1 /*
2 * Copyright (c) 1999, 2019, 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 "os_posix.inline.hpp"
29 #include "utilities/globalDefinitions.hpp"
30 #include "runtime/frame.inline.hpp"
31 #include "runtime/interfaceSupport.inline.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 <dirent.h>
39 #include <dlfcn.h>
40 #include <grp.h>
41 #include <pwd.h>
42 #include <pthread.h>
43 #include <signal.h>
44 #include <sys/mman.h>
45 #include <sys/resource.h>
46 #include <sys/utsname.h>
47 #include <time.h>
48 #include <unistd.h>
49
50 // Todo: provide a os::get_max_process_id() or similar. Number of processes
51 // may have been configured, can be read more accurately from proc fs etc.
52 #ifndef MAX_PID
53 #define MAX_PID INT_MAX
54 #endif
55 #define IS_VALID_PID(p) (p > 0 && p < MAX_PID)
56
57 #define ROOT_UID 0
58
59 #ifndef MAP_ANONYMOUS
60 #define MAP_ANONYMOUS MAP_ANON
61 #endif
62
63 #define check_with_errno(check_type, cond, msg) \
64 do { \
65 int err = errno; \
66 check_type(cond, "%s; error='%s' (errno=%s)", msg, os::strerror(err), \
67 os::errno_name(err)); \
68 } while (false)
69
70 #define assert_with_errno(cond, msg) check_with_errno(assert, cond, msg)
71 #define guarantee_with_errno(cond, msg) check_with_errno(guarantee, cond, msg)
72
73 // Check core dump limit and report possible place where core can be found
74 void os::check_dump_limit(char* buffer, size_t bufferSize) {
75 if (!FLAG_IS_DEFAULT(CreateCoredumpOnCrash) && !CreateCoredumpOnCrash) {
76 jio_snprintf(buffer, bufferSize, "CreateCoredumpOnCrash is disabled from command line");
77 VMError::record_coredump_status(buffer, false);
78 return;
79 }
80
81 int n;
82 struct rlimit rlim;
83 bool success;
84
85 char core_path[PATH_MAX];
86 n = get_core_path(core_path, PATH_MAX);
87
88 if (n <= 0) {
89 jio_snprintf(buffer, bufferSize, "core.%d (may not exist)", current_process_id());
90 success = true;
91 #ifdef LINUX
92 } else if (core_path[0] == '"') { // redirect to user process
93 jio_snprintf(buffer, bufferSize, "Core dumps may be processed with %s", core_path);
94 success = true;
95 #endif
96 } else if (getrlimit(RLIMIT_CORE, &rlim) != 0) {
97 jio_snprintf(buffer, bufferSize, "%s (may not exist)", core_path);
98 success = true;
99 } else {
100 switch(rlim.rlim_cur) {
101 case RLIM_INFINITY:
102 jio_snprintf(buffer, bufferSize, "%s", core_path);
103 success = true;
104 break;
105 case 0:
106 jio_snprintf(buffer, bufferSize, "Core dumps have been disabled. To enable core dumping, try \"ulimit -c unlimited\" before starting Java again");
107 success = false;
108 break;
109 default:
110 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);
111 success = true;
112 break;
113 }
114 }
115
116 VMError::record_coredump_status(buffer, success);
117 }
118
119 int os::get_native_stack(address* stack, int frames, int toSkip) {
120 int frame_idx = 0;
121 int num_of_frames; // number of frames captured
122 frame fr = os::current_frame();
123 while (fr.pc() && frame_idx < frames) {
124 if (toSkip > 0) {
125 toSkip --;
126 } else {
127 stack[frame_idx ++] = fr.pc();
128 }
129 if (fr.fp() == NULL || fr.cb() != NULL ||
130 fr.sender_pc() == NULL || os::is_first_C_frame(&fr)) break;
131
132 if (fr.sender_pc() && !os::is_first_C_frame(&fr)) {
133 fr = os::get_sender_for_C_frame(&fr);
134 } else {
135 break;
136 }
137 }
138 num_of_frames = frame_idx;
139 for (; frame_idx < frames; frame_idx ++) {
140 stack[frame_idx] = NULL;
141 }
142
143 return num_of_frames;
144 }
145
146
147 bool os::unsetenv(const char* name) {
148 assert(name != NULL, "Null pointer");
149 return (::unsetenv(name) == 0);
150 }
151
152 int os::get_last_error() {
153 return errno;
154 }
155
156 size_t os::lasterror(char *buf, size_t len) {
157 if (errno == 0) return 0;
158
159 const char *s = os::strerror(errno);
160 size_t n = ::strlen(s);
161 if (n >= len) {
162 n = len - 1;
163 }
164 ::strncpy(buf, s, n);
165 buf[n] = '\0';
166 return n;
167 }
168
169 bool os::is_debugger_attached() {
170 // not implemented
171 return false;
172 }
173
174 void os::wait_for_keypress_at_exit(void) {
175 // don't do anything on posix platforms
176 return;
177 }
178
179 int os::create_file_for_heap(const char* dir) {
180
181 const char name_template[] = "/jvmheap.XXXXXX";
182
183 size_t fullname_len = strlen(dir) + strlen(name_template);
184 char *fullname = (char*)os::malloc(fullname_len + 1, mtInternal);
185 if (fullname == NULL) {
186 vm_exit_during_initialization(err_msg("Malloc failed during creation of backing file for heap (%s)", os::strerror(errno)));
187 return -1;
188 }
189 int n = snprintf(fullname, fullname_len + 1, "%s%s", dir, name_template);
190 assert((size_t)n == fullname_len, "Unexpected number of characters in string");
191
192 os::native_path(fullname);
193
194 // set the file creation mask.
195 mode_t file_mode = S_IRUSR | S_IWUSR;
196
197 // create a new file.
198 int fd = mkstemp(fullname);
199
200 if (fd < 0) {
201 warning("Could not create file for heap with template %s", fullname);
202 os::free(fullname);
203 return -1;
204 }
205
206 // delete the name from the filesystem. When 'fd' is closed, the file (and space) will be deleted.
207 int ret = unlink(fullname);
208 assert_with_errno(ret == 0, "unlink returned error");
209
210 os::free(fullname);
211 return fd;
212 }
213
214 static char* reserve_mmapped_memory(size_t bytes, char* requested_addr) {
215 char * addr;
216 int flags = MAP_PRIVATE NOT_AIX( | MAP_NORESERVE ) | MAP_ANONYMOUS;
217 if (requested_addr != NULL) {
218 assert((uintptr_t)requested_addr % os::vm_page_size() == 0, "Requested address should be aligned to OS page size");
219 flags |= MAP_FIXED;
220 }
221
222 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
223 // touch an uncommitted page. Otherwise, the read/write might
224 // succeed if we have enough swap space to back the physical page.
225 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
226 flags, -1, 0);
227
228 if (addr != MAP_FAILED) {
229 MemTracker::record_virtual_memory_reserve((address)addr, bytes, CALLER_PC);
230 return addr;
231 }
232 return NULL;
233 }
234
235 static int util_posix_fallocate(int fd, off_t offset, off_t len) {
236 #ifdef __APPLE__
237 fstore_t store = { F_ALLOCATECONTIG, F_PEOFPOSMODE, 0, len };
238 // First we try to get a continuous chunk of disk space
239 int ret = fcntl(fd, F_PREALLOCATE, &store);
240 if (ret == -1) {
241 // Maybe we are too fragmented, try to allocate non-continuous range
242 store.fst_flags = F_ALLOCATEALL;
243 ret = fcntl(fd, F_PREALLOCATE, &store);
244 }
245 if(ret != -1) {
246 return ftruncate(fd, len);
247 }
248 return -1;
249 #else
250 return posix_fallocate(fd, offset, len);
251 #endif
252 }
253
254 // Map the given address range to the provided file descriptor.
255 char* os::map_memory_to_file(char* base, size_t size, int fd) {
256 assert(fd != -1, "File descriptor is not valid");
257
258 // allocate space for the file
259 int ret = util_posix_fallocate(fd, 0, (off_t)size);
260 if (ret != 0) {
261 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory. error(%d)", ret));
262 return NULL;
263 }
264
265 int prot = PROT_READ | PROT_WRITE;
266 int flags = MAP_SHARED;
267 if (base != NULL) {
268 flags |= MAP_FIXED;
269 }
270 char* addr = (char*)mmap(base, size, prot, flags, fd, 0);
271
272 if (addr == MAP_FAILED) {
273 warning("Failed mmap to file. (%s)", os::strerror(errno));
274 return NULL;
275 }
276 if (base != NULL && addr != base) {
277 if (!os::release_memory(addr, size)) {
278 warning("Could not release memory on unsuccessful file mapping");
279 }
280 return NULL;
281 }
282 return addr;
283 }
284
285 char* os::replace_existing_mapping_with_file_mapping(char* base, size_t size, int fd) {
286 assert(fd != -1, "File descriptor is not valid");
287 assert(base != NULL, "Base cannot be NULL");
288
289 return map_memory_to_file(base, size, fd);
290 }
291
292 // Multiple threads can race in this code, and can remap over each other with MAP_FIXED,
293 // so on posix, unmap the section at the start and at the end of the chunk that we mapped
294 // rather than unmapping and remapping the whole chunk to get requested alignment.
295 char* os::reserve_memory_aligned(size_t size, size_t alignment, int file_desc) {
296 assert((alignment & (os::vm_allocation_granularity() - 1)) == 0,
297 "Alignment must be a multiple of allocation granularity (page size)");
298 assert((size & (alignment -1)) == 0, "size must be 'alignment' aligned");
299
300 size_t extra_size = size + alignment;
301 assert(extra_size >= size, "overflow, size is too large to allow alignment");
302
303 char* extra_base;
304 if (file_desc != -1) {
305 // For file mapping, we do not call os:reserve_memory(extra_size, NULL, alignment, file_desc) because
306 // we need to deal with shrinking of the file space later when we release extra memory after alignment.
307 // We also cannot called os:reserve_memory() with file_desc set to -1 because on aix we might get SHM memory.
308 // So here to call a helper function while reserve memory for us. After we have a aligned base,
309 // we will replace anonymous mapping with file mapping.
310 extra_base = reserve_mmapped_memory(extra_size, NULL);
311 if (extra_base != NULL) {
312 MemTracker::record_virtual_memory_reserve((address)extra_base, extra_size, CALLER_PC);
313 }
314 } else {
315 extra_base = os::reserve_memory(extra_size, NULL, alignment);
316 }
317
318 if (extra_base == NULL) {
319 return NULL;
320 }
321
322 // Do manual alignment
323 char* aligned_base = align_up(extra_base, alignment);
324
325 // [ | | ]
326 // ^ extra_base
327 // ^ extra_base + begin_offset == aligned_base
328 // extra_base + begin_offset + size ^
329 // extra_base + extra_size ^
330 // |<>| == begin_offset
331 // end_offset == |<>|
332 size_t begin_offset = aligned_base - extra_base;
333 size_t end_offset = (extra_base + extra_size) - (aligned_base + size);
334
335 if (begin_offset > 0) {
336 os::release_memory(extra_base, begin_offset);
337 }
338
339 if (end_offset > 0) {
340 os::release_memory(extra_base + begin_offset + size, end_offset);
341 }
342
343 if (file_desc != -1) {
344 // After we have an aligned address, we can replace anonymous mapping with file mapping
345 if (replace_existing_mapping_with_file_mapping(aligned_base, size, file_desc) == NULL) {
346 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
347 }
348 MemTracker::record_virtual_memory_commit((address)aligned_base, size, CALLER_PC);
349 }
350 return aligned_base;
351 }
352
353 int os::vsnprintf(char* buf, size_t len, const char* fmt, va_list args) {
354 // All supported POSIX platforms provide C99 semantics.
