1 /*
2 * Copyright (c) 1999, 2012, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "jvm_bsd.h"
34 #include "memory/allocation.inline.hpp"
35 #include "memory/filemap.hpp"
36 #include "mutex_bsd.inline.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "os_share_bsd.hpp"
39 #include "prims/jniFastGetField.hpp"
40 #include "prims/jvm.h"
41 #include "prims/jvm_misc.hpp"
42 #include "runtime/arguments.hpp"
43 #include "runtime/extendedPC.hpp"
44 #include "runtime/globals.hpp"
45 #include "runtime/interfaceSupport.hpp"
46 #include "runtime/java.hpp"
47 #include "runtime/javaCalls.hpp"
48 #include "runtime/mutexLocker.hpp"
49 #include "runtime/objectMonitor.hpp"
50 #include "runtime/osThread.hpp"
51 #include "runtime/perfMemory.hpp"
52 #include "runtime/sharedRuntime.hpp"
53 #include "runtime/statSampler.hpp"
54 #include "runtime/stubRoutines.hpp"
55 #include "runtime/threadCritical.hpp"
56 #include "runtime/timer.hpp"
57 #include "services/attachListener.hpp"
58 #include "services/runtimeService.hpp"
59 #include "thread_bsd.inline.hpp"
60 #include "utilities/decoder.hpp"
61 #include "utilities/defaultStream.hpp"
62 #include "utilities/events.hpp"
63 #include "utilities/growableArray.hpp"
64 #include "utilities/vmError.hpp"
65 #ifdef TARGET_ARCH_x86
66 # include "assembler_x86.inline.hpp"
67 # include "nativeInst_x86.hpp"
68 #endif
69 #ifdef TARGET_ARCH_sparc
70 # include "assembler_sparc.inline.hpp"
71 # include "nativeInst_sparc.hpp"
72 #endif
73 #ifdef TARGET_ARCH_zero
74 # include "assembler_zero.inline.hpp"
75 # include "nativeInst_zero.hpp"
76 #endif
77 #ifdef TARGET_ARCH_arm
78 # include "assembler_arm.inline.hpp"
79 # include "nativeInst_arm.hpp"
80 #endif
81 #ifdef TARGET_ARCH_ppc
82 # include "assembler_ppc.inline.hpp"
83 # include "nativeInst_ppc.hpp"
84 #endif
85 #ifdef COMPILER1
86 #include "c1/c1_Runtime1.hpp"
87 #endif
88 #ifdef COMPILER2
89 #include "opto/runtime.hpp"
90 #endif
91
92 // put OS-includes here
93 # include <sys/types.h>
94 # include <sys/mman.h>
95 # include <sys/stat.h>
96 # include <sys/select.h>
97 # include <pthread.h>
98 # include <signal.h>
99 # include <errno.h>
100 # include <dlfcn.h>
101 # include <stdio.h>
102 # include <unistd.h>
103 # include <sys/resource.h>
104 # include <pthread.h>
105 # include <sys/stat.h>
106 # include <sys/time.h>
107 # include <sys/times.h>
108 # include <sys/utsname.h>
109 # include <sys/socket.h>
110 # include <sys/wait.h>
111 # include <time.h>
112 # include <pwd.h>
113 # include <poll.h>
114 # include <semaphore.h>
115 # include <fcntl.h>
116 # include <string.h>
117 #ifdef _ALLBSD_SOURCE
118 # include <sys/param.h>
119 # include <sys/sysctl.h>
120 #else
121 # include <syscall.h>
122 # include <sys/sysinfo.h>
123 # include <gnu/libc-version.h>
124 #endif
125 # include <sys/ipc.h>
126 # include <sys/shm.h>
127 #ifndef __APPLE__
128 # include <link.h>
129 #endif
130 # include <stdint.h>
131 # include <inttypes.h>
132 # include <sys/ioctl.h>
133
134 #if defined(__FreeBSD__) || defined(__NetBSD__)
135 # include <elf.h>
136 #endif
137
138 #ifdef __APPLE__
139 # include <mach/mach.h> // semaphore_* API
140 # include <mach-o/dyld.h>
141 # include <sys/proc_info.h>
142 # include <objc/objc-auto.h>
143 #endif
144
145 #ifndef MAP_ANONYMOUS
146 #define MAP_ANONYMOUS MAP_ANON
147 #endif
148
149 #define MAX_PATH (2 * K)
150
151 // for timer info max values which include all bits
152 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
153
154 #define LARGEPAGES_BIT (1 << 6)
155 ////////////////////////////////////////////////////////////////////////////////
156 // global variables
157 julong os::Bsd::_physical_memory = 0;
158
159 #ifndef _ALLBSD_SOURCE
160 address os::Bsd::_initial_thread_stack_bottom = NULL;
161 uintptr_t os::Bsd::_initial_thread_stack_size = 0;
162 #endif
163
164 int (*os::Bsd::_clock_gettime)(clockid_t, struct timespec *) = NULL;
165 #ifndef _ALLBSD_SOURCE
166 int (*os::Bsd::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
167 Mutex* os::Bsd::_createThread_lock = NULL;
168 #endif
169 pthread_t os::Bsd::_main_thread;
170 int os::Bsd::_page_size = -1;
171 #ifndef _ALLBSD_SOURCE
172 bool os::Bsd::_is_floating_stack = false;
173 bool os::Bsd::_is_NPTL = false;
174 bool os::Bsd::_supports_fast_thread_cpu_time = false;
175 const char * os::Bsd::_glibc_version = NULL;
176 const char * os::Bsd::_libpthread_version = NULL;
177 #endif
178
179 static jlong initial_time_count=0;
180
181 static int clock_tics_per_sec = 100;
182
183 // For diagnostics to print a message once. see run_periodic_checks
184 static sigset_t check_signal_done;
185 static bool check_signals = true;;
186
187 static pid_t _initial_pid = 0;
188
189 /* Signal number used to suspend/resume a thread */
190
191 /* do not use any signal number less than SIGSEGV, see 4355769 */
192 static int SR_signum = SIGUSR2;
193 sigset_t SR_sigset;
194
195
196 ////////////////////////////////////////////////////////////////////////////////
197 // utility functions
198
199 static int SR_initialize();
200 static int SR_finalize();
201
202 julong os::available_memory() {
203 return Bsd::available_memory();
204 }
205
206 julong os::Bsd::available_memory() {
207 #ifdef _ALLBSD_SOURCE
208 // XXXBSD: this is just a stopgap implementation
209 return physical_memory() >> 2;
210 #else
211 // values in struct sysinfo are "unsigned long"
212 struct sysinfo si;
213 sysinfo(&si);
214
215 return (julong)si.freeram * si.mem_unit;
216 #endif
217 }
218
219 julong os::physical_memory() {
220 return Bsd::physical_memory();
221 }
222
223 julong os::allocatable_physical_memory(julong size) {
224 #ifdef _LP64
225 return size;
226 #else
227 julong result = MIN2(size, (julong)3800*M);
228 if (!is_allocatable(result)) {
229 // See comments under solaris for alignment considerations
230 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
231 result = MIN2(size, reasonable_size);
232 }
233 return result;
234 #endif // _LP64
235 }
236
237 ////////////////////////////////////////////////////////////////////////////////
238 // environment support
239
240 bool os::getenv(const char* name, char* buf, int len) {
241 const char* val = ::getenv(name);
242 if (val != NULL && strlen(val) < (size_t)len) {
243 strcpy(buf, val);
244 return true;
245 }
246 if (len > 0) buf[0] = 0; // return a null string
247 return false;
248 }
249
250
251 // Return true if user is running as root.
252
253 bool os::have_special_privileges() {
254 static bool init = false;
255 static bool privileges = false;
256 if (!init) {
257 privileges = (getuid() != geteuid()) || (getgid() != getegid());
258 init = true;
259 }
260 return privileges;
261 }
262
263
264 #ifndef _ALLBSD_SOURCE
265 #ifndef SYS_gettid
266 // i386: 224, ia64: 1105, amd64: 186, sparc 143
267 #ifdef __ia64__
268 #define SYS_gettid 1105
269 #elif __i386__
270 #define SYS_gettid 224
271 #elif __amd64__
272 #define SYS_gettid 186
273 #elif __sparc__
274 #define SYS_gettid 143
275 #else
276 #error define gettid for the arch
277 #endif
278 #endif
279 #endif
280
281 // Cpu architecture string
282 #if defined(ZERO)
283 static char cpu_arch[] = ZERO_LIBARCH;
284 #elif defined(IA64)
285 static char cpu_arch[] = "ia64";
286 #elif defined(IA32)
287 static char cpu_arch[] = "i386";
288 #elif defined(AMD64)
289 static char cpu_arch[] = "amd64";
290 #elif defined(ARM)
291 static char cpu_arch[] = "arm";
292 #elif defined(PPC)
293 static char cpu_arch[] = "ppc";
294 #elif defined(SPARC)
295 # ifdef _LP64
296 static char cpu_arch[] = "sparcv9";
297 # else
298 static char cpu_arch[] = "sparc";
299 # endif
300 #else
301 #error Add appropriate cpu_arch setting
302 #endif
303
304 // Compiler variant
305 #ifdef COMPILER2
306 #define COMPILER_VARIANT "server"
307 #else
308 #define COMPILER_VARIANT "client"
309 #endif
310
311 #ifndef _ALLBSD_SOURCE
312 // pid_t gettid()
313 //
314 // Returns the kernel thread id of the currently running thread. Kernel
315 // thread id is used to access /proc.
316 //
317 // (Note that getpid() on BsdThreads returns kernel thread id too; but
318 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
319 //
320 pid_t os::Bsd::gettid() {
321 int rslt = syscall(SYS_gettid);
322 if (rslt == -1) {
323 // old kernel, no NPTL support
324 return getpid();
325 } else {
326 return (pid_t)rslt;
327 }
328 }
329
330 // Most versions of bsd have a bug where the number of processors are
331 // determined by looking at the /proc file system. In a chroot environment,
332 // the system call returns 1. This causes the VM to act as if it is
333 // a single processor and elide locking (see is_MP() call).
334 static bool unsafe_chroot_detected = false;
335 static const char *unstable_chroot_error = "/proc file system not found.\n"
336 "Java may be unstable running multithreaded in a chroot "
337 "environment on Bsd when /proc filesystem is not mounted.";
338 #endif
339
340 #ifdef _ALLBSD_SOURCE
341 void os::Bsd::initialize_system_info() {
342 int mib[2];
343 size_t len;
344 int cpu_val;
345 u_long mem_val;
346
347 /* get processors count via hw.ncpus sysctl */
348 mib[0] = CTL_HW;
349 mib[1] = HW_NCPU;
350 len = sizeof(cpu_val);
351 if (sysctl(mib, 2, &cpu_val, &len, NULL, 0) != -1 && cpu_val >= 1) {
352 set_processor_count(cpu_val);
353 }
354 else {
355 set_processor_count(1); // fallback
356 }
357
358 /* get physical memory via hw.usermem sysctl (hw.usermem is used
359 * instead of hw.physmem because we need size of allocatable memory
360 */
361 mib[0] = CTL_HW;
362 mib[1] = HW_USERMEM;
363 len = sizeof(mem_val);
364 if (sysctl(mib, 2, &mem_val, &len, NULL, 0) != -1)
365 _physical_memory = mem_val;
366 else
367 _physical_memory = 256*1024*1024; // fallback (XXXBSD?)
368
369 #ifdef __OpenBSD__
370 {
371 // limit _physical_memory memory view on OpenBSD since
372 // datasize rlimit restricts us anyway.
373 struct rlimit limits;
374 getrlimit(RLIMIT_DATA, &limits);
375 _physical_memory = MIN2(_physical_memory, (julong)limits.rlim_cur);
376 }
377 #endif
378 }
379 #else
380 void os::Bsd::initialize_system_info() {
381 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
382 if (processor_count() == 1) {
383 pid_t pid = os::Bsd::gettid();
384 char fname[32];
385 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
386 FILE *fp = fopen(fname, "r");
387 if (fp == NULL) {
388 unsafe_chroot_detected = true;
389 } else {
390 fclose(fp);
391 }
392 }
393 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
394 assert(processor_count() > 0, "bsd error");
395 }
396 #endif
397
398 #ifdef __APPLE__
399 static const char *get_home() {
400 const char *home_dir = ::getenv("HOME");
401 if ((home_dir == NULL) || (*home_dir == '\0')) {
402 struct passwd *passwd_info = getpwuid(geteuid());
403 if (passwd_info != NULL) {
404 home_dir = passwd_info->pw_dir;
405 }
406 }
407
408 return home_dir;
409 }
410 #endif
411
412 void os::init_system_properties_values() {
413 // char arch[12];
414 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
415
416 // The next steps are taken in the product version:
417 //
418 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
419 // This library should be located at:
420 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
421 //
422 // If "/jre/lib/" appears at the right place in the path, then we
423 // assume libjvm[_g].so is installed in a JDK and we use this path.
424 //
425 // Otherwise exit with message: "Could not create the Java virtual machine."
426 //
427 // The following extra steps are taken in the debugging version:
428 //
429 // If "/jre/lib/" does NOT appear at the right place in the path
430 // instead of exit check for $JAVA_HOME environment variable.
431 //
432 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
433 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
434 // it looks like libjvm[_g].so is installed there
435 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
436 //
437 // Otherwise exit.
438 //
439 // Important note: if the location of libjvm.so changes this
440 // code needs to be changed accordingly.
441
442 // The next few definitions allow the code to be verbatim:
443 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
444 #define getenv(n) ::getenv(n)
445
446 /*
447 * See ld(1):
448 * The linker uses the following search paths to locate required
449 * shared libraries:
450 * 1: ...
451 * ...
452 * 7: The default directories, normally /lib and /usr/lib.
453 */
454 #ifndef DEFAULT_LIBPATH
455 #define DEFAULT_LIBPATH "/lib:/usr/lib"
456 #endif
457
458 #define EXTENSIONS_DIR "/lib/ext"
459 #define ENDORSED_DIR "/lib/endorsed"
460 #define REG_DIR "/usr/java/packages"
461
462 #ifdef __APPLE__
463 #define SYS_EXTENSIONS_DIR "/Library/Java/Extensions"
464 #define SYS_EXTENSIONS_DIRS SYS_EXTENSIONS_DIR ":/Network" SYS_EXTENSIONS_DIR ":/System" SYS_EXTENSIONS_DIR ":/usr/lib/java"
465 const char *user_home_dir = get_home();
466 // the null in SYS_EXTENSIONS_DIRS counts for the size of the colon after user_home_dir
467 int system_ext_size = strlen(user_home_dir) + sizeof(SYS_EXTENSIONS_DIR) +
468 sizeof(SYS_EXTENSIONS_DIRS);
469 #endif
470
471 {
472 /* sysclasspath, java_home, dll_dir */
473 {
474 char *home_path;
475 char *dll_path;
476 char *pslash;
477 char buf[MAXPATHLEN];
478 os::jvm_path(buf, sizeof(buf));
479
480 // Found the full path to libjvm.so.
481 // Now cut the path to <java_home>/jre if we can.
482 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
483 pslash = strrchr(buf, '/');
484 if (pslash != NULL)
485 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
486 dll_path = malloc(strlen(buf) + 1);
487 if (dll_path == NULL)
488 return;
489 strcpy(dll_path, buf);
490 Arguments::set_dll_dir(dll_path);
491
492 if (pslash != NULL) {
493 pslash = strrchr(buf, '/');
494 if (pslash != NULL) {
495 *pslash = '\0'; /* get rid of /<arch> (/lib on macosx) */
496 #ifndef __APPLE__
497 pslash = strrchr(buf, '/');
498 if (pslash != NULL)
499 *pslash = '\0'; /* get rid of /lib */
500 #endif
501 }
502 }
503
504 home_path = malloc(strlen(buf) + 1);
505 if (home_path == NULL)
506 return;
507 strcpy(home_path, buf);
508 Arguments::set_java_home(home_path);
509
510 if (!set_boot_path('/', ':'))
511 return;
512 }
513
514 /*
515 * Where to look for native libraries
516 *
517 * Note: Due to a legacy implementation, most of the library path
518 * is set in the launcher. This was to accomodate linking restrictions
519 * on legacy Bsd implementations (which are no longer supported).
520 * Eventually, all the library path setting will be done here.
521 *
522 * However, to prevent the proliferation of improperly built native
523 * libraries, the new path component /usr/java/packages is added here.
524 * Eventually, all the library path setting will be done here.
525 */
526 {
527 char *ld_library_path;
528
529 /*
530 * Construct the invariant part of ld_library_path. Note that the
531 * space for the colon and the trailing null are provided by the
532 * nulls included by the sizeof operator (so actually we allocate
533 * a byte more than necessary).
534 */
535 #ifdef __APPLE__
536 ld_library_path = (char *) malloc(system_ext_size);
537 sprintf(ld_library_path, "%s" SYS_EXTENSIONS_DIR ":" SYS_EXTENSIONS_DIRS, user_home_dir);
538 #else
539 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
540 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
541 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
542 #endif
543
544 /*
545 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
546 * should always exist (until the legacy problem cited above is
547 * addressed).
548 */
549 #ifdef __APPLE__
550 // Prepend the default path with the JAVA_LIBRARY_PATH so that the app launcher code can specify a directory inside an app wrapper
551 char *l = getenv("JAVA_LIBRARY_PATH");
552 if (l != NULL) {
553 char *t = ld_library_path;
554 /* That's +1 for the colon and +1 for the trailing '\0' */
555 ld_library_path = (char *) malloc(strlen(l) + 1 + strlen(t) + 1);
556 sprintf(ld_library_path, "%s:%s", l, t);
557 free(t);
558 }
559
560 char *v = getenv("DYLD_LIBRARY_PATH");
561 #else
562 char *v = getenv("LD_LIBRARY_PATH");
563 #endif
564 if (v != NULL) {
565 char *t = ld_library_path;
566 /* That's +1 for the colon and +1 for the trailing '\0' */
567 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
568 sprintf(ld_library_path, "%s:%s", v, t);
569 free(t);
570 }
571
572 #ifdef __APPLE__
573 // Apple's Java6 has "." at the beginning of java.library.path.
574 // OpenJDK on Windows has "." at the end of java.library.path.
575 // OpenJDK on Linux and Solaris don't have "." in java.library.path
576 // at all. To ease the transition from Apple's Java6 to OpenJDK7,
577 // "." is appended to the end of java.library.path. Yes, this
578 // could cause a change in behavior, but Apple's Java6 behavior
579 // can be achieved by putting "." at the beginning of the
580 // JAVA_LIBRARY_PATH environment variable.
581 {
582 char *t = ld_library_path;
583 // that's +3 for appending ":." and the trailing '\0'
584 ld_library_path = (char *) malloc(strlen(t) + 3);
585 sprintf(ld_library_path, "%s:%s", t, ".");
586 free(t);
587 }
588 #endif
589
590 Arguments::set_library_path(ld_library_path);
591 }
592
593 /*
594 * Extensions directories.
595 *
596 * Note that the space for the colon and the trailing null are provided
597 * by the nulls included by the sizeof operator (so actually one byte more
598 * than necessary is allocated).
599 */
600 {
601 #ifdef __APPLE__
602 char *buf = malloc(strlen(Arguments::get_java_home()) +
603 sizeof(EXTENSIONS_DIR) + system_ext_size);
604 sprintf(buf, "%s" SYS_EXTENSIONS_DIR ":%s" EXTENSIONS_DIR ":"
605 SYS_EXTENSIONS_DIRS, user_home_dir, Arguments::get_java_home());
606 #else
607 char *buf = malloc(strlen(Arguments::get_java_home()) +
608 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
609 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
610 Arguments::get_java_home());
611 #endif
612
613 Arguments::set_ext_dirs(buf);
614 }
615
616 /* Endorsed standards default directory. */
617 {
618 char * buf;
619 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
620 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
621 Arguments::set_endorsed_dirs(buf);
622 }
623 }
624
625 #ifdef __APPLE__
626 #undef SYS_EXTENSIONS_DIR
627 #endif
628 #undef malloc
629 #undef getenv
630 #undef EXTENSIONS_DIR
631 #undef ENDORSED_DIR
632
633 // Done
634 return;
635 }
636
637 ////////////////////////////////////////////////////////////////////////////////
638 // breakpoint support
639
640 void os::breakpoint() {
641 BREAKPOINT;
642 }
643
644 extern "C" void breakpoint() {
645 // use debugger to set breakpoint here
646 }
647
648 ////////////////////////////////////////////////////////////////////////////////
649 // signal support
650
651 debug_only(static bool signal_sets_initialized = false);
652 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
653
654 bool os::Bsd::is_sig_ignored(int sig) {
655 struct sigaction oact;
656 sigaction(sig, (struct sigaction*)NULL, &oact);
657 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
658 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
659 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
660 return true;
661 else
662 return false;
663 }
664
665 void os::Bsd::signal_sets_init() {
666 // Should also have an assertion stating we are still single-threaded.
667 assert(!signal_sets_initialized, "Already initialized");
668 // Fill in signals that are necessarily unblocked for all threads in
669 // the VM. Currently, we unblock the following signals:
670 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
671 // by -Xrs (=ReduceSignalUsage));
672 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
673 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
674 // the dispositions or masks wrt these signals.
675 // Programs embedding the VM that want to use the above signals for their
676 // own purposes must, at this time, use the "-Xrs" option to prevent
677 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
678 // (See bug 4345157, and other related bugs).
679 // In reality, though, unblocking these signals is really a nop, since
680 // these signals are not blocked by default.
681 sigemptyset(&unblocked_sigs);
682 sigemptyset(&allowdebug_blocked_sigs);
683 sigaddset(&unblocked_sigs, SIGILL);
684 sigaddset(&unblocked_sigs, SIGSEGV);
685 sigaddset(&unblocked_sigs, SIGBUS);
686 sigaddset(&unblocked_sigs, SIGFPE);
687 sigaddset(&unblocked_sigs, SR_signum);
688
689 if (!ReduceSignalUsage) {
690 if (!os::Bsd::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
691 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
692 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
693 }
694 if (!os::Bsd::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
695 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
696 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
697 }
698 if (!os::Bsd::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
699 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
700 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
701 }
702 }
703 // Fill in signals that are blocked by all but the VM thread.
704 sigemptyset(&vm_sigs);
705 if (!ReduceSignalUsage)
706 sigaddset(&vm_sigs, BREAK_SIGNAL);
707 debug_only(signal_sets_initialized = true);
708
709 }
710
711 // These are signals that are unblocked while a thread is running Java.
712 // (For some reason, they get blocked by default.)
713 sigset_t* os::Bsd::unblocked_signals() {
714 assert(signal_sets_initialized, "Not initialized");
715 return &unblocked_sigs;
716 }
717
718 // These are the signals that are blocked while a (non-VM) thread is
719 // running Java. Only the VM thread handles these signals.
720 sigset_t* os::Bsd::vm_signals() {
721 assert(signal_sets_initialized, "Not initialized");
722 return &vm_sigs;
723 }
724
725 // These are signals that are blocked during cond_wait to allow debugger in
726 sigset_t* os::Bsd::allowdebug_blocked_signals() {
727 assert(signal_sets_initialized, "Not initialized");
728 return &allowdebug_blocked_sigs;
729 }
730
731 void os::Bsd::hotspot_sigmask(Thread* thread) {
732
733 //Save caller's signal mask before setting VM signal mask
734 sigset_t caller_sigmask;
735 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
736
737 OSThread* osthread = thread->osthread();
738 osthread->set_caller_sigmask(caller_sigmask);
739
740 pthread_sigmask(SIG_UNBLOCK, os::Bsd::unblocked_signals(), NULL);
741
742 if (!ReduceSignalUsage) {
743 if (thread->is_VM_thread()) {
744 // Only the VM thread handles BREAK_SIGNAL ...
