = 100000;
344  // Find the number of processors online.
345  *id_length = sysconf(_SC_NPROCESSORS_ONLN);
346  // Make up an array to hold their ids.
347  *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
348  // Processors need not be numbered consecutively.
349  long found = 0;
350  processorid_t next = 0;
351  while (found < *id_length && next < MAX_PROCESSOR_ID) {
352    processor_info_t info;
353    if (processor_info(next, &info) == 0) {
354      // NB, PI_NOINTR processors are effectively online ...
355      if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
356        (*id_array)[found] = next;
357        found += 1;
358      }
359    }
360    next += 1;
361  }
362  if (found < *id_length) {
363    // The loop above didn't identify the expected number of processors.
364    // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
365    // and re-running the loop, above, but there's no guarantee of progress
366    // if the system configuration is in flux.  Instead, we just return what
367    // we've got.  Note that in the worst case find_processors_online() could
368    // return an empty set.  (As a fall-back in the case of the empty set we
369    // could just return the ID of the current processor).
370    *id_length = found;
371  }
372
373  return true;
374}
375
376static bool assign_distribution(processorid_t* id_array,
377                                uint           id_length,
378                                uint*          distribution,
379                                uint           distribution_length) {
380  // We assume we can assign processorid_t's to uint's.
381  assert(sizeof(processorid_t) == sizeof(uint),
382         "can't convert processorid_t to uint");
383  // Quick check to see if we won't succeed.
384  if (id_length < distribution_length) {
385    return false;
386  }
387  // Assign processor ids to the distribution.
388  // Try to shuffle processors to distribute work across boards,
389  // assuming 4 processors per board.
390  const uint processors_per_board = ProcessDistributionStride;
391  // Find the maximum processor id.
392  processorid_t max_id = 0;
393  for (uint m = 0; m < id_length; m += 1) {
394    max_id = MAX2(max_id, id_array[m]);
395  }
396  // The next id, to limit loops.
397  const processorid_t limit_id = max_id + 1;
398  // Make up markers for available processors.
399  bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id, mtInternal);
400  for (uint c = 0; c < limit_id; c += 1) {
401    available_id[c] = false;
402  }
403  for (uint a = 0; a < id_length; a += 1) {
404    available_id[id_array[a]] = true;
405  }
406  // Step by "boards", then by "slot", copying to "assigned".
407  // NEEDS_CLEANUP: The assignment of processors should be stateful,
408  //                remembering which processors have been assigned by
409  //                previous calls, etc., so as to distribute several
410  //                independent calls of this method.  What we'd like is
411  //                It would be nice to have an API that let us ask
412  //                how many processes are bound to a processor,
413  //                but we don't have that, either.
414  //                In the short term, "board" is static so that
415  //                subsequent distributions don't all start at board 0.
416  static uint board = 0;
417  uint assigned = 0;
418  // Until we've found enough processors ....
419  while (assigned < distribution_length) {
420    // ... find the next available processor in the board.
421    for (uint slot = 0; slot < processors_per_board; slot += 1) {
422      uint try_id = board * processors_per_board + slot;
423      if ((try_id < limit_id) && (available_id[try_id] == true)) {
424        distribution[assigned] = try_id;
425        available_id[try_id] = false;
426        assigned += 1;
427        break;
428      }
429    }
430    board += 1;
431    if (board * processors_per_board + 0 >= limit_id) {
432      board = 0;
433    }
434  }
435  if (available_id != NULL) {
436    FREE_C_HEAP_ARRAY(bool, available_id);
437  }
438  return true;
439}
440
441void os::set_native_thread_name(const char *name) {
442  // Not yet implemented.
443  return;
444}
445
446bool os::distribute_processes(uint length, uint* distribution) {
447  bool result = false;
448  // Find the processor id's of all the available CPUs.
449  processorid_t* id_array  = NULL;
450  uint           id_length = 0;
451  // There are some races between querying information and using it,
452  // since processor sets can change dynamically.
453  psetid_t pset = PS_NONE;
454  // Are we running in a processor set?
455  if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
456    result = find_processors_in_pset(pset, &id_array, &id_length);
457  } else {
458    result = find_processors_online(&id_array, &id_length);
459  }
460  if (result == true) {
461    if (id_length >= length) {
462      result = assign_distribution(id_array, id_length, distribution, length);
463    } else {
464      result = false;
465    }
466  }
467  if (id_array != NULL) {
468    FREE_C_HEAP_ARRAY(processorid_t, id_array);
469  }
470  return result;
471}
472
473bool os::bind_to_processor(uint processor_id) {
474  // We assume that a processorid_t can be stored in a uint.
475  assert(sizeof(uint) == sizeof(processorid_t),
476         "can't convert uint to processorid_t");
477  int bind_result =
478    processor_bind(P_LWPID,                       // bind LWP.
479                   P_MYID,                        // bind current LWP.
480                   (processorid_t) processor_id,  // id.
481                   NULL);                         // don't return old binding.
482  return (bind_result == 0);
483}
484
485// Return true if user is running as root.
486
487bool os::have_special_privileges() {
488  static bool init = false;
489  static bool privileges = false;
490  if (!init) {
491    privileges = (getuid() != geteuid()) || (getgid() != getegid());
492    init = true;
493  }
494  return privileges;
495}
496
497
498void os::init_system_properties_values() {
499  // The next steps are taken in the product version:
500  //
501  // Obtain the JAVA_HOME value from the location of libjvm.so.
502  // This library should be located at:
503  // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
504  //
505  // If "/jre/lib/" appears at the right place in the path, then we
506  // assume libjvm.so is installed in a JDK and we use this path.
507  //
508  // Otherwise exit with message: "Could not create the Java virtual machine."
509  //
510  // The following extra steps are taken in the debugging version:
511  //
512  // If "/jre/lib/" does NOT appear at the right place in the path
513  // instead of exit check for $JAVA_HOME environment variable.
514  //
515  // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
516  // then we append a fake suffix "hotspot/libjvm.so" to this path so
517  // it looks like libjvm.so is installed there
518  // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
519  //
520  // Otherwise exit.
521  //
522  // Important note: if the location of libjvm.so changes this
523  // code needs to be changed accordingly.
524
525// Base path of extensions installed on the system.
526#define SYS_EXT_DIR     "/usr/jdk/packages"
527#define EXTENSIONS_DIR  "/lib/ext"
528
529  char cpu_arch[12];
530  // Buffer that fits several sprintfs.
531  // Note that the space for the colon and the trailing null are provided
532  // by the nulls included by the sizeof operator.
533  const size_t bufsize =
534    MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
535         sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch), // invariant ld_library_path
536         (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
537  char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
538
539  // sysclasspath, java_home, dll_dir
540  {
541    char *pslash;
542    os::jvm_path(buf, bufsize);
543
544    // Found the full path to libjvm.so.
545    // Now cut the path to <java_home>/jre if we can.
546    *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
547    pslash = strrchr(buf, '/');
548    if (pslash != NULL) {
549      *pslash = '\0';            // Get rid of /{client|server|hotspot}.
550    }
551    Arguments::set_dll_dir(buf);
552
553    if (pslash != NULL) {
554      pslash = strrchr(buf, '/');
555      if (pslash != NULL) {
556        *pslash = '\0';          // Get rid of /<arch>.
557        pslash = strrchr(buf, '/');
558        if (pslash != NULL) {
559          *pslash = '\0';        // Get rid of /lib.
560        }
561      }
562    }
563    Arguments::set_java_home(buf);
564    set_boot_path('/', ':');
565  }
566
567  // Where to look for native libraries.
568  {
569    // Use dlinfo() to determine the correct java.library.path.
570    //
571    // If we're launched by the Java launcher, and the user
572    // does not set java.library.path explicitly on the commandline,
573    // the Java launcher sets LD_LIBRARY_PATH for us and unsets
574    // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
575    // dlinfo returns LD_LIBRARY_PATH + crle settings (including
576    // /usr/lib), which is exactly what we want.
577    //
578    // If the user does set java.library.path, it completely
579    // overwrites this setting, and always has.
580    //
581    // If we're not launched by the Java launcher, we may
582    // get here with any/all of the LD_LIBRARY_PATH[_32|64]
583    // settings.  Again, dlinfo does exactly what we want.
584
585    Dl_serinfo     info_sz, *info = &info_sz;
586    Dl_serpath     *path;
587    char           *library_path;
588    char           *common_path = buf;
589
590    // Determine search path count and required buffer size.
591    if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
592      FREE_C_HEAP_ARRAY(char, buf);
593      vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
594    }
595
596    // Allocate new buffer and initialize.
597    info = (Dl_serinfo*)NEW_C_HEAP_ARRAY(char, info_sz.dls_size, mtInternal);
598    info->dls_size = info_sz.dls_size;
599    info->dls_cnt = info_sz.dls_cnt;
600
601    // Obtain search path information.
602    if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
603      FREE_C_HEAP_ARRAY(char, buf);
604      FREE_C_HEAP_ARRAY(char, info);
605      vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
606    }
607
608    path = &info->dls_serpath[0];
609
610    // Note: Due to a legacy implementation, most of the library path
611    // is set in the launcher. This was to accomodate linking restrictions
612    // on legacy Solaris implementations (which are no longer supported).
613    // Eventually, all the library path setting will be done here.
614    //
615    // However, to prevent the proliferation of improperly built native
616    // libraries, the new path component /usr/jdk/packages is added here.
617
618    // Determine the actual CPU architecture.
619    sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
620#ifdef _LP64
621    // If we are a 64-bit vm, perform the following translations:
622    //   sparc   -> sparcv9
623    //   i386    -> amd64
624    if (strcmp(cpu_arch, "sparc") == 0) {
625      strcat(cpu_arch, "v9");
626    } else if (strcmp(cpu_arch, "i386") == 0) {
627      strcpy(cpu_arch, "amd64");
628    }
629#endif
630
631    // Construct the invariant part of ld_library_path.
632    sprintf(common_path, SYS_EXT_DIR "/lib/%s", cpu_arch);
633
634    // Struct size is more than sufficient for the path components obtained
635    // through the dlinfo() call, so only add additional space for the path
636    // components explicitly added here.
637    size_t library_path_size = info->dls_size + strlen(common_path);
638    library_path = (char *)NEW_C_HEAP_ARRAY(char, library_path_size, mtInternal);
639    library_path[0] = '\0';
640
641    // Construct the desired Java library path from the linker's library
642    // search path.
643    //
644    // For compatibility, it is optimal that we insert the additional path
645    // components specific to the Java VM after those components specified
646    // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
647    // infrastructure.
648    if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it.
649      strcpy(library_path, common_path);
650    } else {
651      int inserted = 0;
652      int i;
653      for (i = 0; i < info->dls_cnt; i++, path++) {
654        uint_t flags = path->dls_flags & LA_SER_MASK;
655        if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
656          strcat(library_path, common_path);
657          strcat(library_path, os::path_separator());
658          inserted = 1;
659        }
660        strcat(library_path, path->dls_name);
661        strcat(library_path, os::path_separator());
662      }
663      // Eliminate trailing path separator.
664      library_path[strlen(library_path)-1] = '\0';
665    }
666
667    // happens before argument parsing - can't use a trace flag
668    // tty->print_raw("init_system_properties_values: native lib path: ");
669    // tty->print_raw_cr(library_path);
670
671    // Callee copies into its own buffer.
672    Arguments::set_library_path(library_path);
673
674    FREE_C_HEAP_ARRAY(char, library_path);
675    FREE_C_HEAP_ARRAY(char, info);
676  }
677
678  // Extensions directories.
679  sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
680  Arguments::set_ext_dirs(buf);
681
682  FREE_C_HEAP_ARRAY(char, buf);
683
684#undef SYS_EXT_DIR
685#undef EXTENSIONS_DIR
686}
687
688void os::breakpoint() {
689  BREAKPOINT;
690}
691
692bool os::obsolete_option(const JavaVMOption *option) {
693  if (!strncmp(option->optionString, "-Xt", 3)) {
694    return true;
695  } else if (!strncmp(option->optionString, "-Xtm", 4)) {
696    return true;
697  } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
698    return true;
699  } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
700    return true;
701  }
702  return false;
703}
704
705bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
706  address  stackStart  = (address)thread->stack_base();
707  address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
708  if (sp < stackStart && sp >= stackEnd) return true;
709  return false;
710}
711
712extern "C" void breakpoint() {
713  // use debugger to set breakpoint here
714}
715
716static thread_t main_thread;
717
718// Thread start routine for all new Java threads
719extern "C" void* java_start(void* thread_addr) {
720  // Try to randomize the cache line index of hot stack frames.
721  // This helps when threads of the same stack traces evict each other's
722  // cache lines. The threads can be either from the same JVM instance, or
723  // from different JVM instances. The benefit is especially true for
724  // processors with hyperthreading technology.
725  static int counter = 0;
726  int pid = os::current_process_id();
727  alloca(((pid ^ counter++) & 7) * 128);
728
729  int prio;
730  Thread* thread = (Thread*)thread_addr;
731
732  thread->initialize_thread_current();
733
734  OSThread* osthr = thread->osthread();
735
736  osthr->set_lwp_id(_lwp_self());  // Store lwp in case we are bound
737  thread->_schedctl = (void *) schedctl_init();
738
739  if (UseNUMA) {
740    int lgrp_id = os::numa_get_group_id();
741    if (lgrp_id != -1) {
742      thread->set_lgrp_id(lgrp_id);
743    }
744  }
745
746  // If the creator called set priority before we started,
747  // we need to call set_native_priority now that we have an lwp.
748  // We used to get the priority from thr_getprio (we called
749  // thr_setprio way back in create_thread) and pass it to
750  // set_native_priority, but Solaris scales the priority
751  // in java_to_os_priority, so when we read it back here,
752  // we pass trash to set_native_priority instead of what's
753  // in java_to_os_priority. So we save the native priority
754  // in the osThread and recall it here.
755
756  if (osthr->thread_id() != -1) {
757    if (UseThreadPriorities) {
758      int prio = osthr->native_priority();
759      if (ThreadPriorityVerbose) {
760        tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
761                      INTPTR_FORMAT ", setting priority: %d\n",
762                      osthr->thread_id(), osthr->lwp_id(), prio);
763      }
764      os::set_native_priority(thread, prio);
765    }
766  } else if (ThreadPriorityVerbose) {
767    warning("Can't set priority in _start routine, thread id hasn't been set\n");
768  }
769
770  assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
771
772  // initialize signal mask for this thread
773  os::Solaris::hotspot_sigmask(thread);
774
775  thread->run();
776
777  // One less thread is executing
778  // When the VMThread gets here, the main thread may have already exited
779  // which frees the CodeHeap containing the Atomic::dec code
780  if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
781    Atomic::dec(&os::Solaris::_os_thread_count);
782  }
783
784  if (UseDetachedThreads) {
785    thr_exit(NULL);
786    ShouldNotReachHere();
787  }
788  return NULL;
789}
790
791static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
792  // Allocate the OSThread object
793  OSThread* osthread = new OSThread(NULL, NULL);
794  if (osthread == NULL) return NULL;
795
796  // Store info on the Solaris thread into the OSThread
797  osthread->set_thread_id(thread_id);
798  osthread->set_lwp_id(_lwp_self());
799  thread->_schedctl = (void *) schedctl_init();
800
801  if (UseNUMA) {
802    int lgrp_id = os::numa_get_group_id();
803    if (lgrp_id != -1) {
804      thread->set_lgrp_id(lgrp_id);
805    }
806  }
807
808  if (ThreadPriorityVerbose) {
809    tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
810                  osthread->thread_id(), osthread->lwp_id());
811  }
812
813  // Initial thread state is INITIALIZED, not SUSPENDED
814  osthread->set_state(INITIALIZED);
815
816  return osthread;
817}
818
819void os::Solaris::hotspot_sigmask(Thread* thread) {
820  //Save caller's signal mask
821  sigset_t sigmask;
822  thr_sigsetmask(SIG_SETMASK, NULL, &sigmask);
823  OSThread *osthread = thread->osthread();
824  osthread->set_caller_sigmask(sigmask);
825
826  thr_sigsetmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
827  if (!ReduceSignalUsage) {
828    if (thread->is_VM_thread()) {
829      // Only the VM thread handles BREAK_SIGNAL ...
830      thr_sigsetmask(SIG_UNBLOCK, vm_signals(), NULL);
831    } else {
832      // ... all other threads block BREAK_SIGNAL
833      assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
834      thr_sigsetmask(SIG_BLOCK, vm_signals(), NULL);
835    }
836  }
837}
838
839bool os::create_attached_thread(JavaThread* thread) {
840#ifdef ASSERT
841  thread->verify_not_published();
842#endif
843  OSThread* osthread = create_os_thread(thread, thr_self());
844  if (osthread == NULL) {
845    return false;
846  }
847
848  // Initial thread state is RUNNABLE
849  osthread->set_state(RUNNABLE);
850  thread->set_osthread(osthread);
851
852  // initialize signal mask for this thread
853  // and save the caller's signal mask
854  os::Solaris::hotspot_sigmask(thread);
855
856  return true;
857}
858
859bool os::create_main_thread(JavaThread* thread) {
860#ifdef ASSERT
861  thread->verify_not_published();
862#endif
863  if (_starting_thread == NULL) {
864    _starting_thread = create_os_thread(thread, main_thread);
865    if (_starting_thread == NULL) {
866      return false;
867    }
868  }
869
870  // The primodial thread is runnable from the start
871  _starting_thread->set_state(RUNNABLE);
872
873  thread->set_osthread(_starting_thread);
874
875  // initialize signal mask for this thread
876  // and save the caller's signal mask
877  os::Solaris::hotspot_sigmask(thread);
878
879  return true;
880}
881
882
883bool os::create_thread(Thread* thread, ThreadType thr_type,
884                       size_t stack_size) {
885  // Allocate the OSThread object
886  OSThread* osthread = new OSThread(NULL, NULL);
887  if (osthread == NULL) {
888    return false;
889  }
890
891  if (ThreadPriorityVerbose) {
892    char *thrtyp;
893    switch (thr_type) {
894    case vm_thread:
895      thrtyp = (char *)"vm";
896      break;
897    case cgc_thread:
898      thrtyp = (char *)"cgc";
899      break;
900    case pgc_thread:
901      thrtyp = (char *)"pgc";
902      break;
903    case java_thread:
904      thrtyp = (char *)"java";
905      break;
906    case compiler_thread:
907      thrtyp = (char *)"compiler";
908      break;
909    case watcher_thread:
910      thrtyp = (char *)"watcher";
911      break;
912    default:
913      thrtyp = (char *)"unknown";
914      break;
915    }
916    tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
917  }
918
919  // Calculate stack size if it's not specified by caller.
920  if (stack_size == 0) {
921    // The default stack size 1M (2M for LP64).
922    stack_size = (BytesPerWord >> 2) * K * K;
923
924    switch (thr_type) {
925    case os::java_thread:
926      // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
927      if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
928      break;
929    case os::compiler_thread:
930      if (CompilerThreadStackSize > 0) {
931        stack_size = (size_t)(CompilerThreadStackSize * K);
932        break;
933      } // else fall through:
934        // use VMThreadStackSize if CompilerThreadStackSize is not defined
935    case os::vm_thread:
936    case os::pgc_thread:
937    case os::cgc_thread:
938    case os::watcher_thread:
939      if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
940      break;
941    }
942  }
943  stack_size = MAX2(stack_size, os::Solaris::min_stack_allowed);
944
945  // Initial state is ALLOCATED but not INITIALIZED
946  osthread->set_state(ALLOCATED);
947
948  if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
949    // We got lots of threads. Check if we still have some address space left.
950    // Need to be at least 5Mb of unreserved address space. We do check by
951    // trying to reserve some.
952    const size_t VirtualMemoryBangSize = 20*K*K;
953    char* mem = os::reserve_memory(VirtualMemoryBangSize);
954    if (mem == NULL) {
955      delete osthread;
956      return false;
957    } else {
958      // Release the memory again
959      os::release_memory(mem, VirtualMemoryBangSize);
960    }
961  }
962
963  // Setup osthread because the child thread may need it.
964  thread->set_osthread(osthread);
965
966  // Create the Solaris thread
967  thread_t tid = 0;
968  long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED;
969  int      status;
970
971  // Mark that we don't have an lwp or thread id yet.
972  // In case we attempt to set the priority before the thread starts.
973  osthread->set_lwp_id(-1);
974  osthread->set_thread_id(-1);
975
976  status = thr_create(NULL, stack_size, java_start, thread, flags, &tid);
977  if (status != 0) {
978    if (PrintMiscellaneous && (Verbose || WizardMode)) {
979      perror("os::create_thread");
980    }
981    thread->set_osthread(NULL);
982    // Need to clean up stuff we've allocated so far
983    delete osthread;
984    return false;
985  }
986
987  Atomic::inc(&os::Solaris::_os_thread_count);
988
989  // Store info on the Solaris thread into the OSThread
990  osthread->set_thread_id(tid);
991
992  // Remember that we created this thread so we can set priority on it
993  osthread->set_vm_created();
994
995  // Initial thread state is INITIALIZED, not SUSPENDED
996  osthread->set_state(INITIALIZED);
997
998  // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
999  return true;
1000}
1001
1002// defined for >= Solaris 10. This allows builds on earlier versions
1003// of Solaris to take advantage of the newly reserved Solaris JVM signals.
1004// With SIGJVM1, SIGJVM2, ASYNC_SIGNAL is SIGJVM2. Previously INTERRUPT_SIGNAL
1005// was SIGJVM1.
1006//
1007#if !defined(SIGJVM1)
1008  #define SIGJVM1 39
1009  #define SIGJVM2 40
1010#endif
1011
1012debug_only(static bool signal_sets_initialized = false);
1013static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
1014
1015int os::Solaris::_SIGasync = ASYNC_SIGNAL;
1016
1017bool os::Solaris::is_sig_ignored(int sig) {
1018  struct sigaction oact;
1019  sigaction(sig, (struct sigaction*)NULL, &oact);
1020  void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1021                                 : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1022  if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
1023    return true;
1024  } else {
1025    return false;
1026  }
1027}
1028
1029// Note: SIGRTMIN is a macro that calls sysconf() so it will
1030// dynamically detect SIGRTMIN value for the system at runtime, not buildtime
1031static bool isJVM1available() {
1032  return SIGJVM1 < SIGRTMIN;
1033}
1034
1035void os::Solaris::signal_sets_init() {
1036  // Should also have an assertion stating we are still single-threaded.
1037  assert(!signal_sets_initialized, "Already initialized");
1038  // Fill in signals that are necessarily unblocked for all threads in
1039  // the VM. Currently, we unblock the following signals:
1040  // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1041  //                         by -Xrs (=ReduceSignalUsage));
1042  // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1043  // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1044  // the dispositions or masks wrt these signals.
