Tue, 26 Apr 2011 11:46:34 -0700
7037939: NUMA: Disable adaptive resizing if SHM large pages are used
Summary: Make the NUMA allocator behave properly with SHM and ISM large pages.
Reviewed-by: ysr
1 /*
2 * Copyright (c) 1997, 2011, 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 */
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 "interpreter/interpreter.hpp"
33 #include "jvm_solaris.h"
34 #include "memory/allocation.inline.hpp"
35 #include "memory/filemap.hpp"
36 #include "mutex_solaris.inline.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "os_share_solaris.hpp"
39 #include "prims/jniFastGetField.hpp"
40 #include "prims/jvm.h"
41 #include "prims/jvm_misc.hpp"
42 #include "runtime/arguments.hpp"
43 #include "runtime/extendedPC.hpp"
44 #include "runtime/globals.hpp"
45 #include "runtime/interfaceSupport.hpp"
46 #include "runtime/java.hpp"
47 #include "runtime/javaCalls.hpp"
48 #include "runtime/mutexLocker.hpp"
49 #include "runtime/objectMonitor.hpp"
50 #include "runtime/osThread.hpp"
51 #include "runtime/perfMemory.hpp"
52 #include "runtime/sharedRuntime.hpp"
53 #include "runtime/statSampler.hpp"
54 #include "runtime/stubRoutines.hpp"
55 #include "runtime/threadCritical.hpp"
56 #include "runtime/timer.hpp"
57 #include "services/attachListener.hpp"
58 #include "services/runtimeService.hpp"
59 #include "thread_solaris.inline.hpp"
60 #include "utilities/decoder.hpp"
61 #include "utilities/defaultStream.hpp"
62 #include "utilities/events.hpp"
63 #include "utilities/growableArray.hpp"
64 #include "utilities/vmError.hpp"
65 #ifdef TARGET_ARCH_x86
66 # include "assembler_x86.inline.hpp"
67 # include "nativeInst_x86.hpp"
68 #endif
69 #ifdef TARGET_ARCH_sparc
70 # include "assembler_sparc.inline.hpp"
71 # include "nativeInst_sparc.hpp"
72 #endif
73 #ifdef COMPILER1
74 #include "c1/c1_Runtime1.hpp"
75 #endif
76 #ifdef COMPILER2
77 #include "opto/runtime.hpp"
78 #endif
80 // put OS-includes here
81 # include <dlfcn.h>
82 # include <errno.h>
83 # include <exception>
84 # include <link.h>
85 # include <poll.h>
86 # include <pthread.h>
87 # include <pwd.h>
88 # include <schedctl.h>
89 # include <setjmp.h>
90 # include <signal.h>
91 # include <stdio.h>
92 # include <alloca.h>
93 # include <sys/filio.h>
94 # include <sys/ipc.h>
95 # include <sys/lwp.h>
96 # include <sys/machelf.h> // for elf Sym structure used by dladdr1
97 # include <sys/mman.h>
98 # include <sys/processor.h>
99 # include <sys/procset.h>
100 # include <sys/pset.h>
101 # include <sys/resource.h>
102 # include <sys/shm.h>
103 # include <sys/socket.h>
104 # include <sys/stat.h>
105 # include <sys/systeminfo.h>
106 # include <sys/time.h>
107 # include <sys/times.h>
108 # include <sys/types.h>
109 # include <sys/wait.h>
110 # include <sys/utsname.h>
111 # include <thread.h>
112 # include <unistd.h>
113 # include <sys/priocntl.h>
114 # include <sys/rtpriocntl.h>
115 # include <sys/tspriocntl.h>
116 # include <sys/iapriocntl.h>
117 # include <sys/loadavg.h>
118 # include <string.h>
119 # include <stdio.h>
121 # define _STRUCTURED_PROC 1 // this gets us the new structured proc interfaces of 5.6 & later
122 # include <sys/procfs.h> // see comment in <sys/procfs.h>
124 #define MAX_PATH (2 * K)
126 // for timer info max values which include all bits
127 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
129 #ifdef _GNU_SOURCE
130 // See bug #6514594
131 extern "C" int madvise(caddr_t, size_t, int);
132 extern "C" int memcntl(caddr_t addr, size_t len, int cmd, caddr_t arg,
133 int attr, int mask);
134 #endif //_GNU_SOURCE
136 /*
137 MPSS Changes Start.
138 The JVM binary needs to be built and run on pre-Solaris 9
139 systems, but the constants needed by MPSS are only in Solaris 9
140 header files. They are textually replicated here to allow
141 building on earlier systems. Once building on Solaris 8 is
142 no longer a requirement, these #defines can be replaced by ordinary
143 system .h inclusion.
145 In earlier versions of the JDK and Solaris, we used ISM for large pages.
146 But ISM requires shared memory to achieve this and thus has many caveats.
147 MPSS is a fully transparent and is a cleaner way to get large pages.
148 Although we still require keeping ISM for backward compatiblitiy as well as
149 giving the opportunity to use large pages on older systems it is
150 recommended that MPSS be used for Solaris 9 and above.
152 */
154 #ifndef MC_HAT_ADVISE
156 struct memcntl_mha {
157 uint_t mha_cmd; /* command(s) */
158 uint_t mha_flags;
159 size_t mha_pagesize;
160 };
161 #define MC_HAT_ADVISE 7 /* advise hat map size */
162 #define MHA_MAPSIZE_VA 0x1 /* set preferred page size */
163 #define MAP_ALIGN 0x200 /* addr specifies alignment */
165 #endif
166 // MPSS Changes End.
169 // Here are some liblgrp types from sys/lgrp_user.h to be able to
170 // compile on older systems without this header file.
172 #ifndef MADV_ACCESS_LWP
173 # define MADV_ACCESS_LWP 7 /* next LWP to access heavily */
174 #endif
175 #ifndef MADV_ACCESS_MANY
176 # define MADV_ACCESS_MANY 8 /* many processes to access heavily */
177 #endif
179 #ifndef LGRP_RSRC_CPU
180 # define LGRP_RSRC_CPU 0 /* CPU resources */
181 #endif
182 #ifndef LGRP_RSRC_MEM
183 # define LGRP_RSRC_MEM 1 /* memory resources */
184 #endif
186 // Some more macros from sys/mman.h that are not present in Solaris 8.
188 #ifndef MAX_MEMINFO_CNT
189 /*
190 * info_req request type definitions for meminfo
191 * request types starting with MEMINFO_V are used for Virtual addresses
192 * and should not be mixed with MEMINFO_PLGRP which is targeted for Physical
193 * addresses
194 */
195 # define MEMINFO_SHIFT 16
196 # define MEMINFO_MASK (0xFF << MEMINFO_SHIFT)
197 # define MEMINFO_VPHYSICAL (0x01 << MEMINFO_SHIFT) /* get physical addr */
198 # define MEMINFO_VLGRP (0x02 << MEMINFO_SHIFT) /* get lgroup */
199 # define MEMINFO_VPAGESIZE (0x03 << MEMINFO_SHIFT) /* size of phys page */
200 # define MEMINFO_VREPLCNT (0x04 << MEMINFO_SHIFT) /* no. of replica */
201 # define MEMINFO_VREPL (0x05 << MEMINFO_SHIFT) /* physical replica */
202 # define MEMINFO_VREPL_LGRP (0x06 << MEMINFO_SHIFT) /* lgrp of replica */
203 # define MEMINFO_PLGRP (0x07 << MEMINFO_SHIFT) /* lgroup for paddr */
205 /* maximum number of addresses meminfo() can process at a time */
206 # define MAX_MEMINFO_CNT 256
208 /* maximum number of request types */
209 # define MAX_MEMINFO_REQ 31
210 #endif
212 // see thr_setprio(3T) for the basis of these numbers
213 #define MinimumPriority 0
214 #define NormalPriority 64
215 #define MaximumPriority 127
217 // Values for ThreadPriorityPolicy == 1
218 int prio_policy1[MaxPriority+1] = { -99999, 0, 16, 32, 48, 64,
219 80, 96, 112, 124, 127 };
221 // System parameters used internally
222 static clock_t clock_tics_per_sec = 100;
224 // Track if we have called enable_extended_FILE_stdio (on Solaris 10u4+)
225 static bool enabled_extended_FILE_stdio = false;
227 // For diagnostics to print a message once. see run_periodic_checks
228 static bool check_addr0_done = false;
229 static sigset_t check_signal_done;
230 static bool check_signals = true;
232 address os::Solaris::handler_start; // start pc of thr_sighndlrinfo
233 address os::Solaris::handler_end; // end pc of thr_sighndlrinfo
235 address os::Solaris::_main_stack_base = NULL; // 4352906 workaround
238 // "default" initializers for missing libc APIs
239 extern "C" {
240 static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
241 static int lwp_mutex_destroy(mutex_t *mx) { return 0; }
243 static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
244 static int lwp_cond_destroy(cond_t *cv) { return 0; }
245 }
247 // "default" initializers for pthread-based synchronization
248 extern "C" {
249 static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
250 static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
251 }
253 // Thread Local Storage
254 // This is common to all Solaris platforms so it is defined here,
255 // in this common file.
256 // The declarations are in the os_cpu threadLS*.hpp files.
257 //
258 // Static member initialization for TLS
259 Thread* ThreadLocalStorage::_get_thread_cache[ThreadLocalStorage::_pd_cache_size] = {NULL};
261 #ifndef PRODUCT
262 #define _PCT(n,d) ((100.0*(double)(n))/(double)(d))
264 int ThreadLocalStorage::_tcacheHit = 0;
265 int ThreadLocalStorage::_tcacheMiss = 0;
267 void ThreadLocalStorage::print_statistics() {
268 int total = _tcacheMiss+_tcacheHit;
269 tty->print_cr("Thread cache hits %d misses %d total %d percent %f\n",
270 _tcacheHit, _tcacheMiss, total, _PCT(_tcacheHit, total));
271 }
272 #undef _PCT
273 #endif // PRODUCT
275 Thread* ThreadLocalStorage::get_thread_via_cache_slowly(uintptr_t raw_id,
276 int index) {
277 Thread *thread = get_thread_slow();
278 if (thread != NULL) {
279 address sp = os::current_stack_pointer();
280 guarantee(thread->_stack_base == NULL ||
281 (sp <= thread->_stack_base &&
282 sp >= thread->_stack_base - thread->_stack_size) ||
283 is_error_reported(),
284 "sp must be inside of selected thread stack");
286 thread->set_self_raw_id(raw_id); // mark for quick retrieval
287 _get_thread_cache[ index ] = thread;
288 }
289 return thread;
290 }
293 static const double all_zero[ sizeof(Thread) / sizeof(double) + 1 ] = {0};
294 #define NO_CACHED_THREAD ((Thread*)all_zero)
296 void ThreadLocalStorage::pd_set_thread(Thread* thread) {
298 // Store the new value before updating the cache to prevent a race
299 // between get_thread_via_cache_slowly() and this store operation.
300 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
302 // Update thread cache with new thread if setting on thread create,
303 // or NO_CACHED_THREAD (zeroed) thread if resetting thread on exit.
304 uintptr_t raw = pd_raw_thread_id();
305 int ix = pd_cache_index(raw);
306 _get_thread_cache[ix] = thread == NULL ? NO_CACHED_THREAD : thread;
307 }
309 void ThreadLocalStorage::pd_init() {
310 for (int i = 0; i < _pd_cache_size; i++) {
311 _get_thread_cache[i] = NO_CACHED_THREAD;
312 }
313 }
315 // Invalidate all the caches (happens to be the same as pd_init).
316 void ThreadLocalStorage::pd_invalidate_all() { pd_init(); }
318 #undef NO_CACHED_THREAD
320 // END Thread Local Storage
322 static inline size_t adjust_stack_size(address base, size_t size) {
323 if ((ssize_t)size < 0) {
324 // 4759953: Compensate for ridiculous stack size.
325 size = max_intx;
326 }
327 if (size > (size_t)base) {
328 // 4812466: Make sure size doesn't allow the stack to wrap the address space.
329 size = (size_t)base;
330 }
331 return size;
332 }
334 static inline stack_t get_stack_info() {
335 stack_t st;
336 int retval = thr_stksegment(&st);
337 st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
338 assert(retval == 0, "incorrect return value from thr_stksegment");
339 assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
340 assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
341 return st;
342 }
344 address os::current_stack_base() {
345 int r = thr_main() ;
346 guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
347 bool is_primordial_thread = r;
349 // Workaround 4352906, avoid calls to thr_stksegment by
350 // thr_main after the first one (it looks like we trash
351 // some data, causing the value for ss_sp to be incorrect).
352 if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
353 stack_t st = get_stack_info();
354 if (is_primordial_thread) {
355 // cache initial value of stack base
356 os::Solaris::_main_stack_base = (address)st.ss_sp;
357 }
358 return (address)st.ss_sp;
359 } else {
360 guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
361 return os::Solaris::_main_stack_base;
362 }
363 }
365 size_t os::current_stack_size() {
366 size_t size;
368 int r = thr_main() ;
369 guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
370 if(!r) {
371 size = get_stack_info().ss_size;
372 } else {
373 struct rlimit limits;
374 getrlimit(RLIMIT_STACK, &limits);
375 size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
376 }
377 // base may not be page aligned
378 address base = current_stack_base();
379 address bottom = (address)align_size_up((intptr_t)(base - size), os::vm_page_size());;
380 return (size_t)(base - bottom);
381 }
383 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
384 return localtime_r(clock, res);
385 }
387 // interruptible infrastructure
389 // setup_interruptible saves the thread state before going into an
390 // interruptible system call.
391 // The saved state is used to restore the thread to
392 // its former state whether or not an interrupt is received.
393 // Used by classloader os::read
394 // os::restartable_read calls skip this layer and stay in _thread_in_native
396 void os::Solaris::setup_interruptible(JavaThread* thread) {
398 JavaThreadState thread_state = thread->thread_state();
400 assert(thread_state != _thread_blocked, "Coming from the wrong thread");
401 assert(thread_state != _thread_in_native, "Native threads skip setup_interruptible");
402 OSThread* osthread = thread->osthread();
403 osthread->set_saved_interrupt_thread_state(thread_state);
404 thread->frame_anchor()->make_walkable(thread);
405 ThreadStateTransition::transition(thread, thread_state, _thread_blocked);
406 }
408 // Version of setup_interruptible() for threads that are already in
409 // _thread_blocked. Used by os_sleep().
410 void os::Solaris::setup_interruptible_already_blocked(JavaThread* thread) {
411 thread->frame_anchor()->make_walkable(thread);
412 }
414 JavaThread* os::Solaris::setup_interruptible() {
415 JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
416 setup_interruptible(thread);
417 return thread;
418 }
420 void os::Solaris::try_enable_extended_io() {
421 typedef int (*enable_extended_FILE_stdio_t)(int, int);
423 if (!UseExtendedFileIO) {
424 return;
425 }
427 enable_extended_FILE_stdio_t enabler =
428 (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
429 "enable_extended_FILE_stdio");
430 if (enabler) {
431 enabler(-1, -1);
432 }
433 }
436 #ifdef ASSERT
438 JavaThread* os::Solaris::setup_interruptible_native() {
439 JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
440 JavaThreadState thread_state = thread->thread_state();
441 assert(thread_state == _thread_in_native, "Assumed thread_in_native");
442 return thread;
443 }
445 void os::Solaris::cleanup_interruptible_native(JavaThread* thread) {
446 JavaThreadState thread_state = thread->thread_state();
447 assert(thread_state == _thread_in_native, "Assumed thread_in_native");
448 }
449 #endif
451 // cleanup_interruptible reverses the effects of setup_interruptible
452 // setup_interruptible_already_blocked() does not need any cleanup.
454 void os::Solaris::cleanup_interruptible(JavaThread* thread) {
455 OSThread* osthread = thread->osthread();
457 ThreadStateTransition::transition(thread, _thread_blocked, osthread->saved_interrupt_thread_state());
458 }
460 // I/O interruption related counters called in _INTERRUPTIBLE
462 void os::Solaris::bump_interrupted_before_count() {
463 RuntimeService::record_interrupted_before_count();
464 }
466 void os::Solaris::bump_interrupted_during_count() {
467 RuntimeService::record_interrupted_during_count();
468 }
470 static int _processors_online = 0;
472 jint os::Solaris::_os_thread_limit = 0;
473 volatile jint os::Solaris::_os_thread_count = 0;
475 julong os::available_memory() {
476 return Solaris::available_memory();
477 }
479 julong os::Solaris::available_memory() {
480 return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
481 }
483 julong os::Solaris::_physical_memory = 0;
485 julong os::physical_memory() {
486 return Solaris::physical_memory();
487 }
489 julong os::allocatable_physical_memory(julong size) {
490 #ifdef _LP64
491 return size;
492 #else
493 julong result = MIN2(size, (julong)3835*M);
494 if (!is_allocatable(result)) {
495 // Memory allocations will be aligned but the alignment
496 // is not known at this point. Alignments will
497 // be at most to LargePageSizeInBytes. Protect
498 // allocations from alignments up to illegal
499 // values. If at this point 2G is illegal.
500 julong reasonable_size = (julong)2*G - 2 * LargePageSizeInBytes;
501 result = MIN2(size, reasonable_size);
502 }
503 return result;
504 #endif
505 }
507 static hrtime_t first_hrtime = 0;
508 static const hrtime_t hrtime_hz = 1000*1000*1000;
509 const int LOCK_BUSY = 1;
510 const int LOCK_FREE = 0;
511 const int LOCK_INVALID = -1;
512 static volatile hrtime_t max_hrtime = 0;
513 static volatile int max_hrtime_lock = LOCK_FREE; // Update counter with LSB as lock-in-progress
516 void os::Solaris::initialize_system_info() {
517 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
518 _processors_online = sysconf (_SC_NPROCESSORS_ONLN);
519 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
520 }
522 int os::active_processor_count() {
523 int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
524 pid_t pid = getpid();
525 psetid_t pset = PS_NONE;
526 // Are we running in a processor set or is there any processor set around?
527 if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
528 uint_t pset_cpus;
529 // Query the number of cpus available to us.
530 if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
531 assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
532 _processors_online = pset_cpus;
533 return pset_cpus;
534 }
535 }
536 // Otherwise return number of online cpus
537 return online_cpus;
538 }
540 static bool find_processors_in_pset(psetid_t pset,
541 processorid_t** id_array,
542 uint_t* id_length) {
543 bool result = false;
544 // Find the number of processors in the processor set.
545 if (pset_info(pset, NULL, id_length, NULL) == 0) {
546 // Make up an array to hold their ids.
547 *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length);
548 // Fill in the array with their processor ids.
549 if (pset_info(pset, NULL, id_length, *id_array) == 0) {
550 result = true;
551 }
552 }
553 return result;
554 }
556 // Callers of find_processors_online() must tolerate imprecise results --
557 // the system configuration can change asynchronously because of DR
558 // or explicit psradm operations.
559 //
560 // We also need to take care that the loop (below) terminates as the
561 // number of processors online can change between the _SC_NPROCESSORS_ONLN
562 // request and the loop that builds the list of processor ids. Unfortunately
563 // there's no reliable way to determine the maximum valid processor id,
564 // so we use a manifest constant, MAX_PROCESSOR_ID, instead. See p_online
565 // man pages, which claim the processor id set is "sparse, but
566 // not too sparse". MAX_PROCESSOR_ID is used to ensure that we eventually
567 // exit the loop.
568 //
569 // In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
570 // not available on S8.0.
572 static bool find_processors_online(processorid_t** id_array,
573 uint* id_length) {
574 const processorid_t MAX_PROCESSOR_ID = 100000 ;
575 // Find the number of processors online.
576 *id_length = sysconf(_SC_NPROCESSORS_ONLN);
577 // Make up an array to hold their ids.
578 *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length);
579 // Processors need not be numbered consecutively.
580 long found = 0;
581 processorid_t next = 0;
582 while (found < *id_length && next < MAX_PROCESSOR_ID) {
583 processor_info_t info;
584 if (processor_info(next, &info) == 0) {
585 // NB, PI_NOINTR processors are effectively online ...
586 if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
587 (*id_array)[found] = next;
588 found += 1;
589 }
590 }
591 next += 1;
592 }
593 if (found < *id_length) {
594 // The loop above didn't identify the expected number of processors.
595 // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
596 // and re-running the loop, above, but there's no guarantee of progress
597 // if the system configuration is in flux. Instead, we just return what
598 // we've got. Note that in the worst case find_processors_online() could
599 // return an empty set. (As a fall-back in the case of the empty set we
600 // could just return the ID of the current processor).
601 *id_length = found ;
602 }
604 return true;
605 }
607 static bool assign_distribution(processorid_t* id_array,
608 uint id_length,
609 uint* distribution,
610 uint distribution_length) {
611 // We assume we can assign processorid_t's to uint's.
612 assert(sizeof(processorid_t) == sizeof(uint),
613 "can't convert processorid_t to uint");
614 // Quick check to see if we won't succeed.
615 if (id_length < distribution_length) {
616 return false;
617 }
618 // Assign processor ids to the distribution.
619 // Try to shuffle processors to distribute work across boards,
620 // assuming 4 processors per board.
621 const uint processors_per_board = ProcessDistributionStride;
622 // Find the maximum processor id.
623 processorid_t max_id = 0;
624 for (uint m = 0; m < id_length; m += 1) {
625 max_id = MAX2(max_id, id_array[m]);
626 }
627 // The next id, to limit loops.
628 const processorid_t limit_id = max_id + 1;
629 // Make up markers for available processors.
630 bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id);
631 for (uint c = 0; c < limit_id; c += 1) {
632 available_id[c] = false;
633 }
634 for (uint a = 0; a < id_length; a += 1) {
635 available_id[id_array[a]] = true;
636 }
637 // Step by "boards", then by "slot", copying to "assigned".
638 // NEEDS_CLEANUP: The assignment of processors should be stateful,
639 // remembering which processors have been assigned by
640 // previous calls, etc., so as to distribute several
641 // independent calls of this method. What we'd like is
642 // It would be nice to have an API that let us ask
643 // how many processes are bound to a processor,
644 // but we don't have that, either.
645 // In the short term, "board" is static so that
646 // subsequent distributions don't all start at board 0.
647 static uint board = 0;
648 uint assigned = 0;
649 // Until we've found enough processors ....
650 while (assigned < distribution_length) {
651 // ... find the next available processor in the board.
652 for (uint slot = 0; slot < processors_per_board; slot += 1) {
653 uint try_id = board * processors_per_board + slot;
654 if ((try_id < limit_id) && (available_id[try_id] == true)) {
655 distribution[assigned] = try_id;
656 available_id[try_id] = false;
657 assigned += 1;
658 break;
659 }
660 }
661 board += 1;
662 if (board * processors_per_board + 0 >= limit_id) {
663 board = 0;
664 }
665 }
666 if (available_id != NULL) {
667 FREE_C_HEAP_ARRAY(bool, available_id);
668 }
669 return true;
670 }
672 bool os::distribute_processes(uint length, uint* distribution) {
673 bool result = false;
674 // Find the processor id's of all the available CPUs.
675 processorid_t* id_array = NULL;
676 uint id_length = 0;
677 // There are some races between querying information and using it,
678 // since processor sets can change dynamically.
679 psetid_t pset = PS_NONE;
680 // Are we running in a processor set?
681 if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
682 result = find_processors_in_pset(pset, &id_array, &id_length);
683 } else {
684 result = find_processors_online(&id_array, &id_length);
685 }
686 if (result == true) {
687 if (id_length >= length) {
688 result = assign_distribution(id_array, id_length, distribution, length);
689 } else {
690 result = false;
691 }
692 }
693 if (id_array != NULL) {
694 FREE_C_HEAP_ARRAY(processorid_t, id_array);
695 }
696 return result;
697 }
699 bool os::bind_to_processor(uint processor_id) {
700 // We assume that a processorid_t can be stored in a uint.
701 assert(sizeof(uint) == sizeof(processorid_t),
702 "can't convert uint to processorid_t");
703 int bind_result =
704 processor_bind(P_LWPID, // bind LWP.
705 P_MYID, // bind current LWP.
706 (processorid_t) processor_id, // id.
707 NULL); // don't return old binding.
708 return (bind_result == 0);
709 }
711 bool os::getenv(const char* name, char* buffer, int len) {
712 char* val = ::getenv( name );
713 if ( val == NULL
714 || strlen(val) + 1 > len ) {
715 if (len > 0) buffer[0] = 0; // return a null string
716 return false;
717 }
718 strcpy( buffer, val );
719 return true;
720 }
723 // Return true if user is running as root.
725 bool os::have_special_privileges() {
726 static bool init = false;
727 static bool privileges = false;
728 if (!init) {
729 privileges = (getuid() != geteuid()) || (getgid() != getegid());
730 init = true;
731 }
732 return privileges;
733 }
736 void os::init_system_properties_values() {
737 char arch[12];
738 sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
740 // The next steps are taken in the product version:
741 //
742 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
743 // This library should be located at:
744 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
745 //
746 // If "/jre/lib/" appears at the right place in the path, then we
747 // assume libjvm[_g].so is installed in a JDK and we use this path.
748 //
749 // Otherwise exit with message: "Could not create the Java virtual machine."
750 //
751 // The following extra steps are taken in the debugging version:
752 //
753 // If "/jre/lib/" does NOT appear at the right place in the path
754 // instead of exit check for $JAVA_HOME environment variable.
755 //
756 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
757 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
758 // it looks like libjvm[_g].so is installed there
759 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
760 //
761 // Otherwise exit.
762 //
763 // Important note: if the location of libjvm.so changes this
764 // code needs to be changed accordingly.
766 // The next few definitions allow the code to be verbatim:
767 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
768 #define free(p) FREE_C_HEAP_ARRAY(char, p)
769 #define getenv(n) ::getenv(n)
771 #define EXTENSIONS_DIR "/lib/ext"
772 #define ENDORSED_DIR "/lib/endorsed"
773 #define COMMON_DIR "/usr/jdk/packages"
775 {
776 /* sysclasspath, java_home, dll_dir */
777 {
778 char *home_path;
779 char *dll_path;
780 char *pslash;
781 char buf[MAXPATHLEN];
782 os::jvm_path(buf, sizeof(buf));
784 // Found the full path to libjvm.so.
785 // Now cut the path to <java_home>/jre if we can.
786 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
787 pslash = strrchr(buf, '/');
788 if (pslash != NULL)
789 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
790 dll_path = malloc(strlen(buf) + 1);
791 if (dll_path == NULL)
792 return;
793 strcpy(dll_path, buf);
794 Arguments::set_dll_dir(dll_path);
796 if (pslash != NULL) {
797 pslash = strrchr(buf, '/');
798 if (pslash != NULL) {
799 *pslash = '\0'; /* get rid of /<arch> */
800 pslash = strrchr(buf, '/');
801 if (pslash != NULL)
802 *pslash = '\0'; /* get rid of /lib */
803 }
804 }
806 home_path = malloc(strlen(buf) + 1);
807 if (home_path == NULL)
808 return;
809 strcpy(home_path, buf);
810 Arguments::set_java_home(home_path);
812 if (!set_boot_path('/', ':'))
813 return;
814 }
816 /*
817 * Where to look for native libraries
818 */
819 {
820 // Use dlinfo() to determine the correct java.library.path.
