1 /*
2 * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "jvm.h"
26 #include "logging/log.hpp"
27 #include "memory/allocation.inline.hpp"
28 #include "utilities/globalDefinitions.hpp"
29 #include "runtime/frame.inline.hpp"
30 #include "runtime/interfaceSupport.inline.hpp"
31 #include "runtime/os.hpp"
32 #include "services/memTracker.hpp"
33 #include "utilities/align.hpp"
34 #include "utilities/formatBuffer.hpp"
35 #include "utilities/macros.hpp"
36 #include "utilities/vmError.hpp"
37
38 #include <dlfcn.h>
39 #include <pthread.h>
40 #include <signal.h>
41 #include <sys/mman.h>
42 #include <sys/resource.h>
43 #include <sys/utsname.h>
44 #include <time.h>
45 #include <unistd.h>
46
47 // Todo: provide a os::get_max_process_id() or similar. Number of processes
48 // may have been configured, can be read more accurately from proc fs etc.
49 #ifndef MAX_PID
50 #define MAX_PID INT_MAX
51 #endif
52 #define IS_VALID_PID(p) (p > 0 && p < MAX_PID)
53
54 #ifndef MAP_ANONYMOUS
55 #define MAP_ANONYMOUS MAP_ANON
56 #endif
57
58 #define check_with_errno(check_type, cond, msg) \
59 do { \
60 int err = errno; \
61 check_type(cond, "%s; error='%s' (errno=%s)", msg, os::strerror(err), \
62 os::errno_name(err)); \
63 } while (false)
64
65 #define assert_with_errno(cond, msg) check_with_errno(assert, cond, msg)
66 #define guarantee_with_errno(cond, msg) check_with_errno(guarantee, cond, msg)
67
68 // Check core dump limit and report possible place where core can be found
69 void os::check_dump_limit(char* buffer, size_t bufferSize) {
70 if (!FLAG_IS_DEFAULT(CreateCoredumpOnCrash) && !CreateCoredumpOnCrash) {
71 jio_snprintf(buffer, bufferSize, "CreateCoredumpOnCrash is disabled from command line");
72 VMError::record_coredump_status(buffer, false);
73 return;
74 }
75
76 int n;
77 struct rlimit rlim;
78 bool success;
79
80 char core_path[PATH_MAX];
81 n = get_core_path(core_path, PATH_MAX);
82
83 if (n <= 0) {
84 jio_snprintf(buffer, bufferSize, "core.%d (may not exist)", current_process_id());
85 success = true;
86 #ifdef LINUX
87 } else if (core_path[0] == '"') { // redirect to user process
88 jio_snprintf(buffer, bufferSize, "Core dumps may be processed with %s", core_path);
89 success = true;
90 #endif
91 } else if (getrlimit(RLIMIT_CORE, &rlim) != 0) {
92 jio_snprintf(buffer, bufferSize, "%s (may not exist)", core_path);
93 success = true;
94 } else {
95 switch(rlim.rlim_cur) {
96 case RLIM_INFINITY:
97 jio_snprintf(buffer, bufferSize, "%s", core_path);
98 success = true;
99 break;
100 case 0:
101 jio_snprintf(buffer, bufferSize, "Core dumps have been disabled. To enable core dumping, try \"ulimit -c unlimited\" before starting Java again");
102 success = false;
103 break;
104 default:
105 jio_snprintf(buffer, bufferSize, "%s (max size " UINT64_FORMAT " kB). To ensure a full core dump, try \"ulimit -c unlimited\" before starting Java again", core_path, uint64_t(rlim.rlim_cur) / 1024);
106 success = true;
107 break;
108 }
109 }
110
111 VMError::record_coredump_status(buffer, success);
112 }
113
114 int os::get_native_stack(address* stack, int frames, int toSkip) {
115 int frame_idx = 0;
116 int num_of_frames; // number of frames captured
117 frame fr = os::current_frame();
118 while (fr.pc() && frame_idx < frames) {
119 if (toSkip > 0) {
120 toSkip --;
121 } else {
122 stack[frame_idx ++] = fr.pc();
123 }
124 if (fr.fp() == NULL || fr.cb() != NULL ||
125 fr.sender_pc() == NULL || os::is_first_C_frame(&fr)) break;
126
127 if (fr.sender_pc() && !os::is_first_C_frame(&fr)) {
128 fr = os::get_sender_for_C_frame(&fr);
129 } else {
130 break;
131 }
132 }
133 num_of_frames = frame_idx;
134 for (; frame_idx < frames; frame_idx ++) {
135 stack[frame_idx] = NULL;
136 }
137
138 return num_of_frames;
139 }
140
141
142 bool os::unsetenv(const char* name) {
143 assert(name != NULL, "Null pointer");
144 return (::unsetenv(name) == 0);
145 }
146
147 int os::get_last_error() {
148 return errno;
149 }
150
151 bool os::is_debugger_attached() {
152 // not implemented
153 return false;
154 }
155
156 void os::wait_for_keypress_at_exit(void) {
157 // don't do anything on posix platforms
158 return;
159 }
160
161 int os::create_file_for_heap(const char* dir) {
162
163 const char name_template[] = "/jvmheap.XXXXXX";
164
165 char *fullname = (char*)os::malloc((strlen(dir) + strlen(name_template) + 1), mtInternal);
166 if (fullname == NULL) {
167 vm_exit_during_initialization(err_msg("Malloc failed during creation of backing file for heap (%s)", os::strerror(errno)));
168 return -1;
169 }
170 (void)strncpy(fullname, dir, strlen(dir)+1);
171 (void)strncat(fullname, name_template, strlen(name_template));
172
173 os::native_path(fullname);
174
175 sigset_t set, oldset;
176 int ret = sigfillset(&set);
177 assert_with_errno(ret == 0, "sigfillset returned error");
178
179 // set the file creation mask.
180 mode_t file_mode = S_IRUSR | S_IWUSR;
181
182 // create a new file.
183 int fd = mkstemp(fullname);
184
185 if (fd < 0) {
186 warning("Could not create file for heap with template %s", fullname);
187 os::free(fullname);
188 return -1;
189 }
190
191 // delete the name from the filesystem. When 'fd' is closed, the file (and space) will be deleted.
192 ret = unlink(fullname);
193 assert_with_errno(ret == 0, "unlink returned error");
194
195 os::free(fullname);
196 return fd;
197 }
198
199 static char* reserve_mmapped_memory(size_t bytes, char* requested_addr) {
200 char * addr;
201 int flags = MAP_PRIVATE NOT_AIX( | MAP_NORESERVE ) | MAP_ANONYMOUS;
202 if (requested_addr != NULL) {
203 assert((uintptr_t)requested_addr % os::vm_page_size() == 0, "Requested address should be aligned to OS page size");
204 flags |= MAP_FIXED;
205 }
206
207 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
208 // touch an uncommitted page. Otherwise, the read/write might
209 // succeed if we have enough swap space to back the physical page.
210 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
211 flags, -1, 0);
212
213 if (addr != MAP_FAILED) {
214 MemTracker::record_virtual_memory_reserve((address)addr, bytes, CALLER_PC);
215 return addr;
216 }
217 return NULL;
218 }
219
220 static int util_posix_fallocate(int fd, off_t offset, off_t len) {
221 #ifdef __APPLE__
222 fstore_t store = { F_ALLOCATECONTIG, F_PEOFPOSMODE, 0, len };
223 // First we try to get a continuous chunk of disk space
224 int ret = fcntl(fd, F_PREALLOCATE, &store);
225 if (ret == -1) {
226 // Maybe we are too fragmented, try to allocate non-continuous range
227 store.fst_flags = F_ALLOCATEALL;
228 ret = fcntl(fd, F_PREALLOCATE, &store);
229 }
230 if(ret != -1) {
231 return ftruncate(fd, len);
232 }
233 return -1;
234 #else
235 return posix_fallocate(fd, offset, len);
236 #endif
237 }
238
239 // Map the given address range to the provided file descriptor.
240 char* os::map_memory_to_file(char* base, size_t size, int fd) {
241 assert(fd != -1, "File descriptor is not valid");
242
243 // allocate space for the file
244 int ret = util_posix_fallocate(fd, 0, (off_t)size);
245 if (ret != 0) {
246 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory. error(%d)", ret));
247 return NULL;
248 }
249
250 int prot = PROT_READ | PROT_WRITE;
251 int flags = MAP_SHARED;
252 if (base != NULL) {
253 flags |= MAP_FIXED;
254 }
255 char* addr = (char*)mmap(base, size, prot, flags, fd, 0);
256
257 if (addr == MAP_FAILED) {
258 warning("Failed mmap to file. (%s)", os::strerror(errno));
259 return NULL;
260 }
261 if (base != NULL && addr != base) {
262 if (!os::release_memory(addr, size)) {
263 warning("Could not release memory on unsuccessful file mapping");
264 }
265 return NULL;
266 }
267 return addr;
268 }
269
270 char* os::replace_existing_mapping_with_file_mapping(char* base, size_t size, int fd) {
271 assert(fd != -1, "File descriptor is not valid");
272 assert(base != NULL, "Base cannot be NULL");
273
274 return map_memory_to_file(base, size, fd);
275 }
276
277 // Multiple threads can race in this code, and can remap over each other with MAP_FIXED,
278 // so on posix, unmap the section at the start and at the end of the chunk that we mapped
279 // rather than unmapping and remapping the whole chunk to get requested alignment.
280 char* os::reserve_memory_aligned(size_t size, size_t alignment, int file_desc) {
281 assert((alignment & (os::vm_allocation_granularity() - 1)) == 0,
282 "Alignment must be a multiple of allocation granularity (page size)");
283 assert((size & (alignment -1)) == 0, "size must be 'alignment' aligned");
284
285 size_t extra_size = size + alignment;
286 assert(extra_size >= size, "overflow, size is too large to allow alignment");
287
288 char* extra_base;
289 if (file_desc != -1) {
290 // For file mapping, we do not call os:reserve_memory(extra_size, NULL, alignment, file_desc) because
291 // we need to deal with shrinking of the file space later when we release extra memory after alignment.
292 // We also cannot called os:reserve_memory() with file_desc set to -1 because on aix we might get SHM memory.
293 // So here to call a helper function while reserve memory for us. After we have a aligned base,
294 // we will replace anonymous mapping with file mapping.