1045  // Programs embedding the VM that want to use the above signals for their
1046  // own purposes must, at this time, use the "-Xrs" option to prevent
1047  // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1048  // (See bug 4345157, and other related bugs).
1049  // In reality, though, unblocking these signals is really a nop, since
1050  // these signals are not blocked by default.
1051  sigemptyset(&unblocked_sigs);
1052  sigemptyset(&allowdebug_blocked_sigs);
1053  sigaddset(&unblocked_sigs, SIGILL);
1054  sigaddset(&unblocked_sigs, SIGSEGV);
1055  sigaddset(&unblocked_sigs, SIGBUS);
1056  sigaddset(&unblocked_sigs, SIGFPE);
1057
1058  // Always true on Solaris 10+
1059  guarantee(isJVM1available(), "SIGJVM1/2 missing!");
1060  os::Solaris::set_SIGasync(SIGJVM2);
1061
1062  sigaddset(&unblocked_sigs, os::Solaris::SIGasync());
1063
1064  if (!ReduceSignalUsage) {
1065    if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1066      sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1067      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
1068    }
1069    if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1070      sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1071      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
1072    }
1073    if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1074      sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1075      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
1076    }
1077  }
1078  // Fill in signals that are blocked by all but the VM thread.
1079  sigemptyset(&vm_sigs);
1080  if (!ReduceSignalUsage) {
1081    sigaddset(&vm_sigs, BREAK_SIGNAL);
1082  }
1083  debug_only(signal_sets_initialized = true);
1084
1085  // For diagnostics only used in run_periodic_checks
1086  sigemptyset(&check_signal_done);
1087}
1088
1089// These are signals that are unblocked while a thread is running Java.
1090// (For some reason, they get blocked by default.)
1091sigset_t* os::Solaris::unblocked_signals() {
1092  assert(signal_sets_initialized, "Not initialized");
1093  return &unblocked_sigs;
1094}
1095
1096// These are the signals that are blocked while a (non-VM) thread is
1097// running Java. Only the VM thread handles these signals.
1098sigset_t* os::Solaris::vm_signals() {
1099  assert(signal_sets_initialized, "Not initialized");
1100  return &vm_sigs;
1101}
1102
1103// These are signals that are blocked during cond_wait to allow debugger in
1104sigset_t* os::Solaris::allowdebug_blocked_signals() {
1105  assert(signal_sets_initialized, "Not initialized");
1106  return &allowdebug_blocked_sigs;
1107}
1108
1109
1110void _handle_uncaught_cxx_exception() {
1111  VMError::report_and_die("An uncaught C++ exception");
1112}
1113
1114
1115// First crack at OS-specific initialization, from inside the new thread.
1116void os::initialize_thread(Thread* thr) {
1117  int r = thr_main();
1118  guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
1119  if (r) {
1120    JavaThread* jt = (JavaThread *)thr;
1121    assert(jt != NULL, "Sanity check");
1122    size_t stack_size;
1123    address base = jt->stack_base();
1124    if (Arguments::created_by_java_launcher()) {
1125      // Use 2MB to allow for Solaris 7 64 bit mode.
1126      stack_size = JavaThread::stack_size_at_create() == 0
1127        ? 2048*K : JavaThread::stack_size_at_create();
1128
1129      // There are rare cases when we may have already used more than
1130      // the basic stack size allotment before this method is invoked.
1131      // Attempt to allow for a normally sized java_stack.
1132      size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1133      stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1134    } else {
1135      // 6269555: If we were not created by a Java launcher, i.e. if we are
1136      // running embedded in a native application, treat the primordial thread
1137      // as much like a native attached thread as possible.  This means using
1138      // the current stack size from thr_stksegment(), unless it is too large
1139      // to reliably setup guard pages.  A reasonable max size is 8MB.
1140      size_t current_size = current_stack_size();
1141      // This should never happen, but just in case....
1142      if (current_size == 0) current_size = 2 * K * K;
1143      stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1144    }
1145    address bottom = (address)align_size_up((intptr_t)(base - stack_size), os::vm_page_size());;
1146    stack_size = (size_t)(base - bottom);
1147
1148    assert(stack_size > 0, "Stack size calculation problem");
1149
1150    if (stack_size > jt->stack_size()) {
1151#ifndef PRODUCT
1152      struct rlimit limits;
1153      getrlimit(RLIMIT_STACK, &limits);
1154      size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1155      assert(size >= jt->stack_size(), "Stack size problem in main thread");
1156#endif
1157      tty->print_cr("Stack size of %d Kb exceeds current limit of %d Kb.\n"
1158                    "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1159                    "See limit(1) to increase the stack size limit.",
1160                    stack_size / K, jt->stack_size() / K);
1161      vm_exit(1);
1162    }
1163    assert(jt->stack_size() >= stack_size,
1164           "Attempt to map more stack than was allocated");
1165    jt->set_stack_size(stack_size);
1166  }
1167
1168  // With the T2 libthread (T1 is no longer supported) threads are always bound
1169  // and we use stackbanging in all cases.
1170
1171  os::Solaris::init_thread_fpu_state();
1172  std::set_terminate(_handle_uncaught_cxx_exception);
1173}
1174
1175
1176
1177// Free Solaris resources related to the OSThread
1178void os::free_thread(OSThread* osthread) {
1179  assert(osthread != NULL, "os::free_thread but osthread not set");
1180
1181
1182  // We are told to free resources of the argument thread,
1183  // but we can only really operate on the current thread.
1184  // The main thread must take the VMThread down synchronously
1185  // before the main thread exits and frees up CodeHeap
1186  guarantee((Thread::current()->osthread() == osthread
1187             || (osthread == VMThread::vm_thread()->osthread())), "os::free_thread but not current thread");
1188  if (Thread::current()->osthread() == osthread) {
1189    // Restore caller's signal mask
1190    sigset_t sigmask = osthread->caller_sigmask();
1191    thr_sigsetmask(SIG_SETMASK, &sigmask, NULL);
1192  }
1193  delete osthread;
1194}
1195
1196void os::pd_start_thread(Thread* thread) {
1197  int status = thr_continue(thread->osthread()->thread_id());
1198  assert_status(status == 0, status, "thr_continue failed");
1199}
1200
1201
1202intx os::current_thread_id() {
1203  return (intx)thr_self();
1204}
1205
1206static pid_t _initial_pid = 0;
1207
1208int os::current_process_id() {
1209  return (int)(_initial_pid ? _initial_pid : getpid());
1210}
1211
1212// gethrtime() should be monotonic according to the documentation,
1213// but some virtualized platforms are known to break this guarantee.
1214// getTimeNanos() must be guaranteed not to move backwards, so we
1215// are forced to add a check here.
1216inline hrtime_t getTimeNanos() {
1217  const hrtime_t now = gethrtime();
1218  const hrtime_t prev = max_hrtime;
1219  if (now <= prev) {
1220    return prev;   // same or retrograde time;
1221  }
1222  const hrtime_t obsv = Atomic::cmpxchg(now, (volatile jlong*)&max_hrtime, prev);
1223  assert(obsv >= prev, "invariant");   // Monotonicity
1224  // If the CAS succeeded then we're done and return "now".
1225  // If the CAS failed and the observed value "obsv" is >= now then
1226  // we should return "obsv".  If the CAS failed and now > obsv > prv then
1227  // some other thread raced this thread and installed a new value, in which case
1228  // we could either (a) retry the entire operation, (b) retry trying to install now
1229  // or (c) just return obsv.  We use (c).   No loop is required although in some cases
1230  // we might discard a higher "now" value in deference to a slightly lower but freshly
1231  // installed obsv value.   That's entirely benign -- it admits no new orderings compared
1232  // to (a) or (b) -- and greatly reduces coherence traffic.
1233  // We might also condition (c) on the magnitude of the delta between obsv and now.
1234  // Avoiding excessive CAS operations to hot RW locations is critical.
1235  // See https://blogs.oracle.com/dave/entry/cas_and_cache_trivia_invalidate
1236  return (prev == obsv) ? now : obsv;
1237}
1238
1239// Time since start-up in seconds to a fine granularity.
1240// Used by VMSelfDestructTimer and the MemProfiler.
1241double os::elapsedTime() {
1242  return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1243}
1244
1245jlong os::elapsed_counter() {
1246  return (jlong)(getTimeNanos() - first_hrtime);
1247}
1248
1249jlong os::elapsed_frequency() {
1250  return hrtime_hz;
1251}
1252
1253// Return the real, user, and system times in seconds from an
1254// arbitrary fixed point in the past.
1255bool os::getTimesSecs(double* process_real_time,
1256                      double* process_user_time,
1257                      double* process_system_time) {
1258  struct tms ticks;
1259  clock_t real_ticks = times(&ticks);
1260
1261  if (real_ticks == (clock_t) (-1)) {
1262    return false;
1263  } else {
1264    double ticks_per_second = (double) clock_tics_per_sec;
1265    *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1266    *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1267    // For consistency return the real time from getTimeNanos()
1268    // converted to seconds.
1269    *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1270
1271    return true;
1272  }
1273}
1274
1275bool os::supports_vtime() { return true; }
1276
1277bool os::enable_vtime() {
1278  int fd = ::open("/proc/self/ctl", O_WRONLY);
1279  if (fd == -1) {
1280    return false;
1281  }
1282
1283  long cmd[] = { PCSET, PR_MSACCT };
1284  int res = ::write(fd, cmd, sizeof(long) * 2);
1285  ::close(fd);
1286  if (res != sizeof(long) * 2) {
1287    return false;
1288  }
1289  return true;
1290}
1291
1292bool os::vtime_enabled() {
1293  int fd = ::open("/proc/self/status", O_RDONLY);
1294  if (fd == -1) {
1295    return false;
1296  }
1297
1298  pstatus_t status;
1299  int res = os::read(fd, (void*) &status, sizeof(pstatus_t));
1300  ::close(fd);
1301  if (res != sizeof(pstatus_t)) {
1302    return false;
1303  }
1304  return status.pr_flags & PR_MSACCT;
1305}
1306
1307double os::elapsedVTime() {
1308  return (double)gethrvtime() / (double)hrtime_hz;
1309}
1310
1311// Used internally for comparisons only
1312// getTimeMillis guaranteed to not move backwards on Solaris
1313jlong getTimeMillis() {
1314  jlong nanotime = getTimeNanos();
1315  return (jlong)(nanotime / NANOSECS_PER_MILLISEC);
1316}
1317
1318// Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1319jlong os::javaTimeMillis() {
1320  timeval t;
1321  if (gettimeofday(&t, NULL) == -1) {
1322    fatal("os::javaTimeMillis: gettimeofday (%s)", strerror(errno));
1323  }
1324  return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1325}
1326
1327void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1328  timeval t;
1329  if (gettimeofday(&t, NULL) == -1) {
1330    fatal("os::javaTimeSystemUTC: gettimeofday (%s)", strerror(errno));
1331  }
1332  seconds = jlong(t.tv_sec);
1333  nanos = jlong(t.tv_usec) * 1000;
1334}
1335
1336
1337jlong os::javaTimeNanos() {
1338  return (jlong)getTimeNanos();
1339}
1340
1341void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1342  info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1343  info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1344  info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1345  info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1346}
1347
1348char * os::local_time_string(char *buf, size_t buflen) {
1349  struct tm t;
1350  time_t long_time;
1351  time(&long_time);
1352  localtime_r(&long_time, &t);
1353  jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1354               t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1355               t.tm_hour, t.tm_min, t.tm_sec);
1356  return buf;
1357}
1358
1359// Note: os::shutdown() might be called very early during initialization, or
1360// called from signal handler. Before adding something to os::shutdown(), make
1361// sure it is async-safe and can handle partially initialized VM.
1362void os::shutdown() {
1363
1364  // allow PerfMemory to attempt cleanup of any persistent resources
1365  perfMemory_exit();
1366
1367  // needs to remove object in file system
1368  AttachListener::abort();
1369
1370  // flush buffered output, finish log files
1371  ostream_abort();
1372
1373  // Check for abort hook
1374  abort_hook_t abort_hook = Arguments::abort_hook();
1375  if (abort_hook != NULL) {
1376    abort_hook();
1377  }
1378}
1379
1380// Note: os::abort() might be called very early during initialization, or
1381// called from signal handler. Before adding something to os::abort(), make
1382// sure it is async-safe and can handle partially initialized VM.
1383void os::abort(bool dump_core, void* siginfo, void* context) {
1384  os::shutdown();
1385  if (dump_core) {
1386#ifndef PRODUCT
1387    fdStream out(defaultStream::output_fd());
1388    out.print_raw("Current thread is ");
1389    char buf[16];
1390    jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1391    out.print_raw_cr(buf);
1392    out.print_raw_cr("Dumping core ...");
1393#endif
1394    ::abort(); // dump core (for debugging)
1395  }
1396
1397  ::exit(1);
1398}
1399
1400// Die immediately, no exit hook, no abort hook, no cleanup.
1401void os::die() {
1402  ::abort(); // dump core (for debugging)
1403}
1404
1405// DLL functions
1406
1407const char* os::dll_file_extension() { return ".so"; }
1408
1409// This must be hard coded because it's the system's temporary
1410// directory not the java application's temp directory, ala java.io.tmpdir.
1411const char* os::get_temp_directory() { return "/tmp"; }
1412
1413static bool file_exists(const char* filename) {
1414  struct stat statbuf;
1415  if (filename == NULL || strlen(filename) == 0) {
1416    return false;
1417  }
1418  return os::stat(filename, &statbuf) == 0;
1419}
1420
1421bool os::dll_build_name(char* buffer, size_t buflen,
1422                        const char* pname, const char* fname) {
1423  bool retval = false;
1424  const size_t pnamelen = pname ? strlen(pname) : 0;
1425
1426  // Return error on buffer overflow.
1427  if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1428    return retval;
1429  }
1430
1431  if (pnamelen == 0) {
1432    snprintf(buffer, buflen, "lib%s.so", fname);
1433    retval = true;
1434  } else if (strchr(pname, *os::path_separator()) != NULL) {
1435    int n;
1436    char** pelements = split_path(pname, &n);
1437    if (pelements == NULL) {
1438      return false;
1439    }
1440    for (int i = 0; i < n; i++) {
1441      // really shouldn't be NULL but what the heck, check can't hurt
1442      if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1443        continue; // skip the empty path values
1444      }
1445      snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1446      if (file_exists(buffer)) {
1447        retval = true;
1448        break;
1449      }
1450    }
1451    // release the storage
1452    for (int i = 0; i < n; i++) {
1453      if (pelements[i] != NULL) {
1454        FREE_C_HEAP_ARRAY(char, pelements[i]);
1455      }
1456    }
1457    if (pelements != NULL) {
1458      FREE_C_HEAP_ARRAY(char*, pelements);
1459    }
1460  } else {
1461    snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1462    retval = true;
1463  }
1464  return retval;
1465}
1466
1467// check if addr is inside libjvm.so
1468bool os::address_is_in_vm(address addr) {
1469  static address libjvm_base_addr;
1470  Dl_info dlinfo;
1471
1472  if (libjvm_base_addr == NULL) {
1473    if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1474      libjvm_base_addr = (address)dlinfo.dli_fbase;
1475    }
1476    assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1477  }
1478
1479  if (dladdr((void *)addr, &dlinfo) != 0) {
1480    if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1481  }
1482
1483  return false;
1484}
1485
1486typedef int (*dladdr1_func_type)(void *, Dl_info *, void **, int);
1487static dladdr1_func_type dladdr1_func = NULL;
1488
1489bool os::dll_address_to_function_name(address addr, char *buf,
1490                                      int buflen, int * offset,
1491                                      bool demangle) {
1492  // buf is not optional, but offset is optional
1493  assert(buf != NULL, "sanity check");
1494
1495  Dl_info dlinfo;
1496
1497  // dladdr1_func was initialized in os::init()
1498  if (dladdr1_func != NULL) {
1499    // yes, we have dladdr1
1500
1501    // Support for dladdr1 is checked at runtime; it may be
1502    // available even if the vm is built on a machine that does
1503    // not have dladdr1 support.  Make sure there is a value for
1504    // RTLD_DL_SYMENT.
1505#ifndef RTLD_DL_SYMENT
1506  #define RTLD_DL_SYMENT 1
1507#endif
1508#ifdef _LP64
1509    Elf64_Sym * info;
1510#else
1511    Elf32_Sym * info;
1512#endif
1513    if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1514                     RTLD_DL_SYMENT) != 0) {
1515      // see if we have a matching symbol that covers our address
1516      if (dlinfo.dli_saddr != NULL &&
1517          (char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1518        if (dlinfo.dli_sname != NULL) {
1519          if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1520            jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1521          }
1522          if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1523          return true;
1524        }
1525      }
1526      // no matching symbol so try for just file info
1527      if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1528        if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1529                            buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1530          return true;
1531        }
1532      }
1533    }
1534    buf[0] = '\0';
1535    if (offset != NULL) *offset  = -1;
1536    return false;
1537  }
1538
1539  // no, only dladdr is available
1540  if (dladdr((void *)addr, &dlinfo) != 0) {
1541    // see if we have a matching symbol
1542    if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1543      if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1544        jio_snprintf(buf, buflen, dlinfo.dli_sname);
1545      }
1546      if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1547      return true;
1548    }
1549    // no matching symbol so try for just file info
1550    if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1551      if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1552                          buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1553        return true;
1554      }
1555    }
1556  }
1557  buf[0] = '\0';
1558  if (offset != NULL) *offset  = -1;
1559  return false;
1560}
1561
1562bool os::dll_address_to_library_name(address addr, char* buf,
1563                                     int buflen, int* offset) {
1564  // buf is not optional, but offset is optional
1565  assert(buf != NULL, "sanity check");
1566
1567  Dl_info dlinfo;
1568
1569  if (dladdr((void*)addr, &dlinfo) != 0) {
1570    if (dlinfo.dli_fname != NULL) {
1571      jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1572    }
1573    if (dlinfo.dli_fbase != NULL && offset != NULL) {
1574      *offset = addr - (address)dlinfo.dli_fbase;
1575    }
1576    return true;
1577  }
1578
1579  buf[0] = '\0';
1580  if (offset) *offset = -1;
1581  return false;
1582}
1583
1584int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1585  Dl_info dli;
1586  // Sanity check?
1587  if (dladdr(CAST_FROM_FN_PTR(void *, os::get_loaded_modules_info), &dli) == 0 ||
1588      dli.dli_fname == NULL) {
1589    return 1;
1590  }
1591
1592  void * handle = dlopen(dli.dli_fname, RTLD_LAZY);
1593  if (handle == NULL) {
1594    return 1;
1595  }
1596
1597  Link_map *map;
1598  dlinfo(handle, RTLD_DI_LINKMAP, &map);
1599  if (map == NULL) {
1600    dlclose(handle);
1601    return 1;
1602  }
1603
1604  while (map->l_prev != NULL) {
1605    map = map->l_prev;
1606  }
1607
1608  while (map != NULL) {
1609    // Iterate through all map entries and call callback with fields of interest
1610    if(callback(map->l_name, (address)map->l_addr, (address)0, param)) {
1611      dlclose(handle);
1612      return 1;
1613    }
1614    map = map->l_next;
1615  }
1616
1617  dlclose(handle);
1618  return 0;
1619}
1620
1621int _print_dll_info_cb(const char * name, address base_address, address top_address, void * param) {
1622  outputStream * out = (outputStream *) param;
1623  out->print_cr(PTR_FORMAT " \t%s", base_address, name);
1624  return 0;
1625}
1626
1627void os::print_dll_info(outputStream * st) {
1628  st->print_cr("Dynamic libraries:"); st->flush();
1629  if (get_loaded_modules_info(_print_dll_info_cb, (void *)st)) {
1630    st->print_cr("Error: Cannot print dynamic libraries.");
1631  }
1632}
1633
1634// Loads .dll/.so and
1635// in case of error it checks if .dll/.so was built for the
1636// same architecture as Hotspot is running on
1637
1638void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1639  void * result= ::dlopen(filename, RTLD_LAZY);
1640  if (result != NULL) {
1641    // Successful loading
1642    return result;
1643  }
1644
1645  Elf32_Ehdr elf_head;
1646
1647  // Read system error message into ebuf
1648  // It may or may not be overwritten below
1649  ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1650  ebuf[ebuflen-1]='\0';
1651  int diag_msg_max_length=ebuflen-strlen(ebuf);
1652  char* diag_msg_buf=ebuf+strlen(ebuf);
1653
1654  if (diag_msg_max_length==0) {
1655    // No more space in ebuf for additional diagnostics message
1656    return NULL;
1657  }
1658
1659
1660  int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1661
1662  if (file_descriptor < 0) {
1663    // Can't open library, report dlerror() message
1664    return NULL;
1665  }
1666
1667  bool failed_to_read_elf_head=
1668    (sizeof(elf_head)!=
1669     (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1670
1671  ::close(file_descriptor);
1672  if (failed_to_read_elf_head) {
1673    // file i/o error - report dlerror() msg
1674    return NULL;
1675  }
1676
1677  typedef struct {
1678    Elf32_Half  code;         // Actual value as defined in elf.h
1679    Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1680    char        elf_class;    // 32 or 64 bit
1681    char        endianess;    // MSB or LSB
1682    char*       name;         // String representation
1683  } arch_t;
1684
1685  static const arch_t arch_array[]={
1686    {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1687    {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1688    {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1689    {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1690    {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1691    {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1692    {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1693    {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1694    {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1695    {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
1696  };
1697
1698#if  (defined IA32)
1699  static  Elf32_Half running_arch_code=EM_386;
1700#elif   (defined AMD64)
1701  static  Elf32_Half running_arch_code=EM_X86_64;
1702#elif  (defined IA64)
1703  static  Elf32_Half running_arch_code=EM_IA_64;
1704#elif  (defined __sparc) && (defined _LP64)
1705  static  Elf32_Half running_arch_code=EM_SPARCV9;
1706#elif  (defined __sparc) && (!defined _LP64)
1707  static  Elf32_Half running_arch_code=EM_SPARC;
1708#elif  (defined __powerpc64__)
1709  static  Elf32_Half running_arch_code=EM_PPC64;
1710#elif  (defined __powerpc__)
1711  static  Elf32_Half running_arch_code=EM_PPC;
1712#elif (defined ARM)
1713  static  Elf32_Half running_arch_code=EM_ARM;
1714#else
1715  #error Method os::dll_load requires that one of following is defined:\
1716       IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
1717#endif
1718
1719  // Identify compatability class for VM's architecture and library's architecture
1720  // Obtain string descriptions for architectures
1721
1722  arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1723  int running_arch_index=-1;
1724
1725  for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1726    if (running_arch_code == arch_array[i].code) {
1727      running_arch_index    = i;
1728    }
1729    if (lib_arch.code == arch_array[i].code) {
1730      lib_arch.compat_class = arch_array[i].compat_class;
1731      lib_arch.name         = arch_array[i].name;
1732    }
1733  }
1734
1735  assert(running_arch_index != -1,
1736         "Didn't find running architecture code (running_arch_code) in arch_array");
1737  if (running_arch_index == -1) {
1738    // Even though running architecture detection failed
1739    // we may still continue with reporting dlerror() message
1740    return NULL;
1741  }
1742
1743  if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1744    ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1745    return NULL;
1746  }
1747
1748  if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1749    ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1750    return NULL;
1751  }
1752
1753  if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1754    if (lib_arch.name!=NULL) {
1755      ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1756                 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1757                 lib_arch.name, arch_array[running_arch_index].name);
1758    } else {
1759      ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1760                 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1761                 lib_arch.code,
1762                 arch_array[running_arch_index].name);
1763    }
1764  }
1765
1766  return NULL;
1767}
1768
1769void* os::dll_lookup(void* handle, const char* name) {
1770  return dlsym(handle, name);
1771}
1772
1773void* os::get_default_process_handle() {
1774  return (void*)::dlopen(NULL, RTLD_LAZY);
1775}
1776
1777int os::stat(const char *path, struct stat *sbuf) {
1778  char pathbuf[MAX_PATH];
1779  if (strlen(path) > MAX_PATH - 1) {
1780    errno = ENAMETOOLONG;
1781    return -1;
1782  }
1783  os::native_path(strcpy(pathbuf, path));
1784  return ::stat(pathbuf, sbuf);
1785}
1786
1787static bool _print_ascii_file(const char* filename, outputStream* st) {
1788  int fd = ::open(filename, O_RDONLY);
1789  if (fd == -1) {
1790    return false;
1791  }
1792
1793  char buf[32];
1794  int bytes;
1795  while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1796    st->print_raw(buf, bytes);
1797  }
1798
1799  ::close(fd);
1800
1801  return true;
1802}
1803
1804void os::print_os_info_brief(outputStream* st) {
1805  os::Solaris::print_distro_info(st);
1806
1807  os::Posix::print_uname_info(st);
1808
1809  os::Solaris::print_libversion_info(st);
1810}
1811
1812void os::print_os_info(outputStream* st) {
1813  st->print("OS:");
1814
1815  os::Solaris::print_distro_info(st);
1816
1817  os::Posix::print_uname_info(st);
1818
1819  os::Solaris::print_libversion_info(st);
1820
1821  os::Posix::print_rlimit_info(st);
1822
1823  os::Posix::print_load_average(st);
1824}
1825
1826void os::Solaris::print_distro_info(outputStream* st) {
1827  if (!_print_ascii_file("/etc/release", st)) {
1828    st->print("Solaris");
1829  }
1830  st->cr();
1831}
1832
1833void os::get_summary_os_info(char* buf, size_t buflen) {
1834  strncpy(buf, "Solaris", buflen);  // default to plain solaris
1835  FILE* fp = fopen("/etc/release", "r");
1836  if (fp != NULL) {
1837    char tmp[256];
1838    // Only get the first line and chop out everything but the os name.