821 //
822 // If we're launched by the Java launcher, and the user
823 // does not set java.library.path explicitly on the commandline,
824 // the Java launcher sets LD_LIBRARY_PATH for us and unsets
825 // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64. In this case
826 // dlinfo returns LD_LIBRARY_PATH + crle settings (including
827 // /usr/lib), which is exactly what we want.
828 //
829 // If the user does set java.library.path, it completely
830 // overwrites this setting, and always has.
831 //
832 // If we're not launched by the Java launcher, we may
833 // get here with any/all of the LD_LIBRARY_PATH[_32|64]
834 // settings. Again, dlinfo does exactly what we want.
836 Dl_serinfo _info, *info = &_info;
837 Dl_serpath *path;
838 char* library_path;
839 char *common_path;
840 int i;
842 // determine search path count and required buffer size
843 if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
844 vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
845 }
847 // allocate new buffer and initialize
848 info = (Dl_serinfo*)malloc(_info.dls_size);
849 if (info == NULL) {
850 vm_exit_out_of_memory(_info.dls_size,
851 "init_system_properties_values info");
852 }
853 info->dls_size = _info.dls_size;
854 info->dls_cnt = _info.dls_cnt;
856 // obtain search path information
857 if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
858 free(info);
859 vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
860 }
862 path = &info->dls_serpath[0];
864 // Note: Due to a legacy implementation, most of the library path
865 // is set in the launcher. This was to accomodate linking restrictions
866 // on legacy Solaris implementations (which are no longer supported).
867 // Eventually, all the library path setting will be done here.
868 //
869 // However, to prevent the proliferation of improperly built native
870 // libraries, the new path component /usr/jdk/packages is added here.
872 // Determine the actual CPU architecture.
873 char cpu_arch[12];
874 sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
875 #ifdef _LP64
876 // If we are a 64-bit vm, perform the following translations:
877 // sparc -> sparcv9
878 // i386 -> amd64
879 if (strcmp(cpu_arch, "sparc") == 0)
880 strcat(cpu_arch, "v9");
881 else if (strcmp(cpu_arch, "i386") == 0)
882 strcpy(cpu_arch, "amd64");
883 #endif
885 // Construct the invariant part of ld_library_path. Note that the
886 // space for the colon and the trailing null are provided by the
887 // nulls included by the sizeof operator.
888 size_t bufsize = sizeof(COMMON_DIR) + sizeof("/lib/") + strlen(cpu_arch);
889 common_path = malloc(bufsize);
890 if (common_path == NULL) {
891 free(info);
892 vm_exit_out_of_memory(bufsize,
893 "init_system_properties_values common_path");
894 }
895 sprintf(common_path, COMMON_DIR "/lib/%s", cpu_arch);
897 // struct size is more than sufficient for the path components obtained
898 // through the dlinfo() call, so only add additional space for the path
899 // components explicitly added here.
900 bufsize = info->dls_size + strlen(common_path);
901 library_path = malloc(bufsize);
902 if (library_path == NULL) {
903 free(info);
904 free(common_path);
905 vm_exit_out_of_memory(bufsize,
906 "init_system_properties_values library_path");
907 }
908 library_path[0] = '\0';
910 // Construct the desired Java library path from the linker's library
911 // search path.
912 //
913 // For compatibility, it is optimal that we insert the additional path
914 // components specific to the Java VM after those components specified
915 // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
916 // infrastructure.
917 if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it
918 strcpy(library_path, common_path);
919 } else {
920 int inserted = 0;
921 for (i = 0; i < info->dls_cnt; i++, path++) {
922 uint_t flags = path->dls_flags & LA_SER_MASK;
923 if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
924 strcat(library_path, common_path);
925 strcat(library_path, os::path_separator());
926 inserted = 1;
927 }
928 strcat(library_path, path->dls_name);
929 strcat(library_path, os::path_separator());
930 }
931 // eliminate trailing path separator
932 library_path[strlen(library_path)-1] = '\0';
933 }
935 // happens before argument parsing - can't use a trace flag
936 // tty->print_raw("init_system_properties_values: native lib path: ");
937 // tty->print_raw_cr(library_path);
939 // callee copies into its own buffer
940 Arguments::set_library_path(library_path);
942 free(common_path);
943 free(library_path);
944 free(info);
945 }
947 /*
948 * Extensions directories.
949 *
950 * Note that the space for the colon and the trailing null are provided
951 * by the nulls included by the sizeof operator (so actually one byte more
952 * than necessary is allocated).
953 */
954 {
955 char *buf = (char *) malloc(strlen(Arguments::get_java_home()) +
956 sizeof(EXTENSIONS_DIR) + sizeof(COMMON_DIR) +
957 sizeof(EXTENSIONS_DIR));
958 sprintf(buf, "%s" EXTENSIONS_DIR ":" COMMON_DIR EXTENSIONS_DIR,
959 Arguments::get_java_home());
960 Arguments::set_ext_dirs(buf);
961 }
963 /* Endorsed standards default directory. */
964 {
965 char * buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
966 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
967 Arguments::set_endorsed_dirs(buf);
968 }
969 }
971 #undef malloc
972 #undef free
973 #undef getenv
974 #undef EXTENSIONS_DIR
975 #undef ENDORSED_DIR
976 #undef COMMON_DIR
978 }
980 void os::breakpoint() {
981 BREAKPOINT;
982 }
984 bool os::obsolete_option(const JavaVMOption *option)
985 {
986 if (!strncmp(option->optionString, "-Xt", 3)) {
987 return true;
988 } else if (!strncmp(option->optionString, "-Xtm", 4)) {
989 return true;
990 } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
991 return true;
992 } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
993 return true;
994 }
995 return false;
996 }
998 bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
999 address stackStart = (address)thread->stack_base();
1000 address stackEnd = (address)(stackStart - (address)thread->stack_size());
1001 if (sp < stackStart && sp >= stackEnd ) return true;
1002 return false;
1003 }
1005 extern "C" void breakpoint() {
1006 // use debugger to set breakpoint here
1007 }
1009 // Returns an estimate of the current stack pointer. Result must be guaranteed to
1010 // point into the calling threads stack, and be no lower than the current stack
1011 // pointer.
1012 address os::current_stack_pointer() {
1013 volatile int dummy;
1014 address sp = (address)&dummy + 8; // %%%% need to confirm if this is right
1015 return sp;
1016 }
1018 static thread_t main_thread;
1020 // Thread start routine for all new Java threads
1021 extern "C" void* java_start(void* thread_addr) {
1022 // Try to randomize the cache line index of hot stack frames.
1023 // This helps when threads of the same stack traces evict each other's
1024 // cache lines. The threads can be either from the same JVM instance, or
1025 // from different JVM instances. The benefit is especially true for
1026 // processors with hyperthreading technology.
1027 static int counter = 0;
1028 int pid = os::current_process_id();
1029 alloca(((pid ^ counter++) & 7) * 128);
1031 int prio;
1032 Thread* thread = (Thread*)thread_addr;
1033 OSThread* osthr = thread->osthread();
1035 osthr->set_lwp_id( _lwp_self() ); // Store lwp in case we are bound
1036 thread->_schedctl = (void *) schedctl_init () ;
1038 if (UseNUMA) {
1039 int lgrp_id = os::numa_get_group_id();
1040 if (lgrp_id != -1) {
1041 thread->set_lgrp_id(lgrp_id);
1042 }
1043 }
1045 // If the creator called set priority before we started,
1046 // we need to call set priority now that we have an lwp.
1047 // Get the priority from libthread and set the priority
1048 // for the new Solaris lwp.
1049 if ( osthr->thread_id() != -1 ) {
1050 if ( UseThreadPriorities ) {
1051 thr_getprio(osthr->thread_id(), &prio);
1052 if (ThreadPriorityVerbose) {
1053 tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT ", setting priority: %d\n",
1054 osthr->thread_id(), osthr->lwp_id(), prio );
1055 }
1056 os::set_native_priority(thread, prio);
1057 }
1058 } else if (ThreadPriorityVerbose) {
1059 warning("Can't set priority in _start routine, thread id hasn't been set\n");
1060 }
1062 assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
1064 // initialize signal mask for this thread
1065 os::Solaris::hotspot_sigmask(thread);
1067 thread->run();
1069 // One less thread is executing
1070 // When the VMThread gets here, the main thread may have already exited
1071 // which frees the CodeHeap containing the Atomic::dec code
1072 if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
1073 Atomic::dec(&os::Solaris::_os_thread_count);
1074 }
1076 if (UseDetachedThreads) {
1077 thr_exit(NULL);
1078 ShouldNotReachHere();
1079 }
1080 return NULL;
1081 }
1083 static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
1084 // Allocate the OSThread object
1085 OSThread* osthread = new OSThread(NULL, NULL);
1086 if (osthread == NULL) return NULL;
1088 // Store info on the Solaris thread into the OSThread
1089 osthread->set_thread_id(thread_id);
1090 osthread->set_lwp_id(_lwp_self());
1091 thread->_schedctl = (void *) schedctl_init () ;
1093 if (UseNUMA) {
1094 int lgrp_id = os::numa_get_group_id();
1095 if (lgrp_id != -1) {
1096 thread->set_lgrp_id(lgrp_id);
1097 }
1098 }
1100 if ( ThreadPriorityVerbose ) {
1101 tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
1102 osthread->thread_id(), osthread->lwp_id() );
1103 }
1105 // Initial thread state is INITIALIZED, not SUSPENDED
1106 osthread->set_state(INITIALIZED);
1108 return osthread;
1109 }
1111 void os::Solaris::hotspot_sigmask(Thread* thread) {
1113 //Save caller's signal mask
1114 sigset_t sigmask;
1115 thr_sigsetmask(SIG_SETMASK, NULL, &sigmask);
1116 OSThread *osthread = thread->osthread();
1117 osthread->set_caller_sigmask(sigmask);
1119 thr_sigsetmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
1120 if (!ReduceSignalUsage) {
1121 if (thread->is_VM_thread()) {
1122 // Only the VM thread handles BREAK_SIGNAL ...
1123 thr_sigsetmask(SIG_UNBLOCK, vm_signals(), NULL);
1124 } else {
1125 // ... all other threads block BREAK_SIGNAL
1126 assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
1127 thr_sigsetmask(SIG_BLOCK, vm_signals(), NULL);
1128 }
1129 }
1130 }
1132 bool os::create_attached_thread(JavaThread* thread) {
1133 #ifdef ASSERT
1134 thread->verify_not_published();
1135 #endif
1136 OSThread* osthread = create_os_thread(thread, thr_self());
1137 if (osthread == NULL) {
1138 return false;
1139 }
1141 // Initial thread state is RUNNABLE
1142 osthread->set_state(RUNNABLE);
1143 thread->set_osthread(osthread);
1145 // initialize signal mask for this thread
1146 // and save the caller's signal mask
1147 os::Solaris::hotspot_sigmask(thread);
1149 return true;
1150 }
1152 bool os::create_main_thread(JavaThread* thread) {
1153 #ifdef ASSERT
1154 thread->verify_not_published();
1155 #endif
1156 if (_starting_thread == NULL) {
1157 _starting_thread = create_os_thread(thread, main_thread);
1158 if (_starting_thread == NULL) {
1159 return false;
1160 }
1161 }
1163 // The primodial thread is runnable from the start
1164 _starting_thread->set_state(RUNNABLE);
1166 thread->set_osthread(_starting_thread);
1168 // initialize signal mask for this thread
1169 // and save the caller's signal mask
1170 os::Solaris::hotspot_sigmask(thread);
1172 return true;
1173 }
1175 // _T2_libthread is true if we believe we are running with the newer
1176 // SunSoft lwp/libthread.so (2.8 patch, 2.9 default)
1177 bool os::Solaris::_T2_libthread = false;
1179 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1180 // Allocate the OSThread object
1181 OSThread* osthread = new OSThread(NULL, NULL);
1182 if (osthread == NULL) {
1183 return false;
1184 }
1186 if ( ThreadPriorityVerbose ) {
1187 char *thrtyp;
1188 switch ( thr_type ) {
1189 case vm_thread:
1190 thrtyp = (char *)"vm";
1191 break;
1192 case cgc_thread:
1193 thrtyp = (char *)"cgc";
1194 break;
1195 case pgc_thread:
1196 thrtyp = (char *)"pgc";
1197 break;
1198 case java_thread:
1199 thrtyp = (char *)"java";
1200 break;
1201 case compiler_thread:
1202 thrtyp = (char *)"compiler";
1203 break;
1204 case watcher_thread:
1205 thrtyp = (char *)"watcher";
1206 break;
1207 default:
1208 thrtyp = (char *)"unknown";
1209 break;
1210 }
1211 tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
1212 }
1214 // Calculate stack size if it's not specified by caller.
1215 if (stack_size == 0) {
1216 // The default stack size 1M (2M for LP64).
1217 stack_size = (BytesPerWord >> 2) * K * K;
1219 switch (thr_type) {
1220 case os::java_thread:
1221 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
1222 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
1223 break;
1224 case os::compiler_thread:
1225 if (CompilerThreadStackSize > 0) {
1226 stack_size = (size_t)(CompilerThreadStackSize * K);
1227 break;
1228 } // else fall through:
1229 // use VMThreadStackSize if CompilerThreadStackSize is not defined
1230 case os::vm_thread:
1231 case os::pgc_thread:
1232 case os::cgc_thread:
1233 case os::watcher_thread:
1234 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1235 break;
1236 }
1237 }
1238 stack_size = MAX2(stack_size, os::Solaris::min_stack_allowed);
1240 // Initial state is ALLOCATED but not INITIALIZED
1241 osthread->set_state(ALLOCATED);
1243 if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
1244 // We got lots of threads. Check if we still have some address space left.
1245 // Need to be at least 5Mb of unreserved address space. We do check by
1246 // trying to reserve some.
1247 const size_t VirtualMemoryBangSize = 20*K*K;
1248 char* mem = os::reserve_memory(VirtualMemoryBangSize);
1249 if (mem == NULL) {
1250 delete osthread;
1251 return false;
1252 } else {
1253 // Release the memory again
1254 os::release_memory(mem, VirtualMemoryBangSize);
1255 }
1256 }
1258 // Setup osthread because the child thread may need it.
1259 thread->set_osthread(osthread);
1261 // Create the Solaris thread
1262 // explicit THR_BOUND for T2_libthread case in case
1263 // that assumption is not accurate, but our alternate signal stack
1264 // handling is based on it which must have bound threads
1265 thread_t tid = 0;
1266 long flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED
1267 | ((UseBoundThreads || os::Solaris::T2_libthread() ||
1268 (thr_type == vm_thread) ||
1269 (thr_type == cgc_thread) ||
1270 (thr_type == pgc_thread) ||
1271 (thr_type == compiler_thread && BackgroundCompilation)) ?
1272 THR_BOUND : 0);
1273 int status;
1275 // 4376845 -- libthread/kernel don't provide enough LWPs to utilize all CPUs.
1276 //
1277 // On multiprocessors systems, libthread sometimes under-provisions our
1278 // process with LWPs. On a 30-way systems, for instance, we could have
1279 // 50 user-level threads in ready state and only 2 or 3 LWPs assigned
1280 // to our process. This can result in under utilization of PEs.
1281 // I suspect the problem is related to libthread's LWP
1282 // pool management and to the kernel's SIGBLOCKING "last LWP parked"
1283 // upcall policy.
1284 //
1285 // The following code is palliative -- it attempts to ensure that our
1286 // process has sufficient LWPs to take advantage of multiple PEs.
1287 // Proper long-term cures include using user-level threads bound to LWPs
1288 // (THR_BOUND) or using LWP-based synchronization. Note that there is a
1289 // slight timing window with respect to sampling _os_thread_count, but
1290 // the race is benign. Also, we should periodically recompute
1291 // _processors_online as the min of SC_NPROCESSORS_ONLN and the
1292 // the number of PEs in our partition. You might be tempted to use
1293 // THR_NEW_LWP here, but I'd recommend against it as that could
1294 // result in undesirable growth of the libthread's LWP pool.
1295 // The fix below isn't sufficient; for instance, it doesn't take into count
1296 // LWPs parked on IO. It does, however, help certain CPU-bound benchmarks.
1297 //
1298 // Some pathologies this scheme doesn't handle:
1299 // * Threads can block, releasing the LWPs. The LWPs can age out.
1300 // When a large number of threads become ready again there aren't
1301 // enough LWPs available to service them. This can occur when the
1302 // number of ready threads oscillates.
1303 // * LWPs/Threads park on IO, thus taking the LWP out of circulation.
1304 //
1305 // Finally, we should call thr_setconcurrency() periodically to refresh
1306 // the LWP pool and thwart the LWP age-out mechanism.
1307 // The "+3" term provides a little slop -- we want to slightly overprovision.
1309 if (AdjustConcurrency && os::Solaris::_os_thread_count < (_processors_online+3)) {
1310 if (!(flags & THR_BOUND)) {
1311 thr_setconcurrency (os::Solaris::_os_thread_count); // avoid starvation
1312 }
1313 }
1314 // Although this doesn't hurt, we should warn of undefined behavior
1315 // when using unbound T1 threads with schedctl(). This should never
1316 // happen, as the compiler and VM threads are always created bound
1317 DEBUG_ONLY(
1318 if ((VMThreadHintNoPreempt || CompilerThreadHintNoPreempt) &&
1319 (!os::Solaris::T2_libthread() && (!(flags & THR_BOUND))) &&
1320 ((thr_type == vm_thread) || (thr_type == cgc_thread) ||
1321 (thr_type == pgc_thread) || (thr_type == compiler_thread && BackgroundCompilation))) {
1322 warning("schedctl behavior undefined when Compiler/VM/GC Threads are Unbound");
1323 }
1324 );
1327 // Mark that we don't have an lwp or thread id yet.
1328 // In case we attempt to set the priority before the thread starts.
1329 osthread->set_lwp_id(-1);
1330 osthread->set_thread_id(-1);
1332 status = thr_create(NULL, stack_size, java_start, thread, flags, &tid);
1333 if (status != 0) {
1334 if (PrintMiscellaneous && (Verbose || WizardMode)) {
1335 perror("os::create_thread");
1336 }
1337 thread->set_osthread(NULL);
1338 // Need to clean up stuff we've allocated so far
1339 delete osthread;
1340 return false;
1341 }
1343 Atomic::inc(&os::Solaris::_os_thread_count);
1345 // Store info on the Solaris thread into the OSThread
1346 osthread->set_thread_id(tid);
1348 // Remember that we created this thread so we can set priority on it
1349 osthread->set_vm_created();
1351 // Set the default thread priority otherwise use NormalPriority
1353 if ( UseThreadPriorities ) {
1354 thr_setprio(tid, (DefaultThreadPriority == -1) ?
1355 java_to_os_priority[NormPriority] :
1356 DefaultThreadPriority);
1357 }
1359 // Initial thread state is INITIALIZED, not SUSPENDED
1360 osthread->set_state(INITIALIZED);
1362 // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1363 return true;
1364 }
1366 /* defined for >= Solaris 10. This allows builds on earlier versions
1367 * of Solaris to take advantage of the newly reserved Solaris JVM signals
1368 * With SIGJVM1, SIGJVM2, INTERRUPT_SIGNAL is SIGJVM1, ASYNC_SIGNAL is SIGJVM2
1369 * and -XX:+UseAltSigs does nothing since these should have no conflict
1370 */
1371 #if !defined(SIGJVM1)
1372 #define SIGJVM1 39
1373 #define SIGJVM2 40
1374 #endif
1376 debug_only(static bool signal_sets_initialized = false);
1377 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
1378 int os::Solaris::_SIGinterrupt = INTERRUPT_SIGNAL;
1379 int os::Solaris::_SIGasync = ASYNC_SIGNAL;
1381 bool os::Solaris::is_sig_ignored(int sig) {
1382 struct sigaction oact;
1383 sigaction(sig, (struct sigaction*)NULL, &oact);
1384 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
1385 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
1386 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
1387 return true;
1388 else
1389 return false;
1390 }
1392 // Note: SIGRTMIN is a macro that calls sysconf() so it will
1393 // dynamically detect SIGRTMIN value for the system at runtime, not buildtime
1394 static bool isJVM1available() {
1395 return SIGJVM1 < SIGRTMIN;
1396 }
1398 void os::Solaris::signal_sets_init() {
1399 // Should also have an assertion stating we are still single-threaded.
1400 assert(!signal_sets_initialized, "Already initialized");
1401 // Fill in signals that are necessarily unblocked for all threads in
1402 // the VM. Currently, we unblock the following signals:
1403 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1404 // by -Xrs (=ReduceSignalUsage));
1405 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1406 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1407 // the dispositions or masks wrt these signals.
1408 // Programs embedding the VM that want to use the above signals for their
1409 // own purposes must, at this time, use the "-Xrs" option to prevent
1410 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1411 // (See bug 4345157, and other related bugs).
1412 // In reality, though, unblocking these signals is really a nop, since
1413 // these signals are not blocked by default.
1414 sigemptyset(&unblocked_sigs);
1415 sigemptyset(&allowdebug_blocked_sigs);
1416 sigaddset(&unblocked_sigs, SIGILL);
1417 sigaddset(&unblocked_sigs, SIGSEGV);
1418 sigaddset(&unblocked_sigs, SIGBUS);
1419 sigaddset(&unblocked_sigs, SIGFPE);
1421 if (isJVM1available) {
1422 os::Solaris::set_SIGinterrupt(SIGJVM1);
1423 os::Solaris::set_SIGasync(SIGJVM2);
1424 } else if (UseAltSigs) {
1425 os::Solaris::set_SIGinterrupt(ALT_INTERRUPT_SIGNAL);
1426 os::Solaris::set_SIGasync(ALT_ASYNC_SIGNAL);
1427 } else {
1428 os::Solaris::set_SIGinterrupt(INTERRUPT_SIGNAL);
1429 os::Solaris::set_SIGasync(ASYNC_SIGNAL);
1430 }
1432 sigaddset(&unblocked_sigs, os::Solaris::SIGinterrupt());
1433 sigaddset(&unblocked_sigs, os::Solaris::SIGasync());
1435 if (!ReduceSignalUsage) {
1436 if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1437 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1438 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
1439 }
1440 if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1441 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1442 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
1443 }
1444 if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1445 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1446 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
1447 }
1448 }
1449 // Fill in signals that are blocked by all but the VM thread.
1450 sigemptyset(&vm_sigs);
1451 if (!ReduceSignalUsage)
1452 sigaddset(&vm_sigs, BREAK_SIGNAL);
1453 debug_only(signal_sets_initialized = true);
1455 // For diagnostics only used in run_periodic_checks
1456 sigemptyset(&check_signal_done);
1457 }
1459 // These are signals that are unblocked while a thread is running Java.
1460 // (For some reason, they get blocked by default.)
1461 sigset_t* os::Solaris::unblocked_signals() {
1462 assert(signal_sets_initialized, "Not initialized");
1463 return &unblocked_sigs;
1464 }
1466 // These are the signals that are blocked while a (non-VM) thread is
1467 // running Java. Only the VM thread handles these signals.
1468 sigset_t* os::Solaris::vm_signals() {
1469 assert(signal_sets_initialized, "Not initialized");
1470 return &vm_sigs;
1471 }
1473 // These are signals that are blocked during cond_wait to allow debugger in
1474 sigset_t* os::Solaris::allowdebug_blocked_signals() {
1475 assert(signal_sets_initialized, "Not initialized");
1476 return &allowdebug_blocked_sigs;
1477 }
1480 void _handle_uncaught_cxx_exception() {
1481 VMError err("An uncaught C++ exception");
1482 err.report_and_die();
1483 }
1486 // First crack at OS-specific initialization, from inside the new thread.
1487 void os::initialize_thread() {
1488 int r = thr_main() ;
1489 guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
1490 if (r) {
1491 JavaThread* jt = (JavaThread *)Thread::current();
1492 assert(jt != NULL,"Sanity check");
1493 size_t stack_size;
1494 address base = jt->stack_base();
1495 if (Arguments::created_by_java_launcher()) {
1496 // Use 2MB to allow for Solaris 7 64 bit mode.
1497 stack_size = JavaThread::stack_size_at_create() == 0
1498 ? 2048*K : JavaThread::stack_size_at_create();
1500 // There are rare cases when we may have already used more than
1501 // the basic stack size allotment before this method is invoked.
1502 // Attempt to allow for a normally sized java_stack.
1503 size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1504 stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1505 } else {
1506 // 6269555: If we were not created by a Java launcher, i.e. if we are
1507 // running embedded in a native application, treat the primordial thread
1508 // as much like a native attached thread as possible. This means using
1509 // the current stack size from thr_stksegment(), unless it is too large
1510 // to reliably setup guard pages. A reasonable max size is 8MB.
1511 size_t current_size = current_stack_size();
1512 // This should never happen, but just in case....
1513 if (current_size == 0) current_size = 2 * K * K;
1514 stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1515 }
1516 address bottom = (address)align_size_up((intptr_t)(base - stack_size), os::vm_page_size());;
1517 stack_size = (size_t)(base - bottom);
1519 assert(stack_size > 0, "Stack size calculation problem");
1521 if (stack_size > jt->stack_size()) {
1522 NOT_PRODUCT(
1523 struct rlimit limits;
1524 getrlimit(RLIMIT_STACK, &limits);
1525 size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1526 assert(size >= jt->stack_size(), "Stack size problem in main thread");
1527 )
1528 tty->print_cr(
1529 "Stack size of %d Kb exceeds current limit of %d Kb.\n"
1530 "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1531 "See limit(1) to increase the stack size limit.",
1532 stack_size / K, jt->stack_size() / K);
1533 vm_exit(1);
1534 }
1535 assert(jt->stack_size() >= stack_size,
1536 "Attempt to map more stack than was allocated");
1537 jt->set_stack_size(stack_size);
1538 }
1540 // 5/22/01: Right now alternate signal stacks do not handle
1541 // throwing stack overflow exceptions, see bug 4463178
1542 // Until a fix is found for this, T2 will NOT imply alternate signal
1543 // stacks.
1544 // If using T2 libthread threads, install an alternate signal stack.
1545 // Because alternate stacks associate with LWPs on Solaris,
1546 // see sigaltstack(2), if using UNBOUND threads, or if UseBoundThreads
1547 // we prefer to explicitly stack bang.
1548 // If not using T2 libthread, but using UseBoundThreads any threads
1549 // (primordial thread, jni_attachCurrentThread) we do not create,
1550 // probably are not bound, therefore they can not have an alternate
1551 // signal stack. Since our stack banging code is generated and
1552 // is shared across threads, all threads must be bound to allow
1553 // using alternate signal stacks. The alternative is to interpose
1554 // on _lwp_create to associate an alt sig stack with each LWP,
1555 // and this could be a problem when the JVM is embedded.
1556 // We would prefer to use alternate signal stacks with T2
1557 // Since there is currently no accurate way to detect T2
1558 // we do not. Assuming T2 when running T1 causes sig 11s or assertions
1559 // on installing alternate signal stacks
1562 // 05/09/03: removed alternate signal stack support for Solaris
1563 // The alternate signal stack mechanism is no longer needed to
1564 // handle stack overflow. This is now handled by allocating
1565 // guard pages (red zone) and stackbanging.
1566 // Initially the alternate signal stack mechanism was removed because
1567 // it did not work with T1 llibthread. Alternate
1568 // signal stacks MUST have all threads bound to lwps. Applications
1569 // can create their own threads and attach them without their being
1570 // bound under T1. This is frequently the case for the primordial thread.
1571 // If we were ever to reenable this mechanism we would need to
1572 // use the dynamic check for T2 libthread.
