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
86 // put OS-includes here
87 # include <sys/types.h>
88 # include <sys/mman.h>
89 # include <sys/stat.h>
90 # include <sys/select.h>
91 # include <pthread.h>
92 # include <signal.h>
93 # include <errno.h>
94 # include <dlfcn.h>
95 # include <stdio.h>
96 # include <unistd.h>
97 # include <sys/resource.h>
98 # include <pthread.h>
99 # include <sys/stat.h>
100 # include <sys/time.h>
101 # include <sys/times.h>
102 # include <sys/utsname.h>
103 # include <sys/socket.h>
104 # include <sys/wait.h>
105 # include <time.h>
106 # include <pwd.h>
107 # include <poll.h>
108 # include <semaphore.h>
109 # include <fcntl.h>
110 # include <string.h>
111 #ifdef _ALLBSD_SOURCE
112 # include <sys/param.h>
113 # include <sys/sysctl.h>
114 #else
115 # include <syscall.h>
116 # include <sys/sysinfo.h>
117 # include <gnu/libc-version.h>
118 #endif
119 # include <sys/ipc.h>
120 # include <sys/shm.h>
121 #ifndef __APPLE__
122 # include <link.h>
123 #endif
124 # include <stdint.h>
125 # include <inttypes.h>
126 # include <sys/ioctl.h>
127
128 #if defined(__FreeBSD__) || defined(__NetBSD__)
129 # include <elf.h>
130 #endif
131
132 #ifdef __APPLE__
133 # include <mach/mach.h> // semaphore_* API
134 # include <mach-o/dyld.h>
135 # include <sys/proc_info.h>
136 # include <objc/objc-auto.h>
137 #endif
138
139 #ifndef MAP_ANONYMOUS
140 #define MAP_ANONYMOUS MAP_ANON
141 #endif
142
143 #define MAX_PATH (2 * K)
144
145 // for timer info max values which include all bits
146 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
147
148 #define LARGEPAGES_BIT (1 << 6)
149 ////////////////////////////////////////////////////////////////////////////////
150 // global variables
151 julong os::Bsd::_physical_memory = 0;
152
153 #ifndef _ALLBSD_SOURCE
154 address os::Bsd::_initial_thread_stack_bottom = NULL;
155 uintptr_t os::Bsd::_initial_thread_stack_size = 0;
156 #endif
157
158 int (*os::Bsd::_clock_gettime)(clockid_t, struct timespec *) = NULL;
159 #ifndef _ALLBSD_SOURCE
160 int (*os::Bsd::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
161 Mutex* os::Bsd::_createThread_lock = NULL;
162 #endif
163 pthread_t os::Bsd::_main_thread;
164 int os::Bsd::_page_size = -1;
165 #ifndef _ALLBSD_SOURCE
166 bool os::Bsd::_is_floating_stack = false;
167 bool os::Bsd::_is_NPTL = false;
168 bool os::Bsd::_supports_fast_thread_cpu_time = false;
169 const char * os::Bsd::_glibc_version = NULL;
170 const char * os::Bsd::_libpthread_version = NULL;
171 #endif
172
173 static jlong initial_time_count=0;
174
175 static int clock_tics_per_sec = 100;
176
177 // For diagnostics to print a message once. see run_periodic_checks
178 static sigset_t check_signal_done;
179 static bool check_signals = true;;
180
181 static pid_t _initial_pid = 0;
182
183 /* Signal number used to suspend/resume a thread */
184
185 /* do not use any signal number less than SIGSEGV, see 4355769 */
186 static int SR_signum = SIGUSR2;
187 sigset_t SR_sigset;
188
189
190 ////////////////////////////////////////////////////////////////////////////////
191 // utility functions
192
193 static int SR_initialize();
194 static int SR_finalize();
195
196 julong os::available_memory() {
197 return Bsd::available_memory();
198 }
199
200 julong os::Bsd::available_memory() {
201 #ifdef _ALLBSD_SOURCE
202 // XXXBSD: this is just a stopgap implementation
203 return physical_memory() >> 2;
204 #else
205 // values in struct sysinfo are "unsigned long"
206 struct sysinfo si;
207 sysinfo(&si);
208
209 return (julong)si.freeram * si.mem_unit;
210 #endif
211 }
212
213 julong os::physical_memory() {
214 return Bsd::physical_memory();
215 }
216
217 julong os::allocatable_physical_memory(julong size) {
218 #ifdef _LP64
219 return size;
220 #else
221 julong result = MIN2(size, (julong)3800*M);
222 if (!is_allocatable(result)) {
223 // See comments under solaris for alignment considerations
224 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
225 result = MIN2(size, reasonable_size);
226 }
227 return result;
228 #endif // _LP64
229 }
230
231 ////////////////////////////////////////////////////////////////////////////////
232 // environment support
233
234 bool os::getenv(const char* name, char* buf, int len) {
235 const char* val = ::getenv(name);
236 if (val != NULL && strlen(val) < (size_t)len) {
237 strcpy(buf, val);
238 return true;
239 }
240 if (len > 0) buf[0] = 0; // return a null string
241 return false;
242 }
243
244
245 // Return true if user is running as root.
