os_solaris.cpp revision 11271:e62bbe48bd66
1/*
2 * Copyright (c) 1997, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25// no precompiled headers
26#include "classfile/classLoader.hpp"
27#include "classfile/systemDictionary.hpp"
28#include "classfile/vmSymbols.hpp"
29#include "code/icBuffer.hpp"
30#include "code/vtableStubs.hpp"
31#include "compiler/compileBroker.hpp"
32#include "compiler/disassembler.hpp"
33#include "interpreter/interpreter.hpp"
34#include "jvm_solaris.h"
35#include "logging/log.hpp"
36#include "memory/allocation.inline.hpp"
37#include "memory/filemap.hpp"
38#include "mutex_solaris.inline.hpp"
39#include "oops/oop.inline.hpp"
40#include "os_share_solaris.hpp"
41#include "os_solaris.inline.hpp"
42#include "prims/jniFastGetField.hpp"
43#include "prims/jvm.h"
44#include "prims/jvm_misc.hpp"
45#include "runtime/arguments.hpp"
46#include "runtime/atomic.inline.hpp"
47#include "runtime/extendedPC.hpp"
48#include "runtime/globals.hpp"
49#include "runtime/interfaceSupport.hpp"
50#include "runtime/java.hpp"
51#include "runtime/javaCalls.hpp"
52#include "runtime/mutexLocker.hpp"
53#include "runtime/objectMonitor.hpp"
54#include "runtime/orderAccess.inline.hpp"
55#include "runtime/osThread.hpp"
56#include "runtime/perfMemory.hpp"
57#include "runtime/sharedRuntime.hpp"
58#include "runtime/statSampler.hpp"
59#include "runtime/stubRoutines.hpp"
60#include "runtime/thread.inline.hpp"
61#include "runtime/threadCritical.hpp"
62#include "runtime/timer.hpp"
63#include "runtime/vm_version.hpp"
64#include "semaphore_posix.hpp"
65#include "services/attachListener.hpp"
66#include "services/memTracker.hpp"
67#include "services/runtimeService.hpp"
68#include "utilities/decoder.hpp"
69#include "utilities/defaultStream.hpp"
70#include "utilities/events.hpp"
71#include "utilities/growableArray.hpp"
72#include "utilities/macros.hpp"
73#include "utilities/vmError.hpp"
74
75// put OS-includes here
76# include <dlfcn.h>
77# include <errno.h>
78# include <exception>
79# include <link.h>
80# include <poll.h>
81# include <pthread.h>
82# include <pwd.h>
83# include <schedctl.h>
84# include <setjmp.h>
85# include <signal.h>
86# include <stdio.h>
87# include <alloca.h>
88# include <sys/filio.h>
89# include <sys/ipc.h>
90# include <sys/lwp.h>
91# include <sys/machelf.h>     // for elf Sym structure used by dladdr1
92# include <sys/mman.h>
93# include <sys/processor.h>
94# include <sys/procset.h>
95# include <sys/pset.h>
96# include <sys/resource.h>
97# include <sys/shm.h>
98# include <sys/socket.h>
99# include <sys/stat.h>
100# include <sys/systeminfo.h>
101# include <sys/time.h>
102# include <sys/times.h>
103# include <sys/types.h>
104# include <sys/wait.h>
105# include <sys/utsname.h>
106# include <thread.h>
107# include <unistd.h>
108# include <sys/priocntl.h>
109# include <sys/rtpriocntl.h>
110# include <sys/tspriocntl.h>
111# include <sys/iapriocntl.h>
112# include <sys/fxpriocntl.h>
113# include <sys/loadavg.h>
114# include <string.h>
115# include <stdio.h>
116
117# define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
118# include <sys/procfs.h>     //  see comment in <sys/procfs.h>
119
120#define MAX_PATH (2 * K)
121
122// for timer info max values which include all bits
123#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
124
125
126// Here are some liblgrp types from sys/lgrp_user.h to be able to
127// compile on older systems without this header file.
128
129#ifndef MADV_ACCESS_LWP
130  #define  MADV_ACCESS_LWP   7       /* next LWP to access heavily */
131#endif
132#ifndef MADV_ACCESS_MANY
133  #define  MADV_ACCESS_MANY  8       /* many processes to access heavily */
134#endif
135
136#ifndef LGRP_RSRC_CPU
137  #define LGRP_RSRC_CPU      0       /* CPU resources */
138#endif
139#ifndef LGRP_RSRC_MEM
140  #define LGRP_RSRC_MEM      1       /* memory resources */
141#endif
142
143// Values for ThreadPriorityPolicy == 1
144int prio_policy1[CriticalPriority+1] = {
145  -99999,  0, 16,  32,  48,  64,
146          80, 96, 112, 124, 127, 127 };
147
148// System parameters used internally
149static clock_t clock_tics_per_sec = 100;
150
151// Track if we have called enable_extended_FILE_stdio (on Solaris 10u4+)
152static bool enabled_extended_FILE_stdio = false;
153
154// For diagnostics to print a message once. see run_periodic_checks
155static bool check_addr0_done = false;
156static sigset_t check_signal_done;
157static bool check_signals = true;
158
159address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
160address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
161
162address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
163
164os::Solaris::pthread_setname_np_func_t os::Solaris::_pthread_setname_np = NULL;
165
166// "default" initializers for missing libc APIs
167extern "C" {
168  static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
169  static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
170
171  static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
172  static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
173}
174
175// "default" initializers for pthread-based synchronization
176extern "C" {
177  static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
178  static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
179}
180
181static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
182
183static inline size_t adjust_stack_size(address base, size_t size) {
184  if ((ssize_t)size < 0) {
185    // 4759953: Compensate for ridiculous stack size.
186    size = max_intx;
187  }
188  if (size > (size_t)base) {
189    // 4812466: Make sure size doesn't allow the stack to wrap the address space.
190    size = (size_t)base;
191  }
192  return size;
193}
194
195static inline stack_t get_stack_info() {
196  stack_t st;
197  int retval = thr_stksegment(&st);
198  st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
199  assert(retval == 0, "incorrect return value from thr_stksegment");
200  assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
201  assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
202  return st;
203}
204
205address os::current_stack_base() {
206  int r = thr_main();
207  guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
208  bool is_primordial_thread = r;
209
210  // Workaround 4352906, avoid calls to thr_stksegment by
211  // thr_main after the first one (it looks like we trash
212  // some data, causing the value for ss_sp to be incorrect).
213  if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
214    stack_t st = get_stack_info();
215    if (is_primordial_thread) {
216      // cache initial value of stack base
217      os::Solaris::_main_stack_base = (address)st.ss_sp;
218    }
219    return (address)st.ss_sp;
220  } else {
221    guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
222    return os::Solaris::_main_stack_base;
223  }
224}
225
226size_t os::current_stack_size() {
227  size_t size;
228
229  int r = thr_main();
230  guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
231  if (!r) {
232    size = get_stack_info().ss_size;
233  } else {
234    struct rlimit limits;
235    getrlimit(RLIMIT_STACK, &limits);
236    size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
237  }
238  // base may not be page aligned
239  address base = current_stack_base();
240  address bottom = (address)align_size_up((intptr_t)(base - size), os::vm_page_size());;
241  return (size_t)(base - bottom);
242}
243
244struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
245  return localtime_r(clock, res);
246}
247
248void os::Solaris::try_enable_extended_io() {
249  typedef int (*enable_extended_FILE_stdio_t)(int, int);
250
251  if (!UseExtendedFileIO) {
252    return;
253  }
254
255  enable_extended_FILE_stdio_t enabler =
256    (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
257                                         "enable_extended_FILE_stdio");
258  if (enabler) {
259    enabler(-1, -1);
260  }
261}
262
263static int _processors_online = 0;
264
265jint os::Solaris::_os_thread_limit = 0;
266volatile jint os::Solaris::_os_thread_count = 0;
267
268julong os::available_memory() {
269  return Solaris::available_memory();
270}
271
272julong os::Solaris::available_memory() {
273  return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
274}
275
276julong os::Solaris::_physical_memory = 0;
277
278julong os::physical_memory() {
279  return Solaris::physical_memory();
280}
281
282static hrtime_t first_hrtime = 0;
283static const hrtime_t hrtime_hz = 1000*1000*1000;
284static volatile hrtime_t max_hrtime = 0;
285
286
287void os::Solaris::initialize_system_info() {
288  set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
289  _processors_online = sysconf(_SC_NPROCESSORS_ONLN);
290  _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) *
291                                     (julong)sysconf(_SC_PAGESIZE);
292}
293
294int os::active_processor_count() {
295  int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
296  pid_t pid = getpid();
297  psetid_t pset = PS_NONE;
298  // Are we running in a processor set or is there any processor set around?
299  if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
300    uint_t pset_cpus;
301    // Query the number of cpus available to us.
302    if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
303      assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
304      _processors_online = pset_cpus;
305      return pset_cpus;
306    }
307  }
308  // Otherwise return number of online cpus
309  return online_cpus;
310}
311
312static bool find_processors_in_pset(psetid_t        pset,
313                                    processorid_t** id_array,
314                                    uint_t*         id_length) {
315  bool result = false;
316  // Find the number of processors in the processor set.
317  if (pset_info(pset, NULL, id_length, NULL) == 0) {
318    // Make up an array to hold their ids.
319    *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
320    // Fill in the array with their processor ids.
321    if (pset_info(pset, NULL, id_length, *id_array) == 0) {
322      result = true;
323    }
324  }
325  return result;
326}
327
328// Callers of find_processors_online() must tolerate imprecise results --
329// the system configuration can change asynchronously because of DR
330// or explicit psradm operations.
331//
332// We also need to take care that the loop (below) terminates as the
333// number of processors online can change between the _SC_NPROCESSORS_ONLN
334// request and the loop that builds the list of processor ids.   Unfortunately
335// there's no reliable way to determine the maximum valid processor id,
336// so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
337// man pages, which claim the processor id set is "sparse, but
338// not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
339// exit the loop.
340//
341// In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
342// not available on S8.0.
343
344static bool find_processors_online(processorid_t** id_array,
345                                   uint*           id_length) {
346  const processorid_t MAX_PROCESSOR_ID = 100000;
347  // Find the number of processors online.
348  *id_length = sysconf(_SC_NPROCESSORS_ONLN);
349  // Make up an array to hold their ids.
350  *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
351  // Processors need not be numbered consecutively.
352  long found = 0;
353  processorid_t next = 0;
354  while (found < *id_length && next < MAX_PROCESSOR_ID) {
355    processor_info_t info;
356    if (processor_info(next, &info) == 0) {
357      // NB, PI_NOINTR processors are effectively online ...
358      if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
359        (*id_array)[found] = next;
360        found += 1;
361      }
362    }
363    next += 1;
364  }
365  if (found < *id_length) {
366    // The loop above didn't identify the expected number of processors.
367    // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
368    // and re-running the loop, above, but there's no guarantee of progress
369    // if the system configuration is in flux.  Instead, we just return what
370    // we've got.  Note that in the worst case find_processors_online() could
371    // return an empty set.  (As a fall-back in the case of the empty set we
372    // could just return the ID of the current processor).
373    *id_length = found;
374  }
375
376  return true;
377}
378
379static bool assign_distribution(processorid_t* id_array,
380                                uint           id_length,
381                                uint*          distribution,
382                                uint           distribution_length) {
383  // We assume we can assign processorid_t's to uint's.
384  assert(sizeof(processorid_t) == sizeof(uint),
385         "can't convert processorid_t to uint");
386  // Quick check to see if we won't succeed.
387  if (id_length < distribution_length) {
388    return false;
389  }
390  // Assign processor ids to the distribution.
391  // Try to shuffle processors to distribute work across boards,
392  // assuming 4 processors per board.
393  const uint processors_per_board = ProcessDistributionStride;
394  // Find the maximum processor id.
395  processorid_t max_id = 0;
396  for (uint m = 0; m < id_length; m += 1) {
397    max_id = MAX2(max_id, id_array[m]);
398  }
399  // The next id, to limit loops.
400  const processorid_t limit_id = max_id + 1;
401  // Make up markers for available processors.
402  bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id, mtInternal);
403  for (uint c = 0; c < limit_id; c += 1) {
404    available_id[c] = false;
405  }
406  for (uint a = 0; a < id_length; a += 1) {
407    available_id[id_array[a]] = true;
408  }
409  // Step by "boards", then by "slot", copying to "assigned".
410  // NEEDS_CLEANUP: The assignment of processors should be stateful,
411  //                remembering which processors have been assigned by
412  //                previous calls, etc., so as to distribute several
413  //                independent calls of this method.  What we'd like is
414  //                It would be nice to have an API that let us ask
415  //                how many processes are bound to a processor,
416  //                but we don't have that, either.
417  //                In the short term, "board" is static so that
418  //                subsequent distributions don't all start at board 0.
419  static uint board = 0;
420  uint assigned = 0;
421  // Until we've found enough processors ....
422  while (assigned < distribution_length) {
423    // ... find the next available processor in the board.
424    for (uint slot = 0; slot < processors_per_board; slot += 1) {
425      uint try_id = board * processors_per_board + slot;
426      if ((try_id < limit_id) && (available_id[try_id] == true)) {
427        distribution[assigned] = try_id;
428        available_id[try_id] = false;
429        assigned += 1;
430        break;
431      }
432    }
433    board += 1;
434    if (board * processors_per_board + 0 >= limit_id) {
435      board = 0;
436    }
437  }
438  if (available_id != NULL) {
439    FREE_C_HEAP_ARRAY(bool, available_id);
440  }
441  return true;
442}
443
444void os::set_native_thread_name(const char *name) {
445  if (Solaris::_pthread_setname_np != NULL) {
446    // Only the first 31 bytes of 'name' are processed by pthread_setname_np
447    // but we explicitly copy into a size-limited buffer to avoid any
448    // possible overflow.
449    char buf[32];
450    snprintf(buf, sizeof(buf), "%s", name);
451    buf[sizeof(buf) - 1] = '\0';
452    Solaris::_pthread_setname_np(pthread_self(), buf);
453  }
454}
455
456bool os::distribute_processes(uint length, uint* distribution) {
457  bool result = false;
458  // Find the processor id's of all the available CPUs.
459  processorid_t* id_array  = NULL;
460  uint           id_length = 0;
461  // There are some races between querying information and using it,
462  // since processor sets can change dynamically.
463  psetid_t pset = PS_NONE;
464  // Are we running in a processor set?
465  if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
466    result = find_processors_in_pset(pset, &id_array, &id_length);
467  } else {
468    result = find_processors_online(&id_array, &id_length);
469  }
470  if (result == true) {
471    if (id_length >= length) {
472      result = assign_distribution(id_array, id_length, distribution, length);
473    } else {
474      result = false;
475    }
476  }
477  if (id_array != NULL) {
478    FREE_C_HEAP_ARRAY(processorid_t, id_array);
479  }
480  return result;
481}
482
483bool os::bind_to_processor(uint processor_id) {
484  // We assume that a processorid_t can be stored in a uint.
485  assert(sizeof(uint) == sizeof(processorid_t),
486         "can't convert uint to processorid_t");
487  int bind_result =
488    processor_bind(P_LWPID,                       // bind LWP.
489                   P_MYID,                        // bind current LWP.
490                   (processorid_t) processor_id,  // id.
491                   NULL);                         // don't return old binding.
492  return (bind_result == 0);
493}
494
495// Return true if user is running as root.
496
497bool os::have_special_privileges() {
498  static bool init = false;
499  static bool privileges = false;
500  if (!init) {
501    privileges = (getuid() != geteuid()) || (getgid() != getegid());
502    init = true;
503  }
504  return privileges;
505}
506
507
508void os::init_system_properties_values() {
509  // The next steps are taken in the product version:
510  //
511  // Obtain the JAVA_HOME value from the location of libjvm.so.
512  // This library should be located at:
513  // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
514  //
515  // If "/jre/lib/" appears at the right place in the path, then we
516  // assume libjvm.so is installed in a JDK and we use this path.
517  //
518  // Otherwise exit with message: "Could not create the Java virtual machine."
519  //
520  // The following extra steps are taken in the debugging version:
521  //
522  // If "/jre/lib/" does NOT appear at the right place in the path
523  // instead of exit check for $JAVA_HOME environment variable.
524  //
525  // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
526  // then we append a fake suffix "hotspot/libjvm.so" to this path so
527  // it looks like libjvm.so is installed there
528  // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
529  //
530  // Otherwise exit.
531  //
532  // Important note: if the location of libjvm.so changes this
533  // code needs to be changed accordingly.
534
535// Base path of extensions installed on the system.
536#define SYS_EXT_DIR     "/usr/jdk/packages"
537#define EXTENSIONS_DIR  "/lib/ext"
538
539  char cpu_arch[12];
540  // Buffer that fits several sprintfs.
541  // Note that the space for the colon and the trailing null are provided
542  // by the nulls included by the sizeof operator.
543  const size_t bufsize =
544    MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
545         sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch), // invariant ld_library_path
546         (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
547  char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
548
549  // sysclasspath, java_home, dll_dir
550  {
551    char *pslash;
552    os::jvm_path(buf, bufsize);
553
554    // Found the full path to libjvm.so.
555    // Now cut the path to <java_home>/jre if we can.
556    *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
557    pslash = strrchr(buf, '/');
558    if (pslash != NULL) {
559      *pslash = '\0';            // Get rid of /{client|server|hotspot}.
560    }
561    Arguments::set_dll_dir(buf);
562
563    if (pslash != NULL) {
564      pslash = strrchr(buf, '/');
565      if (pslash != NULL) {
566        *pslash = '\0';          // Get rid of /<arch>.
567        pslash = strrchr(buf, '/');
568        if (pslash != NULL) {
569          *pslash = '\0';        // Get rid of /lib.
570        }
571      }
572    }
573    Arguments::set_java_home(buf);
574    set_boot_path('/', ':');
575  }
576
577  // Where to look for native libraries.
578  {
579    // Use dlinfo() to determine the correct java.library.path.
580    //
581    // If we're launched by the Java launcher, and the user
582    // does not set java.library.path explicitly on the commandline,
583    // the Java launcher sets LD_LIBRARY_PATH for us and unsets
584    // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
585    // dlinfo returns LD_LIBRARY_PATH + crle settings (including
586    // /usr/lib), which is exactly what we want.
587    //
588    // If the user does set java.library.path, it completely
589    // overwrites this setting, and always has.
590    //
591    // If we're not launched by the Java launcher, we may
592    // get here with any/all of the LD_LIBRARY_PATH[_32|64]
593    // settings.  Again, dlinfo does exactly what we want.
594
595    Dl_serinfo     info_sz, *info = &info_sz;
596    Dl_serpath     *path;
597    char           *library_path;
598    char           *common_path = buf;
599
600    // Determine search path count and required buffer size.
601    if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
602      FREE_C_HEAP_ARRAY(char, buf);
603      vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
604    }
605
606    // Allocate new buffer and initialize.
607    info = (Dl_serinfo*)NEW_C_HEAP_ARRAY(char, info_sz.dls_size, mtInternal);
608    info->dls_size = info_sz.dls_size;
609    info->dls_cnt = info_sz.dls_cnt;
610
611    // Obtain search path information.
612    if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
613      FREE_C_HEAP_ARRAY(char, buf);
614      FREE_C_HEAP_ARRAY(char, info);
615      vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
616    }
617
618    path = &info->dls_serpath[0];
619
620    // Note: Due to a legacy implementation, most of the library path
621    // is set in the launcher. This was to accomodate linking restrictions
622    // on legacy Solaris implementations (which are no longer supported).
623    // Eventually, all the library path setting will be done here.
624    //
625    // However, to prevent the proliferation of improperly built native
626    // libraries, the new path component /usr/jdk/packages is added here.
627
628    // Determine the actual CPU architecture.
629    sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
630#ifdef _LP64
631    // If we are a 64-bit vm, perform the following translations:
632    //   sparc   -> sparcv9
633    //   i386    -> amd64
634    if (strcmp(cpu_arch, "sparc") == 0) {
635      strcat(cpu_arch, "v9");
636    } else if (strcmp(cpu_arch, "i386") == 0) {
637      strcpy(cpu_arch, "amd64");
638    }
639#endif
640
641    // Construct the invariant part of ld_library_path.
642    sprintf(common_path, SYS_EXT_DIR "/lib/%s", cpu_arch);
643
644    // Struct size is more than sufficient for the path components obtained
645    // through the dlinfo() call, so only add additional space for the path
646    // components explicitly added here.
647    size_t library_path_size = info->dls_size + strlen(common_path);
648    library_path = (char *)NEW_C_HEAP_ARRAY(char, library_path_size, mtInternal);
649    library_path[0] = '\0';
650
651    // Construct the desired Java library path from the linker's library
652    // search path.
653    //
654    // For compatibility, it is optimal that we insert the additional path
655    // components specific to the Java VM after those components specified
656    // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
657    // infrastructure.
658    if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it.
659      strcpy(library_path, common_path);
660    } else {
661      int inserted = 0;
662      int i;
663      for (i = 0; i < info->dls_cnt; i++, path++) {
664        uint_t flags = path->dls_flags & LA_SER_MASK;
665        if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
666          strcat(library_path, common_path);
667          strcat(library_path, os::path_separator());
668          inserted = 1;
669        }
670        strcat(library_path, path->dls_name);
671        strcat(library_path, os::path_separator());
672      }
673      // Eliminate trailing path separator.
674      library_path[strlen(library_path)-1] = '\0';
675    }
676
677    // happens before argument parsing - can't use a trace flag
678    // tty->print_raw("init_system_properties_values: native lib path: ");
679    // tty->print_raw_cr(library_path);
680
681    // Callee copies into its own buffer.
682    Arguments::set_library_path(library_path);
683
684    FREE_C_HEAP_ARRAY(char, library_path);
685    FREE_C_HEAP_ARRAY(char, info);
686  }
687
688  // Extensions directories.
689  sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
690  Arguments::set_ext_dirs(buf);
691
692  FREE_C_HEAP_ARRAY(char, buf);
693
694#undef SYS_EXT_DIR
695#undef EXTENSIONS_DIR
696}
697
698void os::breakpoint() {
699  BREAKPOINT;
700}
701
702bool os::obsolete_option(const JavaVMOption *option) {
703  if (!strncmp(option->optionString, "-Xt", 3)) {
704    return true;
705  } else if (!strncmp(option->optionString, "-Xtm", 4)) {
706    return true;
707  } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
708    return true;
709  } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
710    return true;
711  }
712  return false;
713}
714
715bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
716  address  stackStart  = (address)thread->stack_base();
717  address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
718  if (sp < stackStart && sp >= stackEnd) return true;
719  return false;
720}
721
722extern "C" void breakpoint() {
723  // use debugger to set breakpoint here
724}
725
726static thread_t main_thread;
727
728// Thread start routine for all newly created threads
729extern "C" void* thread_native_entry(void* thread_addr) {
730  // Try to randomize the cache line index of hot stack frames.
731  // This helps when threads of the same stack traces evict each other's
732  // cache lines. The threads can be either from the same JVM instance, or
733  // from different JVM instances. The benefit is especially true for
734  // processors with hyperthreading technology.
735  static int counter = 0;
736  int pid = os::current_process_id();
737  alloca(((pid ^ counter++) & 7) * 128);
738
739  int prio;
740  Thread* thread = (Thread*)thread_addr;
741
742  thread->initialize_thread_current();
743
744  OSThread* osthr = thread->osthread();
745
746  osthr->set_lwp_id(_lwp_self());  // Store lwp in case we are bound
747  thread->_schedctl = (void *) schedctl_init();
748
749  log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ").",
750    os::current_thread_id());
751
752  if (UseNUMA) {
753    int lgrp_id = os::numa_get_group_id();
754    if (lgrp_id != -1) {
755      thread->set_lgrp_id(lgrp_id);
756    }
757  }
758
759  // Our priority was set when we were created, and stored in the
760  // osthread, but couldn't be passed through to our LWP until now.
761  // So read back the priority and set it again.