295 extra_base = reserve_mmapped_memory(extra_size, NULL);
296 if (extra_base != NULL) {
297 MemTracker::record_virtual_memory_reserve((address)extra_base, extra_size, CALLER_PC);
298 }
299 } else {
300 extra_base = os::reserve_memory(extra_size, NULL, alignment);
301 }
302
303 if (extra_base == NULL) {
304 return NULL;
305 }
306
307 // Do manual alignment
308 char* aligned_base = align_up(extra_base, alignment);
309
310 // [ | | ]
311 // ^ extra_base
312 // ^ extra_base + begin_offset == aligned_base
313 // extra_base + begin_offset + size ^
314 // extra_base + extra_size ^
315 // |<>| == begin_offset
316 // end_offset == |<>|
317 size_t begin_offset = aligned_base - extra_base;
318 size_t end_offset = (extra_base + extra_size) - (aligned_base + size);
319
320 if (begin_offset > 0) {
321 os::release_memory(extra_base, begin_offset);
322 }
323
324 if (end_offset > 0) {
325 os::release_memory(extra_base + begin_offset + size, end_offset);
326 }
327
328 if (file_desc != -1) {
329 // After we have an aligned address, we can replace anonymous mapping with file mapping
330 if (replace_existing_mapping_with_file_mapping(aligned_base, size, file_desc) == NULL) {
331 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
332 }
333 MemTracker::record_virtual_memory_commit((address)aligned_base, size, CALLER_PC);
334 }
335 return aligned_base;
336 }
337
338 int os::vsnprintf(char* buf, size_t len, const char* fmt, va_list args) {
339 // All supported POSIX platforms provide C99 semantics.
340 int result = ::vsnprintf(buf, len, fmt, args);
341 // If an encoding error occurred (result < 0) then it's not clear
342 // whether the buffer is NUL terminated, so ensure it is.
343 if ((result < 0) && (len > 0)) {
344 buf[len - 1] = '\0';
345 }
346 return result;
347 }
348
349 int os::get_fileno(FILE* fp) {
350 return NOT_AIX(::)fileno(fp);
351 }
352
353 struct tm* os::gmtime_pd(const time_t* clock, struct tm* res) {
354 return gmtime_r(clock, res);
355 }
356
357 void os::Posix::print_load_average(outputStream* st) {
358 st->print("load average:");
359 double loadavg[3];
360 os::loadavg(loadavg, 3);
361 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
362 st->cr();
363 }
364
365 void os::Posix::print_rlimit_info(outputStream* st) {
366 st->print("rlimit:");
367 struct rlimit rlim;
368
369 st->print(" STACK ");
370 getrlimit(RLIMIT_STACK, &rlim);
371 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
372 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
373
374 st->print(", CORE ");
375 getrlimit(RLIMIT_CORE, &rlim);
376 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
377 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
378
379 // Isn't there on solaris
380 #if defined(AIX)
381 st->print(", NPROC ");
382 st->print("%d", sysconf(_SC_CHILD_MAX));
383 #elif !defined(SOLARIS)
384 st->print(", NPROC ");
385 getrlimit(RLIMIT_NPROC, &rlim);
386 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
387 else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
388 #endif
389
390 st->print(", NOFILE ");
391 getrlimit(RLIMIT_NOFILE, &rlim);
392 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
393 else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
394
395 st->print(", AS ");
396 getrlimit(RLIMIT_AS, &rlim);
397 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
398 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
399
400 st->print(", DATA ");
401 getrlimit(RLIMIT_DATA, &rlim);
402 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
403 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
404
405 st->print(", FSIZE ");
406 getrlimit(RLIMIT_FSIZE, &rlim);
407 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
408 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
409
410 st->cr();
411 }
412
413 void os::Posix::print_uname_info(outputStream* st) {
414 // kernel
415 st->print("uname:");
416 struct utsname name;
417 uname(&name);
418 st->print("%s ", name.sysname);
419 #ifdef ASSERT
420 st->print("%s ", name.nodename);
421 #endif
422 st->print("%s ", name.release);
423 st->print("%s ", name.version);
424 st->print("%s", name.machine);
425 st->cr();
426 }
427
428 bool os::get_host_name(char* buf, size_t buflen) {
429 struct utsname name;
430 uname(&name);
431 jio_snprintf(buf, buflen, "%s", name.nodename);
432 return true;
433 }
434
435 bool os::has_allocatable_memory_limit(julong* limit) {
436 struct rlimit rlim;
437 int getrlimit_res = getrlimit(RLIMIT_AS, &rlim);
438 // if there was an error when calling getrlimit, assume that there is no limitation
439 // on virtual memory.
440 bool result;
441 if ((getrlimit_res != 0) || (rlim.rlim_cur == RLIM_INFINITY)) {
442 result = false;
443 } else {
444 *limit = (julong)rlim.rlim_cur;
445 result = true;
446 }
447 #ifdef _LP64
448 return result;
449 #else
450 // arbitrary virtual space limit for 32 bit Unices found by testing. If
451 // getrlimit above returned a limit, bound it with this limit. Otherwise
452 // directly use it.
453 const julong max_virtual_limit = (julong)3800*M;
454 if (result) {
455 *limit = MIN2(*limit, max_virtual_limit);
456 } else {
457 *limit = max_virtual_limit;
458 }
459
460 // bound by actually allocatable memory. The algorithm uses two bounds, an
461 // upper and a lower limit. The upper limit is the current highest amount of
462 // memory that could not be allocated, the lower limit is the current highest
463 // amount of memory that could be allocated.
464 // The algorithm iteratively refines the result by halving the difference
465 // between these limits, updating either the upper limit (if that value could
466 // not be allocated) or the lower limit (if the that value could be allocated)
467 // until the difference between these limits is "small".
468
469 // the minimum amount of memory we care about allocating.
470 const julong min_allocation_size = M;
471
472 julong upper_limit = *limit;
473
474 // first check a few trivial cases
475 if (is_allocatable(upper_limit) || (upper_limit <= min_allocation_size)) {
476 *limit = upper_limit;
477 } else if (!is_allocatable(min_allocation_size)) {
478 // we found that not even min_allocation_size is allocatable. Return it
479 // anyway. There is no point to search for a better value any more.
480 *limit = min_allocation_size;
481 } else {
482 // perform the binary search.
483 julong lower_limit = min_allocation_size;
484 while ((upper_limit - lower_limit) > min_allocation_size) {
485 julong temp_limit = ((upper_limit - lower_limit) / 2) + lower_limit;
486 temp_limit = align_down(temp_limit, min_allocation_size);
487 if (is_allocatable(temp_limit)) {
488 lower_limit = temp_limit;
489 } else {
490 upper_limit = temp_limit;
491 }
492 }
493 *limit = lower_limit;
494 }
495 return true;
496 #endif
497 }
498
499 const char* os::get_current_directory(char *buf, size_t buflen) {
500 return getcwd(buf, buflen);
501 }
502
503 FILE* os::open(int fd, const char* mode) {
504 return ::fdopen(fd, mode);
505 }
506
507 void os::flockfile(FILE* fp) {
508 ::flockfile(fp);
509 }
510
511 void os::funlockfile(FILE* fp) {
512 ::funlockfile(fp);
513 }
514
515 // Builds a platform dependent Agent_OnLoad_<lib_name> function name
516 // which is used to find statically linked in agents.
517 // Parameters:
518 // sym_name: Symbol in library we are looking for
519 // lib_name: Name of library to look in, NULL for shared libs.
520 // is_absolute_path == true if lib_name is absolute path to agent
521 // such as "/a/b/libL.so"
522 // == false if only the base name of the library is passed in
523 // such as "L"
524 char* os::build_agent_function_name(const char *sym_name, const char *lib_name,
525 bool is_absolute_path) {
526 char *agent_entry_name;
527 size_t len;
528 size_t name_len;
529 size_t prefix_len = strlen(JNI_LIB_PREFIX);
530 size_t suffix_len = strlen(JNI_LIB_SUFFIX);
531 const char *start;
532
533 if (lib_name != NULL) {
534 name_len = strlen(lib_name);
535 if (is_absolute_path) {
536 // Need to strip path, prefix and suffix
537 if ((start = strrchr(lib_name, *os::file_separator())) != NULL) {
538 lib_name = ++start;
539 }
540 if (strlen(lib_name) <= (prefix_len + suffix_len)) {
541 return NULL;
542 }
543 lib_name += prefix_len;
544 name_len = strlen(lib_name) - suffix_len;
545 }
546 }
547 len = (lib_name != NULL ? name_len : 0) + strlen(sym_name) + 2;
548 agent_entry_name = NEW_C_HEAP_ARRAY_RETURN_NULL(char, len, mtThread);
549 if (agent_entry_name == NULL) {
550 return NULL;
551 }
552 strcpy(agent_entry_name, sym_name);
553 if (lib_name != NULL) {
554 strcat(agent_entry_name, "_");
555 strncat(agent_entry_name, lib_name, name_len);
556 }
557 return agent_entry_name;
558 }
559
560 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
561 assert(thread == Thread::current(), "thread consistency check");
562
563 ParkEvent * const slp = thread->_SleepEvent ;
564 slp->reset() ;
565 OrderAccess::fence() ;
566
567 if (interruptible) {
568 jlong prevtime = javaTimeNanos();
569
570 for (;;) {
571 if (os::is_interrupted(thread, true)) {
572 return OS_INTRPT;
573 }
574
575 jlong newtime = javaTimeNanos();
576
577 if (newtime - prevtime < 0) {
578 // time moving backwards, should only happen if no monotonic clock
579 // not a guarantee() because JVM should not abort on kernel/glibc bugs
580 assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected in os::sleep(interruptible)");
581 } else {
582 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
583 }
584
585 if (millis <= 0) {
586 return OS_OK;
587 }
588
589 prevtime = newtime;
590
591 {
592 assert(thread->is_Java_thread(), "sanity check");
593 JavaThread *jt = (JavaThread *) thread;
594 ThreadBlockInVM tbivm(jt);
595 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
596
597 jt->set_suspend_equivalent();
598 // cleared by handle_special_suspend_equivalent_condition() or
599 // java_suspend_self() via check_and_wait_while_suspended()
600
601 slp->park(millis);
602
603 // were we externally suspended while we were waiting?