246
247 bool os::have_special_privileges() {
248 static bool init = false;
249 static bool privileges = false;
250 if (!init) {
251 privileges = (getuid() != geteuid()) || (getgid() != getegid());
252 init = true;
253 }
254 return privileges;
255 }
256
257
258 #ifndef _ALLBSD_SOURCE
259 #ifndef SYS_gettid
260 // i386: 224, ia64: 1105, amd64: 186, sparc 143
261 #ifdef __ia64__
262 #define SYS_gettid 1105
263 #elif __i386__
264 #define SYS_gettid 224
265 #elif __amd64__
266 #define SYS_gettid 186
267 #elif __sparc__
268 #define SYS_gettid 143
269 #else
270 #error define gettid for the arch
271 #endif
272 #endif
273 #endif
274
275 // Cpu architecture string
276 #if defined(ZERO)
277 static char cpu_arch[] = ZERO_LIBARCH;
278 #elif defined(IA64)
279 static char cpu_arch[] = "ia64";
280 #elif defined(IA32)
281 static char cpu_arch[] = "i386";
282 #elif defined(AMD64)
283 static char cpu_arch[] = "amd64";
284 #elif defined(ARM)
285 static char cpu_arch[] = "arm";
286 #elif defined(PPC)
287 static char cpu_arch[] = "ppc";
288 #elif defined(SPARC)
289 # ifdef _LP64
290 static char cpu_arch[] = "sparcv9";
291 # else
292 static char cpu_arch[] = "sparc";
293 # endif
294 #else
295 #error Add appropriate cpu_arch setting
296 #endif
297
298 // Compiler variant
299 #ifdef COMPILER2
300 #define COMPILER_VARIANT "server"
301 #else
302 #define COMPILER_VARIANT "client"
303 #endif
304
305 #ifndef _ALLBSD_SOURCE
306 // pid_t gettid()
307 //
308 // Returns the kernel thread id of the currently running thread. Kernel
309 // thread id is used to access /proc.
310 //
311 // (Note that getpid() on BsdThreads returns kernel thread id too; but
312 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
313 //
314 pid_t os::Bsd::gettid() {
315 int rslt = syscall(SYS_gettid);
316 if (rslt == -1) {
317 // old kernel, no NPTL support
318 return getpid();
319 } else {
320 return (pid_t)rslt;
321 }
322 }
323
324 // Most versions of bsd have a bug where the number of processors are
325 // determined by looking at the /proc file system. In a chroot environment,
326 // the system call returns 1. This causes the VM to act as if it is
327 // a single processor and elide locking (see is_MP() call).
328 static bool unsafe_chroot_detected = false;
329 static const char *unstable_chroot_error = "/proc file system not found.\n"
330 "Java may be unstable running multithreaded in a chroot "
331 "environment on Bsd when /proc filesystem is not mounted.";
332 #endif
333
334 #ifdef _ALLBSD_SOURCE
335 void os::Bsd::initialize_system_info() {
336 int mib[2];
337 size_t len;
338 int cpu_val;
339 u_long mem_val;
340
341 /* get processors count via hw.ncpus sysctl */
342 mib[0] = CTL_HW;
343 mib[1] = HW_NCPU;
344 len = sizeof(cpu_val);
345 if (sysctl(mib, 2, &cpu_val, &len, NULL, 0) != -1 && cpu_val >= 1) {
346 set_processor_count(cpu_val);
347 }
348 else {
349 set_processor_count(1); // fallback
350 }
351
352 /* get physical memory via hw.usermem sysctl (hw.usermem is used
353 * instead of hw.physmem because we need size of allocatable memory
354 */
355 mib[0] = CTL_HW;
356 mib[1] = HW_USERMEM;
357 len = sizeof(mem_val);
358 if (sysctl(mib, 2, &mem_val, &len, NULL, 0) != -1)
359 _physical_memory = mem_val;
360 else
361 _physical_memory = 256*1024*1024; // fallback (XXXBSD?)