762
763  if (osthr->thread_id() != -1) {
764    if (UseThreadPriorities) {
765      int prio = osthr->native_priority();
766      if (ThreadPriorityVerbose) {
767        tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
768                      INTPTR_FORMAT ", setting priority: %d\n",
769                      osthr->thread_id(), osthr->lwp_id(), prio);
770      }
771      os::set_native_priority(thread, prio);
772    }
773  } else if (ThreadPriorityVerbose) {
774    warning("Can't set priority in _start routine, thread id hasn't been set\n");
775  }
776
777  assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
778
779  // initialize signal mask for this thread
780  os::Solaris::hotspot_sigmask(thread);
781
782  thread->run();
783
784  // One less thread is executing
785  // When the VMThread gets here, the main thread may have already exited
786  // which frees the CodeHeap containing the Atomic::dec code
787  if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
788    Atomic::dec(&os::Solaris::_os_thread_count);
789  }
790
791  log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ").", os::current_thread_id());
792
793  // If a thread has not deleted itself ("delete this") as part of its
794  // termination sequence, we have to ensure thread-local-storage is
795  // cleared before we actually terminate. No threads should ever be
796  // deleted asynchronously with respect to their termination.
797  if (Thread::current_or_null_safe() != NULL) {
798    assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
799    thread->clear_thread_current();
800  }
801
802  if (UseDetachedThreads) {
803    thr_exit(NULL);
804    ShouldNotReachHere();
805  }
806  return NULL;
807}
808
809static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
810  // Allocate the OSThread object
811  OSThread* osthread = new OSThread(NULL, NULL);
812  if (osthread == NULL) return NULL;
813
814  // Store info on the Solaris thread into the OSThread
815  osthread->set_thread_id(thread_id);
816  osthread->set_lwp_id(_lwp_self());
817  thread->_schedctl = (void *) schedctl_init();
818
819  if (UseNUMA) {
820    int lgrp_id = os::numa_get_group_id();
821    if (lgrp_id != -1) {
822      thread->set_lgrp_id(lgrp_id);
823    }
824  }
825
826  if (ThreadPriorityVerbose) {
827    tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
828                  osthread->thread_id(), osthread->lwp_id());
829  }
830
831  // Initial thread state is INITIALIZED, not SUSPENDED
832  osthread->set_state(INITIALIZED);
833
834  return osthread;
835}
836
837void os::Solaris::hotspot_sigmask(Thread* thread) {
838  //Save caller's signal mask
839  sigset_t sigmask;
840  pthread_sigmask(SIG_SETMASK, NULL, &sigmask);
841  OSThread *osthread = thread->osthread();
842  osthread->set_caller_sigmask(sigmask);
843
844  pthread_sigmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
845  if (!ReduceSignalUsage) {
846    if (thread->is_VM_thread()) {
847      // Only the VM thread handles BREAK_SIGNAL ...
848      pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
849    } else {
850      // ... all other threads block BREAK_SIGNAL
851      assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
852      pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
853    }
854  }
855}
856
857bool os::create_attached_thread(JavaThread* thread) {
858#ifdef ASSERT
859  thread->verify_not_published();
860#endif
861  OSThread* osthread = create_os_thread(thread, thr_self());
862  if (osthread == NULL) {
863    return false;
864  }
865
866  // Initial thread state is RUNNABLE
867  osthread->set_state(RUNNABLE);
868  thread->set_osthread(osthread);
869
870  // initialize signal mask for this thread
871  // and save the caller's signal mask
872  os::Solaris::hotspot_sigmask(thread);
873
874  log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ").",
875    os::current_thread_id());
876
877  return true;
878}
879
880bool os::create_main_thread(JavaThread* thread) {
881#ifdef ASSERT
882  thread->verify_not_published();
883#endif
884  if (_starting_thread == NULL) {
885    _starting_thread = create_os_thread(thread, main_thread);
886    if (_starting_thread == NULL) {
887      return false;
888    }
889  }
890
891  // The primodial thread is runnable from the start
892  _starting_thread->set_state(RUNNABLE);
893
894  thread->set_osthread(_starting_thread);
895
896  // initialize signal mask for this thread
897  // and save the caller's signal mask
898  os::Solaris::hotspot_sigmask(thread);
899
900  return true;
901}
902
903// Helper function to trace thread attributes, similar to os::Posix::describe_pthread_attr()
904static char* describe_thr_create_attributes(char* buf, size_t buflen,
905                                            size_t stacksize, long flags) {
906  stringStream ss(buf, buflen);
907  ss.print("stacksize: " SIZE_FORMAT "k, ", stacksize / 1024);
908  ss.print("flags: ");
909  #define PRINT_FLAG(f) if (flags & f) ss.print( #f " ");
910  #define ALL(X) \
911    X(THR_SUSPENDED) \
912    X(THR_DETACHED) \
913    X(THR_BOUND) \
914    X(THR_NEW_LWP) \
915    X(THR_DAEMON)
916  ALL(PRINT_FLAG)
917  #undef ALL
918  #undef PRINT_FLAG
919  return buf;
920}
921
922bool os::create_thread(Thread* thread, ThreadType thr_type,
923                       size_t stack_size) {
924  // Allocate the OSThread object
925  OSThread* osthread = new OSThread(NULL, NULL);
926  if (osthread == NULL) {
927    return false;
928  }
929
930  if (ThreadPriorityVerbose) {
931    char *thrtyp;
932    switch (thr_type) {
933    case vm_thread:
934      thrtyp = (char *)"vm";
935      break;
936    case cgc_thread:
937      thrtyp = (char *)"cgc";
938      break;
939    case pgc_thread:
940      thrtyp = (char *)"pgc";
941      break;
942    case java_thread:
943      thrtyp = (char *)"java";
944      break;
945    case compiler_thread:
946      thrtyp = (char *)"compiler";
947      break;
948    case watcher_thread:
949      thrtyp = (char *)"watcher";
950      break;
951    default:
952      thrtyp = (char *)"unknown";
953      break;
954    }
955    tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
956  }
957
958  // Calculate stack size if it's not specified by caller.
959  if (stack_size == 0) {
960    // The default stack size 1M (2M for LP64).
961    stack_size = (BytesPerWord >> 2) * K * K;
962
963    switch (thr_type) {
964    case os::java_thread:
965      // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
966      if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
967      break;
968    case os::compiler_thread:
969      if (CompilerThreadStackSize > 0) {
970        stack_size = (size_t)(CompilerThreadStackSize * K);
971        break;
972      } // else fall through:
973        // use VMThreadStackSize if CompilerThreadStackSize is not defined
974    case os::vm_thread:
975    case os::pgc_thread:
976    case os::cgc_thread:
977    case os::watcher_thread:
978      if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
979      break;
980    }
981  }
982  stack_size = MAX2(stack_size, os::Solaris::min_stack_allowed);
983
984  // Initial state is ALLOCATED but not INITIALIZED
985  osthread->set_state(ALLOCATED);
986
987  if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
988    // We got lots of threads. Check if we still have some address space left.
989    // Need to be at least 5Mb of unreserved address space. We do check by
990    // trying to reserve some.
991    const size_t VirtualMemoryBangSize = 20*K*K;
992    char* mem = os::reserve_memory(VirtualMemoryBangSize);
993    if (mem == NULL) {
994      delete osthread;
995      return false;
996    } else {
997      // Release the memory again
998      os::release_memory(mem, VirtualMemoryBangSize);
999    }
1000  }
1001
1002  // Setup osthread because the child thread may need it.
1003  thread->set_osthread(osthread);
1004
1005  // Create the Solaris thread
1006  thread_t tid = 0;
1007  long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED;
1008  int      status;
1009
1010  // Mark that we don't have an lwp or thread id yet.
1011  // In case we attempt to set the priority before the thread starts.
1012  osthread->set_lwp_id(-1);
1013  osthread->set_thread_id(-1);
1014
1015  status = thr_create(NULL, stack_size, thread_native_entry, thread, flags, &tid);
1016
1017  char buf[64];
1018  if (status == 0) {
1019    log_info(os, thread)("Thread started (tid: " UINTX_FORMAT ", attributes: %s). ",
1020      (uintx) tid, describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
1021  } else {
1022    log_warning(os, thread)("Failed to start thread - thr_create failed (%s) for attributes: %s.",
1023      os::errno_name(status), describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
1024  }
1025
1026  if (status != 0) {
1027    thread->set_osthread(NULL);
1028    // Need to clean up stuff we've allocated so far
1029    delete osthread;
1030    return false;
1031  }
1032
1033  Atomic::inc(&os::Solaris::_os_thread_count);
1034
1035  // Store info on the Solaris thread into the OSThread
1036  osthread->set_thread_id(tid);
1037
1038  // Remember that we created this thread so we can set priority on it
1039  osthread->set_vm_created();
1040
1041  // Most thread types will set an explicit priority before starting the thread,
1042  // but for those that don't we need a valid value to read back in thread_native_entry.
1043  osthread->set_native_priority(NormPriority);
1044
1045  // Initial thread state is INITIALIZED, not SUSPENDED
1046  osthread->set_state(INITIALIZED);
1047
1048  // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1049  return true;
1050}
1051
1052// defined for >= Solaris 10. This allows builds on earlier versions
1053// of Solaris to take advantage of the newly reserved Solaris JVM signals.
1054// With SIGJVM1, SIGJVM2, ASYNC_SIGNAL is SIGJVM2. Previously INTERRUPT_SIGNAL
1055// was SIGJVM1.
1056//
1057#if !defined(SIGJVM1)
1058  #define SIGJVM1 39
1059  #define SIGJVM2 40
1060#endif
1061
1062debug_only(static bool signal_sets_initialized = false);
1063static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
1064
1065int os::Solaris::_SIGasync = ASYNC_SIGNAL;
1066
1067bool os::Solaris::is_sig_ignored(int sig) {
1068  struct sigaction oact;
1069  sigaction(sig, (struct sigaction*)NULL, &oact);
1070  void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1071                                 : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1072  if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
1073    return true;
1074  } else {
1075    return false;
1076  }
1077}
1078
1079// Note: SIGRTMIN is a macro that calls sysconf() so it will
1080// dynamically detect SIGRTMIN value for the system at runtime, not buildtime
1081static bool isJVM1available() {
1082  return SIGJVM1 < SIGRTMIN;
1083}
1084
1085void os::Solaris::signal_sets_init() {
1086  // Should also have an assertion stating we are still single-threaded.
1087  assert(!signal_sets_initialized, "Already initialized");
1088  // Fill in signals that are necessarily unblocked for all threads in
1089  // the VM. Currently, we unblock the following signals:
1090  // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1091  //                         by -Xrs (=ReduceSignalUsage));
1092  // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1093  // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1094  // the dispositions or masks wrt these signals.
1095  // Programs embedding the VM that want to use the above signals for their
1096  // own purposes must, at this time, use the "-Xrs" option to prevent
1097  // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1098  // (See bug 4345157, and other related bugs).
1099  // In reality, though, unblocking these signals is really a nop, since
1100  // these signals are not blocked by default.
1101  sigemptyset(&unblocked_sigs);
1102  sigemptyset(&allowdebug_blocked_sigs);
1103  sigaddset(&unblocked_sigs, SIGILL);
1104  sigaddset(&unblocked_sigs, SIGSEGV);
1105  sigaddset(&unblocked_sigs, SIGBUS);
1106  sigaddset(&unblocked_sigs, SIGFPE);
1107
1108  // Always true on Solaris 10+
1109  guarantee(isJVM1available(), "SIGJVM1/2 missing!");
1110  os::Solaris::set_SIGasync(SIGJVM2);
1111
1112  sigaddset(&unblocked_sigs, os::Solaris::SIGasync());
1113
1114  if (!ReduceSignalUsage) {
1115    if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1116      sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1117      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
1118    }
1119    if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1120      sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1121      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
1122    }
1123    if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1124      sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1125      sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
1126    }
1127  }
1128  // Fill in signals that are blocked by all but the VM thread.
1129  sigemptyset(&vm_sigs);
1130  if (!ReduceSignalUsage) {
1131    sigaddset(&vm_sigs, BREAK_SIGNAL);
1132  }
1133  debug_only(signal_sets_initialized = true);
1134
1135  // For diagnostics only used in run_periodic_checks
1136  sigemptyset(&check_signal_done);
1137}
1138
1139// These are signals that are unblocked while a thread is running Java.
1140// (For some reason, they get blocked by default.)
1141sigset_t* os::Solaris::unblocked_signals() {
1142  assert(signal_sets_initialized, "Not initialized");
1143  return &unblocked_sigs;
1144}
1145
1146// These are the signals that are blocked while a (non-VM) thread is
1147// running Java. Only the VM thread handles these signals.
1148sigset_t* os::Solaris::vm_signals() {
1149  assert(signal_sets_initialized, "Not initialized");
1150  return &vm_sigs;
1151}
1152
1153// These are signals that are blocked during cond_wait to allow debugger in
1154sigset_t* os::Solaris::allowdebug_blocked_signals() {
1155  assert(signal_sets_initialized, "Not initialized");
1156  return &allowdebug_blocked_sigs;
1157}
1158
1159
1160void _handle_uncaught_cxx_exception() {
1161  VMError::report_and_die("An uncaught C++ exception");
1162}
1163
1164
1165// First crack at OS-specific initialization, from inside the new thread.
1166void os::initialize_thread(Thread* thr) {
1167  int r = thr_main();
1168  guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
1169  if (r) {
1170    JavaThread* jt = (JavaThread *)thr;
1171    assert(jt != NULL, "Sanity check");
1172    size_t stack_size;
1173    address base = jt->stack_base();
1174    if (Arguments::created_by_java_launcher()) {
1175      // Use 2MB to allow for Solaris 7 64 bit mode.
1176      stack_size = JavaThread::stack_size_at_create() == 0
1177        ? 2048*K : JavaThread::stack_size_at_create();
1178
1179      // There are rare cases when we may have already used more than
1180      // the basic stack size allotment before this method is invoked.
1181      // Attempt to allow for a normally sized java_stack.
1182      size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1183      stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1184    } else {
1185      // 6269555: If we were not created by a Java launcher, i.e. if we are
1186      // running embedded in a native application, treat the primordial thread
1187      // as much like a native attached thread as possible.  This means using
1188      // the current stack size from thr_stksegment(), unless it is too large
1189      // to reliably setup guard pages.  A reasonable max size is 8MB.
1190      size_t current_size = current_stack_size();
1191      // This should never happen, but just in case....
1192      if (current_size == 0) current_size = 2 * K * K;
1193      stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1194    }
1195    address bottom = (address)align_size_up((intptr_t)(base - stack_size), os::vm_page_size());;
1196    stack_size = (size_t)(base - bottom);
1197
1198    assert(stack_size > 0, "Stack size calculation problem");
1199
1200    if (stack_size > jt->stack_size()) {
1201#ifndef PRODUCT
1202      struct rlimit limits;
1203      getrlimit(RLIMIT_STACK, &limits);
1204      size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1205      assert(size >= jt->stack_size(), "Stack size problem in main thread");
1206#endif
1207      tty->print_cr("Stack size of %d Kb exceeds current limit of %d Kb.\n"
1208                    "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1209                    "See limit(1) to increase the stack size limit.",
1210                    stack_size / K, jt->stack_size() / K);
1211      vm_exit(1);
1212    }
1213    assert(jt->stack_size() >= stack_size,
1214           "Attempt to map more stack than was allocated");
1215    jt->set_stack_size(stack_size);
1216  }
1217
1218  // With the T2 libthread (T1 is no longer supported) threads are always bound
1219  // and we use stackbanging in all cases.
1220
1221  os::Solaris::init_thread_fpu_state();
1222  std::set_terminate(_handle_uncaught_cxx_exception);
1223}
1224
1225
1226
1227// Free Solaris resources related to the OSThread
1228void os::free_thread(OSThread* osthread) {
1229  assert(osthread != NULL, "os::free_thread but osthread not set");
1230
1231  // We are told to free resources of the argument thread,
1232  // but we can only really operate on the current thread.
1233  assert(Thread::current()->osthread() == osthread,
1234         "os::free_thread but not current thread");
1235
1236  // Restore caller's signal mask
1237  sigset_t sigmask = osthread->caller_sigmask();
1238  pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1239
1240  delete osthread;
1241}
1242
1243void os::pd_start_thread(Thread* thread) {
1244  int status = thr_continue(thread->osthread()->thread_id());
1245  assert_status(status == 0, status, "thr_continue failed");
1246}
1247
1248
1249intx os::current_thread_id() {
1250  return (intx)thr_self();
1251}
1252
1253static pid_t _initial_pid = 0;
1254
1255int os::current_process_id() {
1256  return (int)(_initial_pid ? _initial_pid : getpid());
1257}
1258
1259// gethrtime() should be monotonic according to the documentation,
1260// but some virtualized platforms are known to break this guarantee.
1261// getTimeNanos() must be guaranteed not to move backwards, so we
1262// are forced to add a check here.
1263inline hrtime_t getTimeNanos() {
1264  const hrtime_t now = gethrtime();
1265  const hrtime_t prev = max_hrtime;
1266  if (now <= prev) {
1267    return prev;   // same or retrograde time;
1268  }
1269  const hrtime_t obsv = Atomic::cmpxchg(now, (volatile jlong*)&max_hrtime, prev);
1270  assert(obsv >= prev, "invariant");   // Monotonicity
1271  // If the CAS succeeded then we're done and return "now".
1272  // If the CAS failed and the observed value "obsv" is >= now then
1273  // we should return "obsv".  If the CAS failed and now > obsv > prv then
1274  // some other thread raced this thread and installed a new value, in which case
1275  // we could either (a) retry the entire operation, (b) retry trying to install now
1276  // or (c) just return obsv.  We use (c).   No loop is required although in some cases
1277  // we might discard a higher "now" value in deference to a slightly lower but freshly
1278  // installed obsv value.   That's entirely benign -- it admits no new orderings compared
1279  // to (a) or (b) -- and greatly reduces coherence traffic.
1280  // We might also condition (c) on the magnitude of the delta between obsv and now.
1281  // Avoiding excessive CAS operations to hot RW locations is critical.
1282  // See https://blogs.oracle.com/dave/entry/cas_and_cache_trivia_invalidate
1283  return (prev == obsv) ? now : obsv;
1284}
1285
1286// Time since start-up in seconds to a fine granularity.
1287// Used by VMSelfDestructTimer and the MemProfiler.
1288double os::elapsedTime() {
1289  return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1290}
1291
1292jlong os::elapsed_counter() {
1293  return (jlong)(getTimeNanos() - first_hrtime);
1294}
1295
1296jlong os::elapsed_frequency() {
1297  return hrtime_hz;
1298}
1299
1300// Return the real, user, and system times in seconds from an
1301// arbitrary fixed point in the past.
1302bool os::getTimesSecs(double* process_real_time,
1303                      double* process_user_time,
1304                      double* process_system_time) {
1305  struct tms ticks;
1306  clock_t real_ticks = times(&ticks);
1307
1308  if (real_ticks == (clock_t) (-1)) {
1309    return false;
1310  } else {
1311    double ticks_per_second = (double) clock_tics_per_sec;
1312    *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1313    *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1314    // For consistency return the real time from getTimeNanos()
1315    // converted to seconds.
1316    *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1317
1318    return true;
1319  }
1320}
1321
1322bool os::supports_vtime() { return true; }
1323
1324bool os::enable_vtime() {
1325  int fd = ::open("/proc/self/ctl", O_WRONLY);
1326  if (fd == -1) {
1327    return false;
1328  }
1329
1330  long cmd[] = { PCSET, PR_MSACCT };
1331  int res = ::write(fd, cmd, sizeof(long) * 2);
1332  ::close(fd);
1333  if (res != sizeof(long) * 2) {
1334    return false;
1335  }
1336  return true;
1337}
1338
1339bool os::vtime_enabled() {
1340  int fd = ::open("/proc/self/status", O_RDONLY);
1341  if (fd == -1) {
1342    return false;
1343  }
1344
1345  pstatus_t status;
1346  int res = os::read(fd, (void*) &status, sizeof(pstatus_t));
1347  ::close(fd);
1348  if (res != sizeof(pstatus_t)) {
1349    return false;
1350  }
1351  return status.pr_flags & PR_MSACCT;
1352}
1353
1354double os::elapsedVTime() {
1355  return (double)gethrvtime() / (double)hrtime_hz;
1356}
1357
1358// in-memory timestamp support - has update accuracy of 1ms
1359typedef void (*_get_nsec_fromepoch_func_t)(hrtime_t*);
1360static _get_nsec_fromepoch_func_t _get_nsec_fromepoch = NULL;
1361
1362// Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1363jlong os::javaTimeMillis() {
1364  if (_get_nsec_fromepoch != NULL) {
1365    hrtime_t now;
1366    _get_nsec_fromepoch(&now);
1367    return now / NANOSECS_PER_MILLISEC;
1368  }
1369  else {
1370    timeval t;
1371    if (gettimeofday(&t, NULL) == -1) {
1372      fatal("os::javaTimeMillis: gettimeofday (%s)", os::strerror(errno));
1373    }
1374    return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1375  }
1376}
1377
1378void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1379  timeval t;
1380  // can't use get_nsec_fromepoch here as we need better accuracy than 1ms
1381  if (gettimeofday(&t, NULL) == -1) {
1382    fatal("os::javaTimeSystemUTC: gettimeofday (%s)", os::strerror(errno));
1383  }
1384  seconds = jlong(t.tv_sec);
1385  nanos = jlong(t.tv_usec) * 1000;
1386}
1387
1388
1389jlong os::javaTimeNanos() {
1390  return (jlong)getTimeNanos();
1391}
1392
1393void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1394  info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1395  info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1396  info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1397  info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1398}
1399
1400char * os::local_time_string(char *buf, size_t buflen) {
1401  struct tm t;
1402  time_t long_time;
1403  time(&long_time);
1404  localtime_r(&long_time, &t);
1405  jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1406               t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1407               t.tm_hour, t.tm_min, t.tm_sec);
1408  return buf;
1409}
1410
1411// Note: os::shutdown() might be called very early during initialization, or
1412// called from signal handler. Before adding something to os::shutdown(), make
1413// sure it is async-safe and can handle partially initialized VM.
1414void os::shutdown() {
1415
1416  // allow PerfMemory to attempt cleanup of any persistent resources
1417  perfMemory_exit();
1418
1419  // needs to remove object in file system
1420  AttachListener::abort();
1421
1422  // flush buffered output, finish log files
1423  ostream_abort();
1424
1425  // Check for abort hook
1426  abort_hook_t abort_hook = Arguments::abort_hook();
1427  if (abort_hook != NULL) {
1428    abort_hook();
1429  }
1430}
1431
1432// Note: os::abort() might be called very early during initialization, or
1433// called from signal handler. Before adding something to os::abort(), make
1434// sure it is async-safe and can handle partially initialized VM.
1435void os::abort(bool dump_core, void* siginfo, const void* context) {
1436  os::shutdown();
1437  if (dump_core) {
1438#ifndef PRODUCT
1439    fdStream out(defaultStream::output_fd());
1440    out.print_raw("Current thread is ");
1441    char buf[16];
1442    jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1443    out.print_raw_cr(buf);
1444    out.print_raw_cr("Dumping core ...");
1445#endif
1446    ::abort(); // dump core (for debugging)
1447  }
1448
1449  ::exit(1);
1450}
1451
1452// Die immediately, no exit hook, no abort hook, no cleanup.
1453void os::die() {
1454  ::abort(); // dump core (for debugging)
1455}
1456
1457// DLL functions
1458
1459const char* os::dll_file_extension() { return ".so"; }
1460
1461// This must be hard coded because it's the system's temporary
1462// directory not the java application's temp directory, ala java.io.tmpdir.
1463const char* os::get_temp_directory() { return "/tmp"; }
1464
1465static bool file_exists(const char* filename) {
1466  struct stat statbuf;
1467  if (filename == NULL || strlen(filename) == 0) {
1468    return false;
1469  }
1470  return os::stat(filename, &statbuf) == 0;
1471}
1472
1473bool os::dll_build_name(char* buffer, size_t buflen,
1474                        const char* pname, const char* fname) {
1475  bool retval = false;
1476  const size_t pnamelen = pname ? strlen(pname) : 0;
1477
1478  // Return error on buffer overflow.