745 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
746 } else {
747 // ... all other threads block BREAK_SIGNAL
748 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
749 }
750 }
751 }
752
753 #ifndef _ALLBSD_SOURCE
754 //////////////////////////////////////////////////////////////////////////////
755 // detecting pthread library
756
757 void os::Bsd::libpthread_init() {
758 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
759 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
760 // generic name for earlier versions.
761 // Define macros here so we can build HotSpot on old systems.
762 # ifndef _CS_GNU_LIBC_VERSION
763 # define _CS_GNU_LIBC_VERSION 2
764 # endif
765 # ifndef _CS_GNU_LIBPTHREAD_VERSION
766 # define _CS_GNU_LIBPTHREAD_VERSION 3
767 # endif
768
769 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
770 if (n > 0) {
771 char *str = (char *)malloc(n);
772 confstr(_CS_GNU_LIBC_VERSION, str, n);
773 os::Bsd::set_glibc_version(str);
774 } else {
775 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
776 static char _gnu_libc_version[32];
777 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
778 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
779 os::Bsd::set_glibc_version(_gnu_libc_version);
780 }
781
782 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
783 if (n > 0) {
784 char *str = (char *)malloc(n);
785 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
786 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
787 // us "NPTL-0.29" even we are running with BsdThreads. Check if this
788 // is the case. BsdThreads has a hard limit on max number of threads.
789 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
790 // On the other hand, NPTL does not have such a limit, sysconf()
791 // will return -1 and errno is not changed. Check if it is really NPTL.
792 if (strcmp(os::Bsd::glibc_version(), "glibc 2.3.2") == 0 &&
793 strstr(str, "NPTL") &&
794 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
795 free(str);
796 os::Bsd::set_libpthread_version("bsdthreads");
797 } else {
798 os::Bsd::set_libpthread_version(str);
799 }
800 } else {
801 // glibc before 2.3.2 only has BsdThreads.
802 os::Bsd::set_libpthread_version("bsdthreads");
803 }
804
805 if (strstr(libpthread_version(), "NPTL")) {
806 os::Bsd::set_is_NPTL();
807 } else {
808 os::Bsd::set_is_BsdThreads();
809 }
810
811 // BsdThreads have two flavors: floating-stack mode, which allows variable
812 // stack size; and fixed-stack mode. NPTL is always floating-stack.
813 if (os::Bsd::is_NPTL() || os::Bsd::supports_variable_stack_size()) {
814 os::Bsd::set_is_floating_stack();
815 }
816 }
817
818 /////////////////////////////////////////////////////////////////////////////
819 // thread stack
820
821 // Force Bsd kernel to expand current thread stack. If "bottom" is close
822 // to the stack guard, caller should block all signals.
823 //
824 // MAP_GROWSDOWN:
825 // A special mmap() flag that is used to implement thread stacks. It tells
826 // kernel that the memory region should extend downwards when needed. This
827 // allows early versions of BsdThreads to only mmap the first few pages
828 // when creating a new thread. Bsd kernel will automatically expand thread
829 // stack as needed (on page faults).
830 //
831 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
832 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
833 // region, it's hard to tell if the fault is due to a legitimate stack
834 // access or because of reading/writing non-exist memory (e.g. buffer
835 // overrun). As a rule, if the fault happens below current stack pointer,
836 // Bsd kernel does not expand stack, instead a SIGSEGV is sent to the
837 // application (see Bsd kernel fault.c).
838 //
839 // This Bsd feature can cause SIGSEGV when VM bangs thread stack for
840 // stack overflow detection.
841 //
842 // Newer version of BsdThreads (since glibc-2.2, or, RH-7.x) and NPTL do
843 // not use this flag. However, the stack of initial thread is not created
844 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
845 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
846 // and then attach the thread to JVM.
847 //
848 // To get around the problem and allow stack banging on Bsd, we need to
849 // manually expand thread stack after receiving the SIGSEGV.
850 //
851 // There are two ways to expand thread stack to address "bottom", we used
852 // both of them in JVM before 1.5:
853 // 1. adjust stack pointer first so that it is below "bottom", and then
854 // touch "bottom"
855 // 2. mmap() the page in question
856 //
857 // Now alternate signal stack is gone, it's harder to use 2. For instance,
858 // if current sp is already near the lower end of page 101, and we need to
859 // call mmap() to map page 100, it is possible that part of the mmap() frame
860 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
861 // That will destroy the mmap() frame and cause VM to crash.
862 //
863 // The following code works by adjusting sp first, then accessing the "bottom"
864 // page to force a page fault. Bsd kernel will then automatically expand the
865 // stack mapping.
866 //
867 // _expand_stack_to() assumes its frame size is less than page size, which
868 // should always be true if the function is not inlined.
869
870 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
871 #define NOINLINE
872 #else
873 #define NOINLINE __attribute__ ((noinline))
874 #endif
875
876 static void _expand_stack_to(address bottom) NOINLINE;
877
878 static void _expand_stack_to(address bottom) {
879 address sp;
880 size_t size;
881 volatile char *p;
882
883 // Adjust bottom to point to the largest address within the same page, it
884 // gives us a one-page buffer if alloca() allocates slightly more memory.
885 bottom = (address)align_size_down((uintptr_t)bottom, os::Bsd::page_size());
886 bottom += os::Bsd::page_size() - 1;
887
888 // sp might be slightly above current stack pointer; if that's the case, we
889 // will alloca() a little more space than necessary, which is OK. Don't use
890 // os::current_stack_pointer(), as its result can be slightly below current
891 // stack pointer, causing us to not alloca enough to reach "bottom".
892 sp = (address)&sp;
893
894 if (sp > bottom) {
895 size = sp - bottom;
896 p = (volatile char *)alloca(size);
897 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
898 p[0] = '\0';
899 }
900 }
901
902 bool os::Bsd::manually_expand_stack(JavaThread * t, address addr) {
903 assert(t!=NULL, "just checking");
904 assert(t->osthread()->expanding_stack(), "expand should be set");
905 assert(t->stack_base() != NULL, "stack_base was not initialized");
906
907 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
908 sigset_t mask_all, old_sigset;
909 sigfillset(&mask_all);
910 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
911 _expand_stack_to(addr);
912 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
913 return true;
914 }
915 return false;
916 }
917 #endif
918
919 //////////////////////////////////////////////////////////////////////////////
920 // create new thread
921
922 static address highest_vm_reserved_address();
923
924 // check if it's safe to start a new thread
925 static bool _thread_safety_check(Thread* thread) {
926 #ifdef _ALLBSD_SOURCE
927 return true;
928 #else
929 if (os::Bsd::is_BsdThreads() && !os::Bsd::is_floating_stack()) {
930 // Fixed stack BsdThreads (SuSE Bsd/x86, and some versions of Redhat)
931 // Heap is mmap'ed at lower end of memory space. Thread stacks are
932 // allocated (MAP_FIXED) from high address space. Every thread stack
933 // occupies a fixed size slot (usually 2Mbytes, but user can change
934 // it to other values if they rebuild BsdThreads).
935 //
936 // Problem with MAP_FIXED is that mmap() can still succeed even part of
937 // the memory region has already been mmap'ed. That means if we have too
938 // many threads and/or very large heap, eventually thread stack will
939 // collide with heap.
940 //
941 // Here we try to prevent heap/stack collision by comparing current
942 // stack bottom with the highest address that has been mmap'ed by JVM
943 // plus a safety margin for memory maps created by native code.
944 //
945 // This feature can be disabled by setting ThreadSafetyMargin to 0
946 //
947 if (ThreadSafetyMargin > 0) {
948 address stack_bottom = os::current_stack_base() - os::current_stack_size();
949
950 // not safe if our stack extends below the safety margin
951 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
952 } else {
953 return true;
954 }
955 } else {
956 // Floating stack BsdThreads or NPTL:
957 // Unlike fixed stack BsdThreads, thread stacks are not MAP_FIXED. When
958 // there's not enough space left, pthread_create() will fail. If we come
959 // here, that means enough space has been reserved for stack.
960 return true;
961 }
962 #endif
963 }
964
965 #ifdef __APPLE__
966 // library handle for calling objc_registerThreadWithCollector()
967 // without static linking to the libobjc library
968 #define OBJC_LIB "/usr/lib/libobjc.dylib"
969 #define OBJC_GCREGISTER "objc_registerThreadWithCollector"
970 typedef void (*objc_registerThreadWithCollector_t)();
971 extern "C" objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction;
972 objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction = NULL;
973 #endif
974
975 // Thread start routine for all newly created threads
976 static void *java_start(Thread *thread) {
977 // Try to randomize the cache line index of hot stack frames.
978 // This helps when threads of the same stack traces evict each other's
979 // cache lines. The threads can be either from the same JVM instance, or
980 // from different JVM instances. The benefit is especially true for
981 // processors with hyperthreading technology.
982 static int counter = 0;
983 int pid = os::current_process_id();
984 alloca(((pid ^ counter++) & 7) * 128);
985
986 ThreadLocalStorage::set_thread(thread);
987
988 OSThread* osthread = thread->osthread();
989 Monitor* sync = osthread->startThread_lock();
990
991 // non floating stack BsdThreads needs extra check, see above
992 if (!_thread_safety_check(thread)) {
993 // notify parent thread
994 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
995 osthread->set_state(ZOMBIE);
996 sync->notify_all();
997 return NULL;
998 }
999
1000 #ifdef _ALLBSD_SOURCE
1001 #ifdef __APPLE__
1002 // thread_id is mach thread on macos
1003 osthread->set_thread_id(::mach_thread_self());
1004 #else
1005 // thread_id is pthread_id on BSD
1006 osthread->set_thread_id(::pthread_self());
1007 #endif
1008 #else
1009 // thread_id is kernel thread id (similar to Solaris LWP id)
1010 osthread->set_thread_id(os::Bsd::gettid());
1011
1012 if (UseNUMA) {
1013 int lgrp_id = os::numa_get_group_id();
1014 if (lgrp_id != -1) {
1015 thread->set_lgrp_id(lgrp_id);
1016 }
1017 }
1018 #endif
1019 // initialize signal mask for this thread
1020 os::Bsd::hotspot_sigmask(thread);
1021
1022 // initialize floating point control register
1023 os::Bsd::init_thread_fpu_state();
1024
1025 #ifdef __APPLE__
1026 // register thread with objc gc
1027 if (objc_registerThreadWithCollectorFunction != NULL) {
1028 objc_registerThreadWithCollectorFunction();
1029 }
1030 #endif
1031
1032 // handshaking with parent thread
1033 {
1034 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
1035
1036 // notify parent thread
1037 osthread->set_state(INITIALIZED);
1038 sync->notify_all();
1039
1040 // wait until os::start_thread()
1041 while (osthread->get_state() == INITIALIZED) {
1042 sync->wait(Mutex::_no_safepoint_check_flag);
1043 }
1044 }
1045
1046 // call one more level start routine
1047 thread->run();
1048
1049 return 0;
1050 }
1051
1052 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1053 assert(thread->osthread() == NULL, "caller responsible");
1054
1055 // Allocate the OSThread object
1056 OSThread* osthread = new OSThread(NULL, NULL);
1057 if (osthread == NULL) {
1058 return false;
1059 }
1060
1061 // set the correct thread state
1062 osthread->set_thread_type(thr_type);
1063
1064 // Initial state is ALLOCATED but not INITIALIZED
1065 osthread->set_state(ALLOCATED);
1066
1067 thread->set_osthread(osthread);
1068
1069 // init thread attributes
1070 pthread_attr_t attr;
1071 pthread_attr_init(&attr);
1072 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1073
1074 // stack size
1075 if (os::Bsd::supports_variable_stack_size()) {
1076 // calculate stack size if it's not specified by caller
1077 if (stack_size == 0) {
1078 stack_size = os::Bsd::default_stack_size(thr_type);
1079
1080 switch (thr_type) {
1081 case os::java_thread:
1082 // Java threads use ThreadStackSize which default value can be
1083 // changed with the flag -Xss
1084 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
1085 stack_size = JavaThread::stack_size_at_create();
1086 break;
1087 case os::compiler_thread:
1088 if (CompilerThreadStackSize > 0) {
1089 stack_size = (size_t)(CompilerThreadStackSize * K);
1090 break;
1091 } // else fall through:
1092 // use VMThreadStackSize if CompilerThreadStackSize is not defined
1093 case os::vm_thread:
1094 case os::pgc_thread:
1095 case os::cgc_thread:
1096 case os::watcher_thread:
1097 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1098 break;
1099 }
1100 }
1101
1102 stack_size = MAX2(stack_size, os::Bsd::min_stack_allowed);
1103 pthread_attr_setstacksize(&attr, stack_size);
1104 } else {
1105 // let pthread_create() pick the default value.
1106 }
1107
1108 #ifndef _ALLBSD_SOURCE
1109 // glibc guard page
1110 pthread_attr_setguardsize(&attr, os::Bsd::default_guard_size(thr_type));
1111 #endif
1112
1113 ThreadState state;
1114
1115 {
1116
1117 #ifndef _ALLBSD_SOURCE
1118 // Serialize thread creation if we are running with fixed stack BsdThreads
1119 bool lock = os::Bsd::is_BsdThreads() && !os::Bsd::is_floating_stack();
1120 if (lock) {
1121 os::Bsd::createThread_lock()->lock_without_safepoint_check();
1122 }
1123 #endif
1124
1125 pthread_t tid;
1126 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
1127
1128 pthread_attr_destroy(&attr);
1129
1130 if (ret != 0) {
1131 if (PrintMiscellaneous && (Verbose || WizardMode)) {
1132 perror("pthread_create()");
1133 }
1134 // Need to clean up stuff we've allocated so far
1135 thread->set_osthread(NULL);
1136 delete osthread;
1137 #ifndef _ALLBSD_SOURCE
1138 if (lock) os::Bsd::createThread_lock()->unlock();
1139 #endif
1140 return false;
1141 }
1142
1143 // Store pthread info into the OSThread
1144 osthread->set_pthread_id(tid);
1145
1146 // Wait until child thread is either initialized or aborted
1147 {
1148 Monitor* sync_with_child = osthread->startThread_lock();
1149 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1150 while ((state = osthread->get_state()) == ALLOCATED) {
1151 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
1152 }
1153 }
1154
1155 #ifndef _ALLBSD_SOURCE
1156 if (lock) {
1157 os::Bsd::createThread_lock()->unlock();
1158 }
1159 #endif
1160 }
1161
1162 // Aborted due to thread limit being reached
1163 if (state == ZOMBIE) {
1164 thread->set_osthread(NULL);
1165 delete osthread;
1166 return false;
1167 }
1168
1169 // The thread is returned suspended (in state INITIALIZED),
1170 // and is started higher up in the call chain
1171 assert(state == INITIALIZED, "race condition");
1172 return true;
1173 }
1174
1175 /////////////////////////////////////////////////////////////////////////////
1176 // attach existing thread
1177
1178 // bootstrap the main thread
1179 bool os::create_main_thread(JavaThread* thread) {
1180 assert(os::Bsd::_main_thread == pthread_self(), "should be called inside main thread");
1181 return create_attached_thread(thread);
1182 }
1183
1184 bool os::create_attached_thread(JavaThread* thread) {
1185 #ifdef ASSERT
1186 thread->verify_not_published();
1187 #endif
1188
1189 // Allocate the OSThread object
1190 OSThread* osthread = new OSThread(NULL, NULL);
1191
1192 if (osthread == NULL) {
1193 return false;
1194 }
1195
1196 // Store pthread info into the OSThread
1197 #ifdef _ALLBSD_SOURCE
1198 #ifdef __APPLE__
1199 osthread->set_thread_id(::mach_thread_self());
1200 #else
1201 osthread->set_thread_id(::pthread_self());
1202 #endif
1203 #else
1204 osthread->set_thread_id(os::Bsd::gettid());
1205 #endif
1206 osthread->set_pthread_id(::pthread_self());
1207
1208 // initialize floating point control register
1209 os::Bsd::init_thread_fpu_state();
1210
1211 // Initial thread state is RUNNABLE
1212 osthread->set_state(RUNNABLE);
1213
1214 thread->set_osthread(osthread);
1215
1216 #ifndef _ALLBSD_SOURCE
1217 if (UseNUMA) {
1218 int lgrp_id = os::numa_get_group_id();
1219 if (lgrp_id != -1) {
1220 thread->set_lgrp_id(lgrp_id);
1221 }
1222 }
1223
1224 if (os::Bsd::is_initial_thread()) {
1225 // If current thread is initial thread, its stack is mapped on demand,
1226 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1227 // the entire stack region to avoid SEGV in stack banging.
1228 // It is also useful to get around the heap-stack-gap problem on SuSE
1229 // kernel (see 4821821 for details). We first expand stack to the top
1230 // of yellow zone, then enable stack yellow zone (order is significant,
1231 // enabling yellow zone first will crash JVM on SuSE Bsd), so there
1232 // is no gap between the last two virtual memory regions.
1233
1234 JavaThread *jt = (JavaThread *)thread;
1235 address addr = jt->stack_yellow_zone_base();
1236 assert(addr != NULL, "initialization problem?");
1237 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1238
1239 osthread->set_expanding_stack();
1240 os::Bsd::manually_expand_stack(jt, addr);
1241 osthread->clear_expanding_stack();
1242 }
1243 #endif
1244
1245 // initialize signal mask for this thread
1246 // and save the caller's signal mask
1247 os::Bsd::hotspot_sigmask(thread);
1248
1249 return true;
1250 }
1251
1252 void os::pd_start_thread(Thread* thread) {
1253 OSThread * osthread = thread->osthread();
1254 assert(osthread->get_state() != INITIALIZED, "just checking");
1255 Monitor* sync_with_child = osthread->startThread_lock();
1256 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1257 sync_with_child->notify();
1258 }
1259
1260 // Free Bsd resources related to the OSThread
1261 void os::free_thread(OSThread* osthread) {
1262 assert(osthread != NULL, "osthread not set");
1263
1264 if (Thread::current()->osthread() == osthread) {
1265 // Restore caller's signal mask
1266 sigset_t sigmask = osthread->caller_sigmask();
1267 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1268 }
1269
1270 delete osthread;
1271 }
1272
1273 //////////////////////////////////////////////////////////////////////////////
1274 // thread local storage
1275
1276 int os::allocate_thread_local_storage() {
1277 pthread_key_t key;
1278 int rslt = pthread_key_create(&key, NULL);
1279 assert(rslt == 0, "cannot allocate thread local storage");
1280 return (int)key;
1281 }
1282
1283 // Note: This is currently not used by VM, as we don't destroy TLS key
1284 // on VM exit.
1285 void os::free_thread_local_storage(int index) {
1286 int rslt = pthread_key_delete((pthread_key_t)index);
1287 assert(rslt == 0, "invalid index");
1288 }
1289
1290 void os::thread_local_storage_at_put(int index, void* value) {
1291 int rslt = pthread_setspecific((pthread_key_t)index, value);
1292 assert(rslt == 0, "pthread_setspecific failed");
1293 }
1294
1295 extern "C" Thread* get_thread() {
1296 return ThreadLocalStorage::thread();
1297 }
1298
1299 //////////////////////////////////////////////////////////////////////////////
1300 // initial thread
1301
1302 #ifndef _ALLBSD_SOURCE
1303 // Check if current thread is the initial thread, similar to Solaris thr_main.
1304 bool os::Bsd::is_initial_thread(void) {
1305 char dummy;
1306 // If called before init complete, thread stack bottom will be null.
1307 // Can be called if fatal error occurs before initialization.
1308 if (initial_thread_stack_bottom() == NULL) return false;
1309 assert(initial_thread_stack_bottom() != NULL &&
1310 initial_thread_stack_size() != 0,
1311 "os::init did not locate initial thread's stack region");
1312 if ((address)&dummy >= initial_thread_stack_bottom() &&
1313 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1314 return true;
1315 else return false;
1316 }
1317
1318 // Find the virtual memory area that contains addr
1319 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1320 FILE *fp = fopen("/proc/self/maps", "r");
1321 if (fp) {
1322 address low, high;
1323 while (!feof(fp)) {
1324 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1325 if (low <= addr && addr < high) {
1326 if (vma_low) *vma_low = low;
1327 if (vma_high) *vma_high = high;
1328 fclose (fp);
1329 return true;
1330 }
1331 }
1332 for (;;) {
1333 int ch = fgetc(fp);
1334 if (ch == EOF || ch == (int)'\n') break;
1335 }
1336 }
1337 fclose(fp);
1338 }
1339 return false;
1340 }
1341
1342 // Locate initial thread stack. This special handling of initial thread stack
1343 // is needed because pthread_getattr_np() on most (all?) Bsd distros returns
1344 // bogus value for initial thread.
1345 void os::Bsd::capture_initial_stack(size_t max_size) {
1346 // stack size is the easy part, get it from RLIMIT_STACK
1347 size_t stack_size;
1348 struct rlimit rlim;
1349 getrlimit(RLIMIT_STACK, &rlim);
1350 stack_size = rlim.rlim_cur;
1351
1352 // 6308388: a bug in ld.so will relocate its own .data section to the
1353 // lower end of primordial stack; reduce ulimit -s value a little bit
1354 // so we won't install guard page on ld.so's data section.
1355 stack_size -= 2 * page_size();
1356
1357 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1358 // 7.1, in both cases we will get 2G in return value.
1359 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1360 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1361 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1362 // in case other parts in glibc still assumes 2M max stack size.
1363 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1364 #ifndef IA64
1365 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1366 #else
1367 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1368 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1369 #endif
1370
1371 // Try to figure out where the stack base (top) is. This is harder.
1372 //
1373 // When an application is started, glibc saves the initial stack pointer in
1374 // a global variable "__libc_stack_end", which is then used by system
1375 // libraries. __libc_stack_end should be pretty close to stack top. The
1376 // variable is available since the very early days. However, because it is
1377 // a private interface, it could disappear in the future.
1378 //
1379 // Bsd kernel saves start_stack information in /proc/<pid>/stat. Similar
1380 // to __libc_stack_end, it is very close to stack top, but isn't the real
1381 // stack top. Note that /proc may not exist if VM is running as a chroot
1382 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1383 // /proc/<pid>/stat could change in the future (though unlikely).
1384 //
1385 // We try __libc_stack_end first. If that doesn't work, look for
1386 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1387 // as a hint, which should work well in most cases.
1388
1389 uintptr_t stack_start;
1390
1391 // try __libc_stack_end first
1392 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1393 if (p && *p) {
1394 stack_start = *p;
1395 } else {
1396 // see if we can get the start_stack field from /proc/self/stat
1397 FILE *fp;
1398 int pid;
1399 char state;
1400 int ppid;
1401 int pgrp;
1402 int session;
1403 int nr;
1404 int tpgrp;
1405 unsigned long flags;
1406 unsigned long minflt;
1407 unsigned long cminflt;
1408 unsigned long majflt;
1409 unsigned long cmajflt;
1410 unsigned long utime;
1411 unsigned long stime;
1412 long cutime;
1413 long cstime;
1414 long prio;
1415 long nice;
1416 long junk;
1417 long it_real;
1418 uintptr_t start;
1419 uintptr_t vsize;
1420 intptr_t rss;
1421 uintptr_t rsslim;
1422 uintptr_t scodes;
1423 uintptr_t ecode;
1424 int i;
1425
1426 // Figure what the primordial thread stack base is. Code is inspired
1427 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1428 // followed by command name surrounded by parentheses, state, etc.