355 int result = ::vsnprintf(buf, len, fmt, args);
356 // If an encoding error occurred (result < 0) then it's not clear
357 // whether the buffer is NUL terminated, so ensure it is.
358 if ((result < 0) && (len > 0)) {
359 buf[len - 1] = '\0';
360 }
361 return result;
362 }
363
364 int os::get_fileno(FILE* fp) {
365 return NOT_AIX(::)fileno(fp);
366 }
367
368 struct tm* os::gmtime_pd(const time_t* clock, struct tm* res) {
369 return gmtime_r(clock, res);
370 }
371
372 void os::Posix::print_load_average(outputStream* st) {
373 st->print("load average:");
374 double loadavg[3];
375 os::loadavg(loadavg, 3);
376 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
377 st->cr();
378 }
379
380 void os::Posix::print_rlimit_info(outputStream* st) {
381 st->print("rlimit:");
382 struct rlimit rlim;
383
384 st->print(" STACK ");
385 getrlimit(RLIMIT_STACK, &rlim);
386 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
387 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
388
389 st->print(", CORE ");
390 getrlimit(RLIMIT_CORE, &rlim);
391 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
392 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
393
394 // Isn't there on solaris
395 #if defined(AIX)
396 st->print(", NPROC ");
397 st->print("%d", sysconf(_SC_CHILD_MAX));
398 #elif !defined(SOLARIS)
399 st->print(", NPROC ");
400 getrlimit(RLIMIT_NPROC, &rlim);
401 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
402 else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
403 #endif
404
405 st->print(", NOFILE ");
406 getrlimit(RLIMIT_NOFILE, &rlim);
407 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
408 else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
409
410 st->print(", AS ");
411 getrlimit(RLIMIT_AS, &rlim);
412 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
413 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
414
415 st->print(", DATA ");
416 getrlimit(RLIMIT_DATA, &rlim);
417 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
418 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
419
420 st->print(", FSIZE ");
421 getrlimit(RLIMIT_FSIZE, &rlim);
422 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
423 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
424
425 st->cr();
426 }
427
428 void os::Posix::print_uname_info(outputStream* st) {
429 // kernel
430 st->print("uname:");
431 struct utsname name;
432 uname(&name);
433 st->print("%s ", name.sysname);
434 #ifdef ASSERT
435 st->print("%s ", name.nodename);
436 #endif
437 st->print("%s ", name.release);
438 st->print("%s ", name.version);
439 st->print("%s", name.machine);
440 st->cr();
441 }
442
443 void os::Posix::print_umask(outputStream* st, mode_t umsk) {
444 st->print((umsk & S_IRUSR) ? "r" : "-");
445 st->print((umsk & S_IWUSR) ? "w" : "-");
446 st->print((umsk & S_IXUSR) ? "x" : "-");
447 st->print((umsk & S_IRGRP) ? "r" : "-");
448 st->print((umsk & S_IWGRP) ? "w" : "-");
449 st->print((umsk & S_IXGRP) ? "x" : "-");
450 st->print((umsk & S_IROTH) ? "r" : "-");
451 st->print((umsk & S_IWOTH) ? "w" : "-");
452 st->print((umsk & S_IXOTH) ? "x" : "-");
453 }
454
455 void os::Posix::print_user_info(outputStream* st) {
456 unsigned id = (unsigned) ::getuid();
457 st->print("uid : %u ", id);
458 id = (unsigned) ::geteuid();
459 st->print("euid : %u ", id);
460 id = (unsigned) ::getgid();
461 st->print("gid : %u ", id);
462 id = (unsigned) ::getegid();
463 st->print_cr("egid : %u", id);
464 st->cr();
465
466 mode_t umsk = ::umask(0);
467 ::umask(umsk);
468 st->print("umask: %04o (", (unsigned) umsk);
469 print_umask(st, umsk);
470 st->print_cr(")");
471 st->cr();
472 }
473
474
475 bool os::get_host_name(char* buf, size_t buflen) {
476 struct utsname name;
477 uname(&name);
478 jio_snprintf(buf, buflen, "%s", name.nodename);
479 return true;
480 }
481
482 bool os::has_allocatable_memory_limit(julong* limit) {
483 struct rlimit rlim;
484 int getrlimit_res = getrlimit(RLIMIT_AS, &rlim);
485 // if there was an error when calling getrlimit, assume that there is no limitation
486 // on virtual memory.
487 bool result;
488 if ((getrlimit_res != 0) || (rlim.rlim_cur == RLIM_INFINITY)) {
489 result = false;
490 } else {
491 *limit = (julong)rlim.rlim_cur;
492 result = true;
493 }
494 #ifdef _LP64
495 return result;
496 #else
497 // arbitrary virtual space limit for 32 bit Unices found by testing. If
498 // getrlimit above returned a limit, bound it with this limit. Otherwise
499 // directly use it.
500 const julong max_virtual_limit = (julong)3800*M;
501 if (result) {
502 *limit = MIN2(*limit, max_virtual_limit);
503 } else {
504 *limit = max_virtual_limit;
505 }
506
507 // bound by actually allocatable memory. The algorithm uses two bounds, an
508 // upper and a lower limit. The upper limit is the current highest amount of
509 // memory that could not be allocated, the lower limit is the current highest
510 // amount of memory that could be allocated.
511 // The algorithm iteratively refines the result by halving the difference
512 // between these limits, updating either the upper limit (if that value could
513 // not be allocated) or the lower limit (if the that value could be allocated)
514 // until the difference between these limits is "small".
515
516 // the minimum amount of memory we care about allocating.
517 const julong min_allocation_size = M;
518
519 julong upper_limit = *limit;
520
521 // first check a few trivial cases
522 if (is_allocatable(upper_limit) || (upper_limit <= min_allocation_size)) {
523 *limit = upper_limit;
524 } else if (!is_allocatable(min_allocation_size)) {
525 // we found that not even min_allocation_size is allocatable. Return it
526 // anyway. There is no point to search for a better value any more.
527 *limit = min_allocation_size;
528 } else {
529 // perform the binary search.
530 julong lower_limit = min_allocation_size;
531 while ((upper_limit - lower_limit) > min_allocation_size) {
532 julong temp_limit = ((upper_limit - lower_limit) / 2) + lower_limit;
533 temp_limit = align_down(temp_limit, min_allocation_size);
534 if (is_allocatable(temp_limit)) {
535 lower_limit = temp_limit;
536 } else {
537 upper_limit = temp_limit;
538 }
539 }
540 *limit = lower_limit;
541 }
542 return true;
543 #endif
544 }
545
546 const char* os::get_current_directory(char *buf, size_t buflen) {
547 return getcwd(buf, buflen);
548 }
549
550 FILE* os::open(int fd, const char* mode) {
551 return ::fdopen(fd, mode);
552 }
553
554 ssize_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
555 return ::pread(fd, buf, nBytes, offset);
556 }
557
558 void os::flockfile(FILE* fp) {
559 ::flockfile(fp);
560 }
561
562 void os::funlockfile(FILE* fp) {
563 ::funlockfile(fp);
564 }
565
566 DIR* os::opendir(const char* dirname) {
567 assert(dirname != NULL, "just checking");
568 return ::opendir(dirname);
569 }
570
571 struct dirent* os::readdir(DIR* dirp) {
572 assert(dirp != NULL, "just checking");
573 return ::readdir(dirp);
574 }
575
576 int os::closedir(DIR *dirp) {
577 assert(dirp != NULL, "just checking");
578 return ::closedir(dirp);
579 }
580
581 // Builds a platform dependent Agent_OnLoad_<lib_name> function name
582 // which is used to find statically linked in agents.
583 // Parameters:
584 // sym_name: Symbol in library we are looking for
585 // lib_name: Name of library to look in, NULL for shared libs.
586 // is_absolute_path == true if lib_name is absolute path to agent
587 // such as "/a/b/libL.so"
588 // == false if only the base name of the library is passed in
589 // such as "L"
590 char* os::build_agent_function_name(const char *sym_name, const char *lib_name,
591 bool is_absolute_path) {
592 char *agent_entry_name;
593 size_t len;
594 size_t name_len;
595 size_t prefix_len = strlen(JNI_LIB_PREFIX);
596 size_t suffix_len = strlen(JNI_LIB_SUFFIX);
597 const char *start;
598
599 if (lib_name != NULL) {
600 name_len = strlen(lib_name);
601 if (is_absolute_path) {
602 // Need to strip path, prefix and suffix
603 if ((start = strrchr(lib_name, *os::file_separator())) != NULL) {
604 lib_name = ++start;
605 }
606 if (strlen(lib_name) <= (prefix_len + suffix_len)) {
607 return NULL;
608 }
609 lib_name += prefix_len;
610 name_len = strlen(lib_name) - suffix_len;
611 }
612 }
613 len = (lib_name != NULL ? name_len : 0) + strlen(sym_name) + 2;
614 agent_entry_name = NEW_C_HEAP_ARRAY_RETURN_NULL(char, len, mtThread);
615 if (agent_entry_name == NULL) {
616 return NULL;
617 }
618 strcpy(agent_entry_name, sym_name);
619 if (lib_name != NULL) {
620 strcat(agent_entry_name, "_");
621 strncat(agent_entry_name, lib_name, name_len);
622 }
623 return agent_entry_name;
624 }
625
626 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
627 assert(thread == Thread::current(), "thread consistency check");
628
629 ParkEvent * const slp = thread->_SleepEvent ;
630 slp->reset() ;
631 OrderAccess::fence() ;
632
633 if (interruptible) {
634 jlong prevtime = javaTimeNanos();
635
636 for (;;) {
637 if (os::is_interrupted(thread, true)) {
638 return OS_INTRPT;
639 }
640
641 jlong newtime = javaTimeNanos();
642
643 if (newtime - prevtime < 0) {
644 // time moving backwards, should only happen if no monotonic clock
645 // not a guarantee() because JVM should not abort on kernel/glibc bugs
646 assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected in os::sleep(interruptible)");
647 } else {
648 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
649 }
650
651 if (millis <= 0) {
652 return OS_OK;
653 }
654
655 prevtime = newtime;
656
657 {
658 assert(thread->is_Java_thread(), "sanity check");
659 JavaThread *jt = (JavaThread *) thread;
660 ThreadBlockInVM tbivm(jt);
661 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
662
663 jt->set_suspend_equivalent();
664 // cleared by handle_special_suspend_equivalent_condition() or
665 // java_suspend_self() via check_and_wait_while_suspended()
666
667 slp->park(millis);
668
669 // were we externally suspended while we were waiting?