604 jt->check_and_wait_while_suspended();
605 }
606 }
607 } else {
608 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
609 jlong prevtime = javaTimeNanos();
610
611 for (;;) {
612 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
613 // the 1st iteration ...
614 jlong newtime = javaTimeNanos();
615
616 if (newtime - prevtime < 0) {
617 // time moving backwards, should only happen if no monotonic clock
618 // not a guarantee() because JVM should not abort on kernel/glibc bugs
619 assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected on os::sleep(!interruptible)");
620 } else {
621 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
622 }
623
624 if (millis <= 0) break ;
625
626 prevtime = newtime;
627 slp->park(millis);
628 }
629 return OS_OK ;
630 }
631 }
632
633 ////////////////////////////////////////////////////////////////////////////////
634 // interrupt support
635
636 void os::interrupt(Thread* thread) {
637 debug_only(Thread::check_for_dangling_thread_pointer(thread);)
638
639 OSThread* osthread = thread->osthread();
640
641 if (!osthread->interrupted()) {
642 osthread->set_interrupted(true);
643 // More than one thread can get here with the same value of osthread,
644 // resulting in multiple notifications. We do, however, want the store
645 // to interrupted() to be visible to other threads before we execute unpark().
646 OrderAccess::fence();
647 ParkEvent * const slp = thread->_SleepEvent ;
648 if (slp != NULL) slp->unpark() ;
649 }
650
651 // For JSR166. Unpark even if interrupt status already was set
652 if (thread->is_Java_thread())
653 ((JavaThread*)thread)->parker()->unpark();
654
655 ParkEvent * ev = thread->_ParkEvent ;
656 if (ev != NULL) ev->unpark() ;
657 }
658
659 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
660 debug_only(Thread::check_for_dangling_thread_pointer(thread);)
661
662 OSThread* osthread = thread->osthread();
663
664 bool interrupted = osthread->interrupted();
665
666 // NOTE that since there is no "lock" around the interrupt and
667 // is_interrupted operations, there is the possibility that the
668 // interrupted flag (in osThread) will be "false" but that the
669 // low-level events will be in the signaled state. This is
670 // intentional. The effect of this is that Object.wait() and
671 // LockSupport.park() will appear to have a spurious wakeup, which
672 // is allowed and not harmful, and the possibility is so rare that
673 // it is not worth the added complexity to add yet another lock.
674 // For the sleep event an explicit reset is performed on entry
675 // to os::sleep, so there is no early return. It has also been
676 // recommended not to put the interrupted flag into the "event"
677 // structure because it hides the issue.
678 if (interrupted && clear_interrupted) {
679 osthread->set_interrupted(false);
680 // consider thread->_SleepEvent->reset() ... optional optimization
681 }
682
683 return interrupted;
684 }
685
686
687
688 static const struct {
689 int sig; const char* name;
690 }
691 g_signal_info[] =
692 {
693 { SIGABRT, "SIGABRT" },
694 #ifdef SIGAIO
695 { SIGAIO, "SIGAIO" },
696 #endif
697 { SIGALRM, "SIGALRM" },
698 #ifdef SIGALRM1
699 { SIGALRM1, "SIGALRM1" },
700 #endif
701 { SIGBUS, "SIGBUS" },
702 #ifdef SIGCANCEL
703 { SIGCANCEL, "SIGCANCEL" },
704 #endif
705 { SIGCHLD, "SIGCHLD" },
706 #ifdef SIGCLD
707 { SIGCLD, "SIGCLD" },
708 #endif
709 { SIGCONT, "SIGCONT" },
710 #ifdef SIGCPUFAIL
711 { SIGCPUFAIL, "SIGCPUFAIL" },
712 #endif
713 #ifdef SIGDANGER
714 { SIGDANGER, "SIGDANGER" },
715 #endif
716 #ifdef SIGDIL
717 { SIGDIL, "SIGDIL" },
718 #endif
719 #ifdef SIGEMT
720 { SIGEMT, "SIGEMT" },
721 #endif
722 { SIGFPE, "SIGFPE" },
723 #ifdef SIGFREEZE
724 { SIGFREEZE, "SIGFREEZE" },
725 #endif
726 #ifdef SIGGFAULT
727 { SIGGFAULT, "SIGGFAULT" },
728 #endif
729 #ifdef SIGGRANT
730 { SIGGRANT, "SIGGRANT" },
731 #endif
732 { SIGHUP, "SIGHUP" },
733 { SIGILL, "SIGILL" },
734 { SIGINT, "SIGINT" },
735 #ifdef SIGIO
736 { SIGIO, "SIGIO" },
737 #endif
738 #ifdef SIGIOINT
739 { SIGIOINT, "SIGIOINT" },
740 #endif
741 #ifdef SIGIOT
742 // SIGIOT is there for BSD compatibility, but on most Unices just a
743 // synonym for SIGABRT. The result should be "SIGABRT", not
744 // "SIGIOT".
745 #if (SIGIOT != SIGABRT )
746 { SIGIOT, "SIGIOT" },
747 #endif
748 #endif
749 #ifdef SIGKAP
750 { SIGKAP, "SIGKAP" },
751 #endif
752 { SIGKILL, "SIGKILL" },
753 #ifdef SIGLOST
754 { SIGLOST, "SIGLOST" },
755 #endif
756 #ifdef SIGLWP
757 { SIGLWP, "SIGLWP" },
758 #endif
759 #ifdef SIGLWPTIMER
760 { SIGLWPTIMER, "SIGLWPTIMER" },
761 #endif
762 #ifdef SIGMIGRATE
763 { SIGMIGRATE, "SIGMIGRATE" },
764 #endif
765 #ifdef SIGMSG
766 { SIGMSG, "SIGMSG" },
767 #endif
768 { SIGPIPE, "SIGPIPE" },
769 #ifdef SIGPOLL
770 { SIGPOLL, "SIGPOLL" },
771 #endif
772 #ifdef SIGPRE
773 { SIGPRE, "SIGPRE" },
774 #endif
775 { SIGPROF, "SIGPROF" },
776 #ifdef SIGPTY
777 { SIGPTY, "SIGPTY" },
778 #endif
779 #ifdef SIGPWR
780 { SIGPWR, "SIGPWR" },
781 #endif
782 { SIGQUIT, "SIGQUIT" },
783 #ifdef SIGRECONFIG
784 { SIGRECONFIG, "SIGRECONFIG" },
785 #endif
786 #ifdef SIGRECOVERY
787 { SIGRECOVERY, "SIGRECOVERY" },
788 #endif
789 #ifdef SIGRESERVE
790 { SIGRESERVE, "SIGRESERVE" },
791 #endif
792 #ifdef SIGRETRACT
793 { SIGRETRACT, "SIGRETRACT" },
794 #endif
795 #ifdef SIGSAK
796 { SIGSAK, "SIGSAK" },
797 #endif
798 { SIGSEGV, "SIGSEGV" },
799 #ifdef SIGSOUND
800 { SIGSOUND, "SIGSOUND" },
801 #endif
802 #ifdef SIGSTKFLT
803 { SIGSTKFLT, "SIGSTKFLT" },
804 #endif
805 { SIGSTOP, "SIGSTOP" },
806 { SIGSYS, "SIGSYS" },
807 #ifdef SIGSYSERROR
808 { SIGSYSERROR, "SIGSYSERROR" },
809 #endif
810 #ifdef SIGTALRM
811 { SIGTALRM, "SIGTALRM" },
812 #endif
813 { SIGTERM, "SIGTERM" },
814 #ifdef SIGTHAW
815 { SIGTHAW, "SIGTHAW" },
816 #endif
817 { SIGTRAP, "SIGTRAP" },
818 #ifdef SIGTSTP
819 { SIGTSTP, "SIGTSTP" },
820 #endif
821 { SIGTTIN, "SIGTTIN" },
822 { SIGTTOU, "SIGTTOU" },
823 #ifdef SIGURG
824 { SIGURG, "SIGURG" },
825 #endif
826 { SIGUSR1, "SIGUSR1" },
827 { SIGUSR2, "SIGUSR2" },
828 #ifdef SIGVIRT
829 { SIGVIRT, "SIGVIRT" },
830 #endif
831 { SIGVTALRM, "SIGVTALRM" },
832 #ifdef SIGWAITING
833 { SIGWAITING, "SIGWAITING" },
834 #endif
835 #ifdef SIGWINCH
836 { SIGWINCH, "SIGWINCH" },
837 #endif
838 #ifdef SIGWINDOW
839 { SIGWINDOW, "SIGWINDOW" },
840 #endif
841 { SIGXCPU, "SIGXCPU" },
842 { SIGXFSZ, "SIGXFSZ" },
843 #ifdef SIGXRES
844 { SIGXRES, "SIGXRES" },
845 #endif
846 { -1, NULL }
847 };
848
849 // Returned string is a constant. For unknown signals "UNKNOWN" is returned.
850 const char* os::Posix::get_signal_name(int sig, char* out, size_t outlen) {
851
852 const char* ret = NULL;
853
854 #ifdef SIGRTMIN
855 if (sig >= SIGRTMIN && sig <= SIGRTMAX) {
856 if (sig == SIGRTMIN) {
857 ret = "SIGRTMIN";
858 } else if (sig == SIGRTMAX) {
859 ret = "SIGRTMAX";
860 } else {
861 jio_snprintf(out, outlen, "SIGRTMIN+%d", sig - SIGRTMIN);
862 return out;
863 }
864 }
865 #endif
866
867 if (sig > 0) {
868 for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
869 if (g_signal_info[idx].sig == sig) {
870 ret = g_signal_info[idx].name;
871 break;
872 }
873 }
874 }
875
876 if (!ret) {
877 if (!is_valid_signal(sig)) {
878 ret = "INVALID";
879 } else {
880 ret = "UNKNOWN";
881 }
882 }
883
884 if (out && outlen > 0) {
885 strncpy(out, ret, outlen);
886 out[outlen - 1] = '\0';
887 }
888 return out;
889 }
890
891 int os::Posix::get_signal_number(const char* signal_name) {
892 char tmp[30];
893 const char* s = signal_name;
894 if (s[0] != 'S' || s[1] != 'I' || s[2] != 'G') {
895 jio_snprintf(tmp, sizeof(tmp), "SIG%s", signal_name);
896 s = tmp;
897 }
898 for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
899 if (strcmp(g_signal_info[idx].name, s) == 0) {
900 return g_signal_info[idx].sig;
901 }
902 }
903 return -1;
904 }
905
906 int os::get_signal_number(const char* signal_name) {
907 return os::Posix::get_signal_number(signal_name);
908 }
909
910 // Returns true if signal number is valid.