1479  if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1480    return retval;
1481  }
1482
1483  if (pnamelen == 0) {
1484    snprintf(buffer, buflen, "lib%s.so", fname);
1485    retval = true;
1486  } else if (strchr(pname, *os::path_separator()) != NULL) {
1487    int n;
1488    char** pelements = split_path(pname, &n);
1489    if (pelements == NULL) {
1490      return false;
1491    }
1492    for (int i = 0; i < n; i++) {
1493      // really shouldn't be NULL but what the heck, check can't hurt
1494      if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1495        continue; // skip the empty path values
1496      }
1497      snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1498      if (file_exists(buffer)) {
1499        retval = true;
1500        break;
1501      }
1502    }
1503    // release the storage
1504    for (int i = 0; i < n; i++) {
1505      if (pelements[i] != NULL) {
1506        FREE_C_HEAP_ARRAY(char, pelements[i]);
1507      }
1508    }
1509    if (pelements != NULL) {
1510      FREE_C_HEAP_ARRAY(char*, pelements);
1511    }
1512  } else {
1513    snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1514    retval = true;
1515  }
1516  return retval;
1517}
1518
1519// check if addr is inside libjvm.so
1520bool os::address_is_in_vm(address addr) {
1521  static address libjvm_base_addr;
1522  Dl_info dlinfo;
1523
1524  if (libjvm_base_addr == NULL) {
1525    if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1526      libjvm_base_addr = (address)dlinfo.dli_fbase;
1527    }
1528    assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1529  }
1530
1531  if (dladdr((void *)addr, &dlinfo) != 0) {
1532    if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1533  }
1534
1535  return false;
1536}
1537
1538typedef int (*dladdr1_func_type)(void *, Dl_info *, void **, int);
1539static dladdr1_func_type dladdr1_func = NULL;
1540
1541bool os::dll_address_to_function_name(address addr, char *buf,
1542                                      int buflen, int * offset,
1543                                      bool demangle) {
1544  // buf is not optional, but offset is optional
1545  assert(buf != NULL, "sanity check");
1546
1547  Dl_info dlinfo;
1548
1549  // dladdr1_func was initialized in os::init()
1550  if (dladdr1_func != NULL) {
1551    // yes, we have dladdr1
1552
1553    // Support for dladdr1 is checked at runtime; it may be
1554    // available even if the vm is built on a machine that does
1555    // not have dladdr1 support.  Make sure there is a value for
1556    // RTLD_DL_SYMENT.
1557#ifndef RTLD_DL_SYMENT
1558  #define RTLD_DL_SYMENT 1
1559#endif
1560#ifdef _LP64
1561    Elf64_Sym * info;
1562#else
1563    Elf32_Sym * info;
1564#endif
1565    if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1566                     RTLD_DL_SYMENT) != 0) {
1567      // see if we have a matching symbol that covers our address
1568      if (dlinfo.dli_saddr != NULL &&
1569          (char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1570        if (dlinfo.dli_sname != NULL) {
1571          if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1572            jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1573          }
1574          if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1575          return true;
1576        }
1577      }
1578      // no matching symbol so try for just file info
1579      if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1580        if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1581                            buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1582          return true;
1583        }
1584      }
1585    }
1586    buf[0] = '\0';
1587    if (offset != NULL) *offset  = -1;
1588    return false;
1589  }
1590
1591  // no, only dladdr is available
1592  if (dladdr((void *)addr, &dlinfo) != 0) {
1593    // see if we have a matching symbol
1594    if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1595      if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1596        jio_snprintf(buf, buflen, dlinfo.dli_sname);
1597      }
1598      if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1599      return true;
1600    }
1601    // no matching symbol so try for just file info
1602    if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1603      if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1604                          buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1605        return true;
1606      }
1607    }
1608  }
1609  buf[0] = '\0';
1610  if (offset != NULL) *offset  = -1;
1611  return false;
1612}
1613
1614bool os::dll_address_to_library_name(address addr, char* buf,
1615                                     int buflen, int* offset) {
1616  // buf is not optional, but offset is optional
1617  assert(buf != NULL, "sanity check");
1618
1619  Dl_info dlinfo;
1620
1621  if (dladdr((void*)addr, &dlinfo) != 0) {
1622    if (dlinfo.dli_fname != NULL) {
1623      jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1624    }
1625    if (dlinfo.dli_fbase != NULL && offset != NULL) {
1626      *offset = addr - (address)dlinfo.dli_fbase;
1627    }
1628    return true;
1629  }
1630
1631  buf[0] = '\0';
1632  if (offset) *offset = -1;
1633  return false;
1634}
1635
1636int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1637  Dl_info dli;
1638  // Sanity check?
1639  if (dladdr(CAST_FROM_FN_PTR(void *, os::get_loaded_modules_info), &dli) == 0 ||
1640      dli.dli_fname == NULL) {
1641    return 1;
1642  }
1643
1644  void * handle = dlopen(dli.dli_fname, RTLD_LAZY);
1645  if (handle == NULL) {
1646    return 1;
1647  }
1648
1649  Link_map *map;
1650  dlinfo(handle, RTLD_DI_LINKMAP, &map);
1651  if (map == NULL) {
1652    dlclose(handle);
1653    return 1;
1654  }
1655
1656  while (map->l_prev != NULL) {
1657    map = map->l_prev;
1658  }
1659
1660  while (map != NULL) {
1661    // Iterate through all map entries and call callback with fields of interest
1662    if(callback(map->l_name, (address)map->l_addr, (address)0, param)) {
1663      dlclose(handle);
1664      return 1;
1665    }
1666    map = map->l_next;
1667  }
1668
1669  dlclose(handle);
1670  return 0;
1671}
1672
1673int _print_dll_info_cb(const char * name, address base_address, address top_address, void * param) {
1674  outputStream * out = (outputStream *) param;
1675  out->print_cr(PTR_FORMAT " \t%s", base_address, name);
1676  return 0;
1677}
1678
1679void os::print_dll_info(outputStream * st) {
1680  st->print_cr("Dynamic libraries:"); st->flush();
1681  if (get_loaded_modules_info(_print_dll_info_cb, (void *)st)) {
1682    st->print_cr("Error: Cannot print dynamic libraries.");
1683  }
1684}
1685
1686// Loads .dll/.so and
1687// in case of error it checks if .dll/.so was built for the
1688// same architecture as Hotspot is running on
1689
1690void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1691  void * result= ::dlopen(filename, RTLD_LAZY);
1692  if (result != NULL) {
1693    // Successful loading
1694    return result;
1695  }
1696
1697  Elf32_Ehdr elf_head;
1698
1699  // Read system error message into ebuf
1700  // It may or may not be overwritten below
1701  ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1702  ebuf[ebuflen-1]='\0';
1703  int diag_msg_max_length=ebuflen-strlen(ebuf);
1704  char* diag_msg_buf=ebuf+strlen(ebuf);
1705
1706  if (diag_msg_max_length==0) {
1707    // No more space in ebuf for additional diagnostics message
1708    return NULL;
1709  }
1710
1711
1712  int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1713
1714  if (file_descriptor < 0) {
1715    // Can't open library, report dlerror() message
1716    return NULL;
1717  }
1718
1719  bool failed_to_read_elf_head=
1720    (sizeof(elf_head)!=
1721     (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1722
1723  ::close(file_descriptor);
1724  if (failed_to_read_elf_head) {
1725    // file i/o error - report dlerror() msg
1726    return NULL;
1727  }
1728
1729  typedef struct {
1730    Elf32_Half  code;         // Actual value as defined in elf.h
1731    Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1732    char        elf_class;    // 32 or 64 bit
1733    char        endianess;    // MSB or LSB
1734    char*       name;         // String representation
1735  } arch_t;
1736
1737  static const arch_t arch_array[]={
1738    {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1739    {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1740    {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1741    {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1742    {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1743    {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1744    {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1745    {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1746    {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1747    {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
1748  };
1749
1750#if  (defined IA32)
1751  static  Elf32_Half running_arch_code=EM_386;
1752#elif   (defined AMD64)
1753  static  Elf32_Half running_arch_code=EM_X86_64;
1754#elif  (defined IA64)
1755  static  Elf32_Half running_arch_code=EM_IA_64;
1756#elif  (defined __sparc) && (defined _LP64)
1757  static  Elf32_Half running_arch_code=EM_SPARCV9;
1758#elif  (defined __sparc) && (!defined _LP64)
1759  static  Elf32_Half running_arch_code=EM_SPARC;
1760#elif  (defined __powerpc64__)
1761  static  Elf32_Half running_arch_code=EM_PPC64;
1762#elif  (defined __powerpc__)
1763  static  Elf32_Half running_arch_code=EM_PPC;
1764#elif (defined ARM)
1765  static  Elf32_Half running_arch_code=EM_ARM;
1766#else
1767  #error Method os::dll_load requires that one of following is defined:\
1768       IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
1769#endif
1770
1771  // Identify compatability class for VM's architecture and library's architecture
1772  // Obtain string descriptions for architectures
1773
1774  arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1775  int running_arch_index=-1;
1776
1777  for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1778    if (running_arch_code == arch_array[i].code) {
1779      running_arch_index    = i;
1780    }
1781    if (lib_arch.code == arch_array[i].code) {
1782      lib_arch.compat_class = arch_array[i].compat_class;
1783      lib_arch.name         = arch_array[i].name;
1784    }
1785  }
1786
1787  assert(running_arch_index != -1,
1788         "Didn't find running architecture code (running_arch_code) in arch_array");
1789  if (running_arch_index == -1) {
1790    // Even though running architecture detection failed
1791    // we may still continue with reporting dlerror() message
1792    return NULL;
1793  }
1794
1795  if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1796    ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1797    return NULL;
1798  }
1799
1800  if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1801    ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1802    return NULL;
1803  }
1804
1805  if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1806    if (lib_arch.name!=NULL) {
1807      ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1808                 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1809                 lib_arch.name, arch_array[running_arch_index].name);
1810    } else {
1811      ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1812                 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1813                 lib_arch.code,
1814                 arch_array[running_arch_index].name);
1815    }
1816  }
1817
1818  return NULL;
1819}
1820
1821void* os::dll_lookup(void* handle, const char* name) {
1822  return dlsym(handle, name);
1823}
1824
1825void* os::get_default_process_handle() {
1826  return (void*)::dlopen(NULL, RTLD_LAZY);
1827}
1828
1829int os::stat(const char *path, struct stat *sbuf) {
1830  char pathbuf[MAX_PATH];
1831  if (strlen(path) > MAX_PATH - 1) {
1832    errno = ENAMETOOLONG;
1833    return -1;
1834  }
1835  os::native_path(strcpy(pathbuf, path));
1836  return ::stat(pathbuf, sbuf);
1837}
1838
1839static inline time_t get_mtime(const char* filename) {
1840  struct stat st;
1841  int ret = os::stat(filename, &st);
1842  assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
1843  return st.st_mtime;
1844}
1845
1846int os::compare_file_modified_times(const char* file1, const char* file2) {
1847  time_t t1 = get_mtime(file1);
1848  time_t t2 = get_mtime(file2);
1849  return t1 - t2;
1850}
1851
1852static bool _print_ascii_file(const char* filename, outputStream* st) {
1853  int fd = ::open(filename, O_RDONLY);
1854  if (fd == -1) {
1855    return false;
1856  }
1857
1858  char buf[32];
1859  int bytes;
1860  while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1861    st->print_raw(buf, bytes);
1862  }
1863
1864  ::close(fd);
1865
1866  return true;
1867}
1868
1869void os::print_os_info_brief(outputStream* st) {
1870  os::Solaris::print_distro_info(st);
1871
1872  os::Posix::print_uname_info(st);
1873
1874  os::Solaris::print_libversion_info(st);
1875}
1876
1877void os::print_os_info(outputStream* st) {
1878  st->print("OS:");
1879
1880  os::Solaris::print_distro_info(st);
1881
1882  os::Posix::print_uname_info(st);
1883
1884  os::Solaris::print_libversion_info(st);
1885
1886  os::Posix::print_rlimit_info(st);
1887
1888  os::Posix::print_load_average(st);
1889}
1890
1891void os::Solaris::print_distro_info(outputStream* st) {
1892  if (!_print_ascii_file("/etc/release", st)) {
1893    st->print("Solaris");
1894  }
1895  st->cr();
1896}
1897
1898void os::get_summary_os_info(char* buf, size_t buflen) {
1899  strncpy(buf, "Solaris", buflen);  // default to plain solaris
1900  FILE* fp = fopen("/etc/release", "r");
1901  if (fp != NULL) {
1902    char tmp[256];
1903    // Only get the first line and chop out everything but the os name.
1904    if (fgets(tmp, sizeof(tmp), fp)) {
1905      char* ptr = tmp;
1906      // skip past whitespace characters
1907      while (*ptr != '\0' && (*ptr == ' ' || *ptr == '\t' || *ptr == '\n')) ptr++;
1908      if (*ptr != '\0') {
1909        char* nl = strchr(ptr, '\n');
1910        if (nl != NULL) *nl = '\0';
1911        strncpy(buf, ptr, buflen);
1912      }
1913    }
1914    fclose(fp);
1915  }
1916}
1917
1918void os::Solaris::print_libversion_info(outputStream* st) {
1919  st->print("  (T2 libthread)");
1920  st->cr();
1921}
1922
1923static bool check_addr0(outputStream* st) {
1924  jboolean status = false;
1925  const int read_chunk = 200;
1926  int ret = 0;
1927  int nmap = 0;
1928  int fd = ::open("/proc/self/map",O_RDONLY);
1929  if (fd >= 0) {
1930    prmap_t *p = NULL;
1931    char *mbuff = (char *) calloc(read_chunk, sizeof(prmap_t));
1932    if (NULL == mbuff) {
1933      ::close(fd);
1934      return status;
1935    }
1936    while ((ret = ::read(fd, mbuff, read_chunk*sizeof(prmap_t))) > 0) {
1937      //check if read() has not read partial data
1938      if( 0 != ret % sizeof(prmap_t)){
1939        break;
1940      }
1941      nmap = ret / sizeof(prmap_t);
1942      p = (prmap_t *)mbuff;
1943      for(int i = 0; i < nmap; i++){
1944        if (p->pr_vaddr == 0x0) {
1945          st->print("Warning: Address: " PTR_FORMAT ", Size: " SIZE_FORMAT "K, ",p->pr_vaddr, p->pr_size/1024);
1946          st->print("Mapped file: %s, ", p->pr_mapname[0] == '\0' ? "None" : p->pr_mapname);
1947          st->print("Access: ");
1948          st->print("%s",(p->pr_mflags & MA_READ)  ? "r" : "-");
1949          st->print("%s",(p->pr_mflags & MA_WRITE) ? "w" : "-");
1950          st->print("%s",(p->pr_mflags & MA_EXEC)  ? "x" : "-");
1951          st->cr();
1952          status = true;
1953        }
1954        p++;
1955      }
1956    }
1957    free(mbuff);
1958    ::close(fd);
1959  }
1960  return status;
1961}
1962
1963void os::get_summary_cpu_info(char* buf, size_t buflen) {
1964  // Get MHz with system call. We don't seem to already have this.
1965  processor_info_t stats;
1966  processorid_t id = getcpuid();
1967  int clock = 0;
1968  if (processor_info(id, &stats) != -1) {
1969    clock = stats.pi_clock;  // pi_processor_type isn't more informative than below
1970  }
1971#ifdef AMD64
1972  snprintf(buf, buflen, "x86 64 bit %d MHz", clock);
1973#else
1974  // must be sparc
1975  snprintf(buf, buflen, "Sparcv9 64 bit %d MHz", clock);
1976#endif
1977}
1978
1979void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
1980  // Nothing to do for now.
1981}
1982
1983void os::print_memory_info(outputStream* st) {
1984  st->print("Memory:");
1985  st->print(" %dk page", os::vm_page_size()>>10);
1986  st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
1987  st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
1988  st->cr();
1989  (void) check_addr0(st);
1990}
1991
1992// Moved from whole group, because we need them here for diagnostic
1993// prints.
1994#define OLDMAXSIGNUM 32
1995static int Maxsignum = 0;
1996static int *ourSigFlags = NULL;
1997
1998int os::Solaris::get_our_sigflags(int sig) {
1999  assert(ourSigFlags!=NULL, "signal data structure not initialized");
2000  assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2001  return ourSigFlags[sig];
2002}
2003
2004void os::Solaris::set_our_sigflags(int sig, int flags) {
2005  assert(ourSigFlags!=NULL, "signal data structure not initialized");
2006  assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2007  ourSigFlags[sig] = flags;
2008}
2009
2010
2011static const char* get_signal_handler_name(address handler,
2012                                           char* buf, int buflen) {
2013  int offset;
2014  bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
2015  if (found) {
2016    // skip directory names
2017    const char *p1, *p2;
2018    p1 = buf;
2019    size_t len = strlen(os::file_separator());
2020    while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
2021    jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
2022  } else {
2023    jio_snprintf(buf, buflen, PTR_FORMAT, handler);
2024  }
2025  return buf;
2026}
2027
2028static void print_signal_handler(outputStream* st, int sig,
2029                                 char* buf, size_t buflen) {
2030  struct sigaction sa;
2031
2032  sigaction(sig, NULL, &sa);
2033
2034  st->print("%s: ", os::exception_name(sig, buf, buflen));
2035
2036  address handler = (sa.sa_flags & SA_SIGINFO)
2037                  ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
2038                  : CAST_FROM_FN_PTR(address, sa.sa_handler);
2039
2040  if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
2041    st->print("SIG_DFL");
2042  } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
2043    st->print("SIG_IGN");
2044  } else {
2045    st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
2046  }
2047
2048  st->print(", sa_mask[0]=");
2049  os::Posix::print_signal_set_short(st, &sa.sa_mask);
2050
2051  address rh = VMError::get_resetted_sighandler(sig);
2052  // May be, handler was resetted by VMError?
2053  if (rh != NULL) {
2054    handler = rh;
2055    sa.sa_flags = VMError::get_resetted_sigflags(sig);
2056  }
2057
2058  st->print(", sa_flags=");
2059  os::Posix::print_sa_flags(st, sa.sa_flags);
2060
2061  // Check: is it our handler?
2062  if (handler == CAST_FROM_FN_PTR(address, signalHandler)) {
2063    // It is our signal handler
2064    // check for flags
2065    if (sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2066      st->print(
2067                ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2068                os::Solaris::get_our_sigflags(sig));
2069    }
2070  }
2071  st->cr();
2072}
2073
2074void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2075  st->print_cr("Signal Handlers:");
2076  print_signal_handler(st, SIGSEGV, buf, buflen);
2077  print_signal_handler(st, SIGBUS , buf, buflen);
2078  print_signal_handler(st, SIGFPE , buf, buflen);
2079  print_signal_handler(st, SIGPIPE, buf, buflen);
2080  print_signal_handler(st, SIGXFSZ, buf, buflen);
2081  print_signal_handler(st, SIGILL , buf, buflen);
2082  print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2083  print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2084  print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2085  print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2086  print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2087  print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2088}
2089
2090static char saved_jvm_path[MAXPATHLEN] = { 0 };
2091
2092// Find the full path to the current module, libjvm.so
2093void os::jvm_path(char *buf, jint buflen) {
2094  // Error checking.
2095  if (buflen < MAXPATHLEN) {
2096    assert(false, "must use a large-enough buffer");
2097    buf[0] = '\0';
2098    return;
2099  }
2100  // Lazy resolve the path to current module.
2101  if (saved_jvm_path[0] != 0) {
2102    strcpy(buf, saved_jvm_path);
2103    return;
2104  }
2105
2106  Dl_info dlinfo;
2107  int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2108  assert(ret != 0, "cannot locate libjvm");
2109  if (ret != 0 && dlinfo.dli_fname != NULL) {
2110    realpath((char *)dlinfo.dli_fname, buf);
2111  } else {
2112    buf[0] = '\0';
2113    return;
2114  }
2115
2116  if (Arguments::sun_java_launcher_is_altjvm()) {
2117    // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2118    // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2119    // If "/jre/lib/" appears at the right place in the string, then
2120    // assume we are installed in a JDK and we're done.  Otherwise, check
2121    // for a JAVA_HOME environment variable and fix up the path so it
2122    // looks like libjvm.so is installed there (append a fake suffix
2123    // hotspot/libjvm.so).
2124    const char *p = buf + strlen(buf) - 1;
2125    for (int count = 0; p > buf && count < 5; ++count) {
2126      for (--p; p > buf && *p != '/'; --p)
2127        /* empty */ ;
2128    }
2129
2130    if (strncmp(p, "/jre/lib/", 9) != 0) {
2131      // Look for JAVA_HOME in the environment.
2132      char* java_home_var = ::getenv("JAVA_HOME");
2133      if (java_home_var != NULL && java_home_var[0] != 0) {
2134        char cpu_arch[12];
2135        char* jrelib_p;
2136        int   len;
2137        sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2138#ifdef _LP64
2139        // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2140        if (strcmp(cpu_arch, "sparc") == 0) {
2141          strcat(cpu_arch, "v9");
2142        } else if (strcmp(cpu_arch, "i386") == 0) {
2143          strcpy(cpu_arch, "amd64");
2144        }
2145#endif
2146        // Check the current module name "libjvm.so".
2147        p = strrchr(buf, '/');
2148        assert(strstr(p, "/libjvm") == p, "invalid library name");
2149
2150        realpath(java_home_var, buf);
2151        // determine if this is a legacy image or modules image
2152        // modules image doesn't have "jre" subdirectory
2153        len = strlen(buf);
2154        assert(len < buflen, "Ran out of buffer space");
2155        jrelib_p = buf + len;
2156        snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2157        if (0 != access(buf, F_OK)) {
2158          snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2159        }
2160
2161        if (0 == access(buf, F_OK)) {
2162          // Use current module name "libjvm.so"
2163          len = strlen(buf);
2164          snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2165        } else {
2166          // Go back to path of .so
2167          realpath((char *)dlinfo.dli_fname, buf);
2168        }
2169      }
2170    }
2171  }
2172
2173  strncpy(saved_jvm_path, buf, MAXPATHLEN);
2174  saved_jvm_path[MAXPATHLEN - 1] = '\0';
2175}
2176
2177
2178void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2179  // no prefix required, not even "_"
2180}
2181
2182
2183void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2184  // no suffix required
2185}
2186
2187// This method is a copy of JDK's sysGetLastErrorString
2188// from src/solaris/hpi/src/system_md.c
2189
2190size_t os::lasterror(char *buf, size_t len) {
2191  if (errno == 0)  return 0;
2192
2193  const char *s = os::strerror(errno);
2194  size_t n = ::strlen(s);
2195  if (n >= len) {
2196    n = len - 1;
2197  }
2198  ::strncpy(buf, s, n);
2199  buf[n] = '\0';
2200  return n;
2201}
2202
2203
2204// sun.misc.Signal
2205
2206extern "C" {
2207  static void UserHandler(int sig, void *siginfo, void *context) {
2208    // Ctrl-C is pressed during error reporting, likely because the error
2209    // handler fails to abort. Let VM die immediately.
2210    if (sig == SIGINT && is_error_reported()) {
2211      os::die();
2212    }
2213
2214    os::signal_notify(sig);
2215    // We do not need to reinstate the signal handler each time...
2216  }
2217}
2218
2219void* os::user_handler() {
2220  return CAST_FROM_FN_PTR(void*, UserHandler);
2221}
2222
2223struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2224  struct timespec ts;
2225  unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2226
2227  return ts;
2228}
2229
2230extern "C" {
2231  typedef void (*sa_handler_t)(int);
2232  typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2233}
2234
2235void* os::signal(int signal_number, void* handler) {
2236  struct sigaction sigAct, oldSigAct;
2237  sigfillset(&(sigAct.sa_mask));
2238  sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2239  sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2240
2241  if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2242    // -1 means registration failed
2243    return (void *)-1;
2244  }
2245
2246  return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2247}
2248
2249void os::signal_raise(int signal_number) {
2250  raise(signal_number);
2251}
2252
2253// The following code is moved from os.cpp for making this
2254// code platform specific, which it is by its very nature.