670 jt->check_and_wait_while_suspended();
671 }
672 }
673 } else {
674 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
675 jlong prevtime = javaTimeNanos();
676
677 for (;;) {
678 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
679 // the 1st iteration ...
680 jlong newtime = javaTimeNanos();
681
682 if (newtime - prevtime < 0) {
683 // time moving backwards, should only happen if no monotonic clock
684 // not a guarantee() because JVM should not abort on kernel/glibc bugs
685 assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected on os::sleep(!interruptible)");
686 } else {
687 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
688 }
689
690 if (millis <= 0) break ;
691
692 prevtime = newtime;
693 slp->park(millis);
694 }
695 return OS_OK ;
696 }
697 }
698
699 void os::naked_short_nanosleep(jlong ns) {
700 struct timespec req;
701 assert(ns > -1 && ns < NANOUNITS, "Un-interruptable sleep, short time use only");
702 req.tv_sec = 0;
703 req.tv_nsec = ns;
704 ::nanosleep(&req, NULL);
705 return;
706 }
707
708 void os::naked_short_sleep(jlong ms) {
709 assert(ms < MILLIUNITS, "Un-interruptable sleep, short time use only");
710 os::naked_short_nanosleep(ms * (NANOUNITS / MILLIUNITS));
711 return;
712 }
713
714 ////////////////////////////////////////////////////////////////////////////////
715 // interrupt support
716
717 void os::interrupt(Thread* thread) {
718 debug_only(Thread::check_for_dangling_thread_pointer(thread);)
719
720 OSThread* osthread = thread->osthread();
721
722 if (!osthread->interrupted()) {
723 osthread->set_interrupted(true);
724 // More than one thread can get here with the same value of osthread,
725 // resulting in multiple notifications. We do, however, want the store
726 // to interrupted() to be visible to other threads before we execute unpark().
727 OrderAccess::fence();
728 ParkEvent * const slp = thread->_SleepEvent ;
729 if (slp != NULL) slp->unpark() ;
730 }
731
732 // For JSR166. Unpark even if interrupt status already was set
733 if (thread->is_Java_thread())
734 ((JavaThread*)thread)->parker()->unpark();
735
736 ParkEvent * ev = thread->_ParkEvent ;
737 if (ev != NULL) ev->unpark() ;
738 }
739
740 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
741 debug_only(Thread::check_for_dangling_thread_pointer(thread);)
742
743 OSThread* osthread = thread->osthread();
744
745 bool interrupted = osthread->interrupted();
746
747 // NOTE that since there is no "lock" around the interrupt and
748 // is_interrupted operations, there is the possibility that the
749 // interrupted flag (in osThread) will be "false" but that the
750 // low-level events will be in the signaled state. This is
751 // intentional. The effect of this is that Object.wait() and
752 // LockSupport.park() will appear to have a spurious wakeup, which
753 // is allowed and not harmful, and the possibility is so rare that
754 // it is not worth the added complexity to add yet another lock.
755 // For the sleep event an explicit reset is performed on entry
756 // to os::sleep, so there is no early return. It has also been
757 // recommended not to put the interrupted flag into the "event"
758 // structure because it hides the issue.
759 if (interrupted && clear_interrupted) {
760 osthread->set_interrupted(false);
761 // consider thread->_SleepEvent->reset() ... optional optimization
762 }
763
764 return interrupted;
765 }
766
767
768
769 static const struct {
770 int sig; const char* name;
771 }
772 g_signal_info[] =
773 {
774 { SIGABRT, "SIGABRT" },
775 #ifdef SIGAIO
776 { SIGAIO, "SIGAIO" },
777 #endif
778 { SIGALRM, "SIGALRM" },
779 #ifdef SIGALRM1
780 { SIGALRM1, "SIGALRM1" },
781 #endif
782 { SIGBUS, "SIGBUS" },
783 #ifdef SIGCANCEL
784 { SIGCANCEL, "SIGCANCEL" },
785 #endif
786 { SIGCHLD, "SIGCHLD" },
787 #ifdef SIGCLD
788 { SIGCLD, "SIGCLD" },
789 #endif
790 { SIGCONT, "SIGCONT" },
791 #ifdef SIGCPUFAIL
792 { SIGCPUFAIL, "SIGCPUFAIL" },
793 #endif
794 #ifdef SIGDANGER
795 { SIGDANGER, "SIGDANGER" },
796 #endif
797 #ifdef SIGDIL
798 { SIGDIL, "SIGDIL" },
799 #endif
800 #ifdef SIGEMT
801 { SIGEMT, "SIGEMT" },
802 #endif
803 { SIGFPE, "SIGFPE" },
804 #ifdef SIGFREEZE
805 { SIGFREEZE, "SIGFREEZE" },
806 #endif
807 #ifdef SIGGFAULT
808 { SIGGFAULT, "SIGGFAULT" },
809 #endif
810 #ifdef SIGGRANT
811 { SIGGRANT, "SIGGRANT" },
812 #endif
813 { SIGHUP, "SIGHUP" },
814 { SIGILL, "SIGILL" },
815 { SIGINT, "SIGINT" },
816 #ifdef SIGIO
817 { SIGIO, "SIGIO" },
818 #endif
819 #ifdef SIGIOINT
820 { SIGIOINT, "SIGIOINT" },
821 #endif
822 #ifdef SIGIOT
823 // SIGIOT is there for BSD compatibility, but on most Unices just a
824 // synonym for SIGABRT. The result should be "SIGABRT", not
825 // "SIGIOT".
826 #if (SIGIOT != SIGABRT )
827 { SIGIOT, "SIGIOT" },
828 #endif
829 #endif
830 #ifdef SIGKAP
831 { SIGKAP, "SIGKAP" },
832 #endif
833 { SIGKILL, "SIGKILL" },
834 #ifdef SIGLOST
835 { SIGLOST, "SIGLOST" },
836 #endif
837 #ifdef SIGLWP
838 { SIGLWP, "SIGLWP" },
839 #endif
840 #ifdef SIGLWPTIMER
841 { SIGLWPTIMER, "SIGLWPTIMER" },
842 #endif
843 #ifdef SIGMIGRATE
844 { SIGMIGRATE, "SIGMIGRATE" },
845 #endif
846 #ifdef SIGMSG
847 { SIGMSG, "SIGMSG" },
848 #endif
849 { SIGPIPE, "SIGPIPE" },
850 #ifdef SIGPOLL
851 { SIGPOLL, "SIGPOLL" },
852 #endif
853 #ifdef SIGPRE
854 { SIGPRE, "SIGPRE" },
855 #endif
856 { SIGPROF, "SIGPROF" },
857 #ifdef SIGPTY
858 { SIGPTY, "SIGPTY" },
859 #endif
860 #ifdef SIGPWR
861 { SIGPWR, "SIGPWR" },
862 #endif
863 { SIGQUIT, "SIGQUIT" },
864 #ifdef SIGRECONFIG
865 { SIGRECONFIG, "SIGRECONFIG" },
866 #endif
867 #ifdef SIGRECOVERY
868 { SIGRECOVERY, "SIGRECOVERY" },
869 #endif
870 #ifdef SIGRESERVE
871 { SIGRESERVE, "SIGRESERVE" },
872 #endif
873 #ifdef SIGRETRACT
874 { SIGRETRACT, "SIGRETRACT" },
875 #endif
876 #ifdef SIGSAK
877 { SIGSAK, "SIGSAK" },
878 #endif
879 { SIGSEGV, "SIGSEGV" },
880 #ifdef SIGSOUND
881 { SIGSOUND, "SIGSOUND" },
882 #endif
883 #ifdef SIGSTKFLT
884 { SIGSTKFLT, "SIGSTKFLT" },
885 #endif
886 { SIGSTOP, "SIGSTOP" },
887 { SIGSYS, "SIGSYS" },
888 #ifdef SIGSYSERROR
889 { SIGSYSERROR, "SIGSYSERROR" },
890 #endif
891 #ifdef SIGTALRM
892 { SIGTALRM, "SIGTALRM" },
893 #endif
894 { SIGTERM, "SIGTERM" },
895 #ifdef SIGTHAW
896 { SIGTHAW, "SIGTHAW" },
897 #endif
898 { SIGTRAP, "SIGTRAP" },
899 #ifdef SIGTSTP
900 { SIGTSTP, "SIGTSTP" },
901 #endif
902 { SIGTTIN, "SIGTTIN" },
903 { SIGTTOU, "SIGTTOU" },
904 #ifdef SIGURG
905 { SIGURG, "SIGURG" },
906 #endif
907 { SIGUSR1, "SIGUSR1" },
908 { SIGUSR2, "SIGUSR2" },
909 #ifdef SIGVIRT
910 { SIGVIRT, "SIGVIRT" },
911 #endif
912 { SIGVTALRM, "SIGVTALRM" },
913 #ifdef SIGWAITING
914 { SIGWAITING, "SIGWAITING" },
915 #endif
916 #ifdef SIGWINCH
917 { SIGWINCH, "SIGWINCH" },
918 #endif
919 #ifdef SIGWINDOW
920 { SIGWINDOW, "SIGWINDOW" },
921 #endif
922 { SIGXCPU, "SIGXCPU" },
923 { SIGXFSZ, "SIGXFSZ" },
924 #ifdef SIGXRES
925 { SIGXRES, "SIGXRES" },
926 #endif
927 { -1, NULL }
928 };
929
930 // Returned string is a constant. For unknown signals "UNKNOWN" is returned.
931 const char* os::Posix::get_signal_name(int sig, char* out, size_t outlen) {
932
933 const char* ret = NULL;
934
935 #ifdef SIGRTMIN
936 if (sig >= SIGRTMIN && sig <= SIGRTMAX) {
937 if (sig == SIGRTMIN) {
938 ret = "SIGRTMIN";
939 } else if (sig == SIGRTMAX) {
940 ret = "SIGRTMAX";
941 } else {
942 jio_snprintf(out, outlen, "SIGRTMIN+%d", sig - SIGRTMIN);
943 return out;
944 }
945 }
946 #endif
947
948 if (sig > 0) {
949 for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
950 if (g_signal_info[idx].sig == sig) {
951 ret = g_signal_info[idx].name;
952 break;
953 }
954 }
955 }
956
957 if (!ret) {
958 if (!is_valid_signal(sig)) {
959 ret = "INVALID";
960 } else {
961 ret = "UNKNOWN";
962 }
963 }
964
965 if (out && outlen > 0) {
966 strncpy(out, ret, outlen);
967 out[outlen - 1] = '\0';
968 }
969 return out;
970 }
971
972 int os::Posix::get_signal_number(const char* signal_name) {
973 char tmp[30];
974 const char* s = signal_name;
975 if (s[0] != 'S' || s[1] != 'I' || s[2] != 'G') {
976 jio_snprintf(tmp, sizeof(tmp), "SIG%s", signal_name);
977 s = tmp;
978 }
979 for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
980 if (strcmp(g_signal_info[idx].name, s) == 0) {
981 return g_signal_info[idx].sig;
982 }
983 }
984 return -1;
985 }
986
987 int os::get_signal_number(const char* signal_name) {
988 return os::Posix::get_signal_number(signal_name);
989 }
990
991 // Returns true if signal number is valid.