1839    if (fgets(tmp, sizeof(tmp), fp)) {
1840      char* ptr = tmp;
1841      // skip past whitespace characters
1842      while (*ptr != '\0' && (*ptr == ' ' || *ptr == '\t' || *ptr == '\n')) ptr++;
1843      if (*ptr != '\0') {
1844        char* nl = strchr(ptr, '\n');
1845        if (nl != NULL) *nl = '\0';
1846        strncpy(buf, ptr, buflen);
1847      }
1848    }
1849    fclose(fp);
1850  }
1851}
1852
1853void os::Solaris::print_libversion_info(outputStream* st) {
1854  st->print("  (T2 libthread)");
1855  st->cr();
1856}
1857
1858static bool check_addr0(outputStream* st) {
1859  jboolean status = false;
1860  int fd = ::open("/proc/self/map",O_RDONLY);
1861  if (fd >= 0) {
1862    prmap_t p;
1863    while (::read(fd, &p, sizeof(p)) > 0) {
1864      if (p.pr_vaddr == 0x0) {
1865        st->print("Warning: Address: 0x%x, Size: %dK, ",p.pr_vaddr, p.pr_size/1024, p.pr_mapname);
1866        st->print("Mapped file: %s, ", p.pr_mapname[0] == '\0' ? "None" : p.pr_mapname);
1867        st->print("Access:");
1868        st->print("%s",(p.pr_mflags & MA_READ)  ? "r" : "-");
1869        st->print("%s",(p.pr_mflags & MA_WRITE) ? "w" : "-");
1870        st->print("%s",(p.pr_mflags & MA_EXEC)  ? "x" : "-");
1871        st->cr();
1872        status = true;
1873      }
1874    }
1875    ::close(fd);
1876  }
1877  return status;
1878}
1879
1880void os::get_summary_cpu_info(char* buf, size_t buflen) {
1881  // Get MHz with system call. We don't seem to already have this.
1882  processor_info_t stats;
1883  processorid_t id = getcpuid();
1884  int clock = 0;
1885  if (processor_info(id, &stats) != -1) {
1886    clock = stats.pi_clock;  // pi_processor_type isn't more informative than below
1887  }
1888#ifdef AMD64
1889  snprintf(buf, buflen, "x86 64 bit %d MHz", clock);
1890#else
1891  // must be sparc
1892  snprintf(buf, buflen, "Sparcv9 64 bit %d MHz", clock);
1893#endif
1894}
1895
1896void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
1897  // Nothing to do for now.
1898}
1899
1900void os::print_memory_info(outputStream* st) {
1901  st->print("Memory:");
1902  st->print(" %dk page", os::vm_page_size()>>10);
1903  st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
1904  st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
1905  st->cr();
1906  (void) check_addr0(st);
1907}
1908
1909void os::print_siginfo(outputStream* st, void* siginfo) {
1910  const siginfo_t* si = (const siginfo_t*)siginfo;
1911
1912  os::Posix::print_siginfo_brief(st, si);
1913
1914  if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
1915      UseSharedSpaces) {
1916    FileMapInfo* mapinfo = FileMapInfo::current_info();
1917    if (mapinfo->is_in_shared_space(si->si_addr)) {
1918      st->print("\n\nError accessing class data sharing archive."   \
1919                " Mapped file inaccessible during execution, "      \
1920                " possible disk/network problem.");
1921    }
1922  }
1923  st->cr();
1924}
1925
1926// Moved from whole group, because we need them here for diagnostic
1927// prints.
1928#define OLDMAXSIGNUM 32
1929static int Maxsignum = 0;
1930static int *ourSigFlags = NULL;
1931
1932int os::Solaris::get_our_sigflags(int sig) {
1933  assert(ourSigFlags!=NULL, "signal data structure not initialized");
1934  assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1935  return ourSigFlags[sig];
1936}
1937
1938void os::Solaris::set_our_sigflags(int sig, int flags) {
1939  assert(ourSigFlags!=NULL, "signal data structure not initialized");
1940  assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1941  ourSigFlags[sig] = flags;
1942}
1943
1944
1945static const char* get_signal_handler_name(address handler,
1946                                           char* buf, int buflen) {
1947  int offset;
1948  bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
1949  if (found) {
1950    // skip directory names
1951    const char *p1, *p2;
1952    p1 = buf;
1953    size_t len = strlen(os::file_separator());
1954    while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
1955    jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
1956  } else {
1957    jio_snprintf(buf, buflen, PTR_FORMAT, handler);
1958  }
1959  return buf;
1960}
1961
1962static void print_signal_handler(outputStream* st, int sig,
1963                                 char* buf, size_t buflen) {
1964  struct sigaction sa;
1965
1966  sigaction(sig, NULL, &sa);
1967
1968  st->print("%s: ", os::exception_name(sig, buf, buflen));
1969
1970  address handler = (sa.sa_flags & SA_SIGINFO)
1971                  ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
1972                  : CAST_FROM_FN_PTR(address, sa.sa_handler);
1973
1974  if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
1975    st->print("SIG_DFL");
1976  } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
1977    st->print("SIG_IGN");
1978  } else {
1979    st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
1980  }
1981
1982  st->print(", sa_mask[0]=");
1983  os::Posix::print_signal_set_short(st, &sa.sa_mask);
1984
1985  address rh = VMError::get_resetted_sighandler(sig);
1986  // May be, handler was resetted by VMError?
1987  if (rh != NULL) {
1988    handler = rh;
1989    sa.sa_flags = VMError::get_resetted_sigflags(sig);
1990  }
1991
1992  st->print(", sa_flags=");
1993  os::Posix::print_sa_flags(st, sa.sa_flags);
1994
1995  // Check: is it our handler?
1996  if (handler == CAST_FROM_FN_PTR(address, signalHandler)) {
1997    // It is our signal handler
1998    // check for flags
1999    if (sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2000      st->print(
2001                ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2002                os::Solaris::get_our_sigflags(sig));
2003    }
2004  }
2005  st->cr();
2006}
2007
2008void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2009  st->print_cr("Signal Handlers:");
2010  print_signal_handler(st, SIGSEGV, buf, buflen);
2011  print_signal_handler(st, SIGBUS , buf, buflen);
2012  print_signal_handler(st, SIGFPE , buf, buflen);
2013  print_signal_handler(st, SIGPIPE, buf, buflen);
2014  print_signal_handler(st, SIGXFSZ, buf, buflen);
2015  print_signal_handler(st, SIGILL , buf, buflen);
2016  print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2017  print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2018  print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2019  print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2020  print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2021  print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2022}
2023
2024static char saved_jvm_path[MAXPATHLEN] = { 0 };
2025
2026// Find the full path to the current module, libjvm.so
2027void os::jvm_path(char *buf, jint buflen) {
2028  // Error checking.
2029  if (buflen < MAXPATHLEN) {
2030    assert(false, "must use a large-enough buffer");
2031    buf[0] = '\0';
2032    return;
2033  }
2034  // Lazy resolve the path to current module.
2035  if (saved_jvm_path[0] != 0) {
2036    strcpy(buf, saved_jvm_path);
2037    return;
2038  }
2039
2040  Dl_info dlinfo;
2041  int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2042  assert(ret != 0, "cannot locate libjvm");
2043  if (ret != 0 && dlinfo.dli_fname != NULL) {
2044    realpath((char *)dlinfo.dli_fname, buf);
2045  } else {
2046    buf[0] = '\0';
2047    return;
2048  }
2049
2050  if (Arguments::sun_java_launcher_is_altjvm()) {
2051    // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2052    // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2053    // If "/jre/lib/" appears at the right place in the string, then
2054    // assume we are installed in a JDK and we're done.  Otherwise, check
2055    // for a JAVA_HOME environment variable and fix up the path so it
2056    // looks like libjvm.so is installed there (append a fake suffix
2057    // hotspot/libjvm.so).
2058    const char *p = buf + strlen(buf) - 1;
2059    for (int count = 0; p > buf && count < 5; ++count) {
2060      for (--p; p > buf && *p != '/'; --p)
2061        /* empty */ ;
2062    }
2063
2064    if (strncmp(p, "/jre/lib/", 9) != 0) {
2065      // Look for JAVA_HOME in the environment.
2066      char* java_home_var = ::getenv("JAVA_HOME");
2067      if (java_home_var != NULL && java_home_var[0] != 0) {
2068        char cpu_arch[12];
2069        char* jrelib_p;
2070        int   len;
2071        sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2072#ifdef _LP64
2073        // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2074        if (strcmp(cpu_arch, "sparc") == 0) {
2075          strcat(cpu_arch, "v9");
2076        } else if (strcmp(cpu_arch, "i386") == 0) {
2077          strcpy(cpu_arch, "amd64");
2078        }
2079#endif
2080        // Check the current module name "libjvm.so".
2081        p = strrchr(buf, '/');
2082        assert(strstr(p, "/libjvm") == p, "invalid library name");
2083
2084        realpath(java_home_var, buf);
2085        // determine if this is a legacy image or modules image
2086        // modules image doesn't have "jre" subdirectory
2087        len = strlen(buf);
2088        assert(len < buflen, "Ran out of buffer space");
2089        jrelib_p = buf + len;
2090        snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2091        if (0 != access(buf, F_OK)) {
2092          snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2093        }
2094
2095        if (0 == access(buf, F_OK)) {
2096          // Use current module name "libjvm.so"
2097          len = strlen(buf);
2098          snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2099        } else {
2100          // Go back to path of .so
2101          realpath((char *)dlinfo.dli_fname, buf);
2102        }
2103      }
2104    }
2105  }
2106
2107  strncpy(saved_jvm_path, buf, MAXPATHLEN);
2108  saved_jvm_path[MAXPATHLEN - 1] = '\0';
2109}
2110
2111
2112void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2113  // no prefix required, not even "_"
2114}
2115
2116
2117void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2118  // no suffix required
2119}
2120
2121// This method is a copy of JDK's sysGetLastErrorString
2122// from src/solaris/hpi/src/system_md.c
2123
2124size_t os::lasterror(char *buf, size_t len) {
2125  if (errno == 0)  return 0;
2126
2127  const char *s = ::strerror(errno);
2128  size_t n = ::strlen(s);
2129  if (n >= len) {
2130    n = len - 1;
2131  }
2132  ::strncpy(buf, s, n);
2133  buf[n] = '\0';
2134  return n;
2135}
2136
2137
2138// sun.misc.Signal
2139
2140extern "C" {
2141  static void UserHandler(int sig, void *siginfo, void *context) {
2142    // Ctrl-C is pressed during error reporting, likely because the error
2143    // handler fails to abort. Let VM die immediately.
2144    if (sig == SIGINT && is_error_reported()) {
2145      os::die();
2146    }
2147
2148    os::signal_notify(sig);
2149    // We do not need to reinstate the signal handler each time...
2150  }
2151}
2152
2153void* os::user_handler() {
2154  return CAST_FROM_FN_PTR(void*, UserHandler);
2155}
2156
2157struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2158  struct timespec ts;
2159  unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2160
2161  return ts;
2162}
2163
2164extern "C" {
2165  typedef void (*sa_handler_t)(int);
2166  typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2167}
2168
2169void* os::signal(int signal_number, void* handler) {
2170  struct sigaction sigAct, oldSigAct;
2171  sigfillset(&(sigAct.sa_mask));
2172  sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2173  sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2174
2175  if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2176    // -1 means registration failed
2177    return (void *)-1;
2178  }
2179
2180  return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2181}
2182
2183void os::signal_raise(int signal_number) {
2184  raise(signal_number);
2185}
2186
2187// The following code is moved from os.cpp for making this
2188// code platform specific, which it is by its very nature.
2189
2190// a counter for each possible signal value
2191static int Sigexit = 0;
2192static int Maxlibjsigsigs;
2193static jint *pending_signals = NULL;
2194static int *preinstalled_sigs = NULL;
2195static struct sigaction *chainedsigactions = NULL;
2196static sema_t sig_sem;
2197typedef int (*version_getting_t)();
2198version_getting_t os::Solaris::get_libjsig_version = NULL;
2199static int libjsigversion = NULL;
2200
2201int os::sigexitnum_pd() {
2202  assert(Sigexit > 0, "signal memory not yet initialized");
2203  return Sigexit;
2204}
2205
2206void os::Solaris::init_signal_mem() {
2207  // Initialize signal structures
2208  Maxsignum = SIGRTMAX;
2209  Sigexit = Maxsignum+1;
2210  assert(Maxsignum >0, "Unable to obtain max signal number");
2211
2212  Maxlibjsigsigs = Maxsignum;
2213
2214  // pending_signals has one int per signal
2215  // The additional signal is for SIGEXIT - exit signal to signal_thread
2216  pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2217  memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2218
2219  if (UseSignalChaining) {
2220    chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2221                                                   * (Maxsignum + 1), mtInternal);
2222    memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2223    preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2224    memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2225  }
2226  ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2227  memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2228}
2229
2230void os::signal_init_pd() {
2231  int ret;
2232
2233  ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2234  assert(ret == 0, "sema_init() failed");
2235}
2236
2237void os::signal_notify(int signal_number) {
2238  int ret;
2239
2240  Atomic::inc(&pending_signals[signal_number]);
2241  ret = ::sema_post(&sig_sem);
2242  assert(ret == 0, "sema_post() failed");
2243}
2244
2245static int check_pending_signals(bool wait_for_signal) {
2246  int ret;
2247  while (true) {
2248    for (int i = 0; i < Sigexit + 1; i++) {
2249      jint n = pending_signals[i];
2250      if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2251        return i;
2252      }
2253    }
2254    if (!wait_for_signal) {
2255      return -1;
2256    }
2257    JavaThread *thread = JavaThread::current();
2258    ThreadBlockInVM tbivm(thread);
2259
2260    bool threadIsSuspended;
2261    do {
2262      thread->set_suspend_equivalent();
2263      // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2264      while ((ret = ::sema_wait(&sig_sem)) == EINTR)
2265        ;
2266      assert(ret == 0, "sema_wait() failed");
2267
2268      // were we externally suspended while we were waiting?
2269      threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2270      if (threadIsSuspended) {
2271        // The semaphore has been incremented, but while we were waiting
2272        // another thread suspended us. We don't want to continue running
2273        // while suspended because that would surprise the thread that
2274        // suspended us.
2275        ret = ::sema_post(&sig_sem);
2276        assert(ret == 0, "sema_post() failed");
2277
2278        thread->java_suspend_self();
2279      }
2280    } while (threadIsSuspended);
2281  }
2282}
2283
2284int os::signal_lookup() {
2285  return check_pending_signals(false);
2286}
2287
2288int os::signal_wait() {
2289  return check_pending_signals(true);
2290}
2291
2292////////////////////////////////////////////////////////////////////////////////
2293// Virtual Memory
2294
2295static int page_size = -1;
2296
2297// The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2298// clear this var if support is not available.
2299static bool has_map_align = true;
2300
2301int os::vm_page_size() {
2302  assert(page_size != -1, "must call os::init");
2303  return page_size;
2304}
2305
2306// Solaris allocates memory by pages.
2307int os::vm_allocation_granularity() {
2308  assert(page_size != -1, "must call os::init");
2309  return page_size;
2310}
2311
2312static bool recoverable_mmap_error(int err) {
2313  // See if the error is one we can let the caller handle. This
2314  // list of errno values comes from the Solaris mmap(2) man page.
2315  switch (err) {
2316  case EBADF:
2317  case EINVAL:
2318  case ENOTSUP:
2319    // let the caller deal with these errors
2320    return true;
2321
2322  default:
2323    // Any remaining errors on this OS can cause our reserved mapping
2324    // to be lost. That can cause confusion where different data
2325    // structures think they have the same memory mapped. The worst
2326    // scenario is if both the VM and a library think they have the
2327    // same memory mapped.
2328    return false;
2329  }
2330}
2331
2332static void warn_fail_commit_memory(char* addr, size_t bytes, bool exec,
2333                                    int err) {
2334  warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2335          ", %d) failed; error='%s' (errno=%d)", addr, bytes, exec,
2336          strerror(err), err);
2337}
2338
2339static void warn_fail_commit_memory(char* addr, size_t bytes,
2340                                    size_t alignment_hint, bool exec,
2341                                    int err) {
2342  warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2343          ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, bytes,
2344          alignment_hint, exec, strerror(err), err);
2345}
2346
2347int os::Solaris::commit_memory_impl(char* addr, size_t bytes, bool exec) {
2348  int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2349  size_t size = bytes;
2350  char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2351  if (res != NULL) {
2352    if (UseNUMAInterleaving) {
2353      numa_make_global(addr, bytes);
2354    }
2355    return 0;
2356  }
2357
2358  int err = errno;  // save errno from mmap() call in mmap_chunk()
2359
2360  if (!recoverable_mmap_error(err)) {
2361    warn_fail_commit_memory(addr, bytes, exec, err);
2362    vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "committing reserved memory.");
2363  }
2364
2365  return err;
2366}
2367
2368bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2369  return Solaris::commit_memory_impl(addr, bytes, exec) == 0;
2370}
2371
2372void os::pd_commit_memory_or_exit(char* addr, size_t bytes, bool exec,
2373                                  const char* mesg) {
2374  assert(mesg != NULL, "mesg must be specified");
2375  int err = os::Solaris::commit_memory_impl(addr, bytes, exec);
2376  if (err != 0) {
2377    // the caller wants all commit errors to exit with the specified mesg:
2378    warn_fail_commit_memory(addr, bytes, exec, err);
2379    vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2380  }
2381}
2382
2383size_t os::Solaris::page_size_for_alignment(size_t alignment) {
2384  assert(is_size_aligned(alignment, (size_t) vm_page_size()),
2385         SIZE_FORMAT " is not aligned to " SIZE_FORMAT,
2386         alignment, (size_t) vm_page_size());
2387
2388  for (int i = 0; _page_sizes[i] != 0; i++) {
2389    if (is_size_aligned(alignment, _page_sizes[i])) {
2390      return _page_sizes[i];
2391    }
2392  }
2393
2394  return (size_t) vm_page_size();
2395}
2396
2397int os::Solaris::commit_memory_impl(char* addr, size_t bytes,
2398                                    size_t alignment_hint, bool exec) {
2399  int err = Solaris::commit_memory_impl(addr, bytes, exec);
2400  if (err == 0 && UseLargePages && alignment_hint > 0) {
2401    assert(is_size_aligned(bytes, alignment_hint),
2402           SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, alignment_hint);
2403
2404    // The syscall memcntl requires an exact page size (see man memcntl for details).
2405    size_t page_size = page_size_for_alignment(alignment_hint);
2406    if (page_size > (size_t) vm_page_size()) {
2407      (void)Solaris::setup_large_pages(addr, bytes, page_size);
2408    }
2409  }
2410  return err;
2411}
2412
2413bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2414                          bool exec) {
2415  return Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec) == 0;
2416}
2417
2418void os::pd_commit_memory_or_exit(char* addr, size_t bytes,
2419                                  size_t alignment_hint, bool exec,
2420                                  const char* mesg) {
2421  assert(mesg != NULL, "mesg must be specified");
2422  int err = os::Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec);
2423  if (err != 0) {
2424    // the caller wants all commit errors to exit with the specified mesg:
2425    warn_fail_commit_memory(addr, bytes, alignment_hint, exec, err);
2426    vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2427  }
2428}
2429
2430// Uncommit the pages in a specified region.