1429 char stat[2048];
1430 int statlen;
1431
1432 fp = fopen("/proc/self/stat", "r");
1433 if (fp) {
1434 statlen = fread(stat, 1, 2047, fp);
1435 stat[statlen] = '\0';
1436 fclose(fp);
1437
1438 // Skip pid and the command string. Note that we could be dealing with
1439 // weird command names, e.g. user could decide to rename java launcher
1440 // to "java 1.4.2 :)", then the stat file would look like
1441 // 1234 (java 1.4.2 :)) R ... ...
1442 // We don't really need to know the command string, just find the last
1443 // occurrence of ")" and then start parsing from there. See bug 4726580.
1444 char * s = strrchr(stat, ')');
1445
1446 i = 0;
1447 if (s) {
1448 // Skip blank chars
1449 do s++; while (isspace(*s));
1450
1451 #define _UFM UINTX_FORMAT
1452 #define _DFM INTX_FORMAT
1453
1454 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1455 /* 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 */
1456 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1457 &state, /* 3 %c */
1458 &ppid, /* 4 %d */
1459 &pgrp, /* 5 %d */
1460 &session, /* 6 %d */
1461 &nr, /* 7 %d */
1462 &tpgrp, /* 8 %d */
1463 &flags, /* 9 %lu */
1464 &minflt, /* 10 %lu */
1465 &cminflt, /* 11 %lu */
1466 &majflt, /* 12 %lu */
1467 &cmajflt, /* 13 %lu */
1468 &utime, /* 14 %lu */
1469 &stime, /* 15 %lu */
1470 &cutime, /* 16 %ld */
1471 &cstime, /* 17 %ld */
1472 &prio, /* 18 %ld */
1473 &nice, /* 19 %ld */
1474 &junk, /* 20 %ld */
1475 &it_real, /* 21 %ld */
1476 &start, /* 22 UINTX_FORMAT */
1477 &vsize, /* 23 UINTX_FORMAT */
1478 &rss, /* 24 INTX_FORMAT */
1479 &rsslim, /* 25 UINTX_FORMAT */
1480 &scodes, /* 26 UINTX_FORMAT */
1481 &ecode, /* 27 UINTX_FORMAT */
1482 &stack_start); /* 28 UINTX_FORMAT */
1483 }
1484
1485 #undef _UFM
1486 #undef _DFM
1487
1488 if (i != 28 - 2) {
1489 assert(false, "Bad conversion from /proc/self/stat");
1490 // product mode - assume we are the initial thread, good luck in the
1491 // embedded case.
1492 warning("Can't detect initial thread stack location - bad conversion");
1493 stack_start = (uintptr_t) &rlim;
1494 }
1495 } else {
1496 // For some reason we can't open /proc/self/stat (for example, running on
1497 // FreeBSD with a Bsd emulator, or inside chroot), this should work for
1498 // most cases, so don't abort:
1499 warning("Can't detect initial thread stack location - no /proc/self/stat");
1500 stack_start = (uintptr_t) &rlim;
1501 }
1502 }
1503
1504 // Now we have a pointer (stack_start) very close to the stack top, the
1505 // next thing to do is to figure out the exact location of stack top. We
1506 // can find out the virtual memory area that contains stack_start by
1507 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1508 // and its upper limit is the real stack top. (again, this would fail if
1509 // running inside chroot, because /proc may not exist.)
1510
1511 uintptr_t stack_top;
1512 address low, high;
1513 if (find_vma((address)stack_start, &low, &high)) {
1514 // success, "high" is the true stack top. (ignore "low", because initial
1515 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1516 stack_top = (uintptr_t)high;
1517 } else {
1518 // failed, likely because /proc/self/maps does not exist
1519 warning("Can't detect initial thread stack location - find_vma failed");
1520 // best effort: stack_start is normally within a few pages below the real
1521 // stack top, use it as stack top, and reduce stack size so we won't put
1522 // guard page outside stack.
1523 stack_top = stack_start;
1524 stack_size -= 16 * page_size();
1525 }
1526
1527 // stack_top could be partially down the page so align it
1528 stack_top = align_size_up(stack_top, page_size());
1529
1530 if (max_size && stack_size > max_size) {
1531 _initial_thread_stack_size = max_size;
1532 } else {
1533 _initial_thread_stack_size = stack_size;
1534 }
1535
1536 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1537 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1538 }
1539 #endif
1540
1541 ////////////////////////////////////////////////////////////////////////////////
1542 // time support
1543
1544 // Time since start-up in seconds to a fine granularity.
1545 // Used by VMSelfDestructTimer and the MemProfiler.
1546 double os::elapsedTime() {
1547
1548 return (double)(os::elapsed_counter()) * 0.000001;
1549 }
1550
1551 jlong os::elapsed_counter() {
1552 timeval time;
1553 int status = gettimeofday(&time, NULL);
1554 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1555 }
1556
1557 jlong os::elapsed_frequency() {
1558 return (1000 * 1000);
1559 }
1560
1561 // XXX: For now, code this as if BSD does not support vtime.
1562 bool os::supports_vtime() { return false; }
1563 bool os::enable_vtime() { return false; }
1564 bool os::vtime_enabled() { return false; }
1565 double os::elapsedVTime() {
1566 // better than nothing, but not much
1567 return elapsedTime();
1568 }
1569
1570 jlong os::javaTimeMillis() {
1571 timeval time;
1572 int status = gettimeofday(&time, NULL);
1573 assert(status != -1, "bsd error");
1574 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1575 }
1576
1577 #ifndef CLOCK_MONOTONIC
1578 #define CLOCK_MONOTONIC (1)
1579 #endif
1580
1581 #ifdef __APPLE__
1582 void os::Bsd::clock_init() {
1583 // XXXDARWIN: Investigate replacement monotonic clock
1584 }
1585 #elif defined(_ALLBSD_SOURCE)
1586 void os::Bsd::clock_init() {
1587 struct timespec res;
1588 struct timespec tp;
1589 if (::clock_getres(CLOCK_MONOTONIC, &res) == 0 &&
1590 ::clock_gettime(CLOCK_MONOTONIC, &tp) == 0) {
1591 // yes, monotonic clock is supported
1592 _clock_gettime = ::clock_gettime;
1593 }
1594 }
1595 #else
1596 void os::Bsd::clock_init() {
1597 // we do dlopen's in this particular order due to bug in bsd
1598 // dynamical loader (see 6348968) leading to crash on exit
1599 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1600 if (handle == NULL) {
1601 handle = dlopen("librt.so", RTLD_LAZY);
1602 }
1603
1604 if (handle) {
1605 int (*clock_getres_func)(clockid_t, struct timespec*) =
1606 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1607 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1608 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1609 if (clock_getres_func && clock_gettime_func) {
1610 // See if monotonic clock is supported by the kernel. Note that some
1611 // early implementations simply return kernel jiffies (updated every
1612 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1613 // for nano time (though the monotonic property is still nice to have).
1614 // It's fixed in newer kernels, however clock_getres() still returns
1615 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1616 // resolution for now. Hopefully as people move to new kernels, this
1617 // won't be a problem.
1618 struct timespec res;
1619 struct timespec tp;
1620 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1621 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1622 // yes, monotonic clock is supported
1623 _clock_gettime = clock_gettime_func;
1624 } else {
1625 // close librt if there is no monotonic clock
1626 dlclose(handle);
1627 }
1628 }
1629 }
1630 }
1631 #endif
1632
1633 #ifndef _ALLBSD_SOURCE
1634 #ifndef SYS_clock_getres
1635
1636 #if defined(IA32) || defined(AMD64)
1637 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1638 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1639 #else
1640 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1641 #define sys_clock_getres(x,y) -1
1642 #endif
1643
1644 #else
1645 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1646 #endif
1647
1648 void os::Bsd::fast_thread_clock_init() {
1649 if (!UseBsdPosixThreadCPUClocks) {
1650 return;
1651 }
1652 clockid_t clockid;
1653 struct timespec tp;
1654 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1655 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1656
1657 // Switch to using fast clocks for thread cpu time if
1658 // the sys_clock_getres() returns 0 error code.
1659 // Note, that some kernels may support the current thread
1660 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1661 // returned by the pthread_getcpuclockid().
1662 // If the fast Posix clocks are supported then the sys_clock_getres()
1663 // must return at least tp.tv_sec == 0 which means a resolution
1664 // better than 1 sec. This is extra check for reliability.
1665
1666 if(pthread_getcpuclockid_func &&
1667 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1668 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1669
1670 _supports_fast_thread_cpu_time = true;
1671 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1672 }
1673 }
1674 #endif
1675
1676 jlong os::javaTimeNanos() {
1677 if (Bsd::supports_monotonic_clock()) {
1678 struct timespec tp;
1679 int status = Bsd::clock_gettime(CLOCK_MONOTONIC, &tp);
1680 assert(status == 0, "gettime error");
1681 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1682 return result;
1683 } else {
1684 timeval time;
1685 int status = gettimeofday(&time, NULL);
1686 assert(status != -1, "bsd error");
1687 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1688 return 1000 * usecs;
1689 }
1690 }
1691
1692 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1693 if (Bsd::supports_monotonic_clock()) {
1694 info_ptr->max_value = ALL_64_BITS;
1695
1696 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1697 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1698 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1699 } else {
1700 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1701 info_ptr->max_value = ALL_64_BITS;
1702
1703 // gettimeofday is a real time clock so it skips
1704 info_ptr->may_skip_backward = true;
1705 info_ptr->may_skip_forward = true;
1706 }
1707
1708 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1709 }
1710
1711 // Return the real, user, and system times in seconds from an
1712 // arbitrary fixed point in the past.
1713 bool os::getTimesSecs(double* process_real_time,
1714 double* process_user_time,
1715 double* process_system_time) {
1716 struct tms ticks;
1717 clock_t real_ticks = times(&ticks);
1718
1719 if (real_ticks == (clock_t) (-1)) {
1720 return false;
1721 } else {
1722 double ticks_per_second = (double) clock_tics_per_sec;
1723 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1724 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1725 *process_real_time = ((double) real_ticks) / ticks_per_second;
1726
1727 return true;
1728 }
1729 }
1730
1731
1732 char * os::local_time_string(char *buf, size_t buflen) {
1733 struct tm t;
1734 time_t long_time;
1735 time(&long_time);
1736 localtime_r(&long_time, &t);
1737 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1738 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1739 t.tm_hour, t.tm_min, t.tm_sec);
1740 return buf;
1741 }
1742
1743 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1744 return localtime_r(clock, res);
1745 }
1746
1747 ////////////////////////////////////////////////////////////////////////////////
1748 // runtime exit support
1749
1750 // Note: os::shutdown() might be called very early during initialization, or
1751 // called from signal handler. Before adding something to os::shutdown(), make
1752 // sure it is async-safe and can handle partially initialized VM.
1753 void os::shutdown() {
1754
1755 // allow PerfMemory to attempt cleanup of any persistent resources
1756 perfMemory_exit();
1757
1758 // needs to remove object in file system
1759 AttachListener::abort();
1760
1761 // flush buffered output, finish log files
1762 ostream_abort();
1763
1764 // Check for abort hook
1765 abort_hook_t abort_hook = Arguments::abort_hook();
1766 if (abort_hook != NULL) {
1767 abort_hook();
1768 }
1769
1770 }
1771
1772 // Note: os::abort() might be called very early during initialization, or
1773 // called from signal handler. Before adding something to os::abort(), make
1774 // sure it is async-safe and can handle partially initialized VM.
1775 void os::abort(bool dump_core) {
1776 os::shutdown();
1777 if (dump_core) {
1778 #ifndef PRODUCT
1779 fdStream out(defaultStream::output_fd());
1780 out.print_raw("Current thread is ");
1781 char buf[16];
1782 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1783 out.print_raw_cr(buf);
1784 out.print_raw_cr("Dumping core ...");
1785 #endif
1786 ::abort(); // dump core
1787 }
1788
1789 ::exit(1);
1790 }
1791
1792 // Die immediately, no exit hook, no abort hook, no cleanup.
1793 void os::die() {
1794 // _exit() on BsdThreads only kills current thread
1795 ::abort();
1796 }
1797
1798 // unused on bsd for now.
1799 void os::set_error_file(const char *logfile) {}
1800
1801
1802 // This method is a copy of JDK's sysGetLastErrorString
1803 // from src/solaris/hpi/src/system_md.c
1804
1805 size_t os::lasterror(char *buf, size_t len) {
1806
1807 if (errno == 0) return 0;
1808
1809 const char *s = ::strerror(errno);
1810 size_t n = ::strlen(s);
1811 if (n >= len) {
1812 n = len - 1;
1813 }
1814 ::strncpy(buf, s, n);
1815 buf[n] = '\0';
1816 return n;
1817 }
1818
1819 intx os::current_thread_id() {
1820 #ifdef __APPLE__
1821 return (intx)::mach_thread_self();
1822 #else
1823 return (intx)::pthread_self();
1824 #endif
1825 }
1826 int os::current_process_id() {
1827
1828 // Under the old bsd thread library, bsd gives each thread
1829 // its own process id. Because of this each thread will return
1830 // a different pid if this method were to return the result
1831 // of getpid(2). Bsd provides no api that returns the pid
1832 // of the launcher thread for the vm. This implementation
1833 // returns a unique pid, the pid of the launcher thread
1834 // that starts the vm 'process'.
1835
1836 // Under the NPTL, getpid() returns the same pid as the
1837 // launcher thread rather than a unique pid per thread.
1838 // Use gettid() if you want the old pre NPTL behaviour.
1839
1840 // if you are looking for the result of a call to getpid() that
1841 // returns a unique pid for the calling thread, then look at the
1842 // OSThread::thread_id() method in osThread_bsd.hpp file
1843
1844 return (int)(_initial_pid ? _initial_pid : getpid());
1845 }
1846
1847 // DLL functions
1848
1849 #define JNI_LIB_PREFIX "lib"
1850 #ifdef __APPLE__
1851 #define JNI_LIB_SUFFIX ".dylib"
1852 #else
1853 #define JNI_LIB_SUFFIX ".so"
1854 #endif
1855
1856 const char* os::dll_file_extension() { return JNI_LIB_SUFFIX; }
1857
1858 // This must be hard coded because it's the system's temporary
1859 // directory not the java application's temp directory, ala java.io.tmpdir.
1860 #ifdef __APPLE__
1861 // macosx has a secure per-user temporary directory
1862 char temp_path_storage[PATH_MAX];
1863 const char* os::get_temp_directory() {
1864 static char *temp_path = NULL;
1865 if (temp_path == NULL) {
1866 int pathSize = confstr(_CS_DARWIN_USER_TEMP_DIR, temp_path_storage, PATH_MAX);
1867 if (pathSize == 0 || pathSize > PATH_MAX) {
1868 strlcpy(temp_path_storage, "/tmp/", sizeof(temp_path_storage));
1869 }
1870 temp_path = temp_path_storage;
1871 }
1872 return temp_path;
1873 }
1874 #else /* __APPLE__ */
1875 const char* os::get_temp_directory() { return "/tmp"; }
1876 #endif /* __APPLE__ */
1877
1878 static bool file_exists(const char* filename) {
1879 struct stat statbuf;
1880 if (filename == NULL || strlen(filename) == 0) {
1881 return false;
1882 }
1883 return os::stat(filename, &statbuf) == 0;
1884 }
1885
1886 void os::dll_build_name(char* buffer, size_t buflen,
1887 const char* pname, const char* fname) {
1888 // Copied from libhpi
1889 const size_t pnamelen = pname ? strlen(pname) : 0;
1890
1891 // Quietly truncate on buffer overflow. Should be an error.
1892 if (pnamelen + strlen(fname) + strlen(JNI_LIB_PREFIX) + strlen(JNI_LIB_SUFFIX) + 2 > buflen) {
1893 *buffer = '\0';
1894 return;
1895 }
1896
1897 if (pnamelen == 0) {
1898 snprintf(buffer, buflen, JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, fname);
1899 } else if (strchr(pname, *os::path_separator()) != NULL) {
1900 int n;
1901 char** pelements = split_path(pname, &n);
1902 for (int i = 0 ; i < n ; i++) {
1903 // Really shouldn't be NULL, but check can't hurt
1904 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1905 continue; // skip the empty path values
1906 }
1907 snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX,
1908 pelements[i], fname);
1909 if (file_exists(buffer)) {
1910 break;
1911 }
1912 }
1913 // release the storage
1914 for (int i = 0 ; i < n ; i++) {
1915 if (pelements[i] != NULL) {
1916 FREE_C_HEAP_ARRAY(char, pelements[i]);
1917 }
1918 }
1919 if (pelements != NULL) {
1920 FREE_C_HEAP_ARRAY(char*, pelements);
1921 }
1922 } else {
1923 snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, pname, fname);
1924 }
1925 }
1926
1927 const char* os::get_current_directory(char *buf, int buflen) {
1928 return getcwd(buf, buflen);
1929 }
1930
1931 // check if addr is inside libjvm[_g].so
1932 bool os::address_is_in_vm(address addr) {
1933 static address libjvm_base_addr;
1934 Dl_info dlinfo;
1935
1936 if (libjvm_base_addr == NULL) {
1937 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1938 libjvm_base_addr = (address)dlinfo.dli_fbase;
1939 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1940 }
1941
1942 if (dladdr((void *)addr, &dlinfo)) {
1943 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1944 }
1945
1946 return false;
1947 }
1948
1949 bool os::dll_address_to_function_name(address addr, char *buf,
1950 int buflen, int *offset) {
1951 Dl_info dlinfo;
1952
1953 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1954 if (buf != NULL) {
1955 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1956 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1957 }
1958 }
1959 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1960 return true;
1961 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1962 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1963 buf, buflen, offset, dlinfo.dli_fname)) {
1964 return true;
1965 }
1966 }
1967
1968 if (buf != NULL) buf[0] = '\0';
1969 if (offset != NULL) *offset = -1;
1970 return false;
1971 }
1972
1973 #ifdef _ALLBSD_SOURCE
1974 // ported from solaris version
1975 bool os::dll_address_to_library_name(address addr, char* buf,
1976 int buflen, int* offset) {
1977 Dl_info dlinfo;
1978
1979 if (dladdr((void*)addr, &dlinfo)){
1980 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1981 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1982 return true;
1983 } else {
1984 if (buf) buf[0] = '\0';
1985 if (offset) *offset = -1;
1986 return false;
1987 }
1988 }
1989 #else
1990 struct _address_to_library_name {
1991 address addr; // input : memory address
1992 size_t buflen; // size of fname
1993 char* fname; // output: library name
1994 address base; // library base addr
1995 };
1996
1997 static int address_to_library_name_callback(struct dl_phdr_info *info,
1998 size_t size, void *data) {
1999 int i;
2000 bool found = false;
2001 address libbase = NULL;
2002 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
2003
2004 // iterate through all loadable segments
2005 for (i = 0; i < info->dlpi_phnum; i++) {
2006 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
2007 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
2008 // base address of a library is the lowest address of its loaded
2009 // segments.
2010 if (libbase == NULL || libbase > segbase) {
2011 libbase = segbase;
2012 }
2013 // see if 'addr' is within current segment
2014 if (segbase <= d->addr &&
2015 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
2016 found = true;
2017 }
2018 }
2019 }
2020
2021 // dlpi_name is NULL or empty if the ELF file is executable, return 0
2022 // so dll_address_to_library_name() can fall through to use dladdr() which
2023 // can figure out executable name from argv[0].
2024 if (found && info->dlpi_name && info->dlpi_name[0]) {
2025 d->base = libbase;
2026 if (d->fname) {
2027 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
2028 }
2029 return 1;
2030 }
2031 return 0;
2032 }
2033
2034 bool os::dll_address_to_library_name(address addr, char* buf,
2035 int buflen, int* offset) {
2036 Dl_info dlinfo;
2037 struct _address_to_library_name data;
2038
2039 // There is a bug in old glibc dladdr() implementation that it could resolve
2040 // to wrong library name if the .so file has a base address != NULL. Here
2041 // we iterate through the program headers of all loaded libraries to find
2042 // out which library 'addr' really belongs to. This workaround can be
2043 // removed once the minimum requirement for glibc is moved to 2.3.x.
2044 data.addr = addr;
2045 data.fname = buf;
2046 data.buflen = buflen;
2047 data.base = NULL;
2048 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
2049
2050 if (rslt) {
2051 // buf already contains library name
2052 if (offset) *offset = addr - data.base;
2053 return true;
2054 } else if (dladdr((void*)addr, &dlinfo)){
2055 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
2056 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
2057 return true;
2058 } else {
2059 if (buf) buf[0] = '\0';
2060 if (offset) *offset = -1;
2061 return false;
2062 }
2063 }
2064 #endif
2065
2066 // Loads .dll/.so and
2067 // in case of error it checks if .dll/.so was built for the
2068 // same architecture as Hotspot is running on
2069
2070 #ifdef __APPLE__
2071 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
2072 void * result= ::dlopen(filename, RTLD_LAZY);
2073 if (result != NULL) {
2074 // Successful loading
2075 return result;
2076 }
2077
2078 // Read system error message into ebuf
2079 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2080 ebuf[ebuflen-1]='\0';
2081
2082 return NULL;
2083 }
2084 #else
2085 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
2086 {
2087 void * result= ::dlopen(filename, RTLD_LAZY);
2088 if (result != NULL) {
2089 // Successful loading
2090 return result;
2091 }
2092
2093 Elf32_Ehdr elf_head;
2094
2095 // Read system error message into ebuf
2096 // It may or may not be overwritten below
2097 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2098 ebuf[ebuflen-1]='\0';
2099 int diag_msg_max_length=ebuflen-strlen(ebuf);
2100 char* diag_msg_buf=ebuf+strlen(ebuf);
2101
2102 if (diag_msg_max_length==0) {
2103 // No more space in ebuf for additional diagnostics message
2104 return NULL;
2105 }
2106
2107
2108 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2109
2110 if (file_descriptor < 0) {
2111 // Can't open library, report dlerror() message
2112 return NULL;
2113 }
2114
2115 bool failed_to_read_elf_head=
2116 (sizeof(elf_head)!=
2117 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2118
2119 ::close(file_descriptor);
2120 if (failed_to_read_elf_head) {
2121 // file i/o error - report dlerror() msg
2122 return NULL;
2123 }
2124
2125 typedef struct {
2126 Elf32_Half code; // Actual value as defined in elf.h
2127 Elf32_Half compat_class; // Compatibility of archs at VM's sense
2128 char elf_class; // 32 or 64 bit
2129 char endianess; // MSB or LSB
2130 char* name; // String representation
2131 } arch_t;
2132
2133 #ifndef EM_486
2134 #define EM_486 6 /* Intel 80486 */
2135 #endif
2136
2137 #ifndef EM_MIPS_RS3_LE
2138 #define EM_MIPS_RS3_LE 10 /* MIPS */
2139 #endif
2140
2141 #ifndef EM_PPC64
2142 #define EM_PPC64 21 /* PowerPC64 */
2143 #endif
2144
2145 #ifndef EM_S390
2146 #define EM_S390 22 /* IBM System/390 */
2147 #endif
2148
2149 #ifndef EM_IA_64
2150 #define EM_IA_64 50 /* HP/Intel IA-64 */
2151 #endif
2152
2153 #ifndef EM_X86_64
2154 #define EM_X86_64 62 /* AMD x86-64 */
2155 #endif
2156
2157 static const arch_t arch_array[]={
2158 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2159 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2160 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2161 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2162 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2163 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2164 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2165 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2166 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
2167 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
2168 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
2169 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
2170 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
2171 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
2172 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
2173 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
2174 };
2175
2176 #if (defined IA32)
2177 static Elf32_Half running_arch_code=EM_386;
2178 #elif (defined AMD64)
2179 static Elf32_Half running_arch_code=EM_X86_64;
2180 #elif (defined IA64)
2181 static Elf32_Half running_arch_code=EM_IA_64;
2182 #elif (defined __sparc) && (defined _LP64)
2183 static Elf32_Half running_arch_code=EM_SPARCV9;
2184 #elif (defined __sparc) && (!defined _LP64)
2185 static Elf32_Half running_arch_code=EM_SPARC;
2186 #elif (defined __powerpc64__)
2187 static Elf32_Half running_arch_code=EM_PPC64;
2188 #elif (defined __powerpc__)
2189 static Elf32_Half running_arch_code=EM_PPC;
2190 #elif (defined ARM)
2191 static Elf32_Half running_arch_code=EM_ARM;
2192 #elif (defined S390)
2193 static Elf32_Half running_arch_code=EM_S390;
2194 #elif (defined ALPHA)
2195 static Elf32_Half running_arch_code=EM_ALPHA;
2196 #elif (defined MIPSEL)
2197 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
2198 #elif (defined PARISC)
2199 static Elf32_Half running_arch_code=EM_PARISC;
2200 #elif (defined MIPS)
2201 static Elf32_Half running_arch_code=EM_MIPS;
2202 #elif (defined M68K)
2203 static Elf32_Half running_arch_code=EM_68K;
2204 #else
2205 #error Method os::dll_load requires that one of following is defined:\
2206 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2207 #endif
2208
2209 // Identify compatability class for VM's architecture and library's architecture
2210 // Obtain string descriptions for architectures
2211
2212 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2213 int running_arch_index=-1;
2214
2215 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2216 if (running_arch_code == arch_array[i].code) {
2217 running_arch_index = i;
2218 }
2219 if (lib_arch.code == arch_array[i].code) {
2220 lib_arch.compat_class = arch_array[i].compat_class;
2221 lib_arch.name = arch_array[i].name;
2222 }
2223 }
2224
2225 assert(running_arch_index != -1,
2226 "Didn't find running architecture code (running_arch_code) in arch_array");
2227 if (running_arch_index == -1) {
2228 // Even though running architecture detection failed
2229 // we may still continue with reporting dlerror() message
2230 return NULL;
2231 }
2232
2233 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2234 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2235 return NULL;
2236 }
2237
2238 #ifndef S390
2239 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2240 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2241 return NULL;
2242 }
2243 #endif // !S390
2244
2245 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2246 if ( lib_arch.name!=NULL ) {
2247 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2248 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2249 lib_arch.name, arch_array[running_arch_index].name);
2250 } else {
2251 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2252 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2253 lib_arch.code,
2254 arch_array[running_arch_index].name);
2255 }
2256 }
2257
2258 return NULL;
2259 }
2260 #endif /* !__APPLE__ */
2261
2262 // XXX: Do we need a lock around this as per Linux?