1574 os::Solaris::init_thread_fpu_state();
1575 std::set_terminate(_handle_uncaught_cxx_exception);
1576 }
1580 // Free Solaris resources related to the OSThread
1581 void os::free_thread(OSThread* osthread) {
1582 assert(osthread != NULL, "os::free_thread but osthread not set");
1585 // We are told to free resources of the argument thread,
1586 // but we can only really operate on the current thread.
1587 // The main thread must take the VMThread down synchronously
1588 // before the main thread exits and frees up CodeHeap
1589 guarantee((Thread::current()->osthread() == osthread
1590 || (osthread == VMThread::vm_thread()->osthread())), "os::free_thread but not current thread");
1591 if (Thread::current()->osthread() == osthread) {
1592 // Restore caller's signal mask
1593 sigset_t sigmask = osthread->caller_sigmask();
1594 thr_sigsetmask(SIG_SETMASK, &sigmask, NULL);
1595 }
1596 delete osthread;
1597 }
1599 void os::pd_start_thread(Thread* thread) {
1600 int status = thr_continue(thread->osthread()->thread_id());
1601 assert_status(status == 0, status, "thr_continue failed");
1602 }
1605 intx os::current_thread_id() {
1606 return (intx)thr_self();
1607 }
1609 static pid_t _initial_pid = 0;
1611 int os::current_process_id() {
1612 return (int)(_initial_pid ? _initial_pid : getpid());
1613 }
1615 int os::allocate_thread_local_storage() {
1616 // %%% in Win32 this allocates a memory segment pointed to by a
1617 // register. Dan Stein can implement a similar feature in
1618 // Solaris. Alternatively, the VM can do the same thing
1619 // explicitly: malloc some storage and keep the pointer in a
1620 // register (which is part of the thread's context) (or keep it
1621 // in TLS).
1622 // %%% In current versions of Solaris, thr_self and TSD can
1623 // be accessed via short sequences of displaced indirections.
1624 // The value of thr_self is available as %g7(36).
1625 // The value of thr_getspecific(k) is stored in %g7(12)(4)(k*4-4),
1626 // assuming that the current thread already has a value bound to k.
1627 // It may be worth experimenting with such access patterns,
1628 // and later having the parameters formally exported from a Solaris
1629 // interface. I think, however, that it will be faster to
1630 // maintain the invariant that %g2 always contains the
1631 // JavaThread in Java code, and have stubs simply
1632 // treat %g2 as a caller-save register, preserving it in a %lN.
1633 thread_key_t tk;
1634 if (thr_keycreate( &tk, NULL ) )
1635 fatal(err_msg("os::allocate_thread_local_storage: thr_keycreate failed "
1636 "(%s)", strerror(errno)));
1637 return int(tk);
1638 }
1640 void os::free_thread_local_storage(int index) {
1641 // %%% don't think we need anything here
1642 // if ( pthread_key_delete((pthread_key_t) tk) )
1643 // fatal("os::free_thread_local_storage: pthread_key_delete failed");
1644 }
1646 #define SMALLINT 32 // libthread allocate for tsd_common is a version specific
1647 // small number - point is NO swap space available
1648 void os::thread_local_storage_at_put(int index, void* value) {
1649 // %%% this is used only in threadLocalStorage.cpp
1650 if (thr_setspecific((thread_key_t)index, value)) {
1651 if (errno == ENOMEM) {
1652 vm_exit_out_of_memory(SMALLINT, "thr_setspecific: out of swap space");
1653 } else {
1654 fatal(err_msg("os::thread_local_storage_at_put: thr_setspecific failed "
1655 "(%s)", strerror(errno)));
1656 }
1657 } else {
1658 ThreadLocalStorage::set_thread_in_slot ((Thread *) value) ;
1659 }
1660 }
1662 // This function could be called before TLS is initialized, for example, when
1663 // VM receives an async signal or when VM causes a fatal error during
1664 // initialization. Return NULL if thr_getspecific() fails.
1665 void* os::thread_local_storage_at(int index) {
1666 // %%% this is used only in threadLocalStorage.cpp
1667 void* r = NULL;
1668 return thr_getspecific((thread_key_t)index, &r) != 0 ? NULL : r;
1669 }
1672 const int NANOSECS_PER_MILLISECS = 1000000;
1673 // gethrtime can move backwards if read from one cpu and then a different cpu
1674 // getTimeNanos is guaranteed to not move backward on Solaris
1675 // local spinloop created as faster for a CAS on an int than
1676 // a CAS on a 64bit jlong. Also Atomic::cmpxchg for jlong is not
1677 // supported on sparc v8 or pre supports_cx8 intel boxes.
1678 // oldgetTimeNanos for systems which do not support CAS on 64bit jlong
1679 // i.e. sparc v8 and pre supports_cx8 (i486) intel boxes
1680 inline hrtime_t oldgetTimeNanos() {
1681 int gotlock = LOCK_INVALID;
1682 hrtime_t newtime = gethrtime();
1684 for (;;) {
1685 // grab lock for max_hrtime
1686 int curlock = max_hrtime_lock;
1687 if (curlock & LOCK_BUSY) continue;
1688 if (gotlock = Atomic::cmpxchg(LOCK_BUSY, &max_hrtime_lock, LOCK_FREE) != LOCK_FREE) continue;
1689 if (newtime > max_hrtime) {
1690 max_hrtime = newtime;
1691 } else {
1692 newtime = max_hrtime;
1693 }
1694 // release lock
1695 max_hrtime_lock = LOCK_FREE;
1696 return newtime;
1697 }
1698 }
1699 // gethrtime can move backwards if read from one cpu and then a different cpu
1700 // getTimeNanos is guaranteed to not move backward on Solaris
1701 inline hrtime_t getTimeNanos() {
1702 if (VM_Version::supports_cx8()) {
1703 const hrtime_t now = gethrtime();
1704 // Use atomic long load since 32-bit x86 uses 2 registers to keep long.
1705 const hrtime_t prev = Atomic::load((volatile jlong*)&max_hrtime);
1706 if (now <= prev) return prev; // same or retrograde time;
1707 const hrtime_t obsv = Atomic::cmpxchg(now, (volatile jlong*)&max_hrtime, prev);
1708 assert(obsv >= prev, "invariant"); // Monotonicity
1709 // If the CAS succeeded then we're done and return "now".
1710 // If the CAS failed and the observed value "obs" is >= now then
1711 // we should return "obs". If the CAS failed and now > obs > prv then
1712 // some other thread raced this thread and installed a new value, in which case
1713 // we could either (a) retry the entire operation, (b) retry trying to install now
1714 // or (c) just return obs. We use (c). No loop is required although in some cases
1715 // we might discard a higher "now" value in deference to a slightly lower but freshly
1716 // installed obs value. That's entirely benign -- it admits no new orderings compared
1717 // to (a) or (b) -- and greatly reduces coherence traffic.
1718 // We might also condition (c) on the magnitude of the delta between obs and now.
1719 // Avoiding excessive CAS operations to hot RW locations is critical.
1720 // See http://blogs.sun.com/dave/entry/cas_and_cache_trivia_invalidate
1721 return (prev == obsv) ? now : obsv ;
1722 } else {
1723 return oldgetTimeNanos();
1724 }
1725 }
1727 // Time since start-up in seconds to a fine granularity.
1728 // Used by VMSelfDestructTimer and the MemProfiler.
1729 double os::elapsedTime() {
1730 return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1731 }
1733 jlong os::elapsed_counter() {
1734 return (jlong)(getTimeNanos() - first_hrtime);
1735 }
1737 jlong os::elapsed_frequency() {
1738 return hrtime_hz;
1739 }
1741 // Return the real, user, and system times in seconds from an
1742 // arbitrary fixed point in the past.
1743 bool os::getTimesSecs(double* process_real_time,
1744 double* process_user_time,
1745 double* process_system_time) {
1746 struct tms ticks;
1747 clock_t real_ticks = times(&ticks);
1749 if (real_ticks == (clock_t) (-1)) {
1750 return false;
1751 } else {
1752 double ticks_per_second = (double) clock_tics_per_sec;
1753 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1754 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1755 // For consistency return the real time from getTimeNanos()
1756 // converted to seconds.
1757 *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1759 return true;
1760 }
1761 }
1763 bool os::supports_vtime() { return true; }
1765 bool os::enable_vtime() {
1766 int fd = ::open("/proc/self/ctl", O_WRONLY);
1767 if (fd == -1)
1768 return false;
1770 long cmd[] = { PCSET, PR_MSACCT };
1771 int res = ::write(fd, cmd, sizeof(long) * 2);
1772 ::close(fd);
1773 if (res != sizeof(long) * 2)
1774 return false;
1776 return true;
1777 }
1779 bool os::vtime_enabled() {
1780 int fd = ::open("/proc/self/status", O_RDONLY);
1781 if (fd == -1)
1782 return false;
1784 pstatus_t status;
1785 int res = os::read(fd, (void*) &status, sizeof(pstatus_t));
1786 ::close(fd);
1787 if (res != sizeof(pstatus_t))
1788 return false;
1790 return status.pr_flags & PR_MSACCT;
1791 }
1793 double os::elapsedVTime() {
1794 return (double)gethrvtime() / (double)hrtime_hz;
1795 }
1797 // Used internally for comparisons only
1798 // getTimeMillis guaranteed to not move backwards on Solaris
1799 jlong getTimeMillis() {
1800 jlong nanotime = getTimeNanos();
1801 return (jlong)(nanotime / NANOSECS_PER_MILLISECS);
1802 }
1804 // Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1805 jlong os::javaTimeMillis() {
1806 timeval t;
1807 if (gettimeofday( &t, NULL) == -1)
1808 fatal(err_msg("os::javaTimeMillis: gettimeofday (%s)", strerror(errno)));
1809 return jlong(t.tv_sec) * 1000 + jlong(t.tv_usec) / 1000;
1810 }
1812 jlong os::javaTimeNanos() {
1813 return (jlong)getTimeNanos();
1814 }
1816 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1817 info_ptr->max_value = ALL_64_BITS; // gethrtime() uses all 64 bits
1818 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1819 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1820 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1821 }
1823 char * os::local_time_string(char *buf, size_t buflen) {
1824 struct tm t;
1825 time_t long_time;
1826 time(&long_time);
1827 localtime_r(&long_time, &t);
1828 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1829 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1830 t.tm_hour, t.tm_min, t.tm_sec);
1831 return buf;
1832 }
1834 // Note: os::shutdown() might be called very early during initialization, or
1835 // called from signal handler. Before adding something to os::shutdown(), make
1836 // sure it is async-safe and can handle partially initialized VM.
1837 void os::shutdown() {
1839 // allow PerfMemory to attempt cleanup of any persistent resources
1840 perfMemory_exit();
1842 // needs to remove object in file system
1843 AttachListener::abort();
1845 // flush buffered output, finish log files
1846 ostream_abort();
1848 // Check for abort hook
1849 abort_hook_t abort_hook = Arguments::abort_hook();
1850 if (abort_hook != NULL) {
1851 abort_hook();
1852 }
1853 }
1855 // Note: os::abort() might be called very early during initialization, or
1856 // called from signal handler. Before adding something to os::abort(), make
1857 // sure it is async-safe and can handle partially initialized VM.
1858 void os::abort(bool dump_core) {
1859 os::shutdown();
1860 if (dump_core) {
1861 #ifndef PRODUCT
1862 fdStream out(defaultStream::output_fd());
1863 out.print_raw("Current thread is ");
1864 char buf[16];
1865 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1866 out.print_raw_cr(buf);
1867 out.print_raw_cr("Dumping core ...");
1868 #endif
1869 ::abort(); // dump core (for debugging)
1870 }
1872 ::exit(1);
1873 }
1875 // Die immediately, no exit hook, no abort hook, no cleanup.
1876 void os::die() {
1877 _exit(-1);
1878 }
1880 // unused
1881 void os::set_error_file(const char *logfile) {}
1883 // DLL functions
1885 const char* os::dll_file_extension() { return ".so"; }
1887 // This must be hard coded because it's the system's temporary
1888 // directory not the java application's temp directory, ala java.io.tmpdir.
1889 const char* os::get_temp_directory() { return "/tmp"; }
1891 static bool file_exists(const char* filename) {
1892 struct stat statbuf;
1893 if (filename == NULL || strlen(filename) == 0) {
1894 return false;
1895 }
1896 return os::stat(filename, &statbuf) == 0;
1897 }
1899 void os::dll_build_name(char* buffer, size_t buflen,
1900 const char* pname, const char* fname) {
1901 const size_t pnamelen = pname ? strlen(pname) : 0;
1903 // Quietly truncate on buffer overflow. Should be an error.
1904 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1905 *buffer = '\0';
1906 return;
1907 }
1909 if (pnamelen == 0) {
1910 snprintf(buffer, buflen, "lib%s.so", fname);
1911 } else if (strchr(pname, *os::path_separator()) != NULL) {
1912 int n;
1913 char** pelements = split_path(pname, &n);
1914 for (int i = 0 ; i < n ; i++) {
1915 // really shouldn't be NULL but what the heck, check can't hurt
1916 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1917 continue; // skip the empty path values
1918 }
1919 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1920 if (file_exists(buffer)) {
1921 break;
1922 }
1923 }
1924 // release the storage
1925 for (int i = 0 ; i < n ; i++) {
1926 if (pelements[i] != NULL) {
1927 FREE_C_HEAP_ARRAY(char, pelements[i]);
1928 }
1929 }
1930 if (pelements != NULL) {
1931 FREE_C_HEAP_ARRAY(char*, pelements);
1932 }
1933 } else {
1934 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1935 }
1936 }
1938 const char* os::get_current_directory(char *buf, int buflen) {
1939 return getcwd(buf, buflen);
1940 }
1942 // check if addr is inside libjvm[_g].so
1943 bool os::address_is_in_vm(address addr) {
1944 static address libjvm_base_addr;
1945 Dl_info dlinfo;
1947 if (libjvm_base_addr == NULL) {
1948 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1949 libjvm_base_addr = (address)dlinfo.dli_fbase;
1950 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1951 }
1953 if (dladdr((void *)addr, &dlinfo)) {
1954 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1955 }
1957 return false;
1958 }
1960 typedef int (*dladdr1_func_type) (void *, Dl_info *, void **, int);
1961 static dladdr1_func_type dladdr1_func = NULL;
1963 bool os::dll_address_to_function_name(address addr, char *buf,
1964 int buflen, int * offset) {
1965 Dl_info dlinfo;
1967 // dladdr1_func was initialized in os::init()
1968 if (dladdr1_func){
1969 // yes, we have dladdr1
1971 // Support for dladdr1 is checked at runtime; it may be
1972 // available even if the vm is built on a machine that does
1973 // not have dladdr1 support. Make sure there is a value for
1974 // RTLD_DL_SYMENT.
1975 #ifndef RTLD_DL_SYMENT
1976 #define RTLD_DL_SYMENT 1
1977 #endif
1978 #ifdef _LP64
1979 Elf64_Sym * info;
1980 #else
1981 Elf32_Sym * info;
1982 #endif
1983 if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1984 RTLD_DL_SYMENT)) {
1985 if ((char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1986 if (buf != NULL) {
1987 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen))
1988 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1989 }
1990 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1991 return true;
1992 }
1993 }
1994 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1995 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1996 dlinfo.dli_fname, buf, buflen, offset) == Decoder::no_error) {
1997 return true;
1998 }
1999 }
2000 if (buf != NULL) buf[0] = '\0';
2001 if (offset != NULL) *offset = -1;
2002 return false;
2003 } else {
2004 // no, only dladdr is available
2005 if (dladdr((void *)addr, &dlinfo)) {
2006 if (buf != NULL) {
2007 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen))
2008 jio_snprintf(buf, buflen, dlinfo.dli_sname);
2009 }
2010 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
2011 return true;
2012 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
2013 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
2014 dlinfo.dli_fname, buf, buflen, offset) == Decoder::no_error) {
2015 return true;
2016 }
2017 }
2018 if (buf != NULL) buf[0] = '\0';
2019 if (offset != NULL) *offset = -1;
2020 return false;
2021 }
2022 }
2024 bool os::dll_address_to_library_name(address addr, char* buf,
2025 int buflen, int* offset) {
2026 Dl_info dlinfo;
2028 if (dladdr((void*)addr, &dlinfo)){
2029 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
2030 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
2031 return true;
2032 } else {
2033 if (buf) buf[0] = '\0';
2034 if (offset) *offset = -1;
2035 return false;
2036 }
2037 }
2039 // Prints the names and full paths of all opened dynamic libraries
2040 // for current process
2041 void os::print_dll_info(outputStream * st) {
2042 Dl_info dli;
2043 void *handle;
2044 Link_map *map;
2045 Link_map *p;
2047 st->print_cr("Dynamic libraries:"); st->flush();
2049 if (!dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli)) {
2050 st->print_cr("Error: Cannot print dynamic libraries.");
2051 return;
2052 }
2053 handle = dlopen(dli.dli_fname, RTLD_LAZY);
2054 if (handle == NULL) {
2055 st->print_cr("Error: Cannot print dynamic libraries.");
2056 return;
2057 }
2058 dlinfo(handle, RTLD_DI_LINKMAP, &map);
2059 if (map == NULL) {
2060 st->print_cr("Error: Cannot print dynamic libraries.");
2061 return;
2062 }
2064 while (map->l_prev != NULL)
2065 map = map->l_prev;
2067 while (map != NULL) {
2068 st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
2069 map = map->l_next;
2070 }
2072 dlclose(handle);
2073 }
2075 // Loads .dll/.so and
2076 // in case of error it checks if .dll/.so was built for the
2077 // same architecture as Hotspot is running on
2079 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
2080 {
2081 void * result= ::dlopen(filename, RTLD_LAZY);
2082 if (result != NULL) {
2083 // Successful loading
2084 return result;
2085 }
2087 Elf32_Ehdr elf_head;
2089 // Read system error message into ebuf
2090 // It may or may not be overwritten below
2091 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2092 ebuf[ebuflen-1]='\0';
2093 int diag_msg_max_length=ebuflen-strlen(ebuf);
2094 char* diag_msg_buf=ebuf+strlen(ebuf);
2096 if (diag_msg_max_length==0) {
2097 // No more space in ebuf for additional diagnostics message
2098 return NULL;
2099 }
2102 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2104 if (file_descriptor < 0) {
2105 // Can't open library, report dlerror() message
2106 return NULL;
2107 }
2109 bool failed_to_read_elf_head=
2110 (sizeof(elf_head)!=
2111 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2113 ::close(file_descriptor);
2114 if (failed_to_read_elf_head) {
2115 // file i/o error - report dlerror() msg
2116 return NULL;
2117 }
2119 typedef struct {
2120 Elf32_Half code; // Actual value as defined in elf.h
2121 Elf32_Half compat_class; // Compatibility of archs at VM's sense
2122 char elf_class; // 32 or 64 bit
2123 char endianess; // MSB or LSB
2124 char* name; // String representation
2125 } arch_t;
2127 static const arch_t arch_array[]={
2128 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2129 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2130 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2131 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2132 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2133 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2134 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2135 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2136 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
2137 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
2138 };
2140 #if (defined IA32)
2141 static Elf32_Half running_arch_code=EM_386;
2142 #elif (defined AMD64)
2143 static Elf32_Half running_arch_code=EM_X86_64;
2144 #elif (defined IA64)
2145 static Elf32_Half running_arch_code=EM_IA_64;
2146 #elif (defined __sparc) && (defined _LP64)
2147 static Elf32_Half running_arch_code=EM_SPARCV9;
2148 #elif (defined __sparc) && (!defined _LP64)
2149 static Elf32_Half running_arch_code=EM_SPARC;
2150 #elif (defined __powerpc64__)
2151 static Elf32_Half running_arch_code=EM_PPC64;
2152 #elif (defined __powerpc__)
2153 static Elf32_Half running_arch_code=EM_PPC;
2154 #elif (defined ARM)
2155 static Elf32_Half running_arch_code=EM_ARM;
2156 #else
2157 #error Method os::dll_load requires that one of following is defined:\
2158 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
2159 #endif
2161 // Identify compatability class for VM's architecture and library's architecture
2162 // Obtain string descriptions for architectures
2164 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2165 int running_arch_index=-1;
2167 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2168 if (running_arch_code == arch_array[i].code) {
2169 running_arch_index = i;
2170 }
2171 if (lib_arch.code == arch_array[i].code) {
2172 lib_arch.compat_class = arch_array[i].compat_class;
2173 lib_arch.name = arch_array[i].name;
2174 }
2175 }
2177 assert(running_arch_index != -1,
2178 "Didn't find running architecture code (running_arch_code) in arch_array");
2179 if (running_arch_index == -1) {
2180 // Even though running architecture detection failed
2181 // we may still continue with reporting dlerror() message
2182 return NULL;
2183 }
2185 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2186 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2187 return NULL;
2188 }
2190 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2191 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2192 return NULL;
2193 }
2195 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2196 if ( lib_arch.name!=NULL ) {
2197 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2198 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2199 lib_arch.name, arch_array[running_arch_index].name);
2200 } else {
2201 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2202 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2203 lib_arch.code,
2204 arch_array[running_arch_index].name);
2205 }
2206 }
2208 return NULL;
2209 }
2211 void* os::dll_lookup(void* handle, const char* name) {
2212 return dlsym(handle, name);
2213 }
2215 int os::stat(const char *path, struct stat *sbuf) {
2216 char pathbuf[MAX_PATH];
2217 if (strlen(path) > MAX_PATH - 1) {
2218 errno = ENAMETOOLONG;
2219 return -1;
2220 }
2221 os::native_path(strcpy(pathbuf, path));
2222 return ::stat(pathbuf, sbuf);
2223 }
2225 static bool _print_ascii_file(const char* filename, outputStream* st) {
2226 int fd = ::open(filename, O_RDONLY);
2227 if (fd == -1) {
2228 return false;
2229 }
2231 char buf[32];
2232 int bytes;
2233 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2234 st->print_raw(buf, bytes);
2235 }
2237 ::close(fd);
2239 return true;
2240 }
2242 void os::print_os_info(outputStream* st) {
2243 st->print("OS:");
2245 if (!_print_ascii_file("/etc/release", st)) {
2246 st->print("Solaris");
2247 }
2248 st->cr();
2250 // kernel
2251 st->print("uname:");
2252 struct utsname name;
2253 uname(&name);
2254 st->print(name.sysname); st->print(" ");
2255 st->print(name.release); st->print(" ");
2256 st->print(name.version); st->print(" ");
2257 st->print(name.machine);
2259 // libthread
2260 if (os::Solaris::T2_libthread()) st->print(" (T2 libthread)");
2261 else st->print(" (T1 libthread)");
2262 st->cr();
2264 // rlimit
2265 st->print("rlimit:");
2266 struct rlimit rlim;
2268 st->print(" STACK ");
2269 getrlimit(RLIMIT_STACK, &rlim);
2270 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2271 else st->print("%uk", rlim.rlim_cur >> 10);
2273 st->print(", CORE ");
2274 getrlimit(RLIMIT_CORE, &rlim);
2275 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2276 else st->print("%uk", rlim.rlim_cur >> 10);
2278 st->print(", NOFILE ");
2279 getrlimit(RLIMIT_NOFILE, &rlim);
2280 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2281 else st->print("%d", rlim.rlim_cur);
2283 st->print(", AS ");
2284 getrlimit(RLIMIT_AS, &rlim);
2285 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2286 else st->print("%uk", rlim.rlim_cur >> 10);
2287 st->cr();
2289 // load average
2290 st->print("load average:");
2291 double loadavg[3];
2292 os::loadavg(loadavg, 3);
2293 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2294 st->cr();
2295 }
2298 static bool check_addr0(outputStream* st) {
2299 jboolean status = false;
2300 int fd = ::open("/proc/self/map",O_RDONLY);
2301 if (fd >= 0) {
2302 prmap_t p;
2303 while(::read(fd, &p, sizeof(p)) > 0) {
2304 if (p.pr_vaddr == 0x0) {
2305 st->print("Warning: Address: 0x%x, Size: %dK, ",p.pr_vaddr, p.pr_size/1024, p.pr_mapname);
2306 st->print("Mapped file: %s, ", p.pr_mapname[0] == '\0' ? "None" : p.pr_mapname);
2307 st->print("Access:");
2308 st->print("%s",(p.pr_mflags & MA_READ) ? "r" : "-");
2309 st->print("%s",(p.pr_mflags & MA_WRITE) ? "w" : "-");
2310 st->print("%s",(p.pr_mflags & MA_EXEC) ? "x" : "-");
2311 st->cr();
2312 status = true;
2313 }
2314 ::close(fd);
2315 }
2316 }
2317 return status;
2318 }
2320 void os::print_memory_info(outputStream* st) {
2321 st->print("Memory:");
2322 st->print(" %dk page", os::vm_page_size()>>10);
2323 st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
2324 st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
2325 st->cr();
2326 (void) check_addr0(st);
2327 }
2329 // Taken from /usr/include/sys/machsig.h Supposed to be architecture specific
2330 // but they're the same for all the solaris architectures that we support.
2331 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2332 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2333 "ILL_COPROC", "ILL_BADSTK" };
2335 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2336 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2337 "FPE_FLTINV", "FPE_FLTSUB" };
2339 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2341 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2343 void os::print_siginfo(outputStream* st, void* siginfo) {
2344 st->print("siginfo:");
2346 const int buflen = 100;
2347 char buf[buflen];
2348 siginfo_t *si = (siginfo_t*)siginfo;
2349 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2350 char *err = strerror(si->si_errno);
2351 if (si->si_errno != 0 && err != NULL) {
2352 st->print("si_errno=%s", err);
2353 } else {
2354 st->print("si_errno=%d", si->si_errno);
2355 }
2356 const int c = si->si_code;
2357 assert(c > 0, "unexpected si_code");
2358 switch (si->si_signo) {
2359 case SIGILL:
2360 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2361 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2362 break;
2363 case SIGFPE:
2364 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2365 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2366 break;
2367 case SIGSEGV:
2368 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2369 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2370 break;
2371 case SIGBUS:
2372 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2373 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2374 break;
2375 default:
2376 st->print(", si_code=%d", si->si_code);
2377 // no si_addr
2378 }
2380 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2381 UseSharedSpaces) {
2382 FileMapInfo* mapinfo = FileMapInfo::current_info();
2383 if (mapinfo->is_in_shared_space(si->si_addr)) {
2384 st->print("\n\nError accessing class data sharing archive." \
2385 " Mapped file inaccessible during execution, " \
2386 " possible disk/network problem.");
2387 }
2388 }
2389 st->cr();
2390 }
2392 // Moved from whole group, because we need them here for diagnostic
2393 // prints.