992 bool os::Posix::is_valid_signal(int sig) {
993 // MacOS not really POSIX compliant: sigaddset does not return
994 // an error for invalid signal numbers. However, MacOS does not
995 // support real time signals and simply seems to have just 33
996 // signals with no holes in the signal range.
997 #ifdef __APPLE__
998 return sig >= 1 && sig < NSIG;
999 #else
1000 // Use sigaddset to check for signal validity.
1001 sigset_t set;
1002 sigemptyset(&set);
1003 if (sigaddset(&set, sig) == -1 && errno == EINVAL) {
1004 return false;
1005 }
1006 return true;
1007 #endif
1008 }
1009
1010 bool os::Posix::is_sig_ignored(int sig) {
1011 struct sigaction oact;
1012 sigaction(sig, (struct sigaction*)NULL, &oact);
1013 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
1014 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
1015 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
1016 return true;
1017 } else {
1018 return false;
1019 }
1020 }
1021
1022 // Returns:
1023 // NULL for an invalid signal number
1024 // "SIG<num>" for a valid but unknown signal number
1025 // signal name otherwise.
1026 const char* os::exception_name(int sig, char* buf, size_t size) {
1027 if (!os::Posix::is_valid_signal(sig)) {
1028 return NULL;
1029 }
1030 const char* const name = os::Posix::get_signal_name(sig, buf, size);
1031 if (strcmp(name, "UNKNOWN") == 0) {
1032 jio_snprintf(buf, size, "SIG%d", sig);
1033 }
1034 return buf;
1035 }
1036
1037 #define NUM_IMPORTANT_SIGS 32
1038 // Returns one-line short description of a signal set in a user provided buffer.
1039 const char* os::Posix::describe_signal_set_short(const sigset_t* set, char* buffer, size_t buf_size) {
1040 assert(buf_size == (NUM_IMPORTANT_SIGS + 1), "wrong buffer size");
1041 // Note: for shortness, just print out the first 32. That should
1042 // cover most of the useful ones, apart from realtime signals.
1043 for (int sig = 1; sig <= NUM_IMPORTANT_SIGS; sig++) {
1044 const int rc = sigismember(set, sig);
1045 if (rc == -1 && errno == EINVAL) {
1046 buffer[sig-1] = '?';
1047 } else {
1048 buffer[sig-1] = rc == 0 ? '0' : '1';
1049 }
1050 }
1051 buffer[NUM_IMPORTANT_SIGS] = 0;
1052 return buffer;
1053 }
1054
1055 // Prints one-line description of a signal set.
1056 void os::Posix::print_signal_set_short(outputStream* st, const sigset_t* set) {
1057 char buf[NUM_IMPORTANT_SIGS + 1];
1058 os::Posix::describe_signal_set_short(set, buf, sizeof(buf));
1059 st->print("%s", buf);
1060 }
1061
1062 // Writes one-line description of a combination of sigaction.sa_flags into a user
1063 // provided buffer. Returns that buffer.
1064 const char* os::Posix::describe_sa_flags(int flags, char* buffer, size_t size) {
1065 char* p = buffer;
1066 size_t remaining = size;
1067 bool first = true;
1068 int idx = 0;
1069
1070 assert(buffer, "invalid argument");
1071
1072 if (size == 0) {
1073 return buffer;
1074 }
1075
1076 strncpy(buffer, "none", size);
1077
1078 const struct {
1079 // NB: i is an unsigned int here because SA_RESETHAND is on some
1080 // systems 0x80000000, which is implicitly unsigned. Assignining
1081 // it to an int field would be an overflow in unsigned-to-signed
1082 // conversion.
1083 unsigned int i;
1084 const char* s;
1085 } flaginfo [] = {
1086 { SA_NOCLDSTOP, "SA_NOCLDSTOP" },
1087 { SA_ONSTACK, "SA_ONSTACK" },
1088 { SA_RESETHAND, "SA_RESETHAND" },
1089 { SA_RESTART, "SA_RESTART" },
1090 { SA_SIGINFO, "SA_SIGINFO" },
1091 { SA_NOCLDWAIT, "SA_NOCLDWAIT" },
1092 { SA_NODEFER, "SA_NODEFER" },
1093 #ifdef AIX
1094 { SA_ONSTACK, "SA_ONSTACK" },
1095 { SA_OLDSTYLE, "SA_OLDSTYLE" },
1096 #endif
1097 { 0, NULL }
1098 };
1099
1100 for (idx = 0; flaginfo[idx].s && remaining > 1; idx++) {
1101 if (flags & flaginfo[idx].i) {
1102 if (first) {
1103 jio_snprintf(p, remaining, "%s", flaginfo[idx].s);
1104 first = false;
1105 } else {
1106 jio_snprintf(p, remaining, "|%s", flaginfo[idx].s);
1107 }
1108 const size_t len = strlen(p);
1109 p += len;
1110 remaining -= len;
1111 }
1112 }
1113
1114 buffer[size - 1] = '\0';
1115
1116 return buffer;
1117 }
1118
1119 // Prints one-line description of a combination of sigaction.sa_flags.
1120 void os::Posix::print_sa_flags(outputStream* st, int flags) {
1121 char buffer[0x100];
1122 os::Posix::describe_sa_flags(flags, buffer, sizeof(buffer));
1123 st->print("%s", buffer);
1124 }
1125
1126 // Helper function for os::Posix::print_siginfo_...():
1127 // return a textual description for signal code.
1128 struct enum_sigcode_desc_t {
1129 const char* s_name;
1130 const char* s_desc;
1131 };
1132
1133 static bool get_signal_code_description(const siginfo_t* si, enum_sigcode_desc_t* out) {
1134
1135 const struct {
1136 int sig; int code; const char* s_code; const char* s_desc;
1137 } t1 [] = {
1138 { SIGILL, ILL_ILLOPC, "ILL_ILLOPC", "Illegal opcode." },
1139 { SIGILL, ILL_ILLOPN, "ILL_ILLOPN", "Illegal operand." },
1140 { SIGILL, ILL_ILLADR, "ILL_ILLADR", "Illegal addressing mode." },
1141 { SIGILL, ILL_ILLTRP, "ILL_ILLTRP", "Illegal trap." },
1142 { SIGILL, ILL_PRVOPC, "ILL_PRVOPC", "Privileged opcode." },
1143 { SIGILL, ILL_PRVREG, "ILL_PRVREG", "Privileged register." },
1144 { SIGILL, ILL_COPROC, "ILL_COPROC", "Coprocessor error." },
1145 { SIGILL, ILL_BADSTK, "ILL_BADSTK", "Internal stack error." },
1146 #if defined(IA64) && defined(LINUX)
1147 { SIGILL, ILL_BADIADDR, "ILL_BADIADDR", "Unimplemented instruction address" },
1148 { SIGILL, ILL_BREAK, "ILL_BREAK", "Application Break instruction" },
1149 #endif
1150 { SIGFPE, FPE_INTDIV, "FPE_INTDIV", "Integer divide by zero." },
1151 { SIGFPE, FPE_INTOVF, "FPE_INTOVF", "Integer overflow." },
1152 { SIGFPE, FPE_FLTDIV, "FPE_FLTDIV", "Floating-point divide by zero." },
1153 { SIGFPE, FPE_FLTOVF, "FPE_FLTOVF", "Floating-point overflow." },
1154 { SIGFPE, FPE_FLTUND, "FPE_FLTUND", "Floating-point underflow." },
1155 { SIGFPE, FPE_FLTRES, "FPE_FLTRES", "Floating-point inexact result." },
1156 { SIGFPE, FPE_FLTINV, "FPE_FLTINV", "Invalid floating-point operation." },
1157 { SIGFPE, FPE_FLTSUB, "FPE_FLTSUB", "Subscript out of range." },
1158 { SIGSEGV, SEGV_MAPERR, "SEGV_MAPERR", "Address not mapped to object." },
1159 { SIGSEGV, SEGV_ACCERR, "SEGV_ACCERR", "Invalid permissions for mapped object." },
1160 #ifdef AIX
1161 // no explanation found what keyerr would be
1162 { SIGSEGV, SEGV_KEYERR, "SEGV_KEYERR", "key error" },
1163 #endif
1164 #if defined(IA64) && !defined(AIX)
1165 { SIGSEGV, SEGV_PSTKOVF, "SEGV_PSTKOVF", "Paragraph stack overflow" },
1166 #endif
1167 #if defined(__sparc) && defined(SOLARIS)
1168 // define Solaris Sparc M7 ADI SEGV signals
1169 #if !defined(SEGV_ACCADI)
1170 #define SEGV_ACCADI 3
1171 #endif
1172 { SIGSEGV, SEGV_ACCADI, "SEGV_ACCADI", "ADI not enabled for mapped object." },
1173 #if !defined(SEGV_ACCDERR)
1174 #define SEGV_ACCDERR 4
1175 #endif
1176 { SIGSEGV, SEGV_ACCDERR, "SEGV_ACCDERR", "ADI disrupting exception." },
1177 #if !defined(SEGV_ACCPERR)
1178 #define SEGV_ACCPERR 5
1179 #endif
1180 { SIGSEGV, SEGV_ACCPERR, "SEGV_ACCPERR", "ADI precise exception." },
1181 #endif // defined(__sparc) && defined(SOLARIS)
1182 { SIGBUS, BUS_ADRALN, "BUS_ADRALN", "Invalid address alignment." },
1183 { SIGBUS, BUS_ADRERR, "BUS_ADRERR", "Nonexistent physical address." },
1184 { SIGBUS, BUS_OBJERR, "BUS_OBJERR", "Object-specific hardware error." },
1185 { SIGTRAP, TRAP_BRKPT, "TRAP_BRKPT", "Process breakpoint." },
1186 { SIGTRAP, TRAP_TRACE, "TRAP_TRACE", "Process trace trap." },
1187 { SIGCHLD, CLD_EXITED, "CLD_EXITED", "Child has exited." },
1188 { SIGCHLD, CLD_KILLED, "CLD_KILLED", "Child has terminated abnormally and did not create a core file." },
1189 { SIGCHLD, CLD_DUMPED, "CLD_DUMPED", "Child has terminated abnormally and created a core file." },
1190 { SIGCHLD, CLD_TRAPPED, "CLD_TRAPPED", "Traced child has trapped." },
1191 { SIGCHLD, CLD_STOPPED, "CLD_STOPPED", "Child has stopped." },
1192 { SIGCHLD, CLD_CONTINUED,"CLD_CONTINUED","Stopped child has continued." },
1193 #ifdef SIGPOLL
1194 { SIGPOLL, POLL_OUT, "POLL_OUT", "Output buffers available." },
1195 { SIGPOLL, POLL_MSG, "POLL_MSG", "Input message available." },
1196 { SIGPOLL, POLL_ERR, "POLL_ERR", "I/O error." },
1197 { SIGPOLL, POLL_PRI, "POLL_PRI", "High priority input available." },
1198 { SIGPOLL, POLL_HUP, "POLL_HUP", "Device disconnected. [Option End]" },
1199 #endif
1200 { -1, -1, NULL, NULL }
1201 };
1202
1203 // Codes valid in any signal context.
