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