2394 #define OLDMAXSIGNUM 32
2395 static int Maxsignum = 0;
2396 static int *ourSigFlags = NULL;
2398 extern "C" void sigINTRHandler(int, siginfo_t*, void*);
2400 int os::Solaris::get_our_sigflags(int sig) {
2401 assert(ourSigFlags!=NULL, "signal data structure not initialized");
2402 assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2403 return ourSigFlags[sig];
2404 }
2406 void os::Solaris::set_our_sigflags(int sig, int flags) {
2407 assert(ourSigFlags!=NULL, "signal data structure not initialized");
2408 assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2409 ourSigFlags[sig] = flags;
2410 }
2413 static const char* get_signal_handler_name(address handler,
2414 char* buf, int buflen) {
2415 int offset;
2416 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
2417 if (found) {
2418 // skip directory names
2419 const char *p1, *p2;
2420 p1 = buf;
2421 size_t len = strlen(os::file_separator());
2422 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
2423 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
2424 } else {
2425 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
2426 }
2427 return buf;
2428 }
2430 static void print_signal_handler(outputStream* st, int sig,
2431 char* buf, size_t buflen) {
2432 struct sigaction sa;
2434 sigaction(sig, NULL, &sa);
2436 st->print("%s: ", os::exception_name(sig, buf, buflen));
2438 address handler = (sa.sa_flags & SA_SIGINFO)
2439 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
2440 : CAST_FROM_FN_PTR(address, sa.sa_handler);
2442 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
2443 st->print("SIG_DFL");
2444 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
2445 st->print("SIG_IGN");
2446 } else {
2447 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
2448 }
2450 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
2452 address rh = VMError::get_resetted_sighandler(sig);
2453 // May be, handler was resetted by VMError?
2454 if(rh != NULL) {
2455 handler = rh;
2456 sa.sa_flags = VMError::get_resetted_sigflags(sig);
2457 }
2459 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
2461 // Check: is it our handler?
2462 if(handler == CAST_FROM_FN_PTR(address, signalHandler) ||
2463 handler == CAST_FROM_FN_PTR(address, sigINTRHandler)) {
2464 // It is our signal handler
2465 // check for flags
2466 if(sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2467 st->print(
2468 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2469 os::Solaris::get_our_sigflags(sig));
2470 }
2471 }
2472 st->cr();
2473 }
2475 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2476 st->print_cr("Signal Handlers:");
2477 print_signal_handler(st, SIGSEGV, buf, buflen);
2478 print_signal_handler(st, SIGBUS , buf, buflen);
2479 print_signal_handler(st, SIGFPE , buf, buflen);
2480 print_signal_handler(st, SIGPIPE, buf, buflen);
2481 print_signal_handler(st, SIGXFSZ, buf, buflen);
2482 print_signal_handler(st, SIGILL , buf, buflen);
2483 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2484 print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2485 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2486 print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2487 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2488 print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2489 print_signal_handler(st, os::Solaris::SIGinterrupt(), buf, buflen);
2490 print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2491 }
2493 static char saved_jvm_path[MAXPATHLEN] = { 0 };
2495 // Find the full path to the current module, libjvm.so or libjvm_g.so
2496 void os::jvm_path(char *buf, jint buflen) {
2497 // Error checking.
2498 if (buflen < MAXPATHLEN) {
2499 assert(false, "must use a large-enough buffer");
2500 buf[0] = '\0';
2501 return;
2502 }
2503 // Lazy resolve the path to current module.
2504 if (saved_jvm_path[0] != 0) {
2505 strcpy(buf, saved_jvm_path);
2506 return;
2507 }
2509 Dl_info dlinfo;
2510 int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2511 assert(ret != 0, "cannot locate libjvm");
2512 realpath((char *)dlinfo.dli_fname, buf);
2514 if (Arguments::created_by_gamma_launcher()) {
2515 // Support for the gamma launcher. Typical value for buf is
2516 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2517 // the right place in the string, then assume we are installed in a JDK and
2518 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2519 // up the path so it looks like libjvm.so is installed there (append a
2520 // fake suffix hotspot/libjvm.so).
2521 const char *p = buf + strlen(buf) - 1;
2522 for (int count = 0; p > buf && count < 5; ++count) {
2523 for (--p; p > buf && *p != '/'; --p)
2524 /* empty */ ;
2525 }
2527 if (strncmp(p, "/jre/lib/", 9) != 0) {
2528 // Look for JAVA_HOME in the environment.
2529 char* java_home_var = ::getenv("JAVA_HOME");
2530 if (java_home_var != NULL && java_home_var[0] != 0) {
2531 char cpu_arch[12];
2532 char* jrelib_p;
2533 int len;
2534 sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2535 #ifdef _LP64
2536 // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2537 if (strcmp(cpu_arch, "sparc") == 0) {
2538 strcat(cpu_arch, "v9");
2539 } else if (strcmp(cpu_arch, "i386") == 0) {
2540 strcpy(cpu_arch, "amd64");
2541 }
2542 #endif
2543 // Check the current module name "libjvm.so" or "libjvm_g.so".
2544 p = strrchr(buf, '/');
2545 assert(strstr(p, "/libjvm") == p, "invalid library name");
2546 p = strstr(p, "_g") ? "_g" : "";
2548 realpath(java_home_var, buf);
2549 // determine if this is a legacy image or modules image
2550 // modules image doesn't have "jre" subdirectory
2551 len = strlen(buf);
2552 jrelib_p = buf + len;
2553 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2554 if (0 != access(buf, F_OK)) {
2555 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2556 }
2558 if (0 == access(buf, F_OK)) {
2559 // Use current module name "libjvm[_g].so" instead of
2560 // "libjvm"debug_only("_g")".so" since for fastdebug version
2561 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2562 len = strlen(buf);
2563 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2564 } else {
2565 // Go back to path of .so
2566 realpath((char *)dlinfo.dli_fname, buf);
2567 }
2568 }
2569 }
2570 }
2572 strcpy(saved_jvm_path, buf);
2573 }
2576 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2577 // no prefix required, not even "_"
2578 }
2581 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2582 // no suffix required
2583 }
2585 // This method is a copy of JDK's sysGetLastErrorString
2586 // from src/solaris/hpi/src/system_md.c
2588 size_t os::lasterror(char *buf, size_t len) {
2590 if (errno == 0) return 0;
2592 const char *s = ::strerror(errno);
2593 size_t n = ::strlen(s);
2594 if (n >= len) {
2595 n = len - 1;
2596 }
2597 ::strncpy(buf, s, n);
2598 buf[n] = '\0';
2599 return n;
2600 }
2603 // sun.misc.Signal
2605 extern "C" {
2606 static void UserHandler(int sig, void *siginfo, void *context) {
2607 // Ctrl-C is pressed during error reporting, likely because the error
2608 // handler fails to abort. Let VM die immediately.
2609 if (sig == SIGINT && is_error_reported()) {
2610 os::die();
2611 }
2613 os::signal_notify(sig);
2614 // We do not need to reinstate the signal handler each time...
2615 }
2616 }
2618 void* os::user_handler() {
2619 return CAST_FROM_FN_PTR(void*, UserHandler);
2620 }
2622 extern "C" {
2623 typedef void (*sa_handler_t)(int);
2624 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2625 }
2627 void* os::signal(int signal_number, void* handler) {
2628 struct sigaction sigAct, oldSigAct;
2629 sigfillset(&(sigAct.sa_mask));
2630 sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2631 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2633 if (sigaction(signal_number, &sigAct, &oldSigAct))
2634 // -1 means registration failed
2635 return (void *)-1;
2637 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2638 }
2640 void os::signal_raise(int signal_number) {
2641 raise(signal_number);
2642 }
2644 /*
2645 * The following code is moved from os.cpp for making this
2646 * code platform specific, which it is by its very nature.
2647 */
2649 // a counter for each possible signal value
2650 static int Sigexit = 0;
2651 static int Maxlibjsigsigs;
2652 static jint *pending_signals = NULL;
2653 static int *preinstalled_sigs = NULL;
2654 static struct sigaction *chainedsigactions = NULL;
2655 static sema_t sig_sem;
2656 typedef int (*version_getting_t)();
2657 version_getting_t os::Solaris::get_libjsig_version = NULL;
2658 static int libjsigversion = NULL;
2660 int os::sigexitnum_pd() {
2661 assert(Sigexit > 0, "signal memory not yet initialized");
2662 return Sigexit;
2663 }
2665 void os::Solaris::init_signal_mem() {
2666 // Initialize signal structures
2667 Maxsignum = SIGRTMAX;
2668 Sigexit = Maxsignum+1;
2669 assert(Maxsignum >0, "Unable to obtain max signal number");
2671 Maxlibjsigsigs = Maxsignum;
2673 // pending_signals has one int per signal
2674 // The additional signal is for SIGEXIT - exit signal to signal_thread
2675 pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1));
2676 memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2678 if (UseSignalChaining) {
2679 chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2680 * (Maxsignum + 1));
2681 memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2682 preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1));
2683 memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2684 }
2685 ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1 ));
2686 memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2687 }
2689 void os::signal_init_pd() {
2690 int ret;
2692 ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2693 assert(ret == 0, "sema_init() failed");
2694 }
2696 void os::signal_notify(int signal_number) {
2697 int ret;
2699 Atomic::inc(&pending_signals[signal_number]);
2700 ret = ::sema_post(&sig_sem);
2701 assert(ret == 0, "sema_post() failed");
2702 }
2704 static int check_pending_signals(bool wait_for_signal) {
2705 int ret;
2706 while (true) {
2707 for (int i = 0; i < Sigexit + 1; i++) {
2708 jint n = pending_signals[i];
2709 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2710 return i;
2711 }
2712 }
2713 if (!wait_for_signal) {
2714 return -1;
2715 }
2716 JavaThread *thread = JavaThread::current();
2717 ThreadBlockInVM tbivm(thread);
2719 bool threadIsSuspended;
2720 do {
2721 thread->set_suspend_equivalent();
2722 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2723 while((ret = ::sema_wait(&sig_sem)) == EINTR)
2724 ;
2725 assert(ret == 0, "sema_wait() failed");
2727 // were we externally suspended while we were waiting?
2728 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2729 if (threadIsSuspended) {
2730 //
2731 // The semaphore has been incremented, but while we were waiting
2732 // another thread suspended us. We don't want to continue running
2733 // while suspended because that would surprise the thread that
2734 // suspended us.
2735 //
2736 ret = ::sema_post(&sig_sem);
2737 assert(ret == 0, "sema_post() failed");
2739 thread->java_suspend_self();
2740 }
2741 } while (threadIsSuspended);
2742 }
2743 }
2745 int os::signal_lookup() {
2746 return check_pending_signals(false);
2747 }
2749 int os::signal_wait() {
2750 return check_pending_signals(true);
2751 }
2753 ////////////////////////////////////////////////////////////////////////////////
2754 // Virtual Memory
2756 static int page_size = -1;
2758 // The mmap MAP_ALIGN flag is supported on Solaris 9 and later. init_2() will
2759 // clear this var if support is not available.
2760 static bool has_map_align = true;
2762 int os::vm_page_size() {
2763 assert(page_size != -1, "must call os::init");
2764 return page_size;
2765 }
2767 // Solaris allocates memory by pages.
2768 int os::vm_allocation_granularity() {
2769 assert(page_size != -1, "must call os::init");
2770 return page_size;
2771 }
2773 bool os::commit_memory(char* addr, size_t bytes, bool exec) {
2774 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2775 size_t size = bytes;
2776 return
2777 NULL != Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2778 }
2780 bool os::commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2781 bool exec) {
2782 if (commit_memory(addr, bytes, exec)) {
2783 if (UseMPSS && alignment_hint > (size_t)vm_page_size()) {
2784 // If the large page size has been set and the VM
2785 // is using large pages, use the large page size
2786 // if it is smaller than the alignment hint. This is
2787 // a case where the VM wants to use a larger alignment size
2788 // for its own reasons but still want to use large pages
2789 // (which is what matters to setting the mpss range.
2790 size_t page_size = 0;
2791 if (large_page_size() < alignment_hint) {
2792 assert(UseLargePages, "Expected to be here for large page use only");
2793 page_size = large_page_size();
2794 } else {
2795 // If the alignment hint is less than the large page
2796 // size, the VM wants a particular alignment (thus the hint)
2797 // for internal reasons. Try to set the mpss range using
2798 // the alignment_hint.
2799 page_size = alignment_hint;
2800 }
2801 // Since this is a hint, ignore any failures.
2802 (void)Solaris::set_mpss_range(addr, bytes, page_size);
2803 }
2804 return true;
2805 }
2806 return false;
2807 }
2809 // Uncommit the pages in a specified region.
2810 void os::free_memory(char* addr, size_t bytes) {
2811 if (madvise(addr, bytes, MADV_FREE) < 0) {
2812 debug_only(warning("MADV_FREE failed."));
2813 return;
2814 }
2815 }
2817 bool os::create_stack_guard_pages(char* addr, size_t size) {
2818 return os::commit_memory(addr, size);
2819 }
2821 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2822 return os::uncommit_memory(addr, size);
2823 }
2825 // Change the page size in a given range.
2826 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2827 assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2828 assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2829 if (UseLargePages && UseMPSS) {
2830 Solaris::set_mpss_range(addr, bytes, alignment_hint);
2831 }
2832 }
2834 // Tell the OS to make the range local to the first-touching LWP
2835 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2836 assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2837 if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2838 debug_only(warning("MADV_ACCESS_LWP failed."));
2839 }
2840 }
2842 // Tell the OS that this range would be accessed from different LWPs.
2843 void os::numa_make_global(char *addr, size_t bytes) {
2844 assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2845 if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2846 debug_only(warning("MADV_ACCESS_MANY failed."));
2847 }
2848 }
2850 // Get the number of the locality groups.
2851 size_t os::numa_get_groups_num() {
2852 size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2853 return n != -1 ? n : 1;
2854 }
2856 // Get a list of leaf locality groups. A leaf lgroup is group that
2857 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2858 // board. An LWP is assigned to one of these groups upon creation.
2859 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2860 if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2861 ids[0] = 0;
2862 return 1;
2863 }
2864 int result_size = 0, top = 1, bottom = 0, cur = 0;
2865 for (int k = 0; k < size; k++) {
2866 int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2867 (Solaris::lgrp_id_t*)&ids[top], size - top);
2868 if (r == -1) {
2869 ids[0] = 0;
2870 return 1;
2871 }
2872 if (!r) {
2873 // That's a leaf node.
2874 assert (bottom <= cur, "Sanity check");
2875 // Check if the node has memory
2876 if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2877 NULL, 0, LGRP_RSRC_MEM) > 0) {
2878 ids[bottom++] = ids[cur];
2879 }
2880 }
2881 top += r;
2882 cur++;
2883 }
2884 if (bottom == 0) {
2885 // Handle a situation, when the OS reports no memory available.
2886 // Assume UMA architecture.
2887 ids[0] = 0;
2888 return 1;
2889 }
2890 return bottom;
2891 }
2893 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2894 bool os::numa_topology_changed() {
2895 int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2896 if (is_stale != -1 && is_stale) {
2897 Solaris::lgrp_fini(Solaris::lgrp_cookie());
2898 Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2899 assert(c != 0, "Failure to initialize LGRP API");
2900 Solaris::set_lgrp_cookie(c);
2901 return true;
2902 }
2903 return false;
2904 }
2906 // Get the group id of the current LWP.
2907 int os::numa_get_group_id() {
2908 int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2909 if (lgrp_id == -1) {
2910 return 0;
2911 }
2912 const int size = os::numa_get_groups_num();
2913 int *ids = (int*)alloca(size * sizeof(int));
2915 // Get the ids of all lgroups with memory; r is the count.
2916 int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2917 (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2918 if (r <= 0) {
2919 return 0;
2920 }
2921 return ids[os::random() % r];
2922 }
2924 // Request information about the page.
2925 bool os::get_page_info(char *start, page_info* info) {
2926 const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2927 uint64_t addr = (uintptr_t)start;
2928 uint64_t outdata[2];
2929 uint_t validity = 0;
2931 if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2932 return false;
2933 }
2935 info->size = 0;
2936 info->lgrp_id = -1;
2938 if ((validity & 1) != 0) {
2939 if ((validity & 2) != 0) {
2940 info->lgrp_id = outdata[0];
2941 }
2942 if ((validity & 4) != 0) {
2943 info->size = outdata[1];
2944 }
2945 return true;
2946 }
2947 return false;
2948 }
2950 // Scan the pages from start to end until a page different than
2951 // the one described in the info parameter is encountered.
2952 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2953 const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2954 const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2955 uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT];
2956 uint_t validity[MAX_MEMINFO_CNT];
2958 size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2959 uint64_t p = (uint64_t)start;
2960 while (p < (uint64_t)end) {
2961 addrs[0] = p;
2962 size_t addrs_count = 1;
2963 while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] < (uint64_t)end) {
2964 addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2965 addrs_count++;
2966 }
2968 if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2969 return NULL;
2970 }
2972 size_t i = 0;
2973 for (; i < addrs_count; i++) {
2974 if ((validity[i] & 1) != 0) {
2975 if ((validity[i] & 4) != 0) {
2976 if (outdata[types * i + 1] != page_expected->size) {
2977 break;
2978 }
2979 } else
2980 if (page_expected->size != 0) {
2981 break;
2982 }
2984 if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2985 if (outdata[types * i] != page_expected->lgrp_id) {
2986 break;
2987 }
2988 }
2989 } else {
2990 return NULL;
2991 }
2992 }
2994 if (i != addrs_count) {
2995 if ((validity[i] & 2) != 0) {
2996 page_found->lgrp_id = outdata[types * i];
2997 } else {
2998 page_found->lgrp_id = -1;
2999 }
3000 if ((validity[i] & 4) != 0) {
3001 page_found->size = outdata[types * i + 1];
3002 } else {
3003 page_found->size = 0;
3004 }
3005 return (char*)addrs[i];
3006 }
3008 p = addrs[addrs_count - 1] + page_size;
3009 }
3010 return end;
3011 }
3013 bool os::uncommit_memory(char* addr, size_t bytes) {
3014 size_t size = bytes;
3015 // Map uncommitted pages PROT_NONE so we fail early if we touch an
3016 // uncommitted page. Otherwise, the read/write might succeed if we
3017 // have enough swap space to back the physical page.
3018 return
3019 NULL != Solaris::mmap_chunk(addr, size,
3020 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
3021 PROT_NONE);
3022 }
3024 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
3025 char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
3027 if (b == MAP_FAILED) {
3028 return NULL;
3029 }
3030 return b;
3031 }
3033 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes, size_t alignment_hint, bool fixed) {
3034 char* addr = requested_addr;
3035 int flags = MAP_PRIVATE | MAP_NORESERVE;
3037 assert(!(fixed && (alignment_hint > 0)), "alignment hint meaningless with fixed mmap");
3039 if (fixed) {
3040 flags |= MAP_FIXED;
3041 } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
3042 flags |= MAP_ALIGN;
3043 addr = (char*) alignment_hint;
3044 }
3046 // Map uncommitted pages PROT_NONE so we fail early if we touch an
3047 // uncommitted page. Otherwise, the read/write might succeed if we
3048 // have enough swap space to back the physical page.
3049 return mmap_chunk(addr, bytes, flags, PROT_NONE);
3050 }
3052 char* os::reserve_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
3053 char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint, (requested_addr != NULL));
3055 guarantee(requested_addr == NULL || requested_addr == addr,
3056 "OS failed to return requested mmap address.");
3057 return addr;
3058 }
3060 // Reserve memory at an arbitrary address, only if that area is
3061 // available (and not reserved for something else).
3063 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3064 const int max_tries = 10;
3065 char* base[max_tries];
3066 size_t size[max_tries];
3068 // Solaris adds a gap between mmap'ed regions. The size of the gap
3069 // is dependent on the requested size and the MMU. Our initial gap
3070 // value here is just a guess and will be corrected later.
3071 bool had_top_overlap = false;
3072 bool have_adjusted_gap = false;
3073 size_t gap = 0x400000;
3075 // Assert only that the size is a multiple of the page size, since
3076 // that's all that mmap requires, and since that's all we really know
3077 // about at this low abstraction level. If we need higher alignment,
3078 // we can either pass an alignment to this method or verify alignment
3079 // in one of the methods further up the call chain. See bug 5044738.
3080 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3082 // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
3083 // Give it a try, if the kernel honors the hint we can return immediately.
3084 char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
3085 volatile int err = errno;
3086 if (addr == requested_addr) {
3087 return addr;
3088 } else if (addr != NULL) {
3089 unmap_memory(addr, bytes);
3090 }
3092 if (PrintMiscellaneous && Verbose) {
3093 char buf[256];
3094 buf[0] = '\0';
3095 if (addr == NULL) {
3096 jio_snprintf(buf, sizeof(buf), ": %s", strerror(err));
3097 }
3098 warning("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
3099 PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
3100 "%s", bytes, requested_addr, addr, buf);
3101 }
3103 // Address hint method didn't work. Fall back to the old method.
3104 // In theory, once SNV becomes our oldest supported platform, this
3105 // code will no longer be needed.
3106 //
3107 // Repeatedly allocate blocks until the block is allocated at the
3108 // right spot. Give up after max_tries.
3109 int i;
3110 for (i = 0; i < max_tries; ++i) {
3111 base[i] = reserve_memory(bytes);
3113 if (base[i] != NULL) {
3114 // Is this the block we wanted?
3115 if (base[i] == requested_addr) {
3116 size[i] = bytes;
3117 break;
3118 }
3120 // check that the gap value is right
3121 if (had_top_overlap && !have_adjusted_gap) {
3122 size_t actual_gap = base[i-1] - base[i] - bytes;
3123 if (gap != actual_gap) {
3124 // adjust the gap value and retry the last 2 allocations
3125 assert(i > 0, "gap adjustment code problem");
3126 have_adjusted_gap = true; // adjust the gap only once, just in case
3127 gap = actual_gap;
3128 if (PrintMiscellaneous && Verbose) {
3129 warning("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
3130 }
3131 unmap_memory(base[i], bytes);
3132 unmap_memory(base[i-1], size[i-1]);
3133 i-=2;
3134 continue;
3135 }
3136 }
3138 // Does this overlap the block we wanted? Give back the overlapped
3139 // parts and try again.
3140 //
3141 // There is still a bug in this code: if top_overlap == bytes,
3142 // the overlap is offset from requested region by the value of gap.
3143 // In this case giving back the overlapped part will not work,
3144 // because we'll give back the entire block at base[i] and
3145 // therefore the subsequent allocation will not generate a new gap.
3146 // This could be fixed with a new algorithm that used larger
3147 // or variable size chunks to find the requested region -
3148 // but such a change would introduce additional complications.
3149 // It's rare enough that the planets align for this bug,
3150 // so we'll just wait for a fix for 6204603/5003415 which
3151 // will provide a mmap flag to allow us to avoid this business.
3153 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3154 if (top_overlap >= 0 && top_overlap < bytes) {
3155 had_top_overlap = true;
3156 unmap_memory(base[i], top_overlap);
3157 base[i] += top_overlap;
3158 size[i] = bytes - top_overlap;
3159 } else {
3160 size_t bottom_overlap = base[i] + bytes - requested_addr;
3161 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3162 if (PrintMiscellaneous && Verbose && bottom_overlap == 0) {
3163 warning("attempt_reserve_memory_at: possible alignment bug");
3164 }
3165 unmap_memory(requested_addr, bottom_overlap);
3166 size[i] = bytes - bottom_overlap;
3167 } else {
3168 size[i] = bytes;
3169 }
3170 }
3171 }
3172 }
3174 // Give back the unused reserved pieces.
3176 for (int j = 0; j < i; ++j) {
3177 if (base[j] != NULL) {
3178 unmap_memory(base[j], size[j]);
3179 }
3180 }
3182 return (i < max_tries) ? requested_addr : NULL;
3183 }
3185 bool os::release_memory(char* addr, size_t bytes) {
3186 size_t size = bytes;
3187 return munmap(addr, size) == 0;
3188 }
3190 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
3191 assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
3192 "addr must be page aligned");
3193 int retVal = mprotect(addr, bytes, prot);
3194 return retVal == 0;
3195 }
3197 // Protect memory (Used to pass readonly pages through
3198 // JNI GetArray<type>Elements with empty arrays.)
3199 // Also, used for serialization page and for compressed oops null pointer
3200 // checking.
3201 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3202 bool is_committed) {
3203 unsigned int p = 0;
3204 switch (prot) {
3205 case MEM_PROT_NONE: p = PROT_NONE; break;
3206 case MEM_PROT_READ: p = PROT_READ; break;
3207 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3208 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3209 default:
3210 ShouldNotReachHere();
3211 }
3212 // is_committed is unused.
3213 return solaris_mprotect(addr, bytes, p);
3214 }
3216 // guard_memory and unguard_memory only happens within stack guard pages.
3217 // Since ISM pertains only to the heap, guard and unguard memory should not
3218 /// happen with an ISM region.
3219 bool os::guard_memory(char* addr, size_t bytes) {
3220 return solaris_mprotect(addr, bytes, PROT_NONE);
3221 }
3223 bool os::unguard_memory(char* addr, size_t bytes) {
3224 return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
3225 }
3227 // Large page support
3229 // UseLargePages is the master flag to enable/disable large page memory.
3230 // UseMPSS and UseISM are supported for compatibility reasons. Their combined
3231 // effects can be described in the following table:
3232 //
3233 // UseLargePages UseMPSS UseISM
3234 // false * * => UseLargePages is the master switch, turning
3235 // it off will turn off both UseMPSS and
3236 // UseISM. VM will not use large page memory
3237 // regardless the settings of UseMPSS/UseISM.
3238 // true false false => Unless future Solaris provides other
3239 // mechanism to use large page memory, this
3240 // combination is equivalent to -UseLargePages,
3241 // VM will not use large page memory
3242 // true true false => JVM will use MPSS for large page memory.
3243 // This is the default behavior.
3244 // true false true => JVM will use ISM for large page memory.
3245 // true true true => JVM will use ISM if it is available.
3246 // Otherwise, JVM will fall back to MPSS.
3247 // Becaues ISM is now available on all
3248 // supported Solaris versions, this combination
3249 // is equivalent to +UseISM -UseMPSS.
3251 typedef int (*getpagesizes_func_type) (size_t[], int);
3252 static size_t _large_page_size = 0;
3254 bool os::Solaris::ism_sanity_check(bool warn, size_t * page_size) {
3255 // x86 uses either 2M or 4M page, depending on whether PAE (Physical Address
3256 // Extensions) mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. Sparc
3257 // can support multiple page sizes.
3259 // Don't bother to probe page size because getpagesizes() comes with MPSS.
3260 // ISM is only recommended on old Solaris where there is no MPSS support.
3261 // Simply choose a conservative value as default.
3262 *page_size = LargePageSizeInBytes ? LargePageSizeInBytes :
3263 SPARC_ONLY(4 * M) IA32_ONLY(4 * M) AMD64_ONLY(2 * M)
3264 ARM_ONLY(2 * M);
3266 // ISM is available on all supported Solaris versions
3267 return true;
3268 }
3270 // Insertion sort for small arrays (descending order).
3271 static void insertion_sort_descending(size_t* array, int len) {
3272 for (int i = 0; i < len; i++) {
3273 size_t val = array[i];
3274 for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
3275 size_t tmp = array[key];
3276 array[key] = array[key - 1];
3277 array[key - 1] = tmp;
3278 }
3279 }
3280 }
3282 bool os::Solaris::mpss_sanity_check(bool warn, size_t * page_size) {
3283 getpagesizes_func_type getpagesizes_func =
3284 CAST_TO_FN_PTR(getpagesizes_func_type, dlsym(RTLD_DEFAULT, "getpagesizes"));
3285 if (getpagesizes_func == NULL) {
3286 if (warn) {
3287 warning("MPSS is not supported by the operating system.");
3288 }
3289 return false;
3290 }
3292 const unsigned int usable_count = VM_Version::page_size_count();
3293 if (usable_count == 1) {
3294 return false;
3295 }
3297 // Fill the array of page sizes.