1204 const struct {
1205 int code; const char* s_code; const char* s_desc;
1206 } t2 [] = {
1207 { SI_USER, "SI_USER", "Signal sent by kill()." },
1208 { SI_QUEUE, "SI_QUEUE", "Signal sent by the sigqueue()." },
1209 { SI_TIMER, "SI_TIMER", "Signal generated by expiration of a timer set by timer_settime()." },
1210 { SI_ASYNCIO, "SI_ASYNCIO", "Signal generated by completion of an asynchronous I/O request." },
1211 { SI_MESGQ, "SI_MESGQ", "Signal generated by arrival of a message on an empty message queue." },
1212 // Linux specific
1213 #ifdef SI_TKILL
1214 { SI_TKILL, "SI_TKILL", "Signal sent by tkill (pthread_kill)" },
1215 #endif
1216 #ifdef SI_DETHREAD
1217 { SI_DETHREAD, "SI_DETHREAD", "Signal sent by execve() killing subsidiary threads" },
1218 #endif
1219 #ifdef SI_KERNEL
1220 { SI_KERNEL, "SI_KERNEL", "Signal sent by kernel." },
1221 #endif
1222 #ifdef SI_SIGIO
1223 { SI_SIGIO, "SI_SIGIO", "Signal sent by queued SIGIO" },
1224 #endif
1225
1226 #ifdef AIX
1227 { SI_UNDEFINED, "SI_UNDEFINED","siginfo contains partial information" },
1228 { SI_EMPTY, "SI_EMPTY", "siginfo contains no useful information" },
1229 #endif
1230
1231 #ifdef __sun
1232 { SI_NOINFO, "SI_NOINFO", "No signal information" },
1233 { SI_RCTL, "SI_RCTL", "kernel generated signal via rctl action" },
1234 { SI_LWP, "SI_LWP", "Signal sent via lwp_kill" },
1235 #endif
1236
1237 { -1, NULL, NULL }
1238 };
1239
1240 const char* s_code = NULL;
1241 const char* s_desc = NULL;
1242
1243 for (int i = 0; t1[i].sig != -1; i ++) {
1244 if (t1[i].sig == si->si_signo && t1[i].code == si->si_code) {
1245 s_code = t1[i].s_code;
1246 s_desc = t1[i].s_desc;
1247 break;
1248 }
1249 }
1250
1251 if (s_code == NULL) {
1252 for (int i = 0; t2[i].s_code != NULL; i ++) {
1253 if (t2[i].code == si->si_code) {
1254 s_code = t2[i].s_code;
1255 s_desc = t2[i].s_desc;
1256 }
1257 }
1258 }
1259
1260 if (s_code == NULL) {
1261 out->s_name = "unknown";
1262 out->s_desc = "unknown";
1263 return false;
1264 }
1265
1266 out->s_name = s_code;
1267 out->s_desc = s_desc;
1268
1269 return true;
1270 }
1271
1272 void os::print_siginfo(outputStream* os, const void* si0) {
1273
1274 const siginfo_t* const si = (const siginfo_t*) si0;
1275
1276 char buf[20];
1277 os->print("siginfo:");
1278
1279 if (!si) {
1280 os->print(" <null>");
1281 return;
1282 }
1283
1284 const int sig = si->si_signo;
1285
1286 os->print(" si_signo: %d (%s)", sig, os::Posix::get_signal_name(sig, buf, sizeof(buf)));
1287
1288 enum_sigcode_desc_t ed;
1289 get_signal_code_description(si, &ed);
1290 os->print(", si_code: %d (%s)", si->si_code, ed.s_name);
1291
1292 if (si->si_errno) {
1293 os->print(", si_errno: %d", si->si_errno);
1294 }
1295
1296 // Output additional information depending on the signal code.
1297
1298 // Note: Many implementations lump si_addr, si_pid, si_uid etc. together as unions,
1299 // so it depends on the context which member to use. For synchronous error signals,
1300 // we print si_addr, unless the signal was sent by another process or thread, in
1301 // which case we print out pid or tid of the sender.
1302 if (si->si_code == SI_USER || si->si_code == SI_QUEUE) {
1303 const pid_t pid = si->si_pid;
1304 os->print(", si_pid: %ld", (long) pid);
1305 if (IS_VALID_PID(pid)) {
1306 const pid_t me = getpid();
1307 if (me == pid) {
1308 os->print(" (current process)");
1309 }
1310 } else {
1311 os->print(" (invalid)");
1312 }
1313 os->print(", si_uid: %ld", (long) si->si_uid);
1314 if (sig == SIGCHLD) {
1315 os->print(", si_status: %d", si->si_status);
1316 }
1317 } else if (sig == SIGSEGV || sig == SIGBUS || sig == SIGILL ||
1318 sig == SIGTRAP || sig == SIGFPE) {
1319 os->print(", si_addr: " PTR_FORMAT, p2i(si->si_addr));
1320 #ifdef SIGPOLL
1321 } else if (sig == SIGPOLL) {
1322 os->print(", si_band: %ld", si->si_band);
1323 #endif
1324 }
1325
1326 }
1327
1328 int os::Posix::unblock_thread_signal_mask(const sigset_t *set) {
1329 return pthread_sigmask(SIG_UNBLOCK, set, NULL);
1330 }
1331
1332 address os::Posix::ucontext_get_pc(const ucontext_t* ctx) {
1333 #if defined(AIX)
1334 return Aix::ucontext_get_pc(ctx);
1335 #elif defined(BSD)
1336 return Bsd::ucontext_get_pc(ctx);
1337 #elif defined(LINUX)
1338 return Linux::ucontext_get_pc(ctx);
1339 #elif defined(SOLARIS)
1340 return Solaris::ucontext_get_pc(ctx);
1341 #else
1342 VMError::report_and_die("unimplemented ucontext_get_pc");
1343 #endif
1344 }
1345
1346 void os::Posix::ucontext_set_pc(ucontext_t* ctx, address pc) {
1347 #if defined(AIX)
1348 Aix::ucontext_set_pc(ctx, pc);
1349 #elif defined(BSD)
1350 Bsd::ucontext_set_pc(ctx, pc);
1351 #elif defined(LINUX)
1352 Linux::ucontext_set_pc(ctx, pc);
1353 #elif defined(SOLARIS)
1354 Solaris::ucontext_set_pc(ctx, pc);
1355 #else
1356 VMError::report_and_die("unimplemented ucontext_get_pc");
1357 #endif
1358 }
1359
1360 char* os::Posix::describe_pthread_attr(char* buf, size_t buflen, const pthread_attr_t* attr) {
1361 size_t stack_size = 0;
1362 size_t guard_size = 0;
1363 int detachstate = 0;
1364 pthread_attr_getstacksize(attr, &stack_size);
1365 pthread_attr_getguardsize(attr, &guard_size);
1366 // Work around linux NPTL implementation error, see also os::create_thread() in os_linux.cpp.
1367 LINUX_ONLY(stack_size -= guard_size);
1368 pthread_attr_getdetachstate(attr, &detachstate);
1369 jio_snprintf(buf, buflen, "stacksize: " SIZE_FORMAT "k, guardsize: " SIZE_FORMAT "k, %s",
1370 stack_size / 1024, guard_size / 1024,
1371 (detachstate == PTHREAD_CREATE_DETACHED ? "detached" : "joinable"));
1372 return buf;
1373 }
1374
1375 char* os::Posix::realpath(const char* filename, char* outbuf, size_t outbuflen) {
1376
1377 if (filename == NULL || outbuf == NULL || outbuflen < 1) {
1378 assert(false, "os::Posix::realpath: invalid arguments.");
1379 errno = EINVAL;
1380 return NULL;
1381 }
1382
1383 char* result = NULL;
1384
1385 // This assumes platform realpath() is implemented according to POSIX.1-2008.
1386 // POSIX.1-2008 allows to specify NULL for the output buffer, in which case
1387 // output buffer is dynamically allocated and must be ::free()'d by the caller.
1388 char* p = ::realpath(filename, NULL);
1389 if (p != NULL) {
1390 if (strlen(p) < outbuflen) {
1391 strcpy(outbuf, p);
1392 result = outbuf;
1393 } else {
1394 errno = ENAMETOOLONG;
1395 }
1396 ::free(p); // *not* os::free
1397 } else {
1398 // Fallback for platforms struggling with modern Posix standards (AIX 5.3, 6.1). If realpath
1399 // returns EINVAL, this may indicate that realpath is not POSIX.1-2008 compatible and
1400 // that it complains about the NULL we handed down as user buffer.
1401 // In this case, use the user provided buffer but at least check whether realpath caused
1402 // a memory overwrite.
1403 if (errno == EINVAL) {
1404 outbuf[outbuflen - 1] = '\0';
1405 p = ::realpath(filename, outbuf);
1406 if (p != NULL) {
1407 guarantee(outbuf[outbuflen - 1] == '\0', "realpath buffer overwrite detected.");
1408 result = p;
1409 }
1410 }
1411 }
1412 return result;
1413
1414 }
1415
1416 int os::stat(const char *path, struct stat *sbuf) {
1417 return ::stat(path, sbuf);
1418 }
1419
1420 char * os::native_path(char *path) {
1421 return path;
1422 }
1423
1424 // Check minimum allowable stack sizes for thread creation and to initialize
1425 // the java system classes, including StackOverflowError - depends on page
1426 // size.
1427 // The space needed for frames during startup is platform dependent. It
1428 // depends on word size, platform calling conventions, C frame layout and
1429 // interpreter/C1/C2 design decisions. Therefore this is given in a
1430 // platform (os/cpu) dependent constant.
1431 // To this, space for guard mechanisms is added, which depends on the
1432 // page size which again depends on the concrete system the VM is running
1433 // on. Space for libc guard pages is not included in this size.
1434 jint os::Posix::set_minimum_stack_sizes() {
1435 size_t os_min_stack_allowed = SOLARIS_ONLY(thr_min_stack()) NOT_SOLARIS(PTHREAD_STACK_MIN);
1436
1437 _java_thread_min_stack_allowed = _java_thread_min_stack_allowed +
1438 JavaThread::stack_guard_zone_size() +
1439 JavaThread::stack_shadow_zone_size();
1440
1441 _java_thread_min_stack_allowed = align_up(_java_thread_min_stack_allowed, vm_page_size());
1442 _java_thread_min_stack_allowed = MAX2(_java_thread_min_stack_allowed, os_min_stack_allowed);
1443
1444 size_t stack_size_in_bytes = ThreadStackSize * K;
1445 if (stack_size_in_bytes != 0 &&
1446 stack_size_in_bytes < _java_thread_min_stack_allowed) {
1447 // The '-Xss' and '-XX:ThreadStackSize=N' options both set
1448 // ThreadStackSize so we go with "Java thread stack size" instead
1449 // of "ThreadStackSize" to be more friendly.