2431void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2432  if (madvise(addr, bytes, MADV_FREE) < 0) {
2433    debug_only(warning("MADV_FREE failed."));
2434    return;
2435  }
2436}
2437
2438bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2439  return os::commit_memory(addr, size, !ExecMem);
2440}
2441
2442bool os::remove_stack_guard_pages(char* addr, size_t size) {
2443  return os::uncommit_memory(addr, size);
2444}
2445
2446// Change the page size in a given range.
2447void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2448  assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2449  assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2450  if (UseLargePages) {
2451    size_t page_size = Solaris::page_size_for_alignment(alignment_hint);
2452    if (page_size > (size_t) vm_page_size()) {
2453      Solaris::setup_large_pages(addr, bytes, page_size);
2454    }
2455  }
2456}
2457
2458// Tell the OS to make the range local to the first-touching LWP
2459void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2460  assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2461  if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2462    debug_only(warning("MADV_ACCESS_LWP failed."));
2463  }
2464}
2465
2466// Tell the OS that this range would be accessed from different LWPs.
2467void os::numa_make_global(char *addr, size_t bytes) {
2468  assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2469  if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2470    debug_only(warning("MADV_ACCESS_MANY failed."));
2471  }
2472}
2473
2474// Get the number of the locality groups.
2475size_t os::numa_get_groups_num() {
2476  size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2477  return n != -1 ? n : 1;
2478}
2479
2480// Get a list of leaf locality groups. A leaf lgroup is group that
2481// doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2482// board. An LWP is assigned to one of these groups upon creation.
2483size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2484  if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2485    ids[0] = 0;
2486    return 1;
2487  }
2488  int result_size = 0, top = 1, bottom = 0, cur = 0;
2489  for (int k = 0; k < size; k++) {
2490    int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2491                                   (Solaris::lgrp_id_t*)&ids[top], size - top);
2492    if (r == -1) {
2493      ids[0] = 0;
2494      return 1;
2495    }
2496    if (!r) {
2497      // That's a leaf node.
2498      assert(bottom <= cur, "Sanity check");
2499      // Check if the node has memory
2500      if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2501                                  NULL, 0, LGRP_RSRC_MEM) > 0) {
2502        ids[bottom++] = ids[cur];
2503      }
2504    }
2505    top += r;
2506    cur++;
2507  }
2508  if (bottom == 0) {
2509    // Handle a situation, when the OS reports no memory available.
2510    // Assume UMA architecture.
2511    ids[0] = 0;
2512    return 1;
2513  }
2514  return bottom;
2515}
2516
2517// Detect the topology change. Typically happens during CPU plugging-unplugging.
2518bool os::numa_topology_changed() {
2519  int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2520  if (is_stale != -1 && is_stale) {
2521    Solaris::lgrp_fini(Solaris::lgrp_cookie());
2522    Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2523    assert(c != 0, "Failure to initialize LGRP API");
2524    Solaris::set_lgrp_cookie(c);
2525    return true;
2526  }
2527  return false;
2528}
2529
2530// Get the group id of the current LWP.
2531int os::numa_get_group_id() {
2532  int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2533  if (lgrp_id == -1) {
2534    return 0;
2535  }
2536  const int size = os::numa_get_groups_num();
2537  int *ids = (int*)alloca(size * sizeof(int));
2538
2539  // Get the ids of all lgroups with memory; r is the count.
2540  int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2541                                  (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2542  if (r <= 0) {
2543    return 0;
2544  }
2545  return ids[os::random() % r];
2546}
2547
2548// Request information about the page.
2549bool os::get_page_info(char *start, page_info* info) {
2550  const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2551  uint64_t addr = (uintptr_t)start;
2552  uint64_t outdata[2];
2553  uint_t validity = 0;
2554
2555  if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2556    return false;
2557  }
2558
2559  info->size = 0;
2560  info->lgrp_id = -1;
2561
2562  if ((validity & 1) != 0) {
2563    if ((validity & 2) != 0) {
2564      info->lgrp_id = outdata[0];
2565    }
2566    if ((validity & 4) != 0) {
2567      info->size = outdata[1];
2568    }
2569    return true;
2570  }
2571  return false;
2572}
2573
2574// Scan the pages from start to end until a page different than
2575// the one described in the info parameter is encountered.
2576char *os::scan_pages(char *start, char* end, page_info* page_expected,
2577                     page_info* page_found) {
2578  const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2579  const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2580  uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT + 1];
2581  uint_t validity[MAX_MEMINFO_CNT];
2582
2583  size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2584  uint64_t p = (uint64_t)start;
2585  while (p < (uint64_t)end) {
2586    addrs[0] = p;
2587    size_t addrs_count = 1;
2588    while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] + page_size < (uint64_t)end) {
2589      addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2590      addrs_count++;
2591    }
2592
2593    if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2594      return NULL;
2595    }
2596
2597    size_t i = 0;
2598    for (; i < addrs_count; i++) {
2599      if ((validity[i] & 1) != 0) {
2600        if ((validity[i] & 4) != 0) {
2601          if (outdata[types * i + 1] != page_expected->size) {
2602            break;
2603          }
2604        } else if (page_expected->size != 0) {
2605          break;
2606        }
2607
2608        if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2609          if (outdata[types * i] != page_expected->lgrp_id) {
2610            break;
2611          }
2612        }
2613      } else {
2614        return NULL;
2615      }
2616    }
2617
2618    if (i < addrs_count) {
2619      if ((validity[i] & 2) != 0) {
2620        page_found->lgrp_id = outdata[types * i];
2621      } else {
2622        page_found->lgrp_id = -1;
2623      }
2624      if ((validity[i] & 4) != 0) {
2625        page_found->size = outdata[types * i + 1];
2626      } else {
2627        page_found->size = 0;
2628      }
2629      return (char*)addrs[i];
2630    }
2631
2632    p = addrs[addrs_count - 1] + page_size;
2633  }
2634  return end;
2635}
2636
2637bool os::pd_uncommit_memory(char* addr, size_t bytes) {
2638  size_t size = bytes;
2639  // Map uncommitted pages PROT_NONE so we fail early if we touch an
2640  // uncommitted page. Otherwise, the read/write might succeed if we
2641  // have enough swap space to back the physical page.
2642  return
2643    NULL != Solaris::mmap_chunk(addr, size,
2644                                MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2645                                PROT_NONE);
2646}
2647
2648char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2649  char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2650
2651  if (b == MAP_FAILED) {
2652    return NULL;
2653  }
2654  return b;
2655}
2656
2657char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes,
2658                             size_t alignment_hint, bool fixed) {
2659  char* addr = requested_addr;
2660  int flags = MAP_PRIVATE | MAP_NORESERVE;
2661
2662  assert(!(fixed && (alignment_hint > 0)),
2663         "alignment hint meaningless with fixed mmap");
2664
2665  if (fixed) {
2666    flags |= MAP_FIXED;
2667  } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
2668    flags |= MAP_ALIGN;
2669    addr = (char*) alignment_hint;
2670  }
2671
2672  // Map uncommitted pages PROT_NONE so we fail early if we touch an
2673  // uncommitted page. Otherwise, the read/write might succeed if we
2674  // have enough swap space to back the physical page.
2675  return mmap_chunk(addr, bytes, flags, PROT_NONE);
2676}
2677
2678char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2679                            size_t alignment_hint) {
2680  char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint,
2681                                  (requested_addr != NULL));
2682
2683  guarantee(requested_addr == NULL || requested_addr == addr,
2684            "OS failed to return requested mmap address.");
2685  return addr;
2686}
2687
2688// Reserve memory at an arbitrary address, only if that area is
2689// available (and not reserved for something else).
2690
2691char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2692  const int max_tries = 10;
2693  char* base[max_tries];
2694  size_t size[max_tries];
2695
2696  // Solaris adds a gap between mmap'ed regions.  The size of the gap
2697  // is dependent on the requested size and the MMU.  Our initial gap
2698  // value here is just a guess and will be corrected later.
2699  bool had_top_overlap = false;
2700  bool have_adjusted_gap = false;
2701  size_t gap = 0x400000;
2702
2703  // Assert only that the size is a multiple of the page size, since
2704  // that's all that mmap requires, and since that's all we really know
2705  // about at this low abstraction level.  If we need higher alignment,
2706  // we can either pass an alignment to this method or verify alignment
2707  // in one of the methods further up the call chain.  See bug 5044738.
2708  assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2709
2710  // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2711  // Give it a try, if the kernel honors the hint we can return immediately.
2712  char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2713
2714  volatile int err = errno;
2715  if (addr == requested_addr) {
2716    return addr;
2717  } else if (addr != NULL) {
2718    pd_unmap_memory(addr, bytes);
2719  }
2720
2721  if (PrintMiscellaneous && Verbose) {
2722    char buf[256];
2723    buf[0] = '\0';
2724    if (addr == NULL) {
2725      jio_snprintf(buf, sizeof(buf), ": %s", strerror(err));
2726    }
2727    warning("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
2728            PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2729            "%s", bytes, requested_addr, addr, buf);
2730  }
2731
2732  // Address hint method didn't work.  Fall back to the old method.
2733  // In theory, once SNV becomes our oldest supported platform, this
2734  // code will no longer be needed.
2735  //
2736  // Repeatedly allocate blocks until the block is allocated at the
2737  // right spot. Give up after max_tries.
2738  int i;
2739  for (i = 0; i < max_tries; ++i) {
2740    base[i] = reserve_memory(bytes);
2741
2742    if (base[i] != NULL) {
2743      // Is this the block we wanted?
2744      if (base[i] == requested_addr) {
2745        size[i] = bytes;
2746        break;
2747      }
2748
2749      // check that the gap value is right
2750      if (had_top_overlap && !have_adjusted_gap) {
2751        size_t actual_gap = base[i-1] - base[i] - bytes;
2752        if (gap != actual_gap) {
2753          // adjust the gap value and retry the last 2 allocations
2754          assert(i > 0, "gap adjustment code problem");
2755          have_adjusted_gap = true;  // adjust the gap only once, just in case
2756          gap = actual_gap;
2757          if (PrintMiscellaneous && Verbose) {
2758            warning("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
2759          }
2760          unmap_memory(base[i], bytes);
2761          unmap_memory(base[i-1], size[i-1]);
2762          i-=2;
2763          continue;
2764        }
2765      }
2766
2767      // Does this overlap the block we wanted? Give back the overlapped
2768      // parts and try again.
2769      //
2770      // There is still a bug in this code: if top_overlap == bytes,
2771      // the overlap is offset from requested region by the value of gap.
2772      // In this case giving back the overlapped part will not work,
2773      // because we'll give back the entire block at base[i] and
2774      // therefore the subsequent allocation will not generate a new gap.
2775      // This could be fixed with a new algorithm that used larger
2776      // or variable size chunks to find the requested region -
2777      // but such a change would introduce additional complications.
2778      // It's rare enough that the planets align for this bug,
2779      // so we'll just wait for a fix for 6204603/5003415 which
2780      // will provide a mmap flag to allow us to avoid this business.
2781
2782      size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2783      if (top_overlap >= 0 && top_overlap < bytes) {
2784        had_top_overlap = true;
2785        unmap_memory(base[i], top_overlap);
2786        base[i] += top_overlap;
2787        size[i] = bytes - top_overlap;
2788      } else {
2789        size_t bottom_overlap = base[i] + bytes - requested_addr;
2790        if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2791          if (PrintMiscellaneous && Verbose && bottom_overlap == 0) {
2792            warning("attempt_reserve_memory_at: possible alignment bug");
2793          }
2794          unmap_memory(requested_addr, bottom_overlap);
2795          size[i] = bytes - bottom_overlap;
2796        } else {
2797          size[i] = bytes;
2798        }
2799      }
2800    }
2801  }
2802
2803  // Give back the unused reserved pieces.
2804
2805  for (int j = 0; j < i; ++j) {
2806    if (base[j] != NULL) {
2807      unmap_memory(base[j], size[j]);
2808    }
2809  }
2810
2811  return (i < max_tries) ? requested_addr : NULL;
2812}
2813
2814bool os::pd_release_memory(char* addr, size_t bytes) {
2815  size_t size = bytes;
2816  return munmap(addr, size) == 0;
2817}
2818
2819static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
2820  assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
2821         "addr must be page aligned");
2822  int retVal = mprotect(addr, bytes, prot);
2823  return retVal == 0;
2824}
2825
2826// Protect memory (Used to pass readonly pages through
2827// JNI GetArray<type>Elements with empty arrays.)
2828// Also, used for serialization page and for compressed oops null pointer
2829// checking.
2830bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2831                        bool is_committed) {
2832  unsigned int p = 0;
2833  switch (prot) {
2834  case MEM_PROT_NONE: p = PROT_NONE; break;
2835  case MEM_PROT_READ: p = PROT_READ; break;
2836  case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2837  case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2838  default:
2839    ShouldNotReachHere();
2840  }
2841  // is_committed is unused.
2842  return solaris_mprotect(addr, bytes, p);
2843}
2844
2845// guard_memory and unguard_memory only happens within stack guard pages.
2846// Since ISM pertains only to the heap, guard and unguard memory should not
2847/// happen with an ISM region.
2848bool os::guard_memory(char* addr, size_t bytes) {
2849  return solaris_mprotect(addr, bytes, PROT_NONE);
2850}
2851
2852bool os::unguard_memory(char* addr, size_t bytes) {
2853  return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
2854}
2855
2856// Large page support
2857static size_t _large_page_size = 0;
2858
2859// Insertion sort for small arrays (descending order).
2860static void insertion_sort_descending(size_t* array, int len) {
2861  for (int i = 0; i < len; i++) {
2862    size_t val = array[i];
2863    for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
2864      size_t tmp = array[key];
2865      array[key] = array[key - 1];
2866      array[key - 1] = tmp;
2867    }
2868  }
2869}
2870
2871bool os::Solaris::mpss_sanity_check(bool warn, size_t* page_size) {
2872  const unsigned int usable_count = VM_Version::page_size_count();
2873  if (usable_count == 1) {
2874    return false;
2875  }
2876
2877  // Find the right getpagesizes interface.  When solaris 11 is the minimum
2878  // build platform, getpagesizes() (without the '2') can be called directly.
2879  typedef int (*gps_t)(size_t[], int);
2880  gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
2881  if (gps_func == NULL) {
2882    gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
2883    if (gps_func == NULL) {
2884      if (warn) {
2885        warning("MPSS is not supported by the operating system.");
2886      }
2887      return false;
2888    }
2889  }
2890
2891  // Fill the array of page sizes.
2892  int n = (*gps_func)(_page_sizes, page_sizes_max);
2893  assert(n > 0, "Solaris bug?");
2894
2895  if (n == page_sizes_max) {
2896    // Add a sentinel value (necessary only if the array was completely filled
2897    // since it is static (zeroed at initialization)).
2898    _page_sizes[--n] = 0;
2899    DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
2900  }
2901  assert(_page_sizes[n] == 0, "missing sentinel");
2902  trace_page_sizes("available page sizes", _page_sizes, n);
2903
2904  if (n == 1) return false;     // Only one page size available.
2905
2906  // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
2907  // select up to usable_count elements.  First sort the array, find the first
2908  // acceptable value, then copy the usable sizes to the top of the array and
2909  // trim the rest.  Make sure to include the default page size :-).
2910  //
2911  // A better policy could get rid of the 4M limit by taking the sizes of the
2912  // important VM memory regions (java heap and possibly the code cache) into
2913  // account.
2914  insertion_sort_descending(_page_sizes, n);
2915  const size_t size_limit =
2916    FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
2917  int beg;
2918  for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */;
2919  const int end = MIN2((int)usable_count, n) - 1;
2920  for (int cur = 0; cur < end; ++cur, ++beg) {
2921    _page_sizes[cur] = _page_sizes[beg];
2922  }
2923  _page_sizes[end] = vm_page_size();
2924  _page_sizes[end + 1] = 0;
2925
2926  if (_page_sizes[end] > _page_sizes[end - 1]) {
2927    // Default page size is not the smallest; sort again.
2928    insertion_sort_descending(_page_sizes, end + 1);
2929  }
2930  *page_size = _page_sizes[0];
2931
2932  trace_page_sizes("usable page sizes", _page_sizes, end + 1);
2933  return true;
2934}
2935
2936void os::large_page_init() {
2937  if (UseLargePages) {
2938    // print a warning if any large page related flag is specified on command line
2939    bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
2940                           !FLAG_IS_DEFAULT(LargePageSizeInBytes);
2941
2942    UseLargePages = Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
2943  }
2944}
2945
2946bool os::Solaris::is_valid_page_size(size_t bytes) {
2947  for (int i = 0; _page_sizes[i] != 0; i++) {
2948    if (_page_sizes[i] == bytes) {
2949      return true;
2950    }
2951  }
2952  return false;
2953}
2954
2955bool os::Solaris::setup_large_pages(caddr_t start, size_t bytes, size_t align) {
2956  assert(is_valid_page_size(align), SIZE_FORMAT " is not a valid page size", align);
2957  assert(is_ptr_aligned((void*) start, align),
2958         PTR_FORMAT " is not aligned to " SIZE_FORMAT, p2i((void*) start), align);
2959  assert(is_size_aligned(bytes, align),
2960         SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, align);
2961
2962  // Signal to OS that we want large pages for addresses
2963  // from addr, addr + bytes
2964  struct memcntl_mha mpss_struct;
2965  mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
2966  mpss_struct.mha_pagesize = align;
2967  mpss_struct.mha_flags = 0;
2968  // Upon successful completion, memcntl() returns 0
2969  if (memcntl(start, bytes, MC_HAT_ADVISE, (caddr_t) &mpss_struct, 0, 0)) {
2970    debug_only(warning("Attempt to use MPSS failed."));
2971    return false;
2972  }
2973  return true;
2974}
2975
2976char* os::reserve_memory_special(size_t size, size_t alignment, char* addr, bool exec) {
2977  fatal("os::reserve_memory_special should not be called on Solaris.");
2978  return NULL;
2979}
2980
2981bool os::release_memory_special(char* base, size_t bytes) {
2982  fatal("os::release_memory_special should not be called on Solaris.");
2983  return false;
2984}
2985
2986size_t os::large_page_size() {
2987  return _large_page_size;
2988}
2989
2990// MPSS allows application to commit large page memory on demand; with ISM
2991// the entire memory region must be allocated as shared memory.
2992bool os::can_commit_large_page_memory() {
2993  return true;
2994}
2995
2996bool os::can_execute_large_page_memory() {
2997  return true;
2998}
2999
3000// Read calls from inside the vm need to perform state transitions
3001size_t os::read(int fd, void *buf, unsigned int nBytes) {
3002  size_t res;
3003  JavaThread* thread = (JavaThread*)Thread::current();
3004  assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
3005  ThreadBlockInVM tbiv(thread);
3006  RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
3007  return res;
3008}
3009
3010size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3011  size_t res;
3012  JavaThread* thread = (JavaThread*)Thread::current();
3013  assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
3014  ThreadBlockInVM tbiv(thread);
3015  RESTARTABLE(::pread(fd, buf, (size_t) nBytes, offset), res);
3016  return res;
3017}
3018
3019size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
3020  size_t res;
3021  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
3022         "Assumed _thread_in_native");
3023  RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
3024  return res;
3025}
3026
3027void os::naked_short_sleep(jlong ms) {
3028  assert(ms < 1000, "Un-interruptable sleep, short time use only");
3029
3030  // usleep is deprecated and removed from POSIX, in favour of nanosleep, but
3031  // Solaris requires -lrt for this.
3032  usleep((ms * 1000));
3033
3034  return;
3035}
3036
3037// Sleep forever; naked call to OS-specific sleep; use with CAUTION
3038void os::infinite_sleep() {
3039  while (true) {    // sleep forever ...
3040    ::sleep(100);   // ... 100 seconds at a time
3041  }
3042}
3043
3044// Used to convert frequent JVM_Yield() to nops
3045bool os::dont_yield() {
3046  if (DontYieldALot) {
3047    static hrtime_t last_time = 0;
3048    hrtime_t diff = getTimeNanos() - last_time;
3049
3050    if (diff < DontYieldALotInterval * 1000000) {
3051      return true;
3052    }
3053
3054    last_time += diff;
3055
3056    return false;
3057  } else {
3058    return false;
3059  }
3060}
3061
3062// Note that yield semantics are defined by the scheduling class to which
3063// the thread currently belongs.  Typically, yield will _not yield to
3064// other equal or higher priority threads that reside on the dispatch queues
3065// of other CPUs.
3066
3067void os::naked_yield() {
3068  thr_yield();
3069}
3070
3071// Interface for setting lwp priorities.  If we are using T2 libthread,
3072// which forces the use of BoundThreads or we manually set UseBoundThreads,
3073// all of our threads will be assigned to real lwp's.  Using the thr_setprio
3074// function is meaningless in this mode so we must adjust the real lwp's priority
3075// The routines below implement the getting and setting of lwp priorities.
3076//
3077// Note: T2 is now the only supported libthread. UseBoundThreads flag is
3078//       being deprecated and all threads are now BoundThreads
3079//
3080// Note: There are three priority scales used on Solaris.  Java priotities
3081//       which range from 1 to 10, libthread "thr_setprio" scale which range
3082//       from 0 to 127, and the current scheduling class of the process we
3083//       are running in.  This is typically from -60 to +60.
3084//       The setting of the lwp priorities in done after a call to thr_setprio
3085//       so Java priorities are mapped to libthread priorities and we map from
3086//       the latter to lwp priorities.  We don't keep priorities stored in
3087//       Java priorities since some of our worker threads want to set priorities
3088//       higher than all Java threads.
3089//
3090// For related information:
3091// (1)  man -s 2 priocntl
3092// (2)  man -s 4 priocntl
3093// (3)  man dispadmin
3094// =    librt.so
3095// =    libthread/common/rtsched.c - thrp_setlwpprio().
3096// =    ps -cL <pid> ... to validate priority.
3097// =    sched_get_priority_min and _max
3098//              pthread_create
3099//              sched_setparam
3100//              pthread_setschedparam
3101//
3102// Assumptions:
3103// +    We assume that all threads in the process belong to the same
3104//              scheduling class.   IE. an homogenous process.
3105// +    Must be root or in IA group to change change "interactive" attribute.
3106//              Priocntl() will fail silently.  The only indication of failure is when
3107//              we read-back the value and notice that it hasn't changed.
3108// +    Interactive threads enter the runq at the head, non-interactive at the tail.
3109// +    For RT, change timeslice as well.  Invariant:
3110//              constant "priority integral"
3111//              Konst == TimeSlice * (60-Priority)
3112//              Given a priority, compute appropriate timeslice.
3113// +    Higher numerical values have higher priority.