2255
2256// a counter for each possible signal value
2257static int Sigexit = 0;
2258static int Maxlibjsigsigs;
2259static jint *pending_signals = NULL;
2260static int *preinstalled_sigs = NULL;
2261static struct sigaction *chainedsigactions = NULL;
2262static sema_t sig_sem;
2263typedef int (*version_getting_t)();
2264version_getting_t os::Solaris::get_libjsig_version = NULL;
2265static int libjsigversion = NULL;
2266
2267int os::sigexitnum_pd() {
2268  assert(Sigexit > 0, "signal memory not yet initialized");
2269  return Sigexit;
2270}
2271
2272void os::Solaris::init_signal_mem() {
2273  // Initialize signal structures
2274  Maxsignum = SIGRTMAX;
2275  Sigexit = Maxsignum+1;
2276  assert(Maxsignum >0, "Unable to obtain max signal number");
2277
2278  Maxlibjsigsigs = Maxsignum;
2279
2280  // pending_signals has one int per signal
2281  // The additional signal is for SIGEXIT - exit signal to signal_thread
2282  pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2283  memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2284
2285  if (UseSignalChaining) {
2286    chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2287                                                   * (Maxsignum + 1), mtInternal);
2288    memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2289    preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2290    memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2291  }
2292  ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2293  memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2294}
2295
2296void os::signal_init_pd() {
2297  int ret;
2298
2299  ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2300  assert(ret == 0, "sema_init() failed");
2301}
2302
2303void os::signal_notify(int signal_number) {
2304  int ret;
2305
2306  Atomic::inc(&pending_signals[signal_number]);
2307  ret = ::sema_post(&sig_sem);
2308  assert(ret == 0, "sema_post() failed");
2309}
2310
2311static int check_pending_signals(bool wait_for_signal) {
2312  int ret;
2313  while (true) {
2314    for (int i = 0; i < Sigexit + 1; i++) {
2315      jint n = pending_signals[i];
2316      if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2317        return i;
2318      }
2319    }
2320    if (!wait_for_signal) {
2321      return -1;
2322    }
2323    JavaThread *thread = JavaThread::current();
2324    ThreadBlockInVM tbivm(thread);
2325
2326    bool threadIsSuspended;
2327    do {
2328      thread->set_suspend_equivalent();
2329      // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2330      while ((ret = ::sema_wait(&sig_sem)) == EINTR)
2331        ;
2332      assert(ret == 0, "sema_wait() failed");
2333
2334      // were we externally suspended while we were waiting?
2335      threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2336      if (threadIsSuspended) {
2337        // The semaphore has been incremented, but while we were waiting
2338        // another thread suspended us. We don't want to continue running
2339        // while suspended because that would surprise the thread that
2340        // suspended us.
2341        ret = ::sema_post(&sig_sem);
2342        assert(ret == 0, "sema_post() failed");
2343
2344        thread->java_suspend_self();
2345      }
2346    } while (threadIsSuspended);
2347  }
2348}
2349
2350int os::signal_lookup() {
2351  return check_pending_signals(false);
2352}
2353
2354int os::signal_wait() {
2355  return check_pending_signals(true);
2356}
2357
2358////////////////////////////////////////////////////////////////////////////////
2359// Virtual Memory
2360
2361static int page_size = -1;
2362
2363// The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2364// clear this var if support is not available.
2365static bool has_map_align = true;
2366
2367int os::vm_page_size() {
2368  assert(page_size != -1, "must call os::init");
2369  return page_size;
2370}
2371
2372// Solaris allocates memory by pages.
2373int os::vm_allocation_granularity() {
2374  assert(page_size != -1, "must call os::init");
2375  return page_size;
2376}
2377
2378static bool recoverable_mmap_error(int err) {
2379  // See if the error is one we can let the caller handle. This
2380  // list of errno values comes from the Solaris mmap(2) man page.
2381  switch (err) {
2382  case EBADF:
2383  case EINVAL:
2384  case ENOTSUP:
2385    // let the caller deal with these errors
2386    return true;
2387
2388  default:
2389    // Any remaining errors on this OS can cause our reserved mapping
2390    // to be lost. That can cause confusion where different data
2391    // structures think they have the same memory mapped. The worst
2392    // scenario is if both the VM and a library think they have the
2393    // same memory mapped.
2394    return false;
2395  }
2396}
2397
2398static void warn_fail_commit_memory(char* addr, size_t bytes, bool exec,
2399                                    int err) {
2400  warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2401          ", %d) failed; error='%s' (errno=%d)", addr, bytes, exec,
2402          os::strerror(err), err);
2403}
2404
2405static void warn_fail_commit_memory(char* addr, size_t bytes,
2406                                    size_t alignment_hint, bool exec,
2407                                    int err) {
2408  warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2409          ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, bytes,
2410          alignment_hint, exec, os::strerror(err), err);
2411}
2412
2413int os::Solaris::commit_memory_impl(char* addr, size_t bytes, bool exec) {
2414  int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2415  size_t size = bytes;
2416  char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2417  if (res != NULL) {
2418    if (UseNUMAInterleaving) {
2419      numa_make_global(addr, bytes);
2420    }
2421    return 0;
2422  }
2423
2424  int err = errno;  // save errno from mmap() call in mmap_chunk()
2425
2426  if (!recoverable_mmap_error(err)) {
2427    warn_fail_commit_memory(addr, bytes, exec, err);
2428    vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "committing reserved memory.");
2429  }
2430
2431  return err;
2432}
2433
2434bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2435  return Solaris::commit_memory_impl(addr, bytes, exec) == 0;
2436}
2437
2438void os::pd_commit_memory_or_exit(char* addr, size_t bytes, bool exec,
2439                                  const char* mesg) {
2440  assert(mesg != NULL, "mesg must be specified");
2441  int err = os::Solaris::commit_memory_impl(addr, bytes, exec);
2442  if (err != 0) {
2443    // the caller wants all commit errors to exit with the specified mesg:
2444    warn_fail_commit_memory(addr, bytes, exec, err);
2445    vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2446  }
2447}
2448
2449size_t os::Solaris::page_size_for_alignment(size_t alignment) {
2450  assert(is_size_aligned(alignment, (size_t) vm_page_size()),
2451         SIZE_FORMAT " is not aligned to " SIZE_FORMAT,
2452         alignment, (size_t) vm_page_size());
2453
2454  for (int i = 0; _page_sizes[i] != 0; i++) {
2455    if (is_size_aligned(alignment, _page_sizes[i])) {
2456      return _page_sizes[i];
2457    }
2458  }
2459
2460  return (size_t) vm_page_size();
2461}
2462
2463int os::Solaris::commit_memory_impl(char* addr, size_t bytes,
2464                                    size_t alignment_hint, bool exec) {
2465  int err = Solaris::commit_memory_impl(addr, bytes, exec);
2466  if (err == 0 && UseLargePages && alignment_hint > 0) {
2467    assert(is_size_aligned(bytes, alignment_hint),
2468           SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, alignment_hint);
2469
2470    // The syscall memcntl requires an exact page size (see man memcntl for details).
2471    size_t page_size = page_size_for_alignment(alignment_hint);
2472    if (page_size > (size_t) vm_page_size()) {
2473      (void)Solaris::setup_large_pages(addr, bytes, page_size);
2474    }
2475  }
2476  return err;
2477}
2478
2479bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2480                          bool exec) {
2481  return Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec) == 0;
2482}
2483
2484void os::pd_commit_memory_or_exit(char* addr, size_t bytes,
2485                                  size_t alignment_hint, bool exec,
2486                                  const char* mesg) {
2487  assert(mesg != NULL, "mesg must be specified");
2488  int err = os::Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec);
2489  if (err != 0) {
2490    // the caller wants all commit errors to exit with the specified mesg:
2491    warn_fail_commit_memory(addr, bytes, alignment_hint, exec, err);
2492    vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2493  }
2494}
2495
2496// Uncommit the pages in a specified region.
2497void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2498  if (madvise(addr, bytes, MADV_FREE) < 0) {
2499    debug_only(warning("MADV_FREE failed."));
2500    return;
2501  }
2502}
2503
2504bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2505  return os::commit_memory(addr, size, !ExecMem);
2506}
2507
2508bool os::remove_stack_guard_pages(char* addr, size_t size) {
2509  return os::uncommit_memory(addr, size);
2510}
2511
2512// Change the page size in a given range.
2513void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2514  assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2515  assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2516  if (UseLargePages) {
2517    size_t page_size = Solaris::page_size_for_alignment(alignment_hint);
2518    if (page_size > (size_t) vm_page_size()) {
2519      Solaris::setup_large_pages(addr, bytes, page_size);
2520    }
2521  }
2522}
2523
2524// Tell the OS to make the range local to the first-touching LWP
2525void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2526  assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2527  if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2528    debug_only(warning("MADV_ACCESS_LWP failed."));
2529  }
2530}
2531
2532// Tell the OS that this range would be accessed from different LWPs.
2533void os::numa_make_global(char *addr, size_t bytes) {
2534  assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2535  if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2536    debug_only(warning("MADV_ACCESS_MANY failed."));
2537  }
2538}
2539
2540// Get the number of the locality groups.
2541size_t os::numa_get_groups_num() {
2542  size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2543  return n != -1 ? n : 1;
2544}
2545
2546// Get a list of leaf locality groups. A leaf lgroup is group that
2547// doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2548// board. An LWP is assigned to one of these groups upon creation.
2549size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2550  if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2551    ids[0] = 0;
2552    return 1;
2553  }
2554  int result_size = 0, top = 1, bottom = 0, cur = 0;
2555  for (int k = 0; k < size; k++) {
2556    int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2557                                   (Solaris::lgrp_id_t*)&ids[top], size - top);
2558    if (r == -1) {
2559      ids[0] = 0;
2560      return 1;
2561    }
2562    if (!r) {
2563      // That's a leaf node.
2564      assert(bottom <= cur, "Sanity check");
2565      // Check if the node has memory
2566      if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2567                                  NULL, 0, LGRP_RSRC_MEM) > 0) {
2568        ids[bottom++] = ids[cur];
2569      }
2570    }
2571    top += r;
2572    cur++;
2573  }
2574  if (bottom == 0) {
2575    // Handle a situation, when the OS reports no memory available.
2576    // Assume UMA architecture.
2577    ids[0] = 0;
2578    return 1;
2579  }
2580  return bottom;
2581}
2582
2583// Detect the topology change. Typically happens during CPU plugging-unplugging.
2584bool os::numa_topology_changed() {
2585  int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2586  if (is_stale != -1 && is_stale) {
2587    Solaris::lgrp_fini(Solaris::lgrp_cookie());
2588    Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2589    assert(c != 0, "Failure to initialize LGRP API");
2590    Solaris::set_lgrp_cookie(c);
2591    return true;
2592  }
2593  return false;
2594}
2595
2596// Get the group id of the current LWP.
2597int os::numa_get_group_id() {
2598  int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2599  if (lgrp_id == -1) {
2600    return 0;
2601  }
2602  const int size = os::numa_get_groups_num();
2603  int *ids = (int*)alloca(size * sizeof(int));
2604
2605  // Get the ids of all lgroups with memory; r is the count.
2606  int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2607                                  (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2608  if (r <= 0) {
2609    return 0;
2610  }
2611  return ids[os::random() % r];
2612}
2613
2614// Request information about the page.
2615bool os::get_page_info(char *start, page_info* info) {
2616  const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2617  uint64_t addr = (uintptr_t)start;
2618  uint64_t outdata[2];
2619  uint_t validity = 0;
2620
2621  if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2622    return false;
2623  }
2624
2625  info->size = 0;
2626  info->lgrp_id = -1;
2627
2628  if ((validity & 1) != 0) {
2629    if ((validity & 2) != 0) {
2630      info->lgrp_id = outdata[0];
2631    }
2632    if ((validity & 4) != 0) {
2633      info->size = outdata[1];
2634    }
2635    return true;
2636  }
2637  return false;
2638}
2639
2640// Scan the pages from start to end until a page different than
2641// the one described in the info parameter is encountered.
2642char *os::scan_pages(char *start, char* end, page_info* page_expected,
2643                     page_info* page_found) {
2644  const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2645  const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2646  uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT + 1];
2647  uint_t validity[MAX_MEMINFO_CNT];
2648
2649  size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2650  uint64_t p = (uint64_t)start;
2651  while (p < (uint64_t)end) {
2652    addrs[0] = p;
2653    size_t addrs_count = 1;
2654    while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] + page_size < (uint64_t)end) {
2655      addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2656      addrs_count++;
2657    }
2658
2659    if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2660      return NULL;
2661    }
2662
2663    size_t i = 0;
2664    for (; i < addrs_count; i++) {
2665      if ((validity[i] & 1) != 0) {
2666        if ((validity[i] & 4) != 0) {
2667          if (outdata[types * i + 1] != page_expected->size) {
2668            break;
2669          }
2670        } else if (page_expected->size != 0) {
2671          break;
2672        }
2673
2674        if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2675          if (outdata[types * i] != page_expected->lgrp_id) {
2676            break;
2677          }
2678        }
2679      } else {
2680        return NULL;
2681      }
2682    }
2683
2684    if (i < addrs_count) {
2685      if ((validity[i] & 2) != 0) {
2686        page_found->lgrp_id = outdata[types * i];
2687      } else {
2688        page_found->lgrp_id = -1;
2689      }
2690      if ((validity[i] & 4) != 0) {
2691        page_found->size = outdata[types * i + 1];
2692      } else {
2693        page_found->size = 0;
2694      }
2695      return (char*)addrs[i];
2696    }
2697
2698    p = addrs[addrs_count - 1] + page_size;
2699  }
2700  return end;
2701}
2702
2703bool os::pd_uncommit_memory(char* addr, size_t bytes) {
2704  size_t size = bytes;
2705  // Map uncommitted pages PROT_NONE so we fail early if we touch an
2706  // uncommitted page. Otherwise, the read/write might succeed if we
2707  // have enough swap space to back the physical page.
2708  return
2709    NULL != Solaris::mmap_chunk(addr, size,
2710                                MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2711                                PROT_NONE);
2712}
2713
2714char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2715  char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2716
2717  if (b == MAP_FAILED) {
2718    return NULL;
2719  }
2720  return b;
2721}
2722
2723char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes,
2724                             size_t alignment_hint, bool fixed) {
2725  char* addr = requested_addr;
2726  int flags = MAP_PRIVATE | MAP_NORESERVE;
2727
2728  assert(!(fixed && (alignment_hint > 0)),
2729         "alignment hint meaningless with fixed mmap");
2730
2731  if (fixed) {
2732    flags |= MAP_FIXED;
2733  } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
2734    flags |= MAP_ALIGN;
2735    addr = (char*) alignment_hint;
2736  }
2737
2738  // Map uncommitted pages PROT_NONE so we fail early if we touch an
2739  // uncommitted page. Otherwise, the read/write might succeed if we
2740  // have enough swap space to back the physical page.
2741  return mmap_chunk(addr, bytes, flags, PROT_NONE);
2742}
2743
2744char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2745                            size_t alignment_hint) {
2746  char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint,
2747                                  (requested_addr != NULL));
2748
2749  guarantee(requested_addr == NULL || requested_addr == addr,
2750            "OS failed to return requested mmap address.");
2751  return addr;
2752}
2753
2754// Reserve memory at an arbitrary address, only if that area is
2755// available (and not reserved for something else).
2756
2757char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2758  const int max_tries = 10;
2759  char* base[max_tries];
2760  size_t size[max_tries];
2761
2762  // Solaris adds a gap between mmap'ed regions.  The size of the gap
2763  // is dependent on the requested size and the MMU.  Our initial gap
2764  // value here is just a guess and will be corrected later.
2765  bool had_top_overlap = false;
2766  bool have_adjusted_gap = false;
2767  size_t gap = 0x400000;
2768
2769  // Assert only that the size is a multiple of the page size, since
2770  // that's all that mmap requires, and since that's all we really know
2771  // about at this low abstraction level.  If we need higher alignment,
2772  // we can either pass an alignment to this method or verify alignment
2773  // in one of the methods further up the call chain.  See bug 5044738.
2774  assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2775
2776  // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2777  // Give it a try, if the kernel honors the hint we can return immediately.
2778  char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2779
2780  volatile int err = errno;
2781  if (addr == requested_addr) {
2782    return addr;
2783  } else if (addr != NULL) {
2784    pd_unmap_memory(addr, bytes);
2785  }
2786
2787  if (log_is_enabled(Warning, os)) {
2788    char buf[256];
2789    buf[0] = '\0';
2790    if (addr == NULL) {
2791      jio_snprintf(buf, sizeof(buf), ": %s", os::strerror(err));
2792    }
2793    log_info(os)("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
2794            PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2795            "%s", bytes, requested_addr, addr, buf);
2796  }
2797
2798  // Address hint method didn't work.  Fall back to the old method.
2799  // In theory, once SNV becomes our oldest supported platform, this
2800  // code will no longer be needed.
2801  //
2802  // Repeatedly allocate blocks until the block is allocated at the
2803  // right spot. Give up after max_tries.
2804  int i;
2805  for (i = 0; i < max_tries; ++i) {
2806    base[i] = reserve_memory(bytes);
2807
2808    if (base[i] != NULL) {
2809      // Is this the block we wanted?
2810      if (base[i] == requested_addr) {
2811        size[i] = bytes;
2812        break;
2813      }
2814
2815      // check that the gap value is right
2816      if (had_top_overlap && !have_adjusted_gap) {
2817        size_t actual_gap = base[i-1] - base[i] - bytes;
2818        if (gap != actual_gap) {
2819          // adjust the gap value and retry the last 2 allocations
2820          assert(i > 0, "gap adjustment code problem");
2821          have_adjusted_gap = true;  // adjust the gap only once, just in case
2822          gap = actual_gap;
2823          log_info(os)("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
2824          unmap_memory(base[i], bytes);
2825          unmap_memory(base[i-1], size[i-1]);
2826          i-=2;
2827          continue;
2828        }
2829      }
2830
2831      // Does this overlap the block we wanted? Give back the overlapped
2832      // parts and try again.
2833      //
2834      // There is still a bug in this code: if top_overlap == bytes,
2835      // the overlap is offset from requested region by the value of gap.
2836      // In this case giving back the overlapped part will not work,
2837      // because we'll give back the entire block at base[i] and
2838      // therefore the subsequent allocation will not generate a new gap.
2839      // This could be fixed with a new algorithm that used larger
2840      // or variable size chunks to find the requested region -
2841      // but such a change would introduce additional complications.
2842      // It's rare enough that the planets align for this bug,
2843      // so we'll just wait for a fix for 6204603/5003415 which
2844      // will provide a mmap flag to allow us to avoid this business.
2845
2846      size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2847      if (top_overlap >= 0 && top_overlap < bytes) {
2848        had_top_overlap = true;
2849        unmap_memory(base[i], top_overlap);
2850        base[i] += top_overlap;
2851        size[i] = bytes - top_overlap;
2852      } else {
2853        size_t bottom_overlap = base[i] + bytes - requested_addr;
2854        if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2855          if (bottom_overlap == 0) {
2856            log_info(os)("attempt_reserve_memory_at: possible alignment bug");
2857          }
2858          unmap_memory(requested_addr, bottom_overlap);
2859          size[i] = bytes - bottom_overlap;
2860        } else {
2861          size[i] = bytes;
2862        }
2863      }
2864    }
2865  }
2866
2867  // Give back the unused reserved pieces.
2868
2869  for (int j = 0; j < i; ++j) {
2870    if (base[j] != NULL) {
2871      unmap_memory(base[j], size[j]);
2872    }
2873  }
2874
2875  return (i < max_tries) ? requested_addr : NULL;
2876}
2877
2878bool os::pd_release_memory(char* addr, size_t bytes) {
2879  size_t size = bytes;
2880  return munmap(addr, size) == 0;
2881}
2882
2883static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
2884  assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
2885         "addr must be page aligned");
2886  int retVal = mprotect(addr, bytes, prot);
2887  return retVal == 0;
2888}
2889
2890// Protect memory (Used to pass readonly pages through
2891// JNI GetArray<type>Elements with empty arrays.)
2892// Also, used for serialization page and for compressed oops null pointer
2893// checking.
2894bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2895                        bool is_committed) {
2896  unsigned int p = 0;
2897  switch (prot) {
2898  case MEM_PROT_NONE: p = PROT_NONE; break;
2899  case MEM_PROT_READ: p = PROT_READ; break;
2900  case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2901  case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2902  default:
2903    ShouldNotReachHere();
2904  }
2905  // is_committed is unused.
2906  return solaris_mprotect(addr, bytes, p);
2907}
2908
2909// guard_memory and unguard_memory only happens within stack guard pages.
2910// Since ISM pertains only to the heap, guard and unguard memory should not
2911/// happen with an ISM region.
2912bool os::guard_memory(char* addr, size_t bytes) {
2913  return solaris_mprotect(addr, bytes, PROT_NONE);
2914}
2915
2916bool os::unguard_memory(char* addr, size_t bytes) {
2917  return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
2918}
2919
2920// Large page support
2921static size_t _large_page_size = 0;
2922
2923// Insertion sort for small arrays (descending order).
2924static void insertion_sort_descending(size_t* array, int len) {
2925  for (int i = 0; i < len; i++) {
2926    size_t val = array[i];
2927    for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
2928      size_t tmp = array[key];
2929      array[key] = array[key - 1];
2930      array[key - 1] = tmp;
2931    }
2932  }
2933}
2934
2935bool os::Solaris::mpss_sanity_check(bool warn, size_t* page_size) {
2936  const unsigned int usable_count = VM_Version::page_size_count();
2937  if (usable_count == 1) {
2938    return false;
2939  }
2940
2941  // Find the right getpagesizes interface.  When solaris 11 is the minimum
2942  // build platform, getpagesizes() (without the '2') can be called directly.
2943  typedef int (*gps_t)(size_t[], int);
2944  gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
2945  if (gps_func == NULL) {
2946    gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
2947    if (gps_func == NULL) {
2948      if (warn) {
2949        warning("MPSS is not supported by the operating system.");
2950      }
2951      return false;
2952    }
2953  }
2954
2955  // Fill the array of page sizes.
2956  int n = (*gps_func)(_page_sizes, page_sizes_max);
2957  assert(n > 0, "Solaris bug?");
2958
2959  if (n == page_sizes_max) {
2960    // Add a sentinel value (necessary only if the array was completely filled
2961    // since it is static (zeroed at initialization)).
2962    _page_sizes[--n] = 0;
2963    DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
2964  }
2965  assert(_page_sizes[n] == 0, "missing sentinel");
2966  trace_page_sizes("available page sizes", _page_sizes, n);
2967
2968  if (n == 1) return false;     // Only one page size available.
2969
2970  // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
2971  // select up to usable_count elements.  First sort the array, find the first
2972  // acceptable value, then copy the usable sizes to the top of the array and
2973  // trim the rest.  Make sure to include the default page size :-).
2974  //
2975  // A better policy could get rid of the 4M limit by taking the sizes of the
2976  // important VM memory regions (java heap and possibly the code cache) into
2977  // account.
2978  insertion_sort_descending(_page_sizes, n);
2979  const size_t size_limit =
2980    FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
2981  int beg;
2982  for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */;
2983  const int end = MIN2((int)usable_count, n) - 1;
2984  for (int cur = 0; cur < end; ++cur, ++beg) {
2985    _page_sizes[cur] = _page_sizes[beg];
2986  }
2987  _page_sizes[end] = vm_page_size();
2988  _page_sizes[end + 1] = 0;
2989
2990  if (_page_sizes[end] > _page_sizes[end - 1]) {
2991    // Default page size is not the smallest; sort again.