1450 tty->print_cr("\nThe Java thread stack size specified is too small. "
1451 "Specify at least " SIZE_FORMAT "k",
1452 _java_thread_min_stack_allowed / K);
1453 return JNI_ERR;
1454 }
1455
1456 // Make the stack size a multiple of the page size so that
1457 // the yellow/red zones can be guarded.
1458 JavaThread::set_stack_size_at_create(align_up(stack_size_in_bytes, vm_page_size()));
1459
1460 // Reminder: a compiler thread is a Java thread.
1461 _compiler_thread_min_stack_allowed = _compiler_thread_min_stack_allowed +
1462 JavaThread::stack_guard_zone_size() +
1463 JavaThread::stack_shadow_zone_size();
1464
1465 _compiler_thread_min_stack_allowed = align_up(_compiler_thread_min_stack_allowed, vm_page_size());
1466 _compiler_thread_min_stack_allowed = MAX2(_compiler_thread_min_stack_allowed, os_min_stack_allowed);
1467
1468 stack_size_in_bytes = CompilerThreadStackSize * K;
1469 if (stack_size_in_bytes != 0 &&
1470 stack_size_in_bytes < _compiler_thread_min_stack_allowed) {
1471 tty->print_cr("\nThe CompilerThreadStackSize specified is too small. "
1472 "Specify at least " SIZE_FORMAT "k",
1473 _compiler_thread_min_stack_allowed / K);
1474 return JNI_ERR;
1475 }
1476
1477 _vm_internal_thread_min_stack_allowed = align_up(_vm_internal_thread_min_stack_allowed, vm_page_size());
1478 _vm_internal_thread_min_stack_allowed = MAX2(_vm_internal_thread_min_stack_allowed, os_min_stack_allowed);
1479
1480 stack_size_in_bytes = VMThreadStackSize * K;
1481 if (stack_size_in_bytes != 0 &&
1482 stack_size_in_bytes < _vm_internal_thread_min_stack_allowed) {
1483 tty->print_cr("\nThe VMThreadStackSize specified is too small. "
1484 "Specify at least " SIZE_FORMAT "k",
1485 _vm_internal_thread_min_stack_allowed / K);
1486 return JNI_ERR;
1487 }
1488 return JNI_OK;
1489 }
1490
1491 // Called when creating the thread. The minimum stack sizes have already been calculated
1492 size_t os::Posix::get_initial_stack_size(ThreadType thr_type, size_t req_stack_size) {
1493 size_t stack_size;
1494 if (req_stack_size == 0) {
1495 stack_size = default_stack_size(thr_type);
1496 } else {
1497 stack_size = req_stack_size;
1498 }
1499
1500 switch (thr_type) {
1501 case os::java_thread:
1502 // Java threads use ThreadStackSize which default value can be
1503 // changed with the flag -Xss
1504 if (req_stack_size == 0 && JavaThread::stack_size_at_create() > 0) {
1505 // no requested size and we have a more specific default value
1506 stack_size = JavaThread::stack_size_at_create();
1507 }
1508 stack_size = MAX2(stack_size,
1509 _java_thread_min_stack_allowed);
1510 break;
1511 case os::compiler_thread:
1512 if (req_stack_size == 0 && CompilerThreadStackSize > 0) {
1513 // no requested size and we have a more specific default value
1514 stack_size = (size_t)(CompilerThreadStackSize * K);
1515 }
1516 stack_size = MAX2(stack_size,
1517 _compiler_thread_min_stack_allowed);
1518 break;
1519 case os::vm_thread:
1520 case os::pgc_thread:
1521 case os::cgc_thread:
1522 case os::watcher_thread:
1523 default: // presume the unknown thr_type is a VM internal
1524 if (req_stack_size == 0 && VMThreadStackSize > 0) {
1525 // no requested size and we have a more specific default value
1526 stack_size = (size_t)(VMThreadStackSize * K);
1527 }
1528
1529 stack_size = MAX2(stack_size,
1530 _vm_internal_thread_min_stack_allowed);
1531 break;
1532 }
1533
1534 // pthread_attr_setstacksize() may require that the size be rounded up to the OS page size.
1535 // Be careful not to round up to 0. Align down in that case.
1536 if (stack_size <= SIZE_MAX - vm_page_size()) {
1537 stack_size = align_up(stack_size, vm_page_size());
1538 } else {
1539 stack_size = align_down(stack_size, vm_page_size());
1540 }
1541
1542 return stack_size;
1543 }
1544
1545 bool os::Posix::is_root(uid_t uid){
1546 return ROOT_UID == uid;
1547 }
1548
1549 bool os::Posix::matches_effective_uid_or_root(uid_t uid) {
1550 return is_root(uid) || geteuid() == uid;
1551 }
1552
1553 bool os::Posix::matches_effective_uid_and_gid_or_root(uid_t uid, gid_t gid) {
1554 return is_root(uid) || (geteuid() == uid && getegid() == gid);
1555 }
1556
1557 Thread* os::ThreadCrashProtection::_protected_thread = NULL;
1558 os::ThreadCrashProtection* os::ThreadCrashProtection::_crash_protection = NULL;
1559 volatile intptr_t os::ThreadCrashProtection::_crash_mux = 0;
1560
1561 os::ThreadCrashProtection::ThreadCrashProtection() {
1562 }
1563
1564 /*
1565 * See the caveats for this class in os_posix.hpp
1566 * Protects the callback call so that SIGSEGV / SIGBUS jumps back into this
1567 * method and returns false. If none of the signals are raised, returns true.
1568 * The callback is supposed to provide the method that should be protected.
1569 */
1570 bool os::ThreadCrashProtection::call(os::CrashProtectionCallback& cb) {
1571 sigset_t saved_sig_mask;
1572
1573 Thread::muxAcquire(&_crash_mux, "CrashProtection");
1574
1575 _protected_thread = Thread::current_or_null();
1576 assert(_protected_thread != NULL, "Cannot crash protect a NULL thread");
1577
1578 // we cannot rely on sigsetjmp/siglongjmp to save/restore the signal mask
1579 // since on at least some systems (OS X) siglongjmp will restore the mask
1580 // for the process, not the thread
1581 pthread_sigmask(0, NULL, &saved_sig_mask);
1582 if (sigsetjmp(_jmpbuf, 0) == 0) {
1583 // make sure we can see in the signal handler that we have crash protection
1584 // installed
1585 _crash_protection = this;
1586 cb.call();
1587 // and clear the crash protection
1588 _crash_protection = NULL;
1589 _protected_thread = NULL;
1590 Thread::muxRelease(&_crash_mux);
1591 return true;
1592 }
1593 // this happens when we siglongjmp() back
1594 pthread_sigmask(SIG_SETMASK, &saved_sig_mask, NULL);
1595 _crash_protection = NULL;
1596 _protected_thread = NULL;
1597 Thread::muxRelease(&_crash_mux);
1598 return false;
1599 }
1600
1601 void os::ThreadCrashProtection::restore() {
1602 assert(_crash_protection != NULL, "must have crash protection");
1603 siglongjmp(_jmpbuf, 1);
1604 }
1605
1606 void os::ThreadCrashProtection::check_crash_protection(int sig,
1607 Thread* thread) {
1608
1609 if (thread != NULL &&
1610 thread == _protected_thread &&
1611 _crash_protection != NULL) {
1612
1613 if (sig == SIGSEGV || sig == SIGBUS) {
1614 _crash_protection->restore();
1615 }
1616 }
1617 }
1618
1619 // Shared clock/time and other supporting routines for pthread_mutex/cond
1620 // initialization. This is enabled on Solaris but only some of the clock/time
1621 // functionality is actually used there.
1622
1623 // Shared condattr object for use with relative timed-waits. Will be associated
1624 // with CLOCK_MONOTONIC if available to avoid issues with time-of-day changes,
1625 // but otherwise whatever default is used by the platform - generally the
1626 // time-of-day clock.
1627 static pthread_condattr_t _condAttr[1];
1628
1629 // Shared mutexattr to explicitly set the type to PTHREAD_MUTEX_NORMAL as not
1630 // all systems (e.g. FreeBSD) map the default to "normal".
1631 static pthread_mutexattr_t _mutexAttr[1];
1632
1633 // common basic initialization that is always supported
1634 static void pthread_init_common(void) {
1635 int status;
1636 if ((status = pthread_condattr_init(_condAttr)) != 0) {
1637 fatal("pthread_condattr_init: %s", os::strerror(status));
1638 }
1639 if ((status = pthread_mutexattr_init(_mutexAttr)) != 0) {
1640 fatal("pthread_mutexattr_init: %s", os::strerror(status));
1641 }
1642 if ((status = pthread_mutexattr_settype(_mutexAttr, PTHREAD_MUTEX_NORMAL)) != 0) {
1643 fatal("pthread_mutexattr_settype: %s", os::strerror(status));
1644 }
1645 }
1646
1647 // Not all POSIX types and API's are available on all notionally "posix"
1648 // platforms. If we have build-time support then we will check for actual
1649 // runtime support via dlopen/dlsym lookup. This allows for running on an
1650 // older OS version compared to the build platform. But if there is no
1651 // build time support then there cannot be any runtime support as we do not
1652 // know what the runtime types would be (for example clockid_t might be an
1653 // int or int64_t).
1654 //
1655 #ifdef SUPPORTS_CLOCK_MONOTONIC
1656
1657 // This means we have clockid_t, clock_gettime et al and CLOCK_MONOTONIC
1658
1659 int (*os::Posix::_clock_gettime)(clockid_t, struct timespec *) = NULL;
1660 int (*os::Posix::_clock_getres)(clockid_t, struct timespec *) = NULL;
1661
1662 static int (*_pthread_condattr_setclock)(pthread_condattr_t *, clockid_t) = NULL;
1663
1664 static bool _use_clock_monotonic_condattr = false;
1665
1666 // Determine what POSIX API's are present and do appropriate
1667 // configuration.
1668 void os::Posix::init(void) {
1669
1670 // NOTE: no logging available when this is called. Put logging
1671 // statements in init_2().
1672
1673 // 1. Check for CLOCK_MONOTONIC support.
1674
1675 void* handle = NULL;
1676
1677 // For linux we need librt, for other OS we can find
1678 // this function in regular libc.
1679 #ifdef NEEDS_LIBRT
1680 // We do dlopen's in this particular order due to bug in linux
1681 // dynamic loader (see 6348968) leading to crash on exit.
1682 handle = dlopen("librt.so.1", RTLD_LAZY);
1683 if (handle == NULL) {
1684 handle = dlopen("librt.so", RTLD_LAZY);
1685 }
1686 #endif
1687
1688 if (handle == NULL) {
1689 handle = RTLD_DEFAULT;
1690 }
1691
1692 int (*clock_getres_func)(clockid_t, struct timespec*) =
1693 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1694 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1695 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1696 if (clock_getres_func != NULL && clock_gettime_func != NULL) {
1697 // We assume that if both clock_gettime and clock_getres support
1698 // CLOCK_MONOTONIC then the OS provides true high-res monotonic clock.
1699 struct timespec res;
1700 struct timespec tp;
1701 if (clock_getres_func(CLOCK_MONOTONIC, &res) == 0 &&
1702 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1703 // Yes, monotonic clock is supported.