3298 int n = getpagesizes_func(_page_sizes, page_sizes_max);
3299 assert(n > 0, "Solaris bug?");
3300 if (n == page_sizes_max) {
3301 // Add a sentinel value (necessary only if the array was completely filled
3302 // since it is static (zeroed at initialization)).
3303 _page_sizes[--n] = 0;
3304 DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
3305 }
3306 assert(_page_sizes[n] == 0, "missing sentinel");
3308 if (n == 1) return false; // Only one page size available.
3310 // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
3311 // select up to usable_count elements. First sort the array, find the first
3312 // acceptable value, then copy the usable sizes to the top of the array and
3313 // trim the rest. Make sure to include the default page size :-).
3314 //
3315 // A better policy could get rid of the 4M limit by taking the sizes of the
3316 // important VM memory regions (java heap and possibly the code cache) into
3317 // account.
3318 insertion_sort_descending(_page_sizes, n);
3319 const size_t size_limit =
3320 FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
3321 int beg;
3322 for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */ ;
3323 const int end = MIN2((int)usable_count, n) - 1;
3324 for (int cur = 0; cur < end; ++cur, ++beg) {
3325 _page_sizes[cur] = _page_sizes[beg];
3326 }
3327 _page_sizes[end] = vm_page_size();
3328 _page_sizes[end + 1] = 0;
3330 if (_page_sizes[end] > _page_sizes[end - 1]) {
3331 // Default page size is not the smallest; sort again.
3332 insertion_sort_descending(_page_sizes, end + 1);
3333 }
3334 *page_size = _page_sizes[0];
3336 return true;
3337 }
3339 bool os::large_page_init() {
3340 if (!UseLargePages) {
3341 UseISM = false;
3342 UseMPSS = false;
3343 return false;
3344 }
3346 // print a warning if any large page related flag is specified on command line
3347 bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages) ||
3348 !FLAG_IS_DEFAULT(UseISM) ||
3349 !FLAG_IS_DEFAULT(UseMPSS) ||
3350 !FLAG_IS_DEFAULT(LargePageSizeInBytes);
3351 UseISM = UseISM &&
3352 Solaris::ism_sanity_check(warn_on_failure, &_large_page_size);
3353 if (UseISM) {
3354 // ISM disables MPSS to be compatible with old JDK behavior
3355 UseMPSS = false;
3356 _page_sizes[0] = _large_page_size;
3357 _page_sizes[1] = vm_page_size();
3358 }
3360 UseMPSS = UseMPSS &&
3361 Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
3363 UseLargePages = UseISM || UseMPSS;
3364 return UseLargePages;
3365 }
3367 bool os::Solaris::set_mpss_range(caddr_t start, size_t bytes, size_t align) {
3368 // Signal to OS that we want large pages for addresses
3369 // from addr, addr + bytes
3370 struct memcntl_mha mpss_struct;
3371 mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
3372 mpss_struct.mha_pagesize = align;
3373 mpss_struct.mha_flags = 0;
3374 if (memcntl(start, bytes, MC_HAT_ADVISE,
3375 (caddr_t) &mpss_struct, 0, 0) < 0) {
3376 debug_only(warning("Attempt to use MPSS failed."));
3377 return false;
3378 }
3379 return true;
3380 }
3382 char* os::reserve_memory_special(size_t bytes, char* addr, bool exec) {
3383 // "exec" is passed in but not used. Creating the shared image for
3384 // the code cache doesn't have an SHM_X executable permission to check.
3385 assert(UseLargePages && UseISM, "only for ISM large pages");
3387 size_t size = bytes;
3388 char* retAddr = NULL;
3389 int shmid;
3390 key_t ismKey;
3392 bool warn_on_failure = UseISM &&
3393 (!FLAG_IS_DEFAULT(UseLargePages) ||
3394 !FLAG_IS_DEFAULT(UseISM) ||
3395 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3396 );
3397 char msg[128];
3399 ismKey = IPC_PRIVATE;
3401 // Create a large shared memory region to attach to based on size.
3402 // Currently, size is the total size of the heap
3403 shmid = shmget(ismKey, size, SHM_R | SHM_W | IPC_CREAT);
3404 if (shmid == -1){
3405 if (warn_on_failure) {
3406 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3407 warning(msg);
3408 }
3409 return NULL;
3410 }
3412 // Attach to the region
3413 retAddr = (char *) shmat(shmid, 0, SHM_SHARE_MMU | SHM_R | SHM_W);
3414 int err = errno;
3416 // Remove shmid. If shmat() is successful, the actual shared memory segment
3417 // will be deleted when it's detached by shmdt() or when the process
3418 // terminates. If shmat() is not successful this will remove the shared
3419 // segment immediately.
3420 shmctl(shmid, IPC_RMID, NULL);
3422 if (retAddr == (char *) -1) {
3423 if (warn_on_failure) {
3424 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3425 warning(msg);
3426 }
3427 return NULL;
3428 }
3430 return retAddr;
3431 }
3433 bool os::release_memory_special(char* base, size_t bytes) {
3434 // detaching the SHM segment will also delete it, see reserve_memory_special()
3435 int rslt = shmdt(base);
3436 return rslt == 0;
3437 }
3439 size_t os::large_page_size() {
3440 return _large_page_size;
3441 }
3443 // MPSS allows application to commit large page memory on demand; with ISM
3444 // the entire memory region must be allocated as shared memory.
3445 bool os::can_commit_large_page_memory() {
3446 return UseISM ? false : true;
3447 }
3449 bool os::can_execute_large_page_memory() {
3450 return UseISM ? false : true;
3451 }
3453 static int os_sleep(jlong millis, bool interruptible) {
3454 const jlong limit = INT_MAX;
3455 jlong prevtime;
3456 int res;
3458 while (millis > limit) {
3459 if ((res = os_sleep(limit, interruptible)) != OS_OK)
3460 return res;
3461 millis -= limit;
3462 }
3464 // Restart interrupted polls with new parameters until the proper delay
3465 // has been completed.
3467 prevtime = getTimeMillis();
3469 while (millis > 0) {
3470 jlong newtime;
3472 if (!interruptible) {
3473 // Following assert fails for os::yield_all:
3474 // assert(!thread->is_Java_thread(), "must not be java thread");
3475 res = poll(NULL, 0, millis);
3476 } else {
3477 JavaThread *jt = JavaThread::current();
3479 INTERRUPTIBLE_NORESTART_VM_ALWAYS(poll(NULL, 0, millis), res, jt,
3480 os::Solaris::clear_interrupted);
3481 }
3483 // INTERRUPTIBLE_NORESTART_VM_ALWAYS returns res == OS_INTRPT for
3484 // thread.Interrupt.
3486 // See c/r 6751923. Poll can return 0 before time
3487 // has elapsed if time is set via clock_settime (as NTP does).
3488 // res == 0 if poll timed out (see man poll RETURN VALUES)
3489 // using the logic below checks that we really did
3490 // sleep at least "millis" if not we'll sleep again.
3491 if( ( res == 0 ) || ((res == OS_ERR) && (errno == EINTR))) {
3492 newtime = getTimeMillis();
3493 assert(newtime >= prevtime, "time moving backwards");
3494 /* Doing prevtime and newtime in microseconds doesn't help precision,
3495 and trying to round up to avoid lost milliseconds can result in a
3496 too-short delay. */
3497 millis -= newtime - prevtime;
3498 if(millis <= 0)
3499 return OS_OK;
3500 prevtime = newtime;
3501 } else
3502 return res;
3503 }
3505 return OS_OK;
3506 }
3508 // Read calls from inside the vm need to perform state transitions
3509 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3510 INTERRUPTIBLE_RETURN_INT_VM(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3511 }
3513 size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
3514 INTERRUPTIBLE_RETURN_INT(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3515 }
3517 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3518 assert(thread == Thread::current(), "thread consistency check");
3520 // TODO-FIXME: this should be removed.
3521 // On Solaris machines (especially 2.5.1) we found that sometimes the VM gets into a live lock
3522 // situation with a JavaThread being starved out of a lwp. The kernel doesn't seem to generate
3523 // a SIGWAITING signal which would enable the threads library to create a new lwp for the starving
3524 // thread. We suspect that because the Watcher thread keeps waking up at periodic intervals the kernel
3525 // is fooled into believing that the system is making progress. In the code below we block the
3526 // the watcher thread while safepoint is in progress so that it would not appear as though the
3527 // system is making progress.
3528 if (!Solaris::T2_libthread() &&
3529 thread->is_Watcher_thread() && SafepointSynchronize::is_synchronizing() && !Arguments::has_profile()) {
3530 // We now try to acquire the threads lock. Since this lock is held by the VM thread during
3531 // the entire safepoint, the watcher thread will line up here during the safepoint.
3532 Threads_lock->lock_without_safepoint_check();
3533 Threads_lock->unlock();
3534 }
3536 if (thread->is_Java_thread()) {
3537 // This is a JavaThread so we honor the _thread_blocked protocol
3538 // even for sleeps of 0 milliseconds. This was originally done
3539 // as a workaround for bug 4338139. However, now we also do it
3540 // to honor the suspend-equivalent protocol.
3542 JavaThread *jt = (JavaThread *) thread;
3543 ThreadBlockInVM tbivm(jt);
3545 jt->set_suspend_equivalent();
3546 // cleared by handle_special_suspend_equivalent_condition() or
3547 // java_suspend_self() via check_and_wait_while_suspended()
3549 int ret_code;
3550 if (millis <= 0) {
3551 thr_yield();
3552 ret_code = 0;
3553 } else {
3554 // The original sleep() implementation did not create an
3555 // OSThreadWaitState helper for sleeps of 0 milliseconds.
3556 // I'm preserving that decision for now.
3557 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3559 ret_code = os_sleep(millis, interruptible);
3560 }
3562 // were we externally suspended while we were waiting?
3563 jt->check_and_wait_while_suspended();
3565 return ret_code;
3566 }
3568 // non-JavaThread from this point on:
3570 if (millis <= 0) {
3571 thr_yield();
3572 return 0;
3573 }
3575 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3577 return os_sleep(millis, interruptible);
3578 }
3580 int os::naked_sleep() {
3581 // %% make the sleep time an integer flag. for now use 1 millisec.
3582 return os_sleep(1, false);
3583 }
3585 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3586 void os::infinite_sleep() {
3587 while (true) { // sleep forever ...
3588 ::sleep(100); // ... 100 seconds at a time
3589 }
3590 }
3592 // Used to convert frequent JVM_Yield() to nops
3593 bool os::dont_yield() {
3594 if (DontYieldALot) {
3595 static hrtime_t last_time = 0;
3596 hrtime_t diff = getTimeNanos() - last_time;
3598 if (diff < DontYieldALotInterval * 1000000)
3599 return true;
3601 last_time += diff;
3603 return false;
3604 }
3605 else {
3606 return false;
3607 }
3608 }
3610 // Caveat: Solaris os::yield() causes a thread-state transition whereas
3611 // the linux and win32 implementations do not. This should be checked.
3613 void os::yield() {
3614 // Yields to all threads with same or greater priority
3615 os::sleep(Thread::current(), 0, false);
3616 }
3618 // Note that yield semantics are defined by the scheduling class to which
3619 // the thread currently belongs. Typically, yield will _not yield to
3620 // other equal or higher priority threads that reside on the dispatch queues
3621 // of other CPUs.
3623 os::YieldResult os::NakedYield() { thr_yield(); return os::YIELD_UNKNOWN; }
3626 // On Solaris we found that yield_all doesn't always yield to all other threads.
3627 // There have been cases where there is a thread ready to execute but it doesn't
3628 // get an lwp as the VM thread continues to spin with sleeps of 1 millisecond.
3629 // The 1 millisecond wait doesn't seem long enough for the kernel to issue a
3630 // SIGWAITING signal which will cause a new lwp to be created. So we count the
3631 // number of times yield_all is called in the one loop and increase the sleep
3632 // time after 8 attempts. If this fails too we increase the concurrency level
3633 // so that the starving thread would get an lwp
3635 void os::yield_all(int attempts) {
3636 // Yields to all threads, including threads with lower priorities
3637 if (attempts == 0) {
3638 os::sleep(Thread::current(), 1, false);
3639 } else {
3640 int iterations = attempts % 30;
3641 if (iterations == 0 && !os::Solaris::T2_libthread()) {
3642 // thr_setconcurrency and _getconcurrency make sense only under T1.
3643 int noofLWPS = thr_getconcurrency();
3644 if (noofLWPS < (Threads::number_of_threads() + 2)) {
3645 thr_setconcurrency(thr_getconcurrency() + 1);
3646 }
3647 } else if (iterations < 25) {
3648 os::sleep(Thread::current(), 1, false);
3649 } else {
3650 os::sleep(Thread::current(), 10, false);
3651 }
3652 }
3653 }
3655 // Called from the tight loops to possibly influence time-sharing heuristics
3656 void os::loop_breaker(int attempts) {
3657 os::yield_all(attempts);
3658 }
3661 // Interface for setting lwp priorities. If we are using T2 libthread,
3662 // which forces the use of BoundThreads or we manually set UseBoundThreads,
3663 // all of our threads will be assigned to real lwp's. Using the thr_setprio
3664 // function is meaningless in this mode so we must adjust the real lwp's priority
3665 // The routines below implement the getting and setting of lwp priorities.
3666 //
3667 // Note: There are three priority scales used on Solaris. Java priotities
3668 // which range from 1 to 10, libthread "thr_setprio" scale which range
3669 // from 0 to 127, and the current scheduling class of the process we
3670 // are running in. This is typically from -60 to +60.
3671 // The setting of the lwp priorities in done after a call to thr_setprio
3672 // so Java priorities are mapped to libthread priorities and we map from
3673 // the latter to lwp priorities. We don't keep priorities stored in
3674 // Java priorities since some of our worker threads want to set priorities
3675 // higher than all Java threads.
3676 //
3677 // For related information:
3678 // (1) man -s 2 priocntl
3679 // (2) man -s 4 priocntl
3680 // (3) man dispadmin
3681 // = librt.so
3682 // = libthread/common/rtsched.c - thrp_setlwpprio().
3683 // = ps -cL <pid> ... to validate priority.
3684 // = sched_get_priority_min and _max
3685 // pthread_create
3686 // sched_setparam
3687 // pthread_setschedparam
3688 //
3689 // Assumptions:
3690 // + We assume that all threads in the process belong to the same
3691 // scheduling class. IE. an homogenous process.
3692 // + Must be root or in IA group to change change "interactive" attribute.
3693 // Priocntl() will fail silently. The only indication of failure is when
3694 // we read-back the value and notice that it hasn't changed.
3695 // + Interactive threads enter the runq at the head, non-interactive at the tail.
3696 // + For RT, change timeslice as well. Invariant:
3697 // constant "priority integral"
3698 // Konst == TimeSlice * (60-Priority)
3699 // Given a priority, compute appropriate timeslice.
3700 // + Higher numerical values have higher priority.
3702 // sched class attributes
3703 typedef struct {
3704 int schedPolicy; // classID
3705 int maxPrio;
3706 int minPrio;
3707 } SchedInfo;
3710 static SchedInfo tsLimits, iaLimits, rtLimits;
3712 #ifdef ASSERT
3713 static int ReadBackValidate = 1;
3714 #endif
3715 static int myClass = 0;
3716 static int myMin = 0;
3717 static int myMax = 0;
3718 static int myCur = 0;
3719 static bool priocntl_enable = false;
3722 // Call the version of priocntl suitable for all supported versions
3723 // of Solaris. We need to call through this wrapper so that we can
3724 // build on Solaris 9 and run on Solaris 8, 9 and 10.
3725 //
3726 // This code should be removed if we ever stop supporting Solaris 8
3727 // and earlier releases.
3729 static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3730 typedef long (*priocntl_type)(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3731 static priocntl_type priocntl_ptr = priocntl_stub;
3733 // Stub to set the value of the real pointer, and then call the real
3734 // function.
3736 static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg) {
3737 // Try Solaris 8- name only.
3738 priocntl_type tmp = (priocntl_type)dlsym(RTLD_DEFAULT, "__priocntl");
3739 guarantee(tmp != NULL, "priocntl function not found.");
3740 priocntl_ptr = tmp;
3741 return (*priocntl_ptr)(PC_VERSION, idtype, id, cmd, arg);
3742 }
3745 // lwp_priocntl_init
3746 //
3747 // Try to determine the priority scale for our process.
3748 //
3749 // Return errno or 0 if OK.
3750 //
3751 static
3752 int lwp_priocntl_init ()
3753 {
3754 int rslt;
3755 pcinfo_t ClassInfo;
3756 pcparms_t ParmInfo;
3757 int i;
3759 if (!UseThreadPriorities) return 0;
3761 // We are using Bound threads, we need to determine our priority ranges
3762 if (os::Solaris::T2_libthread() || UseBoundThreads) {
3763 // If ThreadPriorityPolicy is 1, switch tables
3764 if (ThreadPriorityPolicy == 1) {
3765 for (i = 0 ; i < MaxPriority+1; i++)
3766 os::java_to_os_priority[i] = prio_policy1[i];
3767 }
3768 }
3769 // Not using Bound Threads, set to ThreadPolicy 1
3770 else {
3771 for ( i = 0 ; i < MaxPriority+1; i++ ) {
3772 os::java_to_os_priority[i] = prio_policy1[i];
3773 }
3774 return 0;
3775 }
3778 // Get IDs for a set of well-known scheduling classes.
3779 // TODO-FIXME: GETCLINFO returns the current # of classes in the
3780 // the system. We should have a loop that iterates over the
3781 // classID values, which are known to be "small" integers.
3783 strcpy(ClassInfo.pc_clname, "TS");
3784 ClassInfo.pc_cid = -1;
3785 rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3786 if (rslt < 0) return errno;
3787 assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3788 tsLimits.schedPolicy = ClassInfo.pc_cid;
3789 tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3790 tsLimits.minPrio = -tsLimits.maxPrio;
3792 strcpy(ClassInfo.pc_clname, "IA");
3793 ClassInfo.pc_cid = -1;
3794 rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3795 if (rslt < 0) return errno;
3796 assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3797 iaLimits.schedPolicy = ClassInfo.pc_cid;
3798 iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3799 iaLimits.minPrio = -iaLimits.maxPrio;
3801 strcpy(ClassInfo.pc_clname, "RT");
3802 ClassInfo.pc_cid = -1;
3803 rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3804 if (rslt < 0) return errno;
3805 assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3806 rtLimits.schedPolicy = ClassInfo.pc_cid;
3807 rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3808 rtLimits.minPrio = 0;
3811 // Query our "current" scheduling class.
3812 // This will normally be IA,TS or, rarely, RT.
3813 memset (&ParmInfo, 0, sizeof(ParmInfo));
3814 ParmInfo.pc_cid = PC_CLNULL;
3815 rslt = (*priocntl_ptr) (PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo );
3816 if ( rslt < 0 ) return errno;
3817 myClass = ParmInfo.pc_cid;
3819 // We now know our scheduling classId, get specific information
3820 // the class.
3821 ClassInfo.pc_cid = myClass;
3822 ClassInfo.pc_clname[0] = 0;
3823 rslt = (*priocntl_ptr) (PC_VERSION, (idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo );
3824 if ( rslt < 0 ) return errno;
3826 if (ThreadPriorityVerbose)
3827 tty->print_cr ("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3829 memset(&ParmInfo, 0, sizeof(pcparms_t));
3830 ParmInfo.pc_cid = PC_CLNULL;
3831 rslt = (*priocntl_ptr)(PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3832 if (rslt < 0) return errno;
3834 if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3835 myMin = rtLimits.minPrio;
3836 myMax = rtLimits.maxPrio;
3837 } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3838 iaparms_t *iaInfo = (iaparms_t*)ParmInfo.pc_clparms;
3839 myMin = iaLimits.minPrio;
3840 myMax = iaLimits.maxPrio;
3841 myMax = MIN2(myMax, (int)iaInfo->ia_uprilim); // clamp - restrict
3842 } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3843 tsparms_t *tsInfo = (tsparms_t*)ParmInfo.pc_clparms;
3844 myMin = tsLimits.minPrio;
3845 myMax = tsLimits.maxPrio;
3846 myMax = MIN2(myMax, (int)tsInfo->ts_uprilim); // clamp - restrict
3847 } else {
3848 // No clue - punt
3849 if (ThreadPriorityVerbose)
3850 tty->print_cr ("Unknown scheduling class: %s ... \n", ClassInfo.pc_clname);
3851 return EINVAL; // no clue, punt
3852 }
3854 if (ThreadPriorityVerbose)
3855 tty->print_cr ("Thread priority Range: [%d..%d]\n", myMin, myMax);
3857 priocntl_enable = true; // Enable changing priorities
3858 return 0;
3859 }
3861 #define IAPRI(x) ((iaparms_t *)((x).pc_clparms))
3862 #define RTPRI(x) ((rtparms_t *)((x).pc_clparms))
3863 #define TSPRI(x) ((tsparms_t *)((x).pc_clparms))
3866 // scale_to_lwp_priority
3867 //
3868 // Convert from the libthread "thr_setprio" scale to our current
3869 // lwp scheduling class scale.
3870 //
3871 static
3872 int scale_to_lwp_priority (int rMin, int rMax, int x)
3873 {
3874 int v;
3876 if (x == 127) return rMax; // avoid round-down
3877 v = (((x*(rMax-rMin)))/128)+rMin;
3878 return v;
3879 }
3882 // set_lwp_priority
3883 //
3884 // Set the priority of the lwp. This call should only be made
3885 // when using bound threads (T2 threads are bound by default).
3886 //
3887 int set_lwp_priority (int ThreadID, int lwpid, int newPrio )
3888 {
3889 int rslt;
3890 int Actual, Expected, prv;
3891 pcparms_t ParmInfo; // for GET-SET
3892 #ifdef ASSERT
3893 pcparms_t ReadBack; // for readback
3894 #endif
3896 // Set priority via PC_GETPARMS, update, PC_SETPARMS
3897 // Query current values.
3898 // TODO: accelerate this by eliminating the PC_GETPARMS call.
3899 // Cache "pcparms_t" in global ParmCache.
3900 // TODO: elide set-to-same-value
3902 // If something went wrong on init, don't change priorities.
3903 if ( !priocntl_enable ) {
3904 if (ThreadPriorityVerbose)
3905 tty->print_cr("Trying to set priority but init failed, ignoring");
3906 return EINVAL;
3907 }
3910 // If lwp hasn't started yet, just return
3911 // the _start routine will call us again.
3912 if ( lwpid <= 0 ) {
3913 if (ThreadPriorityVerbose) {
3914 tty->print_cr ("deferring the set_lwp_priority of thread " INTPTR_FORMAT " to %d, lwpid not set",
3915 ThreadID, newPrio);
3916 }
3917 return 0;
3918 }
3920 if (ThreadPriorityVerbose) {
3921 tty->print_cr ("set_lwp_priority(" INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3922 ThreadID, lwpid, newPrio);
3923 }
3925 memset(&ParmInfo, 0, sizeof(pcparms_t));
3926 ParmInfo.pc_cid = PC_CLNULL;
3927 rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3928 if (rslt < 0) return errno;
3930 if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3931 rtparms_t *rtInfo = (rtparms_t*)ParmInfo.pc_clparms;
3932 rtInfo->rt_pri = scale_to_lwp_priority (rtLimits.minPrio, rtLimits.maxPrio, newPrio);
3933 rtInfo->rt_tqsecs = RT_NOCHANGE;
3934 rtInfo->rt_tqnsecs = RT_NOCHANGE;
3935 if (ThreadPriorityVerbose) {
3936 tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3937 }
3938 } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3939 iaparms_t *iaInfo = (iaparms_t*)ParmInfo.pc_clparms;
3940 int maxClamped = MIN2(iaLimits.maxPrio, (int)iaInfo->ia_uprilim);
3941 iaInfo->ia_upri = scale_to_lwp_priority(iaLimits.minPrio, maxClamped, newPrio);
3942 iaInfo->ia_uprilim = IA_NOCHANGE;
3943 iaInfo->ia_mode = IA_NOCHANGE;
3944 if (ThreadPriorityVerbose) {
3945 tty->print_cr ("IA: [%d...%d] %d->%d\n",
3946 iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3947 }
3948 } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3949 tsparms_t *tsInfo = (tsparms_t*)ParmInfo.pc_clparms;
3950 int maxClamped = MIN2(tsLimits.maxPrio, (int)tsInfo->ts_uprilim);
3951 prv = tsInfo->ts_upri;
3952 tsInfo->ts_upri = scale_to_lwp_priority(tsLimits.minPrio, maxClamped, newPrio);
3953 tsInfo->ts_uprilim = IA_NOCHANGE;
3954 if (ThreadPriorityVerbose) {
3955 tty->print_cr ("TS: %d [%d...%d] %d->%d\n",
3956 prv, tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3957 }
3958 if (prv == tsInfo->ts_upri) return 0;
3959 } else {
3960 if ( ThreadPriorityVerbose ) {
3961 tty->print_cr ("Unknown scheduling class\n");
3962 }
3963 return EINVAL; // no clue, punt
3964 }
3966 rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3967 if (ThreadPriorityVerbose && rslt) {
3968 tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3969 }
3970 if (rslt < 0) return errno;
3972 #ifdef ASSERT
3973 // Sanity check: read back what we just attempted to set.
3974 // In theory it could have changed in the interim ...
3975 //
3976 // The priocntl system call is tricky.
3977 // Sometimes it'll validate the priority value argument and
3978 // return EINVAL if unhappy. At other times it fails silently.
3979 // Readbacks are prudent.
3981 if (!ReadBackValidate) return 0;
3983 memset(&ReadBack, 0, sizeof(pcparms_t));
3984 ReadBack.pc_cid = PC_CLNULL;
3985 rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3986 assert(rslt >= 0, "priocntl failed");
3987 Actual = Expected = 0xBAD;
3988 assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3989 if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3990 Actual = RTPRI(ReadBack)->rt_pri;
3991 Expected = RTPRI(ParmInfo)->rt_pri;
3992 } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3993 Actual = IAPRI(ReadBack)->ia_upri;
3994 Expected = IAPRI(ParmInfo)->ia_upri;
3995 } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3996 Actual = TSPRI(ReadBack)->ts_upri;
3997 Expected = TSPRI(ParmInfo)->ts_upri;
3998 } else {
3999 if ( ThreadPriorityVerbose ) {
4000 tty->print_cr("set_lwp_priority: unexpected class in readback: %d\n", ParmInfo.pc_cid);
4001 }
4002 }
4004 if (Actual != Expected) {
4005 if ( ThreadPriorityVerbose ) {
4006 tty->print_cr ("set_lwp_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
4007 lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
4008 }
4009 }
4010 #endif
4012 return 0;
4013 }
4017 // Solaris only gives access to 128 real priorities at a time,
4018 // so we expand Java's ten to fill this range. This would be better
4019 // if we dynamically adjusted relative priorities.
4020 //
4021 // The ThreadPriorityPolicy option allows us to select 2 different
4022 // priority scales.
4023 //
4024 // ThreadPriorityPolicy=0
4025 // Since the Solaris' default priority is MaximumPriority, we do not
4026 // set a priority lower than Max unless a priority lower than
4027 // NormPriority is requested.
4028 //
4029 // ThreadPriorityPolicy=1
4030 // This mode causes the priority table to get filled with
4031 // linear values. NormPriority get's mapped to 50% of the
4032 // Maximum priority an so on. This will cause VM threads
4033 // to get unfair treatment against other Solaris processes
4034 // which do not explicitly alter their thread priorities.