2263 void* os::dll_lookup(void* handle, const char* name) {
2264 return dlsym(handle, name);
2265 }
2266
2267
2268 static bool _print_ascii_file(const char* filename, outputStream* st) {
2269 int fd = ::open(filename, O_RDONLY);
2270 if (fd == -1) {
2271 return false;
2272 }
2273
2274 char buf[32];
2275 int bytes;
2276 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2277 st->print_raw(buf, bytes);
2278 }
2279
2280 ::close(fd);
2281
2282 return true;
2283 }
2284
2285 void os::print_dll_info(outputStream *st) {
2286 st->print_cr("Dynamic libraries:");
2287 #ifdef _ALLBSD_SOURCE
2288 #ifdef RTLD_DI_LINKMAP
2289 Dl_info dli;
2290 void *handle;
2291 Link_map *map;
2292 Link_map *p;
2293
2294 if (!dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli)) {
2295 st->print_cr("Error: Cannot print dynamic libraries.");
2296 return;
2297 }
2298 handle = dlopen(dli.dli_fname, RTLD_LAZY);
2299 if (handle == NULL) {
2300 st->print_cr("Error: Cannot print dynamic libraries.");
2301 return;
2302 }
2303 dlinfo(handle, RTLD_DI_LINKMAP, &map);
2304 if (map == NULL) {
2305 st->print_cr("Error: Cannot print dynamic libraries.");
2306 return;
2307 }
2308
2309 while (map->l_prev != NULL)
2310 map = map->l_prev;
2311
2312 while (map != NULL) {
2313 st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
2314 map = map->l_next;
2315 }
2316
2317 dlclose(handle);
2318 #elif defined(__APPLE__)
2319 uint32_t count;
2320 uint32_t i;
2321
2322 count = _dyld_image_count();
2323 for (i = 1; i < count; i++) {
2324 const char *name = _dyld_get_image_name(i);
2325 intptr_t slide = _dyld_get_image_vmaddr_slide(i);
2326 st->print_cr(PTR_FORMAT " \t%s", slide, name);
2327 }
2328 #else
2329 st->print_cr("Error: Cannot print dynamic libraries.");
2330 #endif
2331 #else
2332 char fname[32];
2333 pid_t pid = os::Bsd::gettid();
2334
2335 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2336
2337 if (!_print_ascii_file(fname, st)) {
2338 st->print("Can not get library information for pid = %d\n", pid);
2339 }
2340 #endif
2341 }
2342
2343 void os::print_os_info_brief(outputStream* st) {
2344 st->print("Bsd");
2345
2346 os::Posix::print_uname_info(st);
2347 }
2348
2349 void os::print_os_info(outputStream* st) {
2350 st->print("OS:");
2351 st->print("Bsd");
2352
2353 os::Posix::print_uname_info(st);
2354
2355 os::Posix::print_rlimit_info(st);
2356
2357 os::Posix::print_load_average(st);
2358 }
2359
2360 void os::pd_print_cpu_info(outputStream* st) {
2361 // Nothing to do for now.
2362 }
2363
2364 void os::print_memory_info(outputStream* st) {
2365
2366 st->print("Memory:");
2367 st->print(" %dk page", os::vm_page_size()>>10);
2368
2369 #ifndef _ALLBSD_SOURCE
2370 // values in struct sysinfo are "unsigned long"
2371 struct sysinfo si;
2372 sysinfo(&si);
2373 #endif
2374
2375 st->print(", physical " UINT64_FORMAT "k",
2376 os::physical_memory() >> 10);
2377 st->print("(" UINT64_FORMAT "k free)",
2378 os::available_memory() >> 10);
2379 #ifndef _ALLBSD_SOURCE
2380 st->print(", swap " UINT64_FORMAT "k",
2381 ((jlong)si.totalswap * si.mem_unit) >> 10);
2382 st->print("(" UINT64_FORMAT "k free)",
2383 ((jlong)si.freeswap * si.mem_unit) >> 10);
2384 #endif
2385 st->cr();
2386
2387 // meminfo
2388 st->print("\n/proc/meminfo:\n");
2389 _print_ascii_file("/proc/meminfo", st);
2390 st->cr();
2391 }
2392
2393 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2394 // but they're the same for all the bsd arch that we support
2395 // and they're the same for solaris but there's no common place to put this.
2396 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2397 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2398 "ILL_COPROC", "ILL_BADSTK" };
2399
2400 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2401 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2402 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2403
2404 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2405
2406 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2407
2408 void os::print_siginfo(outputStream* st, void* siginfo) {
2409 st->print("siginfo:");
2410
2411 const int buflen = 100;
2412 char buf[buflen];
2413 siginfo_t *si = (siginfo_t*)siginfo;
2414 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2415 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2416 st->print("si_errno=%s", buf);
2417 } else {
2418 st->print("si_errno=%d", si->si_errno);
2419 }
2420 const int c = si->si_code;
2421 assert(c > 0, "unexpected si_code");
2422 switch (si->si_signo) {
2423 case SIGILL:
2424 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2425 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2426 break;
2427 case SIGFPE:
2428 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2429 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2430 break;
2431 case SIGSEGV:
2432 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2433 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2434 break;
2435 case SIGBUS:
2436 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2437 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2438 break;
2439 default:
2440 st->print(", si_code=%d", si->si_code);
2441 // no si_addr
2442 }
2443
2444 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2445 UseSharedSpaces) {
2446 FileMapInfo* mapinfo = FileMapInfo::current_info();
2447 if (mapinfo->is_in_shared_space(si->si_addr)) {
2448 st->print("\n\nError accessing class data sharing archive." \
2449 " Mapped file inaccessible during execution, " \
2450 " possible disk/network problem.");
2451 }
2452 }
2453 st->cr();
2454 }
2455
2456
2457 static void print_signal_handler(outputStream* st, int sig,
2458 char* buf, size_t buflen);
2459
2460 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2461 st->print_cr("Signal Handlers:");
2462 print_signal_handler(st, SIGSEGV, buf, buflen);
2463 print_signal_handler(st, SIGBUS , buf, buflen);
2464 print_signal_handler(st, SIGFPE , buf, buflen);
2465 print_signal_handler(st, SIGPIPE, buf, buflen);
2466 print_signal_handler(st, SIGXFSZ, buf, buflen);
2467 print_signal_handler(st, SIGILL , buf, buflen);
2468 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2469 print_signal_handler(st, SR_signum, buf, buflen);
2470 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2471 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2472 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2473 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2474 }
2475
2476 static char saved_jvm_path[MAXPATHLEN] = {0};
2477
2478 // Find the full path to the current module, libjvm or libjvm_g
2479 void os::jvm_path(char *buf, jint buflen) {
2480 // Error checking.
2481 if (buflen < MAXPATHLEN) {
2482 assert(false, "must use a large-enough buffer");
2483 buf[0] = '\0';
2484 return;
2485 }
2486 // Lazy resolve the path to current module.
2487 if (saved_jvm_path[0] != 0) {
2488 strcpy(buf, saved_jvm_path);
2489 return;
2490 }
2491
2492 char dli_fname[MAXPATHLEN];
2493 bool ret = dll_address_to_library_name(
2494 CAST_FROM_FN_PTR(address, os::jvm_path),
2495 dli_fname, sizeof(dli_fname), NULL);
2496 assert(ret != 0, "cannot locate libjvm");
2497 char *rp = realpath(dli_fname, buf);
2498 if (rp == NULL)
2499 return;
2500
2501 if (Arguments::created_by_gamma_launcher()) {
2502 // Support for the gamma launcher. Typical value for buf is
2503 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm". If "/jre/lib/" appears at
2504 // the right place in the string, then assume we are installed in a JDK and
2505 // we're done. Otherwise, check for a JAVA_HOME environment variable and
2506 // construct a path to the JVM being overridden.
2507
2508 const char *p = buf + strlen(buf) - 1;
2509 for (int count = 0; p > buf && count < 5; ++count) {
2510 for (--p; p > buf && *p != '/'; --p)
2511 /* empty */ ;
2512 }
2513
2514 if (strncmp(p, "/jre/lib/", 9) != 0) {
2515 // Look for JAVA_HOME in the environment.
2516 char* java_home_var = ::getenv("JAVA_HOME");
2517 if (java_home_var != NULL && java_home_var[0] != 0) {
2518 char* jrelib_p;
2519 int len;
2520
2521 // Check the current module name "libjvm" or "libjvm_g".
2522 p = strrchr(buf, '/');
2523 assert(strstr(p, "/libjvm") == p, "invalid library name");
2524 p = strstr(p, "_g") ? "_g" : "";
2525
2526 rp = realpath(java_home_var, buf);
2527 if (rp == NULL)
2528 return;
2529
2530 // determine if this is a legacy image or modules image
2531 // modules image doesn't have "jre" subdirectory
2532 len = strlen(buf);
2533 jrelib_p = buf + len;
2534
2535 // Add the appropriate library subdir
2536 snprintf(jrelib_p, buflen-len, "/jre/lib");
2537 if (0 != access(buf, F_OK)) {
2538 snprintf(jrelib_p, buflen-len, "/lib");
2539 }
2540
2541 // Add the appropriate client or server subdir
2542 len = strlen(buf);
2543 jrelib_p = buf + len;
2544 snprintf(jrelib_p, buflen-len, "/%s", COMPILER_VARIANT);
2545 if (0 != access(buf, F_OK)) {
2546 snprintf(jrelib_p, buflen-len, "");
2547 }
2548
2549 // If the path exists within JAVA_HOME, add the JVM library name
2550 // to complete the path to JVM being overridden. Otherwise fallback
2551 // to the path to the current library.
2552 if (0 == access(buf, F_OK)) {
2553 // Use current module name "libjvm[_g]" instead of
2554 // "libjvm"debug_only("_g")"" since for fastdebug version
2555 // we should have "libjvm" but debug_only("_g") adds "_g"!
2556 len = strlen(buf);
2557 snprintf(buf + len, buflen-len, "/libjvm%s%s", p, JNI_LIB_SUFFIX);
2558 } else {
2559 // Fall back to path of current library
2560 rp = realpath(dli_fname, buf);
2561 if (rp == NULL)
2562 return;
2563 }
2564 }
2565 }
2566 }
2567
2568 strcpy(saved_jvm_path, buf);
2569 }
2570
2571 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2572 // no prefix required, not even "_"
2573 }
2574
2575 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2576 // no suffix required
2577 }
2578
2579 ////////////////////////////////////////////////////////////////////////////////
2580 // sun.misc.Signal support
2581
2582 static volatile jint sigint_count = 0;
2583
2584 static void
2585 UserHandler(int sig, void *siginfo, void *context) {
2586 // 4511530 - sem_post is serialized and handled by the manager thread. When
2587 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2588 // don't want to flood the manager thread with sem_post requests.
2589 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2590 return;
2591
2592 // Ctrl-C is pressed during error reporting, likely because the error
2593 // handler fails to abort. Let VM die immediately.
2594 if (sig == SIGINT && is_error_reported()) {
2595 os::die();
2596 }
2597
2598 os::signal_notify(sig);
2599 }
2600
2601 void* os::user_handler() {
2602 return CAST_FROM_FN_PTR(void*, UserHandler);
2603 }
2604
2605 extern "C" {
2606 typedef void (*sa_handler_t)(int);
2607 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2608 }
2609
2610 void* os::signal(int signal_number, void* handler) {
2611 struct sigaction sigAct, oldSigAct;
2612
2613 sigfillset(&(sigAct.sa_mask));
2614 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2615 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2616
2617 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2618 // -1 means registration failed
2619 return (void *)-1;
2620 }
2621
2622 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2623 }
2624
2625 void os::signal_raise(int signal_number) {
2626 ::raise(signal_number);
2627 }
2628
2629 /*
2630 * The following code is moved from os.cpp for making this
2631 * code platform specific, which it is by its very nature.
2632 */
2633
2634 // Will be modified when max signal is changed to be dynamic
2635 int os::sigexitnum_pd() {
2636 return NSIG;
2637 }
2638
2639 // a counter for each possible signal value
2640 static volatile jint pending_signals[NSIG+1] = { 0 };
2641
2642 // Bsd(POSIX) specific hand shaking semaphore.
2643 #ifdef __APPLE__
2644 static semaphore_t sig_sem;
2645 #define SEM_INIT(sem, value) semaphore_create(mach_task_self(), &sem, SYNC_POLICY_FIFO, value)
2646 #define SEM_WAIT(sem) semaphore_wait(sem);
2647 #define SEM_POST(sem) semaphore_signal(sem);
2648 #else
2649 static sem_t sig_sem;
2650 #define SEM_INIT(sem, value) sem_init(&sem, 0, value)
2651 #define SEM_WAIT(sem) sem_wait(&sem);
2652 #define SEM_POST(sem) sem_post(&sem);
2653 #endif
2654
2655 void os::signal_init_pd() {
2656 // Initialize signal structures
2657 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2658
2659 // Initialize signal semaphore
2660 ::SEM_INIT(sig_sem, 0);
2661 }
2662
2663 void os::signal_notify(int sig) {
2664 Atomic::inc(&pending_signals[sig]);
2665 ::SEM_POST(sig_sem);
2666 }
2667
2668 static int check_pending_signals(bool wait) {
2669 Atomic::store(0, &sigint_count);
2670 for (;;) {
2671 for (int i = 0; i < NSIG + 1; i++) {
2672 jint n = pending_signals[i];
2673 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2674 return i;
2675 }
2676 }
2677 if (!wait) {
2678 return -1;
2679 }
2680 JavaThread *thread = JavaThread::current();
2681 ThreadBlockInVM tbivm(thread);
2682
2683 bool threadIsSuspended;
2684 do {
2685 thread->set_suspend_equivalent();
2686 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2687 ::SEM_WAIT(sig_sem);
2688
2689 // were we externally suspended while we were waiting?
2690 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2691 if (threadIsSuspended) {
2692 //
2693 // The semaphore has been incremented, but while we were waiting
2694 // another thread suspended us. We don't want to continue running
2695 // while suspended because that would surprise the thread that
2696 // suspended us.
2697 //
2698 ::SEM_POST(sig_sem);
2699
2700 thread->java_suspend_self();
2701 }
2702 } while (threadIsSuspended);
2703 }
2704 }
2705
2706 int os::signal_lookup() {
2707 return check_pending_signals(false);
2708 }
2709
2710 int os::signal_wait() {
2711 return check_pending_signals(true);
2712 }
2713
2714 ////////////////////////////////////////////////////////////////////////////////
2715 // Virtual Memory
2716
2717 int os::vm_page_size() {
2718 // Seems redundant as all get out
2719 assert(os::Bsd::page_size() != -1, "must call os::init");
2720 return os::Bsd::page_size();
2721 }
2722
2723 // Solaris allocates memory by pages.
2724 int os::vm_allocation_granularity() {
2725 assert(os::Bsd::page_size() != -1, "must call os::init");
2726 return os::Bsd::page_size();
2727 }
2728
2729 // Rationale behind this function:
2730 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2731 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2732 // samples for JITted code. Here we create private executable mapping over the code cache
2733 // and then we can use standard (well, almost, as mapping can change) way to provide
2734 // info for the reporting script by storing timestamp and location of symbol
2735 void bsd_wrap_code(char* base, size_t size) {
2736 static volatile jint cnt = 0;
2737
2738 if (!UseOprofile) {
2739 return;
2740 }
2741
2742 char buf[PATH_MAX + 1];
2743 int num = Atomic::add(1, &cnt);
2744
2745 snprintf(buf, PATH_MAX + 1, "%s/hs-vm-%d-%d",
2746 os::get_temp_directory(), os::current_process_id(), num);
2747 unlink(buf);
2748
2749 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2750
2751 if (fd != -1) {
2752 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2753 if (rv != (off_t)-1) {
2754 if (::write(fd, "", 1) == 1) {
2755 mmap(base, size,
2756 PROT_READ|PROT_WRITE|PROT_EXEC,
2757 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2758 }
2759 }
2760 ::close(fd);
2761 unlink(buf);
2762 }
2763 }
2764
2765 // NOTE: Bsd kernel does not really reserve the pages for us.
2766 // All it does is to check if there are enough free pages
2767 // left at the time of mmap(). This could be a potential
2768 // problem.
2769 bool os::commit_memory(char* addr, size_t size, bool exec) {
2770 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2771 #ifdef __OpenBSD__
2772 // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
2773 return ::mprotect(addr, size, prot) == 0;
2774 #else
2775 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2776 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2777 return res != (uintptr_t) MAP_FAILED;
2778 #endif
2779 }
2780
2781 #ifndef _ALLBSD_SOURCE
2782 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2783 #ifndef MAP_HUGETLB
2784 #define MAP_HUGETLB 0x40000
2785 #endif
2786
2787 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2788 #ifndef MADV_HUGEPAGE
2789 #define MADV_HUGEPAGE 14
2790 #endif
2791 #endif
2792
2793 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2794 bool exec) {
2795 #ifndef _ALLBSD_SOURCE
2796 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2797 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2798 uintptr_t res =
2799 (uintptr_t) ::mmap(addr, size, prot,
2800 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2801 -1, 0);
2802 return res != (uintptr_t) MAP_FAILED;
2803 }
2804 #endif
2805
2806 return commit_memory(addr, size, exec);
2807 }
2808
2809 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2810 #ifndef _ALLBSD_SOURCE
2811 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2812 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2813 // be supported or the memory may already be backed by huge pages.
2814 ::madvise(addr, bytes, MADV_HUGEPAGE);
2815 }
2816 #endif
2817 }
2818
2819 void os::free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2820 ::madvise(addr, bytes, MADV_DONTNEED);
2821 }
2822
2823 void os::numa_make_global(char *addr, size_t bytes) {
2824 }
2825
2826 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2827 }
2828
2829 bool os::numa_topology_changed() { return false; }
2830
2831 size_t os::numa_get_groups_num() {
2832 return 1;
2833 }
2834
2835 int os::numa_get_group_id() {
2836 return 0;
2837 }
2838
2839 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2840 if (size > 0) {
2841 ids[0] = 0;
2842 return 1;
2843 }
2844 return 0;
2845 }
2846
2847 bool os::get_page_info(char *start, page_info* info) {
2848 return false;
2849 }
2850
2851 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2852 return end;
2853 }
2854
2855 #ifndef _ALLBSD_SOURCE
2856 // Something to do with the numa-aware allocator needs these symbols
2857 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2858 extern "C" JNIEXPORT void numa_error(char *where) { }
2859 extern "C" JNIEXPORT int fork1() { return fork(); }
2860
2861
2862 // If we are running with libnuma version > 2, then we should
2863 // be trying to use symbols with versions 1.1
2864 // If we are running with earlier version, which did not have symbol versions,
2865 // we should use the base version.
2866 void* os::Bsd::libnuma_dlsym(void* handle, const char *name) {
2867 void *f = dlvsym(handle, name, "libnuma_1.1");
2868 if (f == NULL) {
2869 f = dlsym(handle, name);
2870 }
2871 return f;
2872 }
2873
2874 bool os::Bsd::libnuma_init() {
2875 // sched_getcpu() should be in libc.
2876 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2877 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2878
2879 if (sched_getcpu() != -1) { // Does it work?
2880 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2881 if (handle != NULL) {
2882 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2883 libnuma_dlsym(handle, "numa_node_to_cpus")));
2884 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2885 libnuma_dlsym(handle, "numa_max_node")));
2886 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2887 libnuma_dlsym(handle, "numa_available")));
2888 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2889 libnuma_dlsym(handle, "numa_tonode_memory")));
2890 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2891 libnuma_dlsym(handle, "numa_interleave_memory")));
2892
2893
2894 if (numa_available() != -1) {
2895 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2896 // Create a cpu -> node mapping
2897 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2898 rebuild_cpu_to_node_map();
2899 return true;
2900 }
2901 }
2902 }
2903 return false;
2904 }
2905
2906 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2907 // The table is later used in get_node_by_cpu().
2908 void os::Bsd::rebuild_cpu_to_node_map() {
2909 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2910 // in libnuma (possible values are starting from 16,
2911 // and continuing up with every other power of 2, but less
2912 // than the maximum number of CPUs supported by kernel), and
2913 // is a subject to change (in libnuma version 2 the requirements
2914 // are more reasonable) we'll just hardcode the number they use
2915 // in the library.