1704 _clock_gettime = clock_gettime_func;
1705 _clock_getres = clock_getres_func;
1706 } else {
1707 #ifdef NEEDS_LIBRT
1708 // Close librt if there is no monotonic clock.
1709 if (handle != RTLD_DEFAULT) {
1710 dlclose(handle);
1711 }
1712 #endif
1713 }
1714 }
1715
1716 // 2. Check for pthread_condattr_setclock support.
1717
1718 // libpthread is already loaded.
1719 int (*condattr_setclock_func)(pthread_condattr_t*, clockid_t) =
1720 (int (*)(pthread_condattr_t*, clockid_t))dlsym(RTLD_DEFAULT,
1721 "pthread_condattr_setclock");
1722 if (condattr_setclock_func != NULL) {
1723 _pthread_condattr_setclock = condattr_setclock_func;
1724 }
1725
1726 // Now do general initialization.
1727
1728 pthread_init_common();
1729
1730 #ifndef SOLARIS
1731 int status;
1732 if (_pthread_condattr_setclock != NULL && _clock_gettime != NULL) {
1733 if ((status = _pthread_condattr_setclock(_condAttr, CLOCK_MONOTONIC)) != 0) {
1734 if (status == EINVAL) {
1735 _use_clock_monotonic_condattr = false;
1736 warning("Unable to use monotonic clock with relative timed-waits" \
1737 " - changes to the time-of-day clock may have adverse affects");
1738 } else {
1739 fatal("pthread_condattr_setclock: %s", os::strerror(status));
1740 }
1741 } else {
1742 _use_clock_monotonic_condattr = true;
1743 }
1744 }
1745 #endif // !SOLARIS
1746
1747 }
1748
1749 void os::Posix::init_2(void) {
1750 #ifndef SOLARIS
1751 log_info(os)("Use of CLOCK_MONOTONIC is%s supported",
1752 (_clock_gettime != NULL ? "" : " not"));
1753 log_info(os)("Use of pthread_condattr_setclock is%s supported",
1754 (_pthread_condattr_setclock != NULL ? "" : " not"));
1755 log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with %s",
1756 _use_clock_monotonic_condattr ? "CLOCK_MONOTONIC" : "the default clock");
1757 #endif // !SOLARIS
1758 }
1759
1760 #else // !SUPPORTS_CLOCK_MONOTONIC
1761
1762 void os::Posix::init(void) {
1763 pthread_init_common();
1764 }
1765
1766 void os::Posix::init_2(void) {
1767 #ifndef SOLARIS
1768 log_info(os)("Use of CLOCK_MONOTONIC is not supported");
1769 log_info(os)("Use of pthread_condattr_setclock is not supported");
1770 log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with the default clock");
1771 #endif // !SOLARIS
1772 }
1773
1774 #endif // SUPPORTS_CLOCK_MONOTONIC
1775
1776 // Utility to convert the given timeout to an absolute timespec
1777 // (based on the appropriate clock) to use with pthread_cond_timewait,
1778 // and sem_timedwait().
1779 // The clock queried here must be the clock used to manage the
1780 // timeout of the condition variable or semaphore.
1781 //
1782 // The passed in timeout value is either a relative time in nanoseconds
1783 // or an absolute time in milliseconds. A relative timeout will be
1784 // associated with CLOCK_MONOTONIC if available, unless the real-time clock
1785 // is explicitly requested; otherwise, or if absolute,
1786 // the default time-of-day clock will be used.
1787
1788 // Given time is a 64-bit value and the time_t used in the timespec is
1789 // sometimes a signed-32-bit value we have to watch for overflow if times
1790 // way in the future are given. Further on Solaris versions
1791 // prior to 10 there is a restriction (see cond_timedwait) that the specified
1792 // number of seconds, in abstime, is less than current_time + 100000000.
1793 // As it will be over 20 years before "now + 100000000" will overflow we can
1794 // ignore overflow and just impose a hard-limit on seconds using the value
1795 // of "now + 100000000". This places a limit on the timeout of about 3.17
1796 // years from "now".
1797 //
1798 #define MAX_SECS 100000000
1799
1800 // Calculate a new absolute time that is "timeout" nanoseconds from "now".
1801 // "unit" indicates the unit of "now_part_sec" (may be nanos or micros depending
1802 // on which clock API is being used).
1803 static void calc_rel_time(timespec* abstime, jlong timeout, jlong now_sec,
1804 jlong now_part_sec, jlong unit) {
1805 time_t max_secs = now_sec + MAX_SECS;
1806
1807 jlong seconds = timeout / NANOUNITS;
1808 timeout %= NANOUNITS; // remaining nanos
1809
1810 if (seconds >= MAX_SECS) {
1811 // More seconds than we can add, so pin to max_secs.
1812 abstime->tv_sec = max_secs;
1813 abstime->tv_nsec = 0;
1814 } else {
1815 abstime->tv_sec = now_sec + seconds;
1816 long nanos = (now_part_sec * (NANOUNITS / unit)) + timeout;
1817 if (nanos >= NANOUNITS) { // overflow
1818 abstime->tv_sec += 1;
1819 nanos -= NANOUNITS;
1820 }
1821 abstime->tv_nsec = nanos;
1822 }
1823 }
1824
1825 // Unpack the given deadline in milliseconds since the epoch, into the given timespec.
1826 // The current time in seconds is also passed in to enforce an upper bound as discussed above.
1827 // This is only used with gettimeofday, when clock_gettime is not available.
1828 static void unpack_abs_time(timespec* abstime, jlong deadline, jlong now_sec) {
1829 time_t max_secs = now_sec + MAX_SECS;
1830
1831 jlong seconds = deadline / MILLIUNITS;
1832 jlong millis = deadline % MILLIUNITS;
1833
1834 if (seconds >= max_secs) {
1835 // Absolute seconds exceeds allowed max, so pin to max_secs.
1836 abstime->tv_sec = max_secs;
1837 abstime->tv_nsec = 0;
1838 } else {
1839 abstime->tv_sec = seconds;
1840 abstime->tv_nsec = millis * (NANOUNITS / MILLIUNITS);
1841 }
1842 }
1843
1844 static jlong millis_to_nanos(jlong millis) {
1845 // We have to watch for overflow when converting millis to nanos,
1846 // but if millis is that large then we will end up limiting to
1847 // MAX_SECS anyway, so just do that here.
1848 if (millis / MILLIUNITS > MAX_SECS) {
1849 millis = jlong(MAX_SECS) * MILLIUNITS;
1850 }
1851 return millis * (NANOUNITS / MILLIUNITS);
1852 }
1853
1854 static void to_abstime(timespec* abstime, jlong timeout,
1855 bool isAbsolute, bool isRealtime) {
1856 DEBUG_ONLY(int max_secs = MAX_SECS;)
1857
1858 if (timeout < 0) {
1859 timeout = 0;
1860 }
1861
1862 #ifdef SUPPORTS_CLOCK_MONOTONIC
1863
1864 clockid_t clock = CLOCK_MONOTONIC;
1865 // need to ensure we have a runtime check for clock_gettime support
1866 if (!isAbsolute && os::Posix::supports_monotonic_clock()) {
1867 if (!_use_clock_monotonic_condattr || isRealtime) {
1868 clock = CLOCK_REALTIME;
1869 }
1870 struct timespec now;
1871 int status = os::Posix::clock_gettime(clock, &now);
1872 assert_status(status == 0, status, "clock_gettime");
1873 calc_rel_time(abstime, timeout, now.tv_sec, now.tv_nsec, NANOUNITS);
1874 DEBUG_ONLY(max_secs += now.tv_sec;)
1875 } else {
1876
1877 #else
1878
1879 { // Match the block scope.
1880
1881 #endif // SUPPORTS_CLOCK_MONOTONIC
1882
1883 // Time-of-day clock is all we can reliably use.
1884 struct timeval now;
1885 int status = gettimeofday(&now, NULL);
1886 assert_status(status == 0, errno, "gettimeofday");
1887 if (isAbsolute) {
1888 unpack_abs_time(abstime, timeout, now.tv_sec);
1889 } else {
1890 calc_rel_time(abstime, timeout, now.tv_sec, now.tv_usec, MICROUNITS);
1891 }
1892 DEBUG_ONLY(max_secs += now.tv_sec;)
1893 }
1894
1895 assert(abstime->tv_sec >= 0, "tv_sec < 0");
1896 assert(abstime->tv_sec <= max_secs, "tv_sec > max_secs");
1897 assert(abstime->tv_nsec >= 0, "tv_nsec < 0");
1898 assert(abstime->tv_nsec < NANOUNITS, "tv_nsec >= NANOUNITS");
1899 }
1900
1901 // Create an absolute time 'millis' milliseconds in the future, using the
1902 // real-time (time-of-day) clock. Used by PosixSemaphore.
1903 void os::Posix::to_RTC_abstime(timespec* abstime, int64_t millis) {
1904 to_abstime(abstime, millis_to_nanos(millis),
1905 false /* not absolute */,
1906 true /* use real-time clock */);
1907 }
1908
1909 // Shared pthread_mutex/cond based PlatformEvent implementation.
1910 // Not currently usable by Solaris.
1911
1912 #ifndef SOLARIS
1913
1914 // PlatformEvent
1915 //
1916 // Assumption:
1917 // Only one parker can exist on an event, which is why we allocate
1918 // them per-thread. Multiple unparkers can coexist.
1919 //
1920 // _event serves as a restricted-range semaphore.
1921 // -1 : thread is blocked, i.e. there is a waiter
1922 // 0 : neutral: thread is running or ready,
1923 // could have been signaled after a wait started
1924 // 1 : signaled - thread is running or ready
1925 //
1926 // Having three states allows for some detection of bad usage - see
1927 // comments on unpark().
1928
1929 os::PlatformEvent::PlatformEvent() {
1930 int status = pthread_cond_init(_cond, _condAttr);
1931 assert_status(status == 0, status, "cond_init");
1932 status = pthread_mutex_init(_mutex, _mutexAttr);
1933 assert_status(status == 0, status, "mutex_init");
1934 _event = 0;
1935 _nParked = 0;
1936 }
1937
1938 void os::PlatformEvent::park() { // AKA "down()"
1939 // Transitions for _event:
1940 // -1 => -1 : illegal
1941 // 1 => 0 : pass - return immediately
1942 // 0 => -1 : block; then set _event to 0 before returning
1943
1944 // Invariant: Only the thread associated with the PlatformEvent
1945 // may call park().
1946 assert(_nParked == 0, "invariant");
1947
1948 int v;
1949
1950 // atomically decrement _event
1951 for (;;) {
1952 v = _event;
1953 if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1954 }
1955 guarantee(v >= 0, "invariant");
1956
1957 if (v == 0) { // Do this the hard way by blocking ...