3114
3115// sched class attributes
3116typedef struct {
3117  int   schedPolicy;              // classID
3118  int   maxPrio;
3119  int   minPrio;
3120} SchedInfo;
3121
3122
3123static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3124
3125#ifdef ASSERT
3126static int  ReadBackValidate = 1;
3127#endif
3128static int  myClass     = 0;
3129static int  myMin       = 0;
3130static int  myMax       = 0;
3131static int  myCur       = 0;
3132static bool priocntl_enable = false;
3133
3134static const int criticalPrio = FXCriticalPriority;
3135static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3136
3137
3138// lwp_priocntl_init
3139//
3140// Try to determine the priority scale for our process.
3141//
3142// Return errno or 0 if OK.
3143//
3144static int lwp_priocntl_init() {
3145  int rslt;
3146  pcinfo_t ClassInfo;
3147  pcparms_t ParmInfo;
3148  int i;
3149
3150  if (!UseThreadPriorities) return 0;
3151
3152  // If ThreadPriorityPolicy is 1, switch tables
3153  if (ThreadPriorityPolicy == 1) {
3154    for (i = 0; i < CriticalPriority+1; i++)
3155      os::java_to_os_priority[i] = prio_policy1[i];
3156  }
3157  if (UseCriticalJavaThreadPriority) {
3158    // MaxPriority always maps to the FX scheduling class and criticalPrio.
3159    // See set_native_priority() and set_lwp_class_and_priority().
3160    // Save original MaxPriority mapping in case attempt to
3161    // use critical priority fails.
3162    java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3163    // Set negative to distinguish from other priorities
3164    os::java_to_os_priority[MaxPriority] = -criticalPrio;
3165  }
3166
3167  // Get IDs for a set of well-known scheduling classes.
3168  // TODO-FIXME: GETCLINFO returns the current # of classes in the
3169  // the system.  We should have a loop that iterates over the
3170  // classID values, which are known to be "small" integers.
3171
3172  strcpy(ClassInfo.pc_clname, "TS");
3173  ClassInfo.pc_cid = -1;
3174  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3175  if (rslt < 0) return errno;
3176  assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3177  tsLimits.schedPolicy = ClassInfo.pc_cid;
3178  tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3179  tsLimits.minPrio = -tsLimits.maxPrio;
3180
3181  strcpy(ClassInfo.pc_clname, "IA");
3182  ClassInfo.pc_cid = -1;
3183  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3184  if (rslt < 0) return errno;
3185  assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3186  iaLimits.schedPolicy = ClassInfo.pc_cid;
3187  iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3188  iaLimits.minPrio = -iaLimits.maxPrio;
3189
3190  strcpy(ClassInfo.pc_clname, "RT");
3191  ClassInfo.pc_cid = -1;
3192  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3193  if (rslt < 0) return errno;
3194  assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3195  rtLimits.schedPolicy = ClassInfo.pc_cid;
3196  rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3197  rtLimits.minPrio = 0;
3198
3199  strcpy(ClassInfo.pc_clname, "FX");
3200  ClassInfo.pc_cid = -1;
3201  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3202  if (rslt < 0) return errno;
3203  assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3204  fxLimits.schedPolicy = ClassInfo.pc_cid;
3205  fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3206  fxLimits.minPrio = 0;
3207
3208  // Query our "current" scheduling class.
3209  // This will normally be IA, TS or, rarely, FX or RT.
3210  memset(&ParmInfo, 0, sizeof(ParmInfo));
3211  ParmInfo.pc_cid = PC_CLNULL;
3212  rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3213  if (rslt < 0) return errno;
3214  myClass = ParmInfo.pc_cid;
3215
3216  // We now know our scheduling classId, get specific information
3217  // about the class.
3218  ClassInfo.pc_cid = myClass;
3219  ClassInfo.pc_clname[0] = 0;
3220  rslt = priocntl((idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3221  if (rslt < 0) return errno;
3222
3223  if (ThreadPriorityVerbose) {
3224    tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3225  }
3226
3227  memset(&ParmInfo, 0, sizeof(pcparms_t));
3228  ParmInfo.pc_cid = PC_CLNULL;
3229  rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3230  if (rslt < 0) return errno;
3231
3232  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3233    myMin = rtLimits.minPrio;
3234    myMax = rtLimits.maxPrio;
3235  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3236    iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3237    myMin = iaLimits.minPrio;
3238    myMax = iaLimits.maxPrio;
3239    myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3240  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3241    tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3242    myMin = tsLimits.minPrio;
3243    myMax = tsLimits.maxPrio;
3244    myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3245  } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3246    fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3247    myMin = fxLimits.minPrio;
3248    myMax = fxLimits.maxPrio;
3249    myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3250  } else {
3251    // No clue - punt
3252    if (ThreadPriorityVerbose) {
3253      tty->print_cr("Unknown scheduling class: %s ... \n",
3254                    ClassInfo.pc_clname);
3255    }
3256    return EINVAL;      // no clue, punt
3257  }
3258
3259  if (ThreadPriorityVerbose) {
3260    tty->print_cr("Thread priority Range: [%d..%d]\n", myMin, myMax);
3261  }
3262
3263  priocntl_enable = true;  // Enable changing priorities
3264  return 0;
3265}
3266
3267#define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3268#define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3269#define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3270#define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3271
3272
3273// scale_to_lwp_priority
3274//
3275// Convert from the libthread "thr_setprio" scale to our current
3276// lwp scheduling class scale.
3277//
3278static int scale_to_lwp_priority(int rMin, int rMax, int x) {
3279  int v;
3280
3281  if (x == 127) return rMax;            // avoid round-down
3282  v = (((x*(rMax-rMin)))/128)+rMin;
3283  return v;
3284}
3285
3286
3287// set_lwp_class_and_priority
3288int set_lwp_class_and_priority(int ThreadID, int lwpid,
3289                               int newPrio, int new_class, bool scale) {
3290  int rslt;
3291  int Actual, Expected, prv;
3292  pcparms_t ParmInfo;                   // for GET-SET
3293#ifdef ASSERT
3294  pcparms_t ReadBack;                   // for readback
3295#endif
3296
3297  // Set priority via PC_GETPARMS, update, PC_SETPARMS
3298  // Query current values.
3299  // TODO: accelerate this by eliminating the PC_GETPARMS call.
3300  // Cache "pcparms_t" in global ParmCache.
3301  // TODO: elide set-to-same-value
3302
3303  // If something went wrong on init, don't change priorities.
3304  if (!priocntl_enable) {
3305    if (ThreadPriorityVerbose) {
3306      tty->print_cr("Trying to set priority but init failed, ignoring");
3307    }
3308    return EINVAL;
3309  }
3310
3311  // If lwp hasn't started yet, just return
3312  // the _start routine will call us again.
3313  if (lwpid <= 0) {
3314    if (ThreadPriorityVerbose) {
3315      tty->print_cr("deferring the set_lwp_class_and_priority of thread "
3316                    INTPTR_FORMAT " to %d, lwpid not set",
3317                    ThreadID, newPrio);
3318    }
3319    return 0;
3320  }
3321
3322  if (ThreadPriorityVerbose) {
3323    tty->print_cr ("set_lwp_class_and_priority("
3324                   INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3325                   ThreadID, lwpid, newPrio);
3326  }
3327
3328  memset(&ParmInfo, 0, sizeof(pcparms_t));
3329  ParmInfo.pc_cid = PC_CLNULL;
3330  rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3331  if (rslt < 0) return errno;
3332
3333  int cur_class = ParmInfo.pc_cid;
3334  ParmInfo.pc_cid = (id_t)new_class;
3335
3336  if (new_class == rtLimits.schedPolicy) {
3337    rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3338    rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3339                                                       rtLimits.maxPrio, newPrio)
3340                               : newPrio;
3341    rtInfo->rt_tqsecs  = RT_NOCHANGE;
3342    rtInfo->rt_tqnsecs = RT_NOCHANGE;
3343    if (ThreadPriorityVerbose) {
3344      tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3345    }
3346  } else if (new_class == iaLimits.schedPolicy) {
3347    iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3348    int maxClamped     = MIN2(iaLimits.maxPrio,
3349                              cur_class == new_class
3350                              ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3351    iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3352                                                       maxClamped, newPrio)
3353                               : newPrio;
3354    iaInfo->ia_uprilim = cur_class == new_class
3355                           ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3356    iaInfo->ia_mode    = IA_NOCHANGE;
3357    if (ThreadPriorityVerbose) {
3358      tty->print_cr("IA: [%d...%d] %d->%d\n",
3359                    iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3360    }
3361  } else if (new_class == tsLimits.schedPolicy) {
3362    tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3363    int maxClamped     = MIN2(tsLimits.maxPrio,
3364                              cur_class == new_class
3365                              ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3366    tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3367                                                       maxClamped, newPrio)
3368                               : newPrio;
3369    tsInfo->ts_uprilim = cur_class == new_class
3370                           ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
3371    if (ThreadPriorityVerbose) {
3372      tty->print_cr("TS: [%d...%d] %d->%d\n",
3373                    tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3374    }
3375  } else if (new_class == fxLimits.schedPolicy) {
3376    fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
3377    int maxClamped     = MIN2(fxLimits.maxPrio,
3378                              cur_class == new_class
3379                              ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
3380    fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
3381                                                       maxClamped, newPrio)
3382                               : newPrio;
3383    fxInfo->fx_uprilim = cur_class == new_class
3384                           ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
3385    fxInfo->fx_tqsecs  = FX_NOCHANGE;
3386    fxInfo->fx_tqnsecs = FX_NOCHANGE;
3387    if (ThreadPriorityVerbose) {
3388      tty->print_cr("FX: [%d...%d] %d->%d\n",
3389                    fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
3390    }
3391  } else {
3392    if (ThreadPriorityVerbose) {
3393      tty->print_cr("Unknown new scheduling class %d\n", new_class);
3394    }
3395    return EINVAL;    // no clue, punt
3396  }
3397
3398  rslt = priocntl(P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3399  if (ThreadPriorityVerbose && rslt) {
3400    tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3401  }
3402  if (rslt < 0) return errno;
3403
3404#ifdef ASSERT
3405  // Sanity check: read back what we just attempted to set.
3406  // In theory it could have changed in the interim ...
3407  //
3408  // The priocntl system call is tricky.
3409  // Sometimes it'll validate the priority value argument and
3410  // return EINVAL if unhappy.  At other times it fails silently.
3411  // Readbacks are prudent.
3412
3413  if (!ReadBackValidate) return 0;
3414
3415  memset(&ReadBack, 0, sizeof(pcparms_t));
3416  ReadBack.pc_cid = PC_CLNULL;
3417  rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3418  assert(rslt >= 0, "priocntl failed");
3419  Actual = Expected = 0xBAD;
3420  assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3421  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3422    Actual   = RTPRI(ReadBack)->rt_pri;
3423    Expected = RTPRI(ParmInfo)->rt_pri;
3424  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3425    Actual   = IAPRI(ReadBack)->ia_upri;
3426    Expected = IAPRI(ParmInfo)->ia_upri;
3427  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3428    Actual   = TSPRI(ReadBack)->ts_upri;
3429    Expected = TSPRI(ParmInfo)->ts_upri;
3430  } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3431    Actual   = FXPRI(ReadBack)->fx_upri;
3432    Expected = FXPRI(ParmInfo)->fx_upri;
3433  } else {
3434    if (ThreadPriorityVerbose) {
3435      tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
3436                    ParmInfo.pc_cid);
3437    }
3438  }
3439
3440  if (Actual != Expected) {
3441    if (ThreadPriorityVerbose) {
3442      tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3443                     lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3444    }
3445  }
3446#endif
3447
3448  return 0;
3449}
3450
3451// Solaris only gives access to 128 real priorities at a time,
3452// so we expand Java's ten to fill this range.  This would be better
3453// if we dynamically adjusted relative priorities.
3454//
3455// The ThreadPriorityPolicy option allows us to select 2 different
3456// priority scales.
3457//
3458// ThreadPriorityPolicy=0
3459// Since the Solaris' default priority is MaximumPriority, we do not
3460// set a priority lower than Max unless a priority lower than
3461// NormPriority is requested.
3462//
3463// ThreadPriorityPolicy=1
3464// This mode causes the priority table to get filled with
3465// linear values.  NormPriority get's mapped to 50% of the
3466// Maximum priority an so on.  This will cause VM threads
3467// to get unfair treatment against other Solaris processes
3468// which do not explicitly alter their thread priorities.
3469
3470int os::java_to_os_priority[CriticalPriority + 1] = {
3471  -99999,         // 0 Entry should never be used
3472
3473  0,              // 1 MinPriority
3474  32,             // 2
3475  64,             // 3
3476
3477  96,             // 4
3478  127,            // 5 NormPriority
3479  127,            // 6
3480
3481  127,            // 7
3482  127,            // 8
3483  127,            // 9 NearMaxPriority
3484
3485  127,            // 10 MaxPriority
3486
3487  -criticalPrio   // 11 CriticalPriority
3488};
3489
3490OSReturn os::set_native_priority(Thread* thread, int newpri) {
3491  OSThread* osthread = thread->osthread();
3492
3493  // Save requested priority in case the thread hasn't been started
3494  osthread->set_native_priority(newpri);
3495
3496  // Check for critical priority request
3497  bool fxcritical = false;
3498  if (newpri == -criticalPrio) {
3499    fxcritical = true;
3500    newpri = criticalPrio;
3501  }
3502
3503  assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3504  if (!UseThreadPriorities) return OS_OK;
3505
3506  int status = 0;
3507
3508  if (!fxcritical) {
3509    // Use thr_setprio only if we have a priority that thr_setprio understands
3510    status = thr_setprio(thread->osthread()->thread_id(), newpri);
3511  }
3512
3513  int lwp_status =
3514          set_lwp_class_and_priority(osthread->thread_id(),
3515                                     osthread->lwp_id(),
3516                                     newpri,
3517                                     fxcritical ? fxLimits.schedPolicy : myClass,
3518                                     !fxcritical);
3519  if (lwp_status != 0 && fxcritical) {
3520    // Try again, this time without changing the scheduling class
3521    newpri = java_MaxPriority_to_os_priority;
3522    lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
3523                                            osthread->lwp_id(),
3524                                            newpri, myClass, false);
3525  }
3526  status |= lwp_status;
3527  return (status == 0) ? OS_OK : OS_ERR;
3528}
3529
3530
3531OSReturn os::get_native_priority(const Thread* const thread,
3532                                 int *priority_ptr) {
3533  int p;
3534  if (!UseThreadPriorities) {
3535    *priority_ptr = NormalPriority;
3536    return OS_OK;
3537  }
3538  int status = thr_getprio(thread->osthread()->thread_id(), &p);
3539  if (status != 0) {
3540    return OS_ERR;
3541  }
3542  *priority_ptr = p;
3543  return OS_OK;
3544}
3545
3546
3547// Hint to the underlying OS that a task switch would not be good.
3548// Void return because it's a hint and can fail.
3549void os::hint_no_preempt() {
3550  schedctl_start(schedctl_init());
3551}
3552
3553static void resume_clear_context(OSThread *osthread) {
3554  osthread->set_ucontext(NULL);
3555}
3556
3557static void suspend_save_context(OSThread *osthread, ucontext_t* context) {
3558  osthread->set_ucontext(context);
3559}
3560
3561static PosixSemaphore sr_semaphore;
3562
3563void os::Solaris::SR_handler(Thread* thread, ucontext_t* uc) {
3564  // Save and restore errno to avoid confusing native code with EINTR
3565  // after sigsuspend.
3566  int old_errno = errno;
3567
3568  OSThread* osthread = thread->osthread();
3569  assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3570
3571  os::SuspendResume::State current = osthread->sr.state();
3572  if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3573    suspend_save_context(osthread, uc);
3574
3575    // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3576    os::SuspendResume::State state = osthread->sr.suspended();
3577    if (state == os::SuspendResume::SR_SUSPENDED) {
3578      sigset_t suspend_set;  // signals for sigsuspend()
3579
3580      // get current set of blocked signals and unblock resume signal
3581      thr_sigsetmask(SIG_BLOCK, NULL, &suspend_set);
3582      sigdelset(&suspend_set, os::Solaris::SIGasync());
3583
3584      sr_semaphore.signal();
3585      // wait here until we are resumed
3586      while (1) {
3587        sigsuspend(&suspend_set);
3588
3589        os::SuspendResume::State result = osthread->sr.running();
3590        if (result == os::SuspendResume::SR_RUNNING) {
3591          sr_semaphore.signal();
3592          break;
3593        }
3594      }
3595
3596    } else if (state == os::SuspendResume::SR_RUNNING) {
3597      // request was cancelled, continue
3598    } else {
3599      ShouldNotReachHere();
3600    }
3601
3602    resume_clear_context(osthread);
3603  } else if (current == os::SuspendResume::SR_RUNNING) {
3604    // request was cancelled, continue
3605  } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3606    // ignore
3607  } else {
3608    // ignore
3609  }
3610
3611  errno = old_errno;
3612}
3613
3614void os::print_statistics() {
3615}
3616
3617bool os::message_box(const char* title, const char* message) {
3618  int i;
3619  fdStream err(defaultStream::error_fd());
3620  for (i = 0; i < 78; i++) err.print_raw("=");
3621  err.cr();
3622  err.print_raw_cr(title);
3623  for (i = 0; i < 78; i++) err.print_raw("-");
3624  err.cr();
3625  err.print_raw_cr(message);
3626  for (i = 0; i < 78; i++) err.print_raw("=");
3627  err.cr();
3628
3629  char buf[16];
3630  // Prevent process from exiting upon "read error" without consuming all CPU
3631  while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3632
3633  return buf[0] == 'y' || buf[0] == 'Y';
3634}
3635
3636static int sr_notify(OSThread* osthread) {
3637  int status = thr_kill(osthread->thread_id(), os::Solaris::SIGasync());
3638  assert_status(status == 0, status, "thr_kill");
3639  return status;
3640}
3641
3642// "Randomly" selected value for how long we want to spin
3643// before bailing out on suspending a thread, also how often
3644// we send a signal to a thread we want to resume
3645static const int RANDOMLY_LARGE_INTEGER = 1000000;
3646static const int RANDOMLY_LARGE_INTEGER2 = 100;
3647
3648static bool do_suspend(OSThread* osthread) {
3649  assert(osthread->sr.is_running(), "thread should be running");
3650  assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3651
3652  // mark as suspended and send signal
3653  if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3654    // failed to switch, state wasn't running?
3655    ShouldNotReachHere();
3656    return false;
3657  }
3658
3659  if (sr_notify(osthread) != 0) {
3660    ShouldNotReachHere();
3661  }
3662
3663  // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3664  while (true) {
3665    if (sr_semaphore.timedwait(0, 2000 * NANOSECS_PER_MILLISEC)) {
3666      break;
3667    } else {
3668      // timeout
3669      os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3670      if (cancelled == os::SuspendResume::SR_RUNNING) {
3671        return false;
3672      } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3673        // make sure that we consume the signal on the semaphore as well
3674        sr_semaphore.wait();
3675        break;
3676      } else {
3677        ShouldNotReachHere();
3678        return false;
3679      }
3680    }
3681  }
3682
3683  guarantee(osthread->sr.is_suspended(), "Must be suspended");
3684  return true;
3685}
3686
3687static void do_resume(OSThread* osthread) {
3688  assert(osthread->sr.is_suspended(), "thread should be suspended");
3689  assert(!sr_semaphore.trywait(), "invalid semaphore state");
3690
3691  if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3692    // failed to switch to WAKEUP_REQUEST
3693    ShouldNotReachHere();
3694    return;
3695  }
3696
3697  while (true) {
3698    if (sr_notify(osthread) == 0) {
3699      if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3700        if (osthread->sr.is_running()) {
3701          return;
3702        }
3703      }
3704    } else {
3705      ShouldNotReachHere();
3706    }
3707  }
3708
3709  guarantee(osthread->sr.is_running(), "Must be running!");
3710}
3711
3712void os::SuspendedThreadTask::internal_do_task() {
3713  if (do_suspend(_thread->osthread())) {
3714    SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3715    do_task(context);
3716    do_resume(_thread->osthread());
3717  }
3718}
3719
3720class PcFetcher : public os::SuspendedThreadTask {
3721 public:
3722  PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
3723  ExtendedPC result();
3724 protected:
3725  void do_task(const os::SuspendedThreadTaskContext& context);
3726 private:
3727  ExtendedPC _epc;
3728};
3729
3730ExtendedPC PcFetcher::result() {
3731  guarantee(is_done(), "task is not done yet.");
3732  return _epc;
3733}
3734
3735void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
3736  Thread* thread = context.thread();
3737  OSThread* osthread = thread->osthread();
3738  if (osthread->ucontext() != NULL) {
3739    _epc = os::Solaris::ucontext_get_pc((ucontext_t *) context.ucontext());
3740  } else {
3741    // NULL context is unexpected, double-check this is the VMThread
3742    guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3743  }
3744}
3745
3746// A lightweight implementation that does not suspend the target thread and
3747// thus returns only a hint. Used for profiling only!
3748ExtendedPC os::get_thread_pc(Thread* thread) {
3749  // Make sure that it is called by the watcher and the Threads lock is owned.
3750  assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
3751  // For now, is only used to profile the VM Thread
3752  assert(thread->is_VM_thread(), "Can only be called for VMThread");
3753  PcFetcher fetcher(thread);
3754  fetcher.run();
3755  return fetcher.result();
3756}
3757
3758
3759// This does not do anything on Solaris. This is basically a hook for being
3760// able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
3761void os::os_exception_wrapper(java_call_t f, JavaValue* value,
3762                              const methodHandle& method, JavaCallArguments* args,
3763                              Thread* thread) {
3764  f(value, method, args, thread);
3765}
3766
3767// This routine may be used by user applications as a "hook" to catch signals.
3768// The user-defined signal handler must pass unrecognized signals to this
3769// routine, and if it returns true (non-zero), then the signal handler must
3770// return immediately.  If the flag "abort_if_unrecognized" is true, then this
3771// routine will never retun false (zero), but instead will execute a VM panic
3772// routine kill the process.
3773//
3774// If this routine returns false, it is OK to call it again.  This allows
3775// the user-defined signal handler to perform checks either before or after
3776// the VM performs its own checks.  Naturally, the user code would be making
3777// a serious error if it tried to handle an exception (such as a null check
3778// or breakpoint) that the VM was generating for its own correct operation.
3779//
3780// This routine may recognize any of the following kinds of signals:
3781// SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
3782// os::Solaris::SIGasync
3783// It should be consulted by handlers for any of those signals.