4035 //
4038 int os::java_to_os_priority[MaxPriority + 1] = {
4039 -99999, // 0 Entry should never be used
4041 0, // 1 MinPriority
4042 32, // 2
4043 64, // 3
4045 96, // 4
4046 127, // 5 NormPriority
4047 127, // 6
4049 127, // 7
4050 127, // 8
4051 127, // 9 NearMaxPriority
4053 127 // 10 MaxPriority
4054 };
4057 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4058 assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
4059 if ( !UseThreadPriorities ) return OS_OK;
4060 int status = thr_setprio(thread->osthread()->thread_id(), newpri);
4061 if ( os::Solaris::T2_libthread() || (UseBoundThreads && thread->osthread()->is_vm_created()) )
4062 status |= (set_lwp_priority (thread->osthread()->thread_id(),
4063 thread->osthread()->lwp_id(), newpri ));
4064 return (status == 0) ? OS_OK : OS_ERR;
4065 }
4068 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4069 int p;
4070 if ( !UseThreadPriorities ) {
4071 *priority_ptr = NormalPriority;
4072 return OS_OK;
4073 }
4074 int status = thr_getprio(thread->osthread()->thread_id(), &p);
4075 if (status != 0) {
4076 return OS_ERR;
4077 }
4078 *priority_ptr = p;
4079 return OS_OK;
4080 }
4083 // Hint to the underlying OS that a task switch would not be good.
4084 // Void return because it's a hint and can fail.
4085 void os::hint_no_preempt() {
4086 schedctl_start(schedctl_init());
4087 }
4089 void os::interrupt(Thread* thread) {
4090 assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4092 OSThread* osthread = thread->osthread();
4094 int isInterrupted = osthread->interrupted();
4095 if (!isInterrupted) {
4096 osthread->set_interrupted(true);
4097 OrderAccess::fence();
4098 // os::sleep() is implemented with either poll (NULL,0,timeout) or
4099 // by parking on _SleepEvent. If the former, thr_kill will unwedge
4100 // the sleeper by SIGINTR, otherwise the unpark() will wake the sleeper.
4101 ParkEvent * const slp = thread->_SleepEvent ;
4102 if (slp != NULL) slp->unpark() ;
4103 }
4105 // For JSR166: unpark after setting status but before thr_kill -dl
4106 if (thread->is_Java_thread()) {
4107 ((JavaThread*)thread)->parker()->unpark();
4108 }
4110 // Handle interruptible wait() ...
4111 ParkEvent * const ev = thread->_ParkEvent ;
4112 if (ev != NULL) ev->unpark() ;
4114 // When events are used everywhere for os::sleep, then this thr_kill
4115 // will only be needed if UseVMInterruptibleIO is true.
4117 if (!isInterrupted) {
4118 int status = thr_kill(osthread->thread_id(), os::Solaris::SIGinterrupt());
4119 assert_status(status == 0, status, "thr_kill");
4121 // Bump thread interruption counter
4122 RuntimeService::record_thread_interrupt_signaled_count();
4123 }
4124 }
4127 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4128 assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4130 OSThread* osthread = thread->osthread();
4132 bool res = osthread->interrupted();
4134 // NOTE that since there is no "lock" around these two operations,
4135 // there is the possibility that the interrupted flag will be
4136 // "false" but that the interrupt event will be set. This is
4137 // intentional. The effect of this is that Object.wait() will appear
4138 // to have a spurious wakeup, which is not harmful, and the
4139 // possibility is so rare that it is not worth the added complexity
4140 // to add yet another lock. It has also been recommended not to put
4141 // the interrupted flag into the os::Solaris::Event structure,
4142 // because it hides the issue.
4143 if (res && clear_interrupted) {
4144 osthread->set_interrupted(false);
4145 }
4146 return res;
4147 }
4150 void os::print_statistics() {
4151 }
4153 int os::message_box(const char* title, const char* message) {
4154 int i;
4155 fdStream err(defaultStream::error_fd());
4156 for (i = 0; i < 78; i++) err.print_raw("=");
4157 err.cr();
4158 err.print_raw_cr(title);
4159 for (i = 0; i < 78; i++) err.print_raw("-");
4160 err.cr();
4161 err.print_raw_cr(message);
4162 for (i = 0; i < 78; i++) err.print_raw("=");
4163 err.cr();
4165 char buf[16];
4166 // Prevent process from exiting upon "read error" without consuming all CPU
4167 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4169 return buf[0] == 'y' || buf[0] == 'Y';
4170 }
4172 // A lightweight implementation that does not suspend the target thread and
4173 // thus returns only a hint. Used for profiling only!
4174 ExtendedPC os::get_thread_pc(Thread* thread) {
4175 // Make sure that it is called by the watcher and the Threads lock is owned.
4176 assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
4177 // For now, is only used to profile the VM Thread
4178 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4179 ExtendedPC epc;
4181 GetThreadPC_Callback cb(ProfileVM_lock);
4182 OSThread *osthread = thread->osthread();
4183 const int time_to_wait = 400; // 400ms wait for initial response
4184 int status = cb.interrupt(thread, time_to_wait);
4186 if (cb.is_done() ) {
4187 epc = cb.addr();
4188 } else {
4189 DEBUG_ONLY(tty->print_cr("Failed to get pc for thread: %d got %d status",
4190 osthread->thread_id(), status););
4191 // epc is already NULL
4192 }
4193 return epc;
4194 }
4197 // This does not do anything on Solaris. This is basically a hook for being
4198 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
4199 void os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread) {
4200 f(value, method, args, thread);
4201 }
4203 // This routine may be used by user applications as a "hook" to catch signals.
4204 // The user-defined signal handler must pass unrecognized signals to this
4205 // routine, and if it returns true (non-zero), then the signal handler must
4206 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4207 // routine will never retun false (zero), but instead will execute a VM panic
4208 // routine kill the process.
4209 //
4210 // If this routine returns false, it is OK to call it again. This allows
4211 // the user-defined signal handler to perform checks either before or after
4212 // the VM performs its own checks. Naturally, the user code would be making
4213 // a serious error if it tried to handle an exception (such as a null check
4214 // or breakpoint) that the VM was generating for its own correct operation.
4215 //
4216 // This routine may recognize any of the following kinds of signals:
4217 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
4218 // os::Solaris::SIGasync
4219 // It should be consulted by handlers for any of those signals.
4220 // It explicitly does not recognize os::Solaris::SIGinterrupt
4221 //
4222 // The caller of this routine must pass in the three arguments supplied
4223 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4224 // field of the structure passed to sigaction(). This routine assumes that
4225 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4226 //
4227 // Note that the VM will print warnings if it detects conflicting signal
4228 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4229 //
4230 extern "C" JNIEXPORT int
4231 JVM_handle_solaris_signal(int signo, siginfo_t* siginfo, void* ucontext,
4232 int abort_if_unrecognized);
4235 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
4236 JVM_handle_solaris_signal(sig, info, ucVoid, true);
4237 }
4239 /* Do not delete - if guarantee is ever removed, a signal handler (even empty)
4240 is needed to provoke threads blocked on IO to return an EINTR
4241 Note: this explicitly does NOT call JVM_handle_solaris_signal and
4242 does NOT participate in signal chaining due to requirement for
4243 NOT setting SA_RESTART to make EINTR work. */
4244 extern "C" void sigINTRHandler(int sig, siginfo_t* info, void* ucVoid) {
4245 if (UseSignalChaining) {
4246 struct sigaction *actp = os::Solaris::get_chained_signal_action(sig);
4247 if (actp && actp->sa_handler) {
4248 vm_exit_during_initialization("Signal chaining detected for VM interrupt signal, try -XX:+UseAltSigs");
4249 }
4250 }
4251 }
4253 // This boolean allows users to forward their own non-matching signals
4254 // to JVM_handle_solaris_signal, harmlessly.
4255 bool os::Solaris::signal_handlers_are_installed = false;
4257 // For signal-chaining
4258 bool os::Solaris::libjsig_is_loaded = false;
4259 typedef struct sigaction *(*get_signal_t)(int);
4260 get_signal_t os::Solaris::get_signal_action = NULL;
4262 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
4263 struct sigaction *actp = NULL;
4265 if ((libjsig_is_loaded) && (sig <= Maxlibjsigsigs)) {
4266 // Retrieve the old signal handler from libjsig
4267 actp = (*get_signal_action)(sig);
4268 }
4269 if (actp == NULL) {
4270 // Retrieve the preinstalled signal handler from jvm
4271 actp = get_preinstalled_handler(sig);
4272 }
4274 return actp;
4275 }
4277 static bool call_chained_handler(struct sigaction *actp, int sig,
4278 siginfo_t *siginfo, void *context) {
4279 // Call the old signal handler
4280 if (actp->sa_handler == SIG_DFL) {
4281 // It's more reasonable to let jvm treat it as an unexpected exception
4282 // instead of taking the default action.
4283 return false;
4284 } else if (actp->sa_handler != SIG_IGN) {
4285 if ((actp->sa_flags & SA_NODEFER) == 0) {
4286 // automaticlly block the signal
4287 sigaddset(&(actp->sa_mask), sig);
4288 }
4290 sa_handler_t hand;
4291 sa_sigaction_t sa;
4292 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4293 // retrieve the chained handler
4294 if (siginfo_flag_set) {
4295 sa = actp->sa_sigaction;
4296 } else {
4297 hand = actp->sa_handler;
4298 }
4300 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4301 actp->sa_handler = SIG_DFL;
4302 }
4304 // try to honor the signal mask
4305 sigset_t oset;
4306 thr_sigsetmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4308 // call into the chained handler
4309 if (siginfo_flag_set) {
4310 (*sa)(sig, siginfo, context);
4311 } else {
4312 (*hand)(sig);
4313 }
4315 // restore the signal mask
4316 thr_sigsetmask(SIG_SETMASK, &oset, 0);
4317 }
4318 // Tell jvm's signal handler the signal is taken care of.
4319 return true;
4320 }
4322 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4323 bool chained = false;
4324 // signal-chaining
4325 if (UseSignalChaining) {
4326 struct sigaction *actp = get_chained_signal_action(sig);
4327 if (actp != NULL) {
4328 chained = call_chained_handler(actp, sig, siginfo, context);
4329 }
4330 }
4331 return chained;
4332 }
4334 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
4335 assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4336 if (preinstalled_sigs[sig] != 0) {
4337 return &chainedsigactions[sig];
4338 }
4339 return NULL;
4340 }
4342 void os::Solaris::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4344 assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
4345 assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4346 chainedsigactions[sig] = oldAct;
4347 preinstalled_sigs[sig] = 1;
4348 }
4350 void os::Solaris::set_signal_handler(int sig, bool set_installed, bool oktochain) {
4351 // Check for overwrite.
4352 struct sigaction oldAct;
4353 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4354 void* oldhand = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4355 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4356 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4357 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4358 oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
4359 if (AllowUserSignalHandlers || !set_installed) {
4360 // Do not overwrite; user takes responsibility to forward to us.
4361 return;
4362 } else if (UseSignalChaining) {
4363 if (oktochain) {
4364 // save the old handler in jvm
4365 save_preinstalled_handler(sig, oldAct);
4366 } else {
4367 vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal, try -XX:+UseAltSigs.");
4368 }
4369 // libjsig also interposes the sigaction() call below and saves the
4370 // old sigaction on it own.
4371 } else {
4372 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4373 "%#lx for signal %d.", (long)oldhand, sig));
4374 }
4375 }
4377 struct sigaction sigAct;
4378 sigfillset(&(sigAct.sa_mask));
4379 sigAct.sa_handler = SIG_DFL;
4381 sigAct.sa_sigaction = signalHandler;
4382 // Handle SIGSEGV on alternate signal stack if
4383 // not using stack banging
4384 if (!UseStackBanging && sig == SIGSEGV) {
4385 sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
4386 // Interruptible i/o requires SA_RESTART cleared so EINTR
4387 // is returned instead of restarting system calls
4388 } else if (sig == os::Solaris::SIGinterrupt()) {
4389 sigemptyset(&sigAct.sa_mask);
4390 sigAct.sa_handler = NULL;
4391 sigAct.sa_flags = SA_SIGINFO;
4392 sigAct.sa_sigaction = sigINTRHandler;
4393 } else {
4394 sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
4395 }
4396 os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
4398 sigaction(sig, &sigAct, &oldAct);
4400 void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4401 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4402 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4403 }
4406 #define DO_SIGNAL_CHECK(sig) \
4407 if (!sigismember(&check_signal_done, sig)) \
4408 os::Solaris::check_signal_handler(sig)
4410 // This method is a periodic task to check for misbehaving JNI applications
4411 // under CheckJNI, we can add any periodic checks here
4413 void os::run_periodic_checks() {
4414 // A big source of grief is hijacking virt. addr 0x0 on Solaris,
4415 // thereby preventing a NULL checks.
4416 if(!check_addr0_done) check_addr0_done = check_addr0(tty);
4418 if (check_signals == false) return;
4420 // SEGV and BUS if overridden could potentially prevent
4421 // generation of hs*.log in the event of a crash, debugging
4422 // such a case can be very challenging, so we absolutely
4423 // check for the following for a good measure:
4424 DO_SIGNAL_CHECK(SIGSEGV);
4425 DO_SIGNAL_CHECK(SIGILL);
4426 DO_SIGNAL_CHECK(SIGFPE);
4427 DO_SIGNAL_CHECK(SIGBUS);
4428 DO_SIGNAL_CHECK(SIGPIPE);
4429 DO_SIGNAL_CHECK(SIGXFSZ);
4431 // ReduceSignalUsage allows the user to override these handlers
4432 // see comments at the very top and jvm_solaris.h
4433 if (!ReduceSignalUsage) {
4434 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4435 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4436 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4437 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4438 }
4440 // See comments above for using JVM1/JVM2 and UseAltSigs
4441 DO_SIGNAL_CHECK(os::Solaris::SIGinterrupt());
4442 DO_SIGNAL_CHECK(os::Solaris::SIGasync());
4444 }
4446 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4448 static os_sigaction_t os_sigaction = NULL;
4450 void os::Solaris::check_signal_handler(int sig) {
4451 char buf[O_BUFLEN];
4452 address jvmHandler = NULL;
4454 struct sigaction act;
4455 if (os_sigaction == NULL) {
4456 // only trust the default sigaction, in case it has been interposed
4457 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4458 if (os_sigaction == NULL) return;
4459 }
4461 os_sigaction(sig, (struct sigaction*)NULL, &act);
4463 address thisHandler = (act.sa_flags & SA_SIGINFO)
4464 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4465 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4468 switch(sig) {
4469 case SIGSEGV:
4470 case SIGBUS:
4471 case SIGFPE:
4472 case SIGPIPE:
4473 case SIGXFSZ:
4474 case SIGILL:
4475 jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4476 break;
4478 case SHUTDOWN1_SIGNAL:
4479 case SHUTDOWN2_SIGNAL:
4480 case SHUTDOWN3_SIGNAL:
4481 case BREAK_SIGNAL:
4482 jvmHandler = (address)user_handler();
4483 break;
4485 default:
4486 int intrsig = os::Solaris::SIGinterrupt();
4487 int asynsig = os::Solaris::SIGasync();
4489 if (sig == intrsig) {
4490 jvmHandler = CAST_FROM_FN_PTR(address, sigINTRHandler);
4491 } else if (sig == asynsig) {
4492 jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4493 } else {
4494 return;
4495 }
4496 break;
4497 }
4500 if (thisHandler != jvmHandler) {
4501 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4502 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4503 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4504 // No need to check this sig any longer
4505 sigaddset(&check_signal_done, sig);
4506 } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4507 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4508 tty->print("expected:" PTR32_FORMAT, os::Solaris::get_our_sigflags(sig));
4509 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4510 // No need to check this sig any longer
4511 sigaddset(&check_signal_done, sig);
4512 }
4514 // Print all the signal handler state
4515 if (sigismember(&check_signal_done, sig)) {
4516 print_signal_handlers(tty, buf, O_BUFLEN);
4517 }
4519 }
4521 void os::Solaris::install_signal_handlers() {
4522 bool libjsigdone = false;
4523 signal_handlers_are_installed = true;
4525 // signal-chaining
4526 typedef void (*signal_setting_t)();
4527 signal_setting_t begin_signal_setting = NULL;
4528 signal_setting_t end_signal_setting = NULL;
4529 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4530 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4531 if (begin_signal_setting != NULL) {
4532 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4533 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4534 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4535 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4536 get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4537 dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4538 libjsig_is_loaded = true;
4539 if (os::Solaris::get_libjsig_version != NULL) {
4540 libjsigversion = (*os::Solaris::get_libjsig_version)();
4541 }
4542 assert(UseSignalChaining, "should enable signal-chaining");
4543 }
4544 if (libjsig_is_loaded) {
4545 // Tell libjsig jvm is setting signal handlers
4546 (*begin_signal_setting)();
4547 }
4549 set_signal_handler(SIGSEGV, true, true);
4550 set_signal_handler(SIGPIPE, true, true);
4551 set_signal_handler(SIGXFSZ, true, true);
4552 set_signal_handler(SIGBUS, true, true);
4553 set_signal_handler(SIGILL, true, true);
4554 set_signal_handler(SIGFPE, true, true);
4557 if (os::Solaris::SIGinterrupt() > OLDMAXSIGNUM || os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4559 // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4560 // can not register overridable signals which might be > 32
4561 if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4562 // Tell libjsig jvm has finished setting signal handlers
4563 (*end_signal_setting)();
4564 libjsigdone = true;
4565 }
4566 }
4568 // Never ok to chain our SIGinterrupt
4569 set_signal_handler(os::Solaris::SIGinterrupt(), true, false);
4570 set_signal_handler(os::Solaris::SIGasync(), true, true);
4572 if (libjsig_is_loaded && !libjsigdone) {
4573 // Tell libjsig jvm finishes setting signal handlers
4574 (*end_signal_setting)();
4575 }
4577 // We don't activate signal checker if libjsig is in place, we trust ourselves
4578 // and if UserSignalHandler is installed all bets are off
4579 if (CheckJNICalls) {
4580 if (libjsig_is_loaded) {
4581 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4582 check_signals = false;
4583 }
4584 if (AllowUserSignalHandlers) {
4585 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4586 check_signals = false;
4587 }
4588 }
4589 }
4592 void report_error(const char* file_name, int line_no, const char* title, const char* format, ...);
4594 const char * signames[] = {
4595 "SIG0",
4596 "SIGHUP", "SIGINT", "SIGQUIT", "SIGILL", "SIGTRAP",
4597 "SIGABRT", "SIGEMT", "SIGFPE", "SIGKILL", "SIGBUS",
4598 "SIGSEGV", "SIGSYS", "SIGPIPE", "SIGALRM", "SIGTERM",
4599 "SIGUSR1", "SIGUSR2", "SIGCLD", "SIGPWR", "SIGWINCH",
4600 "SIGURG", "SIGPOLL", "SIGSTOP", "SIGTSTP", "SIGCONT",
4601 "SIGTTIN", "SIGTTOU", "SIGVTALRM", "SIGPROF", "SIGXCPU",
4602 "SIGXFSZ", "SIGWAITING", "SIGLWP", "SIGFREEZE", "SIGTHAW",
4603 "SIGCANCEL", "SIGLOST"
4604 };
4606 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4607 if (0 < exception_code && exception_code <= SIGRTMAX) {
4608 // signal
4609 if (exception_code < sizeof(signames)/sizeof(const char*)) {
4610 jio_snprintf(buf, size, "%s", signames[exception_code]);
4611 } else {
4612 jio_snprintf(buf, size, "SIG%d", exception_code);
4613 }
4614 return buf;
4615 } else {
4616 return NULL;
4617 }
4618 }
4620 // (Static) wrappers for the new libthread API
4621 int_fnP_thread_t_iP_uP_stack_tP_gregset_t os::Solaris::_thr_getstate;
4622 int_fnP_thread_t_i_gregset_t os::Solaris::_thr_setstate;
4623 int_fnP_thread_t_i os::Solaris::_thr_setmutator;
4624 int_fnP_thread_t os::Solaris::_thr_suspend_mutator;
4625 int_fnP_thread_t os::Solaris::_thr_continue_mutator;
4627 // (Static) wrapper for getisax(2) call.
4628 os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4630 // (Static) wrappers for the liblgrp API
4631 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4632 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4633 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4634 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4635 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4636 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4637 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4638 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4639 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4641 // (Static) wrapper for meminfo() call.
4642 os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4644 static address resolve_symbol_lazy(const char* name) {
4645 address addr = (address) dlsym(RTLD_DEFAULT, name);
4646 if(addr == NULL) {
4647 // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4648 addr = (address) dlsym(RTLD_NEXT, name);
4649 }
4650 return addr;
4651 }
4653 static address resolve_symbol(const char* name) {
4654 address addr = resolve_symbol_lazy(name);
4655 if(addr == NULL) {
4656 fatal(dlerror());
4657 }
4658 return addr;
4659 }
4663 // isT2_libthread()
4664 //
4665 // Routine to determine if we are currently using the new T2 libthread.
4666 //
4667 // We determine if we are using T2 by reading /proc/self/lstatus and
4668 // looking for a thread with the ASLWP bit set. If we find this status
4669 // bit set, we must assume that we are NOT using T2. The T2 team
4670 // has approved this algorithm.
4671 //
4672 // We need to determine if we are running with the new T2 libthread
4673 // since setting native thread priorities is handled differently
4674 // when using this library. All threads created using T2 are bound
4675 // threads. Calling thr_setprio is meaningless in this case.
4676 //
4677 bool isT2_libthread() {
4678 static prheader_t * lwpArray = NULL;
4679 static int lwpSize = 0;
4680 static int lwpFile = -1;
4681 lwpstatus_t * that;
4682 char lwpName [128];
4683 bool isT2 = false;
4685 #define ADR(x) ((uintptr_t)(x))
4686 #define LWPINDEX(ary,ix) ((lwpstatus_t *)(((ary)->pr_entsize * (ix)) + (ADR((ary) + 1))))
4688 lwpFile = ::open("/proc/self/lstatus", O_RDONLY, 0);
4689 if (lwpFile < 0) {
4690 if (ThreadPriorityVerbose) warning ("Couldn't open /proc/self/lstatus\n");
4691 return false;
4692 }
4693 lwpSize = 16*1024;
4694 for (;;) {
4695 ::lseek64 (lwpFile, 0, SEEK_SET);
4696 lwpArray = (prheader_t *)NEW_C_HEAP_ARRAY(char, lwpSize);
4697 if (::read(lwpFile, lwpArray, lwpSize) < 0) {
4698 if (ThreadPriorityVerbose) warning("Error reading /proc/self/lstatus\n");
4699 break;
4700 }
4701 if ((lwpArray->pr_nent * lwpArray->pr_entsize) <= lwpSize) {
4702 // We got a good snapshot - now iterate over the list.
4703 int aslwpcount = 0;
4704 for (int i = 0; i < lwpArray->pr_nent; i++ ) {
4705 that = LWPINDEX(lwpArray,i);
4706 if (that->pr_flags & PR_ASLWP) {
4707 aslwpcount++;
4708 }
4709 }
4710 if (aslwpcount == 0) isT2 = true;
4711 break;
4712 }
4713 lwpSize = lwpArray->pr_nent * lwpArray->pr_entsize;
4714 FREE_C_HEAP_ARRAY(char, lwpArray); // retry.
4715 }
4717 FREE_C_HEAP_ARRAY(char, lwpArray);
4718 ::close (lwpFile);
4719 if (ThreadPriorityVerbose) {
4720 if (isT2) tty->print_cr("We are running with a T2 libthread\n");
4721 else tty->print_cr("We are not running with a T2 libthread\n");
4722 }
4723 return isT2;
4724 }
4727 void os::Solaris::libthread_init() {
4728 address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4730 // Determine if we are running with the new T2 libthread
4731 os::Solaris::set_T2_libthread(isT2_libthread());
4733 lwp_priocntl_init();
4735 // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4736 if(func == NULL) {
4737 func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4738 // Guarantee that this VM is running on an new enough OS (5.6 or
4739 // later) that it will have a new enough libthread.so.
4740 guarantee(func != NULL, "libthread.so is too old.");
4741 }
4743 // Initialize the new libthread getstate API wrappers
4744 func = resolve_symbol("thr_getstate");
4745 os::Solaris::set_thr_getstate(CAST_TO_FN_PTR(int_fnP_thread_t_iP_uP_stack_tP_gregset_t, func));
4747 func = resolve_symbol("thr_setstate");
4748 os::Solaris::set_thr_setstate(CAST_TO_FN_PTR(int_fnP_thread_t_i_gregset_t, func));
4750 func = resolve_symbol("thr_setmutator");
4751 os::Solaris::set_thr_setmutator(CAST_TO_FN_PTR(int_fnP_thread_t_i, func));
4753 func = resolve_symbol("thr_suspend_mutator");
4754 os::Solaris::set_thr_suspend_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4756 func = resolve_symbol("thr_continue_mutator");
4757 os::Solaris::set_thr_continue_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4759 int size;
4760 void (*handler_info_func)(address *, int *);
4761 handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4762 handler_info_func(&handler_start, &size);
4763 handler_end = handler_start + size;
4764 }
4767 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4768 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4769 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4770 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4771 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4772 int os::Solaris::_mutex_scope = USYNC_THREAD;
4774 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4775 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4776 int_fnP_cond_tP os::Solaris::_cond_signal;
4777 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4778 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4779 int_fnP_cond_tP os::Solaris::_cond_destroy;
4780 int os::Solaris::_cond_scope = USYNC_THREAD;
4782 void os::Solaris::synchronization_init() {
4783 if(UseLWPSynchronization) {
4784 os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4785 os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4786 os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4787 os::Solaris::set_mutex_init(lwp_mutex_init);
4788 os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4789 os::Solaris::set_mutex_scope(USYNC_THREAD);
4791 os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4792 os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4793 os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4794 os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4795 os::Solaris::set_cond_init(lwp_cond_init);
4796 os::Solaris::set_cond_destroy(lwp_cond_destroy);
4797 os::Solaris::set_cond_scope(USYNC_THREAD);
4798 }
4799 else {
4800 os::Solaris::set_mutex_scope(USYNC_THREAD);
4801 os::Solaris::set_cond_scope(USYNC_THREAD);
4803 if(UsePthreads) {
4804 os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4805 os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4806 os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4807 os::Solaris::set_mutex_init(pthread_mutex_default_init);
4808 os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4810 os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4811 os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4812 os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4813 os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4814 os::Solaris::set_cond_init(pthread_cond_default_init);
4815 os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4816 }
4817 else {
4818 os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4819 os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4820 os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4821 os::Solaris::set_mutex_init(::mutex_init);
4822 os::Solaris::set_mutex_destroy(::mutex_destroy);
4824 os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4825 os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4826 os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4827 os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4828 os::Solaris::set_cond_init(::cond_init);
4829 os::Solaris::set_cond_destroy(::cond_destroy);
4830 }
4831 }
4832 }
4834 bool os::Solaris::liblgrp_init() {
4835 void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4836 if (handle != NULL) {
4837 os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4838 os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4839 os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4840 os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4841 os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4842 os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4843 os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4844 os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4845 dlsym(handle, "lgrp_cookie_stale")));
4847 lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4848 set_lgrp_cookie(c);
4849 return true;
4850 }
4851 return false;
4852 }
4854 void os::Solaris::misc_sym_init() {
4855 address func;
4857 // getisax
4858 func = resolve_symbol_lazy("getisax");
4859 if (func != NULL) {
4860 os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
4861 }
4863 // meminfo
4864 func = resolve_symbol_lazy("meminfo");
4865 if (func != NULL) {
4866 os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
4867 }
4868 }
4870 uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
4871 assert(_getisax != NULL, "_getisax not set");
4872 return _getisax(array, n);
4873 }
4875 // Symbol doesn't exist in Solaris 8 pset.h
4876 #ifndef PS_MYID
4877 #define PS_MYID -3
4878 #endif
4880 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4881 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4882 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4884 void init_pset_getloadavg_ptr(void) {
4885 pset_getloadavg_ptr =
4886 (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4887 if (PrintMiscellaneous && Verbose && pset_getloadavg_ptr == NULL) {
4888 warning("pset_getloadavg function not found");
4889 }
4890 }
4892 int os::Solaris::_dev_zero_fd = -1;
4894 // this is called _before_ the global arguments have been parsed
4895 void os::init(void) {
4896 _initial_pid = getpid();
4898 max_hrtime = first_hrtime = gethrtime();
4900 init_random(1234567);
4902 page_size = sysconf(_SC_PAGESIZE);
4903 if (page_size == -1)
4904 fatal(err_msg("os_solaris.cpp: os::init: sysconf failed (%s)",
4905 strerror(errno)));
4906 init_page_sizes((size_t) page_size);
4908 Solaris::initialize_system_info();
4910 // Initialize misc. symbols as soon as possible, so we can use them
4911 // if we need them.