2992    insertion_sort_descending(_page_sizes, end + 1);
2993  }
2994  *page_size = _page_sizes[0];
2995
2996  trace_page_sizes("usable page sizes", _page_sizes, end + 1);
2997  return true;
2998}
2999
3000void os::large_page_init() {
3001  if (UseLargePages) {
3002    // print a warning if any large page related flag is specified on command line
3003    bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
3004                           !FLAG_IS_DEFAULT(LargePageSizeInBytes);
3005
3006    UseLargePages = Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
3007  }
3008}
3009
3010bool os::Solaris::is_valid_page_size(size_t bytes) {
3011  for (int i = 0; _page_sizes[i] != 0; i++) {
3012    if (_page_sizes[i] == bytes) {
3013      return true;
3014    }
3015  }
3016  return false;
3017}
3018
3019bool os::Solaris::setup_large_pages(caddr_t start, size_t bytes, size_t align) {
3020  assert(is_valid_page_size(align), SIZE_FORMAT " is not a valid page size", align);
3021  assert(is_ptr_aligned((void*) start, align),
3022         PTR_FORMAT " is not aligned to " SIZE_FORMAT, p2i((void*) start), align);
3023  assert(is_size_aligned(bytes, align),
3024         SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, align);
3025
3026  // Signal to OS that we want large pages for addresses
3027  // from addr, addr + bytes
3028  struct memcntl_mha mpss_struct;
3029  mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
3030  mpss_struct.mha_pagesize = align;
3031  mpss_struct.mha_flags = 0;
3032  // Upon successful completion, memcntl() returns 0
3033  if (memcntl(start, bytes, MC_HAT_ADVISE, (caddr_t) &mpss_struct, 0, 0)) {
3034    debug_only(warning("Attempt to use MPSS failed."));
3035    return false;
3036  }
3037  return true;
3038}
3039
3040char* os::reserve_memory_special(size_t size, size_t alignment, char* addr, bool exec) {
3041  fatal("os::reserve_memory_special should not be called on Solaris.");
3042  return NULL;
3043}
3044
3045bool os::release_memory_special(char* base, size_t bytes) {
3046  fatal("os::release_memory_special should not be called on Solaris.");
3047  return false;
3048}
3049
3050size_t os::large_page_size() {
3051  return _large_page_size;
3052}
3053
3054// MPSS allows application to commit large page memory on demand; with ISM
3055// the entire memory region must be allocated as shared memory.
3056bool os::can_commit_large_page_memory() {
3057  return true;
3058}
3059
3060bool os::can_execute_large_page_memory() {
3061  return true;
3062}
3063
3064// Read calls from inside the vm need to perform state transitions
3065size_t os::read(int fd, void *buf, unsigned int nBytes) {
3066  size_t res;
3067  JavaThread* thread = (JavaThread*)Thread::current();
3068  assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
3069  ThreadBlockInVM tbiv(thread);
3070  RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
3071  return res;
3072}
3073
3074size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3075  size_t res;
3076  JavaThread* thread = (JavaThread*)Thread::current();
3077  assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
3078  ThreadBlockInVM tbiv(thread);
3079  RESTARTABLE(::pread(fd, buf, (size_t) nBytes, offset), res);
3080  return res;
3081}
3082
3083size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
3084  size_t res;
3085  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
3086         "Assumed _thread_in_native");
3087  RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
3088  return res;
3089}
3090
3091void os::naked_short_sleep(jlong ms) {
3092  assert(ms < 1000, "Un-interruptable sleep, short time use only");
3093
3094  // usleep is deprecated and removed from POSIX, in favour of nanosleep, but
3095  // Solaris requires -lrt for this.
3096  usleep((ms * 1000));
3097
3098  return;
3099}
3100
3101// Sleep forever; naked call to OS-specific sleep; use with CAUTION
3102void os::infinite_sleep() {
3103  while (true) {    // sleep forever ...
3104    ::sleep(100);   // ... 100 seconds at a time
3105  }
3106}
3107
3108// Used to convert frequent JVM_Yield() to nops
3109bool os::dont_yield() {
3110  if (DontYieldALot) {
3111    static hrtime_t last_time = 0;
3112    hrtime_t diff = getTimeNanos() - last_time;
3113
3114    if (diff < DontYieldALotInterval * 1000000) {
3115      return true;
3116    }
3117
3118    last_time += diff;
3119
3120    return false;
3121  } else {
3122    return false;
3123  }
3124}
3125
3126// Note that yield semantics are defined by the scheduling class to which
3127// the thread currently belongs.  Typically, yield will _not yield to
3128// other equal or higher priority threads that reside on the dispatch queues
3129// of other CPUs.
3130
3131void os::naked_yield() {
3132  thr_yield();
3133}
3134
3135// Interface for setting lwp priorities.  If we are using T2 libthread,
3136// which forces the use of BoundThreads or we manually set UseBoundThreads,
3137// all of our threads will be assigned to real lwp's.  Using the thr_setprio
3138// function is meaningless in this mode so we must adjust the real lwp's priority
3139// The routines below implement the getting and setting of lwp priorities.
3140//
3141// Note: T2 is now the only supported libthread. UseBoundThreads flag is
3142//       being deprecated and all threads are now BoundThreads
3143//
3144// Note: There are three priority scales used on Solaris.  Java priotities
3145//       which range from 1 to 10, libthread "thr_setprio" scale which range
3146//       from 0 to 127, and the current scheduling class of the process we
3147//       are running in.  This is typically from -60 to +60.
3148//       The setting of the lwp priorities in done after a call to thr_setprio
3149//       so Java priorities are mapped to libthread priorities and we map from
3150//       the latter to lwp priorities.  We don't keep priorities stored in
3151//       Java priorities since some of our worker threads want to set priorities
3152//       higher than all Java threads.
3153//
3154// For related information:
3155// (1)  man -s 2 priocntl
3156// (2)  man -s 4 priocntl
3157// (3)  man dispadmin
3158// =    librt.so
3159// =    libthread/common/rtsched.c - thrp_setlwpprio().
3160// =    ps -cL <pid> ... to validate priority.
3161// =    sched_get_priority_min and _max
3162//              pthread_create
3163//              sched_setparam
3164//              pthread_setschedparam
3165//
3166// Assumptions:
3167// +    We assume that all threads in the process belong to the same
3168//              scheduling class.   IE. an homogenous process.
3169// +    Must be root or in IA group to change change "interactive" attribute.
3170//              Priocntl() will fail silently.  The only indication of failure is when
3171//              we read-back the value and notice that it hasn't changed.
3172// +    Interactive threads enter the runq at the head, non-interactive at the tail.
3173// +    For RT, change timeslice as well.  Invariant:
3174//              constant "priority integral"
3175//              Konst == TimeSlice * (60-Priority)
3176//              Given a priority, compute appropriate timeslice.
3177// +    Higher numerical values have higher priority.
3178
3179// sched class attributes
3180typedef struct {
3181  int   schedPolicy;              // classID
3182  int   maxPrio;
3183  int   minPrio;
3184} SchedInfo;
3185
3186
3187static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3188
3189#ifdef ASSERT
3190static int  ReadBackValidate = 1;
3191#endif
3192static int  myClass     = 0;
3193static int  myMin       = 0;
3194static int  myMax       = 0;
3195static int  myCur       = 0;
3196static bool priocntl_enable = false;
3197
3198static const int criticalPrio = FXCriticalPriority;
3199static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3200
3201
3202// lwp_priocntl_init
3203//
3204// Try to determine the priority scale for our process.
3205//
3206// Return errno or 0 if OK.
3207//
3208static int lwp_priocntl_init() {
3209  int rslt;
3210  pcinfo_t ClassInfo;
3211  pcparms_t ParmInfo;
3212  int i;
3213
3214  if (!UseThreadPriorities) return 0;
3215
3216  // If ThreadPriorityPolicy is 1, switch tables
3217  if (ThreadPriorityPolicy == 1) {
3218    for (i = 0; i < CriticalPriority+1; i++)
3219      os::java_to_os_priority[i] = prio_policy1[i];
3220  }
3221  if (UseCriticalJavaThreadPriority) {
3222    // MaxPriority always maps to the FX scheduling class and criticalPrio.
3223    // See set_native_priority() and set_lwp_class_and_priority().
3224    // Save original MaxPriority mapping in case attempt to
3225    // use critical priority fails.
3226    java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3227    // Set negative to distinguish from other priorities
3228    os::java_to_os_priority[MaxPriority] = -criticalPrio;
3229  }
3230
3231  // Get IDs for a set of well-known scheduling classes.
3232  // TODO-FIXME: GETCLINFO returns the current # of classes in the
3233  // the system.  We should have a loop that iterates over the
3234  // classID values, which are known to be "small" integers.
3235
3236  strcpy(ClassInfo.pc_clname, "TS");
3237  ClassInfo.pc_cid = -1;
3238  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3239  if (rslt < 0) return errno;
3240  assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3241  tsLimits.schedPolicy = ClassInfo.pc_cid;
3242  tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3243  tsLimits.minPrio = -tsLimits.maxPrio;
3244
3245  strcpy(ClassInfo.pc_clname, "IA");
3246  ClassInfo.pc_cid = -1;
3247  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3248  if (rslt < 0) return errno;
3249  assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3250  iaLimits.schedPolicy = ClassInfo.pc_cid;
3251  iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3252  iaLimits.minPrio = -iaLimits.maxPrio;
3253
3254  strcpy(ClassInfo.pc_clname, "RT");
3255  ClassInfo.pc_cid = -1;
3256  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3257  if (rslt < 0) return errno;
3258  assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3259  rtLimits.schedPolicy = ClassInfo.pc_cid;
3260  rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3261  rtLimits.minPrio = 0;
3262
3263  strcpy(ClassInfo.pc_clname, "FX");
3264  ClassInfo.pc_cid = -1;
3265  rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3266  if (rslt < 0) return errno;
3267  assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3268  fxLimits.schedPolicy = ClassInfo.pc_cid;
3269  fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3270  fxLimits.minPrio = 0;
3271
3272  // Query our "current" scheduling class.
3273  // This will normally be IA, TS or, rarely, FX or RT.
3274  memset(&ParmInfo, 0, sizeof(ParmInfo));
3275  ParmInfo.pc_cid = PC_CLNULL;
3276  rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3277  if (rslt < 0) return errno;
3278  myClass = ParmInfo.pc_cid;
3279
3280  // We now know our scheduling classId, get specific information
3281  // about the class.
3282  ClassInfo.pc_cid = myClass;
3283  ClassInfo.pc_clname[0] = 0;
3284  rslt = priocntl((idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3285  if (rslt < 0) return errno;
3286
3287  if (ThreadPriorityVerbose) {
3288    tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3289  }
3290
3291  memset(&ParmInfo, 0, sizeof(pcparms_t));
3292  ParmInfo.pc_cid = PC_CLNULL;
3293  rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3294  if (rslt < 0) return errno;
3295
3296  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3297    myMin = rtLimits.minPrio;
3298    myMax = rtLimits.maxPrio;
3299  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3300    iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3301    myMin = iaLimits.minPrio;
3302    myMax = iaLimits.maxPrio;
3303    myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3304  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3305    tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3306    myMin = tsLimits.minPrio;
3307    myMax = tsLimits.maxPrio;
3308    myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3309  } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3310    fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3311    myMin = fxLimits.minPrio;
3312    myMax = fxLimits.maxPrio;
3313    myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3314  } else {
3315    // No clue - punt
3316    if (ThreadPriorityVerbose) {
3317      tty->print_cr("Unknown scheduling class: %s ... \n",
3318                    ClassInfo.pc_clname);
3319    }
3320    return EINVAL;      // no clue, punt
3321  }
3322
3323  if (ThreadPriorityVerbose) {
3324    tty->print_cr("Thread priority Range: [%d..%d]\n", myMin, myMax);
3325  }
3326
3327  priocntl_enable = true;  // Enable changing priorities
3328  return 0;
3329}
3330
3331#define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3332#define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3333#define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3334#define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3335
3336
3337// scale_to_lwp_priority
3338//
3339// Convert from the libthread "thr_setprio" scale to our current
3340// lwp scheduling class scale.
3341//
3342static int scale_to_lwp_priority(int rMin, int rMax, int x) {
3343  int v;
3344
3345  if (x == 127) return rMax;            // avoid round-down
3346  v = (((x*(rMax-rMin)))/128)+rMin;
3347  return v;
3348}
3349
3350
3351// set_lwp_class_and_priority
3352int set_lwp_class_and_priority(int ThreadID, int lwpid,
3353                               int newPrio, int new_class, bool scale) {
3354  int rslt;
3355  int Actual, Expected, prv;
3356  pcparms_t ParmInfo;                   // for GET-SET
3357#ifdef ASSERT
3358  pcparms_t ReadBack;                   // for readback
3359#endif
3360
3361  // Set priority via PC_GETPARMS, update, PC_SETPARMS
3362  // Query current values.
3363  // TODO: accelerate this by eliminating the PC_GETPARMS call.
3364  // Cache "pcparms_t" in global ParmCache.
3365  // TODO: elide set-to-same-value
3366
3367  // If something went wrong on init, don't change priorities.
3368  if (!priocntl_enable) {
3369    if (ThreadPriorityVerbose) {
3370      tty->print_cr("Trying to set priority but init failed, ignoring");
3371    }
3372    return EINVAL;
3373  }
3374
3375  // If lwp hasn't started yet, just return
3376  // the _start routine will call us again.
3377  if (lwpid <= 0) {
3378    if (ThreadPriorityVerbose) {
3379      tty->print_cr("deferring the set_lwp_class_and_priority of thread "
3380                    INTPTR_FORMAT " to %d, lwpid not set",
3381                    ThreadID, newPrio);
3382    }
3383    return 0;
3384  }
3385
3386  if (ThreadPriorityVerbose) {
3387    tty->print_cr ("set_lwp_class_and_priority("
3388                   INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3389                   ThreadID, lwpid, newPrio);
3390  }
3391
3392  memset(&ParmInfo, 0, sizeof(pcparms_t));
3393  ParmInfo.pc_cid = PC_CLNULL;
3394  rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3395  if (rslt < 0) return errno;
3396
3397  int cur_class = ParmInfo.pc_cid;
3398  ParmInfo.pc_cid = (id_t)new_class;
3399
3400  if (new_class == rtLimits.schedPolicy) {
3401    rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3402    rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3403                                                       rtLimits.maxPrio, newPrio)
3404                               : newPrio;
3405    rtInfo->rt_tqsecs  = RT_NOCHANGE;
3406    rtInfo->rt_tqnsecs = RT_NOCHANGE;
3407    if (ThreadPriorityVerbose) {
3408      tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3409    }
3410  } else if (new_class == iaLimits.schedPolicy) {
3411    iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3412    int maxClamped     = MIN2(iaLimits.maxPrio,
3413                              cur_class == new_class
3414                              ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3415    iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3416                                                       maxClamped, newPrio)
3417                               : newPrio;
3418    iaInfo->ia_uprilim = cur_class == new_class
3419                           ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3420    iaInfo->ia_mode    = IA_NOCHANGE;
3421    if (ThreadPriorityVerbose) {
3422      tty->print_cr("IA: [%d...%d] %d->%d\n",
3423                    iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3424    }
3425  } else if (new_class == tsLimits.schedPolicy) {
3426    tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3427    int maxClamped     = MIN2(tsLimits.maxPrio,
3428                              cur_class == new_class
3429                              ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3430    tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3431                                                       maxClamped, newPrio)
3432                               : newPrio;
3433    tsInfo->ts_uprilim = cur_class == new_class
3434                           ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
3435    if (ThreadPriorityVerbose) {
3436      tty->print_cr("TS: [%d...%d] %d->%d\n",
3437                    tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3438    }
3439  } else if (new_class == fxLimits.schedPolicy) {
3440    fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
3441    int maxClamped     = MIN2(fxLimits.maxPrio,
3442                              cur_class == new_class
3443                              ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
3444    fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
3445                                                       maxClamped, newPrio)
3446                               : newPrio;
3447    fxInfo->fx_uprilim = cur_class == new_class
3448                           ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
3449    fxInfo->fx_tqsecs  = FX_NOCHANGE;
3450    fxInfo->fx_tqnsecs = FX_NOCHANGE;
3451    if (ThreadPriorityVerbose) {
3452      tty->print_cr("FX: [%d...%d] %d->%d\n",
3453                    fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
3454    }
3455  } else {
3456    if (ThreadPriorityVerbose) {
3457      tty->print_cr("Unknown new scheduling class %d\n", new_class);
3458    }
3459    return EINVAL;    // no clue, punt
3460  }
3461
3462  rslt = priocntl(P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3463  if (ThreadPriorityVerbose && rslt) {
3464    tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3465  }
3466  if (rslt < 0) return errno;
3467
3468#ifdef ASSERT
3469  // Sanity check: read back what we just attempted to set.
3470  // In theory it could have changed in the interim ...
3471  //
3472  // The priocntl system call is tricky.
3473  // Sometimes it'll validate the priority value argument and
3474  // return EINVAL if unhappy.  At other times it fails silently.
3475  // Readbacks are prudent.
3476
3477  if (!ReadBackValidate) return 0;
3478
3479  memset(&ReadBack, 0, sizeof(pcparms_t));
3480  ReadBack.pc_cid = PC_CLNULL;
3481  rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3482  assert(rslt >= 0, "priocntl failed");
3483  Actual = Expected = 0xBAD;
3484  assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3485  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3486    Actual   = RTPRI(ReadBack)->rt_pri;
3487    Expected = RTPRI(ParmInfo)->rt_pri;
3488  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3489    Actual   = IAPRI(ReadBack)->ia_upri;
3490    Expected = IAPRI(ParmInfo)->ia_upri;
3491  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3492    Actual   = TSPRI(ReadBack)->ts_upri;
3493    Expected = TSPRI(ParmInfo)->ts_upri;
3494  } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3495    Actual   = FXPRI(ReadBack)->fx_upri;
3496    Expected = FXPRI(ParmInfo)->fx_upri;
3497  } else {
3498    if (ThreadPriorityVerbose) {
3499      tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
3500                    ParmInfo.pc_cid);
3501    }
3502  }
3503
3504  if (Actual != Expected) {
3505    if (ThreadPriorityVerbose) {
3506      tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3507                     lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3508    }
3509  }
3510#endif
3511
3512  return 0;
3513}
3514
3515// Solaris only gives access to 128 real priorities at a time,
3516// so we expand Java's ten to fill this range.  This would be better
3517// if we dynamically adjusted relative priorities.
3518//
3519// The ThreadPriorityPolicy option allows us to select 2 different
3520// priority scales.
3521//
3522// ThreadPriorityPolicy=0
3523// Since the Solaris' default priority is MaximumPriority, we do not
3524// set a priority lower than Max unless a priority lower than
3525// NormPriority is requested.
3526//
3527// ThreadPriorityPolicy=1
3528// This mode causes the priority table to get filled with
3529// linear values.  NormPriority get's mapped to 50% of the
3530// Maximum priority an so on.  This will cause VM threads
3531// to get unfair treatment against other Solaris processes
3532// which do not explicitly alter their thread priorities.
3533
3534int os::java_to_os_priority[CriticalPriority + 1] = {
3535  -99999,         // 0 Entry should never be used
3536
3537  0,              // 1 MinPriority
3538  32,             // 2
3539  64,             // 3
3540
3541  96,             // 4
3542  127,            // 5 NormPriority
3543  127,            // 6
3544
3545  127,            // 7
3546  127,            // 8
3547  127,            // 9 NearMaxPriority
3548
3549  127,            // 10 MaxPriority
3550
3551  -criticalPrio   // 11 CriticalPriority
3552};
3553
3554OSReturn os::set_native_priority(Thread* thread, int newpri) {
3555  OSThread* osthread = thread->osthread();
3556
3557  // Save requested priority in case the thread hasn't been started
3558  osthread->set_native_priority(newpri);
3559
3560  // Check for critical priority request
3561  bool fxcritical = false;
3562  if (newpri == -criticalPrio) {
3563    fxcritical = true;
3564    newpri = criticalPrio;
3565  }
3566
3567  assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3568  if (!UseThreadPriorities) return OS_OK;
3569
3570  int status = 0;
3571
3572  if (!fxcritical) {
3573    // Use thr_setprio only if we have a priority that thr_setprio understands
3574    status = thr_setprio(thread->osthread()->thread_id(), newpri);
3575  }
3576
3577  int lwp_status =
3578          set_lwp_class_and_priority(osthread->thread_id(),
3579                                     osthread->lwp_id(),
3580                                     newpri,
3581                                     fxcritical ? fxLimits.schedPolicy : myClass,
3582                                     !fxcritical);
3583  if (lwp_status != 0 && fxcritical) {
3584    // Try again, this time without changing the scheduling class
3585    newpri = java_MaxPriority_to_os_priority;
3586    lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
3587                                            osthread->lwp_id(),
3588                                            newpri, myClass, false);
3589  }
3590  status |= lwp_status;
3591  return (status == 0) ? OS_OK : OS_ERR;
3592}
3593
3594
3595OSReturn os::get_native_priority(const Thread* const thread,
3596                                 int *priority_ptr) {
3597  int p;
3598  if (!UseThreadPriorities) {
3599    *priority_ptr = NormalPriority;
3600    return OS_OK;
3601  }
3602  int status = thr_getprio(thread->osthread()->thread_id(), &p);
3603  if (status != 0) {
3604    return OS_ERR;
3605  }
3606  *priority_ptr = p;
3607  return OS_OK;
3608}
3609
3610
3611// Hint to the underlying OS that a task switch would not be good.
3612// Void return because it's a hint and can fail.
3613void os::hint_no_preempt() {
3614  schedctl_start(schedctl_init());
3615}
3616
3617static void resume_clear_context(OSThread *osthread) {
3618  osthread->set_ucontext(NULL);
3619}
3620
3621static void suspend_save_context(OSThread *osthread, ucontext_t* context) {
3622  osthread->set_ucontext(context);
3623}
3624
3625static PosixSemaphore sr_semaphore;
3626
3627void os::Solaris::SR_handler(Thread* thread, ucontext_t* uc) {
3628  // Save and restore errno to avoid confusing native code with EINTR
3629  // after sigsuspend.
3630  int old_errno = errno;
3631
3632  OSThread* osthread = thread->osthread();
3633  assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3634
3635  os::SuspendResume::State current = osthread->sr.state();
3636  if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3637    suspend_save_context(osthread, uc);
3638
3639    // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3640    os::SuspendResume::State state = osthread->sr.suspended();
3641    if (state == os::SuspendResume::SR_SUSPENDED) {
3642      sigset_t suspend_set;  // signals for sigsuspend()
3643
3644      // get current set of blocked signals and unblock resume signal
3645      pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3646      sigdelset(&suspend_set, os::Solaris::SIGasync());
3647
3648      sr_semaphore.signal();
3649      // wait here until we are resumed
3650      while (1) {
3651        sigsuspend(&suspend_set);
3652
3653        os::SuspendResume::State result = osthread->sr.running();
3654        if (result == os::SuspendResume::SR_RUNNING) {
3655          sr_semaphore.signal();
3656          break;
3657        }
3658      }
3659
3660    } else if (state == os::SuspendResume::SR_RUNNING) {
3661      // request was cancelled, continue
3662    } else {
3663      ShouldNotReachHere();
3664    }
3665
3666    resume_clear_context(osthread);
3667  } else if (current == os::SuspendResume::SR_RUNNING) {
3668    // request was cancelled, continue
3669  } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3670    // ignore
3671  } else {
3672    // ignore
3673  }
3674
3675  errno = old_errno;
3676}
3677
3678void os::print_statistics() {
3679}
3680
3681bool os::message_box(const char* title, const char* message) {
3682  int i;
3683  fdStream err(defaultStream::error_fd());
3684  for (i = 0; i < 78; i++) err.print_raw("=");
3685  err.cr();
3686  err.print_raw_cr(title);
3687  for (i = 0; i < 78; i++) err.print_raw("-");
3688  err.cr();
3689  err.print_raw_cr(message);
3690  for (i = 0; i < 78; i++) err.print_raw("=");
3691  err.cr();
3692
3693  char buf[16];
3694  // Prevent process from exiting upon "read error" without consuming all CPU
3695  while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3696
3697  return buf[0] == 'y' || buf[0] == 'Y';
3698}
3699
3700static int sr_notify(OSThread* osthread) {
3701  int status = thr_kill(osthread->thread_id(), os::Solaris::SIGasync());
3702  assert_status(status == 0, status, "thr_kill");
3703  return status;
3704}
3705
3706// "Randomly" selected value for how long we want to spin
3707// before bailing out on suspending a thread, also how often
3708// we send a signal to a thread we want to resume
3709static const int RANDOMLY_LARGE_INTEGER = 1000000;
3710static const int RANDOMLY_LARGE_INTEGER2 = 100;
3711
3712static bool do_suspend(OSThread* osthread) {
3713  assert(osthread->sr.is_running(), "thread should be running");
3714  assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3715
3716  // mark as suspended and send signal
3717  if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3718    // failed to switch, state wasn't running?