911 bool os::Posix::is_valid_signal(int sig) {
912 // MacOS not really POSIX compliant: sigaddset does not return
913 // an error for invalid signal numbers. However, MacOS does not
914 // support real time signals and simply seems to have just 33
915 // signals with no holes in the signal range.
916 #ifdef __APPLE__
917 return sig >= 1 && sig < NSIG;
918 #else
919 // Use sigaddset to check for signal validity.
920 sigset_t set;
921 sigemptyset(&set);
922 if (sigaddset(&set, sig) == -1 && errno == EINVAL) {
923 return false;
924 }
925 return true;
926 #endif
927 }
928
929 // Returns:
930 // NULL for an invalid signal number
931 // "SIG<num>" for a valid but unknown signal number
932 // signal name otherwise.
933 const char* os::exception_name(int sig, char* buf, size_t size) {
934 if (!os::Posix::is_valid_signal(sig)) {
935 return NULL;
936 }
937 const char* const name = os::Posix::get_signal_name(sig, buf, size);
938 if (strcmp(name, "UNKNOWN") == 0) {
939 jio_snprintf(buf, size, "SIG%d", sig);
940 }
941 return buf;
942 }
943
944 #define NUM_IMPORTANT_SIGS 32
945 // Returns one-line short description of a signal set in a user provided buffer.
946 const char* os::Posix::describe_signal_set_short(const sigset_t* set, char* buffer, size_t buf_size) {
947 assert(buf_size == (NUM_IMPORTANT_SIGS + 1), "wrong buffer size");
948 // Note: for shortness, just print out the first 32. That should
949 // cover most of the useful ones, apart from realtime signals.
950 for (int sig = 1; sig <= NUM_IMPORTANT_SIGS; sig++) {
951 const int rc = sigismember(set, sig);
952 if (rc == -1 && errno == EINVAL) {
953 buffer[sig-1] = '?';
954 } else {
955 buffer[sig-1] = rc == 0 ? '0' : '1';
956 }
957 }
958 buffer[NUM_IMPORTANT_SIGS] = 0;
959 return buffer;
960 }
961
962 // Prints one-line description of a signal set.
963 void os::Posix::print_signal_set_short(outputStream* st, const sigset_t* set) {
964 char buf[NUM_IMPORTANT_SIGS + 1];
965 os::Posix::describe_signal_set_short(set, buf, sizeof(buf));
966 st->print("%s", buf);
967 }
968
969 // Writes one-line description of a combination of sigaction.sa_flags into a user
970 // provided buffer. Returns that buffer.
971 const char* os::Posix::describe_sa_flags(int flags, char* buffer, size_t size) {
972 char* p = buffer;
973 size_t remaining = size;
974 bool first = true;
975 int idx = 0;
976
977 assert(buffer, "invalid argument");
978
979 if (size == 0) {
980 return buffer;
981 }
982
983 strncpy(buffer, "none", size);
984
985 const struct {
986 // NB: i is an unsigned int here because SA_RESETHAND is on some
987 // systems 0x80000000, which is implicitly unsigned. Assignining
988 // it to an int field would be an overflow in unsigned-to-signed
989 // conversion.
990 unsigned int i;
991 const char* s;
992 } flaginfo [] = {
993 { SA_NOCLDSTOP, "SA_NOCLDSTOP" },
994 { SA_ONSTACK, "SA_ONSTACK" },
995 { SA_RESETHAND, "SA_RESETHAND" },
996 { SA_RESTART, "SA_RESTART" },
997 { SA_SIGINFO, "SA_SIGINFO" },
998 { SA_NOCLDWAIT, "SA_NOCLDWAIT" },
999 { SA_NODEFER, "SA_NODEFER" },
1000 #ifdef AIX
1001 { SA_ONSTACK, "SA_ONSTACK" },
1002 { SA_OLDSTYLE, "SA_OLDSTYLE" },
1003 #endif
1004 { 0, NULL }
1005 };
1006
1007 for (idx = 0; flaginfo[idx].s && remaining > 1; idx++) {
1008 if (flags & flaginfo[idx].i) {
1009 if (first) {
1010 jio_snprintf(p, remaining, "%s", flaginfo[idx].s);
1011 first = false;
1012 } else {
1013 jio_snprintf(p, remaining, "|%s", flaginfo[idx].s);
1014 }
1015 const size_t len = strlen(p);
1016 p += len;
1017 remaining -= len;
1018 }
1019 }
1020
1021 buffer[size - 1] = '\0';
1022
1023 return buffer;
1024 }
1025
1026 // Prints one-line description of a combination of sigaction.sa_flags.
1027 void os::Posix::print_sa_flags(outputStream* st, int flags) {
1028 char buffer[0x100];
1029 os::Posix::describe_sa_flags(flags, buffer, sizeof(buffer));
1030 st->print("%s", buffer);
1031 }
1032
1033 // Helper function for os::Posix::print_siginfo_...():
1034 // return a textual description for signal code.
1035 struct enum_sigcode_desc_t {
1036 const char* s_name;
1037 const char* s_desc;
1038 };
1039
1040 static bool get_signal_code_description(const siginfo_t* si, enum_sigcode_desc_t* out) {
1041
1042 const struct {
1043 int sig; int code; const char* s_code; const char* s_desc;
1044 } t1 [] = {
1045 { SIGILL, ILL_ILLOPC, "ILL_ILLOPC", "Illegal opcode." },
1046 { SIGILL, ILL_ILLOPN, "ILL_ILLOPN", "Illegal operand." },
1047 { SIGILL, ILL_ILLADR, "ILL_ILLADR", "Illegal addressing mode." },
1048 { SIGILL, ILL_ILLTRP, "ILL_ILLTRP", "Illegal trap." },
1049 { SIGILL, ILL_PRVOPC, "ILL_PRVOPC", "Privileged opcode." },
1050 { SIGILL, ILL_PRVREG, "ILL_PRVREG", "Privileged register." },
1051 { SIGILL, ILL_COPROC, "ILL_COPROC", "Coprocessor error." },
1052 { SIGILL, ILL_BADSTK, "ILL_BADSTK", "Internal stack error." },
1053 #if defined(IA64) && defined(LINUX)
1054 { SIGILL, ILL_BADIADDR, "ILL_BADIADDR", "Unimplemented instruction address" },
1055 { SIGILL, ILL_BREAK, "ILL_BREAK", "Application Break instruction" },
1056 #endif
1057 { SIGFPE, FPE_INTDIV, "FPE_INTDIV", "Integer divide by zero." },
1058 { SIGFPE, FPE_INTOVF, "FPE_INTOVF", "Integer overflow." },
1059 { SIGFPE, FPE_FLTDIV, "FPE_FLTDIV", "Floating-point divide by zero." },
1060 { SIGFPE, FPE_FLTOVF, "FPE_FLTOVF", "Floating-point overflow." },
1061 { SIGFPE, FPE_FLTUND, "FPE_FLTUND", "Floating-point underflow." },
1062 { SIGFPE, FPE_FLTRES, "FPE_FLTRES", "Floating-point inexact result." },
1063 { SIGFPE, FPE_FLTINV, "FPE_FLTINV", "Invalid floating-point operation." },
1064 { SIGFPE, FPE_FLTSUB, "FPE_FLTSUB", "Subscript out of range." },
1065 { SIGSEGV, SEGV_MAPERR, "SEGV_MAPERR", "Address not mapped to object." },
1066 { SIGSEGV, SEGV_ACCERR, "SEGV_ACCERR", "Invalid permissions for mapped object." },
1067 #ifdef AIX
1068 // no explanation found what keyerr would be
1069 { SIGSEGV, SEGV_KEYERR, "SEGV_KEYERR", "key error" },
1070 #endif
1071 #if defined(IA64) && !defined(AIX)
1072 { SIGSEGV, SEGV_PSTKOVF, "SEGV_PSTKOVF", "Paragraph stack overflow" },
1073 #endif
1074 #if defined(__sparc) && defined(SOLARIS)
1075 // define Solaris Sparc M7 ADI SEGV signals
1076 #if !defined(SEGV_ACCADI)
1077 #define SEGV_ACCADI 3
1078 #endif
1079 { SIGSEGV, SEGV_ACCADI, "SEGV_ACCADI", "ADI not enabled for mapped object." },
1080 #if !defined(SEGV_ACCDERR)
1081 #define SEGV_ACCDERR 4
1082 #endif
1083 { SIGSEGV, SEGV_ACCDERR, "SEGV_ACCDERR", "ADI disrupting exception." },
1084 #if !defined(SEGV_ACCPERR)
1085 #define SEGV_ACCPERR 5
1086 #endif
1087 { SIGSEGV, SEGV_ACCPERR, "SEGV_ACCPERR", "ADI precise exception." },
1088 #endif // defined(__sparc) && defined(SOLARIS)
1089 { SIGBUS, BUS_ADRALN, "BUS_ADRALN", "Invalid address alignment." },
1090 { SIGBUS, BUS_ADRERR, "BUS_ADRERR", "Nonexistent physical address." },
1091 { SIGBUS, BUS_OBJERR, "BUS_OBJERR", "Object-specific hardware error." },
1092 { SIGTRAP, TRAP_BRKPT, "TRAP_BRKPT", "Process breakpoint." },
1093 { SIGTRAP, TRAP_TRACE, "TRAP_TRACE", "Process trace trap." },
1094 { SIGCHLD, CLD_EXITED, "CLD_EXITED", "Child has exited." },
1095 { SIGCHLD, CLD_KILLED, "CLD_KILLED", "Child has terminated abnormally and did not create a core file." },
1096 { SIGCHLD, CLD_DUMPED, "CLD_DUMPED", "Child has terminated abnormally and created a core file." },
1097 { SIGCHLD, CLD_TRAPPED, "CLD_TRAPPED", "Traced child has trapped." },
1098 { SIGCHLD, CLD_STOPPED, "CLD_STOPPED", "Child has stopped." },
1099 { SIGCHLD, CLD_CONTINUED,"CLD_CONTINUED","Stopped child has continued." },
1100 #ifdef SIGPOLL
1101 { SIGPOLL, POLL_OUT, "POLL_OUT", "Output buffers available." },
1102 { SIGPOLL, POLL_MSG, "POLL_MSG", "Input message available." },
1103 { SIGPOLL, POLL_ERR, "POLL_ERR", "I/O error." },
1104 { SIGPOLL, POLL_PRI, "POLL_PRI", "High priority input available." },
1105 { SIGPOLL, POLL_HUP, "POLL_HUP", "Device disconnected. [Option End]" },
1106 #endif
1107 { -1, -1, NULL, NULL }
1108 };
1109
1110 // Codes valid in any signal context.