2916 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2917
2918 size_t cpu_num = os::active_processor_count();
2919 size_t cpu_map_size = NCPUS / BitsPerCLong;
2920 size_t cpu_map_valid_size =
2921 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2922
2923 cpu_to_node()->clear();
2924 cpu_to_node()->at_grow(cpu_num - 1);
2925 size_t node_num = numa_get_groups_num();
2926
2927 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2928 for (size_t i = 0; i < node_num; i++) {
2929 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2930 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2931 if (cpu_map[j] != 0) {
2932 for (size_t k = 0; k < BitsPerCLong; k++) {
2933 if (cpu_map[j] & (1UL << k)) {
2934 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2935 }
2936 }
2937 }
2938 }
2939 }
2940 }
2941 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2942 }
2943
2944 int os::Bsd::get_node_by_cpu(int cpu_id) {
2945 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2946 return cpu_to_node()->at(cpu_id);
2947 }
2948 return -1;
2949 }
2950
2951 GrowableArray<int>* os::Bsd::_cpu_to_node;
2952 os::Bsd::sched_getcpu_func_t os::Bsd::_sched_getcpu;
2953 os::Bsd::numa_node_to_cpus_func_t os::Bsd::_numa_node_to_cpus;
2954 os::Bsd::numa_max_node_func_t os::Bsd::_numa_max_node;
2955 os::Bsd::numa_available_func_t os::Bsd::_numa_available;
2956 os::Bsd::numa_tonode_memory_func_t os::Bsd::_numa_tonode_memory;
2957 os::Bsd::numa_interleave_memory_func_t os::Bsd::_numa_interleave_memory;
2958 unsigned long* os::Bsd::_numa_all_nodes;
2959 #endif
2960
2961 bool os::uncommit_memory(char* addr, size_t size) {
2962 #ifdef __OpenBSD__
2963 // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
2964 return ::mprotect(addr, size, PROT_NONE) == 0;
2965 #else
2966 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2967 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2968 return res != (uintptr_t) MAP_FAILED;
2969 #endif
2970 }
2971
2972 bool os::create_stack_guard_pages(char* addr, size_t size) {
2973 return os::commit_memory(addr, size);
2974 }
2975
2976 // If this is a growable mapping, remove the guard pages entirely by
2977 // munmap()ping them. If not, just call uncommit_memory().
2978 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2979 return os::uncommit_memory(addr, size);
2980 }
2981
2982 static address _highest_vm_reserved_address = NULL;
2983
2984 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2985 // at 'requested_addr'. If there are existing memory mappings at the same
2986 // location, however, they will be overwritten. If 'fixed' is false,
2987 // 'requested_addr' is only treated as a hint, the return value may or
2988 // may not start from the requested address. Unlike Bsd mmap(), this
2989 // function returns NULL to indicate failure.
2990 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2991 char * addr;
2992 int flags;
2993
2994 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2995 if (fixed) {
2996 assert((uintptr_t)requested_addr % os::Bsd::page_size() == 0, "unaligned address");
2997 flags |= MAP_FIXED;
2998 }
2999
3000 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
3001 // to PROT_EXEC if executable when we commit the page.
3002 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
3003 flags, -1, 0);
3004
3005 if (addr != MAP_FAILED) {
3006 // anon_mmap() should only get called during VM initialization,
3007 // don't need lock (actually we can skip locking even it can be called
3008 // from multiple threads, because _highest_vm_reserved_address is just a
3009 // hint about the upper limit of non-stack memory regions.)
3010 if ((address)addr + bytes > _highest_vm_reserved_address) {
3011 _highest_vm_reserved_address = (address)addr + bytes;
3012 }
3013 }
3014
3015 return addr == MAP_FAILED ? NULL : addr;
3016 }
3017
3018 // Don't update _highest_vm_reserved_address, because there might be memory
3019 // regions above addr + size. If so, releasing a memory region only creates
3020 // a hole in the address space, it doesn't help prevent heap-stack collision.
3021 //
3022 static int anon_munmap(char * addr, size_t size) {
3023 return ::munmap(addr, size) == 0;
3024 }
3025
3026 char* os::reserve_memory(size_t bytes, char* requested_addr,
3027 size_t alignment_hint) {
3028 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3029 }
3030
3031 bool os::release_memory(char* addr, size_t size) {
3032 return anon_munmap(addr, size);
3033 }
3034
3035 static address highest_vm_reserved_address() {
3036 return _highest_vm_reserved_address;
3037 }
3038
3039 static bool bsd_mprotect(char* addr, size_t size, int prot) {
3040 // Bsd wants the mprotect address argument to be page aligned.
3041 char* bottom = (char*)align_size_down((intptr_t)addr, os::Bsd::page_size());
3042
3043 // According to SUSv3, mprotect() should only be used with mappings
3044 // established by mmap(), and mmap() always maps whole pages. Unaligned
3045 // 'addr' likely indicates problem in the VM (e.g. trying to change
3046 // protection of malloc'ed or statically allocated memory). Check the
3047 // caller if you hit this assert.
3048 assert(addr == bottom, "sanity check");
3049
3050 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Bsd::page_size());
3051 return ::mprotect(bottom, size, prot) == 0;
3052 }
3053
3054 // Set protections specified
3055 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3056 bool is_committed) {
3057 unsigned int p = 0;
3058 switch (prot) {
3059 case MEM_PROT_NONE: p = PROT_NONE; break;
3060 case MEM_PROT_READ: p = PROT_READ; break;
3061 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3062 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3063 default:
3064 ShouldNotReachHere();
3065 }
3066 // is_committed is unused.
3067 return bsd_mprotect(addr, bytes, p);
3068 }
3069
3070 bool os::guard_memory(char* addr, size_t size) {
3071 return bsd_mprotect(addr, size, PROT_NONE);
3072 }
3073
3074 bool os::unguard_memory(char* addr, size_t size) {
3075 return bsd_mprotect(addr, size, PROT_READ|PROT_WRITE);
3076 }
3077
3078 bool os::Bsd::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3079 bool result = false;
3080 #ifndef _ALLBSD_SOURCE
3081 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
3082 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3083 -1, 0);
3084
3085 if (p != (void *) -1) {
3086 // We don't know if this really is a huge page or not.
3087 FILE *fp = fopen("/proc/self/maps", "r");
3088 if (fp) {
3089 while (!feof(fp)) {
3090 char chars[257];
3091 long x = 0;
3092 if (fgets(chars, sizeof(chars), fp)) {
3093 if (sscanf(chars, "%lx-%*x", &x) == 1
3094 && x == (long)p) {
3095 if (strstr (chars, "hugepage")) {
3096 result = true;
3097 break;
3098 }
3099 }
3100 }
3101 }
3102 fclose(fp);
3103 }
3104 munmap (p, page_size);
3105 if (result)
3106 return true;
3107 }
3108
3109 if (warn) {
3110 warning("HugeTLBFS is not supported by the operating system.");
3111 }
3112 #endif
3113
3114 return result;
3115 }
3116
3117 /*
3118 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3119 *
3120 * From the coredump_filter documentation:
3121 *
3122 * - (bit 0) anonymous private memory
3123 * - (bit 1) anonymous shared memory
3124 * - (bit 2) file-backed private memory
3125 * - (bit 3) file-backed shared memory
3126 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3127 * effective only if the bit 2 is cleared)
3128 * - (bit 5) hugetlb private memory
3129 * - (bit 6) hugetlb shared memory
3130 */
3131 static void set_coredump_filter(void) {
3132 FILE *f;
3133 long cdm;
3134
3135 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3136 return;
3137 }
3138
3139 if (fscanf(f, "%lx", &cdm) != 1) {
3140 fclose(f);
3141 return;
3142 }
3143
3144 rewind(f);
3145
3146 if ((cdm & LARGEPAGES_BIT) == 0) {
3147 cdm |= LARGEPAGES_BIT;
3148 fprintf(f, "%#lx", cdm);
3149 }
3150
3151 fclose(f);
3152 }
3153
3154 // Large page support
3155
3156 static size_t _large_page_size = 0;
3157
3158 void os::large_page_init() {
3159 #ifndef _ALLBSD_SOURCE
3160 if (!UseLargePages) {
3161 UseHugeTLBFS = false;
3162 UseSHM = false;
3163 return;
3164 }
3165
3166 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3167 // If UseLargePages is specified on the command line try both methods,
3168 // if it's default, then try only HugeTLBFS.
3169 if (FLAG_IS_DEFAULT(UseLargePages)) {
3170 UseHugeTLBFS = true;
3171 } else {
3172 UseHugeTLBFS = UseSHM = true;
3173 }
3174 }
3175
3176 if (LargePageSizeInBytes) {
3177 _large_page_size = LargePageSizeInBytes;
3178 } else {
3179 // large_page_size on Bsd is used to round up heap size. x86 uses either
3180 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3181 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3182 // page as large as 256M.
3183 //
3184 // Here we try to figure out page size by parsing /proc/meminfo and looking
3185 // for a line with the following format:
3186 // Hugepagesize: 2048 kB
3187 //
3188 // If we can't determine the value (e.g. /proc is not mounted, or the text
3189 // format has been changed), we'll use the largest page size supported by
3190 // the processor.
3191
3192 #ifndef ZERO
3193 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3194 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3195 #endif // ZERO
3196
3197 FILE *fp = fopen("/proc/meminfo", "r");
3198 if (fp) {
3199 while (!feof(fp)) {
3200 int x = 0;
3201 char buf[16];
3202 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3203 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3204 _large_page_size = x * K;
3205 break;
3206 }
3207 } else {
3208 // skip to next line
3209 for (;;) {
3210 int ch = fgetc(fp);
3211 if (ch == EOF || ch == (int)'\n') break;
3212 }
3213 }
3214 }
3215 fclose(fp);
3216 }
3217 }
3218
3219 // print a warning if any large page related flag is specified on command line
3220 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3221
3222 const size_t default_page_size = (size_t)Bsd::page_size();
3223 if (_large_page_size > default_page_size) {
3224 _page_sizes[0] = _large_page_size;
3225 _page_sizes[1] = default_page_size;
3226 _page_sizes[2] = 0;
3227 }
3228 UseHugeTLBFS = UseHugeTLBFS &&
3229 Bsd::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3230
3231 if (UseHugeTLBFS)
3232 UseSHM = false;
3233
3234 UseLargePages = UseHugeTLBFS || UseSHM;
3235
3236 set_coredump_filter();
3237 #endif
3238 }
3239
3240 #ifndef _ALLBSD_SOURCE
3241 #ifndef SHM_HUGETLB
3242 #define SHM_HUGETLB 04000
3243 #endif
3244 #endif
3245
3246 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3247 // "exec" is passed in but not used. Creating the shared image for
3248 // the code cache doesn't have an SHM_X executable permission to check.
3249 assert(UseLargePages && UseSHM, "only for SHM large pages");
3250
3251 key_t key = IPC_PRIVATE;
3252 char *addr;
3253
3254 bool warn_on_failure = UseLargePages &&
3255 (!FLAG_IS_DEFAULT(UseLargePages) ||
3256 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3257 );
3258 char msg[128];
3259
3260 // Create a large shared memory region to attach to based on size.
3261 // Currently, size is the total size of the heap
3262 #ifndef _ALLBSD_SOURCE
3263 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3264 #else
3265 int shmid = shmget(key, bytes, IPC_CREAT|SHM_R|SHM_W);
3266 #endif
3267 if (shmid == -1) {
3268 // Possible reasons for shmget failure:
3269 // 1. shmmax is too small for Java heap.
3270 // > check shmmax value: cat /proc/sys/kernel/shmmax
3271 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3272 // 2. not enough large page memory.
3273 // > check available large pages: cat /proc/meminfo
3274 // > increase amount of large pages:
3275 // echo new_value > /proc/sys/vm/nr_hugepages
3276 // Note 1: different Bsd may use different name for this property,
3277 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3278 // Note 2: it's possible there's enough physical memory available but
3279 // they are so fragmented after a long run that they can't
3280 // coalesce into large pages. Try to reserve large pages when
3281 // the system is still "fresh".
3282 if (warn_on_failure) {
3283 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3284 warning(msg);
3285 }
3286 return NULL;
3287 }
3288
3289 // attach to the region
3290 addr = (char*)shmat(shmid, req_addr, 0);
3291 int err = errno;
3292
3293 // Remove shmid. If shmat() is successful, the actual shared memory segment
3294 // will be deleted when it's detached by shmdt() or when the process
3295 // terminates. If shmat() is not successful this will remove the shared
3296 // segment immediately.
3297 shmctl(shmid, IPC_RMID, NULL);
3298
3299 if ((intptr_t)addr == -1) {
3300 if (warn_on_failure) {
3301 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3302 warning(msg);
3303 }
3304 return NULL;
3305 }
3306
3307 return addr;
3308 }
3309
3310 bool os::release_memory_special(char* base, size_t bytes) {
3311 // detaching the SHM segment will also delete it, see reserve_memory_special()
3312 int rslt = shmdt(base);
3313 return rslt == 0;
3314 }
3315
3316 size_t os::large_page_size() {
3317 return _large_page_size;
3318 }
3319
3320 // HugeTLBFS allows application to commit large page memory on demand;
3321 // with SysV SHM the entire memory region must be allocated as shared
3322 // memory.
3323 bool os::can_commit_large_page_memory() {
3324 return UseHugeTLBFS;
3325 }
3326
3327 bool os::can_execute_large_page_memory() {
3328 return UseHugeTLBFS;
3329 }
3330
3331 // Reserve memory at an arbitrary address, only if that area is
3332 // available (and not reserved for something else).
3333
3334 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3335 const int max_tries = 10;
3336 char* base[max_tries];
3337 size_t size[max_tries];
3338 const size_t gap = 0x000000;
3339
3340 // Assert only that the size is a multiple of the page size, since
3341 // that's all that mmap requires, and since that's all we really know
3342 // about at this low abstraction level. If we need higher alignment,
3343 // we can either pass an alignment to this method or verify alignment
3344 // in one of the methods further up the call chain. See bug 5044738.
3345 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3346
3347 // Repeatedly allocate blocks until the block is allocated at the
3348 // right spot. Give up after max_tries. Note that reserve_memory() will
3349 // automatically update _highest_vm_reserved_address if the call is
3350 // successful. The variable tracks the highest memory address every reserved
3351 // by JVM. It is used to detect heap-stack collision if running with
3352 // fixed-stack BsdThreads. Because here we may attempt to reserve more
3353 // space than needed, it could confuse the collision detecting code. To
3354 // solve the problem, save current _highest_vm_reserved_address and
3355 // calculate the correct value before return.
3356 address old_highest = _highest_vm_reserved_address;
3357
3358 // Bsd mmap allows caller to pass an address as hint; give it a try first,
3359 // if kernel honors the hint then we can return immediately.
3360 char * addr = anon_mmap(requested_addr, bytes, false);
3361 if (addr == requested_addr) {
3362 return requested_addr;
3363 }
3364
3365 if (addr != NULL) {
3366 // mmap() is successful but it fails to reserve at the requested address
3367 anon_munmap(addr, bytes);
3368 }
3369
3370 int i;
3371 for (i = 0; i < max_tries; ++i) {
3372 base[i] = reserve_memory(bytes);
3373
3374 if (base[i] != NULL) {
3375 // Is this the block we wanted?
3376 if (base[i] == requested_addr) {
3377 size[i] = bytes;
3378 break;
3379 }
3380
3381 // Does this overlap the block we wanted? Give back the overlapped
3382 // parts and try again.
3383
3384 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3385 if (top_overlap >= 0 && top_overlap < bytes) {
3386 unmap_memory(base[i], top_overlap);
3387 base[i] += top_overlap;
3388 size[i] = bytes - top_overlap;
3389 } else {
3390 size_t bottom_overlap = base[i] + bytes - requested_addr;
3391 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3392 unmap_memory(requested_addr, bottom_overlap);
3393 size[i] = bytes - bottom_overlap;
3394 } else {
3395 size[i] = bytes;
3396 }
3397 }
3398 }
3399 }
3400
3401 // Give back the unused reserved pieces.
3402
3403 for (int j = 0; j < i; ++j) {
3404 if (base[j] != NULL) {
3405 unmap_memory(base[j], size[j]);
3406 }
3407 }
3408
3409 if (i < max_tries) {
3410 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3411 return requested_addr;
3412 } else {
3413 _highest_vm_reserved_address = old_highest;
3414 return NULL;
3415 }
3416 }
3417
3418 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3419 RESTARTABLE_RETURN_INT(::read(fd, buf, nBytes));
3420 }
3421
3422 // TODO-FIXME: reconcile Solaris' os::sleep with the bsd variation.
3423 // Solaris uses poll(), bsd uses park().
3424 // Poll() is likely a better choice, assuming that Thread.interrupt()
3425 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3426 // SIGSEGV, see 4355769.
3427
3428 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3429 assert(thread == Thread::current(), "thread consistency check");
3430
3431 ParkEvent * const slp = thread->_SleepEvent ;
3432 slp->reset() ;
3433 OrderAccess::fence() ;
3434
3435 if (interruptible) {
3436 jlong prevtime = javaTimeNanos();
3437
3438 for (;;) {
3439 if (os::is_interrupted(thread, true)) {
3440 return OS_INTRPT;
3441 }
3442
3443 jlong newtime = javaTimeNanos();
3444
3445 if (newtime - prevtime < 0) {
3446 // time moving backwards, should only happen if no monotonic clock
3447 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3448 assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
3449 } else {
3450 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3451 }
3452
3453 if(millis <= 0) {
3454 return OS_OK;
3455 }
3456
3457 prevtime = newtime;
3458
3459 {
3460 assert(thread->is_Java_thread(), "sanity check");
3461 JavaThread *jt = (JavaThread *) thread;
3462 ThreadBlockInVM tbivm(jt);
3463 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3464
3465 jt->set_suspend_equivalent();
3466 // cleared by handle_special_suspend_equivalent_condition() or
3467 // java_suspend_self() via check_and_wait_while_suspended()
3468
3469 slp->park(millis);
3470
3471 // were we externally suspended while we were waiting?
3472 jt->check_and_wait_while_suspended();
3473 }
3474 }
3475 } else {
3476 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3477 jlong prevtime = javaTimeNanos();
3478
3479 for (;;) {
3480 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3481 // the 1st iteration ...
3482 jlong newtime = javaTimeNanos();
3483
3484 if (newtime - prevtime < 0) {
3485 // time moving backwards, should only happen if no monotonic clock
3486 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3487 assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
3488 } else {
3489 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3490 }
3491
3492 if(millis <= 0) break ;
3493
3494 prevtime = newtime;
3495 slp->park(millis);
3496 }
3497 return OS_OK ;
3498 }
3499 }
3500
3501 int os::naked_sleep() {
3502 // %% make the sleep time an integer flag. for now use 1 millisec.
3503 return os::sleep(Thread::current(), 1, false);
3504 }
3505
3506 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3507 void os::infinite_sleep() {
3508 while (true) { // sleep forever ...
3509 ::sleep(100); // ... 100 seconds at a time
3510 }
3511 }
3512
3513 // Used to convert frequent JVM_Yield() to nops
3514 bool os::dont_yield() {
3515 return DontYieldALot;
3516 }
3517
3518 void os::yield() {
3519 sched_yield();
3520 }
3521
3522 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3523
3524 void os::yield_all(int attempts) {
3525 // Yields to all threads, including threads with lower priorities
3526 // Threads on Bsd are all with same priority. The Solaris style
3527 // os::yield_all() with nanosleep(1ms) is not necessary.
3528 sched_yield();
3529 }
3530
3531 // Called from the tight loops to possibly influence time-sharing heuristics
3532 void os::loop_breaker(int attempts) {
3533 os::yield_all(attempts);
3534 }
3535
3536 ////////////////////////////////////////////////////////////////////////////////
3537 // thread priority support
3538
3539 // Note: Normal Bsd applications are run with SCHED_OTHER policy. SCHED_OTHER
3540 // only supports dynamic priority, static priority must be zero. For real-time
3541 // applications, Bsd supports SCHED_RR which allows static priority (1-99).
3542 // However, for large multi-threaded applications, SCHED_RR is not only slower
3543 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3544 // of 5 runs - Sep 2005).
3545 //
3546 // The following code actually changes the niceness of kernel-thread/LWP. It
3547 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3548 // not the entire user process, and user level threads are 1:1 mapped to kernel
3549 // threads. It has always been the case, but could change in the future. For
3550 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3551 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3552
3553 #if defined(_ALLBSD_SOURCE) && !defined(__APPLE__)
3554 int os::java_to_os_priority[CriticalPriority + 1] = {
3555 19, // 0 Entry should never be used
3556
3557 0, // 1 MinPriority
3558 3, // 2
3559 6, // 3
3560
3561 10, // 4
3562 15, // 5 NormPriority
3563 18, // 6
3564
3565 21, // 7
3566 25, // 8
3567 28, // 9 NearMaxPriority
3568
3569 31, // 10 MaxPriority
3570
3571 31 // 11 CriticalPriority
3572 };
3573 #elif defined(__APPLE__)
3574 /* Using Mach high-level priority assignments */
3575 int os::java_to_os_priority[CriticalPriority + 1] = {
3576 0, // 0 Entry should never be used (MINPRI_USER)
3577
3578 27, // 1 MinPriority
3579 28, // 2
3580 29, // 3
3581
3582 30, // 4
3583 31, // 5 NormPriority (BASEPRI_DEFAULT)
3584 32, // 6
3585
3586 33, // 7
3587 34, // 8
3588 35, // 9 NearMaxPriority
3589
3590 36, // 10 MaxPriority
3591
3592 36 // 11 CriticalPriority
3593 };
3594 #else
3595 int os::java_to_os_priority[CriticalPriority + 1] = {
3596 19, // 0 Entry should never be used
3597
3598 4, // 1 MinPriority
3599 3, // 2
3600 2, // 3
3601
3602 1, // 4
3603 0, // 5 NormPriority
3604 -1, // 6
3605
3606 -2, // 7
3607 -3, // 8
3608 -4, // 9 NearMaxPriority
3609
3610 -5, // 10 MaxPriority
3611
3612 -5 // 11 CriticalPriority
3613 };
3614 #endif
3615
3616 static int prio_init() {
3617 if (ThreadPriorityPolicy == 1) {
3618 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3619 // if effective uid is not root. Perhaps, a more elegant way of doing
3620 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3621 if (geteuid() != 0) {
3622 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3623 warning("-XX:ThreadPriorityPolicy requires root privilege on Bsd");
3624 }
3625 ThreadPriorityPolicy = 0;
3626 }
3627 }
3628 if (UseCriticalJavaThreadPriority) {
3629 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3630 }
3631 return 0;
3632 }
3633
3634 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3635 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3636
3637 #ifdef __OpenBSD__
3638 // OpenBSD pthread_setprio starves low priority threads
3639 return OS_OK;
3640 #elif defined(__FreeBSD__)
3641 int ret = pthread_setprio(thread->osthread()->pthread_id(), newpri);
3642 #elif defined(__APPLE__) || defined(__NetBSD__)
3643 struct sched_param sp;
3644 int policy;
3645 pthread_t self = pthread_self();
3646
3647 if (pthread_getschedparam(self, &policy, &sp) != 0)
3648 return OS_ERR;
3649
3650 sp.sched_priority = newpri;
3651 if (pthread_setschedparam(self, policy, &sp) != 0)
3652 return OS_ERR;
3653
3654 return OS_OK;
3655 #else
3656 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3657 return (ret == 0) ? OS_OK : OS_ERR;
3658 #endif
3659 }
3660
3661 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3662 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3663 *priority_ptr = java_to_os_priority[NormPriority];
3664 return OS_OK;
3665 }
3666
3667 errno = 0;
3668 #if defined(__OpenBSD__) || defined(__FreeBSD__)
3669 *priority_ptr = pthread_getprio(thread->osthread()->pthread_id());
3670 #elif defined(__APPLE__) || defined(__NetBSD__)
3671 int policy;
3672 struct sched_param sp;
3673
3674 pthread_getschedparam(pthread_self(), &policy, &sp);
3675 *priority_ptr = sp.sched_priority;
3676 #else
3677 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3678 #endif
3679 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3680 }
3681
3682 // Hint to the underlying OS that a task switch would not be good.