3719    ShouldNotReachHere();
3720    return false;
3721  }
3722
3723  if (sr_notify(osthread) != 0) {
3724    ShouldNotReachHere();
3725  }
3726
3727  // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3728  while (true) {
3729    if (sr_semaphore.timedwait(0, 2000 * NANOSECS_PER_MILLISEC)) {
3730      break;
3731    } else {
3732      // timeout
3733      os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3734      if (cancelled == os::SuspendResume::SR_RUNNING) {
3735        return false;
3736      } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3737        // make sure that we consume the signal on the semaphore as well
3738        sr_semaphore.wait();
3739        break;
3740      } else {
3741        ShouldNotReachHere();
3742        return false;
3743      }
3744    }
3745  }
3746
3747  guarantee(osthread->sr.is_suspended(), "Must be suspended");
3748  return true;
3749}
3750
3751static void do_resume(OSThread* osthread) {
3752  assert(osthread->sr.is_suspended(), "thread should be suspended");
3753  assert(!sr_semaphore.trywait(), "invalid semaphore state");
3754
3755  if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3756    // failed to switch to WAKEUP_REQUEST
3757    ShouldNotReachHere();
3758    return;
3759  }
3760
3761  while (true) {
3762    if (sr_notify(osthread) == 0) {
3763      if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3764        if (osthread->sr.is_running()) {
3765          return;
3766        }
3767      }
3768    } else {
3769      ShouldNotReachHere();
3770    }
3771  }
3772
3773  guarantee(osthread->sr.is_running(), "Must be running!");
3774}
3775
3776void os::SuspendedThreadTask::internal_do_task() {
3777  if (do_suspend(_thread->osthread())) {
3778    SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3779    do_task(context);
3780    do_resume(_thread->osthread());
3781  }
3782}
3783
3784class PcFetcher : public os::SuspendedThreadTask {
3785 public:
3786  PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
3787  ExtendedPC result();
3788 protected:
3789  void do_task(const os::SuspendedThreadTaskContext& context);
3790 private:
3791  ExtendedPC _epc;
3792};
3793
3794ExtendedPC PcFetcher::result() {
3795  guarantee(is_done(), "task is not done yet.");
3796  return _epc;
3797}
3798
3799void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
3800  Thread* thread = context.thread();
3801  OSThread* osthread = thread->osthread();
3802  if (osthread->ucontext() != NULL) {
3803    _epc = os::Solaris::ucontext_get_pc((const ucontext_t *) context.ucontext());
3804  } else {
3805    // NULL context is unexpected, double-check this is the VMThread
3806    guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3807  }
3808}
3809
3810// A lightweight implementation that does not suspend the target thread and
3811// thus returns only a hint. Used for profiling only!
3812ExtendedPC os::get_thread_pc(Thread* thread) {
3813  // Make sure that it is called by the watcher and the Threads lock is owned.
3814  assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
3815  // For now, is only used to profile the VM Thread
3816  assert(thread->is_VM_thread(), "Can only be called for VMThread");
3817  PcFetcher fetcher(thread);
3818  fetcher.run();
3819  return fetcher.result();
3820}
3821
3822
3823// This does not do anything on Solaris. This is basically a hook for being
3824// able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
3825void os::os_exception_wrapper(java_call_t f, JavaValue* value,
3826                              const methodHandle& method, JavaCallArguments* args,
3827                              Thread* thread) {
3828  f(value, method, args, thread);
3829}
3830
3831// This routine may be used by user applications as a "hook" to catch signals.
3832// The user-defined signal handler must pass unrecognized signals to this
3833// routine, and if it returns true (non-zero), then the signal handler must
3834// return immediately.  If the flag "abort_if_unrecognized" is true, then this
3835// routine will never retun false (zero), but instead will execute a VM panic
3836// routine kill the process.
3837//
3838// If this routine returns false, it is OK to call it again.  This allows
3839// the user-defined signal handler to perform checks either before or after
3840// the VM performs its own checks.  Naturally, the user code would be making
3841// a serious error if it tried to handle an exception (such as a null check
3842// or breakpoint) that the VM was generating for its own correct operation.
3843//
3844// This routine may recognize any of the following kinds of signals:
3845// SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
3846// os::Solaris::SIGasync
3847// It should be consulted by handlers for any of those signals.
3848//
3849// The caller of this routine must pass in the three arguments supplied
3850// to the function referred to in the "sa_sigaction" (not the "sa_handler")
3851// field of the structure passed to sigaction().  This routine assumes that
3852// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3853//
3854// Note that the VM will print warnings if it detects conflicting signal
3855// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3856//
3857extern "C" JNIEXPORT int JVM_handle_solaris_signal(int signo,
3858                                                   siginfo_t* siginfo,
3859                                                   void* ucontext,
3860                                                   int abort_if_unrecognized);
3861
3862
3863void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
3864  int orig_errno = errno;  // Preserve errno value over signal handler.
3865  JVM_handle_solaris_signal(sig, info, ucVoid, true);
3866  errno = orig_errno;
3867}
3868
3869// This boolean allows users to forward their own non-matching signals
3870// to JVM_handle_solaris_signal, harmlessly.
3871bool os::Solaris::signal_handlers_are_installed = false;
3872
3873// For signal-chaining
3874bool os::Solaris::libjsig_is_loaded = false;
3875typedef struct sigaction *(*get_signal_t)(int);
3876get_signal_t os::Solaris::get_signal_action = NULL;
3877
3878struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
3879  struct sigaction *actp = NULL;
3880
3881  if ((libjsig_is_loaded)  && (sig <= Maxlibjsigsigs)) {
3882    // Retrieve the old signal handler from libjsig
3883    actp = (*get_signal_action)(sig);
3884  }
3885  if (actp == NULL) {
3886    // Retrieve the preinstalled signal handler from jvm
3887    actp = get_preinstalled_handler(sig);
3888  }
3889
3890  return actp;
3891}
3892
3893static bool call_chained_handler(struct sigaction *actp, int sig,
3894                                 siginfo_t *siginfo, void *context) {
3895  // Call the old signal handler
3896  if (actp->sa_handler == SIG_DFL) {
3897    // It's more reasonable to let jvm treat it as an unexpected exception
3898    // instead of taking the default action.
3899    return false;
3900  } else if (actp->sa_handler != SIG_IGN) {
3901    if ((actp->sa_flags & SA_NODEFER) == 0) {
3902      // automaticlly block the signal
3903      sigaddset(&(actp->sa_mask), sig);
3904    }
3905
3906    sa_handler_t hand;
3907    sa_sigaction_t sa;
3908    bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3909    // retrieve the chained handler
3910    if (siginfo_flag_set) {
3911      sa = actp->sa_sigaction;
3912    } else {
3913      hand = actp->sa_handler;
3914    }
3915
3916    if ((actp->sa_flags & SA_RESETHAND) != 0) {
3917      actp->sa_handler = SIG_DFL;
3918    }
3919
3920    // try to honor the signal mask
3921    sigset_t oset;
3922    pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3923
3924    // call into the chained handler
3925    if (siginfo_flag_set) {
3926      (*sa)(sig, siginfo, context);
3927    } else {
3928      (*hand)(sig);
3929    }
3930
3931    // restore the signal mask
3932    pthread_sigmask(SIG_SETMASK, &oset, 0);
3933  }
3934  // Tell jvm's signal handler the signal is taken care of.
3935  return true;
3936}
3937
3938bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3939  bool chained = false;
3940  // signal-chaining
3941  if (UseSignalChaining) {
3942    struct sigaction *actp = get_chained_signal_action(sig);
3943    if (actp != NULL) {
3944      chained = call_chained_handler(actp, sig, siginfo, context);
3945    }
3946  }
3947  return chained;
3948}
3949
3950struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
3951  assert((chainedsigactions != (struct sigaction *)NULL) &&
3952         (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3953  if (preinstalled_sigs[sig] != 0) {
3954    return &chainedsigactions[sig];
3955  }
3956  return NULL;
3957}
3958
3959void os::Solaris::save_preinstalled_handler(int sig,
3960                                            struct sigaction& oldAct) {
3961  assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
3962  assert((chainedsigactions != (struct sigaction *)NULL) &&
3963         (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3964  chainedsigactions[sig] = oldAct;
3965  preinstalled_sigs[sig] = 1;
3966}
3967
3968void os::Solaris::set_signal_handler(int sig, bool set_installed,
3969                                     bool oktochain) {
3970  // Check for overwrite.
3971  struct sigaction oldAct;
3972  sigaction(sig, (struct sigaction*)NULL, &oldAct);
3973  void* oldhand =
3974      oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3975                          : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3976  if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3977      oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3978      oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
3979    if (AllowUserSignalHandlers || !set_installed) {
3980      // Do not overwrite; user takes responsibility to forward to us.
3981      return;
3982    } else if (UseSignalChaining) {
3983      if (oktochain) {
3984        // save the old handler in jvm
3985        save_preinstalled_handler(sig, oldAct);
3986      } else {
3987        vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal.");
3988      }
3989      // libjsig also interposes the sigaction() call below and saves the
3990      // old sigaction on it own.
3991    } else {
3992      fatal("Encountered unexpected pre-existing sigaction handler "
3993            "%#lx for signal %d.", (long)oldhand, sig);
3994    }
3995  }
3996
3997  struct sigaction sigAct;
3998  sigfillset(&(sigAct.sa_mask));
3999  sigAct.sa_handler = SIG_DFL;
4000
4001  sigAct.sa_sigaction = signalHandler;
4002  // Handle SIGSEGV on alternate signal stack if
4003  // not using stack banging
4004  if (!UseStackBanging && sig == SIGSEGV) {
4005    sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
4006  } else {
4007    sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
4008  }
4009  os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
4010
4011  sigaction(sig, &sigAct, &oldAct);
4012
4013  void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4014                                       : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4015  assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4016}
4017
4018
4019#define DO_SIGNAL_CHECK(sig)                      \
4020  do {                                            \
4021    if (!sigismember(&check_signal_done, sig)) {  \
4022      os::Solaris::check_signal_handler(sig);     \
4023    }                                             \
4024  } while (0)
4025
4026// This method is a periodic task to check for misbehaving JNI applications
4027// under CheckJNI, we can add any periodic checks here
4028
4029void os::run_periodic_checks() {
4030  // A big source of grief is hijacking virt. addr 0x0 on Solaris,
4031  // thereby preventing a NULL checks.
4032  if (!check_addr0_done) check_addr0_done = check_addr0(tty);
4033
4034  if (check_signals == false) return;
4035
4036  // SEGV and BUS if overridden could potentially prevent
4037  // generation of hs*.log in the event of a crash, debugging
4038  // such a case can be very challenging, so we absolutely
4039  // check for the following for a good measure:
4040  DO_SIGNAL_CHECK(SIGSEGV);
4041  DO_SIGNAL_CHECK(SIGILL);
4042  DO_SIGNAL_CHECK(SIGFPE);
4043  DO_SIGNAL_CHECK(SIGBUS);
4044  DO_SIGNAL_CHECK(SIGPIPE);
4045  DO_SIGNAL_CHECK(SIGXFSZ);
4046
4047  // ReduceSignalUsage allows the user to override these handlers
4048  // see comments at the very top and jvm_solaris.h
4049  if (!ReduceSignalUsage) {
4050    DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4051    DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4052    DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4053    DO_SIGNAL_CHECK(BREAK_SIGNAL);
4054  }
4055
4056  // See comments above for using JVM1/JVM2
4057  DO_SIGNAL_CHECK(os::Solaris::SIGasync());
4058
4059}
4060
4061typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4062
4063static os_sigaction_t os_sigaction = NULL;
4064
4065void os::Solaris::check_signal_handler(int sig) {
4066  char buf[O_BUFLEN];
4067  address jvmHandler = NULL;
4068
4069  struct sigaction act;
4070  if (os_sigaction == NULL) {
4071    // only trust the default sigaction, in case it has been interposed
4072    os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4073    if (os_sigaction == NULL) return;
4074  }
4075
4076  os_sigaction(sig, (struct sigaction*)NULL, &act);
4077
4078  address thisHandler = (act.sa_flags & SA_SIGINFO)
4079    ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4080    : CAST_FROM_FN_PTR(address, act.sa_handler);
4081
4082
4083  switch (sig) {
4084  case SIGSEGV:
4085  case SIGBUS:
4086  case SIGFPE:
4087  case SIGPIPE:
4088  case SIGXFSZ:
4089  case SIGILL:
4090    jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4091    break;
4092
4093  case SHUTDOWN1_SIGNAL:
4094  case SHUTDOWN2_SIGNAL:
4095  case SHUTDOWN3_SIGNAL:
4096  case BREAK_SIGNAL:
4097    jvmHandler = (address)user_handler();
4098    break;
4099
4100  default:
4101    int asynsig = os::Solaris::SIGasync();
4102
4103    if (sig == asynsig) {
4104      jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4105    } else {
4106      return;
4107    }
4108    break;
4109  }
4110
4111
4112  if (thisHandler != jvmHandler) {
4113    tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4114    tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4115    tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4116    // No need to check this sig any longer
4117    sigaddset(&check_signal_done, sig);
4118    // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4119    if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4120      tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4121                    exception_name(sig, buf, O_BUFLEN));
4122    }
4123  } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4124    tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4125    tty->print("expected:");
4126    os::Posix::print_sa_flags(tty, os::Solaris::get_our_sigflags(sig));
4127    tty->cr();
4128    tty->print("  found:");
4129    os::Posix::print_sa_flags(tty, act.sa_flags);
4130    tty->cr();
4131    // No need to check this sig any longer
4132    sigaddset(&check_signal_done, sig);
4133  }
4134
4135  // Print all the signal handler state
4136  if (sigismember(&check_signal_done, sig)) {
4137    print_signal_handlers(tty, buf, O_BUFLEN);
4138  }
4139
4140}
4141
4142void os::Solaris::install_signal_handlers() {
4143  bool libjsigdone = false;
4144  signal_handlers_are_installed = true;
4145
4146  // signal-chaining
4147  typedef void (*signal_setting_t)();
4148  signal_setting_t begin_signal_setting = NULL;
4149  signal_setting_t end_signal_setting = NULL;
4150  begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4151                                        dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4152  if (begin_signal_setting != NULL) {
4153    end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4154                                        dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4155    get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4156                                       dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4157    get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4158                                         dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4159    libjsig_is_loaded = true;
4160    if (os::Solaris::get_libjsig_version != NULL) {
4161      libjsigversion =  (*os::Solaris::get_libjsig_version)();
4162    }
4163    assert(UseSignalChaining, "should enable signal-chaining");
4164  }
4165  if (libjsig_is_loaded) {
4166    // Tell libjsig jvm is setting signal handlers
4167    (*begin_signal_setting)();
4168  }
4169
4170  set_signal_handler(SIGSEGV, true, true);
4171  set_signal_handler(SIGPIPE, true, true);
4172  set_signal_handler(SIGXFSZ, true, true);
4173  set_signal_handler(SIGBUS, true, true);
4174  set_signal_handler(SIGILL, true, true);
4175  set_signal_handler(SIGFPE, true, true);
4176
4177
4178  if (os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4179    // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4180    // can not register overridable signals which might be > 32
4181    if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4182      // Tell libjsig jvm has finished setting signal handlers
4183      (*end_signal_setting)();
4184      libjsigdone = true;
4185    }
4186  }
4187
4188  set_signal_handler(os::Solaris::SIGasync(), true, true);
4189
4190  if (libjsig_is_loaded && !libjsigdone) {
4191    // Tell libjsig jvm finishes setting signal handlers
4192    (*end_signal_setting)();
4193  }
4194
4195  // We don't activate signal checker if libjsig is in place, we trust ourselves
4196  // and if UserSignalHandler is installed all bets are off.
4197  // Log that signal checking is off only if -verbose:jni is specified.
4198  if (CheckJNICalls) {
4199    if (libjsig_is_loaded) {
4200      if (PrintJNIResolving) {
4201        tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4202      }
4203      check_signals = false;
4204    }
4205    if (AllowUserSignalHandlers) {
4206      if (PrintJNIResolving) {
4207        tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4208      }
4209      check_signals = false;
4210    }
4211  }
4212}
4213
4214
4215void report_error(const char* file_name, int line_no, const char* title,
4216                  const char* format, ...);
4217
4218// (Static) wrapper for getisax(2) call.
4219os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4220
4221// (Static) wrappers for the liblgrp API
4222os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4223os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4224os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4225os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4226os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4227os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4228os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4229os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4230os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4231
4232// (Static) wrapper for meminfo() call.
4233os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4234
4235static address resolve_symbol_lazy(const char* name) {
4236  address addr = (address) dlsym(RTLD_DEFAULT, name);
4237  if (addr == NULL) {
4238    // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4239    addr = (address) dlsym(RTLD_NEXT, name);
4240  }
4241  return addr;
4242}
4243
4244static address resolve_symbol(const char* name) {
4245  address addr = resolve_symbol_lazy(name);
4246  if (addr == NULL) {
4247    fatal(dlerror());
4248  }
4249  return addr;
4250}
4251
4252void os::Solaris::libthread_init() {
4253  address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4254
4255  lwp_priocntl_init();
4256
4257  // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4258  if (func == NULL) {
4259    func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4260    // Guarantee that this VM is running on an new enough OS (5.6 or
4261    // later) that it will have a new enough libthread.so.
4262    guarantee(func != NULL, "libthread.so is too old.");
4263  }
4264
4265  int size;
4266  void (*handler_info_func)(address *, int *);
4267  handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4268  handler_info_func(&handler_start, &size);
4269  handler_end = handler_start + size;
4270}
4271
4272
4273int_fnP_mutex_tP os::Solaris::_mutex_lock;
4274int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4275int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4276int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4277int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4278int os::Solaris::_mutex_scope = USYNC_THREAD;
4279
4280int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4281int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4282int_fnP_cond_tP os::Solaris::_cond_signal;
4283int_fnP_cond_tP os::Solaris::_cond_broadcast;
4284int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4285int_fnP_cond_tP os::Solaris::_cond_destroy;
4286int os::Solaris::_cond_scope = USYNC_THREAD;
4287
4288void os::Solaris::synchronization_init() {
4289  if (UseLWPSynchronization) {
4290    os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4291    os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4292    os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4293    os::Solaris::set_mutex_init(lwp_mutex_init);
4294    os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4295    os::Solaris::set_mutex_scope(USYNC_THREAD);
4296
4297    os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4298    os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4299    os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4300    os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4301    os::Solaris::set_cond_init(lwp_cond_init);
4302    os::Solaris::set_cond_destroy(lwp_cond_destroy);
4303    os::Solaris::set_cond_scope(USYNC_THREAD);
4304  } else {
4305    os::Solaris::set_mutex_scope(USYNC_THREAD);
4306    os::Solaris::set_cond_scope(USYNC_THREAD);
4307
4308    if (UsePthreads) {
4309      os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4310      os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4311      os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4312      os::Solaris::set_mutex_init(pthread_mutex_default_init);
4313      os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4314
4315      os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4316      os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4317      os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4318      os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4319      os::Solaris::set_cond_init(pthread_cond_default_init);
4320      os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4321    } else {
4322      os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4323      os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4324      os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4325      os::Solaris::set_mutex_init(::mutex_init);
4326      os::Solaris::set_mutex_destroy(::mutex_destroy);
4327
4328      os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4329      os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4330      os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4331      os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4332      os::Solaris::set_cond_init(::cond_init);
4333      os::Solaris::set_cond_destroy(::cond_destroy);
4334    }
4335  }
4336}
4337
4338bool os::Solaris::liblgrp_init() {
4339  void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4340  if (handle != NULL) {
4341    os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4342    os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4343    os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4344    os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4345    os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4346    os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4347    os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4348    os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4349                                                      dlsym(handle, "lgrp_cookie_stale")));
4350
4351    lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4352    set_lgrp_cookie(c);
4353    return true;
4354  }
4355  return false;
4356}
4357
4358void os::Solaris::misc_sym_init() {
4359  address func;
4360
4361  // getisax
4362  func = resolve_symbol_lazy("getisax");
4363  if (func != NULL) {
4364    os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
4365  }
4366
4367  // meminfo
4368  func = resolve_symbol_lazy("meminfo");
4369  if (func != NULL) {
4370    os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
4371  }
4372}
4373
4374uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
4375  assert(_getisax != NULL, "_getisax not set");
4376  return _getisax(array, n);
4377}
4378
4379// int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4380typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4381static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4382
4383void init_pset_getloadavg_ptr(void) {
4384  pset_getloadavg_ptr =
4385    (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4386  if (pset_getloadavg_ptr == NULL) {
4387    log_warning(os)("pset_getloadavg function not found");
4388  }
4389}
4390
4391int os::Solaris::_dev_zero_fd = -1;
4392
4393// this is called _before_ the global arguments have been parsed
4394void os::init(void) {
4395  _initial_pid = getpid();
4396
4397  max_hrtime = first_hrtime = gethrtime();
4398
4399  init_random(1234567);
4400
4401  page_size = sysconf(_SC_PAGESIZE);
4402  if (page_size == -1) {
4403    fatal("os_solaris.cpp: os::init: sysconf failed (%s)", os::strerror(errno));
4404  }
4405  init_page_sizes((size_t) page_size);
4406
4407  Solaris::initialize_system_info();
4408
4409  // Initialize misc. symbols as soon as possible, so we can use them
4410  // if we need them.
4411  Solaris::misc_sym_init();
4412
4413  int fd = ::open("/dev/zero", O_RDWR);
4414  if (fd < 0) {
4415    fatal("os::init: cannot open /dev/zero (%s)", os::strerror(errno));
4416  } else {
4417    Solaris::set_dev_zero_fd(fd);
4418
4419    // Close on exec, child won't inherit.
4420    fcntl(fd, F_SETFD, FD_CLOEXEC);
4421  }
4422
4423  clock_tics_per_sec = CLK_TCK;
4424
4425  // check if dladdr1() exists; dladdr1 can provide more information than
4426  // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4427  // and is available on linker patches for 5.7 and 5.8.
4428  // libdl.so must have been loaded, this call is just an entry lookup
4429  void * hdl = dlopen("libdl.so", RTLD_NOW);
4430  if (hdl) {
4431    dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4432  }
4433
4434  // (Solaris only) this switches to calls that actually do locking.
4435  ThreadCritical::initialize();
4436
4437  main_thread = thr_self();
4438
4439  // Constant minimum stack size allowed. It must be at least
4440  // the minimum of what the OS supports (thr_min_stack()), and
4441  // enough to allow the thread to get to user bytecode execution.
4442  Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
4443
4444  // dynamic lookup of functions that may not be available in our lowest
4445  // supported Solaris release
4446  void * handle = dlopen("libc.so.1", RTLD_LAZY);
4447  if (handle != NULL) {
4448    Solaris::_pthread_setname_np =  // from 11.3
4449        (Solaris::pthread_setname_np_func_t)dlsym(handle, "pthread_setname_np");
4450
4451    _get_nsec_fromepoch =           // from 11.3.6
4452        (_get_nsec_fromepoch_func_t) dlsym(handle, "get_nsec_fromepoch");
4453  }
4454}
4455
4456// To install functions for atexit system call
4457extern "C" {
4458  static void perfMemory_exit_helper() {
4459    perfMemory_exit();
4460  }
4461}
4462
4463// this is called _after_ the global arguments have been parsed
4464jint os::init_2(void) {
4465  // try to enable extended file IO ASAP, see 6431278
4466  os::Solaris::try_enable_extended_io();
4467
4468  // Allocate a single page and mark it as readable for safepoint polling.  Also
4469  // use this first mmap call to check support for MAP_ALIGN.