1111 const struct {
1112 int code; const char* s_code; const char* s_desc;
1113 } t2 [] = {
1114 { SI_USER, "SI_USER", "Signal sent by kill()." },
1115 { SI_QUEUE, "SI_QUEUE", "Signal sent by the sigqueue()." },
1116 { SI_TIMER, "SI_TIMER", "Signal generated by expiration of a timer set by timer_settime()." },
1117 { SI_ASYNCIO, "SI_ASYNCIO", "Signal generated by completion of an asynchronous I/O request." },
1118 { SI_MESGQ, "SI_MESGQ", "Signal generated by arrival of a message on an empty message queue." },
1119 // Linux specific
1120 #ifdef SI_TKILL
1121 { SI_TKILL, "SI_TKILL", "Signal sent by tkill (pthread_kill)" },
1122 #endif
1123 #ifdef SI_DETHREAD
1124 { SI_DETHREAD, "SI_DETHREAD", "Signal sent by execve() killing subsidiary threads" },
1125 #endif
1126 #ifdef SI_KERNEL
1127 { SI_KERNEL, "SI_KERNEL", "Signal sent by kernel." },
1128 #endif
1129 #ifdef SI_SIGIO
1130 { SI_SIGIO, "SI_SIGIO", "Signal sent by queued SIGIO" },
1131 #endif
1132
1133 #ifdef AIX
1134 { SI_UNDEFINED, "SI_UNDEFINED","siginfo contains partial information" },
1135 { SI_EMPTY, "SI_EMPTY", "siginfo contains no useful information" },
1136 #endif
1137
1138 #ifdef __sun
1139 { SI_NOINFO, "SI_NOINFO", "No signal information" },
1140 { SI_RCTL, "SI_RCTL", "kernel generated signal via rctl action" },
1141 { SI_LWP, "SI_LWP", "Signal sent via lwp_kill" },
1142 #endif
1143
1144 { -1, NULL, NULL }
1145 };
1146
1147 const char* s_code = NULL;
1148 const char* s_desc = NULL;
1149
1150 for (int i = 0; t1[i].sig != -1; i ++) {
1151 if (t1[i].sig == si->si_signo && t1[i].code == si->si_code) {
1152 s_code = t1[i].s_code;
1153 s_desc = t1[i].s_desc;
1154 break;
1155 }
1156 }
1157
1158 if (s_code == NULL) {
1159 for (int i = 0; t2[i].s_code != NULL; i ++) {
1160 if (t2[i].code == si->si_code) {
1161 s_code = t2[i].s_code;
1162 s_desc = t2[i].s_desc;
1163 }
1164 }
1165 }
1166
1167 if (s_code == NULL) {
1168 out->s_name = "unknown";
1169 out->s_desc = "unknown";
1170 return false;
1171 }
1172
1173 out->s_name = s_code;
1174 out->s_desc = s_desc;
1175
1176 return true;
1177 }
1178
1179 void os::print_siginfo(outputStream* os, const void* si0) {
1180
1181 const siginfo_t* const si = (const siginfo_t*) si0;
1182
1183 char buf[20];
1184 os->print("siginfo:");
1185
1186 if (!si) {
1187 os->print(" <null>");
1188 return;
1189 }
1190
1191 const int sig = si->si_signo;
1192
1193 os->print(" si_signo: %d (%s)", sig, os::Posix::get_signal_name(sig, buf, sizeof(buf)));
1194
1195 enum_sigcode_desc_t ed;
1196 get_signal_code_description(si, &ed);
1197 os->print(", si_code: %d (%s)", si->si_code, ed.s_name);
1198
1199 if (si->si_errno) {
1200 os->print(", si_errno: %d", si->si_errno);
1201 }
1202
1203 // Output additional information depending on the signal code.
1204
1205 // Note: Many implementations lump si_addr, si_pid, si_uid etc. together as unions,
1206 // so it depends on the context which member to use. For synchronous error signals,
1207 // we print si_addr, unless the signal was sent by another process or thread, in
1208 // which case we print out pid or tid of the sender.
1209 if (si->si_code == SI_USER || si->si_code == SI_QUEUE) {
1210 const pid_t pid = si->si_pid;
1211 os->print(", si_pid: %ld", (long) pid);
1212 if (IS_VALID_PID(pid)) {
1213 const pid_t me = getpid();
1214 if (me == pid) {
1215 os->print(" (current process)");
1216 }
1217 } else {
1218 os->print(" (invalid)");
1219 }
1220 os->print(", si_uid: %ld", (long) si->si_uid);
1221 if (sig == SIGCHLD) {
1222 os->print(", si_status: %d", si->si_status);
1223 }
1224 } else if (sig == SIGSEGV || sig == SIGBUS || sig == SIGILL ||
1225 sig == SIGTRAP || sig == SIGFPE) {
1226 os->print(", si_addr: " PTR_FORMAT, p2i(si->si_addr));
1227 #ifdef SIGPOLL
1228 } else if (sig == SIGPOLL) {
1229 os->print(", si_band: %ld", si->si_band);
1230 #endif
1231 }
1232
1233 }
1234
1235 int os::Posix::unblock_thread_signal_mask(const sigset_t *set) {
1236 return pthread_sigmask(SIG_UNBLOCK, set, NULL);
1237 }
1238
1239 address os::Posix::ucontext_get_pc(const ucontext_t* ctx) {
1240 #if defined(AIX)
1241 return Aix::ucontext_get_pc(ctx);
1242 #elif defined(BSD)
1243 return Bsd::ucontext_get_pc(ctx);
1244 #elif defined(LINUX)
1245 return Linux::ucontext_get_pc(ctx);
1246 #elif defined(SOLARIS)
1247 return Solaris::ucontext_get_pc(ctx);
1248 #else
1249 VMError::report_and_die("unimplemented ucontext_get_pc");
1250 #endif
1251 }
1252
1253 void os::Posix::ucontext_set_pc(ucontext_t* ctx, address pc) {
1254 #if defined(AIX)
1255 Aix::ucontext_set_pc(ctx, pc);
1256 #elif defined(BSD)
1257 Bsd::ucontext_set_pc(ctx, pc);
1258 #elif defined(LINUX)
1259 Linux::ucontext_set_pc(ctx, pc);
1260 #elif defined(SOLARIS)
1261 Solaris::ucontext_set_pc(ctx, pc);
1262 #else
1263 VMError::report_and_die("unimplemented ucontext_get_pc");
1264 #endif
1265 }
1266
1267 char* os::Posix::describe_pthread_attr(char* buf, size_t buflen, const pthread_attr_t* attr) {
1268 size_t stack_size = 0;
1269 size_t guard_size = 0;
1270 int detachstate = 0;
1271 pthread_attr_getstacksize(attr, &stack_size);
1272 pthread_attr_getguardsize(attr, &guard_size);
1273 // Work around linux NPTL implementation error, see also os::create_thread() in os_linux.cpp.
1274 LINUX_ONLY(stack_size -= guard_size);
1275 pthread_attr_getdetachstate(attr, &detachstate);
1276 jio_snprintf(buf, buflen, "stacksize: " SIZE_FORMAT "k, guardsize: " SIZE_FORMAT "k, %s",
1277 stack_size / 1024, guard_size / 1024,
1278 (detachstate == PTHREAD_CREATE_DETACHED ? "detached" : "joinable"));
1279 return buf;
1280 }
1281
1282 char* os::Posix::realpath(const char* filename, char* outbuf, size_t outbuflen) {
1283
1284 if (filename == NULL || outbuf == NULL || outbuflen < 1) {
1285 assert(false, "os::Posix::realpath: invalid arguments.");
1286 errno = EINVAL;
1287 return NULL;
1288 }
1289
1290 char* result = NULL;
1291
1292 // This assumes platform realpath() is implemented according to POSIX.1-2008.
1293 // POSIX.1-2008 allows to specify NULL for the output buffer, in which case
1294 // output buffer is dynamically allocated and must be ::free()'d by the caller.
1295 char* p = ::realpath(filename, NULL);
1296 if (p != NULL) {
1297 if (strlen(p) < outbuflen) {
1298 strcpy(outbuf, p);
1299 result = outbuf;
1300 } else {
1301 errno = ENAMETOOLONG;
1302 }
1303 ::free(p); // *not* os::free
1304 } else {
1305 // Fallback for platforms struggling with modern Posix standards (AIX 5.3, 6.1). If realpath
1306 // returns EINVAL, this may indicate that realpath is not POSIX.1-2008 compatible and
1307 // that it complains about the NULL we handed down as user buffer.
1308 // In this case, use the user provided buffer but at least check whether realpath caused
1309 // a memory overwrite.
1310 if (errno == EINVAL) {
1311 outbuf[outbuflen - 1] = '\0';
1312 p = ::realpath(filename, outbuf);
1313 if (p != NULL) {
1314 guarantee(outbuf[outbuflen - 1] == '\0', "realpath buffer overwrite detected.");
1315 result = p;
1316 }
1317 }
1318 }
1319 return result;
1320
1321 }
1322
1323
1324 // Check minimum allowable stack sizes for thread creation and to initialize
1325 // the java system classes, including StackOverflowError - depends on page
1326 // size.
1327 // The space needed for frames during startup is platform dependent. It
1328 // depends on word size, platform calling conventions, C frame layout and
1329 // interpreter/C1/C2 design decisions. Therefore this is given in a
1330 // platform (os/cpu) dependent constant.
1331 // To this, space for guard mechanisms is added, which depends on the
1332 // page size which again depends on the concrete system the VM is running
1333 // on. Space for libc guard pages is not included in this size.