3784//
3785// The caller of this routine must pass in the three arguments supplied
3786// to the function referred to in the "sa_sigaction" (not the "sa_handler")
3787// field of the structure passed to sigaction().  This routine assumes that
3788// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3789//
3790// Note that the VM will print warnings if it detects conflicting signal
3791// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3792//
3793extern "C" JNIEXPORT int JVM_handle_solaris_signal(int signo,
3794                                                   siginfo_t* siginfo,
3795                                                   void* ucontext,
3796                                                   int abort_if_unrecognized);
3797
3798
3799void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
3800  int orig_errno = errno;  // Preserve errno value over signal handler.
3801  JVM_handle_solaris_signal(sig, info, ucVoid, true);
3802  errno = orig_errno;
3803}
3804
3805// This boolean allows users to forward their own non-matching signals
3806// to JVM_handle_solaris_signal, harmlessly.
3807bool os::Solaris::signal_handlers_are_installed = false;
3808
3809// For signal-chaining
3810bool os::Solaris::libjsig_is_loaded = false;
3811typedef struct sigaction *(*get_signal_t)(int);
3812get_signal_t os::Solaris::get_signal_action = NULL;
3813
3814struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
3815  struct sigaction *actp = NULL;
3816
3817  if ((libjsig_is_loaded)  && (sig <= Maxlibjsigsigs)) {
3818    // Retrieve the old signal handler from libjsig
3819    actp = (*get_signal_action)(sig);
3820  }
3821  if (actp == NULL) {
3822    // Retrieve the preinstalled signal handler from jvm
3823    actp = get_preinstalled_handler(sig);
3824  }
3825
3826  return actp;
3827}
3828
3829static bool call_chained_handler(struct sigaction *actp, int sig,
3830                                 siginfo_t *siginfo, void *context) {
3831  // Call the old signal handler
3832  if (actp->sa_handler == SIG_DFL) {
3833    // It's more reasonable to let jvm treat it as an unexpected exception
3834    // instead of taking the default action.
3835    return false;
3836  } else if (actp->sa_handler != SIG_IGN) {
3837    if ((actp->sa_flags & SA_NODEFER) == 0) {
3838      // automaticlly block the signal
3839      sigaddset(&(actp->sa_mask), sig);
3840    }
3841
3842    sa_handler_t hand;
3843    sa_sigaction_t sa;
3844    bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3845    // retrieve the chained handler
3846    if (siginfo_flag_set) {
3847      sa = actp->sa_sigaction;
3848    } else {
3849      hand = actp->sa_handler;
3850    }
3851
3852    if ((actp->sa_flags & SA_RESETHAND) != 0) {
3853      actp->sa_handler = SIG_DFL;
3854    }
3855
3856    // try to honor the signal mask
3857    sigset_t oset;
3858    thr_sigsetmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3859
3860    // call into the chained handler
3861    if (siginfo_flag_set) {
3862      (*sa)(sig, siginfo, context);
3863    } else {
3864      (*hand)(sig);
3865    }
3866
3867    // restore the signal mask
3868    thr_sigsetmask(SIG_SETMASK, &oset, 0);
3869  }
3870  // Tell jvm's signal handler the signal is taken care of.
3871  return true;
3872}
3873
3874bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3875  bool chained = false;
3876  // signal-chaining
3877  if (UseSignalChaining) {
3878    struct sigaction *actp = get_chained_signal_action(sig);
3879    if (actp != NULL) {
3880      chained = call_chained_handler(actp, sig, siginfo, context);
3881    }
3882  }
3883  return chained;
3884}
3885
3886struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
3887  assert((chainedsigactions != (struct sigaction *)NULL) &&
3888         (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3889  if (preinstalled_sigs[sig] != 0) {
3890    return &chainedsigactions[sig];
3891  }
3892  return NULL;
3893}
3894
3895void os::Solaris::save_preinstalled_handler(int sig,
3896                                            struct sigaction& oldAct) {
3897  assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
3898  assert((chainedsigactions != (struct sigaction *)NULL) &&
3899         (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3900  chainedsigactions[sig] = oldAct;
3901  preinstalled_sigs[sig] = 1;
3902}
3903
3904void os::Solaris::set_signal_handler(int sig, bool set_installed,
3905                                     bool oktochain) {
3906  // Check for overwrite.
3907  struct sigaction oldAct;
3908  sigaction(sig, (struct sigaction*)NULL, &oldAct);
3909  void* oldhand =
3910      oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3911                          : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3912  if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3913      oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3914      oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
3915    if (AllowUserSignalHandlers || !set_installed) {
3916      // Do not overwrite; user takes responsibility to forward to us.
3917      return;
3918    } else if (UseSignalChaining) {
3919      if (oktochain) {
3920        // save the old handler in jvm
3921        save_preinstalled_handler(sig, oldAct);
3922      } else {
3923        vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal.");
3924      }
3925      // libjsig also interposes the sigaction() call below and saves the
3926      // old sigaction on it own.
3927    } else {
3928      fatal("Encountered unexpected pre-existing sigaction handler "
3929            "%#lx for signal %d.", (long)oldhand, sig);
3930    }
3931  }
3932
3933  struct sigaction sigAct;
3934  sigfillset(&(sigAct.sa_mask));
3935  sigAct.sa_handler = SIG_DFL;
3936
3937  sigAct.sa_sigaction = signalHandler;
3938  // Handle SIGSEGV on alternate signal stack if
3939  // not using stack banging
3940  if (!UseStackBanging && sig == SIGSEGV) {
3941    sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
3942  } else {
3943    sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
3944  }
3945  os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
3946
3947  sigaction(sig, &sigAct, &oldAct);
3948
3949  void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3950                                       : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3951  assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3952}
3953
3954
3955#define DO_SIGNAL_CHECK(sig)                      \
3956  do {                                            \
3957    if (!sigismember(&check_signal_done, sig)) {  \
3958      os::Solaris::check_signal_handler(sig);     \
3959    }                                             \
3960  } while (0)
3961
3962// This method is a periodic task to check for misbehaving JNI applications
3963// under CheckJNI, we can add any periodic checks here
3964
3965void os::run_periodic_checks() {
3966  // A big source of grief is hijacking virt. addr 0x0 on Solaris,
3967  // thereby preventing a NULL checks.
3968  if (!check_addr0_done) check_addr0_done = check_addr0(tty);
3969
3970  if (check_signals == false) return;
3971
3972  // SEGV and BUS if overridden could potentially prevent
3973  // generation of hs*.log in the event of a crash, debugging
3974  // such a case can be very challenging, so we absolutely
3975  // check for the following for a good measure:
3976  DO_SIGNAL_CHECK(SIGSEGV);
3977  DO_SIGNAL_CHECK(SIGILL);
3978  DO_SIGNAL_CHECK(SIGFPE);
3979  DO_SIGNAL_CHECK(SIGBUS);
3980  DO_SIGNAL_CHECK(SIGPIPE);
3981  DO_SIGNAL_CHECK(SIGXFSZ);
3982
3983  // ReduceSignalUsage allows the user to override these handlers
3984  // see comments at the very top and jvm_solaris.h
3985  if (!ReduceSignalUsage) {
3986    DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3987    DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3988    DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3989    DO_SIGNAL_CHECK(BREAK_SIGNAL);
3990  }
3991
3992  // See comments above for using JVM1/JVM2
3993  DO_SIGNAL_CHECK(os::Solaris::SIGasync());
3994
3995}
3996
3997typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3998
3999static os_sigaction_t os_sigaction = NULL;
4000
4001void os::Solaris::check_signal_handler(int sig) {
4002  char buf[O_BUFLEN];
4003  address jvmHandler = NULL;
4004
4005  struct sigaction act;
4006  if (os_sigaction == NULL) {
4007    // only trust the default sigaction, in case it has been interposed
4008    os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4009    if (os_sigaction == NULL) return;
4010  }
4011
4012  os_sigaction(sig, (struct sigaction*)NULL, &act);
4013
4014  address thisHandler = (act.sa_flags & SA_SIGINFO)
4015    ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4016    : CAST_FROM_FN_PTR(address, act.sa_handler);
4017
4018
4019  switch (sig) {
4020  case SIGSEGV:
4021  case SIGBUS:
4022  case SIGFPE:
4023  case SIGPIPE:
4024  case SIGXFSZ:
4025  case SIGILL:
4026    jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4027    break;
4028
4029  case SHUTDOWN1_SIGNAL:
4030  case SHUTDOWN2_SIGNAL:
4031  case SHUTDOWN3_SIGNAL:
4032  case BREAK_SIGNAL:
4033    jvmHandler = (address)user_handler();
4034    break;
4035
4036  default:
4037    int asynsig = os::Solaris::SIGasync();
4038
4039    if (sig == asynsig) {
4040      jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4041    } else {
4042      return;
4043    }
4044    break;
4045  }
4046
4047
4048  if (thisHandler != jvmHandler) {
4049    tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4050    tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4051    tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4052    // No need to check this sig any longer
4053    sigaddset(&check_signal_done, sig);
4054    // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4055    if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4056      tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4057                    exception_name(sig, buf, O_BUFLEN));
4058    }
4059  } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4060    tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4061    tty->print("expected:");
4062    os::Posix::print_sa_flags(tty, os::Solaris::get_our_sigflags(sig));
4063    tty->cr();
4064    tty->print("  found:");
4065    os::Posix::print_sa_flags(tty, act.sa_flags);
4066    tty->cr();
4067    // No need to check this sig any longer
4068    sigaddset(&check_signal_done, sig);
4069  }
4070
4071  // Print all the signal handler state
4072  if (sigismember(&check_signal_done, sig)) {
4073    print_signal_handlers(tty, buf, O_BUFLEN);
4074  }
4075
4076}
4077
4078void os::Solaris::install_signal_handlers() {
4079  bool libjsigdone = false;
4080  signal_handlers_are_installed = true;
4081
4082  // signal-chaining
4083  typedef void (*signal_setting_t)();
4084  signal_setting_t begin_signal_setting = NULL;
4085  signal_setting_t end_signal_setting = NULL;
4086  begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4087                                        dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4088  if (begin_signal_setting != NULL) {
4089    end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4090                                        dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4091    get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4092                                       dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4093    get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4094                                         dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4095    libjsig_is_loaded = true;
4096    if (os::Solaris::get_libjsig_version != NULL) {
4097      libjsigversion =  (*os::Solaris::get_libjsig_version)();
4098    }
4099    assert(UseSignalChaining, "should enable signal-chaining");
4100  }
4101  if (libjsig_is_loaded) {
4102    // Tell libjsig jvm is setting signal handlers
4103    (*begin_signal_setting)();
4104  }
4105
4106  set_signal_handler(SIGSEGV, true, true);
4107  set_signal_handler(SIGPIPE, true, true);
4108  set_signal_handler(SIGXFSZ, true, true);
4109  set_signal_handler(SIGBUS, true, true);
4110  set_signal_handler(SIGILL, true, true);
4111  set_signal_handler(SIGFPE, true, true);
4112
4113
4114  if (os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4115    // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4116    // can not register overridable signals which might be > 32
4117    if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4118      // Tell libjsig jvm has finished setting signal handlers
4119      (*end_signal_setting)();
4120      libjsigdone = true;
4121    }
4122  }
4123
4124  set_signal_handler(os::Solaris::SIGasync(), true, true);
4125
4126  if (libjsig_is_loaded && !libjsigdone) {
4127    // Tell libjsig jvm finishes setting signal handlers
4128    (*end_signal_setting)();
4129  }
4130
4131  // We don't activate signal checker if libjsig is in place, we trust ourselves
4132  // and if UserSignalHandler is installed all bets are off.
4133  // Log that signal checking is off only if -verbose:jni is specified.
4134  if (CheckJNICalls) {
4135    if (libjsig_is_loaded) {
4136      if (PrintJNIResolving) {
4137        tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4138      }
4139      check_signals = false;
4140    }
4141    if (AllowUserSignalHandlers) {
4142      if (PrintJNIResolving) {
4143        tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4144      }
4145      check_signals = false;
4146    }
4147  }
4148}
4149
4150
4151void report_error(const char* file_name, int line_no, const char* title,
4152                  const char* format, ...);
4153
4154// (Static) wrapper for getisax(2) call.
4155os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4156
4157// (Static) wrappers for the liblgrp API
4158os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4159os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4160os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4161os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4162os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4163os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4164os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4165os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4166os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4167
4168// (Static) wrapper for meminfo() call.
4169os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4170
4171static address resolve_symbol_lazy(const char* name) {
4172  address addr = (address) dlsym(RTLD_DEFAULT, name);
4173  if (addr == NULL) {
4174    // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4175    addr = (address) dlsym(RTLD_NEXT, name);
4176  }
4177  return addr;
4178}
4179
4180static address resolve_symbol(const char* name) {
4181  address addr = resolve_symbol_lazy(name);
4182  if (addr == NULL) {
4183    fatal(dlerror());
4184  }
4185  return addr;
4186}
4187
4188void os::Solaris::libthread_init() {
4189  address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4190
4191  lwp_priocntl_init();
4192
4193  // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4194  if (func == NULL) {
4195    func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4196    // Guarantee that this VM is running on an new enough OS (5.6 or
4197    // later) that it will have a new enough libthread.so.
4198    guarantee(func != NULL, "libthread.so is too old.");
4199  }
4200
4201  int size;
4202  void (*handler_info_func)(address *, int *);
4203  handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4204  handler_info_func(&handler_start, &size);
4205  handler_end = handler_start + size;
4206}
4207
4208
4209int_fnP_mutex_tP os::Solaris::_mutex_lock;
4210int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4211int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4212int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4213int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4214int os::Solaris::_mutex_scope = USYNC_THREAD;
4215
4216int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4217int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4218int_fnP_cond_tP os::Solaris::_cond_signal;
4219int_fnP_cond_tP os::Solaris::_cond_broadcast;
4220int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4221int_fnP_cond_tP os::Solaris::_cond_destroy;
4222int os::Solaris::_cond_scope = USYNC_THREAD;
4223
4224void os::Solaris::synchronization_init() {
4225  if (UseLWPSynchronization) {
4226    os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4227    os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4228    os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4229    os::Solaris::set_mutex_init(lwp_mutex_init);
4230    os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4231    os::Solaris::set_mutex_scope(USYNC_THREAD);
4232
4233    os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4234    os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4235    os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4236    os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4237    os::Solaris::set_cond_init(lwp_cond_init);
4238    os::Solaris::set_cond_destroy(lwp_cond_destroy);
4239    os::Solaris::set_cond_scope(USYNC_THREAD);
4240  } else {
4241    os::Solaris::set_mutex_scope(USYNC_THREAD);
4242    os::Solaris::set_cond_scope(USYNC_THREAD);
4243
4244    if (UsePthreads) {
4245      os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4246      os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4247      os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4248      os::Solaris::set_mutex_init(pthread_mutex_default_init);
4249      os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4250
4251      os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4252      os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4253      os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4254      os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4255      os::Solaris::set_cond_init(pthread_cond_default_init);
4256      os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4257    } else {
4258      os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4259      os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4260      os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4261      os::Solaris::set_mutex_init(::mutex_init);
4262      os::Solaris::set_mutex_destroy(::mutex_destroy);
4263
4264      os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4265      os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4266      os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4267      os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4268      os::Solaris::set_cond_init(::cond_init);
4269      os::Solaris::set_cond_destroy(::cond_destroy);
4270    }
4271  }
4272}
4273
4274bool os::Solaris::liblgrp_init() {
4275  void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4276  if (handle != NULL) {
4277    os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4278    os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4279    os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4280    os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4281    os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4282    os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4283    os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4284    os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4285                                                      dlsym(handle, "lgrp_cookie_stale")));
4286
4287    lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4288    set_lgrp_cookie(c);
4289    return true;
4290  }
4291  return false;
4292}
4293
4294void os::Solaris::misc_sym_init() {
4295  address func;
4296
4297  // getisax
4298  func = resolve_symbol_lazy("getisax");
4299  if (func != NULL) {
4300    os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
4301  }
4302
4303  // meminfo
4304  func = resolve_symbol_lazy("meminfo");
4305  if (func != NULL) {
4306    os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
4307  }
4308}
4309
4310uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
4311  assert(_getisax != NULL, "_getisax not set");
4312  return _getisax(array, n);
4313}
4314
4315// int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4316typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4317static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4318
4319void init_pset_getloadavg_ptr(void) {
4320  pset_getloadavg_ptr =
4321    (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4322  if (PrintMiscellaneous && Verbose && pset_getloadavg_ptr == NULL) {
4323    warning("pset_getloadavg function not found");
4324  }
4325}
4326
4327int os::Solaris::_dev_zero_fd = -1;
4328
4329// this is called _before_ the global arguments have been parsed
4330void os::init(void) {
4331  _initial_pid = getpid();
4332
4333  max_hrtime = first_hrtime = gethrtime();
4334
4335  init_random(1234567);
4336
4337  page_size = sysconf(_SC_PAGESIZE);
4338  if (page_size == -1) {
4339    fatal("os_solaris.cpp: os::init: sysconf failed (%s)", strerror(errno));
4340  }
4341  init_page_sizes((size_t) page_size);
4342
4343  Solaris::initialize_system_info();
4344
4345  // Initialize misc. symbols as soon as possible, so we can use them
4346  // if we need them.
4347  Solaris::misc_sym_init();
4348
4349  int fd = ::open("/dev/zero", O_RDWR);
4350  if (fd < 0) {
4351    fatal("os::init: cannot open /dev/zero (%s)", strerror(errno));
4352  } else {
4353    Solaris::set_dev_zero_fd(fd);
4354
4355    // Close on exec, child won't inherit.
4356    fcntl(fd, F_SETFD, FD_CLOEXEC);
4357  }
4358
4359  clock_tics_per_sec = CLK_TCK;
4360
4361  // check if dladdr1() exists; dladdr1 can provide more information than
4362  // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4363  // and is available on linker patches for 5.7 and 5.8.
4364  // libdl.so must have been loaded, this call is just an entry lookup
4365  void * hdl = dlopen("libdl.so", RTLD_NOW);
4366  if (hdl) {
4367    dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4368  }
4369
4370  // (Solaris only) this switches to calls that actually do locking.
4371  ThreadCritical::initialize();
4372
4373  main_thread = thr_self();
4374
4375  // Constant minimum stack size allowed. It must be at least
4376  // the minimum of what the OS supports (thr_min_stack()), and
4377  // enough to allow the thread to get to user bytecode execution.
4378  Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
4379  // If the pagesize of the VM is greater than 8K determine the appropriate
4380  // number of initial guard pages.  The user can change this with the
4381  // command line arguments, if needed.
4382  if (vm_page_size() > 8*K) {
4383    StackYellowPages = 1;
4384    StackRedPages = 1;
4385    StackReservedPages = 1;
4386    StackShadowPages = round_to((StackShadowPages*8*K), vm_page_size()) / vm_page_size();
4387  }
4388}
4389
4390// To install functions for atexit system call
4391extern "C" {
4392  static void perfMemory_exit_helper() {
4393    perfMemory_exit();
4394  }
4395}
4396
4397// this is called _after_ the global arguments have been parsed
4398jint os::init_2(void) {
4399  // try to enable extended file IO ASAP, see 6431278
4400  os::Solaris::try_enable_extended_io();
4401
4402  // Allocate a single page and mark it as readable for safepoint polling.  Also
4403  // use this first mmap call to check support for MAP_ALIGN.
4404  address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
4405                                                      page_size,
4406                                                      MAP_PRIVATE | MAP_ALIGN,
4407                                                      PROT_READ);
4408  if (polling_page == NULL) {
4409    has_map_align = false;
4410    polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
4411                                                PROT_READ);
4412  }
4413
4414  os::set_polling_page(polling_page);
4415
4416#ifndef PRODUCT
4417  if (Verbose && PrintMiscellaneous) {
4418    tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4419               (intptr_t)polling_page);
4420  }
4421#endif
4422
4423  if (!UseMembar) {
4424    address mem_serialize_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE);
4425    guarantee(mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4426    os::set_memory_serialize_page(mem_serialize_page);
4427
4428#ifndef PRODUCT
4429    if (Verbose && PrintMiscellaneous) {
4430      tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n",
4431                 (intptr_t)mem_serialize_page);
4432    }
4433#endif
4434  }
4435
4436  // Check minimum allowable stack size for thread creation and to initialize
4437  // the java system classes, including StackOverflowError - depends on page
4438  // size.  Add a page for compiler2 recursion in main thread.
4439  // Add in 2*BytesPerWord times page size to account for VM stack during
4440  // class initialization depending on 32 or 64 bit VM.
4441  os::Solaris::min_stack_allowed = MAX2(os::Solaris::min_stack_allowed,
4442                                        (size_t)(StackReservedPages+StackYellowPages+StackRedPages+StackShadowPages+
4443                                        2*BytesPerWord COMPILER2_PRESENT(+1)) * page_size);
4444
4445  size_t threadStackSizeInBytes = ThreadStackSize * K;
4446  if (threadStackSizeInBytes != 0 &&
4447      threadStackSizeInBytes < os::Solaris::min_stack_allowed) {
4448    tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
4449                  os::Solaris::min_stack_allowed/K);
4450    return JNI_ERR;
4451  }
4452
4453  // For 64kbps there will be a 64kb page size, which makes
4454  // the usable default stack size quite a bit less.  Increase the
4455  // stack for 64kb (or any > than 8kb) pages, this increases
4456  // virtual memory fragmentation (since we're not creating the
4457  // stack on a power of 2 boundary.  The real fix for this
4458  // should be to fix the guard page mechanism.
4459
4460  if (vm_page_size() > 8*K) {
4461    threadStackSizeInBytes = (threadStackSizeInBytes != 0)
4462       ? threadStackSizeInBytes +
4463         ((StackYellowPages + StackRedPages) * vm_page_size())
4464       : 0;
4465    ThreadStackSize = threadStackSizeInBytes/K;
4466  }
4467
4468  // Make the stack size a multiple of the page size so that
4469  // the yellow/red zones can be guarded.
4470  JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4471                                                vm_page_size()));
4472
4473  Solaris::libthread_init();
4474
4475  if (UseNUMA) {
4476    if (!Solaris::liblgrp_init()) {
4477      UseNUMA = false;
4478    } else {
4479      size_t lgrp_limit = os::numa_get_groups_num();
4480      int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
4481      size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4482      FREE_C_HEAP_ARRAY(int, lgrp_ids);
4483      if (lgrp_num < 2) {
4484        // There's only one locality group, disable NUMA.