4470  address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
4471                                                      page_size,
4472                                                      MAP_PRIVATE | MAP_ALIGN,
4473                                                      PROT_READ);
4474  if (polling_page == NULL) {
4475    has_map_align = false;
4476    polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
4477                                                PROT_READ);
4478  }
4479
4480  os::set_polling_page(polling_page);
4481  log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4482
4483  if (!UseMembar) {
4484    address mem_serialize_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE);
4485    guarantee(mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4486    os::set_memory_serialize_page(mem_serialize_page);
4487    log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4488  }
4489
4490  // Check minimum allowable stack size for thread creation and to initialize
4491  // the java system classes, including StackOverflowError - depends on page
4492  // size.  Add a page for compiler2 recursion in main thread.
4493  // Add in 2*BytesPerWord times page size to account for VM stack during
4494  // class initialization depending on 32 or 64 bit VM.
4495  os::Solaris::min_stack_allowed = MAX2(os::Solaris::min_stack_allowed,
4496                                        JavaThread::stack_guard_zone_size() +
4497                                        JavaThread::stack_shadow_zone_size() +
4498                                        (2*BytesPerWord COMPILER2_PRESENT(+1)) * page_size);
4499
4500  size_t threadStackSizeInBytes = ThreadStackSize * K;
4501  if (threadStackSizeInBytes != 0 &&
4502      threadStackSizeInBytes < os::Solaris::min_stack_allowed) {
4503    tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
4504                  os::Solaris::min_stack_allowed/K);
4505    return JNI_ERR;
4506  }
4507
4508  // For 64kbps there will be a 64kb page size, which makes
4509  // the usable default stack size quite a bit less.  Increase the
4510  // stack for 64kb (or any > than 8kb) pages, this increases
4511  // virtual memory fragmentation (since we're not creating the
4512  // stack on a power of 2 boundary.  The real fix for this
4513  // should be to fix the guard page mechanism.
4514
4515  if (vm_page_size() > 8*K) {
4516    threadStackSizeInBytes = (threadStackSizeInBytes != 0)
4517       ? threadStackSizeInBytes +
4518         JavaThread::stack_red_zone_size() +
4519         JavaThread::stack_yellow_zone_size()
4520       : 0;
4521    ThreadStackSize = threadStackSizeInBytes/K;
4522  }
4523
4524  // Make the stack size a multiple of the page size so that
4525  // the yellow/red zones can be guarded.
4526  JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4527                                                vm_page_size()));
4528
4529  Solaris::libthread_init();
4530
4531  if (UseNUMA) {
4532    if (!Solaris::liblgrp_init()) {
4533      UseNUMA = false;
4534    } else {
4535      size_t lgrp_limit = os::numa_get_groups_num();
4536      int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
4537      size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4538      FREE_C_HEAP_ARRAY(int, lgrp_ids);
4539      if (lgrp_num < 2) {
4540        // There's only one locality group, disable NUMA.
4541        UseNUMA = false;
4542      }
4543    }
4544    if (!UseNUMA && ForceNUMA) {
4545      UseNUMA = true;
4546    }
4547  }
4548
4549  Solaris::signal_sets_init();
4550  Solaris::init_signal_mem();
4551  Solaris::install_signal_handlers();
4552
4553  if (libjsigversion < JSIG_VERSION_1_4_1) {
4554    Maxlibjsigsigs = OLDMAXSIGNUM;
4555  }
4556
4557  // initialize synchronization primitives to use either thread or
4558  // lwp synchronization (controlled by UseLWPSynchronization)
4559  Solaris::synchronization_init();
4560
4561  if (MaxFDLimit) {
4562    // set the number of file descriptors to max. print out error
4563    // if getrlimit/setrlimit fails but continue regardless.
4564    struct rlimit nbr_files;
4565    int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4566    if (status != 0) {
4567      log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4568    } else {
4569      nbr_files.rlim_cur = nbr_files.rlim_max;
4570      status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4571      if (status != 0) {
4572        log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4573      }
4574    }
4575  }
4576
4577  // Calculate theoretical max. size of Threads to guard gainst
4578  // artifical out-of-memory situations, where all available address-
4579  // space has been reserved by thread stacks. Default stack size is 1Mb.
4580  size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4581    JavaThread::stack_size_at_create() : (1*K*K);
4582  assert(pre_thread_stack_size != 0, "Must have a stack");
4583  // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4584  // we should start doing Virtual Memory banging. Currently when the threads will
4585  // have used all but 200Mb of space.
4586  size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4587  Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4588
4589  // at-exit methods are called in the reverse order of their registration.
4590  // In Solaris 7 and earlier, atexit functions are called on return from
4591  // main or as a result of a call to exit(3C). There can be only 32 of
4592  // these functions registered and atexit() does not set errno. In Solaris
4593  // 8 and later, there is no limit to the number of functions registered
4594  // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4595  // functions are called upon dlclose(3DL) in addition to return from main
4596  // and exit(3C).
4597
4598  if (PerfAllowAtExitRegistration) {
4599    // only register atexit functions if PerfAllowAtExitRegistration is set.
4600    // atexit functions can be delayed until process exit time, which
4601    // can be problematic for embedded VM situations. Embedded VMs should
4602    // call DestroyJavaVM() to assure that VM resources are released.
4603
4604    // note: perfMemory_exit_helper atexit function may be removed in
4605    // the future if the appropriate cleanup code can be added to the
4606    // VM_Exit VMOperation's doit method.
4607    if (atexit(perfMemory_exit_helper) != 0) {
4608      warning("os::init2 atexit(perfMemory_exit_helper) failed");
4609    }
4610  }
4611
4612  // Init pset_loadavg function pointer
4613  init_pset_getloadavg_ptr();
4614
4615  return JNI_OK;
4616}
4617
4618// Mark the polling page as unreadable
4619void os::make_polling_page_unreadable(void) {
4620  if (mprotect((char *)_polling_page, page_size, PROT_NONE) != 0) {
4621    fatal("Could not disable polling page");
4622  }
4623}
4624
4625// Mark the polling page as readable
4626void os::make_polling_page_readable(void) {
4627  if (mprotect((char *)_polling_page, page_size, PROT_READ) != 0) {
4628    fatal("Could not enable polling page");
4629  }
4630}
4631
4632// OS interface.
4633
4634bool os::check_heap(bool force) { return true; }
4635
4636// Is a (classpath) directory empty?
4637bool os::dir_is_empty(const char* path) {
4638  DIR *dir = NULL;
4639  struct dirent *ptr;
4640
4641  dir = opendir(path);
4642  if (dir == NULL) return true;
4643
4644  // Scan the directory
4645  bool result = true;
4646  char buf[sizeof(struct dirent) + MAX_PATH];
4647  struct dirent *dbuf = (struct dirent *) buf;
4648  while (result && (ptr = readdir(dir, dbuf)) != NULL) {
4649    if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4650      result = false;
4651    }
4652  }
4653  closedir(dir);
4654  return result;
4655}
4656
4657// This code originates from JDK's sysOpen and open64_w
4658// from src/solaris/hpi/src/system_md.c
4659
4660int os::open(const char *path, int oflag, int mode) {
4661  if (strlen(path) > MAX_PATH - 1) {
4662    errno = ENAMETOOLONG;
4663    return -1;
4664  }
4665  int fd;
4666
4667  fd = ::open64(path, oflag, mode);
4668  if (fd == -1) return -1;
4669
4670  // If the open succeeded, the file might still be a directory
4671  {
4672    struct stat64 buf64;
4673    int ret = ::fstat64(fd, &buf64);
4674    int st_mode = buf64.st_mode;
4675
4676    if (ret != -1) {
4677      if ((st_mode & S_IFMT) == S_IFDIR) {
4678        errno = EISDIR;
4679        ::close(fd);
4680        return -1;
4681      }
4682    } else {
4683      ::close(fd);
4684      return -1;
4685    }
4686  }
4687
4688  // 32-bit Solaris systems suffer from:
4689  //
4690  // - an historical default soft limit of 256 per-process file
4691  //   descriptors that is too low for many Java programs.
4692  //
4693  // - a design flaw where file descriptors created using stdio
4694  //   fopen must be less than 256, _even_ when the first limit above
4695  //   has been raised.  This can cause calls to fopen (but not calls to
4696  //   open, for example) to fail mysteriously, perhaps in 3rd party
4697  //   native code (although the JDK itself uses fopen).  One can hardly
4698  //   criticize them for using this most standard of all functions.
4699  //
4700  // We attempt to make everything work anyways by:
4701  //
4702  // - raising the soft limit on per-process file descriptors beyond
4703  //   256
4704  //
4705  // - As of Solaris 10u4, we can request that Solaris raise the 256
4706  //   stdio fopen limit by calling function enable_extended_FILE_stdio.
4707  //   This is done in init_2 and recorded in enabled_extended_FILE_stdio
4708  //
4709  // - If we are stuck on an old (pre 10u4) Solaris system, we can
4710  //   workaround the bug by remapping non-stdio file descriptors below
4711  //   256 to ones beyond 256, which is done below.
4712  //
4713  // See:
4714  // 1085341: 32-bit stdio routines should support file descriptors >255
4715  // 6533291: Work around 32-bit Solaris stdio limit of 256 open files
4716  // 6431278: Netbeans crash on 32 bit Solaris: need to call
4717  //          enable_extended_FILE_stdio() in VM initialisation
4718  // Giri Mandalika's blog
4719  // http://technopark02.blogspot.com/2005_05_01_archive.html
4720  //
4721#ifndef  _LP64
4722  if ((!enabled_extended_FILE_stdio) && fd < 256) {
4723    int newfd = ::fcntl(fd, F_DUPFD, 256);
4724    if (newfd != -1) {
4725      ::close(fd);
4726      fd = newfd;
4727    }
4728  }
4729#endif // 32-bit Solaris
4730
4731  // All file descriptors that are opened in the JVM and not
4732  // specifically destined for a subprocess should have the
4733  // close-on-exec flag set.  If we don't set it, then careless 3rd
4734  // party native code might fork and exec without closing all
4735  // appropriate file descriptors (e.g. as we do in closeDescriptors in
4736  // UNIXProcess.c), and this in turn might:
4737  //
4738  // - cause end-of-file to fail to be detected on some file
4739  //   descriptors, resulting in mysterious hangs, or
4740  //
4741  // - might cause an fopen in the subprocess to fail on a system
4742  //   suffering from bug 1085341.
4743  //
4744  // (Yes, the default setting of the close-on-exec flag is a Unix
4745  // design flaw)
4746  //
4747  // See:
4748  // 1085341: 32-bit stdio routines should support file descriptors >255
4749  // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4750  // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4751  //
4752#ifdef FD_CLOEXEC
4753  {
4754    int flags = ::fcntl(fd, F_GETFD);
4755    if (flags != -1) {
4756      ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4757    }
4758  }
4759#endif
4760
4761  return fd;
4762}
4763
4764// create binary file, rewriting existing file if required
4765int os::create_binary_file(const char* path, bool rewrite_existing) {
4766  int oflags = O_WRONLY | O_CREAT;
4767  if (!rewrite_existing) {
4768    oflags |= O_EXCL;
4769  }
4770  return ::open64(path, oflags, S_IREAD | S_IWRITE);
4771}
4772
4773// return current position of file pointer
4774jlong os::current_file_offset(int fd) {
4775  return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4776}
4777
4778// move file pointer to the specified offset
4779jlong os::seek_to_file_offset(int fd, jlong offset) {
4780  return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4781}
4782
4783jlong os::lseek(int fd, jlong offset, int whence) {
4784  return (jlong) ::lseek64(fd, offset, whence);
4785}
4786
4787char * os::native_path(char *path) {
4788  return path;
4789}
4790
4791int os::ftruncate(int fd, jlong length) {
4792  return ::ftruncate64(fd, length);
4793}
4794
4795int os::fsync(int fd)  {
4796  RESTARTABLE_RETURN_INT(::fsync(fd));
4797}
4798
4799int os::available(int fd, jlong *bytes) {
4800  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
4801         "Assumed _thread_in_native");
4802  jlong cur, end;
4803  int mode;
4804  struct stat64 buf64;
4805
4806  if (::fstat64(fd, &buf64) >= 0) {
4807    mode = buf64.st_mode;
4808    if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4809      int n,ioctl_return;
4810
4811      RESTARTABLE(::ioctl(fd, FIONREAD, &n), ioctl_return);
4812      if (ioctl_return>= 0) {
4813        *bytes = n;
4814        return 1;
4815      }
4816    }
4817  }
4818  if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4819    return 0;
4820  } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4821    return 0;
4822  } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4823    return 0;
4824  }
4825  *bytes = end - cur;
4826  return 1;
4827}
4828
4829// Map a block of memory.
4830char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4831                        char *addr, size_t bytes, bool read_only,
4832                        bool allow_exec) {
4833  int prot;
4834  int flags;
4835
4836  if (read_only) {
4837    prot = PROT_READ;
4838    flags = MAP_SHARED;
4839  } else {
4840    prot = PROT_READ | PROT_WRITE;
4841    flags = MAP_PRIVATE;
4842  }
4843
4844  if (allow_exec) {
4845    prot |= PROT_EXEC;
4846  }
4847
4848  if (addr != NULL) {
4849    flags |= MAP_FIXED;
4850  }
4851
4852  char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4853                                     fd, file_offset);
4854  if (mapped_address == MAP_FAILED) {
4855    return NULL;
4856  }
4857  return mapped_address;
4858}
4859
4860
4861// Remap a block of memory.
4862char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4863                          char *addr, size_t bytes, bool read_only,
4864                          bool allow_exec) {
4865  // same as map_memory() on this OS
4866  return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4867                        allow_exec);
4868}
4869
4870
4871// Unmap a block of memory.
4872bool os::pd_unmap_memory(char* addr, size_t bytes) {
4873  return munmap(addr, bytes) == 0;
4874}
4875
4876void os::pause() {
4877  char filename[MAX_PATH];
4878  if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4879    jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4880  } else {
4881    jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4882  }
4883
4884  int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4885  if (fd != -1) {
4886    struct stat buf;
4887    ::close(fd);
4888    while (::stat(filename, &buf) == 0) {
4889      (void)::poll(NULL, 0, 100);
4890    }
4891  } else {
4892    jio_fprintf(stderr,
4893                "Could not open pause file '%s', continuing immediately.\n", filename);
4894  }
4895}
4896
4897#ifndef PRODUCT
4898#ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4899// Turn this on if you need to trace synch operations.
4900// Set RECORD_SYNCH_LIMIT to a large-enough value,
4901// and call record_synch_enable and record_synch_disable
4902// around the computation of interest.
4903
4904void record_synch(char* name, bool returning);  // defined below
4905
4906class RecordSynch {
4907  char* _name;
4908 public:
4909  RecordSynch(char* name) :_name(name) { record_synch(_name, false); }
4910  ~RecordSynch()                       { record_synch(_name, true); }
4911};
4912
4913#define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
4914extern "C" ret name params {                                    \
4915  typedef ret name##_t params;                                  \
4916  static name##_t* implem = NULL;                               \
4917  static int callcount = 0;                                     \
4918  if (implem == NULL) {                                         \
4919    implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
4920    if (implem == NULL)  fatal(dlerror());                      \
4921  }                                                             \
4922  ++callcount;                                                  \
4923  RecordSynch _rs(#name);                                       \
4924  inner;                                                        \
4925  return implem args;                                           \
4926}
4927// in dbx, examine callcounts this way:
4928// for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
4929
4930#define CHECK_POINTER_OK(p) \
4931  (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
4932#define CHECK_MU \
4933  if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
4934#define CHECK_CV \
4935  if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
4936#define CHECK_P(p) \
4937  if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
4938
4939#define CHECK_MUTEX(mutex_op) \
4940  CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
4941
4942CHECK_MUTEX(   mutex_lock)
4943CHECK_MUTEX(  _mutex_lock)
4944CHECK_MUTEX( mutex_unlock)
4945CHECK_MUTEX(_mutex_unlock)
4946CHECK_MUTEX( mutex_trylock)
4947CHECK_MUTEX(_mutex_trylock)
4948
4949#define CHECK_COND(cond_op) \
4950  CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU; CHECK_CV);
4951
4952CHECK_COND( cond_wait);
4953CHECK_COND(_cond_wait);
4954CHECK_COND(_cond_wait_cancel);
4955
4956#define CHECK_COND2(cond_op) \
4957  CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU; CHECK_CV);
4958
4959CHECK_COND2( cond_timedwait);
4960CHECK_COND2(_cond_timedwait);
4961CHECK_COND2(_cond_timedwait_cancel);
4962
4963// do the _lwp_* versions too
4964#define mutex_t lwp_mutex_t
4965#define cond_t  lwp_cond_t
4966CHECK_MUTEX(  _lwp_mutex_lock)
4967CHECK_MUTEX(  _lwp_mutex_unlock)
4968CHECK_MUTEX(  _lwp_mutex_trylock)
4969CHECK_MUTEX( __lwp_mutex_lock)
4970CHECK_MUTEX( __lwp_mutex_unlock)
4971CHECK_MUTEX( __lwp_mutex_trylock)
4972CHECK_MUTEX(___lwp_mutex_lock)
4973CHECK_MUTEX(___lwp_mutex_unlock)
4974
4975CHECK_COND(  _lwp_cond_wait);
4976CHECK_COND( __lwp_cond_wait);
4977CHECK_COND(___lwp_cond_wait);
4978
4979CHECK_COND2(  _lwp_cond_timedwait);
4980CHECK_COND2( __lwp_cond_timedwait);
4981#undef mutex_t
4982#undef cond_t
4983
4984CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4985CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4986CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4987CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4988CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4989CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4990CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4991CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4992
4993
4994// recording machinery:
4995
4996enum { RECORD_SYNCH_LIMIT = 200 };
4997char* record_synch_name[RECORD_SYNCH_LIMIT];
4998void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
4999bool record_synch_returning[RECORD_SYNCH_LIMIT];
5000thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
5001int record_synch_count = 0;
5002bool record_synch_enabled = false;
5003
5004// in dbx, examine recorded data this way:
5005// for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
5006
5007void record_synch(char* name, bool returning) {
5008  if (record_synch_enabled) {
5009    if (record_synch_count < RECORD_SYNCH_LIMIT) {
5010      record_synch_name[record_synch_count] = name;
5011      record_synch_returning[record_synch_count] = returning;
5012      record_synch_thread[record_synch_count] = thr_self();
5013      record_synch_arg0ptr[record_synch_count] = &name;
5014      record_synch_count++;
5015    }
5016    // put more checking code here:
5017    // ...
5018  }
5019}
5020
5021void record_synch_enable() {
5022  // start collecting trace data, if not already doing so
5023  if (!record_synch_enabled)  record_synch_count = 0;
5024  record_synch_enabled = true;
5025}
5026
5027void record_synch_disable() {
5028  // stop collecting trace data
5029  record_synch_enabled = false;
5030}
5031
5032#endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5033#endif // PRODUCT
5034
5035const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5036const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
5037                               (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5038
5039
5040// JVMTI & JVM monitoring and management support
5041// The thread_cpu_time() and current_thread_cpu_time() are only
5042// supported if is_thread_cpu_time_supported() returns true.
5043// They are not supported on Solaris T1.
5044
5045// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5046// are used by JVM M&M and JVMTI to get user+sys or user CPU time
5047// of a thread.
5048//
5049// current_thread_cpu_time() and thread_cpu_time(Thread *)
5050// returns the fast estimate available on the platform.
5051
5052// hrtime_t gethrvtime() return value includes
5053// user time but does not include system time
5054jlong os::current_thread_cpu_time() {
5055  return (jlong) gethrvtime();
5056}
5057
5058jlong os::thread_cpu_time(Thread *thread) {
5059  // return user level CPU time only to be consistent with
5060  // what current_thread_cpu_time returns.
5061  // thread_cpu_time_info() must be changed if this changes
5062  return os::thread_cpu_time(thread, false /* user time only */);
5063}
5064
5065jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5066  if (user_sys_cpu_time) {
5067    return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5068  } else {
5069    return os::current_thread_cpu_time();
5070  }
5071}
5072
5073jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5074  char proc_name[64];
5075  int count;
5076  prusage_t prusage;
5077  jlong lwp_time;
5078  int fd;
5079
5080  sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5081          getpid(),
5082          thread->osthread()->lwp_id());
5083  fd = ::open(proc_name, O_RDONLY);
5084  if (fd == -1) return -1;
5085
5086  do {
5087    count = ::pread(fd,
5088                    (void *)&prusage.pr_utime,
5089                    thr_time_size,
5090                    thr_time_off);
5091  } while (count < 0 && errno == EINTR);
5092  ::close(fd);
5093  if (count < 0) return -1;
5094
5095  if (user_sys_cpu_time) {
5096    // user + system CPU time
5097    lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5098                 (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5099                 (jlong)prusage.pr_stime.tv_nsec +
5100                 (jlong)prusage.pr_utime.tv_nsec;
5101  } else {
5102    // user level CPU time only
5103    lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5104                (jlong)prusage.pr_utime.tv_nsec;
5105  }
5106
5107  return (lwp_time);
5108}
5109
5110void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5111  info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5112  info_ptr->may_skip_backward = false;    // elapsed time not wall time
5113  info_ptr->may_skip_forward = false;     // elapsed time not wall time
5114  info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5115}
5116
5117void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5118  info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5119  info_ptr->may_skip_backward = false;    // elapsed time not wall time
5120  info_ptr->may_skip_forward = false;     // elapsed time not wall time
5121  info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5122}
5123
5124bool os::is_thread_cpu_time_supported() {
5125  return true;
5126}
5127
5128// System loadavg support.  Returns -1 if load average cannot be obtained.
5129// Return the load average for our processor set if the primitive exists
5130// (Solaris 9 and later).  Otherwise just return system wide loadavg.
5131int os::loadavg(double loadavg[], int nelem) {
5132  if (pset_getloadavg_ptr != NULL) {
5133    return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5134  } else {
5135    return ::getloadavg(loadavg, nelem);
5136  }
5137}
5138
5139//---------------------------------------------------------------------------------
5140
5141bool os::find(address addr, outputStream* st) {
5142  Dl_info dlinfo;
5143  memset(&dlinfo, 0, sizeof(dlinfo));
5144  if (dladdr(addr, &dlinfo) != 0) {
5145    st->print(PTR_FORMAT ": ", addr);
5146    if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5147      st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5148    } else if (dlinfo.dli_fbase != NULL) {
5149      st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5150    } else {
5151      st->print("<absolute address>");
5152    }
5153    if (dlinfo.dli_fname != NULL) {
5154      st->print(" in %s", dlinfo.dli_fname);
5155    }
5156    if (dlinfo.dli_fbase != NULL) {
5157      st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5158    }
5159    st->cr();
5160
5161    if (Verbose) {
5162      // decode some bytes around the PC
5163      address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5164      address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5165      address       lowest = (address) dlinfo.dli_sname;
5166      if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5167      if (begin < lowest)  begin = lowest;
5168      Dl_info dlinfo2;
5169      if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5170          && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5171        end = (address) dlinfo2.dli_saddr;
5172      }
5173      Disassembler::decode(begin, end, st);
5174    }
5175    return true;
5176  }
5177  return false;
5178}
5179
5180// Following function has been added to support HotSparc's libjvm.so running
5181// under Solaris production JDK 1.2.2 / 1.3.0.  These came from
5182// src/solaris/hpi/native_threads in the EVM codebase.
5183//
5184// NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5185// libraries and should thus be removed. We will leave it behind for a while
5186// until we no longer want to able to run on top of 1.3.0 Solaris production
5187// JDK. See 4341971.
5188
5189#define STACK_SLACK 0x800
5190
5191extern "C" {
5192  intptr_t sysThreadAvailableStackWithSlack() {
5193    stack_t st;
5194    intptr_t retval, stack_top;
5195    retval = thr_stksegment(&st);
5196    assert(retval == 0, "incorrect return value from thr_stksegment");
5197    assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5198    assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5199    stack_top=(intptr_t)st.ss_sp-st.ss_size;
5200    return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5201  }
5202}
5203
5204// ObjectMonitor park-unpark infrastructure ...