1334 jint os::Posix::set_minimum_stack_sizes() {
1335 size_t os_min_stack_allowed = SOLARIS_ONLY(thr_min_stack()) NOT_SOLARIS(PTHREAD_STACK_MIN);
1336
1337 _java_thread_min_stack_allowed = _java_thread_min_stack_allowed +
1338 JavaThread::stack_guard_zone_size() +
1339 JavaThread::stack_shadow_zone_size();
1340
1341 _java_thread_min_stack_allowed = align_up(_java_thread_min_stack_allowed, vm_page_size());
1342 _java_thread_min_stack_allowed = MAX2(_java_thread_min_stack_allowed, os_min_stack_allowed);
1343
1344 size_t stack_size_in_bytes = ThreadStackSize * K;
1345 if (stack_size_in_bytes != 0 &&
1346 stack_size_in_bytes < _java_thread_min_stack_allowed) {
1347 // The '-Xss' and '-XX:ThreadStackSize=N' options both set
1348 // ThreadStackSize so we go with "Java thread stack size" instead
1349 // of "ThreadStackSize" to be more friendly.
1350 tty->print_cr("\nThe Java thread stack size specified is too small. "
1351 "Specify at least " SIZE_FORMAT "k",
1352 _java_thread_min_stack_allowed / K);
1353 return JNI_ERR;
1354 }
1355
1356 // Make the stack size a multiple of the page size so that
1357 // the yellow/red zones can be guarded.
1358 JavaThread::set_stack_size_at_create(align_up(stack_size_in_bytes, vm_page_size()));
1359
1360 // Reminder: a compiler thread is a Java thread.
1361 _compiler_thread_min_stack_allowed = _compiler_thread_min_stack_allowed +
1362 JavaThread::stack_guard_zone_size() +
1363 JavaThread::stack_shadow_zone_size();
1364
1365 _compiler_thread_min_stack_allowed = align_up(_compiler_thread_min_stack_allowed, vm_page_size());
1366 _compiler_thread_min_stack_allowed = MAX2(_compiler_thread_min_stack_allowed, os_min_stack_allowed);
1367
1368 stack_size_in_bytes = CompilerThreadStackSize * K;
1369 if (stack_size_in_bytes != 0 &&
1370 stack_size_in_bytes < _compiler_thread_min_stack_allowed) {
1371 tty->print_cr("\nThe CompilerThreadStackSize specified is too small. "
1372 "Specify at least " SIZE_FORMAT "k",
1373 _compiler_thread_min_stack_allowed / K);
1374 return JNI_ERR;
1375 }
1376
1377 _vm_internal_thread_min_stack_allowed = align_up(_vm_internal_thread_min_stack_allowed, vm_page_size());
1378 _vm_internal_thread_min_stack_allowed = MAX2(_vm_internal_thread_min_stack_allowed, os_min_stack_allowed);
1379
1380 stack_size_in_bytes = VMThreadStackSize * K;
1381 if (stack_size_in_bytes != 0 &&
1382 stack_size_in_bytes < _vm_internal_thread_min_stack_allowed) {
1383 tty->print_cr("\nThe VMThreadStackSize specified is too small. "
1384 "Specify at least " SIZE_FORMAT "k",
1385 _vm_internal_thread_min_stack_allowed / K);
1386 return JNI_ERR;
1387 }
1388 return JNI_OK;
1389 }
1390
1391 // Called when creating the thread. The minimum stack sizes have already been calculated
1392 size_t os::Posix::get_initial_stack_size(ThreadType thr_type, size_t req_stack_size) {
1393 size_t stack_size;
1394 if (req_stack_size == 0) {
1395 stack_size = default_stack_size(thr_type);
1396 } else {
1397 stack_size = req_stack_size;
1398 }
1399
1400 switch (thr_type) {
1401 case os::java_thread:
1402 // Java threads use ThreadStackSize which default value can be
1403 // changed with the flag -Xss
1404 if (req_stack_size == 0 && JavaThread::stack_size_at_create() > 0) {
1405 // no requested size and we have a more specific default value
1406 stack_size = JavaThread::stack_size_at_create();
1407 }
1408 stack_size = MAX2(stack_size,
1409 _java_thread_min_stack_allowed);
1410 break;
1411 case os::compiler_thread:
1412 if (req_stack_size == 0 && CompilerThreadStackSize > 0) {
1413 // no requested size and we have a more specific default value
1414 stack_size = (size_t)(CompilerThreadStackSize * K);
1415 }
1416 stack_size = MAX2(stack_size,
1417 _compiler_thread_min_stack_allowed);
1418 break;
1419 case os::vm_thread:
1420 case os::pgc_thread:
1421 case os::cgc_thread:
1422 case os::watcher_thread:
1423 default: // presume the unknown thr_type is a VM internal
1424 if (req_stack_size == 0 && VMThreadStackSize > 0) {
1425 // no requested size and we have a more specific default value
1426 stack_size = (size_t)(VMThreadStackSize * K);
1427 }
1428
1429 stack_size = MAX2(stack_size,
1430 _vm_internal_thread_min_stack_allowed);
1431 break;
1432 }
1433
1434 // pthread_attr_setstacksize() may require that the size be rounded up to the OS page size.
1435 // Be careful not to round up to 0. Align down in that case.
1436 if (stack_size <= SIZE_MAX - vm_page_size()) {
1437 stack_size = align_up(stack_size, vm_page_size());
1438 } else {
1439 stack_size = align_down(stack_size, vm_page_size());
1440 }
1441
1442 return stack_size;
1443 }
1444
1445 Thread* os::ThreadCrashProtection::_protected_thread = NULL;
1446 os::ThreadCrashProtection* os::ThreadCrashProtection::_crash_protection = NULL;
1447 volatile intptr_t os::ThreadCrashProtection::_crash_mux = 0;
1448
1449 os::ThreadCrashProtection::ThreadCrashProtection() {
1450 }
1451
1452 /*
1453 * See the caveats for this class in os_posix.hpp
1454 * Protects the callback call so that SIGSEGV / SIGBUS jumps back into this
1455 * method and returns false. If none of the signals are raised, returns true.
1456 * The callback is supposed to provide the method that should be protected.
1457 */
1458 bool os::ThreadCrashProtection::call(os::CrashProtectionCallback& cb) {
1459 sigset_t saved_sig_mask;
1460
1461 Thread::muxAcquire(&_crash_mux, "CrashProtection");
1462
1463 _protected_thread = Thread::current_or_null();
1464 assert(_protected_thread != NULL, "Cannot crash protect a NULL thread");
1465
1466 // we cannot rely on sigsetjmp/siglongjmp to save/restore the signal mask
1467 // since on at least some systems (OS X) siglongjmp will restore the mask
1468 // for the process, not the thread
1469 pthread_sigmask(0, NULL, &saved_sig_mask);
1470 if (sigsetjmp(_jmpbuf, 0) == 0) {
1471 // make sure we can see in the signal handler that we have crash protection
1472 // installed
1473 _crash_protection = this;
1474 cb.call();
1475 // and clear the crash protection
1476 _crash_protection = NULL;
1477 _protected_thread = NULL;
1478 Thread::muxRelease(&_crash_mux);
1479 return true;
1480 }
1481 // this happens when we siglongjmp() back
1482 pthread_sigmask(SIG_SETMASK, &saved_sig_mask, NULL);
1483 _crash_protection = NULL;
1484 _protected_thread = NULL;
1485 Thread::muxRelease(&_crash_mux);
1486 return false;
1487 }
1488
1489 void os::ThreadCrashProtection::restore() {
1490 assert(_crash_protection != NULL, "must have crash protection");
1491 siglongjmp(_jmpbuf, 1);
1492 }
1493
1494 void os::ThreadCrashProtection::check_crash_protection(int sig,
1495 Thread* thread) {
1496
1497 if (thread != NULL &&
1498 thread == _protected_thread &&
1499 _crash_protection != NULL) {
1500
1501 if (sig == SIGSEGV || sig == SIGBUS) {
1502 _crash_protection->restore();
1503 }
1504 }
1505 }
1506
1507
1508 // Shared pthread_mutex/cond based PlatformEvent implementation.
1509 // Not currently usable by Solaris.
1510
1511 #ifndef SOLARIS
1512
1513 // Shared condattr object for use with relative timed-waits. Will be associated
1514 // with CLOCK_MONOTONIC if available to avoid issues with time-of-day changes,
1515 // but otherwise whatever default is used by the platform - generally the
1516 // time-of-day clock.
1517 static pthread_condattr_t _condAttr[1];
1518
1519 // Shared mutexattr to explicitly set the type to PTHREAD_MUTEX_NORMAL as not
1520 // all systems (e.g. FreeBSD) map the default to "normal".
1521 static pthread_mutexattr_t _mutexAttr[1];
1522
1523 // common basic initialization that is always supported
1524 static void pthread_init_common(void) {
1525 int status;
1526 if ((status = pthread_condattr_init(_condAttr)) != 0) {
1527 fatal("pthread_condattr_init: %s", os::strerror(status));
1528 }
1529 if ((status = pthread_mutexattr_init(_mutexAttr)) != 0) {
1530 fatal("pthread_mutexattr_init: %s", os::strerror(status));
1531 }
1532 if ((status = pthread_mutexattr_settype(_mutexAttr, PTHREAD_MUTEX_NORMAL)) != 0) {
1533 fatal("pthread_mutexattr_settype: %s", os::strerror(status));
1534 }
1535 }
1536
1537 // Not all POSIX types and API's are available on all notionally "posix"
1538 // platforms. If we have build-time support then we will check for actual
1539 // runtime support via dlopen/dlsym lookup. This allows for running on an
1540 // older OS version compared to the build platform. But if there is no
1541 // build time support then there cannot be any runtime support as we do not
1542 // know what the runtime types would be (for example clockid_t might be an
1543 // int or int64_t).
1544 //
1545 #ifdef SUPPORTS_CLOCK_MONOTONIC
1546
1547 // This means we have clockid_t, clock_gettime et al and CLOCK_MONOTONIC
1548
1549 static int (*_clock_gettime)(clockid_t, struct timespec *);
1550 static int (*_pthread_condattr_setclock)(pthread_condattr_t *, clockid_t);
1551
1552 static bool _use_clock_monotonic_condattr;
1553
1554 // Determine what POSIX API's are present and do appropriate
1555 // configuration.