3683 // Void return because it's a hint and can fail.
3684 void os::hint_no_preempt() {}
3685
3686 ////////////////////////////////////////////////////////////////////////////////
3687 // suspend/resume support
3688
3689 // the low-level signal-based suspend/resume support is a remnant from the
3690 // old VM-suspension that used to be for java-suspension, safepoints etc,
3691 // within hotspot. Now there is a single use-case for this:
3692 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3693 // that runs in the watcher thread.
3694 // The remaining code is greatly simplified from the more general suspension
3695 // code that used to be used.
3696 //
3697 // The protocol is quite simple:
3698 // - suspend:
3699 // - sends a signal to the target thread
3700 // - polls the suspend state of the osthread using a yield loop
3701 // - target thread signal handler (SR_handler) sets suspend state
3702 // and blocks in sigsuspend until continued
3703 // - resume:
3704 // - sets target osthread state to continue
3705 // - sends signal to end the sigsuspend loop in the SR_handler
3706 //
3707 // Note that the SR_lock plays no role in this suspend/resume protocol.
3708 //
3709
3710 static void resume_clear_context(OSThread *osthread) {
3711 osthread->set_ucontext(NULL);
3712 osthread->set_siginfo(NULL);
3713
3714 // notify the suspend action is completed, we have now resumed
3715 osthread->sr.clear_suspended();
3716 }
3717
3718 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3719 osthread->set_ucontext(context);
3720 osthread->set_siginfo(siginfo);
3721 }
3722
3723 //
3724 // Handler function invoked when a thread's execution is suspended or
3725 // resumed. We have to be careful that only async-safe functions are
3726 // called here (Note: most pthread functions are not async safe and
3727 // should be avoided.)
3728 //
3729 // Note: sigwait() is a more natural fit than sigsuspend() from an
3730 // interface point of view, but sigwait() prevents the signal hander
3731 // from being run. libpthread would get very confused by not having
3732 // its signal handlers run and prevents sigwait()'s use with the
3733 // mutex granting granting signal.
3734 //
3735 // Currently only ever called on the VMThread
3736 //
3737 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3738 // Save and restore errno to avoid confusing native code with EINTR
3739 // after sigsuspend.
3740 int old_errno = errno;
3741
3742 Thread* thread = Thread::current();
3743 OSThread* osthread = thread->osthread();
3744 assert(thread->is_VM_thread(), "Must be VMThread");
3745 // read current suspend action
3746 int action = osthread->sr.suspend_action();
3747 if (action == SR_SUSPEND) {
3748 suspend_save_context(osthread, siginfo, context);
3749
3750 // Notify the suspend action is about to be completed. do_suspend()
3751 // waits until SR_SUSPENDED is set and then returns. We will wait
3752 // here for a resume signal and that completes the suspend-other
3753 // action. do_suspend/do_resume is always called as a pair from
3754 // the same thread - so there are no races
3755
3756 // notify the caller
3757 osthread->sr.set_suspended();
3758
3759 sigset_t suspend_set; // signals for sigsuspend()
3760
3761 // get current set of blocked signals and unblock resume signal
3762 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3763 sigdelset(&suspend_set, SR_signum);
3764
3765 // wait here until we are resumed
3766 do {
3767 sigsuspend(&suspend_set);
3768 // ignore all returns until we get a resume signal
3769 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3770
3771 resume_clear_context(osthread);
3772
3773 } else {
3774 assert(action == SR_CONTINUE, "unexpected sr action");
3775 // nothing special to do - just leave the handler
3776 }
3777
3778 errno = old_errno;
3779 }
3780
3781
3782 static int SR_initialize() {
3783 struct sigaction act;
3784 char *s;
3785 /* Get signal number to use for suspend/resume */
3786 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3787 int sig = ::strtol(s, 0, 10);
3788 if (sig > 0 || sig < NSIG) {
3789 SR_signum = sig;
3790 }
3791 }
3792
3793 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3794 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3795
3796 sigemptyset(&SR_sigset);
3797 sigaddset(&SR_sigset, SR_signum);
3798
3799 /* Set up signal handler for suspend/resume */
3800 act.sa_flags = SA_RESTART|SA_SIGINFO;
3801 act.sa_handler = (void (*)(int)) SR_handler;
3802
3803 // SR_signum is blocked by default.
3804 // 4528190 - We also need to block pthread restart signal (32 on all
3805 // supported Bsd platforms). Note that BsdThreads need to block
3806 // this signal for all threads to work properly. So we don't have
3807 // to use hard-coded signal number when setting up the mask.
3808 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3809
3810 if (sigaction(SR_signum, &act, 0) == -1) {
3811 return -1;
3812 }
3813
3814 // Save signal flag
3815 os::Bsd::set_our_sigflags(SR_signum, act.sa_flags);
3816 return 0;
3817 }
3818
3819 static int SR_finalize() {
3820 return 0;
3821 }
3822
3823
3824 // returns true on success and false on error - really an error is fatal
3825 // but this seems the normal response to library errors
3826 static bool do_suspend(OSThread* osthread) {
3827 // mark as suspended and send signal
3828 osthread->sr.set_suspend_action(SR_SUSPEND);
3829 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3830 assert_status(status == 0, status, "pthread_kill");
3831
3832 // check status and wait until notified of suspension
3833 if (status == 0) {
3834 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3835 os::yield_all(i);
3836 }
3837 osthread->sr.set_suspend_action(SR_NONE);
3838 return true;
3839 }
3840 else {
3841 osthread->sr.set_suspend_action(SR_NONE);
3842 return false;
3843 }
3844 }
3845
3846 static void do_resume(OSThread* osthread) {
3847 assert(osthread->sr.is_suspended(), "thread should be suspended");
3848 osthread->sr.set_suspend_action(SR_CONTINUE);
3849
3850 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3851 assert_status(status == 0, status, "pthread_kill");
3852 // check status and wait unit notified of resumption
3853 if (status == 0) {
3854 for (int i = 0; osthread->sr.is_suspended(); i++) {
3855 os::yield_all(i);
3856 }
3857 }
3858 osthread->sr.set_suspend_action(SR_NONE);
3859 }
3860
3861 ////////////////////////////////////////////////////////////////////////////////
3862 // interrupt support
3863
3864 void os::interrupt(Thread* thread) {
3865 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3866 "possibility of dangling Thread pointer");
3867
3868 OSThread* osthread = thread->osthread();
3869
3870 if (!osthread->interrupted()) {
3871 osthread->set_interrupted(true);
3872 // More than one thread can get here with the same value of osthread,
3873 // resulting in multiple notifications. We do, however, want the store
3874 // to interrupted() to be visible to other threads before we execute unpark().
3875 OrderAccess::fence();
3876 ParkEvent * const slp = thread->_SleepEvent ;
3877 if (slp != NULL) slp->unpark() ;
3878 }
3879
3880 // For JSR166. Unpark even if interrupt status already was set
3881 if (thread->is_Java_thread())
3882 ((JavaThread*)thread)->parker()->unpark();
3883
3884 ParkEvent * ev = thread->_ParkEvent ;
3885 if (ev != NULL) ev->unpark() ;
3886
3887 }
3888
3889 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3890 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3891 "possibility of dangling Thread pointer");
3892
3893 OSThread* osthread = thread->osthread();
3894
3895 bool interrupted = osthread->interrupted();
3896
3897 if (interrupted && clear_interrupted) {
3898 osthread->set_interrupted(false);
3899 // consider thread->_SleepEvent->reset() ... optional optimization
3900 }
3901
3902 return interrupted;
3903 }
3904
3905 ///////////////////////////////////////////////////////////////////////////////////
3906 // signal handling (except suspend/resume)
3907
3908 // This routine may be used by user applications as a "hook" to catch signals.
3909 // The user-defined signal handler must pass unrecognized signals to this
3910 // routine, and if it returns true (non-zero), then the signal handler must
3911 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3912 // routine will never retun false (zero), but instead will execute a VM panic
3913 // routine kill the process.
3914 //
3915 // If this routine returns false, it is OK to call it again. This allows
3916 // the user-defined signal handler to perform checks either before or after
3917 // the VM performs its own checks. Naturally, the user code would be making
3918 // a serious error if it tried to handle an exception (such as a null check
3919 // or breakpoint) that the VM was generating for its own correct operation.
3920 //
3921 // This routine may recognize any of the following kinds of signals:
3922 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3923 // It should be consulted by handlers for any of those signals.
3924 //
3925 // The caller of this routine must pass in the three arguments supplied
3926 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3927 // field of the structure passed to sigaction(). This routine assumes that
3928 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3929 //
3930 // Note that the VM will print warnings if it detects conflicting signal
3931 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3932 //
3933 extern "C" JNIEXPORT int
3934 JVM_handle_bsd_signal(int signo, siginfo_t* siginfo,
3935 void* ucontext, int abort_if_unrecognized);
3936
3937 void signalHandler(int sig, siginfo_t* info, void* uc) {
3938 assert(info != NULL && uc != NULL, "it must be old kernel");
3939 JVM_handle_bsd_signal(sig, info, uc, true);
3940 }
3941
3942
3943 // This boolean allows users to forward their own non-matching signals
3944 // to JVM_handle_bsd_signal, harmlessly.
3945 bool os::Bsd::signal_handlers_are_installed = false;
3946
3947 // For signal-chaining
3948 struct sigaction os::Bsd::sigact[MAXSIGNUM];
3949 unsigned int os::Bsd::sigs = 0;
3950 bool os::Bsd::libjsig_is_loaded = false;
3951 typedef struct sigaction *(*get_signal_t)(int);
3952 get_signal_t os::Bsd::get_signal_action = NULL;
3953
3954 struct sigaction* os::Bsd::get_chained_signal_action(int sig) {
3955 struct sigaction *actp = NULL;
3956
3957 if (libjsig_is_loaded) {
3958 // Retrieve the old signal handler from libjsig
3959 actp = (*get_signal_action)(sig);
3960 }
3961 if (actp == NULL) {
3962 // Retrieve the preinstalled signal handler from jvm
3963 actp = get_preinstalled_handler(sig);
3964 }
3965
3966 return actp;
3967 }
3968
3969 static bool call_chained_handler(struct sigaction *actp, int sig,
3970 siginfo_t *siginfo, void *context) {
3971 // Call the old signal handler
3972 if (actp->sa_handler == SIG_DFL) {
3973 // It's more reasonable to let jvm treat it as an unexpected exception
3974 // instead of taking the default action.
3975 return false;
3976 } else if (actp->sa_handler != SIG_IGN) {
3977 if ((actp->sa_flags & SA_NODEFER) == 0) {
3978 // automaticlly block the signal
3979 sigaddset(&(actp->sa_mask), sig);
3980 }
3981
3982 sa_handler_t hand;
3983 sa_sigaction_t sa;
3984 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3985 // retrieve the chained handler
3986 if (siginfo_flag_set) {
3987 sa = actp->sa_sigaction;
3988 } else {
3989 hand = actp->sa_handler;
3990 }
3991
3992 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3993 actp->sa_handler = SIG_DFL;
3994 }
3995
3996 // try to honor the signal mask
3997 sigset_t oset;
3998 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3999
4000 // call into the chained handler
4001 if (siginfo_flag_set) {
4002 (*sa)(sig, siginfo, context);
4003 } else {
4004 (*hand)(sig);
4005 }
4006
4007 // restore the signal mask
4008 pthread_sigmask(SIG_SETMASK, &oset, 0);
4009 }
4010 // Tell jvm's signal handler the signal is taken care of.
4011 return true;
4012 }
4013
4014 bool os::Bsd::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4015 bool chained = false;
4016 // signal-chaining
4017 if (UseSignalChaining) {
4018 struct sigaction *actp = get_chained_signal_action(sig);
4019 if (actp != NULL) {
4020 chained = call_chained_handler(actp, sig, siginfo, context);
4021 }
4022 }
4023 return chained;
4024 }
4025
4026 struct sigaction* os::Bsd::get_preinstalled_handler(int sig) {
4027 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4028 return &sigact[sig];
4029 }
4030 return NULL;
4031 }
4032
4033 void os::Bsd::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4034 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4035 sigact[sig] = oldAct;
4036 sigs |= (unsigned int)1 << sig;
4037 }
4038
4039 // for diagnostic
4040 int os::Bsd::sigflags[MAXSIGNUM];
4041
4042 int os::Bsd::get_our_sigflags(int sig) {
4043 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4044 return sigflags[sig];
4045 }
4046
4047 void os::Bsd::set_our_sigflags(int sig, int flags) {
4048 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4049 sigflags[sig] = flags;
4050 }
4051
4052 void os::Bsd::set_signal_handler(int sig, bool set_installed) {
4053 // Check for overwrite.
4054 struct sigaction oldAct;
4055 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4056
4057 void* oldhand = oldAct.sa_sigaction
4058 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4059 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4060 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4061 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4062 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4063 if (AllowUserSignalHandlers || !set_installed) {
4064 // Do not overwrite; user takes responsibility to forward to us.
4065 return;
4066 } else if (UseSignalChaining) {
4067 // save the old handler in jvm
4068 save_preinstalled_handler(sig, oldAct);
4069 // libjsig also interposes the sigaction() call below and saves the
4070 // old sigaction on it own.
4071 } else {
4072 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4073 "%#lx for signal %d.", (long)oldhand, sig));
4074 }
4075 }
4076
4077 struct sigaction sigAct;
4078 sigfillset(&(sigAct.sa_mask));
4079 sigAct.sa_handler = SIG_DFL;
4080 if (!set_installed) {
4081 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4082 } else {
4083 sigAct.sa_sigaction = signalHandler;
4084 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4085 }
4086 // Save flags, which are set by ours
4087 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4088 sigflags[sig] = sigAct.sa_flags;
4089
4090 int ret = sigaction(sig, &sigAct, &oldAct);
4091 assert(ret == 0, "check");
4092
4093 void* oldhand2 = oldAct.sa_sigaction
4094 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4095 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4096 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4097 }
4098
4099 // install signal handlers for signals that HotSpot needs to
4100 // handle in order to support Java-level exception handling.
4101
4102 void os::Bsd::install_signal_handlers() {
4103 if (!signal_handlers_are_installed) {
4104 signal_handlers_are_installed = true;
4105
4106 // signal-chaining
4107 typedef void (*signal_setting_t)();
4108 signal_setting_t begin_signal_setting = NULL;
4109 signal_setting_t end_signal_setting = NULL;
4110 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4111 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4112 if (begin_signal_setting != NULL) {
4113 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4114 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4115 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4116 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4117 libjsig_is_loaded = true;
4118 assert(UseSignalChaining, "should enable signal-chaining");
4119 }
4120 if (libjsig_is_loaded) {
4121 // Tell libjsig jvm is setting signal handlers
4122 (*begin_signal_setting)();
4123 }
4124
4125 set_signal_handler(SIGSEGV, true);
4126 set_signal_handler(SIGPIPE, true);
4127 set_signal_handler(SIGBUS, true);
4128 set_signal_handler(SIGILL, true);
4129 set_signal_handler(SIGFPE, true);
4130 set_signal_handler(SIGXFSZ, true);
4131
4132 #if defined(__APPLE__)
4133 // In Mac OS X 10.4, CrashReporter will write a crash log for all 'fatal' signals, including
4134 // signals caught and handled by the JVM. To work around this, we reset the mach task
4135 // signal handler that's placed on our process by CrashReporter. This disables
4136 // CrashReporter-based reporting.
4137 //
4138 // This work-around is not necessary for 10.5+, as CrashReporter no longer intercedes
4139 // on caught fatal signals.
4140 //
4141 // Additionally, gdb installs both standard BSD signal handlers, and mach exception
4142 // handlers. By replacing the existing task exception handler, we disable gdb's mach
4143 // exception handling, while leaving the standard BSD signal handlers functional.
4144 kern_return_t kr;
4145 kr = task_set_exception_ports(mach_task_self(),
4146 EXC_MASK_BAD_ACCESS | EXC_MASK_ARITHMETIC,
4147 MACH_PORT_NULL,
4148 EXCEPTION_STATE_IDENTITY,
4149 MACHINE_THREAD_STATE);
4150
4151 assert(kr == KERN_SUCCESS, "could not set mach task signal handler");
4152 #endif
4153
4154 if (libjsig_is_loaded) {
4155 // Tell libjsig jvm finishes setting signal handlers
4156 (*end_signal_setting)();
4157 }
4158
4159 // We don't activate signal checker if libjsig is in place, we trust ourselves
4160 // and if UserSignalHandler is installed all bets are off
4161 if (CheckJNICalls) {
4162 if (libjsig_is_loaded) {
4163 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4164 check_signals = false;
4165 }
4166 if (AllowUserSignalHandlers) {
4167 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4168 check_signals = false;
4169 }
4170 }
4171 }
4172 }
4173
4174 #ifndef _ALLBSD_SOURCE
4175 // This is the fastest way to get thread cpu time on Bsd.
4176 // Returns cpu time (user+sys) for any thread, not only for current.
4177 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4178 // It might work on 2.6.10+ with a special kernel/glibc patch.
4179 // For reference, please, see IEEE Std 1003.1-2004:
4180 // http://www.unix.org/single_unix_specification
4181
4182 jlong os::Bsd::fast_thread_cpu_time(clockid_t clockid) {
4183 struct timespec tp;
4184 int rc = os::Bsd::clock_gettime(clockid, &tp);
4185 assert(rc == 0, "clock_gettime is expected to return 0 code");
4186
4187 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4188 }
4189 #endif
4190
4191 /////
4192 // glibc on Bsd platform uses non-documented flag
4193 // to indicate, that some special sort of signal
4194 // trampoline is used.
4195 // We will never set this flag, and we should
4196 // ignore this flag in our diagnostic
4197 #ifdef SIGNIFICANT_SIGNAL_MASK
4198 #undef SIGNIFICANT_SIGNAL_MASK
4199 #endif
4200 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4201
4202 static const char* get_signal_handler_name(address handler,
4203 char* buf, int buflen) {
4204 int offset;
4205 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4206 if (found) {
4207 // skip directory names
4208 const char *p1, *p2;
4209 p1 = buf;
4210 size_t len = strlen(os::file_separator());
4211 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4212 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4213 } else {
4214 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4215 }
4216 return buf;
4217 }
4218
4219 static void print_signal_handler(outputStream* st, int sig,
4220 char* buf, size_t buflen) {
4221 struct sigaction sa;
4222
4223 sigaction(sig, NULL, &sa);
4224
4225 // See comment for SIGNIFICANT_SIGNAL_MASK define
4226 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4227
4228 st->print("%s: ", os::exception_name(sig, buf, buflen));
4229
4230 address handler = (sa.sa_flags & SA_SIGINFO)
4231 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4232 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4233
4234 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4235 st->print("SIG_DFL");
4236 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4237 st->print("SIG_IGN");
4238 } else {
4239 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4240 }
4241
4242 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4243
4244 address rh = VMError::get_resetted_sighandler(sig);
4245 // May be, handler was resetted by VMError?
4246 if(rh != NULL) {
4247 handler = rh;
4248 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4249 }
4250
4251 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
4252
4253 // Check: is it our handler?
4254 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4255 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4256 // It is our signal handler
4257 // check for flags, reset system-used one!
4258 if((int)sa.sa_flags != os::Bsd::get_our_sigflags(sig)) {
4259 st->print(
4260 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4261 os::Bsd::get_our_sigflags(sig));
4262 }
4263 }
4264 st->cr();
4265 }
4266
4267
4268 #define DO_SIGNAL_CHECK(sig) \
4269 if (!sigismember(&check_signal_done, sig)) \
4270 os::Bsd::check_signal_handler(sig)
4271
4272 // This method is a periodic task to check for misbehaving JNI applications
4273 // under CheckJNI, we can add any periodic checks here
4274
4275 void os::run_periodic_checks() {
4276
4277 if (check_signals == false) return;
4278
4279 // SEGV and BUS if overridden could potentially prevent
4280 // generation of hs*.log in the event of a crash, debugging
4281 // such a case can be very challenging, so we absolutely
4282 // check the following for a good measure:
4283 DO_SIGNAL_CHECK(SIGSEGV);
4284 DO_SIGNAL_CHECK(SIGILL);
4285 DO_SIGNAL_CHECK(SIGFPE);
4286 DO_SIGNAL_CHECK(SIGBUS);
4287 DO_SIGNAL_CHECK(SIGPIPE);
4288 DO_SIGNAL_CHECK(SIGXFSZ);
4289
4290
4291 // ReduceSignalUsage allows the user to override these handlers
4292 // see comments at the very top and jvm_solaris.h
4293 if (!ReduceSignalUsage) {
4294 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4295 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4296 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4297 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4298 }
4299
4300 DO_SIGNAL_CHECK(SR_signum);
4301 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4302 }
4303
4304 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4305
4306 static os_sigaction_t os_sigaction = NULL;
4307
4308 void os::Bsd::check_signal_handler(int sig) {
4309 char buf[O_BUFLEN];
4310 address jvmHandler = NULL;
4311
4312
4313 struct sigaction act;
4314 if (os_sigaction == NULL) {
4315 // only trust the default sigaction, in case it has been interposed
4316 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4317 if (os_sigaction == NULL) return;
4318 }
4319
4320 os_sigaction(sig, (struct sigaction*)NULL, &act);
4321
4322
4323 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4324
4325 address thisHandler = (act.sa_flags & SA_SIGINFO)
4326 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4327 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4328
4329
4330 switch(sig) {
4331 case SIGSEGV:
4332 case SIGBUS:
4333 case SIGFPE:
4334 case SIGPIPE:
4335 case SIGILL:
4336 case SIGXFSZ:
4337 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4338 break;
4339
4340 case SHUTDOWN1_SIGNAL:
4341 case SHUTDOWN2_SIGNAL:
4342 case SHUTDOWN3_SIGNAL:
4343 case BREAK_SIGNAL:
4344 jvmHandler = (address)user_handler();
4345 break;
4346
4347 case INTERRUPT_SIGNAL:
4348 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4349 break;
4350
4351 default:
4352 if (sig == SR_signum) {
4353 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4354 } else {
4355 return;
4356 }
4357 break;
4358 }
4359
4360 if (thisHandler != jvmHandler) {
4361 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4362 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4363 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4364 // No need to check this sig any longer
4365 sigaddset(&check_signal_done, sig);
4366 } else if(os::Bsd::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Bsd::get_our_sigflags(sig)) {
4367 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4368 tty->print("expected:" PTR32_FORMAT, os::Bsd::get_our_sigflags(sig));
4369 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4370 // No need to check this sig any longer
4371 sigaddset(&check_signal_done, sig);
4372 }
4373
4374 // Dump all the signal
4375 if (sigismember(&check_signal_done, sig)) {
4376 print_signal_handlers(tty, buf, O_BUFLEN);
4377 }
4378 }
4379
4380 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4381
4382 extern bool signal_name(int signo, char* buf, size_t len);
4383
4384 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4385 if (0 < exception_code && exception_code <= SIGRTMAX) {
4386 // signal
4387 if (!signal_name(exception_code, buf, size)) {
4388 jio_snprintf(buf, size, "SIG%d", exception_code);
4389 }
4390 return buf;
4391 } else {
4392 return NULL;
4393 }
4394 }
4395
4396 // this is called _before_ the most of global arguments have been parsed
4397 void os::init(void) {
4398 char dummy; /* used to get a guess on initial stack address */
4399 // first_hrtime = gethrtime();
4400
4401 // With BsdThreads the JavaMain thread pid (primordial thread)
4402 // is different than the pid of the java launcher thread.