5205//
5206// We implement Solaris and Linux PlatformEvents with the
5207// obvious condvar-mutex-flag triple.
5208// Another alternative that works quite well is pipes:
5209// Each PlatformEvent consists of a pipe-pair.
5210// The thread associated with the PlatformEvent
5211// calls park(), which reads from the input end of the pipe.
5212// Unpark() writes into the other end of the pipe.
5213// The write-side of the pipe must be set NDELAY.
5214// Unfortunately pipes consume a large # of handles.
5215// Native solaris lwp_park() and lwp_unpark() work nicely, too.
5216// Using pipes for the 1st few threads might be workable, however.
5217//
5218// park() is permitted to return spuriously.
5219// Callers of park() should wrap the call to park() in
5220// an appropriate loop.  A litmus test for the correct
5221// usage of park is the following: if park() were modified
5222// to immediately return 0 your code should still work,
5223// albeit degenerating to a spin loop.
5224//
5225// In a sense, park()-unpark() just provides more polite spinning
5226// and polling with the key difference over naive spinning being
5227// that a parked thread needs to be explicitly unparked() in order
5228// to wake up and to poll the underlying condition.
5229//
5230// Assumption:
5231//    Only one parker can exist on an event, which is why we allocate
5232//    them per-thread. Multiple unparkers can coexist.
5233//
5234// _Event transitions in park()
5235//   -1 => -1 : illegal
5236//    1 =>  0 : pass - return immediately
5237//    0 => -1 : block; then set _Event to 0 before returning
5238//
5239// _Event transitions in unpark()
5240//    0 => 1 : just return
5241//    1 => 1 : just return
5242//   -1 => either 0 or 1; must signal target thread
5243//         That is, we can safely transition _Event from -1 to either
5244//         0 or 1.
5245//
5246// _Event serves as a restricted-range semaphore.
5247//   -1 : thread is blocked, i.e. there is a waiter
5248//    0 : neutral: thread is running or ready,
5249//        could have been signaled after a wait started
5250//    1 : signaled - thread is running or ready
5251//
5252// Another possible encoding of _Event would be with
5253// explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5254//
5255// TODO-FIXME: add DTRACE probes for:
5256// 1.   Tx parks
5257// 2.   Ty unparks Tx
5258// 3.   Tx resumes from park
5259
5260
5261// value determined through experimentation
5262#define ROUNDINGFIX 11
5263
5264// utility to compute the abstime argument to timedwait.
5265// TODO-FIXME: switch from compute_abstime() to unpackTime().
5266
5267static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5268  // millis is the relative timeout time
5269  // abstime will be the absolute timeout time
5270  if (millis < 0)  millis = 0;
5271  struct timeval now;
5272  int status = gettimeofday(&now, NULL);
5273  assert(status == 0, "gettimeofday");
5274  jlong seconds = millis / 1000;
5275  jlong max_wait_period;
5276
5277  if (UseLWPSynchronization) {
5278    // forward port of fix for 4275818 (not sleeping long enough)
5279    // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5280    // _lwp_cond_timedwait() used a round_down algorithm rather
5281    // than a round_up. For millis less than our roundfactor
5282    // it rounded down to 0 which doesn't meet the spec.
5283    // For millis > roundfactor we may return a bit sooner, but
5284    // since we can not accurately identify the patch level and
5285    // this has already been fixed in Solaris 9 and 8 we will
5286    // leave it alone rather than always rounding down.
5287
5288    if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5289    // It appears that when we go directly through Solaris _lwp_cond_timedwait()
5290    // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
5291    max_wait_period = 21000000;
5292  } else {
5293    max_wait_period = 50000000;
5294  }
5295  millis %= 1000;
5296  if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
5297    seconds = max_wait_period;
5298  }
5299  abstime->tv_sec = now.tv_sec  + seconds;
5300  long       usec = now.tv_usec + millis * 1000;
5301  if (usec >= 1000000) {
5302    abstime->tv_sec += 1;
5303    usec -= 1000000;
5304  }
5305  abstime->tv_nsec = usec * 1000;
5306  return abstime;
5307}
5308
5309void os::PlatformEvent::park() {           // AKA: down()
5310  // Transitions for _Event:
5311  //   -1 => -1 : illegal
5312  //    1 =>  0 : pass - return immediately
5313  //    0 => -1 : block; then set _Event to 0 before returning
5314
5315  // Invariant: Only the thread associated with the Event/PlatformEvent
5316  // may call park().
5317  assert(_nParked == 0, "invariant");
5318
5319  int v;
5320  for (;;) {
5321    v = _Event;
5322    if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5323  }
5324  guarantee(v >= 0, "invariant");
5325  if (v == 0) {
5326    // Do this the hard way by blocking ...
5327    // See http://monaco.sfbay/detail.jsf?cr=5094058.
5328    // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5329    // Only for SPARC >= V8PlusA
5330#if defined(__sparc) && defined(COMPILER2)
5331    if (ClearFPUAtPark) { _mark_fpu_nosave(); }
5332#endif
5333    int status = os::Solaris::mutex_lock(_mutex);
5334    assert_status(status == 0, status, "mutex_lock");
5335    guarantee(_nParked == 0, "invariant");
5336    ++_nParked;
5337    while (_Event < 0) {
5338      // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5339      // Treat this the same as if the wait was interrupted
5340      // With usr/lib/lwp going to kernel, always handle ETIME
5341      status = os::Solaris::cond_wait(_cond, _mutex);
5342      if (status == ETIME) status = EINTR;
5343      assert_status(status == 0 || status == EINTR, status, "cond_wait");
5344    }
5345    --_nParked;
5346    _Event = 0;
5347    status = os::Solaris::mutex_unlock(_mutex);
5348    assert_status(status == 0, status, "mutex_unlock");
5349    // Paranoia to ensure our locked and lock-free paths interact
5350    // correctly with each other.
5351    OrderAccess::fence();
5352  }
5353}
5354
5355int os::PlatformEvent::park(jlong millis) {
5356  // Transitions for _Event:
5357  //   -1 => -1 : illegal
5358  //    1 =>  0 : pass - return immediately
5359  //    0 => -1 : block; then set _Event to 0 before returning
5360
5361  guarantee(_nParked == 0, "invariant");
5362  int v;
5363  for (;;) {
5364    v = _Event;
5365    if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5366  }
5367  guarantee(v >= 0, "invariant");
5368  if (v != 0) return OS_OK;
5369
5370  int ret = OS_TIMEOUT;
5371  timestruc_t abst;
5372  compute_abstime(&abst, millis);
5373
5374  // See http://monaco.sfbay/detail.jsf?cr=5094058.
5375  // For Solaris SPARC set fprs.FEF=0 prior to parking.
5376  // Only for SPARC >= V8PlusA
5377#if defined(__sparc) && defined(COMPILER2)
5378  if (ClearFPUAtPark) { _mark_fpu_nosave(); }
5379#endif
5380  int status = os::Solaris::mutex_lock(_mutex);
5381  assert_status(status == 0, status, "mutex_lock");
5382  guarantee(_nParked == 0, "invariant");
5383  ++_nParked;
5384  while (_Event < 0) {
5385    int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5386    assert_status(status == 0 || status == EINTR ||
5387                  status == ETIME || status == ETIMEDOUT,
5388                  status, "cond_timedwait");
5389    if (!FilterSpuriousWakeups) break;                // previous semantics
5390    if (status == ETIME || status == ETIMEDOUT) break;
5391    // We consume and ignore EINTR and spurious wakeups.
5392  }
5393  --_nParked;
5394  if (_Event >= 0) ret = OS_OK;
5395  _Event = 0;
5396  status = os::Solaris::mutex_unlock(_mutex);
5397  assert_status(status == 0, status, "mutex_unlock");
5398  // Paranoia to ensure our locked and lock-free paths interact
5399  // correctly with each other.
5400  OrderAccess::fence();
5401  return ret;
5402}
5403
5404void os::PlatformEvent::unpark() {
5405  // Transitions for _Event:
5406  //    0 => 1 : just return
5407  //    1 => 1 : just return
5408  //   -1 => either 0 or 1; must signal target thread
5409  //         That is, we can safely transition _Event from -1 to either
5410  //         0 or 1.
5411  // See also: "Semaphores in Plan 9" by Mullender & Cox
5412  //
5413  // Note: Forcing a transition from "-1" to "1" on an unpark() means
5414  // that it will take two back-to-back park() calls for the owning
5415  // thread to block. This has the benefit of forcing a spurious return
5416  // from the first park() call after an unpark() call which will help
5417  // shake out uses of park() and unpark() without condition variables.
5418
5419  if (Atomic::xchg(1, &_Event) >= 0) return;
5420
5421  // If the thread associated with the event was parked, wake it.
5422  // Wait for the thread assoc with the PlatformEvent to vacate.
5423  int status = os::Solaris::mutex_lock(_mutex);
5424  assert_status(status == 0, status, "mutex_lock");
5425  int AnyWaiters = _nParked;
5426  status = os::Solaris::mutex_unlock(_mutex);
5427  assert_status(status == 0, status, "mutex_unlock");
5428  guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5429  if (AnyWaiters != 0) {
5430    // Note that we signal() *after* dropping the lock for "immortal" Events.
5431    // This is safe and avoids a common class of  futile wakeups.  In rare
5432    // circumstances this can cause a thread to return prematurely from
5433    // cond_{timed}wait() but the spurious wakeup is benign and the victim
5434    // will simply re-test the condition and re-park itself.
5435    // This provides particular benefit if the underlying platform does not
5436    // provide wait morphing.
5437    status = os::Solaris::cond_signal(_cond);
5438    assert_status(status == 0, status, "cond_signal");
5439  }
5440}
5441
5442// JSR166
5443// -------------------------------------------------------
5444
5445// The solaris and linux implementations of park/unpark are fairly
5446// conservative for now, but can be improved. They currently use a
5447// mutex/condvar pair, plus _counter.
5448// Park decrements _counter if > 0, else does a condvar wait.  Unpark
5449// sets count to 1 and signals condvar.  Only one thread ever waits
5450// on the condvar. Contention seen when trying to park implies that someone
5451// is unparking you, so don't wait. And spurious returns are fine, so there
5452// is no need to track notifications.
5453
5454#define MAX_SECS 100000000
5455
5456// This code is common to linux and solaris and will be moved to a
5457// common place in dolphin.
5458//
5459// The passed in time value is either a relative time in nanoseconds
5460// or an absolute time in milliseconds. Either way it has to be unpacked
5461// into suitable seconds and nanoseconds components and stored in the
5462// given timespec structure.
5463// Given time is a 64-bit value and the time_t used in the timespec is only
5464// a signed-32-bit value (except on 64-bit Linux) we have to watch for
5465// overflow if times way in the future are given. Further on Solaris versions
5466// prior to 10 there is a restriction (see cond_timedwait) that the specified
5467// number of seconds, in abstime, is less than current_time  + 100,000,000.
5468// As it will be 28 years before "now + 100000000" will overflow we can
5469// ignore overflow and just impose a hard-limit on seconds using the value
5470// of "now + 100,000,000". This places a limit on the timeout of about 3.17
5471// years from "now".
5472//
5473static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5474  assert(time > 0, "convertTime");
5475
5476  struct timeval now;
5477  int status = gettimeofday(&now, NULL);
5478  assert(status == 0, "gettimeofday");
5479
5480  time_t max_secs = now.tv_sec + MAX_SECS;
5481
5482  if (isAbsolute) {
5483    jlong secs = time / 1000;
5484    if (secs > max_secs) {
5485      absTime->tv_sec = max_secs;
5486    } else {
5487      absTime->tv_sec = secs;
5488    }
5489    absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5490  } else {
5491    jlong secs = time / NANOSECS_PER_SEC;
5492    if (secs >= MAX_SECS) {
5493      absTime->tv_sec = max_secs;
5494      absTime->tv_nsec = 0;
5495    } else {
5496      absTime->tv_sec = now.tv_sec + secs;
5497      absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5498      if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5499        absTime->tv_nsec -= NANOSECS_PER_SEC;
5500        ++absTime->tv_sec; // note: this must be <= max_secs
5501      }
5502    }
5503  }
5504  assert(absTime->tv_sec >= 0, "tv_sec < 0");
5505  assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5506  assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5507  assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5508}
5509
5510void Parker::park(bool isAbsolute, jlong time) {
5511  // Ideally we'd do something useful while spinning, such
5512  // as calling unpackTime().
5513
5514  // Optional fast-path check:
5515  // Return immediately if a permit is available.
5516  // We depend on Atomic::xchg() having full barrier semantics
5517  // since we are doing a lock-free update to _counter.
5518  if (Atomic::xchg(0, &_counter) > 0) return;
5519
5520  // Optional fast-exit: Check interrupt before trying to wait
5521  Thread* thread = Thread::current();
5522  assert(thread->is_Java_thread(), "Must be JavaThread");
5523  JavaThread *jt = (JavaThread *)thread;
5524  if (Thread::is_interrupted(thread, false)) {
5525    return;
5526  }
5527
5528  // First, demultiplex/decode time arguments
5529  timespec absTime;
5530  if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5531    return;
5532  }
5533  if (time > 0) {
5534    // Warning: this code might be exposed to the old Solaris time
5535    // round-down bugs.  Grep "roundingFix" for details.
5536    unpackTime(&absTime, isAbsolute, time);
5537  }
5538
5539  // Enter safepoint region
5540  // Beware of deadlocks such as 6317397.
5541  // The per-thread Parker:: _mutex is a classic leaf-lock.
5542  // In particular a thread must never block on the Threads_lock while
5543  // holding the Parker:: mutex.  If safepoints are pending both the
5544  // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5545  ThreadBlockInVM tbivm(jt);
5546
5547  // Don't wait if cannot get lock since interference arises from
5548  // unblocking.  Also. check interrupt before trying wait
5549  if (Thread::is_interrupted(thread, false) ||
5550      os::Solaris::mutex_trylock(_mutex) != 0) {
5551    return;
5552  }
5553
5554  int status;
5555
5556  if (_counter > 0)  { // no wait needed
5557    _counter = 0;
5558    status = os::Solaris::mutex_unlock(_mutex);
5559    assert(status == 0, "invariant");
5560    // Paranoia to ensure our locked and lock-free paths interact
5561    // correctly with each other and Java-level accesses.
5562    OrderAccess::fence();
5563    return;
5564  }
5565
5566#ifdef ASSERT
5567  // Don't catch signals while blocked; let the running threads have the signals.
5568  // (This allows a debugger to break into the running thread.)
5569  sigset_t oldsigs;
5570  sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
5571  pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5572#endif
5573
5574  OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5575  jt->set_suspend_equivalent();
5576  // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5577
5578  // Do this the hard way by blocking ...
5579  // See http://monaco.sfbay/detail.jsf?cr=5094058.
5580  // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5581  // Only for SPARC >= V8PlusA
5582#if defined(__sparc) && defined(COMPILER2)
5583  if (ClearFPUAtPark) { _mark_fpu_nosave(); }
5584#endif
5585
5586  if (time == 0) {
5587    status = os::Solaris::cond_wait(_cond, _mutex);
5588  } else {
5589    status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5590  }
5591  // Note that an untimed cond_wait() can sometimes return ETIME on older
5592  // versions of the Solaris.
5593  assert_status(status == 0 || status == EINTR ||
5594                status == ETIME || status == ETIMEDOUT,
5595                status, "cond_timedwait");
5596
5597#ifdef ASSERT
5598  pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5599#endif
5600  _counter = 0;
5601  status = os::Solaris::mutex_unlock(_mutex);
5602  assert_status(status == 0, status, "mutex_unlock");
5603  // Paranoia to ensure our locked and lock-free paths interact
5604  // correctly with each other and Java-level accesses.
5605  OrderAccess::fence();
5606
5607  // If externally suspended while waiting, re-suspend
5608  if (jt->handle_special_suspend_equivalent_condition()) {
5609    jt->java_suspend_self();
5610  }
5611}
5612
5613void Parker::unpark() {
5614  int status = os::Solaris::mutex_lock(_mutex);
5615  assert(status == 0, "invariant");
5616  const int s = _counter;
5617  _counter = 1;
5618  status = os::Solaris::mutex_unlock(_mutex);
5619  assert(status == 0, "invariant");
5620
5621  if (s < 1) {
5622    status = os::Solaris::cond_signal(_cond);
5623    assert(status == 0, "invariant");
5624  }
5625}
5626
5627extern char** environ;
5628
5629// Run the specified command in a separate process. Return its exit value,
5630// or -1 on failure (e.g. can't fork a new process).
5631// Unlike system(), this function can be called from signal handler. It
5632// doesn't block SIGINT et al.
5633int os::fork_and_exec(char* cmd) {
5634  char * argv[4];
5635  argv[0] = (char *)"sh";
5636  argv[1] = (char *)"-c";
5637  argv[2] = cmd;
5638  argv[3] = NULL;
5639
5640  // fork is async-safe, fork1 is not so can't use in signal handler
5641  pid_t pid;
5642  Thread* t = Thread::current_or_null_safe();
5643  if (t != NULL && t->is_inside_signal_handler()) {
5644    pid = fork();
5645  } else {
5646    pid = fork1();
5647  }
5648
5649  if (pid < 0) {
5650    // fork failed
5651    warning("fork failed: %s", os::strerror(errno));
5652    return -1;
5653
5654  } else if (pid == 0) {
5655    // child process
5656
5657    // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5658    execve("/usr/bin/sh", argv, environ);
5659
5660    // execve failed
5661    _exit(-1);
5662
5663  } else  {
5664    // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5665    // care about the actual exit code, for now.
5666
5667    int status;
5668
5669    // Wait for the child process to exit.  This returns immediately if
5670    // the child has already exited. */
5671    while (waitpid(pid, &status, 0) < 0) {
5672      switch (errno) {
5673      case ECHILD: return 0;
5674      case EINTR: break;
5675      default: return -1;
5676      }
5677    }
5678
5679    if (WIFEXITED(status)) {
5680      // The child exited normally; get its exit code.
5681      return WEXITSTATUS(status);
5682    } else if (WIFSIGNALED(status)) {
5683      // The child exited because of a signal
5684      // The best value to return is 0x80 + signal number,
5685      // because that is what all Unix shells do, and because
5686      // it allows callers to distinguish between process exit and
5687      // process death by signal.
5688      return 0x80 + WTERMSIG(status);
5689    } else {
5690      // Unknown exit code; pass it through
5691      return status;
5692    }
5693  }
5694}
5695
5696// is_headless_jre()
5697//
5698// Test for the existence of xawt/libmawt.so or libawt_xawt.so
5699// in order to report if we are running in a headless jre
5700//
5701// Since JDK8 xawt/libmawt.so was moved into the same directory
5702// as libawt.so, and renamed libawt_xawt.so
5703//
5704bool os::is_headless_jre() {
5705  struct stat statbuf;
5706  char buf[MAXPATHLEN];
5707  char libmawtpath[MAXPATHLEN];
5708  const char *xawtstr  = "/xawt/libmawt.so";
5709  const char *new_xawtstr = "/libawt_xawt.so";
5710  char *p;
5711
5712  // Get path to libjvm.so
5713  os::jvm_path(buf, sizeof(buf));
5714
5715  // Get rid of libjvm.so
5716  p = strrchr(buf, '/');
5717  if (p == NULL) {
5718    return false;
5719  } else {
5720    *p = '\0';
5721  }
5722
5723  // Get rid of client or server
5724  p = strrchr(buf, '/');
5725  if (p == NULL) {
5726    return false;
5727  } else {
5728    *p = '\0';
5729  }
5730
5731  // check xawt/libmawt.so
5732  strcpy(libmawtpath, buf);
5733  strcat(libmawtpath, xawtstr);
5734  if (::stat(libmawtpath, &statbuf) == 0) return false;
5735
5736  // check libawt_xawt.so
5737  strcpy(libmawtpath, buf);
5738  strcat(libmawtpath, new_xawtstr);
5739  if (::stat(libmawtpath, &statbuf) == 0) return false;
5740
5741  return true;
5742}
5743
5744size_t os::write(int fd, const void *buf, unsigned int nBytes) {
5745  size_t res;
5746  RESTARTABLE((size_t) ::write(fd, buf, (size_t) nBytes), res);
5747  return res;
5748}
5749
5750int os::close(int fd) {
5751  return ::close(fd);
5752}
5753
5754int os::socket_close(int fd) {
5755  return ::close(fd);
5756}
5757
5758int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
5759  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5760         "Assumed _thread_in_native");
5761  RESTARTABLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags));
5762}
5763
5764int os::send(int fd, char* buf, size_t nBytes, uint flags) {
5765  assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5766         "Assumed _thread_in_native");
5767  RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5768}
5769
5770int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
5771  RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5772}
5773
5774// As both poll and select can be interrupted by signals, we have to be
5775// prepared to restart the system call after updating the timeout, unless
5776// a poll() is done with timeout == -1, in which case we repeat with this
5777// "wait forever" value.
5778
5779int os::connect(int fd, struct sockaddr *him, socklen_t len) {
5780  int _result;
5781  _result = ::connect(fd, him, len);
5782
5783  // On Solaris, when a connect() call is interrupted, the connection
5784  // can be established asynchronously (see 6343810). Subsequent calls
5785  // to connect() must check the errno value which has the semantic
5786  // described below (copied from the connect() man page). Handling
5787  // of asynchronously established connections is required for both
5788  // blocking and non-blocking sockets.
5789  //     EINTR            The  connection  attempt  was   interrupted
5790  //                      before  any data arrived by the delivery of
5791  //                      a signal. The connection, however, will  be
5792  //                      established asynchronously.
5793  //
5794  //     EINPROGRESS      The socket is non-blocking, and the connec-
5795  //                      tion  cannot  be completed immediately.
5796  //
5797  //     EALREADY         The socket is non-blocking,  and a previous
5798  //                      connection  attempt  has  not yet been com-
5799  //                      pleted.
5800  //
5801  //     EISCONN          The socket is already connected.
5802  if (_result == OS_ERR && errno == EINTR) {
5803    // restarting a connect() changes its errno semantics
5804    RESTARTABLE(::connect(fd, him, len), _result);
5805    // undo these changes
5806    if (_result == OS_ERR) {
5807      if (errno == EALREADY) {
5808        errno = EINPROGRESS; // fall through
5809      } else if (errno == EISCONN) {
5810        errno = 0;
5811        return OS_OK;
5812      }
5813    }
5814  }
5815  return _result;
5816}
5817
5818// Get the default path to the core file
5819// Returns the length of the string
5820int os::get_core_path(char* buffer, size_t bufferSize) {
5821  const char* p = get_current_directory(buffer, bufferSize);
5822
5823  if (p == NULL) {
5824    assert(p != NULL, "failed to get current directory");
5825    return 0;
5826  }
5827
5828  jio_snprintf(buffer, bufferSize, "%s/core or core.%d",
5829                                              p, current_process_id());
5830
5831  return strlen(buffer);
5832}
5833
5834#ifndef PRODUCT
5835void TestReserveMemorySpecial_test() {
5836  // No tests available for this platform
5837}
5838#endif
5839
5840bool os::start_debugging(char *buf, int buflen) {
5841  int len = (int)strlen(buf);
5842  char *p = &buf[len];
5843
5844  jio_snprintf(p, buflen-len,
5845               "\n\n"
5846               "Do you want to debug the problem?\n\n"
5847               "To debug, run 'dbx - %d'; then switch to thread " INTX_FORMAT "\n"
5848               "Enter 'yes' to launch dbx automatically (PATH must include dbx)\n"
5849               "Otherwise, press RETURN to abort...",
5850               os::current_process_id(), os::current_thread_id());
5851
5852  bool yes = os::message_box("Unexpected Error", buf);
5853
5854  if (yes) {
5855    // yes, user asked VM to launch debugger
5856    jio_snprintf(buf, sizeof(buf), "dbx - %d", os::current_process_id());
5857
5858    os::fork_and_exec(buf);
5859    yes = false;
5860  }
5861  return yes;
5862}
5863