1556 void os::Posix::init(void) {
1557
1558 // NOTE: no logging available when this is called. Put logging
1559 // statements in init_2().
1560
1561 // Copied from os::Linux::clock_init(). The duplication is temporary.
1562
1563 // 1. Check for CLOCK_MONOTONIC support.
1564
1565 void* handle = NULL;
1566
1567 // For linux we need librt, for other OS we can find
1568 // this function in regular libc.
1569 #ifdef NEEDS_LIBRT
1570 // We do dlopen's in this particular order due to bug in linux
1571 // dynamic loader (see 6348968) leading to crash on exit.
1572 handle = dlopen("librt.so.1", RTLD_LAZY);
1573 if (handle == NULL) {
1574 handle = dlopen("librt.so", RTLD_LAZY);
1575 }
1576 #endif
1577
1578 if (handle == NULL) {
1579 handle = RTLD_DEFAULT;
1580 }
1581
1582 _clock_gettime = NULL;
1583
1584 int (*clock_getres_func)(clockid_t, struct timespec*) =
1585 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1586 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1587 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1588 if (clock_getres_func != NULL && clock_gettime_func != NULL) {
1589 // We assume that if both clock_gettime and clock_getres support
1590 // CLOCK_MONOTONIC then the OS provides true high-res monotonic clock.
1591 struct timespec res;
1592 struct timespec tp;
1593 if (clock_getres_func(CLOCK_MONOTONIC, &res) == 0 &&
1594 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1595 // Yes, monotonic clock is supported.
1596 _clock_gettime = clock_gettime_func;
1597 } else {
1598 #ifdef NEEDS_LIBRT
1599 // Close librt if there is no monotonic clock.
1600 if (handle != RTLD_DEFAULT) {
1601 dlclose(handle);
1602 }
1603 #endif
1604 }
1605 }
1606
1607 // 2. Check for pthread_condattr_setclock support.
1608
1609 _pthread_condattr_setclock = NULL;
1610
1611 // libpthread is already loaded.
1612 int (*condattr_setclock_func)(pthread_condattr_t*, clockid_t) =
1613 (int (*)(pthread_condattr_t*, clockid_t))dlsym(RTLD_DEFAULT,
1614 "pthread_condattr_setclock");
1615 if (condattr_setclock_func != NULL) {
1616 _pthread_condattr_setclock = condattr_setclock_func;
1617 }
1618
1619 // Now do general initialization.
1620
1621 pthread_init_common();
1622
1623 int status;
1624 if (_pthread_condattr_setclock != NULL && _clock_gettime != NULL) {
1625 if ((status = _pthread_condattr_setclock(_condAttr, CLOCK_MONOTONIC)) != 0) {
1626 if (status == EINVAL) {
1627 _use_clock_monotonic_condattr = false;
1628 warning("Unable to use monotonic clock with relative timed-waits" \
1629 " - changes to the time-of-day clock may have adverse affects");
1630 } else {
1631 fatal("pthread_condattr_setclock: %s", os::strerror(status));
1632 }
1633 } else {
1634 _use_clock_monotonic_condattr = true;
1635 }
1636 } else {
1637 _use_clock_monotonic_condattr = false;
1638 }
1639 }
1640
1641 void os::Posix::init_2(void) {
1642 log_info(os)("Use of CLOCK_MONOTONIC is%s supported",
1643 (_clock_gettime != NULL ? "" : " not"));
1644 log_info(os)("Use of pthread_condattr_setclock is%s supported",
1645 (_pthread_condattr_setclock != NULL ? "" : " not"));
1646 log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with %s",
1647 _use_clock_monotonic_condattr ? "CLOCK_MONOTONIC" : "the default clock");
1648 }
1649
1650 #else // !SUPPORTS_CLOCK_MONOTONIC
1651
1652 void os::Posix::init(void) {
1653 pthread_init_common();
1654 }
1655
1656 void os::Posix::init_2(void) {
1657 log_info(os)("Use of CLOCK_MONOTONIC is not supported");
1658 log_info(os)("Use of pthread_condattr_setclock is not supported");
1659 log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with the default clock");
1660 }
1661
1662 #endif // SUPPORTS_CLOCK_MONOTONIC
1663
1664 os::PlatformEvent::PlatformEvent() {
1665 int status = pthread_cond_init(_cond, _condAttr);
1666 assert_status(status == 0, status, "cond_init");
1667 status = pthread_mutex_init(_mutex, _mutexAttr);
1668 assert_status(status == 0, status, "mutex_init");
1669 _event = 0;
1670 _nParked = 0;
1671 }
1672
1673 // Utility to convert the given timeout to an absolute timespec
1674 // (based on the appropriate clock) to use with pthread_cond_timewait.
1675 // The clock queried here must be the clock used to manage the
1676 // timeout of the condition variable.
1677 //
1678 // The passed in timeout value is either a relative time in nanoseconds
1679 // or an absolute time in milliseconds. A relative timeout will be
1680 // associated with CLOCK_MONOTONIC if available; otherwise, or if absolute,
1681 // the default time-of-day clock will be used.
1682
1683 // Given time is a 64-bit value and the time_t used in the timespec is
1684 // sometimes a signed-32-bit value we have to watch for overflow if times
1685 // way in the future are given. Further on Solaris versions
1686 // prior to 10 there is a restriction (see cond_timedwait) that the specified
1687 // number of seconds, in abstime, is less than current_time + 100000000.
1688 // As it will be over 20 years before "now + 100000000" will overflow we can
1689 // ignore overflow and just impose a hard-limit on seconds using the value
1690 // of "now + 100000000". This places a limit on the timeout of about 3.17
1691 // years from "now".
1692 //
1693 #define MAX_SECS 100000000
1694
1695 // Calculate a new absolute time that is "timeout" nanoseconds from "now".
1696 // "unit" indicates the unit of "now_part_sec" (may be nanos or micros depending
1697 // on which clock is being used).
1698 static void calc_rel_time(timespec* abstime, jlong timeout, jlong now_sec,
1699 jlong now_part_sec, jlong unit) {
1700 time_t max_secs = now_sec + MAX_SECS;
1701
1702 jlong seconds = timeout / NANOUNITS;
1703 timeout %= NANOUNITS; // remaining nanos
1704
1705 if (seconds >= MAX_SECS) {
1706 // More seconds than we can add, so pin to max_secs.
1707 abstime->tv_sec = max_secs;
1708 abstime->tv_nsec = 0;
1709 } else {
1710 abstime->tv_sec = now_sec + seconds;
1711 long nanos = (now_part_sec * (NANOUNITS / unit)) + timeout;
1712 if (nanos >= NANOUNITS) { // overflow
1713 abstime->tv_sec += 1;
1714 nanos -= NANOUNITS;
1715 }
1716 abstime->tv_nsec = nanos;
1717 }
1718 }
1719
1720 // Unpack the given deadline in milliseconds since the epoch, into the given timespec.
1721 // The current time in seconds is also passed in to enforce an upper bound as discussed above.
1722 static void unpack_abs_time(timespec* abstime, jlong deadline, jlong now_sec) {
1723 time_t max_secs = now_sec + MAX_SECS;
1724
1725 jlong seconds = deadline / MILLIUNITS;
1726 jlong millis = deadline % MILLIUNITS;
1727
1728 if (seconds >= max_secs) {
1729 // Absolute seconds exceeds allowed max, so pin to max_secs.
1730 abstime->tv_sec = max_secs;
1731 abstime->tv_nsec = 0;
1732 } else {
1733 abstime->tv_sec = seconds;
1734 abstime->tv_nsec = millis * (NANOUNITS / MILLIUNITS);
1735 }
1736 }
1737
1738 static void to_abstime(timespec* abstime, jlong timeout, bool isAbsolute) {
1739 DEBUG_ONLY(int max_secs = MAX_SECS;)
1740
1741 if (timeout < 0) {
1742 timeout = 0;
1743 }
1744
1745 #ifdef SUPPORTS_CLOCK_MONOTONIC
1746
1747 if (_use_clock_monotonic_condattr && !isAbsolute) {
1748 struct timespec now;
1749 int status = _clock_gettime(CLOCK_MONOTONIC, &now);
1750 assert_status(status == 0, status, "clock_gettime");
1751 calc_rel_time(abstime, timeout, now.tv_sec, now.tv_nsec, NANOUNITS);
1752 DEBUG_ONLY(max_secs += now.tv_sec;)
1753 } else {
1754
1755 #else
1756
1757 { // Match the block scope.
1758
1759 #endif // SUPPORTS_CLOCK_MONOTONIC
1760
1761 // Time-of-day clock is all we can reliably use.
1762 struct timeval now;
1763 int status = gettimeofday(&now, NULL);
1764 assert_status(status == 0, errno, "gettimeofday");
1765 if (isAbsolute) {
1766 unpack_abs_time(abstime, timeout, now.tv_sec);
1767 } else {
1768 calc_rel_time(abstime, timeout, now.tv_sec, now.tv_usec, MICROUNITS);
1769 }
1770 DEBUG_ONLY(max_secs += now.tv_sec;)
1771 }
1772
1773 assert(abstime->tv_sec >= 0, "tv_sec < 0");
1774 assert(abstime->tv_sec <= max_secs, "tv_sec > max_secs");
1775 assert(abstime->tv_nsec >= 0, "tv_nsec < 0");
1776 assert(abstime->tv_nsec < NANOUNITS, "tv_nsec >= NANOUNITS");
1777 }
1778
1779 // PlatformEvent
1780 //
1781 // Assumption:
1782 // Only one parker can exist on an event, which is why we allocate
1783 // them per-thread. Multiple unparkers can coexist.
1784 //
1785 // _event serves as a restricted-range semaphore.
1786 // -1 : thread is blocked, i.e. there is a waiter
1787 // 0 : neutral: thread is running or ready,
1788 // could have been signaled after a wait started
1789 // 1 : signaled - thread is running or ready
1790 //
1791 // Having three states allows for some detection of bad usage - see
1792 // comments on unpark().
1793
1794 void os::PlatformEvent::park() { // AKA "down()"
1795 // Transitions for _event:
1796 // -1 => -1 : illegal
1797 // 1 => 0 : pass - return immediately
1798 // 0 => -1 : block; then set _event to 0 before returning
1799
1800 // Invariant: Only the thread associated with the PlatformEvent
1801 // may call park().