4485        UseNUMA = false;
4486      }
4487    }
4488    if (!UseNUMA && ForceNUMA) {
4489      UseNUMA = true;
4490    }
4491  }
4492
4493  Solaris::signal_sets_init();
4494  Solaris::init_signal_mem();
4495  Solaris::install_signal_handlers();
4496
4497  if (libjsigversion < JSIG_VERSION_1_4_1) {
4498    Maxlibjsigsigs = OLDMAXSIGNUM;
4499  }
4500
4501  // initialize synchronization primitives to use either thread or
4502  // lwp synchronization (controlled by UseLWPSynchronization)
4503  Solaris::synchronization_init();
4504
4505  if (MaxFDLimit) {
4506    // set the number of file descriptors to max. print out error
4507    // if getrlimit/setrlimit fails but continue regardless.
4508    struct rlimit nbr_files;
4509    int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4510    if (status != 0) {
4511      if (PrintMiscellaneous && (Verbose || WizardMode)) {
4512        perror("os::init_2 getrlimit failed");
4513      }
4514    } else {
4515      nbr_files.rlim_cur = nbr_files.rlim_max;
4516      status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4517      if (status != 0) {
4518        if (PrintMiscellaneous && (Verbose || WizardMode)) {
4519          perror("os::init_2 setrlimit failed");
4520        }
4521      }
4522    }
4523  }
4524
4525  // Calculate theoretical max. size of Threads to guard gainst
4526  // artifical out-of-memory situations, where all available address-
4527  // space has been reserved by thread stacks. Default stack size is 1Mb.
4528  size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4529    JavaThread::stack_size_at_create() : (1*K*K);
4530  assert(pre_thread_stack_size != 0, "Must have a stack");
4531  // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4532  // we should start doing Virtual Memory banging. Currently when the threads will
4533  // have used all but 200Mb of space.
4534  size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4535  Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4536
4537  // at-exit methods are called in the reverse order of their registration.
4538  // In Solaris 7 and earlier, atexit functions are called on return from
4539  // main or as a result of a call to exit(3C). There can be only 32 of
4540  // these functions registered and atexit() does not set errno. In Solaris
4541  // 8 and later, there is no limit to the number of functions registered
4542  // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4543  // functions are called upon dlclose(3DL) in addition to return from main
4544  // and exit(3C).
4545
4546  if (PerfAllowAtExitRegistration) {
4547    // only register atexit functions if PerfAllowAtExitRegistration is set.
4548    // atexit functions can be delayed until process exit time, which
4549    // can be problematic for embedded VM situations. Embedded VMs should
4550    // call DestroyJavaVM() to assure that VM resources are released.
4551
4552    // note: perfMemory_exit_helper atexit function may be removed in
4553    // the future if the appropriate cleanup code can be added to the
4554    // VM_Exit VMOperation's doit method.
4555    if (atexit(perfMemory_exit_helper) != 0) {
4556      warning("os::init2 atexit(perfMemory_exit_helper) failed");
4557    }
4558  }
4559
4560  // Init pset_loadavg function pointer
4561  init_pset_getloadavg_ptr();
4562
4563  return JNI_OK;
4564}
4565
4566// Mark the polling page as unreadable
4567void os::make_polling_page_unreadable(void) {
4568  if (mprotect((char *)_polling_page, page_size, PROT_NONE) != 0) {
4569    fatal("Could not disable polling page");
4570  }
4571}
4572
4573// Mark the polling page as readable
4574void os::make_polling_page_readable(void) {
4575  if (mprotect((char *)_polling_page, page_size, PROT_READ) != 0) {
4576    fatal("Could not enable polling page");
4577  }
4578}
4579
4580// OS interface.
4581
4582bool os::check_heap(bool force) { return true; }
4583
4584// Is a (classpath) directory empty?
4585bool os::dir_is_empty(const char* path) {
4586  DIR *dir = NULL;
4587  struct dirent *ptr;
4588
4589  dir = opendir(path);
4590  if (dir == NULL) return true;
4591
4592  // Scan the directory
4593  bool result = true;
4594  char buf[sizeof(struct dirent) + MAX_PATH];
4595  struct dirent *dbuf = (struct dirent *) buf;
4596  while (result && (ptr = readdir(dir, dbuf)) != NULL) {
4597    if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4598      result = false;
4599    }
4600  }
4601  closedir(dir);
4602  return result;
4603}
4604
4605// This code originates from JDK's sysOpen and open64_w
4606// from src/solaris/hpi/src/system_md.c
4607
4608int os::open(const char *path, int oflag, int mode) {
4609  if (strlen(path) > MAX_PATH - 1) {
4610    errno = ENAMETOOLONG;
4611    return -1;
4612  }
4613  int fd;
4614
4615  fd = ::open64(path, oflag, mode);
4616  if (fd == -1) return -1;
4617
4618  // If the open succeeded, the file might still be a directory
4619  {
4620    struct stat64 buf64;
4621    int ret = ::fstat64(fd, &buf64);
4622    int st_mode = buf64.st_mode;
4623
4624    if (ret != -1) {
4625      if ((st_mode & S_IFMT) == S_IFDIR) {
4626        errno = EISDIR;
4627        ::close(fd);
4628        return -1;
4629      }
4630    } else {
4631      ::close(fd);
4632      return -1;
4633    }
4634  }
4635
4636  // 32-bit Solaris systems suffer from:
4637  //
4638  // - an historical default soft limit of 256 per-process file
4639  //   descriptors that is too low for many Java programs.
4640  //
4641  // - a design flaw where file descriptors created using stdio
4642  //   fopen must be less than 256, _even_ when the first limit above
4643  //   has been raised.  This can cause calls to fopen (but not calls to
4644  //   open, for example) to fail mysteriously, perhaps in 3rd party
4645  //   native code (although the JDK itself uses fopen).  One can hardly
4646  //   criticize them for using this most standard of all functions.
4647  //
4648  // We attempt to make everything work anyways by:
4649  //
4650  // - raising the soft limit on per-process file descriptors beyond
4651  //   256
4652  //
4653  // - As of Solaris 10u4, we can request that Solaris raise the 256
4654  //   stdio fopen limit by calling function enable_extended_FILE_stdio.
4655  //   This is done in init_2 and recorded in enabled_extended_FILE_stdio
4656  //
4657  // - If we are stuck on an old (pre 10u4) Solaris system, we can
4658  //   workaround the bug by remapping non-stdio file descriptors below
4659  //   256 to ones beyond 256, which is done below.
4660  //
4661  // See:
4662  // 1085341: 32-bit stdio routines should support file descriptors >255
4663  // 6533291: Work around 32-bit Solaris stdio limit of 256 open files
4664  // 6431278: Netbeans crash on 32 bit Solaris: need to call
4665  //          enable_extended_FILE_stdio() in VM initialisation
4666  // Giri Mandalika's blog
4667  // http://technopark02.blogspot.com/2005_05_01_archive.html
4668  //
4669#ifndef  _LP64
4670  if ((!enabled_extended_FILE_stdio) && fd < 256) {
4671    int newfd = ::fcntl(fd, F_DUPFD, 256);
4672    if (newfd != -1) {
4673      ::close(fd);
4674      fd = newfd;
4675    }
4676  }
4677#endif // 32-bit Solaris
4678
4679  // All file descriptors that are opened in the JVM and not
4680  // specifically destined for a subprocess should have the
4681  // close-on-exec flag set.  If we don't set it, then careless 3rd
4682  // party native code might fork and exec without closing all
4683  // appropriate file descriptors (e.g. as we do in closeDescriptors in
4684  // UNIXProcess.c), and this in turn might:
4685  //
4686  // - cause end-of-file to fail to be detected on some file
4687  //   descriptors, resulting in mysterious hangs, or
4688  //
4689  // - might cause an fopen in the subprocess to fail on a system
4690  //   suffering from bug 1085341.
4691  //
4692  // (Yes, the default setting of the close-on-exec flag is a Unix
4693  // design flaw)
4694  //
4695  // See:
4696  // 1085341: 32-bit stdio routines should support file descriptors >255
4697  // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4698  // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4699  //
4700#ifdef FD_CLOEXEC
4701  {
4702    int flags = ::fcntl(fd, F_GETFD);
4703    if (flags != -1) {
4704      ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4705    }
4706  }
4707#endif
4708
4709  return fd;
4710}
4711
4712// create binary file, rewriting existing file if required
4713int os::create_binary_file(const char* path, bool rewrite_existing) {
4714  int oflags = O_WRONLY | O_CREAT;
4715  if (!rewrite_existing) {
4716    oflags |= O_EXCL;
4717  }
4718  return ::open64(path, oflags, S_IREAD | S_IWRITE);
4719}
4720
4721// return current position of file pointer
4722jlong os::current_file_offset(int fd) {
4723  return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4724}
4725
4726// move file pointer to the specified offset
4727jlong os::seek_to_file_offset(int fd, jlong offset) {
4728  return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4729}
4730
4731jlong os::lseek(int fd, jlong offset, int whence) {
4732  return (jlong) ::lseek64(fd, offset, whence);
4733}
4734
4735char * os::native_path(char *path) {
4736  return path;
4737}
4738
4739int os::ftruncate(int fd, jlong length) {
4740  return ::ftruncate64(fd, length);
4741}
4742
4743int os::fsync(int fd)  {
4744  RESTARTABLE_RETURN_INT(::fsync(fd));
4745}
4746
4747int os::available(int fd, jlong *bytes) {
4748  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
4749         "Assumed _thread_in_native");
4750  jlong cur, end;
4751  int mode;
4752  struct stat64 buf64;
4753
4754  if (::fstat64(fd, &buf64) >= 0) {
4755    mode = buf64.st_mode;
4756    if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4757      int n,ioctl_return;
4758
4759      RESTARTABLE(::ioctl(fd, FIONREAD, &n), ioctl_return);
4760      if (ioctl_return>= 0) {
4761        *bytes = n;
4762        return 1;
4763      }
4764    }
4765  }
4766  if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4767    return 0;
4768  } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4769    return 0;
4770  } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4771    return 0;
4772  }
4773  *bytes = end - cur;
4774  return 1;
4775}
4776
4777// Map a block of memory.
4778char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4779                        char *addr, size_t bytes, bool read_only,
4780                        bool allow_exec) {
4781  int prot;
4782  int flags;
4783
4784  if (read_only) {
4785    prot = PROT_READ;
4786    flags = MAP_SHARED;
4787  } else {
4788    prot = PROT_READ | PROT_WRITE;
4789    flags = MAP_PRIVATE;
4790  }
4791
4792  if (allow_exec) {
4793    prot |= PROT_EXEC;
4794  }
4795
4796  if (addr != NULL) {
4797    flags |= MAP_FIXED;
4798  }
4799
4800  char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4801                                     fd, file_offset);
4802  if (mapped_address == MAP_FAILED) {
4803    return NULL;
4804  }
4805  return mapped_address;
4806}
4807
4808
4809// Remap a block of memory.
4810char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4811                          char *addr, size_t bytes, bool read_only,
4812                          bool allow_exec) {
4813  // same as map_memory() on this OS
4814  return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4815                        allow_exec);
4816}
4817
4818
4819// Unmap a block of memory.
4820bool os::pd_unmap_memory(char* addr, size_t bytes) {
4821  return munmap(addr, bytes) == 0;
4822}
4823
4824void os::pause() {
4825  char filename[MAX_PATH];
4826  if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4827    jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4828  } else {
4829    jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4830  }
4831
4832  int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4833  if (fd != -1) {
4834    struct stat buf;
4835    ::close(fd);
4836    while (::stat(filename, &buf) == 0) {
4837      (void)::poll(NULL, 0, 100);
4838    }
4839  } else {
4840    jio_fprintf(stderr,
4841                "Could not open pause file '%s', continuing immediately.\n", filename);
4842  }
4843}
4844
4845#ifndef PRODUCT
4846#ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4847// Turn this on if you need to trace synch operations.
4848// Set RECORD_SYNCH_LIMIT to a large-enough value,
4849// and call record_synch_enable and record_synch_disable
4850// around the computation of interest.
4851
4852void record_synch(char* name, bool returning);  // defined below
4853
4854class RecordSynch {
4855  char* _name;
4856 public:
4857  RecordSynch(char* name) :_name(name) { record_synch(_name, false); }
4858  ~RecordSynch()                       { record_synch(_name, true); }
4859};
4860
4861#define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
4862extern "C" ret name params {                                    \
4863  typedef ret name##_t params;                                  \
4864  static name##_t* implem = NULL;                               \
4865  static int callcount = 0;                                     \
4866  if (implem == NULL) {                                         \
4867    implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
4868    if (implem == NULL)  fatal(dlerror());                      \
4869  }                                                             \
4870  ++callcount;                                                  \
4871  RecordSynch _rs(#name);                                       \
4872  inner;                                                        \
4873  return implem args;                                           \
4874}
4875// in dbx, examine callcounts this way:
4876// for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
4877
4878#define CHECK_POINTER_OK(p) \
4879  (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
4880#define CHECK_MU \
4881  if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
4882#define CHECK_CV \
4883  if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
4884#define CHECK_P(p) \
4885  if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
4886
4887#define CHECK_MUTEX(mutex_op) \
4888  CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
4889
4890CHECK_MUTEX(   mutex_lock)
4891CHECK_MUTEX(  _mutex_lock)
4892CHECK_MUTEX( mutex_unlock)
4893CHECK_MUTEX(_mutex_unlock)
4894CHECK_MUTEX( mutex_trylock)
4895CHECK_MUTEX(_mutex_trylock)
4896
4897#define CHECK_COND(cond_op) \
4898  CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU; CHECK_CV);
4899
4900CHECK_COND( cond_wait);
4901CHECK_COND(_cond_wait);
4902CHECK_COND(_cond_wait_cancel);
4903
4904#define CHECK_COND2(cond_op) \
4905  CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU; CHECK_CV);
4906
4907CHECK_COND2( cond_timedwait);
4908CHECK_COND2(_cond_timedwait);
4909CHECK_COND2(_cond_timedwait_cancel);
4910
4911// do the _lwp_* versions too
4912#define mutex_t lwp_mutex_t
4913#define cond_t  lwp_cond_t
4914CHECK_MUTEX(  _lwp_mutex_lock)
4915CHECK_MUTEX(  _lwp_mutex_unlock)
4916CHECK_MUTEX(  _lwp_mutex_trylock)
4917CHECK_MUTEX( __lwp_mutex_lock)
4918CHECK_MUTEX( __lwp_mutex_unlock)
4919CHECK_MUTEX( __lwp_mutex_trylock)
4920CHECK_MUTEX(___lwp_mutex_lock)
4921CHECK_MUTEX(___lwp_mutex_unlock)
4922
4923CHECK_COND(  _lwp_cond_wait);
4924CHECK_COND( __lwp_cond_wait);
4925CHECK_COND(___lwp_cond_wait);
4926
4927CHECK_COND2(  _lwp_cond_timedwait);
4928CHECK_COND2( __lwp_cond_timedwait);
4929#undef mutex_t
4930#undef cond_t
4931
4932CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4933CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4934CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4935CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4936CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4937CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4938CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4939CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4940
4941
4942// recording machinery:
4943
4944enum { RECORD_SYNCH_LIMIT = 200 };
4945char* record_synch_name[RECORD_SYNCH_LIMIT];
4946void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
4947bool record_synch_returning[RECORD_SYNCH_LIMIT];
4948thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
4949int record_synch_count = 0;
4950bool record_synch_enabled = false;
4951
4952// in dbx, examine recorded data this way:
4953// for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
4954
4955void record_synch(char* name, bool returning) {
4956  if (record_synch_enabled) {
4957    if (record_synch_count < RECORD_SYNCH_LIMIT) {
4958      record_synch_name[record_synch_count] = name;
4959      record_synch_returning[record_synch_count] = returning;
4960      record_synch_thread[record_synch_count] = thr_self();
4961      record_synch_arg0ptr[record_synch_count] = &name;
4962      record_synch_count++;
4963    }
4964    // put more checking code here:
4965    // ...
4966  }
4967}
4968
4969void record_synch_enable() {
4970  // start collecting trace data, if not already doing so
4971  if (!record_synch_enabled)  record_synch_count = 0;
4972  record_synch_enabled = true;
4973}
4974
4975void record_synch_disable() {
4976  // stop collecting trace data
4977  record_synch_enabled = false;
4978}
4979
4980#endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4981#endif // PRODUCT
4982
4983const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4984const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
4985                               (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4986
4987
4988// JVMTI & JVM monitoring and management support
4989// The thread_cpu_time() and current_thread_cpu_time() are only
4990// supported if is_thread_cpu_time_supported() returns true.
4991// They are not supported on Solaris T1.
4992
4993// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4994// are used by JVM M&M and JVMTI to get user+sys or user CPU time
4995// of a thread.
4996//
4997// current_thread_cpu_time() and thread_cpu_time(Thread *)
4998// returns the fast estimate available on the platform.
4999
5000// hrtime_t gethrvtime() return value includes
5001// user time but does not include system time
5002jlong os::current_thread_cpu_time() {
5003  return (jlong) gethrvtime();
5004}
5005
5006jlong os::thread_cpu_time(Thread *thread) {
5007  // return user level CPU time only to be consistent with
5008  // what current_thread_cpu_time returns.
5009  // thread_cpu_time_info() must be changed if this changes
5010  return os::thread_cpu_time(thread, false /* user time only */);
5011}
5012
5013jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5014  if (user_sys_cpu_time) {
5015    return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5016  } else {
5017    return os::current_thread_cpu_time();
5018  }
5019}
5020
5021jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5022  char proc_name[64];
5023  int count;
5024  prusage_t prusage;
5025  jlong lwp_time;
5026  int fd;
5027
5028  sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5029          getpid(),
5030          thread->osthread()->lwp_id());
5031  fd = ::open(proc_name, O_RDONLY);
5032  if (fd == -1) return -1;
5033
5034  do {
5035    count = ::pread(fd,
5036                    (void *)&prusage.pr_utime,
5037                    thr_time_size,
5038                    thr_time_off);
5039  } while (count < 0 && errno == EINTR);
5040  ::close(fd);
5041  if (count < 0) return -1;
5042
5043  if (user_sys_cpu_time) {
5044    // user + system CPU time
5045    lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5046                 (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5047                 (jlong)prusage.pr_stime.tv_nsec +
5048                 (jlong)prusage.pr_utime.tv_nsec;
5049  } else {
5050    // user level CPU time only
5051    lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5052                (jlong)prusage.pr_utime.tv_nsec;
5053  }
5054
5055  return (lwp_time);
5056}
5057
5058void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5059  info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5060  info_ptr->may_skip_backward = false;    // elapsed time not wall time
5061  info_ptr->may_skip_forward = false;     // elapsed time not wall time
5062  info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5063}
5064
5065void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5066  info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5067  info_ptr->may_skip_backward = false;    // elapsed time not wall time
5068  info_ptr->may_skip_forward = false;     // elapsed time not wall time
5069  info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5070}
5071
5072bool os::is_thread_cpu_time_supported() {
5073  return true;
5074}
5075
5076// System loadavg support.  Returns -1 if load average cannot be obtained.
5077// Return the load average for our processor set if the primitive exists
5078// (Solaris 9 and later).  Otherwise just return system wide loadavg.
5079int os::loadavg(double loadavg[], int nelem) {
5080  if (pset_getloadavg_ptr != NULL) {
5081    return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5082  } else {
5083    return ::getloadavg(loadavg, nelem);
5084  }
5085}
5086
5087//---------------------------------------------------------------------------------
5088
5089bool os::find(address addr, outputStream* st) {
5090  Dl_info dlinfo;
5091  memset(&dlinfo, 0, sizeof(dlinfo));
5092  if (dladdr(addr, &dlinfo) != 0) {
5093    st->print(PTR_FORMAT ": ", addr);
5094    if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5095      st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5096    } else if (dlinfo.dli_fbase != NULL) {
5097      st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5098    } else {
5099      st->print("<absolute address>");
5100    }
5101    if (dlinfo.dli_fname != NULL) {
5102      st->print(" in %s", dlinfo.dli_fname);
5103    }
5104    if (dlinfo.dli_fbase != NULL) {
5105      st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5106    }
5107    st->cr();
5108
5109    if (Verbose) {
5110      // decode some bytes around the PC
5111      address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5112      address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5113      address       lowest = (address) dlinfo.dli_sname;
5114      if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5115      if (begin < lowest)  begin = lowest;
5116      Dl_info dlinfo2;
5117      if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5118          && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5119        end = (address) dlinfo2.dli_saddr;
5120      }
5121      Disassembler::decode(begin, end, st);
5122    }
5123    return true;
5124  }
5125  return false;
5126}
5127
5128// Following function has been added to support HotSparc's libjvm.so running
5129// under Solaris production JDK 1.2.2 / 1.3.0.  These came from
5130// src/solaris/hpi/native_threads in the EVM codebase.
5131//
5132// NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5133// libraries and should thus be removed. We will leave it behind for a while
5134// until we no longer want to able to run on top of 1.3.0 Solaris production
5135// JDK. See 4341971.
5136
5137#define STACK_SLACK 0x800
5138
5139extern "C" {
5140  intptr_t sysThreadAvailableStackWithSlack() {
5141    stack_t st;
5142    intptr_t retval, stack_top;
5143    retval = thr_stksegment(&st);
5144    assert(retval == 0, "incorrect return value from thr_stksegment");
5145    assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5146    assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5147    stack_top=(intptr_t)st.ss_sp-st.ss_size;
5148    return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5149  }
5150}
5151
5152// ObjectMonitor park-unpark infrastructure ...
5153//
5154// We implement Solaris and Linux PlatformEvents with the
5155// obvious condvar-mutex-flag triple.
5156// Another alternative that works quite well is pipes:
5157// Each PlatformEvent consists of a pipe-pair.
5158// The thread associated with the PlatformEvent
5159// calls park(), which reads from the input end of the pipe.
5160// Unpark() writes into the other end of the pipe.
5161// The write-side of the pipe must be set NDELAY.
5162// Unfortunately pipes consume a large # of handles.
5163// Native solaris lwp_park() and lwp_unpark() work nicely, too.
5164// Using pipes for the 1st few threads might be workable, however.
5165//
5166// park() is permitted to return spuriously.
5167// Callers of park() should wrap the call to park() in
5168// an appropriate loop.  A litmus test for the correct
5169// usage of park is the following: if park() were modified
5170// to immediately return 0 your code should still work,
5171// albeit degenerating to a spin loop.
5172//
5173// In a sense, park()-unpark() just provides more polite spinning
5174// and polling with the key difference over naive spinning being
5175// that a parked thread needs to be explicitly unparked() in order
5176// to wake up and to poll the underlying condition.
5177//
5178// Assumption:
5179//    Only one parker can exist on an event, which is why we allocate
5180//    them per-thread. Multiple unparkers can coexist.