1958 int status = pthread_mutex_lock(_mutex);
1959 assert_status(status == 0, status, "mutex_lock");
1960 guarantee(_nParked == 0, "invariant");
1961 ++_nParked;
1962 while (_event < 0) {
1963 // OS-level "spurious wakeups" are ignored
1964 status = pthread_cond_wait(_cond, _mutex);
1965 assert_status(status == 0, status, "cond_wait");
1966 }
1967 --_nParked;
1968
1969 _event = 0;
1970 status = pthread_mutex_unlock(_mutex);
1971 assert_status(status == 0, status, "mutex_unlock");
1972 // Paranoia to ensure our locked and lock-free paths interact
1973 // correctly with each other.
1974 OrderAccess::fence();
1975 }
1976 guarantee(_event >= 0, "invariant");
1977 }
1978
1979 int os::PlatformEvent::park(jlong millis) {
1980 // Transitions for _event:
1981 // -1 => -1 : illegal
1982 // 1 => 0 : pass - return immediately
1983 // 0 => -1 : block; then set _event to 0 before returning
1984
1985 // Invariant: Only the thread associated with the Event/PlatformEvent
1986 // may call park().
1987 assert(_nParked == 0, "invariant");
1988
1989 int v;
1990 // atomically decrement _event
1991 for (;;) {
1992 v = _event;
1993 if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1994 }
1995 guarantee(v >= 0, "invariant");
1996
1997 if (v == 0) { // Do this the hard way by blocking ...
1998 struct timespec abst;
1999 to_abstime(&abst, millis_to_nanos(millis), false, false);
2000
2001 int ret = OS_TIMEOUT;
2002 int status = pthread_mutex_lock(_mutex);
2003 assert_status(status == 0, status, "mutex_lock");
2004 guarantee(_nParked == 0, "invariant");
2005 ++_nParked;
2006
2007 while (_event < 0) {
2008 status = pthread_cond_timedwait(_cond, _mutex, &abst);
2009 assert_status(status == 0 || status == ETIMEDOUT,
2010 status, "cond_timedwait");
2011 // OS-level "spurious wakeups" are ignored unless the archaic
2012 // FilterSpuriousWakeups is set false. That flag should be obsoleted.
2013 if (!FilterSpuriousWakeups) break;
2014 if (status == ETIMEDOUT) break;
2015 }
2016 --_nParked;
2017
2018 if (_event >= 0) {
2019 ret = OS_OK;
2020 }
2021
2022 _event = 0;
2023 status = pthread_mutex_unlock(_mutex);
2024 assert_status(status == 0, status, "mutex_unlock");
2025 // Paranoia to ensure our locked and lock-free paths interact
2026 // correctly with each other.
2027 OrderAccess::fence();
2028 return ret;
2029 }
2030 return OS_OK;
2031 }
2032
2033 void os::PlatformEvent::unpark() {
2034 // Transitions for _event:
2035 // 0 => 1 : just return
2036 // 1 => 1 : just return
2037 // -1 => either 0 or 1; must signal target thread
2038 // That is, we can safely transition _event from -1 to either
2039 // 0 or 1.
2040 // See also: "Semaphores in Plan 9" by Mullender & Cox
2041 //
2042 // Note: Forcing a transition from "-1" to "1" on an unpark() means
2043 // that it will take two back-to-back park() calls for the owning
2044 // thread to block. This has the benefit of forcing a spurious return
2045 // from the first park() call after an unpark() call which will help
2046 // shake out uses of park() and unpark() without checking state conditions
2047 // properly. This spurious return doesn't manifest itself in any user code
2048 // but only in the correctly written condition checking loops of ObjectMonitor,
2049 // Mutex/Monitor, Thread::muxAcquire and os::sleep
2050
2051 if (Atomic::xchg(1, &_event) >= 0) return;
2052
2053 int status = pthread_mutex_lock(_mutex);
2054 assert_status(status == 0, status, "mutex_lock");
2055 int anyWaiters = _nParked;
2056 assert(anyWaiters == 0 || anyWaiters == 1, "invariant");
2057 status = pthread_mutex_unlock(_mutex);
2058 assert_status(status == 0, status, "mutex_unlock");
2059
2060 // Note that we signal() *after* dropping the lock for "immortal" Events.
2061 // This is safe and avoids a common class of futile wakeups. In rare
2062 // circumstances this can cause a thread to return prematurely from
2063 // cond_{timed}wait() but the spurious wakeup is benign and the victim
2064 // will simply re-test the condition and re-park itself.
2065 // This provides particular benefit if the underlying platform does not
2066 // provide wait morphing.
2067
2068 if (anyWaiters != 0) {
2069 status = pthread_cond_signal(_cond);
2070 assert_status(status == 0, status, "cond_signal");
2071 }
2072 }
2073
2074 // JSR166 support
2075
2076 os::PlatformParker::PlatformParker() {
2077 int status;
2078 status = pthread_cond_init(&_cond[REL_INDEX], _condAttr);
2079 assert_status(status == 0, status, "cond_init rel");
2080 status = pthread_cond_init(&_cond[ABS_INDEX], NULL);
2081 assert_status(status == 0, status, "cond_init abs");
2082 status = pthread_mutex_init(_mutex, _mutexAttr);
2083 assert_status(status == 0, status, "mutex_init");
2084 _cur_index = -1; // mark as unused
2085 }
2086
2087 // Parker::park decrements count if > 0, else does a condvar wait. Unpark
2088 // sets count to 1 and signals condvar. Only one thread ever waits
2089 // on the condvar. Contention seen when trying to park implies that someone
2090 // is unparking you, so don't wait. And spurious returns are fine, so there
2091 // is no need to track notifications.
2092
2093 void Parker::park(bool isAbsolute, jlong time) {
2094
2095 // Optional fast-path check:
2096 // Return immediately if a permit is available.
2097 // We depend on Atomic::xchg() having full barrier semantics
2098 // since we are doing a lock-free update to _counter.
2099 if (Atomic::xchg(0, &_counter) > 0) return;
2100
2101 Thread* thread = Thread::current();
2102 assert(thread->is_Java_thread(), "Must be JavaThread");
2103 JavaThread *jt = (JavaThread *)thread;
2104
2105 // Optional optimization -- avoid state transitions if there's
2106 // an interrupt pending.
2107 if (Thread::is_interrupted(thread, false)) {
2108 return;
2109 }
2110
2111 // Next, demultiplex/decode time arguments
2112 struct timespec absTime;
2113 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
2114 return;
2115 }
2116 if (time > 0) {
2117 to_abstime(&absTime, time, isAbsolute, false);
2118 }
2119
2120 // Enter safepoint region
2121 // Beware of deadlocks such as 6317397.
2122 // The per-thread Parker:: mutex is a classic leaf-lock.
2123 // In particular a thread must never block on the Threads_lock while
2124 // holding the Parker:: mutex. If safepoints are pending both the
2125 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
2126 ThreadBlockInVM tbivm(jt);
2127
2128 // Don't wait if cannot get lock since interference arises from
2129 // unparking. Also re-check interrupt before trying wait.
2130 if (Thread::is_interrupted(thread, false) ||
2131 pthread_mutex_trylock(_mutex) != 0) {
2132 return;
2133 }
2134
2135 int status;
2136 if (_counter > 0) { // no wait needed
2137 _counter = 0;
2138 status = pthread_mutex_unlock(_mutex);
2139 assert_status(status == 0, status, "invariant");
2140 // Paranoia to ensure our locked and lock-free paths interact
2141 // correctly with each other and Java-level accesses.
2142 OrderAccess::fence();
2143 return;
2144 }
2145
2146 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2147 jt->set_suspend_equivalent();
2148 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2149
2150 assert(_cur_index == -1, "invariant");
2151 if (time == 0) {
2152 _cur_index = REL_INDEX; // arbitrary choice when not timed
2153 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
2154 assert_status(status == 0, status, "cond_timedwait");
2155 }
2156 else {
2157 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
2158 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
2159 assert_status(status == 0 || status == ETIMEDOUT,
2160 status, "cond_timedwait");
2161 }
2162 _cur_index = -1;
2163
2164 _counter = 0;
2165 status = pthread_mutex_unlock(_mutex);
2166 assert_status(status == 0, status, "invariant");
2167 // Paranoia to ensure our locked and lock-free paths interact
2168 // correctly with each other and Java-level accesses.
2169 OrderAccess::fence();
2170
2171 // If externally suspended while waiting, re-suspend
2172 if (jt->handle_special_suspend_equivalent_condition()) {
2173 jt->java_suspend_self();
2174 }
2175 }
2176
2177 void Parker::unpark() {
2178 int status = pthread_mutex_lock(_mutex);
2179 assert_status(status == 0, status, "invariant");
2180 const int s = _counter;
2181 _counter = 1;
2182 // must capture correct index before unlocking
2183 int index = _cur_index;
2184 status = pthread_mutex_unlock(_mutex);
2185 assert_status(status == 0, status, "invariant");
2186
2187 // Note that we signal() *after* dropping the lock for "immortal" Events.
2188 // This is safe and avoids a common class of futile wakeups. In rare
2189 // circumstances this can cause a thread to return prematurely from
2190 // cond_{timed}wait() but the spurious wakeup is benign and the victim
2191 // will simply re-test the condition and re-park itself.
2192 // This provides particular benefit if the underlying platform does not
2193 // provide wait morphing.
2194
2195 if (s < 1 && index != -1) {
2196 // thread is definitely parked
2197 status = pthread_cond_signal(&_cond[index]);
2198 assert_status(status == 0, status, "invariant");
2199 }
2200 }
2201
2202 // Platform Monitor implementation
2203
2204 os::PlatformMonitor::PlatformMonitor() {
2205 int status = pthread_cond_init(&_cond, _condAttr);
2206 assert_status(status == 0, status, "cond_init");
2207 status = pthread_mutex_init(&_mutex, _mutexAttr);
2208 assert_status(status == 0, status, "mutex_init");
2209 }
2210
2211 os::PlatformMonitor::~PlatformMonitor() {
2212 int status = pthread_cond_destroy(&_cond);
2213 assert_status(status == 0, status, "cond_destroy");
2214 status = pthread_mutex_destroy(&_mutex);
2215 assert_status(status == 0, status, "mutex_destroy");
2216 }
2217
2218 // Must already be locked
2219 int os::PlatformMonitor::wait(jlong millis) {
2220 assert(millis >= 0, "negative timeout");
2221 if (millis > 0) {
2222 struct timespec abst;
2223 // We have to watch for overflow when converting millis to nanos,
2224 // but if millis is that large then we will end up limiting to
2225 // MAX_SECS anyway, so just do that here.
2226 if (millis / MILLIUNITS > MAX_SECS) {
2227 millis = jlong(MAX_SECS) * MILLIUNITS;
2228 }
2229 to_abstime(&abst, millis * (NANOUNITS / MILLIUNITS), false, false);
2230
2231 int ret = OS_TIMEOUT;
2232 int status = pthread_cond_timedwait(&_cond, &_mutex, &abst);
2233 assert_status(status == 0 || status == ETIMEDOUT,
2234 status, "cond_timedwait");
2235 if (status == 0) {
2236 ret = OS_OK;
2237 }
2238 return ret;
2239 } else {
2240 int status = pthread_cond_wait(&_cond, &_mutex);
2241 assert_status(status == 0, status, "cond_wait");
2242 return OS_OK;
2243 }
2244 }
2245
2246 #endif // !SOLARIS