4912 Solaris::misc_sym_init();
4914 int fd = ::open("/dev/zero", O_RDWR);
4915 if (fd < 0) {
4916 fatal(err_msg("os::init: cannot open /dev/zero (%s)", strerror(errno)));
4917 } else {
4918 Solaris::set_dev_zero_fd(fd);
4920 // Close on exec, child won't inherit.
4921 fcntl(fd, F_SETFD, FD_CLOEXEC);
4922 }
4924 clock_tics_per_sec = CLK_TCK;
4926 // check if dladdr1() exists; dladdr1 can provide more information than
4927 // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4928 // and is available on linker patches for 5.7 and 5.8.
4929 // libdl.so must have been loaded, this call is just an entry lookup
4930 void * hdl = dlopen("libdl.so", RTLD_NOW);
4931 if (hdl)
4932 dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4934 // (Solaris only) this switches to calls that actually do locking.
4935 ThreadCritical::initialize();
4937 main_thread = thr_self();
4939 // Constant minimum stack size allowed. It must be at least
4940 // the minimum of what the OS supports (thr_min_stack()), and
4941 // enough to allow the thread to get to user bytecode execution.
4942 Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
4943 // If the pagesize of the VM is greater than 8K determine the appropriate
4944 // number of initial guard pages. The user can change this with the
4945 // command line arguments, if needed.
4946 if (vm_page_size() > 8*K) {
4947 StackYellowPages = 1;
4948 StackRedPages = 1;
4949 StackShadowPages = round_to((StackShadowPages*8*K), vm_page_size()) / vm_page_size();
4950 }
4951 }
4953 // To install functions for atexit system call
4954 extern "C" {
4955 static void perfMemory_exit_helper() {
4956 perfMemory_exit();
4957 }
4958 }
4960 // this is called _after_ the global arguments have been parsed
4961 jint os::init_2(void) {
4962 // try to enable extended file IO ASAP, see 6431278
4963 os::Solaris::try_enable_extended_io();
4965 // Allocate a single page and mark it as readable for safepoint polling. Also
4966 // use this first mmap call to check support for MAP_ALIGN.
4967 address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
4968 page_size,
4969 MAP_PRIVATE | MAP_ALIGN,
4970 PROT_READ);
4971 if (polling_page == NULL) {
4972 has_map_align = false;
4973 polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
4974 PROT_READ);
4975 }
4977 os::set_polling_page(polling_page);
4979 #ifndef PRODUCT
4980 if( Verbose && PrintMiscellaneous )
4981 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4982 #endif
4984 if (!UseMembar) {
4985 address mem_serialize_page = (address)Solaris::mmap_chunk( NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE );
4986 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4987 os::set_memory_serialize_page( mem_serialize_page );
4989 #ifndef PRODUCT
4990 if(Verbose && PrintMiscellaneous)
4991 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4992 #endif
4993 }
4995 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
4997 // Check minimum allowable stack size for thread creation and to initialize
4998 // the java system classes, including StackOverflowError - depends on page
4999 // size. Add a page for compiler2 recursion in main thread.
5000 // Add in 2*BytesPerWord times page size to account for VM stack during
5001 // class initialization depending on 32 or 64 bit VM.
5002 os::Solaris::min_stack_allowed = MAX2(os::Solaris::min_stack_allowed,
5003 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
5004 2*BytesPerWord COMPILER2_PRESENT(+1)) * page_size);
5006 size_t threadStackSizeInBytes = ThreadStackSize * K;
5007 if (threadStackSizeInBytes != 0 &&
5008 threadStackSizeInBytes < os::Solaris::min_stack_allowed) {
5009 tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
5010 os::Solaris::min_stack_allowed/K);
5011 return JNI_ERR;
5012 }
5014 // For 64kbps there will be a 64kb page size, which makes
5015 // the usable default stack size quite a bit less. Increase the
5016 // stack for 64kb (or any > than 8kb) pages, this increases
5017 // virtual memory fragmentation (since we're not creating the
5018 // stack on a power of 2 boundary. The real fix for this
5019 // should be to fix the guard page mechanism.
5021 if (vm_page_size() > 8*K) {
5022 threadStackSizeInBytes = (threadStackSizeInBytes != 0)
5023 ? threadStackSizeInBytes +
5024 ((StackYellowPages + StackRedPages) * vm_page_size())
5025 : 0;
5026 ThreadStackSize = threadStackSizeInBytes/K;
5027 }
5029 // Make the stack size a multiple of the page size so that
5030 // the yellow/red zones can be guarded.
5031 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5032 vm_page_size()));
5034 Solaris::libthread_init();
5036 if (UseNUMA) {
5037 if (!Solaris::liblgrp_init()) {
5038 UseNUMA = false;
5039 } else {
5040 size_t lgrp_limit = os::numa_get_groups_num();
5041 int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit);
5042 size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
5043 FREE_C_HEAP_ARRAY(int, lgrp_ids);
5044 if (lgrp_num < 2) {
5045 // There's only one locality group, disable NUMA.
5046 UseNUMA = false;
5047 }
5048 }
5049 // ISM is not compatible with the NUMA allocator - it always allocates
5050 // pages round-robin across the lgroups.
5051 if (UseNUMA && UseLargePages && UseISM) {
5052 if (!FLAG_IS_DEFAULT(UseNUMA)) {
5053 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseISM)) {
5054 UseLargePages = false;
5055 } else {
5056 warning("UseNUMA is not compatible with ISM large pages, disabling NUMA allocator");
5057 UseNUMA = false;
5058 }
5059 } else {
5060 UseNUMA = false;
5061 }
5062 }
5063 if (!UseNUMA && ForceNUMA) {
5064 UseNUMA = true;
5065 }
5066 }
5068 Solaris::signal_sets_init();
5069 Solaris::init_signal_mem();
5070 Solaris::install_signal_handlers();
5072 if (libjsigversion < JSIG_VERSION_1_4_1) {
5073 Maxlibjsigsigs = OLDMAXSIGNUM;
5074 }
5076 // initialize synchronization primitives to use either thread or
5077 // lwp synchronization (controlled by UseLWPSynchronization)
5078 Solaris::synchronization_init();
5080 if (MaxFDLimit) {
5081 // set the number of file descriptors to max. print out error
5082 // if getrlimit/setrlimit fails but continue regardless.
5083 struct rlimit nbr_files;
5084 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5085 if (status != 0) {
5086 if (PrintMiscellaneous && (Verbose || WizardMode))
5087 perror("os::init_2 getrlimit failed");
5088 } else {
5089 nbr_files.rlim_cur = nbr_files.rlim_max;
5090 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5091 if (status != 0) {
5092 if (PrintMiscellaneous && (Verbose || WizardMode))
5093 perror("os::init_2 setrlimit failed");
5094 }
5095 }
5096 }
5098 // Calculate theoretical max. size of Threads to guard gainst
5099 // artifical out-of-memory situations, where all available address-
5100 // space has been reserved by thread stacks. Default stack size is 1Mb.
5101 size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
5102 JavaThread::stack_size_at_create() : (1*K*K);
5103 assert(pre_thread_stack_size != 0, "Must have a stack");
5104 // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
5105 // we should start doing Virtual Memory banging. Currently when the threads will
5106 // have used all but 200Mb of space.
5107 size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
5108 Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
5110 // at-exit methods are called in the reverse order of their registration.
5111 // In Solaris 7 and earlier, atexit functions are called on return from
5112 // main or as a result of a call to exit(3C). There can be only 32 of
5113 // these functions registered and atexit() does not set errno. In Solaris
5114 // 8 and later, there is no limit to the number of functions registered
5115 // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
5116 // functions are called upon dlclose(3DL) in addition to return from main
5117 // and exit(3C).
5119 if (PerfAllowAtExitRegistration) {
5120 // only register atexit functions if PerfAllowAtExitRegistration is set.
5121 // atexit functions can be delayed until process exit time, which
5122 // can be problematic for embedded VM situations. Embedded VMs should
5123 // call DestroyJavaVM() to assure that VM resources are released.
5125 // note: perfMemory_exit_helper atexit function may be removed in
5126 // the future if the appropriate cleanup code can be added to the
5127 // VM_Exit VMOperation's doit method.
5128 if (atexit(perfMemory_exit_helper) != 0) {
5129 warning("os::init2 atexit(perfMemory_exit_helper) failed");
5130 }
5131 }
5133 // Init pset_loadavg function pointer
5134 init_pset_getloadavg_ptr();
5136 return JNI_OK;
5137 }
5139 void os::init_3(void) {
5140 return;
5141 }
5143 // Mark the polling page as unreadable
5144 void os::make_polling_page_unreadable(void) {
5145 if( mprotect((char *)_polling_page, page_size, PROT_NONE) != 0 )
5146 fatal("Could not disable polling page");
5147 };
5149 // Mark the polling page as readable
5150 void os::make_polling_page_readable(void) {
5151 if( mprotect((char *)_polling_page, page_size, PROT_READ) != 0 )
5152 fatal("Could not enable polling page");
5153 };
5155 // OS interface.
5157 bool os::check_heap(bool force) { return true; }
5159 typedef int (*vsnprintf_t)(char* buf, size_t count, const char* fmt, va_list argptr);
5160 static vsnprintf_t sol_vsnprintf = NULL;
5162 int local_vsnprintf(char* buf, size_t count, const char* fmt, va_list argptr) {
5163 if (!sol_vsnprintf) {
5164 //search for the named symbol in the objects that were loaded after libjvm
5165 void* where = RTLD_NEXT;
5166 if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5167 sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5168 if (!sol_vsnprintf){
5169 //search for the named symbol in the objects that were loaded before libjvm
5170 where = RTLD_DEFAULT;
5171 if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5172 sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5173 assert(sol_vsnprintf != NULL, "vsnprintf not found");
5174 }
5175 }
5176 return (*sol_vsnprintf)(buf, count, fmt, argptr);
5177 }
5180 // Is a (classpath) directory empty?
5181 bool os::dir_is_empty(const char* path) {
5182 DIR *dir = NULL;
5183 struct dirent *ptr;
5185 dir = opendir(path);
5186 if (dir == NULL) return true;
5188 /* Scan the directory */
5189 bool result = true;
5190 char buf[sizeof(struct dirent) + MAX_PATH];
5191 struct dirent *dbuf = (struct dirent *) buf;
5192 while (result && (ptr = readdir(dir, dbuf)) != NULL) {
5193 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5194 result = false;
5195 }
5196 }
5197 closedir(dir);
5198 return result;
5199 }
5201 // This code originates from JDK's sysOpen and open64_w
5202 // from src/solaris/hpi/src/system_md.c
5204 #ifndef O_DELETE
5205 #define O_DELETE 0x10000
5206 #endif
5208 // Open a file. Unlink the file immediately after open returns
5209 // if the specified oflag has the O_DELETE flag set.
5210 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5212 int os::open(const char *path, int oflag, int mode) {
5213 if (strlen(path) > MAX_PATH - 1) {
5214 errno = ENAMETOOLONG;
5215 return -1;
5216 }
5217 int fd;
5218 int o_delete = (oflag & O_DELETE);
5219 oflag = oflag & ~O_DELETE;
5221 fd = ::open64(path, oflag, mode);
5222 if (fd == -1) return -1;
5224 //If the open succeeded, the file might still be a directory
5225 {
5226 struct stat64 buf64;
5227 int ret = ::fstat64(fd, &buf64);
5228 int st_mode = buf64.st_mode;
5230 if (ret != -1) {
5231 if ((st_mode & S_IFMT) == S_IFDIR) {
5232 errno = EISDIR;
5233 ::close(fd);
5234 return -1;
5235 }
5236 } else {
5237 ::close(fd);
5238 return -1;
5239 }
5240 }
5241 /*
5242 * 32-bit Solaris systems suffer from:
5243 *
5244 * - an historical default soft limit of 256 per-process file
5245 * descriptors that is too low for many Java programs.
5246 *
5247 * - a design flaw where file descriptors created using stdio
5248 * fopen must be less than 256, _even_ when the first limit above
5249 * has been raised. This can cause calls to fopen (but not calls to
5250 * open, for example) to fail mysteriously, perhaps in 3rd party
5251 * native code (although the JDK itself uses fopen). One can hardly
5252 * criticize them for using this most standard of all functions.
5253 *
5254 * We attempt to make everything work anyways by:
5255 *
5256 * - raising the soft limit on per-process file descriptors beyond
5257 * 256
5258 *
5259 * - As of Solaris 10u4, we can request that Solaris raise the 256
5260 * stdio fopen limit by calling function enable_extended_FILE_stdio.
5261 * This is done in init_2 and recorded in enabled_extended_FILE_stdio
5262 *
5263 * - If we are stuck on an old (pre 10u4) Solaris system, we can
5264 * workaround the bug by remapping non-stdio file descriptors below
5265 * 256 to ones beyond 256, which is done below.
5266 *
5267 * See:
5268 * 1085341: 32-bit stdio routines should support file descriptors >255
5269 * 6533291: Work around 32-bit Solaris stdio limit of 256 open files
5270 * 6431278: Netbeans crash on 32 bit Solaris: need to call
5271 * enable_extended_FILE_stdio() in VM initialisation
5272 * Giri Mandalika's blog
5273 * http://technopark02.blogspot.com/2005_05_01_archive.html
5274 */
5275 #ifndef _LP64
5276 if ((!enabled_extended_FILE_stdio) && fd < 256) {
5277 int newfd = ::fcntl(fd, F_DUPFD, 256);
5278 if (newfd != -1) {
5279 ::close(fd);
5280 fd = newfd;
5281 }
5282 }
5283 #endif // 32-bit Solaris
5284 /*
5285 * All file descriptors that are opened in the JVM and not
5286 * specifically destined for a subprocess should have the
5287 * close-on-exec flag set. If we don't set it, then careless 3rd
5288 * party native code might fork and exec without closing all
5289 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5290 * UNIXProcess.c), and this in turn might:
5291 *
5292 * - cause end-of-file to fail to be detected on some file
5293 * descriptors, resulting in mysterious hangs, or
5294 *
5295 * - might cause an fopen in the subprocess to fail on a system
5296 * suffering from bug 1085341.
5297 *
5298 * (Yes, the default setting of the close-on-exec flag is a Unix
5299 * design flaw)
5300 *
5301 * See:
5302 * 1085341: 32-bit stdio routines should support file descriptors >255
5303 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5304 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5305 */
5306 #ifdef FD_CLOEXEC
5307 {
5308 int flags = ::fcntl(fd, F_GETFD);
5309 if (flags != -1)
5310 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5311 }
5312 #endif
5314 if (o_delete != 0) {
5315 ::unlink(path);
5316 }
5317 return fd;
5318 }
5320 // create binary file, rewriting existing file if required
5321 int os::create_binary_file(const char* path, bool rewrite_existing) {
5322 int oflags = O_WRONLY | O_CREAT;
5323 if (!rewrite_existing) {
5324 oflags |= O_EXCL;
5325 }
5326 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5327 }
5329 // return current position of file pointer
5330 jlong os::current_file_offset(int fd) {
5331 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5332 }
5334 // move file pointer to the specified offset
5335 jlong os::seek_to_file_offset(int fd, jlong offset) {
5336 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5337 }
5339 jlong os::lseek(int fd, jlong offset, int whence) {
5340 return (jlong) ::lseek64(fd, offset, whence);
5341 }
5343 char * os::native_path(char *path) {
5344 return path;
5345 }
5347 int os::ftruncate(int fd, jlong length) {
5348 return ::ftruncate64(fd, length);
5349 }
5351 int os::fsync(int fd) {
5352 RESTARTABLE_RETURN_INT(::fsync(fd));
5353 }
5355 int os::available(int fd, jlong *bytes) {
5356 jlong cur, end;
5357 int mode;
5358 struct stat64 buf64;
5360 if (::fstat64(fd, &buf64) >= 0) {
5361 mode = buf64.st_mode;
5362 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5363 /*
5364 * XXX: is the following call interruptible? If so, this might
5365 * need to go through the INTERRUPT_IO() wrapper as for other
5366 * blocking, interruptible calls in this file.
5367 */
5368 int n,ioctl_return;
5370 INTERRUPTIBLE(::ioctl(fd, FIONREAD, &n),ioctl_return,os::Solaris::clear_interrupted);
5371 if (ioctl_return>= 0) {
5372 *bytes = n;
5373 return 1;
5374 }
5375 }
5376 }
5377 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5378 return 0;
5379 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5380 return 0;
5381 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5382 return 0;
5383 }
5384 *bytes = end - cur;
5385 return 1;
5386 }
5388 // Map a block of memory.
5389 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
5390 char *addr, size_t bytes, bool read_only,
5391 bool allow_exec) {
5392 int prot;
5393 int flags;
5395 if (read_only) {
5396 prot = PROT_READ;
5397 flags = MAP_SHARED;
5398 } else {
5399 prot = PROT_READ | PROT_WRITE;
5400 flags = MAP_PRIVATE;
5401 }
5403 if (allow_exec) {
5404 prot |= PROT_EXEC;
5405 }
5407 if (addr != NULL) {
5408 flags |= MAP_FIXED;
5409 }
5411 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5412 fd, file_offset);
5413 if (mapped_address == MAP_FAILED) {
5414 return NULL;
5415 }
5416 return mapped_address;
5417 }
5420 // Remap a block of memory.
5421 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
5422 char *addr, size_t bytes, bool read_only,
5423 bool allow_exec) {
5424 // same as map_memory() on this OS
5425 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5426 allow_exec);
5427 }
5430 // Unmap a block of memory.
5431 bool os::unmap_memory(char* addr, size_t bytes) {
5432 return munmap(addr, bytes) == 0;
5433 }
5435 void os::pause() {
5436 char filename[MAX_PATH];
5437 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5438 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5439 } else {
5440 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5441 }
5443 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5444 if (fd != -1) {
5445 struct stat buf;
5446 ::close(fd);
5447 while (::stat(filename, &buf) == 0) {
5448 (void)::poll(NULL, 0, 100);
5449 }
5450 } else {
5451 jio_fprintf(stderr,
5452 "Could not open pause file '%s', continuing immediately.\n", filename);
5453 }
5454 }
5456 #ifndef PRODUCT
5457 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5458 // Turn this on if you need to trace synch operations.
5459 // Set RECORD_SYNCH_LIMIT to a large-enough value,
5460 // and call record_synch_enable and record_synch_disable
5461 // around the computation of interest.
5463 void record_synch(char* name, bool returning); // defined below
5465 class RecordSynch {
5466 char* _name;
5467 public:
5468 RecordSynch(char* name) :_name(name)
5469 { record_synch(_name, false); }
5470 ~RecordSynch() { record_synch(_name, true); }
5471 };
5473 #define CHECK_SYNCH_OP(ret, name, params, args, inner) \
5474 extern "C" ret name params { \
5475 typedef ret name##_t params; \
5476 static name##_t* implem = NULL; \
5477 static int callcount = 0; \
5478 if (implem == NULL) { \
5479 implem = (name##_t*) dlsym(RTLD_NEXT, #name); \
5480 if (implem == NULL) fatal(dlerror()); \
5481 } \
5482 ++callcount; \
5483 RecordSynch _rs(#name); \
5484 inner; \
5485 return implem args; \
5486 }
5487 // in dbx, examine callcounts this way:
5488 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
5490 #define CHECK_POINTER_OK(p) \
5491 (Universe::perm_gen() == NULL || !Universe::is_reserved_heap((oop)(p)))
5492 #define CHECK_MU \
5493 if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
5494 #define CHECK_CV \
5495 if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
5496 #define CHECK_P(p) \
5497 if (!CHECK_POINTER_OK(p)) fatal(false, "Pointer must be in C heap only.");
5499 #define CHECK_MUTEX(mutex_op) \
5500 CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
5502 CHECK_MUTEX( mutex_lock)
5503 CHECK_MUTEX( _mutex_lock)
5504 CHECK_MUTEX( mutex_unlock)
5505 CHECK_MUTEX(_mutex_unlock)
5506 CHECK_MUTEX( mutex_trylock)
5507 CHECK_MUTEX(_mutex_trylock)
5509 #define CHECK_COND(cond_op) \
5510 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU;CHECK_CV);
5512 CHECK_COND( cond_wait);
5513 CHECK_COND(_cond_wait);
5514 CHECK_COND(_cond_wait_cancel);
5516 #define CHECK_COND2(cond_op) \
5517 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU;CHECK_CV);
5519 CHECK_COND2( cond_timedwait);
5520 CHECK_COND2(_cond_timedwait);
5521 CHECK_COND2(_cond_timedwait_cancel);
5523 // do the _lwp_* versions too
5524 #define mutex_t lwp_mutex_t
5525 #define cond_t lwp_cond_t
5526 CHECK_MUTEX( _lwp_mutex_lock)
5527 CHECK_MUTEX( _lwp_mutex_unlock)
5528 CHECK_MUTEX( _lwp_mutex_trylock)
5529 CHECK_MUTEX( __lwp_mutex_lock)
5530 CHECK_MUTEX( __lwp_mutex_unlock)
5531 CHECK_MUTEX( __lwp_mutex_trylock)
5532 CHECK_MUTEX(___lwp_mutex_lock)
5533 CHECK_MUTEX(___lwp_mutex_unlock)
5535 CHECK_COND( _lwp_cond_wait);
5536 CHECK_COND( __lwp_cond_wait);
5537 CHECK_COND(___lwp_cond_wait);
5539 CHECK_COND2( _lwp_cond_timedwait);
5540 CHECK_COND2( __lwp_cond_timedwait);
5541 #undef mutex_t
5542 #undef cond_t
5544 CHECK_SYNCH_OP(int, _lwp_suspend2, (int lwp, int *n), (lwp, n), 0);
5545 CHECK_SYNCH_OP(int,__lwp_suspend2, (int lwp, int *n), (lwp, n), 0);
5546 CHECK_SYNCH_OP(int, _lwp_kill, (int lwp, int n), (lwp, n), 0);
5547 CHECK_SYNCH_OP(int,__lwp_kill, (int lwp, int n), (lwp, n), 0);
5548 CHECK_SYNCH_OP(int, _lwp_sema_wait, (lwp_sema_t* p), (p), CHECK_P(p));
5549 CHECK_SYNCH_OP(int,__lwp_sema_wait, (lwp_sema_t* p), (p), CHECK_P(p));
5550 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv), (cv), CHECK_CV);
5551 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv), (cv), CHECK_CV);
5554 // recording machinery:
5556 enum { RECORD_SYNCH_LIMIT = 200 };
5557 char* record_synch_name[RECORD_SYNCH_LIMIT];
5558 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
5559 bool record_synch_returning[RECORD_SYNCH_LIMIT];
5560 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
5561 int record_synch_count = 0;
5562 bool record_synch_enabled = false;
5564 // in dbx, examine recorded data this way:
5565 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
5567 void record_synch(char* name, bool returning) {
5568 if (record_synch_enabled) {
5569 if (record_synch_count < RECORD_SYNCH_LIMIT) {
5570 record_synch_name[record_synch_count] = name;
5571 record_synch_returning[record_synch_count] = returning;
5572 record_synch_thread[record_synch_count] = thr_self();
5573 record_synch_arg0ptr[record_synch_count] = &name;
5574 record_synch_count++;
5575 }
5576 // put more checking code here:
5577 // ...
5578 }
5579 }
5581 void record_synch_enable() {
5582 // start collecting trace data, if not already doing so
5583 if (!record_synch_enabled) record_synch_count = 0;
5584 record_synch_enabled = true;
5585 }
5587 void record_synch_disable() {
5588 // stop collecting trace data
5589 record_synch_enabled = false;
5590 }
5592 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5593 #endif // PRODUCT
5595 const intptr_t thr_time_off = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5596 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
5597 (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5600 // JVMTI & JVM monitoring and management support
5601 // The thread_cpu_time() and current_thread_cpu_time() are only
5602 // supported if is_thread_cpu_time_supported() returns true.
5603 // They are not supported on Solaris T1.
5605 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5606 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5607 // of a thread.
5608 //
5609 // current_thread_cpu_time() and thread_cpu_time(Thread *)
5610 // returns the fast estimate available on the platform.
5612 // hrtime_t gethrvtime() return value includes
5613 // user time but does not include system time
5614 jlong os::current_thread_cpu_time() {
5615 return (jlong) gethrvtime();
5616 }
5618 jlong os::thread_cpu_time(Thread *thread) {
5619 // return user level CPU time only to be consistent with
5620 // what current_thread_cpu_time returns.
5621 // thread_cpu_time_info() must be changed if this changes
5622 return os::thread_cpu_time(thread, false /* user time only */);
5623 }
5625 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5626 if (user_sys_cpu_time) {
5627 return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5628 } else {
5629 return os::current_thread_cpu_time();
5630 }
5631 }
5633 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5634 char proc_name[64];
5635 int count;
5636 prusage_t prusage;
5637 jlong lwp_time;
5638 int fd;
5640 sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5641 getpid(),
5642 thread->osthread()->lwp_id());
5643 fd = ::open(proc_name, O_RDONLY);
5644 if ( fd == -1 ) return -1;
5646 do {
5647 count = ::pread(fd,
5648 (void *)&prusage.pr_utime,
5649 thr_time_size,
5650 thr_time_off);
5651 } while (count < 0 && errno == EINTR);
5652 ::close(fd);
5653 if ( count < 0 ) return -1;
5655 if (user_sys_cpu_time) {
5656 // user + system CPU time
5657 lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5658 (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5659 (jlong)prusage.pr_stime.tv_nsec +
5660 (jlong)prusage.pr_utime.tv_nsec;
5661 } else {
5662 // user level CPU time only
5663 lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5664 (jlong)prusage.pr_utime.tv_nsec;
5665 }
5667 return(lwp_time);
5668 }
5670 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5671 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5672 info_ptr->may_skip_backward = false; // elapsed time not wall time
5673 info_ptr->may_skip_forward = false; // elapsed time not wall time
5674 info_ptr->kind = JVMTI_TIMER_USER_CPU; // only user time is returned
5675 }
5677 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5678 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5679 info_ptr->may_skip_backward = false; // elapsed time not wall time
5680 info_ptr->may_skip_forward = false; // elapsed time not wall time
5681 info_ptr->kind = JVMTI_TIMER_USER_CPU; // only user time is returned
5682 }
5684 bool os::is_thread_cpu_time_supported() {
5685 if ( os::Solaris::T2_libthread() || UseBoundThreads ) {
5686 return true;
5687 } else {
5688 return false;
5689 }
5690 }
5692 // System loadavg support. Returns -1 if load average cannot be obtained.