5181//
5182// _Event transitions in park()
5183//   -1 => -1 : illegal
5184//    1 =>  0 : pass - return immediately
5185//    0 => -1 : block; then set _Event to 0 before returning
5186//
5187// _Event transitions in unpark()
5188//    0 => 1 : just return
5189//    1 => 1 : just return
5190//   -1 => either 0 or 1; must signal target thread
5191//         That is, we can safely transition _Event from -1 to either
5192//         0 or 1.
5193//
5194// _Event serves as a restricted-range semaphore.
5195//   -1 : thread is blocked, i.e. there is a waiter
5196//    0 : neutral: thread is running or ready,
5197//        could have been signaled after a wait started
5198//    1 : signaled - thread is running or ready
5199//
5200// Another possible encoding of _Event would be with
5201// explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5202//
5203// TODO-FIXME: add DTRACE probes for:
5204// 1.   Tx parks
5205// 2.   Ty unparks Tx
5206// 3.   Tx resumes from park
5207
5208
5209// value determined through experimentation
5210#define ROUNDINGFIX 11
5211
5212// utility to compute the abstime argument to timedwait.
5213// TODO-FIXME: switch from compute_abstime() to unpackTime().
5214
5215static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5216  // millis is the relative timeout time
5217  // abstime will be the absolute timeout time
5218  if (millis < 0)  millis = 0;
5219  struct timeval now;
5220  int status = gettimeofday(&now, NULL);
5221  assert(status == 0, "gettimeofday");
5222  jlong seconds = millis / 1000;
5223  jlong max_wait_period;
5224
5225  if (UseLWPSynchronization) {
5226    // forward port of fix for 4275818 (not sleeping long enough)
5227    // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5228    // _lwp_cond_timedwait() used a round_down algorithm rather
5229    // than a round_up. For millis less than our roundfactor
5230    // it rounded down to 0 which doesn't meet the spec.
5231    // For millis > roundfactor we may return a bit sooner, but
5232    // since we can not accurately identify the patch level and
5233    // this has already been fixed in Solaris 9 and 8 we will
5234    // leave it alone rather than always rounding down.
5235
5236    if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5237    // It appears that when we go directly through Solaris _lwp_cond_timedwait()
5238    // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
5239    max_wait_period = 21000000;
5240  } else {
5241    max_wait_period = 50000000;
5242  }
5243  millis %= 1000;
5244  if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
5245    seconds = max_wait_period;
5246  }
5247  abstime->tv_sec = now.tv_sec  + seconds;
5248  long       usec = now.tv_usec + millis * 1000;
5249  if (usec >= 1000000) {
5250    abstime->tv_sec += 1;
5251    usec -= 1000000;
5252  }
5253  abstime->tv_nsec = usec * 1000;
5254  return abstime;
5255}
5256
5257void os::PlatformEvent::park() {           // AKA: down()
5258  // Transitions for _Event:
5259  //   -1 => -1 : illegal
5260  //    1 =>  0 : pass - return immediately
5261  //    0 => -1 : block; then set _Event to 0 before returning
5262
5263  // Invariant: Only the thread associated with the Event/PlatformEvent
5264  // may call park().
5265  assert(_nParked == 0, "invariant");
5266
5267  int v;
5268  for (;;) {
5269    v = _Event;
5270    if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5271  }
5272  guarantee(v >= 0, "invariant");
5273  if (v == 0) {
5274    // Do this the hard way by blocking ...
5275    // See http://monaco.sfbay/detail.jsf?cr=5094058.
5276    // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5277    // Only for SPARC >= V8PlusA
5278#if defined(__sparc) && defined(COMPILER2)
5279    if (ClearFPUAtPark) { _mark_fpu_nosave(); }
5280#endif
5281    int status = os::Solaris::mutex_lock(_mutex);
5282    assert_status(status == 0, status, "mutex_lock");
5283    guarantee(_nParked == 0, "invariant");
5284    ++_nParked;
5285    while (_Event < 0) {
5286      // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5287      // Treat this the same as if the wait was interrupted
5288      // With usr/lib/lwp going to kernel, always handle ETIME
5289      status = os::Solaris::cond_wait(_cond, _mutex);
5290      if (status == ETIME) status = EINTR;
5291      assert_status(status == 0 || status == EINTR, status, "cond_wait");
5292    }
5293    --_nParked;
5294    _Event = 0;
5295    status = os::Solaris::mutex_unlock(_mutex);
5296    assert_status(status == 0, status, "mutex_unlock");
5297    // Paranoia to ensure our locked and lock-free paths interact
5298    // correctly with each other.
5299    OrderAccess::fence();
5300  }
5301}
5302
5303int os::PlatformEvent::park(jlong millis) {
5304  // Transitions for _Event:
5305  //   -1 => -1 : illegal
5306  //    1 =>  0 : pass - return immediately
5307  //    0 => -1 : block; then set _Event to 0 before returning
5308
5309  guarantee(_nParked == 0, "invariant");
5310  int v;
5311  for (;;) {
5312    v = _Event;
5313    if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5314  }
5315  guarantee(v >= 0, "invariant");
5316  if (v != 0) return OS_OK;
5317
5318  int ret = OS_TIMEOUT;
5319  timestruc_t abst;
5320  compute_abstime(&abst, millis);
5321
5322  // See http://monaco.sfbay/detail.jsf?cr=5094058.
5323  // For Solaris SPARC set fprs.FEF=0 prior to parking.
5324  // Only for SPARC >= V8PlusA
5325#if defined(__sparc) && defined(COMPILER2)
5326  if (ClearFPUAtPark) { _mark_fpu_nosave(); }
5327#endif
5328  int status = os::Solaris::mutex_lock(_mutex);
5329  assert_status(status == 0, status, "mutex_lock");
5330  guarantee(_nParked == 0, "invariant");
5331  ++_nParked;
5332  while (_Event < 0) {
5333    int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5334    assert_status(status == 0 || status == EINTR ||
5335                  status == ETIME || status == ETIMEDOUT,
5336                  status, "cond_timedwait");
5337    if (!FilterSpuriousWakeups) break;                // previous semantics
5338    if (status == ETIME || status == ETIMEDOUT) break;
5339    // We consume and ignore EINTR and spurious wakeups.
5340  }
5341  --_nParked;
5342  if (_Event >= 0) ret = OS_OK;
5343  _Event = 0;
5344  status = os::Solaris::mutex_unlock(_mutex);
5345  assert_status(status == 0, status, "mutex_unlock");
5346  // Paranoia to ensure our locked and lock-free paths interact
5347  // correctly with each other.
5348  OrderAccess::fence();
5349  return ret;
5350}
5351
5352void os::PlatformEvent::unpark() {
5353  // Transitions for _Event:
5354  //    0 => 1 : just return
5355  //    1 => 1 : just return
5356  //   -1 => either 0 or 1; must signal target thread
5357  //         That is, we can safely transition _Event from -1 to either
5358  //         0 or 1.
5359  // See also: "Semaphores in Plan 9" by Mullender & Cox
5360  //
5361  // Note: Forcing a transition from "-1" to "1" on an unpark() means
5362  // that it will take two back-to-back park() calls for the owning
5363  // thread to block. This has the benefit of forcing a spurious return
5364  // from the first park() call after an unpark() call which will help
5365  // shake out uses of park() and unpark() without condition variables.
5366
5367  if (Atomic::xchg(1, &_Event) >= 0) return;
5368
5369  // If the thread associated with the event was parked, wake it.
5370  // Wait for the thread assoc with the PlatformEvent to vacate.
5371  int status = os::Solaris::mutex_lock(_mutex);
5372  assert_status(status == 0, status, "mutex_lock");
5373  int AnyWaiters = _nParked;
5374  status = os::Solaris::mutex_unlock(_mutex);
5375  assert_status(status == 0, status, "mutex_unlock");
5376  guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5377  if (AnyWaiters != 0) {
5378    // Note that we signal() *after* dropping the lock for "immortal" Events.
5379    // This is safe and avoids a common class of  futile wakeups.  In rare
5380    // circumstances this can cause a thread to return prematurely from
5381    // cond_{timed}wait() but the spurious wakeup is benign and the victim
5382    // will simply re-test the condition and re-park itself.
5383    // This provides particular benefit if the underlying platform does not
5384    // provide wait morphing.
5385    status = os::Solaris::cond_signal(_cond);
5386    assert_status(status == 0, status, "cond_signal");
5387  }
5388}
5389
5390// JSR166
5391// -------------------------------------------------------
5392
5393// The solaris and linux implementations of park/unpark are fairly
5394// conservative for now, but can be improved. They currently use a
5395// mutex/condvar pair, plus _counter.
5396// Park decrements _counter if > 0, else does a condvar wait.  Unpark
5397// sets count to 1 and signals condvar.  Only one thread ever waits
5398// on the condvar. Contention seen when trying to park implies that someone
5399// is unparking you, so don't wait. And spurious returns are fine, so there
5400// is no need to track notifications.
5401
5402#define MAX_SECS 100000000
5403
5404// This code is common to linux and solaris and will be moved to a
5405// common place in dolphin.
5406//
5407// The passed in time value is either a relative time in nanoseconds
5408// or an absolute time in milliseconds. Either way it has to be unpacked
5409// into suitable seconds and nanoseconds components and stored in the
5410// given timespec structure.
5411// Given time is a 64-bit value and the time_t used in the timespec is only
5412// a signed-32-bit value (except on 64-bit Linux) we have to watch for
5413// overflow if times way in the future are given. Further on Solaris versions
5414// prior to 10 there is a restriction (see cond_timedwait) that the specified
5415// number of seconds, in abstime, is less than current_time  + 100,000,000.
5416// As it will be 28 years before "now + 100000000" will overflow we can
5417// ignore overflow and just impose a hard-limit on seconds using the value
5418// of "now + 100,000,000". This places a limit on the timeout of about 3.17
5419// years from "now".
5420//
5421static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5422  assert(time > 0, "convertTime");
5423
5424  struct timeval now;
5425  int status = gettimeofday(&now, NULL);
5426  assert(status == 0, "gettimeofday");
5427
5428  time_t max_secs = now.tv_sec + MAX_SECS;
5429
5430  if (isAbsolute) {
5431    jlong secs = time / 1000;
5432    if (secs > max_secs) {
5433      absTime->tv_sec = max_secs;
5434    } else {
5435      absTime->tv_sec = secs;
5436    }
5437    absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5438  } else {
5439    jlong secs = time / NANOSECS_PER_SEC;
5440    if (secs >= MAX_SECS) {
5441      absTime->tv_sec = max_secs;
5442      absTime->tv_nsec = 0;
5443    } else {
5444      absTime->tv_sec = now.tv_sec + secs;
5445      absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5446      if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5447        absTime->tv_nsec -= NANOSECS_PER_SEC;
5448        ++absTime->tv_sec; // note: this must be <= max_secs
5449      }
5450    }
5451  }
5452  assert(absTime->tv_sec >= 0, "tv_sec < 0");
5453  assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5454  assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5455  assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5456}
5457
5458void Parker::park(bool isAbsolute, jlong time) {
5459  // Ideally we'd do something useful while spinning, such
5460  // as calling unpackTime().
5461
5462  // Optional fast-path check:
5463  // Return immediately if a permit is available.
5464  // We depend on Atomic::xchg() having full barrier semantics
5465  // since we are doing a lock-free update to _counter.
5466  if (Atomic::xchg(0, &_counter) > 0) return;
5467
5468  // Optional fast-exit: Check interrupt before trying to wait
5469  Thread* thread = Thread::current();
5470  assert(thread->is_Java_thread(), "Must be JavaThread");
5471  JavaThread *jt = (JavaThread *)thread;
5472  if (Thread::is_interrupted(thread, false)) {
5473    return;
5474  }
5475
5476  // First, demultiplex/decode time arguments
5477  timespec absTime;
5478  if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5479    return;
5480  }
5481  if (time > 0) {
5482    // Warning: this code might be exposed to the old Solaris time
5483    // round-down bugs.  Grep "roundingFix" for details.
5484    unpackTime(&absTime, isAbsolute, time);
5485  }
5486
5487  // Enter safepoint region
5488  // Beware of deadlocks such as 6317397.
5489  // The per-thread Parker:: _mutex is a classic leaf-lock.
5490  // In particular a thread must never block on the Threads_lock while
5491  // holding the Parker:: mutex.  If safepoints are pending both the
5492  // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5493  ThreadBlockInVM tbivm(jt);
5494
5495  // Don't wait if cannot get lock since interference arises from
5496  // unblocking.  Also. check interrupt before trying wait
5497  if (Thread::is_interrupted(thread, false) ||
5498      os::Solaris::mutex_trylock(_mutex) != 0) {
5499    return;
5500  }
5501
5502  int status;
5503
5504  if (_counter > 0)  { // no wait needed
5505    _counter = 0;
5506    status = os::Solaris::mutex_unlock(_mutex);
5507    assert(status == 0, "invariant");
5508    // Paranoia to ensure our locked and lock-free paths interact
5509    // correctly with each other and Java-level accesses.
5510    OrderAccess::fence();
5511    return;
5512  }
5513
5514#ifdef ASSERT
5515  // Don't catch signals while blocked; let the running threads have the signals.
5516  // (This allows a debugger to break into the running thread.)
5517  sigset_t oldsigs;
5518  sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
5519  thr_sigsetmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5520#endif
5521
5522  OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5523  jt->set_suspend_equivalent();
5524  // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5525
5526  // Do this the hard way by blocking ...
5527  // See http://monaco.sfbay/detail.jsf?cr=5094058.
5528  // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5529  // Only for SPARC >= V8PlusA
5530#if defined(__sparc) && defined(COMPILER2)
5531  if (ClearFPUAtPark) { _mark_fpu_nosave(); }
5532#endif
5533
5534  if (time == 0) {
5535    status = os::Solaris::cond_wait(_cond, _mutex);
5536  } else {
5537    status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5538  }
5539  // Note that an untimed cond_wait() can sometimes return ETIME on older
5540  // versions of the Solaris.
5541  assert_status(status == 0 || status == EINTR ||
5542                status == ETIME || status == ETIMEDOUT,
5543                status, "cond_timedwait");
5544
5545#ifdef ASSERT
5546  thr_sigsetmask(SIG_SETMASK, &oldsigs, NULL);
5547#endif
5548  _counter = 0;
5549  status = os::Solaris::mutex_unlock(_mutex);
5550  assert_status(status == 0, status, "mutex_unlock");
5551  // Paranoia to ensure our locked and lock-free paths interact
5552  // correctly with each other and Java-level accesses.
5553  OrderAccess::fence();
5554
5555  // If externally suspended while waiting, re-suspend
5556  if (jt->handle_special_suspend_equivalent_condition()) {
5557    jt->java_suspend_self();
5558  }
5559}
5560
5561void Parker::unpark() {
5562  int status = os::Solaris::mutex_lock(_mutex);
5563  assert(status == 0, "invariant");
5564  const int s = _counter;
5565  _counter = 1;
5566  status = os::Solaris::mutex_unlock(_mutex);
5567  assert(status == 0, "invariant");
5568
5569  if (s < 1) {
5570    status = os::Solaris::cond_signal(_cond);
5571    assert(status == 0, "invariant");
5572  }
5573}
5574
5575extern char** environ;
5576
5577// Run the specified command in a separate process. Return its exit value,
5578// or -1 on failure (e.g. can't fork a new process).
5579// Unlike system(), this function can be called from signal handler. It
5580// doesn't block SIGINT et al.
5581int os::fork_and_exec(char* cmd) {
5582  char * argv[4];
5583  argv[0] = (char *)"sh";
5584  argv[1] = (char *)"-c";
5585  argv[2] = cmd;
5586  argv[3] = NULL;
5587
5588  // fork is async-safe, fork1 is not so can't use in signal handler
5589  pid_t pid;
5590  Thread* t = Thread::current_or_null_safe();
5591  if (t != NULL && t->is_inside_signal_handler()) {
5592    pid = fork();
5593  } else {
5594    pid = fork1();
5595  }
5596
5597  if (pid < 0) {
5598    // fork failed
5599    warning("fork failed: %s", strerror(errno));
5600    return -1;
5601
5602  } else if (pid == 0) {
5603    // child process
5604
5605    // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5606    execve("/usr/bin/sh", argv, environ);
5607
5608    // execve failed
5609    _exit(-1);
5610
5611  } else  {
5612    // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5613    // care about the actual exit code, for now.
5614
5615    int status;
5616
5617    // Wait for the child process to exit.  This returns immediately if
5618    // the child has already exited. */
5619    while (waitpid(pid, &status, 0) < 0) {
5620      switch (errno) {
5621      case ECHILD: return 0;
5622      case EINTR: break;
5623      default: return -1;
5624      }
5625    }
5626
5627    if (WIFEXITED(status)) {
5628      // The child exited normally; get its exit code.
5629      return WEXITSTATUS(status);
5630    } else if (WIFSIGNALED(status)) {
5631      // The child exited because of a signal
5632      // The best value to return is 0x80 + signal number,
5633      // because that is what all Unix shells do, and because
5634      // it allows callers to distinguish between process exit and
5635      // process death by signal.
5636      return 0x80 + WTERMSIG(status);
5637    } else {
5638      // Unknown exit code; pass it through
5639      return status;
5640    }
5641  }
5642}
5643
5644// is_headless_jre()
5645//
5646// Test for the existence of xawt/libmawt.so or libawt_xawt.so
5647// in order to report if we are running in a headless jre
5648//
5649// Since JDK8 xawt/libmawt.so was moved into the same directory
5650// as libawt.so, and renamed libawt_xawt.so
5651//
5652bool os::is_headless_jre() {
5653  struct stat statbuf;
5654  char buf[MAXPATHLEN];
5655  char libmawtpath[MAXPATHLEN];
5656  const char *xawtstr  = "/xawt/libmawt.so";
5657  const char *new_xawtstr = "/libawt_xawt.so";
5658  char *p;
5659
5660  // Get path to libjvm.so
5661  os::jvm_path(buf, sizeof(buf));
5662
5663  // Get rid of libjvm.so
5664  p = strrchr(buf, '/');
5665  if (p == NULL) {
5666    return false;
5667  } else {
5668    *p = '\0';
5669  }
5670
5671  // Get rid of client or server
5672  p = strrchr(buf, '/');
5673  if (p == NULL) {
5674    return false;
5675  } else {
5676    *p = '\0';
5677  }
5678
5679  // check xawt/libmawt.so
5680  strcpy(libmawtpath, buf);
5681  strcat(libmawtpath, xawtstr);
5682  if (::stat(libmawtpath, &statbuf) == 0) return false;
5683
5684  // check libawt_xawt.so
5685  strcpy(libmawtpath, buf);
5686  strcat(libmawtpath, new_xawtstr);
5687  if (::stat(libmawtpath, &statbuf) == 0) return false;
5688
5689  return true;
5690}
5691
5692size_t os::write(int fd, const void *buf, unsigned int nBytes) {
5693  size_t res;
5694  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5695         "Assumed _thread_in_native");
5696  RESTARTABLE((size_t) ::write(fd, buf, (size_t) nBytes), res);
5697  return res;
5698}
5699
5700int os::close(int fd) {
5701  return ::close(fd);
5702}
5703
5704int os::socket_close(int fd) {
5705  return ::close(fd);
5706}
5707
5708int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
5709  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5710         "Assumed _thread_in_native");
5711  RESTARTABLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags));
5712}
5713
5714int os::send(int fd, char* buf, size_t nBytes, uint flags) {
5715  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5716         "Assumed _thread_in_native");
5717  RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5718}
5719
5720int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
5721  RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5722}
5723
5724// As both poll and select can be interrupted by signals, we have to be
5725// prepared to restart the system call after updating the timeout, unless
5726// a poll() is done with timeout == -1, in which case we repeat with this
5727// "wait forever" value.
5728
5729int os::connect(int fd, struct sockaddr *him, socklen_t len) {
5730  int _result;
5731  _result = ::connect(fd, him, len);
5732
5733  // On Solaris, when a connect() call is interrupted, the connection
5734  // can be established asynchronously (see 6343810). Subsequent calls
5735  // to connect() must check the errno value which has the semantic
5736  // described below (copied from the connect() man page). Handling
5737  // of asynchronously established connections is required for both
5738  // blocking and non-blocking sockets.
5739  //     EINTR            The  connection  attempt  was   interrupted
5740  //                      before  any data arrived by the delivery of
5741  //                      a signal. The connection, however, will  be
5742  //                      established asynchronously.
5743  //
5744  //     EINPROGRESS      The socket is non-blocking, and the connec-
5745  //                      tion  cannot  be completed immediately.
5746  //
5747  //     EALREADY         The socket is non-blocking,  and a previous
5748  //                      connection  attempt  has  not yet been com-
5749  //                      pleted.
5750  //
5751  //     EISCONN          The socket is already connected.
5752  if (_result == OS_ERR && errno == EINTR) {
5753    // restarting a connect() changes its errno semantics
5754    RESTARTABLE(::connect(fd, him, len), _result);
5755    // undo these changes
5756    if (_result == OS_ERR) {
5757      if (errno == EALREADY) {
5758        errno = EINPROGRESS; // fall through
5759      } else if (errno == EISCONN) {
5760        errno = 0;
5761        return OS_OK;
5762      }
5763    }
5764  }
5765  return _result;
5766}
5767
5768// Get the default path to the core file
5769// Returns the length of the string
5770int os::get_core_path(char* buffer, size_t bufferSize) {
5771  const char* p = get_current_directory(buffer, bufferSize);
5772
5773  if (p == NULL) {
5774    assert(p != NULL, "failed to get current directory");
5775    return 0;
5776  }
5777
5778  jio_snprintf(buffer, bufferSize, "%s/core or core.%d",
5779                                              p, current_process_id());
5780
5781  return strlen(buffer);
5782}
5783
5784#ifndef PRODUCT
5785void TestReserveMemorySpecial_test() {
5786  // No tests available for this platform
5787}
5788#endif
5789
5790bool os::start_debugging(char *buf, int buflen) {
5791  int len = (int)strlen(buf);
5792  char *p = &buf[len];
5793
5794  jio_snprintf(p, buflen-len,
5795               "\n\n"
5796               "Do you want to debug the problem?\n\n"
5797               "To debug, run 'dbx - %d'; then switch to thread " INTX_FORMAT "\n"
5798               "Enter 'yes' to launch dbx automatically (PATH must include dbx)\n"
5799               "Otherwise, press RETURN to abort...",
5800               os::current_process_id(), os::current_thread_id());
5801
5802  bool yes = os::message_box("Unexpected Error", buf);
5803
5804  if (yes) {
5805    // yes, user asked VM to launch debugger
5806    jio_snprintf(buf, sizeof(buf), "dbx - %d", os::current_process_id());
5807
5808    os::fork_and_exec(buf);
5809    yes = false;
5810  }
5811  return yes;
5812}
5813