4403 // So, on Bsd, the launcher thread pid is passed to the VM
4404 // via the sun.java.launcher.pid property.
4405 // Use this property instead of getpid() if it was correctly passed.
4406 // See bug 6351349.
4407 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4408
4409 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4410
4411 clock_tics_per_sec = CLK_TCK;
4412
4413 init_random(1234567);
4414
4415 ThreadCritical::initialize();
4416
4417 Bsd::set_page_size(getpagesize());
4418 if (Bsd::page_size() == -1) {
4419 fatal(err_msg("os_bsd.cpp: os::init: sysconf failed (%s)",
4420 strerror(errno)));
4421 }
4422 init_page_sizes((size_t) Bsd::page_size());
4423
4424 Bsd::initialize_system_info();
4425
4426 // main_thread points to the aboriginal thread
4427 Bsd::_main_thread = pthread_self();
4428
4429 Bsd::clock_init();
4430 initial_time_count = os::elapsed_counter();
4431
4432 #ifdef __APPLE__
4433 // XXXDARWIN
4434 // Work around the unaligned VM callbacks in hotspot's
4435 // sharedRuntime. The callbacks don't use SSE2 instructions, and work on
4436 // Linux, Solaris, and FreeBSD. On Mac OS X, dyld (rightly so) enforces
4437 // alignment when doing symbol lookup. To work around this, we force early
4438 // binding of all symbols now, thus binding when alignment is known-good.
4439 _dyld_bind_fully_image_containing_address((const void *) &os::init);
4440 #endif
4441 }
4442
4443 // To install functions for atexit system call
4444 extern "C" {
4445 static void perfMemory_exit_helper() {
4446 perfMemory_exit();
4447 }
4448 }
4449
4450 // this is called _after_ the global arguments have been parsed
4451 jint os::init_2(void)
4452 {
4453 #ifndef _ALLBSD_SOURCE
4454 Bsd::fast_thread_clock_init();
4455 #endif
4456
4457 // Allocate a single page and mark it as readable for safepoint polling
4458 address polling_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4459 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4460
4461 os::set_polling_page( polling_page );
4462
4463 #ifndef PRODUCT
4464 if(Verbose && PrintMiscellaneous)
4465 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4466 #endif
4467
4468 if (!UseMembar) {
4469 address mem_serialize_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4470 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4471 os::set_memory_serialize_page( mem_serialize_page );
4472
4473 #ifndef PRODUCT
4474 if(Verbose && PrintMiscellaneous)
4475 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4476 #endif
4477 }
4478
4479 os::large_page_init();
4480
4481 // initialize suspend/resume support - must do this before signal_sets_init()
4482 if (SR_initialize() != 0) {
4483 perror("SR_initialize failed");
4484 return JNI_ERR;
4485 }
4486
4487 Bsd::signal_sets_init();
4488 Bsd::install_signal_handlers();
4489
4490 // Check minimum allowable stack size for thread creation and to initialize
4491 // the java system classes, including StackOverflowError - depends on page
4492 // size. Add a page for compiler2 recursion in main thread.
4493 // Add in 2*BytesPerWord times page size to account for VM stack during
4494 // class initialization depending on 32 or 64 bit VM.
4495 os::Bsd::min_stack_allowed = MAX2(os::Bsd::min_stack_allowed,
4496 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4497 2*BytesPerWord COMPILER2_PRESENT(+1)) * Bsd::page_size());
4498
4499 size_t threadStackSizeInBytes = ThreadStackSize * K;
4500 if (threadStackSizeInBytes != 0 &&
4501 threadStackSizeInBytes < os::Bsd::min_stack_allowed) {
4502 tty->print_cr("\nThe stack size specified is too small, "
4503 "Specify at least %dk",
4504 os::Bsd::min_stack_allowed/ K);
4505 return JNI_ERR;
4506 }
4507
4508 // Make the stack size a multiple of the page size so that
4509 // the yellow/red zones can be guarded.
4510 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4511 vm_page_size()));
4512
4513 #ifndef _ALLBSD_SOURCE
4514 Bsd::capture_initial_stack(JavaThread::stack_size_at_create());
4515
4516 Bsd::libpthread_init();
4517 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4518 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4519 Bsd::glibc_version(), Bsd::libpthread_version(),
4520 Bsd::is_floating_stack() ? "floating stack" : "fixed stack");
4521 }
4522
4523 if (UseNUMA) {
4524 if (!Bsd::libnuma_init()) {
4525 UseNUMA = false;
4526 } else {
4527 if ((Bsd::numa_max_node() < 1)) {
4528 // There's only one node(they start from 0), disable NUMA.
4529 UseNUMA = false;
4530 }
4531 }
4532 // With SHM large pages we cannot uncommit a page, so there's not way
4533 // we can make the adaptive lgrp chunk resizing work. If the user specified
4534 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4535 // disable adaptive resizing.
4536 if (UseNUMA && UseLargePages && UseSHM) {
4537 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4538 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4539 UseLargePages = false;
4540 } else {
4541 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4542 UseAdaptiveSizePolicy = false;
4543 UseAdaptiveNUMAChunkSizing = false;
4544 }
4545 } else {
4546 UseNUMA = false;
4547 }
4548 }
4549 if (!UseNUMA && ForceNUMA) {
4550 UseNUMA = true;
4551 }
4552 }
4553 #endif
4554
4555 if (MaxFDLimit) {
4556 // set the number of file descriptors to max. print out error
4557 // if getrlimit/setrlimit fails but continue regardless.
4558 struct rlimit nbr_files;
4559 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4560 if (status != 0) {
4561 if (PrintMiscellaneous && (Verbose || WizardMode))
4562 perror("os::init_2 getrlimit failed");
4563 } else {
4564 nbr_files.rlim_cur = nbr_files.rlim_max;
4565
4566 #ifdef __APPLE__
4567 // Darwin returns RLIM_INFINITY for rlim_max, but fails with EINVAL if
4568 // you attempt to use RLIM_INFINITY. As per setrlimit(2), OPEN_MAX must
4569 // be used instead
4570 nbr_files.rlim_cur = MIN(OPEN_MAX, nbr_files.rlim_cur);
4571 #endif
4572
4573 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4574 if (status != 0) {
4575 if (PrintMiscellaneous && (Verbose || WizardMode))
4576 perror("os::init_2 setrlimit failed");
4577 }
4578 }
4579 }
4580
4581 #ifndef _ALLBSD_SOURCE
4582 // Initialize lock used to serialize thread creation (see os::create_thread)
4583 Bsd::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4584 #endif
4585
4586 // at-exit methods are called in the reverse order of their registration.
4587 // atexit functions are called on return from main or as a result of a
4588 // call to exit(3C). There can be only 32 of these functions registered
4589 // and atexit() does not set errno.
4590
4591 if (PerfAllowAtExitRegistration) {
4592 // only register atexit functions if PerfAllowAtExitRegistration is set.
4593 // atexit functions can be delayed until process exit time, which
4594 // can be problematic for embedded VM situations. Embedded VMs should
4595 // call DestroyJavaVM() to assure that VM resources are released.
4596
4597 // note: perfMemory_exit_helper atexit function may be removed in
4598 // the future if the appropriate cleanup code can be added to the
4599 // VM_Exit VMOperation's doit method.
4600 if (atexit(perfMemory_exit_helper) != 0) {
4601 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4602 }
4603 }
4604
4605 // initialize thread priority policy
4606 prio_init();
4607
4608 #ifdef __APPLE__
4609 // dynamically link to objective c gc registration
4610 void *handleLibObjc = dlopen(OBJC_LIB, RTLD_LAZY);
4611 if (handleLibObjc != NULL) {
4612 objc_registerThreadWithCollectorFunction = (objc_registerThreadWithCollector_t) dlsym(handleLibObjc, OBJC_GCREGISTER);
4613 }
4614 #endif
4615
4616 return JNI_OK;
4617 }
4618
4619 // this is called at the end of vm_initialization
4620 void os::init_3(void) { }
4621
4622 // Mark the polling page as unreadable
4623 void os::make_polling_page_unreadable(void) {
4624 if( !guard_memory((char*)_polling_page, Bsd::page_size()) )
4625 fatal("Could not disable polling page");
4626 };
4627
4628 // Mark the polling page as readable
4629 void os::make_polling_page_readable(void) {
4630 if( !bsd_mprotect((char *)_polling_page, Bsd::page_size(), PROT_READ)) {
4631 fatal("Could not enable polling page");
4632 }
4633 };
4634
4635 int os::active_processor_count() {
4636 #ifdef _ALLBSD_SOURCE
4637 return _processor_count;
4638 #else
4639 // Bsd doesn't yet have a (official) notion of processor sets,
4640 // so just return the number of online processors.
4641 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4642 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4643 return online_cpus;
4644 #endif
4645 }
4646
4647 void os::set_native_thread_name(const char *name) {
4648 #if defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED > MAC_OS_X_VERSION_10_5
4649 // This is only supported in Snow Leopard and beyond
4650 if (name != NULL) {
4651 // Add a "Java: " prefix to the name
4652 char buf[MAXTHREADNAMESIZE];
4653 snprintf(buf, sizeof(buf), "Java: %s", name);
4654 pthread_setname_np(buf);
4655 }
4656 #endif
4657 }
4658
4659 bool os::distribute_processes(uint length, uint* distribution) {
4660 // Not yet implemented.
4661 return false;
4662 }
4663
4664 bool os::bind_to_processor(uint processor_id) {
4665 // Not yet implemented.
4666 return false;
4667 }
4668
4669 ///
4670
4671 // Suspends the target using the signal mechanism and then grabs the PC before
4672 // resuming the target. Used by the flat-profiler only
4673 ExtendedPC os::get_thread_pc(Thread* thread) {
4674 // Make sure that it is called by the watcher for the VMThread
4675 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4676 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4677
4678 ExtendedPC epc;
4679
4680 OSThread* osthread = thread->osthread();
4681 if (do_suspend(osthread)) {
4682 if (osthread->ucontext() != NULL) {
4683 epc = os::Bsd::ucontext_get_pc(osthread->ucontext());
4684 } else {
4685 // NULL context is unexpected, double-check this is the VMThread
4686 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4687 }
4688 do_resume(osthread);
4689 }
4690 // failure means pthread_kill failed for some reason - arguably this is
4691 // a fatal problem, but such problems are ignored elsewhere
4692
4693 return epc;
4694 }
4695
4696 int os::Bsd::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4697 {
4698 #ifdef _ALLBSD_SOURCE
4699 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4700 #else
4701 if (is_NPTL()) {
4702 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4703 } else {
4704 #ifndef IA64
4705 // 6292965: BsdThreads pthread_cond_timedwait() resets FPU control
4706 // word back to default 64bit precision if condvar is signaled. Java
4707 // wants 53bit precision. Save and restore current value.
4708 int fpu = get_fpu_control_word();
4709 #endif // IA64
4710 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4711 #ifndef IA64
4712 set_fpu_control_word(fpu);
4713 #endif // IA64
4714 return status;
4715 }
4716 #endif
4717 }
4718
4719 ////////////////////////////////////////////////////////////////////////////////
4720 // debug support
4721
4722 static address same_page(address x, address y) {
4723 int page_bits = -os::vm_page_size();
4724 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4725 return x;
4726 else if (x > y)
4727 return (address)(intptr_t(y) | ~page_bits) + 1;
4728 else
4729 return (address)(intptr_t(y) & page_bits);
4730 }
4731
4732 bool os::find(address addr, outputStream* st) {
4733 Dl_info dlinfo;
4734 memset(&dlinfo, 0, sizeof(dlinfo));
4735 if (dladdr(addr, &dlinfo)) {
4736 st->print(PTR_FORMAT ": ", addr);
4737 if (dlinfo.dli_sname != NULL) {
4738 st->print("%s+%#x", dlinfo.dli_sname,
4739 addr - (intptr_t)dlinfo.dli_saddr);
4740 } else if (dlinfo.dli_fname) {
4741 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4742 } else {
4743 st->print("<absolute address>");
4744 }
4745 if (dlinfo.dli_fname) {
4746 st->print(" in %s", dlinfo.dli_fname);
4747 }
4748 if (dlinfo.dli_fbase) {
4749 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4750 }
4751 st->cr();
4752
4753 if (Verbose) {
4754 // decode some bytes around the PC
4755 address begin = same_page(addr-40, addr);
4756 address end = same_page(addr+40, addr);
4757 address lowest = (address) dlinfo.dli_sname;
4758 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4759 if (begin < lowest) begin = lowest;
4760 Dl_info dlinfo2;
4761 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4762 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4763 end = (address) dlinfo2.dli_saddr;
4764 Disassembler::decode(begin, end, st);
4765 }
4766 return true;
4767 }
4768 return false;
4769 }
4770
4771 ////////////////////////////////////////////////////////////////////////////////
4772 // misc
4773
4774 // This does not do anything on Bsd. This is basically a hook for being
4775 // able to use structured exception handling (thread-local exception filters)
4776 // on, e.g., Win32.
4777 void
4778 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4779 JavaCallArguments* args, Thread* thread) {
4780 f(value, method, args, thread);
4781 }
4782
4783 void os::print_statistics() {
4784 }
4785
4786 int os::message_box(const char* title, const char* message) {
4787 int i;
4788 fdStream err(defaultStream::error_fd());
4789 for (i = 0; i < 78; i++) err.print_raw("=");
4790 err.cr();
4791 err.print_raw_cr(title);
4792 for (i = 0; i < 78; i++) err.print_raw("-");
4793 err.cr();
4794 err.print_raw_cr(message);
4795 for (i = 0; i < 78; i++) err.print_raw("=");
4796 err.cr();
4797
4798 char buf[16];
4799 // Prevent process from exiting upon "read error" without consuming all CPU
4800 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4801
4802 return buf[0] == 'y' || buf[0] == 'Y';
4803 }
4804
4805 int os::stat(const char *path, struct stat *sbuf) {
4806 char pathbuf[MAX_PATH];
4807 if (strlen(path) > MAX_PATH - 1) {
4808 errno = ENAMETOOLONG;
4809 return -1;
4810 }
4811 os::native_path(strcpy(pathbuf, path));
4812 return ::stat(pathbuf, sbuf);
4813 }
4814
4815 bool os::check_heap(bool force) {
4816 return true;
4817 }
4818
4819 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4820 return ::vsnprintf(buf, count, format, args);
4821 }
4822
4823 // Is a (classpath) directory empty?
4824 bool os::dir_is_empty(const char* path) {
4825 DIR *dir = NULL;
4826 struct dirent *ptr;
4827
4828 dir = opendir(path);
4829 if (dir == NULL) return true;
4830
4831 /* Scan the directory */
4832 bool result = true;
4833 char buf[sizeof(struct dirent) + MAX_PATH];
4834 while (result && (ptr = ::readdir(dir)) != NULL) {
4835 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4836 result = false;
4837 }
4838 }
4839 closedir(dir);
4840 return result;
4841 }
4842
4843 // This code originates from JDK's sysOpen and open64_w
4844 // from src/solaris/hpi/src/system_md.c
4845
4846 #ifndef O_DELETE
4847 #define O_DELETE 0x10000
4848 #endif
4849
4850 // Open a file. Unlink the file immediately after open returns
4851 // if the specified oflag has the O_DELETE flag set.
4852 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4853
4854 int os::open(const char *path, int oflag, int mode) {
4855
4856 if (strlen(path) > MAX_PATH - 1) {
4857 errno = ENAMETOOLONG;
4858 return -1;
4859 }
4860 int fd;
4861 int o_delete = (oflag & O_DELETE);
4862 oflag = oflag & ~O_DELETE;
4863
4864 fd = ::open(path, oflag, mode);
4865 if (fd == -1) return -1;
4866
4867 //If the open succeeded, the file might still be a directory
4868 {
4869 struct stat buf;
4870 int ret = ::fstat(fd, &buf);
4871 int st_mode = buf.st_mode;
4872
4873 if (ret != -1) {
4874 if ((st_mode & S_IFMT) == S_IFDIR) {
4875 errno = EISDIR;
4876 ::close(fd);
4877 return -1;
4878 }
4879 } else {
4880 ::close(fd);
4881 return -1;
4882 }
4883 }
4884
4885 /*
4886 * All file descriptors that are opened in the JVM and not
4887 * specifically destined for a subprocess should have the
4888 * close-on-exec flag set. If we don't set it, then careless 3rd
4889 * party native code might fork and exec without closing all
4890 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4891 * UNIXProcess.c), and this in turn might:
4892 *
4893 * - cause end-of-file to fail to be detected on some file
4894 * descriptors, resulting in mysterious hangs, or
4895 *
4896 * - might cause an fopen in the subprocess to fail on a system
4897 * suffering from bug 1085341.
4898 *
4899 * (Yes, the default setting of the close-on-exec flag is a Unix
4900 * design flaw)
4901 *
4902 * See:
4903 * 1085341: 32-bit stdio routines should support file descriptors >255
4904 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4905 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4906 */
4907 #ifdef FD_CLOEXEC
4908 {
4909 int flags = ::fcntl(fd, F_GETFD);
4910 if (flags != -1)
4911 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4912 }
4913 #endif
4914
4915 if (o_delete != 0) {
4916 ::unlink(path);
4917 }
4918 return fd;
4919 }
4920
4921
4922 // create binary file, rewriting existing file if required
4923 int os::create_binary_file(const char* path, bool rewrite_existing) {
4924 int oflags = O_WRONLY | O_CREAT;
4925 if (!rewrite_existing) {
4926 oflags |= O_EXCL;
4927 }
4928 return ::open(path, oflags, S_IREAD | S_IWRITE);
4929 }
4930
4931 // return current position of file pointer
4932 jlong os::current_file_offset(int fd) {
4933 return (jlong)::lseek(fd, (off_t)0, SEEK_CUR);
4934 }
4935
4936 // move file pointer to the specified offset
4937 jlong os::seek_to_file_offset(int fd, jlong offset) {
4938 return (jlong)::lseek(fd, (off_t)offset, SEEK_SET);
4939 }
4940
4941 // This code originates from JDK's sysAvailable
4942 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4943
4944 int os::available(int fd, jlong *bytes) {
4945 jlong cur, end;
4946 int mode;
4947 struct stat buf;
4948
4949 if (::fstat(fd, &buf) >= 0) {
4950 mode = buf.st_mode;
4951 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4952 /*
4953 * XXX: is the following call interruptible? If so, this might
4954 * need to go through the INTERRUPT_IO() wrapper as for other
4955 * blocking, interruptible calls in this file.
4956 */
4957 int n;
4958 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4959 *bytes = n;
4960 return 1;
4961 }
4962 }
4963 }
4964 if ((cur = ::lseek(fd, 0L, SEEK_CUR)) == -1) {
4965 return 0;
4966 } else if ((end = ::lseek(fd, 0L, SEEK_END)) == -1) {
4967 return 0;
4968 } else if (::lseek(fd, cur, SEEK_SET) == -1) {
4969 return 0;
4970 }
4971 *bytes = end - cur;
4972 return 1;
4973 }
4974
4975 int os::socket_available(int fd, jint *pbytes) {
4976 if (fd < 0)
4977 return OS_OK;
4978
4979 int ret;
4980
4981 RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
4982
4983 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4984 // is expected to return 0 on failure and 1 on success to the jdk.
4985
4986 return (ret == OS_ERR) ? 0 : 1;
4987 }
4988
4989 // Map a block of memory.
4990 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4991 char *addr, size_t bytes, bool read_only,
4992 bool allow_exec) {
4993 int prot;
4994 int flags;
4995
4996 if (read_only) {
4997 prot = PROT_READ;
4998 flags = MAP_SHARED;
4999 } else {
5000 prot = PROT_READ | PROT_WRITE;
5001 flags = MAP_PRIVATE;
5002 }
5003
5004 if (allow_exec) {
5005 prot |= PROT_EXEC;
5006 }
5007
5008 if (addr != NULL) {
5009 flags |= MAP_FIXED;
5010 }
5011
5012 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5013 fd, file_offset);
5014 if (mapped_address == MAP_FAILED) {
5015 return NULL;
5016 }
5017 return mapped_address;
5018 }
5019
5020
5021 // Remap a block of memory.
5022 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
5023 char *addr, size_t bytes, bool read_only,
5024 bool allow_exec) {
5025 // same as map_memory() on this OS
5026 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5027 allow_exec);
5028 }
5029
5030
5031 // Unmap a block of memory.
5032 bool os::unmap_memory(char* addr, size_t bytes) {
5033 return munmap(addr, bytes) == 0;
5034 }
5035
5036 #ifndef _ALLBSD_SOURCE
5037 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5038
5039 static clockid_t thread_cpu_clockid(Thread* thread) {
5040 pthread_t tid = thread->osthread()->pthread_id();
5041 clockid_t clockid;
5042
5043 // Get thread clockid
5044 int rc = os::Bsd::pthread_getcpuclockid(tid, &clockid);
5045 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5046 return clockid;
5047 }
5048 #endif
5049
5050 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5051 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5052 // of a thread.
5053 //
5054 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5055 // the fast estimate available on the platform.
5056
5057 jlong os::current_thread_cpu_time() {
5058 #ifdef __APPLE__
5059 return os::thread_cpu_time(Thread::current(), true /* user + sys */);
5060 #elif !defined(_ALLBSD_SOURCE)
5061 if (os::Bsd::supports_fast_thread_cpu_time()) {
5062 return os::Bsd::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5063 } else {
5064 // return user + sys since the cost is the same
5065 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5066 }
5067 #endif
5068 }
5069
5070 jlong os::thread_cpu_time(Thread* thread) {
5071 #ifndef _ALLBSD_SOURCE
5072 // consistent with what current_thread_cpu_time() returns
5073 if (os::Bsd::supports_fast_thread_cpu_time()) {
5074 return os::Bsd::fast_thread_cpu_time(thread_cpu_clockid(thread));
5075 } else {
5076 return slow_thread_cpu_time(thread, true /* user + sys */);
5077 }
5078 #endif
5079 }
5080
5081 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5082 #ifdef __APPLE__
5083 return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5084 #elif !defined(_ALLBSD_SOURCE)
5085 if (user_sys_cpu_time && os::Bsd::supports_fast_thread_cpu_time()) {
5086 return os::Bsd::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5087 } else {
5088 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5089 }
5090 #endif
5091 }
5092
5093 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5094 #ifdef __APPLE__
5095 struct thread_basic_info tinfo;
5096 mach_msg_type_number_t tcount = THREAD_INFO_MAX;
5097 kern_return_t kr;
5098 thread_t mach_thread;
5099
5100 mach_thread = thread->osthread()->thread_id();
5101 kr = thread_info(mach_thread, THREAD_BASIC_INFO, (thread_info_t)&tinfo, &tcount);
5102 if (kr != KERN_SUCCESS)
5103 return -1;
5104
5105 if (user_sys_cpu_time) {
5106 jlong nanos;
5107 nanos = ((jlong) tinfo.system_time.seconds + tinfo.user_time.seconds) * (jlong)1000000000;
5108 nanos += ((jlong) tinfo.system_time.microseconds + (jlong) tinfo.user_time.microseconds) * (jlong)1000;
5109 return nanos;
5110 } else {
5111 return ((jlong)tinfo.user_time.seconds * 1000000000) + ((jlong)tinfo.user_time.microseconds * (jlong)1000);
5112 }
5113 #elif !defined(_ALLBSD_SOURCE)
5114 if (user_sys_cpu_time && os::Bsd::supports_fast_thread_cpu_time()) {
5115 return os::Bsd::fast_thread_cpu_time(thread_cpu_clockid(thread));
5116 } else {
5117 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5118 }
5119 #endif
5120 }
5121
5122 #ifndef _ALLBSD_SOURCE
5123 //
5124 // -1 on error.