1802 assert(_nParked == 0, "invariant");
1803
1804 int v;
1805
1806 // atomically decrement _event
1807 for (;;) {
1808 v = _event;
1809 if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1810 }
1811 guarantee(v >= 0, "invariant");
1812
1813 if (v == 0) { // Do this the hard way by blocking ...
1814 int status = pthread_mutex_lock(_mutex);
1815 assert_status(status == 0, status, "mutex_lock");
1816 guarantee(_nParked == 0, "invariant");
1817 ++_nParked;
1818 while (_event < 0) {
1819 // OS-level "spurious wakeups" are ignored
1820 status = pthread_cond_wait(_cond, _mutex);
1821 assert_status(status == 0, status, "cond_wait");
1822 }
1823 --_nParked;
1824
1825 _event = 0;
1826 status = pthread_mutex_unlock(_mutex);
1827 assert_status(status == 0, status, "mutex_unlock");
1828 // Paranoia to ensure our locked and lock-free paths interact
1829 // correctly with each other.
1830 OrderAccess::fence();
1831 }
1832 guarantee(_event >= 0, "invariant");
1833 }
1834
1835 int os::PlatformEvent::park(jlong millis) {
1836 // Transitions for _event:
1837 // -1 => -1 : illegal
1838 // 1 => 0 : pass - return immediately
1839 // 0 => -1 : block; then set _event to 0 before returning
1840
1841 // Invariant: Only the thread associated with the Event/PlatformEvent
1842 // may call park().
1843 assert(_nParked == 0, "invariant");
1844
1845 int v;
1846 // atomically decrement _event
1847 for (;;) {
1848 v = _event;
1849 if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1850 }
1851 guarantee(v >= 0, "invariant");
1852
1853 if (v == 0) { // Do this the hard way by blocking ...
1854 struct timespec abst;
1855 // We have to watch for overflow when converting millis to nanos,
1856 // but if millis is that large then we will end up limiting to
1857 // MAX_SECS anyway, so just do that here.
1858 if (millis / MILLIUNITS > MAX_SECS) {
1859 millis = jlong(MAX_SECS) * MILLIUNITS;
1860 }
1861 to_abstime(&abst, millis * (NANOUNITS / MILLIUNITS), false);
1862
1863 int ret = OS_TIMEOUT;
1864 int status = pthread_mutex_lock(_mutex);
1865 assert_status(status == 0, status, "mutex_lock");
1866 guarantee(_nParked == 0, "invariant");
1867 ++_nParked;
1868
1869 while (_event < 0) {
1870 status = pthread_cond_timedwait(_cond, _mutex, &abst);
1871 assert_status(status == 0 || status == ETIMEDOUT,
1872 status, "cond_timedwait");
1873 // OS-level "spurious wakeups" are ignored unless the archaic
1874 // FilterSpuriousWakeups is set false. That flag should be obsoleted.
1875 if (!FilterSpuriousWakeups) break;
1876 if (status == ETIMEDOUT) break;
1877 }
1878 --_nParked;
1879
1880 if (_event >= 0) {
1881 ret = OS_OK;
1882 }
1883
1884 _event = 0;
1885 status = pthread_mutex_unlock(_mutex);
1886 assert_status(status == 0, status, "mutex_unlock");
1887 // Paranoia to ensure our locked and lock-free paths interact
1888 // correctly with each other.
1889 OrderAccess::fence();
1890 return ret;
1891 }
1892 return OS_OK;
1893 }
1894
1895 void os::PlatformEvent::unpark() {
1896 // Transitions for _event:
1897 // 0 => 1 : just return
1898 // 1 => 1 : just return
1899 // -1 => either 0 or 1; must signal target thread
1900 // That is, we can safely transition _event from -1 to either
1901 // 0 or 1.
1902 // See also: "Semaphores in Plan 9" by Mullender & Cox
1903 //
1904 // Note: Forcing a transition from "-1" to "1" on an unpark() means
1905 // that it will take two back-to-back park() calls for the owning
1906 // thread to block. This has the benefit of forcing a spurious return
1907 // from the first park() call after an unpark() call which will help
1908 // shake out uses of park() and unpark() without checking state conditions
1909 // properly. This spurious return doesn't manifest itself in any user code
1910 // but only in the correctly written condition checking loops of ObjectMonitor,
1911 // Mutex/Monitor, Thread::muxAcquire and os::sleep
1912
1913 if (Atomic::xchg(1, &_event) >= 0) return;
1914
1915 int status = pthread_mutex_lock(_mutex);
1916 assert_status(status == 0, status, "mutex_lock");
1917 int anyWaiters = _nParked;
1918 assert(anyWaiters == 0 || anyWaiters == 1, "invariant");
1919 status = pthread_mutex_unlock(_mutex);
1920 assert_status(status == 0, status, "mutex_unlock");
1921
1922 // Note that we signal() *after* dropping the lock for "immortal" Events.
1923 // This is safe and avoids a common class of futile wakeups. In rare
1924 // circumstances this can cause a thread to return prematurely from
1925 // cond_{timed}wait() but the spurious wakeup is benign and the victim
1926 // will simply re-test the condition and re-park itself.
1927 // This provides particular benefit if the underlying platform does not
1928 // provide wait morphing.
1929
1930 if (anyWaiters != 0) {
1931 status = pthread_cond_signal(_cond);
1932 assert_status(status == 0, status, "cond_signal");
1933 }
1934 }
1935
1936 // JSR166 support
1937
1938 os::PlatformParker::PlatformParker() {
1939 int status;
1940 status = pthread_cond_init(&_cond[REL_INDEX], _condAttr);
1941 assert_status(status == 0, status, "cond_init rel");
1942 status = pthread_cond_init(&_cond[ABS_INDEX], NULL);
1943 assert_status(status == 0, status, "cond_init abs");
1944 status = pthread_mutex_init(_mutex, _mutexAttr);
1945 assert_status(status == 0, status, "mutex_init");
1946 _cur_index = -1; // mark as unused
1947 }
1948
1949 // Parker::park decrements count if > 0, else does a condvar wait. Unpark
1950 // sets count to 1 and signals condvar. Only one thread ever waits
1951 // on the condvar. Contention seen when trying to park implies that someone
1952 // is unparking you, so don't wait. And spurious returns are fine, so there
1953 // is no need to track notifications.
1954
1955 void Parker::park(bool isAbsolute, jlong time) {
1956
1957 // Optional fast-path check:
1958 // Return immediately if a permit is available.
1959 // We depend on Atomic::xchg() having full barrier semantics
1960 // since we are doing a lock-free update to _counter.
1961 if (Atomic::xchg(0, &_counter) > 0) return;
1962
1963 Thread* thread = Thread::current();
1964 assert(thread->is_Java_thread(), "Must be JavaThread");
1965 JavaThread *jt = (JavaThread *)thread;
1966
1967 // Optional optimization -- avoid state transitions if there's
1968 // an interrupt pending.
1969 if (Thread::is_interrupted(thread, false)) {
1970 return;
1971 }
1972
1973 // Next, demultiplex/decode time arguments
1974 struct timespec absTime;
1975 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
1976 return;
1977 }
1978 if (time > 0) {
1979 to_abstime(&absTime, time, isAbsolute);
1980 }
1981
1982 // Enter safepoint region
1983 // Beware of deadlocks such as 6317397.
1984 // The per-thread Parker:: mutex is a classic leaf-lock.
1985 // In particular a thread must never block on the Threads_lock while
1986 // holding the Parker:: mutex. If safepoints are pending both the
1987 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
1988 ThreadBlockInVM tbivm(jt);
1989
1990 // Don't wait if cannot get lock since interference arises from
1991 // unparking. Also re-check interrupt before trying wait.
1992 if (Thread::is_interrupted(thread, false) ||
1993 pthread_mutex_trylock(_mutex) != 0) {
1994 return;
1995 }
1996
1997 int status;
1998 if (_counter > 0) { // no wait needed
1999 _counter = 0;
2000 status = pthread_mutex_unlock(_mutex);
2001 assert_status(status == 0, status, "invariant");
2002 // Paranoia to ensure our locked and lock-free paths interact
2003 // correctly with each other and Java-level accesses.
2004 OrderAccess::fence();
2005 return;
2006 }
2007
2008 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2009 jt->set_suspend_equivalent();
2010 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2011
2012 assert(_cur_index == -1, "invariant");
2013 if (time == 0) {
2014 _cur_index = REL_INDEX; // arbitrary choice when not timed
2015 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
2016 assert_status(status == 0, status, "cond_timedwait");
2017 }
2018 else {
2019 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
2020 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
2021 assert_status(status == 0 || status == ETIMEDOUT,
2022 status, "cond_timedwait");
2023 }
2024 _cur_index = -1;
2025
2026 _counter = 0;
2027 status = pthread_mutex_unlock(_mutex);
2028 assert_status(status == 0, status, "invariant");
2029 // Paranoia to ensure our locked and lock-free paths interact
2030 // correctly with each other and Java-level accesses.
2031 OrderAccess::fence();
2032
2033 // If externally suspended while waiting, re-suspend
2034 if (jt->handle_special_suspend_equivalent_condition()) {
2035 jt->java_suspend_self();
2036 }
2037 }
2038
2039 void Parker::unpark() {
2040 int status = pthread_mutex_lock(_mutex);
2041 assert_status(status == 0, status, "invariant");
2042 const int s = _counter;
2043 _counter = 1;
2044 // must capture correct index before unlocking
2045 int index = _cur_index;
2046 status = pthread_mutex_unlock(_mutex);
2047 assert_status(status == 0, status, "invariant");
2048
2049 // Note that we signal() *after* dropping the lock for "immortal" Events.
2050 // This is safe and avoids a common class of futile wakeups. In rare
2051 // circumstances this can cause a thread to return prematurely from
2052 // cond_{timed}wait() but the spurious wakeup is benign and the victim
2053 // will simply re-test the condition and re-park itself.
2054 // This provides particular benefit if the underlying platform does not
2055 // provide wait morphing.
2056
2057 if (s < 1 && index != -1) {
2058 // thread is definitely parked
2059 status = pthread_cond_signal(&_cond[index]);
2060 assert_status(status == 0, status, "invariant");
2061 }
2062 }
2063
2064
2065 #endif // !SOLARIS