5693 // Return the load average for our processor set if the primitive exists
5694 // (Solaris 9 and later). Otherwise just return system wide loadavg.
5695 int os::loadavg(double loadavg[], int nelem) {
5696 if (pset_getloadavg_ptr != NULL) {
5697 return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5698 } else {
5699 return ::getloadavg(loadavg, nelem);
5700 }
5701 }
5703 //---------------------------------------------------------------------------------
5705 static address same_page(address x, address y) {
5706 intptr_t page_bits = -os::vm_page_size();
5707 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
5708 return x;
5709 else if (x > y)
5710 return (address)(intptr_t(y) | ~page_bits) + 1;
5711 else
5712 return (address)(intptr_t(y) & page_bits);
5713 }
5715 bool os::find(address addr, outputStream* st) {
5716 Dl_info dlinfo;
5717 memset(&dlinfo, 0, sizeof(dlinfo));
5718 if (dladdr(addr, &dlinfo)) {
5719 #ifdef _LP64
5720 st->print("0x%016lx: ", addr);
5721 #else
5722 st->print("0x%08x: ", addr);
5723 #endif
5724 if (dlinfo.dli_sname != NULL)
5725 st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5726 else if (dlinfo.dli_fname)
5727 st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5728 else
5729 st->print("<absolute address>");
5730 if (dlinfo.dli_fname) st->print(" in %s", dlinfo.dli_fname);
5731 #ifdef _LP64
5732 if (dlinfo.dli_fbase) st->print(" at 0x%016lx", dlinfo.dli_fbase);
5733 #else
5734 if (dlinfo.dli_fbase) st->print(" at 0x%08x", dlinfo.dli_fbase);
5735 #endif
5736 st->cr();
5738 if (Verbose) {
5739 // decode some bytes around the PC
5740 address begin = same_page(addr-40, addr);
5741 address end = same_page(addr+40, addr);
5742 address lowest = (address) dlinfo.dli_sname;
5743 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5744 if (begin < lowest) begin = lowest;
5745 Dl_info dlinfo2;
5746 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
5747 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5748 end = (address) dlinfo2.dli_saddr;
5749 Disassembler::decode(begin, end, st);
5750 }
5751 return true;
5752 }
5753 return false;
5754 }
5756 // Following function has been added to support HotSparc's libjvm.so running
5757 // under Solaris production JDK 1.2.2 / 1.3.0. These came from
5758 // src/solaris/hpi/native_threads in the EVM codebase.
5759 //
5760 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5761 // libraries and should thus be removed. We will leave it behind for a while
5762 // until we no longer want to able to run on top of 1.3.0 Solaris production
5763 // JDK. See 4341971.
5765 #define STACK_SLACK 0x800
5767 extern "C" {
5768 intptr_t sysThreadAvailableStackWithSlack() {
5769 stack_t st;
5770 intptr_t retval, stack_top;
5771 retval = thr_stksegment(&st);
5772 assert(retval == 0, "incorrect return value from thr_stksegment");
5773 assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5774 assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5775 stack_top=(intptr_t)st.ss_sp-st.ss_size;
5776 return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5777 }
5778 }
5780 // Just to get the Kernel build to link on solaris for testing.
5782 extern "C" {
5783 class ASGCT_CallTrace;
5784 void AsyncGetCallTrace(ASGCT_CallTrace *trace, jint depth, void* ucontext)
5785 KERNEL_RETURN;
5786 }
5789 // ObjectMonitor park-unpark infrastructure ...
5790 //
5791 // We implement Solaris and Linux PlatformEvents with the
5792 // obvious condvar-mutex-flag triple.
5793 // Another alternative that works quite well is pipes:
5794 // Each PlatformEvent consists of a pipe-pair.
5795 // The thread associated with the PlatformEvent
5796 // calls park(), which reads from the input end of the pipe.
5797 // Unpark() writes into the other end of the pipe.
5798 // The write-side of the pipe must be set NDELAY.
5799 // Unfortunately pipes consume a large # of handles.
5800 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
5801 // Using pipes for the 1st few threads might be workable, however.
5802 //
5803 // park() is permitted to return spuriously.
5804 // Callers of park() should wrap the call to park() in
5805 // an appropriate loop. A litmus test for the correct
5806 // usage of park is the following: if park() were modified
5807 // to immediately return 0 your code should still work,
5808 // albeit degenerating to a spin loop.
5809 //
5810 // An interesting optimization for park() is to use a trylock()
5811 // to attempt to acquire the mutex. If the trylock() fails
5812 // then we know that a concurrent unpark() operation is in-progress.
5813 // in that case the park() code could simply set _count to 0
5814 // and return immediately. The subsequent park() operation *might*
5815 // return immediately. That's harmless as the caller of park() is
5816 // expected to loop. By using trylock() we will have avoided a
5817 // avoided a context switch caused by contention on the per-thread mutex.
5818 //
5819 // TODO-FIXME:
5820 // 1. Reconcile Doug's JSR166 j.u.c park-unpark with the
5821 // objectmonitor implementation.
5822 // 2. Collapse the JSR166 parker event, and the
5823 // objectmonitor ParkEvent into a single "Event" construct.
5824 // 3. In park() and unpark() add:
5825 // assert (Thread::current() == AssociatedWith).
5826 // 4. add spurious wakeup injection on a -XX:EarlyParkReturn=N switch.
5827 // 1-out-of-N park() operations will return immediately.
5828 //
5829 // _Event transitions in park()
5830 // -1 => -1 : illegal
5831 // 1 => 0 : pass - return immediately
5832 // 0 => -1 : block
5833 //
5834 // _Event serves as a restricted-range semaphore.
5835 //
5836 // Another possible encoding of _Event would be with
5837 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5838 //
5839 // TODO-FIXME: add DTRACE probes for:
5840 // 1. Tx parks
5841 // 2. Ty unparks Tx
5842 // 3. Tx resumes from park
5845 // value determined through experimentation
5846 #define ROUNDINGFIX 11
5848 // utility to compute the abstime argument to timedwait.
5849 // TODO-FIXME: switch from compute_abstime() to unpackTime().
5851 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5852 // millis is the relative timeout time
5853 // abstime will be the absolute timeout time
5854 if (millis < 0) millis = 0;
5855 struct timeval now;
5856 int status = gettimeofday(&now, NULL);
5857 assert(status == 0, "gettimeofday");
5858 jlong seconds = millis / 1000;
5859 jlong max_wait_period;
5861 if (UseLWPSynchronization) {
5862 // forward port of fix for 4275818 (not sleeping long enough)
5863 // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5864 // _lwp_cond_timedwait() used a round_down algorithm rather
5865 // than a round_up. For millis less than our roundfactor
5866 // it rounded down to 0 which doesn't meet the spec.
5867 // For millis > roundfactor we may return a bit sooner, but
5868 // since we can not accurately identify the patch level and
5869 // this has already been fixed in Solaris 9 and 8 we will
5870 // leave it alone rather than always rounding down.
5872 if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5873 // It appears that when we go directly through Solaris _lwp_cond_timedwait()
5874 // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
5875 max_wait_period = 21000000;
5876 } else {
5877 max_wait_period = 50000000;
5878 }
5879 millis %= 1000;
5880 if (seconds > max_wait_period) { // see man cond_timedwait(3T)
5881 seconds = max_wait_period;
5882 }
5883 abstime->tv_sec = now.tv_sec + seconds;
5884 long usec = now.tv_usec + millis * 1000;
5885 if (usec >= 1000000) {
5886 abstime->tv_sec += 1;
5887 usec -= 1000000;
5888 }
5889 abstime->tv_nsec = usec * 1000;
5890 return abstime;
5891 }
5893 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5894 // Conceptually TryPark() should be equivalent to park(0).
5896 int os::PlatformEvent::TryPark() {
5897 for (;;) {
5898 const int v = _Event ;
5899 guarantee ((v == 0) || (v == 1), "invariant") ;
5900 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5901 }
5902 }
5904 void os::PlatformEvent::park() { // AKA: down()
5905 // Invariant: Only the thread associated with the Event/PlatformEvent
5906 // may call park().
5907 int v ;
5908 for (;;) {
5909 v = _Event ;
5910 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5911 }
5912 guarantee (v >= 0, "invariant") ;
5913 if (v == 0) {
5914 // Do this the hard way by blocking ...
5915 // See http://monaco.sfbay/detail.jsf?cr=5094058.
5916 // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5917 // Only for SPARC >= V8PlusA
5918 #if defined(__sparc) && defined(COMPILER2)
5919 if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
5920 #endif
5921 int status = os::Solaris::mutex_lock(_mutex);
5922 assert_status(status == 0, status, "mutex_lock");
5923 guarantee (_nParked == 0, "invariant") ;
5924 ++ _nParked ;
5925 while (_Event < 0) {
5926 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5927 // Treat this the same as if the wait was interrupted
5928 // With usr/lib/lwp going to kernel, always handle ETIME
5929 status = os::Solaris::cond_wait(_cond, _mutex);
5930 if (status == ETIME) status = EINTR ;
5931 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5932 }
5933 -- _nParked ;
5934 _Event = 0 ;
5935 status = os::Solaris::mutex_unlock(_mutex);
5936 assert_status(status == 0, status, "mutex_unlock");
5937 }
5938 }
5940 int os::PlatformEvent::park(jlong millis) {
5941 guarantee (_nParked == 0, "invariant") ;
5942 int v ;
5943 for (;;) {
5944 v = _Event ;
5945 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5946 }
5947 guarantee (v >= 0, "invariant") ;
5948 if (v != 0) return OS_OK ;
5950 int ret = OS_TIMEOUT;
5951 timestruc_t abst;
5952 compute_abstime (&abst, millis);
5954 // See http://monaco.sfbay/detail.jsf?cr=5094058.
5955 // For Solaris SPARC set fprs.FEF=0 prior to parking.
5956 // Only for SPARC >= V8PlusA
5957 #if defined(__sparc) && defined(COMPILER2)
5958 if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
5959 #endif
5960 int status = os::Solaris::mutex_lock(_mutex);
5961 assert_status(status == 0, status, "mutex_lock");
5962 guarantee (_nParked == 0, "invariant") ;
5963 ++ _nParked ;
5964 while (_Event < 0) {
5965 int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5966 assert_status(status == 0 || status == EINTR ||
5967 status == ETIME || status == ETIMEDOUT,
5968 status, "cond_timedwait");
5969 if (!FilterSpuriousWakeups) break ; // previous semantics
5970 if (status == ETIME || status == ETIMEDOUT) break ;
5971 // We consume and ignore EINTR and spurious wakeups.
5972 }
5973 -- _nParked ;
5974 if (_Event >= 0) ret = OS_OK ;
5975 _Event = 0 ;
5976 status = os::Solaris::mutex_unlock(_mutex);
5977 assert_status(status == 0, status, "mutex_unlock");
5978 return ret;
5979 }
5981 void os::PlatformEvent::unpark() {
5982 int v, AnyWaiters;
5984 // Increment _Event.
5985 // Another acceptable implementation would be to simply swap 1
5986 // into _Event:
5987 // if (Swap (&_Event, 1) < 0) {
5988 // mutex_lock (_mutex) ; AnyWaiters = nParked; mutex_unlock (_mutex) ;
5989 // if (AnyWaiters) cond_signal (_cond) ;
5990 // }
5992 for (;;) {
5993 v = _Event ;
5994 if (v > 0) {
5995 // The LD of _Event could have reordered or be satisfied
5996 // by a read-aside from this processor's write buffer.
5997 // To avoid problems execute a barrier and then
5998 // ratify the value. A degenerate CAS() would also work.
5999 // Viz., CAS (v+0, &_Event, v) == v).
6000 OrderAccess::fence() ;
6001 if (_Event == v) return ;
6002 continue ;
6003 }
6004 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
6005 }
6007 // If the thread associated with the event was parked, wake it.
6008 if (v < 0) {
6009 int status ;
6010 // Wait for the thread assoc with the PlatformEvent to vacate.
6011 status = os::Solaris::mutex_lock(_mutex);
6012 assert_status(status == 0, status, "mutex_lock");
6013 AnyWaiters = _nParked ;
6014 status = os::Solaris::mutex_unlock(_mutex);
6015 assert_status(status == 0, status, "mutex_unlock");
6016 guarantee (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
6017 if (AnyWaiters != 0) {
6018 // We intentional signal *after* dropping the lock
6019 // to avoid a common class of futile wakeups.
6020 status = os::Solaris::cond_signal(_cond);
6021 assert_status(status == 0, status, "cond_signal");
6022 }
6023 }
6024 }
6026 // JSR166
6027 // -------------------------------------------------------
6029 /*
6030 * The solaris and linux implementations of park/unpark are fairly
6031 * conservative for now, but can be improved. They currently use a
6032 * mutex/condvar pair, plus _counter.
6033 * Park decrements _counter if > 0, else does a condvar wait. Unpark
6034 * sets count to 1 and signals condvar. Only one thread ever waits
6035 * on the condvar. Contention seen when trying to park implies that someone
6036 * is unparking you, so don't wait. And spurious returns are fine, so there
6037 * is no need to track notifications.
6038 */
6040 #define NANOSECS_PER_SEC 1000000000
6041 #define NANOSECS_PER_MILLISEC 1000000
6042 #define MAX_SECS 100000000
6044 /*
6045 * This code is common to linux and solaris and will be moved to a
6046 * common place in dolphin.
6047 *
6048 * The passed in time value is either a relative time in nanoseconds
6049 * or an absolute time in milliseconds. Either way it has to be unpacked
6050 * into suitable seconds and nanoseconds components and stored in the
6051 * given timespec structure.
6052 * Given time is a 64-bit value and the time_t used in the timespec is only
6053 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6054 * overflow if times way in the future are given. Further on Solaris versions
6055 * prior to 10 there is a restriction (see cond_timedwait) that the specified
6056 * number of seconds, in abstime, is less than current_time + 100,000,000.
6057 * As it will be 28 years before "now + 100000000" will overflow we can
6058 * ignore overflow and just impose a hard-limit on seconds using the value
6059 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6060 * years from "now".
6061 */
6062 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6063 assert (time > 0, "convertTime");
6065 struct timeval now;
6066 int status = gettimeofday(&now, NULL);
6067 assert(status == 0, "gettimeofday");
6069 time_t max_secs = now.tv_sec + MAX_SECS;
6071 if (isAbsolute) {
6072 jlong secs = time / 1000;
6073 if (secs > max_secs) {
6074 absTime->tv_sec = max_secs;
6075 }
6076 else {
6077 absTime->tv_sec = secs;
6078 }
6079 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6080 }
6081 else {
6082 jlong secs = time / NANOSECS_PER_SEC;
6083 if (secs >= MAX_SECS) {
6084 absTime->tv_sec = max_secs;
6085 absTime->tv_nsec = 0;
6086 }
6087 else {
6088 absTime->tv_sec = now.tv_sec + secs;
6089 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6090 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6091 absTime->tv_nsec -= NANOSECS_PER_SEC;
6092 ++absTime->tv_sec; // note: this must be <= max_secs
6093 }
6094 }
6095 }
6096 assert(absTime->tv_sec >= 0, "tv_sec < 0");
6097 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6098 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6099 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6100 }
6102 void Parker::park(bool isAbsolute, jlong time) {
6104 // Optional fast-path check:
6105 // Return immediately if a permit is available.
6106 if (_counter > 0) {
6107 _counter = 0 ;
6108 OrderAccess::fence();
6109 return ;
6110 }
6112 // Optional fast-exit: Check interrupt before trying to wait
6113 Thread* thread = Thread::current();
6114 assert(thread->is_Java_thread(), "Must be JavaThread");
6115 JavaThread *jt = (JavaThread *)thread;
6116 if (Thread::is_interrupted(thread, false)) {
6117 return;
6118 }
6120 // First, demultiplex/decode time arguments
6121 timespec absTime;
6122 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6123 return;
6124 }
6125 if (time > 0) {
6126 // Warning: this code might be exposed to the old Solaris time
6127 // round-down bugs. Grep "roundingFix" for details.
6128 unpackTime(&absTime, isAbsolute, time);
6129 }
6131 // Enter safepoint region
6132 // Beware of deadlocks such as 6317397.
6133 // The per-thread Parker:: _mutex is a classic leaf-lock.
6134 // In particular a thread must never block on the Threads_lock while
6135 // holding the Parker:: mutex. If safepoints are pending both the
6136 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6137 ThreadBlockInVM tbivm(jt);
6139 // Don't wait if cannot get lock since interference arises from
6140 // unblocking. Also. check interrupt before trying wait
6141 if (Thread::is_interrupted(thread, false) ||
6142 os::Solaris::mutex_trylock(_mutex) != 0) {
6143 return;
6144 }
6146 int status ;
6148 if (_counter > 0) { // no wait needed
6149 _counter = 0;
6150 status = os::Solaris::mutex_unlock(_mutex);
6151 assert (status == 0, "invariant") ;
6152 OrderAccess::fence();
6153 return;
6154 }
6156 #ifdef ASSERT
6157 // Don't catch signals while blocked; let the running threads have the signals.
6158 // (This allows a debugger to break into the running thread.)
6159 sigset_t oldsigs;
6160 sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
6161 thr_sigsetmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6162 #endif
6164 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6165 jt->set_suspend_equivalent();
6166 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6168 // Do this the hard way by blocking ...
6169 // See http://monaco.sfbay/detail.jsf?cr=5094058.
6170 // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
6171 // Only for SPARC >= V8PlusA
6172 #if defined(__sparc) && defined(COMPILER2)
6173 if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6174 #endif
6176 if (time == 0) {
6177 status = os::Solaris::cond_wait (_cond, _mutex) ;
6178 } else {
6179 status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
6180 }
6181 // Note that an untimed cond_wait() can sometimes return ETIME on older
6182 // versions of the Solaris.
6183 assert_status(status == 0 || status == EINTR ||
6184 status == ETIME || status == ETIMEDOUT,
6185 status, "cond_timedwait");
6187 #ifdef ASSERT
6188 thr_sigsetmask(SIG_SETMASK, &oldsigs, NULL);
6189 #endif
6190 _counter = 0 ;
6191 status = os::Solaris::mutex_unlock(_mutex);
6192 assert_status(status == 0, status, "mutex_unlock") ;
6194 // If externally suspended while waiting, re-suspend
6195 if (jt->handle_special_suspend_equivalent_condition()) {
6196 jt->java_suspend_self();
6197 }
6198 OrderAccess::fence();
6199 }
6201 void Parker::unpark() {
6202 int s, status ;
6203 status = os::Solaris::mutex_lock (_mutex) ;
6204 assert (status == 0, "invariant") ;
6205 s = _counter;
6206 _counter = 1;
6207 status = os::Solaris::mutex_unlock (_mutex) ;
6208 assert (status == 0, "invariant") ;
6210 if (s < 1) {
6211 status = os::Solaris::cond_signal (_cond) ;
6212 assert (status == 0, "invariant") ;
6213 }
6214 }
6216 extern char** environ;
6218 // Run the specified command in a separate process. Return its exit value,
6219 // or -1 on failure (e.g. can't fork a new process).
6220 // Unlike system(), this function can be called from signal handler. It
6221 // doesn't block SIGINT et al.
6222 int os::fork_and_exec(char* cmd) {
6223 char * argv[4];
6224 argv[0] = (char *)"sh";
6225 argv[1] = (char *)"-c";
6226 argv[2] = cmd;
6227 argv[3] = NULL;
6229 // fork is async-safe, fork1 is not so can't use in signal handler
6230 pid_t pid;
6231 Thread* t = ThreadLocalStorage::get_thread_slow();
6232 if (t != NULL && t->is_inside_signal_handler()) {
6233 pid = fork();
6234 } else {
6235 pid = fork1();
6236 }
6238 if (pid < 0) {
6239 // fork failed
6240 warning("fork failed: %s", strerror(errno));
6241 return -1;
6243 } else if (pid == 0) {
6244 // child process
6246 // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
6247 execve("/usr/bin/sh", argv, environ);
6249 // execve failed
6250 _exit(-1);
6252 } else {
6253 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6254 // care about the actual exit code, for now.
6256 int status;
6258 // Wait for the child process to exit. This returns immediately if
6259 // the child has already exited. */
6260 while (waitpid(pid, &status, 0) < 0) {
6261 switch (errno) {
6262 case ECHILD: return 0;
6263 case EINTR: break;
6264 default: return -1;
6265 }
6266 }
6268 if (WIFEXITED(status)) {
6269 // The child exited normally; get its exit code.
6270 return WEXITSTATUS(status);
6271 } else if (WIFSIGNALED(status)) {
6272 // The child exited because of a signal
6273 // The best value to return is 0x80 + signal number,
6274 // because that is what all Unix shells do, and because
6275 // it allows callers to distinguish between process exit and
6276 // process death by signal.
6277 return 0x80 + WTERMSIG(status);
6278 } else {
6279 // Unknown exit code; pass it through
6280 return status;
6281 }
6282 }
6283 }
6285 // is_headless_jre()
6286 //
6287 // Test for the existence of libmawt in motif21 or xawt directories
6288 // in order to report if we are running in a headless jre
6289 //
6290 bool os::is_headless_jre() {
6291 struct stat statbuf;
6292 char buf[MAXPATHLEN];
6293 char libmawtpath[MAXPATHLEN];
6294 const char *xawtstr = "/xawt/libmawt.so";
6295 const char *motifstr = "/motif21/libmawt.so";
6296 char *p;
6298 // Get path to libjvm.so
6299 os::jvm_path(buf, sizeof(buf));
6301 // Get rid of libjvm.so
6302 p = strrchr(buf, '/');
6303 if (p == NULL) return false;
6304 else *p = '\0';
6306 // Get rid of client or server
6307 p = strrchr(buf, '/');
6308 if (p == NULL) return false;
6309 else *p = '\0';
6311 // check xawt/libmawt.so
6312 strcpy(libmawtpath, buf);
6313 strcat(libmawtpath, xawtstr);
6314 if (::stat(libmawtpath, &statbuf) == 0) return false;
6316 // check motif21/libmawt.so
6317 strcpy(libmawtpath, buf);
6318 strcat(libmawtpath, motifstr);
6319 if (::stat(libmawtpath, &statbuf) == 0) return false;
6321 return true;
6322 }
6324 size_t os::write(int fd, const void *buf, unsigned int nBytes) {
6325 INTERRUPTIBLE_RETURN_INT(::write(fd, buf, nBytes), os::Solaris::clear_interrupted);
6326 }
6328 int os::close(int fd) {
6329 RESTARTABLE_RETURN_INT(::close(fd));
6330 }
6332 int os::socket_close(int fd) {
6333 RESTARTABLE_RETURN_INT(::close(fd));
6334 }
6336 int os::recv(int fd, char *buf, int nBytes, int flags) {
6337 INTERRUPTIBLE_RETURN_INT(::recv(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6338 }
6341 int os::send(int fd, char *buf, int nBytes, int flags) {
6342 INTERRUPTIBLE_RETURN_INT(::send(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6343 }
6345 int os::raw_send(int fd, char *buf, int nBytes, int flags) {
6346 RESTARTABLE_RETURN_INT(::send(fd, buf, nBytes, flags));
6347 }
6349 // As both poll and select can be interrupted by signals, we have to be
6350 // prepared to restart the system call after updating the timeout, unless
6351 // a poll() is done with timeout == -1, in which case we repeat with this
6352 // "wait forever" value.
6354 int os::timeout(int fd, long timeout) {
6355 int res;
6356 struct timeval t;
6357 julong prevtime, newtime;
6358 static const char* aNull = 0;
6359 struct pollfd pfd;
6360 pfd.fd = fd;
6361 pfd.events = POLLIN;
6363 gettimeofday(&t, &aNull);
6364 prevtime = ((julong)t.tv_sec * 1000) + t.tv_usec / 1000;
6366 for(;;) {
6367 INTERRUPTIBLE_NORESTART(::poll(&pfd, 1, timeout), res, os::Solaris::clear_interrupted);
6368 if(res == OS_ERR && errno == EINTR) {
6369 if(timeout != -1) {
6370 gettimeofday(&t, &aNull);
6371 newtime = ((julong)t.tv_sec * 1000) + t.tv_usec /1000;
6372 timeout -= newtime - prevtime;
6373 if(timeout <= 0)
6374 return OS_OK;
6375 prevtime = newtime;
6376 }
6377 } else return res;
6378 }
6379 }
6381 int os::connect(int fd, struct sockaddr *him, int len) {
6382 int _result;
6383 INTERRUPTIBLE_NORESTART(::connect(fd, him, len), _result,
6384 os::Solaris::clear_interrupted);
6386 // Depending on when thread interruption is reset, _result could be
6387 // one of two values when errno == EINTR
6389 if (((_result == OS_INTRPT) || (_result == OS_ERR))
6390 && (errno == EINTR)) {
6391 /* restarting a connect() changes its errno semantics */
6392 INTERRUPTIBLE(::connect(fd, him, len), _result,
6393 os::Solaris::clear_interrupted);
6394 /* undo these changes */
6395 if (_result == OS_ERR) {
6396 if (errno == EALREADY) {
6397 errno = EINPROGRESS; /* fall through */
6398 } else if (errno == EISCONN) {
6399 errno = 0;
6400 return OS_OK;
6401 }
6402 }
6403 }
6404 return _result;
6405 }
6407 int os::accept(int fd, struct sockaddr *him, int *len) {
6408 if (fd < 0)
6409 return OS_ERR;
6410 INTERRUPTIBLE_RETURN_INT((int)::accept(fd, him,\
6411 (socklen_t*) len), os::Solaris::clear_interrupted);
6412 }
6414 int os::recvfrom(int fd, char *buf, int nBytes, int flags,
6415 sockaddr *from, int *fromlen) {
6416 //%%note jvm_r11
6417 INTERRUPTIBLE_RETURN_INT((int)::recvfrom(fd, buf, nBytes,\
6418 flags, from, fromlen), os::Solaris::clear_interrupted);
6419 }
6421 int os::sendto(int fd, char *buf, int len, int flags,
6422 struct sockaddr *to, int tolen) {
6423 //%%note jvm_r11
6424 INTERRUPTIBLE_RETURN_INT((int)::sendto(fd, buf, len, flags,\
6425 to, tolen), os::Solaris::clear_interrupted);
6426 }
6428 int os::socket_available(int fd, jint *pbytes) {
6429 if (fd < 0)
6430 return OS_OK;
6432 int ret;
6434 RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
6436 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
6437 // is expected to return 0 on failure and 1 on success to the jdk.
6439 return (ret == OS_ERR) ? 0 : 1;
6440 }
6443 int os::bind(int fd, struct sockaddr *him, int len) {
6444 INTERRUPTIBLE_RETURN_INT_NORESTART(::bind(fd, him, len),\
6445 os::Solaris::clear_interrupted);
6446 }