5125 //
5126
5127 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5128 static bool proc_pid_cpu_avail = true;
5129 static bool proc_task_unchecked = true;
5130 static const char *proc_stat_path = "/proc/%d/stat";
5131 pid_t tid = thread->osthread()->thread_id();
5132 int i;
5133 char *s;
5134 char stat[2048];
5135 int statlen;
5136 char proc_name[64];
5137 int count;
5138 long sys_time, user_time;
5139 char string[64];
5140 char cdummy;
5141 int idummy;
5142 long ldummy;
5143 FILE *fp;
5144
5145 // We first try accessing /proc/<pid>/cpu since this is faster to
5146 // process. If this file is not present (bsd kernels 2.5 and above)
5147 // then we open /proc/<pid>/stat.
5148 if ( proc_pid_cpu_avail ) {
5149 sprintf(proc_name, "/proc/%d/cpu", tid);
5150 fp = fopen(proc_name, "r");
5151 if ( fp != NULL ) {
5152 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
5153 fclose(fp);
5154 if ( count != 3 ) return -1;
5155
5156 if (user_sys_cpu_time) {
5157 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5158 } else {
5159 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5160 }
5161 }
5162 else proc_pid_cpu_avail = false;
5163 }
5164
5165 // The /proc/<tid>/stat aggregates per-process usage on
5166 // new Bsd kernels 2.6+ where NPTL is supported.
5167 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5168 // See bug 6328462.
5169 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
5170 // and possibly in some other cases, so we check its availability.
5171 if (proc_task_unchecked && os::Bsd::is_NPTL()) {
5172 // This is executed only once
5173 proc_task_unchecked = false;
5174 fp = fopen("/proc/self/task", "r");
5175 if (fp != NULL) {
5176 proc_stat_path = "/proc/self/task/%d/stat";
5177 fclose(fp);
5178 }
5179 }
5180
5181 sprintf(proc_name, proc_stat_path, tid);
5182 fp = fopen(proc_name, "r");
5183 if ( fp == NULL ) return -1;
5184 statlen = fread(stat, 1, 2047, fp);
5185 stat[statlen] = '\0';
5186 fclose(fp);
5187
5188 // Skip pid and the command string. Note that we could be dealing with
5189 // weird command names, e.g. user could decide to rename java launcher
5190 // to "java 1.4.2 :)", then the stat file would look like
5191 // 1234 (java 1.4.2 :)) R ... ...
5192 // We don't really need to know the command string, just find the last
5193 // occurrence of ")" and then start parsing from there. See bug 4726580.
5194 s = strrchr(stat, ')');
5195 i = 0;
5196 if (s == NULL ) return -1;
5197
5198 // Skip blank chars
5199 do s++; while (isspace(*s));
5200
5201 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5202 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5203 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5204 &user_time, &sys_time);
5205 if ( count != 13 ) return -1;
5206 if (user_sys_cpu_time) {
5207 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5208 } else {
5209 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5210 }
5211 }
5212 #endif
5213
5214 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5215 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5216 info_ptr->may_skip_backward = false; // elapsed time not wall time
5217 info_ptr->may_skip_forward = false; // elapsed time not wall time
5218 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5219 }
5220
5221 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5222 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5223 info_ptr->may_skip_backward = false; // elapsed time not wall time
5224 info_ptr->may_skip_forward = false; // elapsed time not wall time
5225 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5226 }
5227
5228 bool os::is_thread_cpu_time_supported() {
5229 #ifdef __APPLE__
5230 return true;
5231 #elif defined(_ALLBSD_SOURCE)
5232 return false;
5233 #else
5234 return true;
5235 #endif
5236 }
5237
5238 // System loadavg support. Returns -1 if load average cannot be obtained.
5239 // Bsd doesn't yet have a (official) notion of processor sets,
5240 // so just return the system wide load average.
5241 int os::loadavg(double loadavg[], int nelem) {
5242 return ::getloadavg(loadavg, nelem);
5243 }
5244
5245 void os::pause() {
5246 char filename[MAX_PATH];
5247 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5248 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5249 } else {
5250 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5251 }
5252
5253 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5254 if (fd != -1) {
5255 struct stat buf;
5256 ::close(fd);
5257 while (::stat(filename, &buf) == 0) {
5258 (void)::poll(NULL, 0, 100);
5259 }
5260 } else {
5261 jio_fprintf(stderr,
5262 "Could not open pause file '%s', continuing immediately.\n", filename);
5263 }
5264 }
5265
5266
5267 // Refer to the comments in os_solaris.cpp park-unpark.
5268 //
5269 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5270 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5271 // For specifics regarding the bug see GLIBC BUGID 261237 :
5272 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5273 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5274 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5275 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5276 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5277 // and monitorenter when we're using 1-0 locking. All those operations may result in
5278 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5279 // of libpthread avoids the problem, but isn't practical.
5280 //
5281 // Possible remedies:
5282 //
5283 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5284 // This is palliative and probabilistic, however. If the thread is preempted
5285 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5286 // than the minimum period may have passed, and the abstime may be stale (in the
5287 // past) resultin in a hang. Using this technique reduces the odds of a hang
5288 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5289 //
5290 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5291 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5292 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5293 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5294 // thread.
5295 //
5296 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5297 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5298 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5299 // This also works well. In fact it avoids kernel-level scalability impediments
5300 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5301 // timers in a graceful fashion.
5302 //
5303 // 4. When the abstime value is in the past it appears that control returns
5304 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5305 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5306 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5307 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5308 // It may be possible to avoid reinitialization by checking the return
5309 // value from pthread_cond_timedwait(). In addition to reinitializing the
5310 // condvar we must establish the invariant that cond_signal() is only called
5311 // within critical sections protected by the adjunct mutex. This prevents
5312 // cond_signal() from "seeing" a condvar that's in the midst of being
5313 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5314 // desirable signal-after-unlock optimization that avoids futile context switching.
5315 //
5316 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5317 // structure when a condvar is used or initialized. cond_destroy() would
5318 // release the helper structure. Our reinitialize-after-timedwait fix
5319 // put excessive stress on malloc/free and locks protecting the c-heap.
5320 //
5321 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5322 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5323 // and only enabling the work-around for vulnerable environments.
5324
5325 // utility to compute the abstime argument to timedwait:
5326 // millis is the relative timeout time
5327 // abstime will be the absolute timeout time
5328 // TODO: replace compute_abstime() with unpackTime()
5329
5330 static struct timespec* compute_abstime(struct timespec* abstime, jlong millis) {
5331 if (millis < 0) millis = 0;
5332 struct timeval now;
5333 int status = gettimeofday(&now, NULL);
5334 assert(status == 0, "gettimeofday");
5335 jlong seconds = millis / 1000;
5336 millis %= 1000;
5337 if (seconds > 50000000) { // see man cond_timedwait(3T)
5338 seconds = 50000000;
5339 }
5340 abstime->tv_sec = now.tv_sec + seconds;
5341 long usec = now.tv_usec + millis * 1000;
5342 if (usec >= 1000000) {
5343 abstime->tv_sec += 1;
5344 usec -= 1000000;
5345 }
5346 abstime->tv_nsec = usec * 1000;
5347 return abstime;
5348 }
5349
5350
5351 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5352 // Conceptually TryPark() should be equivalent to park(0).
5353
5354 int os::PlatformEvent::TryPark() {
5355 for (;;) {
5356 const int v = _Event ;
5357 guarantee ((v == 0) || (v == 1), "invariant") ;
5358 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5359 }
5360 }
5361
5362 void os::PlatformEvent::park() { // AKA "down()"
5363 // Invariant: Only the thread associated with the Event/PlatformEvent
5364 // may call park().
5365 // TODO: assert that _Assoc != NULL or _Assoc == Self
5366 int v ;
5367 for (;;) {
5368 v = _Event ;
5369 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5370 }
5371 guarantee (v >= 0, "invariant") ;
5372 if (v == 0) {
5373 // Do this the hard way by blocking ...
5374 int status = pthread_mutex_lock(_mutex);
5375 assert_status(status == 0, status, "mutex_lock");
5376 guarantee (_nParked == 0, "invariant") ;
5377 ++ _nParked ;
5378 while (_Event < 0) {
5379 status = pthread_cond_wait(_cond, _mutex);
5380 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5381 // Treat this the same as if the wait was interrupted
5382 if (status == ETIMEDOUT) { status = EINTR; }
5383 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5384 }
5385 -- _nParked ;
5386
5387 // In theory we could move the ST of 0 into _Event past the unlock(),
5388 // but then we'd need a MEMBAR after the ST.
5389 _Event = 0 ;
5390 status = pthread_mutex_unlock(_mutex);
5391 assert_status(status == 0, status, "mutex_unlock");
5392 }
5393 guarantee (_Event >= 0, "invariant") ;
5394 }
5395
5396 int os::PlatformEvent::park(jlong millis) {
5397 guarantee (_nParked == 0, "invariant") ;
5398
5399 int v ;
5400 for (;;) {
5401 v = _Event ;
5402 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5403 }
5404 guarantee (v >= 0, "invariant") ;
5405 if (v != 0) return OS_OK ;
5406
5407 // We do this the hard way, by blocking the thread.
5408 // Consider enforcing a minimum timeout value.
5409 struct timespec abst;
5410 compute_abstime(&abst, millis);
5411
5412 int ret = OS_TIMEOUT;
5413 int status = pthread_mutex_lock(_mutex);
5414 assert_status(status == 0, status, "mutex_lock");
5415 guarantee (_nParked == 0, "invariant") ;
5416 ++_nParked ;
5417
5418 // Object.wait(timo) will return because of
5419 // (a) notification
5420 // (b) timeout
5421 // (c) thread.interrupt
5422 //
5423 // Thread.interrupt and object.notify{All} both call Event::set.
5424 // That is, we treat thread.interrupt as a special case of notification.
5425 // The underlying Solaris implementation, cond_timedwait, admits
5426 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5427 // JVM from making those visible to Java code. As such, we must
5428 // filter out spurious wakeups. We assume all ETIME returns are valid.
5429 //
5430 // TODO: properly differentiate simultaneous notify+interrupt.
5431 // In that case, we should propagate the notify to another waiter.
5432
5433 while (_Event < 0) {
5434 status = os::Bsd::safe_cond_timedwait(_cond, _mutex, &abst);
5435 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5436 pthread_cond_destroy (_cond);
5437 pthread_cond_init (_cond, NULL) ;
5438 }
5439 assert_status(status == 0 || status == EINTR ||
5440 status == ETIMEDOUT,
5441 status, "cond_timedwait");
5442 if (!FilterSpuriousWakeups) break ; // previous semantics
5443 if (status == ETIMEDOUT) break ;
5444 // We consume and ignore EINTR and spurious wakeups.
5445 }
5446 --_nParked ;
5447 if (_Event >= 0) {
5448 ret = OS_OK;
5449 }
5450 _Event = 0 ;
5451 status = pthread_mutex_unlock(_mutex);
5452 assert_status(status == 0, status, "mutex_unlock");
5453 assert (_nParked == 0, "invariant") ;
5454 return ret;
5455 }
5456
5457 void os::PlatformEvent::unpark() {
5458 int v, AnyWaiters ;
5459 for (;;) {
5460 v = _Event ;
5461 if (v > 0) {
5462 // The LD of _Event could have reordered or be satisfied
5463 // by a read-aside from this processor's write buffer.
5464 // To avoid problems execute a barrier and then
5465 // ratify the value.
5466 OrderAccess::fence() ;
5467 if (_Event == v) return ;
5468 continue ;
5469 }
5470 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5471 }
5472 if (v < 0) {
5473 // Wait for the thread associated with the event to vacate
5474 int status = pthread_mutex_lock(_mutex);
5475 assert_status(status == 0, status, "mutex_lock");
5476 AnyWaiters = _nParked ;
5477 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5478 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5479 AnyWaiters = 0 ;
5480 pthread_cond_signal (_cond);
5481 }
5482 status = pthread_mutex_unlock(_mutex);
5483 assert_status(status == 0, status, "mutex_unlock");
5484 if (AnyWaiters != 0) {
5485 status = pthread_cond_signal(_cond);
5486 assert_status(status == 0, status, "cond_signal");
5487 }
5488 }
5489
5490 // Note that we signal() _after dropping the lock for "immortal" Events.
5491 // This is safe and avoids a common class of futile wakeups. In rare
5492 // circumstances this can cause a thread to return prematurely from
5493 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5494 // simply re-test the condition and re-park itself.
5495 }
5496
5497
5498 // JSR166
5499 // -------------------------------------------------------
5500
5501 /*
5502 * The solaris and bsd implementations of park/unpark are fairly
5503 * conservative for now, but can be improved. They currently use a
5504 * mutex/condvar pair, plus a a count.
5505 * Park decrements count if > 0, else does a condvar wait. Unpark
5506 * sets count to 1 and signals condvar. Only one thread ever waits
5507 * on the condvar. Contention seen when trying to park implies that someone
5508 * is unparking you, so don't wait. And spurious returns are fine, so there
5509 * is no need to track notifications.
5510 */
5511
5512 #define MAX_SECS 100000000
5513 /*
5514 * This code is common to bsd and solaris and will be moved to a
5515 * common place in dolphin.
5516 *
5517 * The passed in time value is either a relative time in nanoseconds
5518 * or an absolute time in milliseconds. Either way it has to be unpacked
5519 * into suitable seconds and nanoseconds components and stored in the
5520 * given timespec structure.
5521 * Given time is a 64-bit value and the time_t used in the timespec is only
5522 * a signed-32-bit value (except on 64-bit Bsd) we have to watch for
5523 * overflow if times way in the future are given. Further on Solaris versions
5524 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5525 * number of seconds, in abstime, is less than current_time + 100,000,000.
5526 * As it will be 28 years before "now + 100000000" will overflow we can
5527 * ignore overflow and just impose a hard-limit on seconds using the value
5528 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5529 * years from "now".
5530 */
5531
5532 static void unpackTime(struct timespec* absTime, bool isAbsolute, jlong time) {
5533 assert (time > 0, "convertTime");
5534
5535 struct timeval now;
5536 int status = gettimeofday(&now, NULL);
5537 assert(status == 0, "gettimeofday");
5538
5539 time_t max_secs = now.tv_sec + MAX_SECS;
5540
5541 if (isAbsolute) {
5542 jlong secs = time / 1000;
5543 if (secs > max_secs) {
5544 absTime->tv_sec = max_secs;
5545 }
5546 else {
5547 absTime->tv_sec = secs;
5548 }
5549 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5550 }
5551 else {
5552 jlong secs = time / NANOSECS_PER_SEC;
5553 if (secs >= MAX_SECS) {
5554 absTime->tv_sec = max_secs;
5555 absTime->tv_nsec = 0;
5556 }
5557 else {
5558 absTime->tv_sec = now.tv_sec + secs;
5559 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5560 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5561 absTime->tv_nsec -= NANOSECS_PER_SEC;
5562 ++absTime->tv_sec; // note: this must be <= max_secs
5563 }
5564 }
5565 }
5566 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5567 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5568 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5569 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5570 }
5571
5572 void Parker::park(bool isAbsolute, jlong time) {
5573 // Optional fast-path check:
5574 // Return immediately if a permit is available.
5575 if (_counter > 0) {
5576 _counter = 0 ;
5577 OrderAccess::fence();
5578 return ;
5579 }
5580
5581 Thread* thread = Thread::current();
5582 assert(thread->is_Java_thread(), "Must be JavaThread");
5583 JavaThread *jt = (JavaThread *)thread;
5584
5585 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5586 // Check interrupt before trying to wait
5587 if (Thread::is_interrupted(thread, false)) {
5588 return;
5589 }
5590
5591 // Next, demultiplex/decode time arguments
5592 struct timespec absTime;
5593 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5594 return;
5595 }
5596 if (time > 0) {
5597 unpackTime(&absTime, isAbsolute, time);
5598 }
5599
5600
5601 // Enter safepoint region
5602 // Beware of deadlocks such as 6317397.
5603 // The per-thread Parker:: mutex is a classic leaf-lock.
5604 // In particular a thread must never block on the Threads_lock while
5605 // holding the Parker:: mutex. If safepoints are pending both the
5606 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5607 ThreadBlockInVM tbivm(jt);
5608
5609 // Don't wait if cannot get lock since interference arises from
5610 // unblocking. Also. check interrupt before trying wait
5611 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5612 return;
5613 }
5614
5615 int status ;
5616 if (_counter > 0) { // no wait needed
5617 _counter = 0;
5618 status = pthread_mutex_unlock(_mutex);
5619 assert (status == 0, "invariant") ;
5620 OrderAccess::fence();
5621 return;
5622 }
5623
5624 #ifdef ASSERT
5625 // Don't catch signals while blocked; let the running threads have the signals.
5626 // (This allows a debugger to break into the running thread.)
5627 sigset_t oldsigs;
5628 sigset_t* allowdebug_blocked = os::Bsd::allowdebug_blocked_signals();
5629 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5630 #endif
5631
5632 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5633 jt->set_suspend_equivalent();
5634 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5635
5636 if (time == 0) {
5637 status = pthread_cond_wait (_cond, _mutex) ;
5638 } else {
5639 status = os::Bsd::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5640 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5641 pthread_cond_destroy (_cond) ;
5642 pthread_cond_init (_cond, NULL);
5643 }
5644 }
5645 assert_status(status == 0 || status == EINTR ||
5646 status == ETIMEDOUT,
5647 status, "cond_timedwait");
5648
5649 #ifdef ASSERT
5650 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5651 #endif
5652
5653 _counter = 0 ;
5654 status = pthread_mutex_unlock(_mutex) ;
5655 assert_status(status == 0, status, "invariant") ;
5656 // If externally suspended while waiting, re-suspend
5657 if (jt->handle_special_suspend_equivalent_condition()) {
5658 jt->java_suspend_self();
5659 }
5660
5661 OrderAccess::fence();
5662 }
5663
5664 void Parker::unpark() {
5665 int s, status ;
5666 status = pthread_mutex_lock(_mutex);
5667 assert (status == 0, "invariant") ;
5668 s = _counter;
5669 _counter = 1;
5670 if (s < 1) {
5671 if (WorkAroundNPTLTimedWaitHang) {
5672 status = pthread_cond_signal (_cond) ;
5673 assert (status == 0, "invariant") ;
5674 status = pthread_mutex_unlock(_mutex);
5675 assert (status == 0, "invariant") ;
5676 } else {
5677 status = pthread_mutex_unlock(_mutex);
5678 assert (status == 0, "invariant") ;
5679 status = pthread_cond_signal (_cond) ;
5680 assert (status == 0, "invariant") ;
5681 }
5682 } else {
5683 pthread_mutex_unlock(_mutex);
5684 assert (status == 0, "invariant") ;
5685 }
5686 }
5687
5688
5689 /* Darwin has no "environ" in a dynamic library. */
5690 #ifdef __APPLE__
5691 #include <crt_externs.h>
5692 #define environ (*_NSGetEnviron())
5693 #else
5694 extern char** environ;
5695 #endif
5696
5697 // Run the specified command in a separate process. Return its exit value,
5698 // or -1 on failure (e.g. can't fork a new process).
5699 // Unlike system(), this function can be called from signal handler. It
5700 // doesn't block SIGINT et al.
5701 int os::fork_and_exec(char* cmd) {
5702 const char * argv[4] = {"sh", "-c", cmd, NULL};
5703
5704 // fork() in BsdThreads/NPTL is not async-safe. It needs to run
5705 // pthread_atfork handlers and reset pthread library. All we need is a
5706 // separate process to execve. Make a direct syscall to fork process.
5707 // On IA64 there's no fork syscall, we have to use fork() and hope for
5708 // the best...
5709 pid_t pid = fork();
5710
5711 if (pid < 0) {
5712 // fork failed
5713 return -1;
5714
5715 } else if (pid == 0) {
5716 // child process
5717
5718 // execve() in BsdThreads will call pthread_kill_other_threads_np()
5719 // first to kill every thread on the thread list. Because this list is
5720 // not reset by fork() (see notes above), execve() will instead kill
5721 // every thread in the parent process. We know this is the only thread
5722 // in the new process, so make a system call directly.
5723 // IA64 should use normal execve() from glibc to match the glibc fork()
5724 // above.
5725 execve("/bin/sh", (char* const*)argv, environ);
5726
5727 // execve failed
5728 _exit(-1);
5729
5730 } else {
5731 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5732 // care about the actual exit code, for now.
5733
5734 int status;
5735
5736 // Wait for the child process to exit. This returns immediately if
5737 // the child has already exited. */
5738 while (waitpid(pid, &status, 0) < 0) {
5739 switch (errno) {
5740 case ECHILD: return 0;
5741 case EINTR: break;
5742 default: return -1;
5743 }
5744 }
5745
5746 if (WIFEXITED(status)) {
5747 // The child exited normally; get its exit code.
5748 return WEXITSTATUS(status);
5749 } else if (WIFSIGNALED(status)) {
5750 // The child exited because of a signal
5751 // The best value to return is 0x80 + signal number,
5752 // because that is what all Unix shells do, and because
5753 // it allows callers to distinguish between process exit and
5754 // process death by signal.
5755 return 0x80 + WTERMSIG(status);
5756 } else {
5757 // Unknown exit code; pass it through
5758 return status;
5759 }
5760 }
5761 }
5762
5763 // is_headless_jre()
5764 //
5765 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5766 // in order to report if we are running in a headless jre
5767 //
5768 // Since JDK8 xawt/libmawt.so was moved into the same directory
5769 // as libawt.so, and renamed libawt_xawt.so
5770 //
5771 bool os::is_headless_jre() {
5772 struct stat statbuf;
5773 char buf[MAXPATHLEN];
5774 char libmawtpath[MAXPATHLEN];
5775 const char *xawtstr = "/xawt/libmawt" JNI_LIB_SUFFIX;
5776 const char *new_xawtstr = "/libawt_xawt" JNI_LIB_SUFFIX;
5777 char *p;
5778
5779 // Get path to libjvm.so
5780 os::jvm_path(buf, sizeof(buf));
5781
5782 // Get rid of libjvm.so
5783 p = strrchr(buf, '/');
5784 if (p == NULL) return false;
5785 else *p = '\0';
5786
5787 // Get rid of client or server
5788 p = strrchr(buf, '/');
5789 if (p == NULL) return false;
5790 else *p = '\0';
5791
5792 // check xawt/libmawt.so
5793 strcpy(libmawtpath, buf);
5794 strcat(libmawtpath, xawtstr);
5795 if (::stat(libmawtpath, &statbuf) == 0) return false;
5796
5797 // check libawt_xawt.so
5798 strcpy(libmawtpath, buf);
5799 strcat(libmawtpath, new_xawtstr);
5800 if (::stat(libmawtpath, &statbuf) == 0) return false;
5801
5802 return true;
5803 }
5804
5805 // Generate the candidate core paths into the supplied buffer
5806 // Return the number of characters actually generated
5807 int os::get_core_paths(char* buffer, size_t bufferSize) {
5808 int n;
5809
5810 n = jio_snprintf(buffer, bufferSize, "/cores/core.%d", current_process_id());
5811
5812 // Truncate if theoretical string was longer than bufferSize
5813 n = MIN2(n, (int)bufferSize);
5814
5815 return n;
5816 }