362
363 #ifdef __OpenBSD__
364 {
365 // limit _physical_memory memory view on OpenBSD since
366 // datasize rlimit restricts us anyway.
367 struct rlimit limits;
368 getrlimit(RLIMIT_DATA, &limits);
369 _physical_memory = MIN2(_physical_memory, (julong)limits.rlim_cur);
370 }
371 #endif
372 }
373 #else
374 void os::Bsd::initialize_system_info() {
375 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
376 if (processor_count() == 1) {
377 pid_t pid = os::Bsd::gettid();
378 char fname[32];
379 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
380 FILE *fp = fopen(fname, "r");
381 if (fp == NULL) {
382 unsafe_chroot_detected = true;
383 } else {
384 fclose(fp);
385 }
386 }
387 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
388 assert(processor_count() > 0, "bsd error");
389 }
390 #endif
391
392 #ifdef __APPLE__
393 static const char *get_home() {
394 const char *home_dir = ::getenv("HOME");
395 if ((home_dir == NULL) || (*home_dir == '\0')) {
396 struct passwd *passwd_info = getpwuid(geteuid());
397 if (passwd_info != NULL) {
398 home_dir = passwd_info->pw_dir;
399 }
400 }
401
402 return home_dir;
403 }
404 #endif
405
406 void os::init_system_properties_values() {
407 // char arch[12];
408 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
409
410 // The next steps are taken in the product version:
411 //
412 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
413 // This library should be located at:
414 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
415 //
416 // If "/jre/lib/" appears at the right place in the path, then we
417 // assume libjvm[_g].so is installed in a JDK and we use this path.
418 //
419 // Otherwise exit with message: "Could not create the Java virtual machine."
420 //
421 // The following extra steps are taken in the debugging version:
422 //
423 // If "/jre/lib/" does NOT appear at the right place in the path
424 // instead of exit check for $JAVA_HOME environment variable.
425 //
426 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
427 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
428 // it looks like libjvm[_g].so is installed there
429 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
430 //
431 // Otherwise exit.
432 //
433 // Important note: if the location of libjvm.so changes this
434 // code needs to be changed accordingly.
435
436 // The next few definitions allow the code to be verbatim:
437 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
438 #define getenv(n) ::getenv(n)
439
440 /*
441 * See ld(1):
442 * The linker uses the following search paths to locate required
443 * shared libraries:
444 * 1: ...
445 * ...
446 * 7: The default directories, normally /lib and /usr/lib.
447 */
448 #ifndef DEFAULT_LIBPATH
449 #define DEFAULT_LIBPATH "/lib:/usr/lib"
450 #endif
451
452 #define EXTENSIONS_DIR "/lib/ext"
453 #define ENDORSED_DIR "/lib/endorsed"
454 #define REG_DIR "/usr/java/packages"
455
456 #ifdef __APPLE__
457 #define SYS_EXTENSIONS_DIR "/Library/Java/Extensions"
458 #define SYS_EXTENSIONS_DIRS SYS_EXTENSIONS_DIR ":/Network" SYS_EXTENSIONS_DIR ":/System" SYS_EXTENSIONS_DIR ":/usr/lib/java"
459 const char *user_home_dir = get_home();
460 // the null in SYS_EXTENSIONS_DIRS counts for the size of the colon after user_home_dir
461 int system_ext_size = strlen(user_home_dir) + sizeof(SYS_EXTENSIONS_DIR) +
462 sizeof(SYS_EXTENSIONS_DIRS);
463 #endif
464
465 {
466 /* sysclasspath, java_home, dll_dir */
467 {
468 char *home_path;
469 char *dll_path;
470 char *pslash;
471 char buf[MAXPATHLEN];
472 os::jvm_path(buf, sizeof(buf));
473
474 // Found the full path to libjvm.so.
475 // Now cut the path to <java_home>/jre if we can.
476 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
477 pslash = strrchr(buf, '/');
478 if (pslash != NULL)
479 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
480 dll_path = malloc(strlen(buf) + 1);
481 if (dll_path == NULL)
482 return;
483 strcpy(dll_path, buf);
484 Arguments::set_dll_dir(dll_path);
485
486 if (pslash != NULL) {
487 pslash = strrchr(buf, '/');
488 if (pslash != NULL) {
489 *pslash = '\0'; /* get rid of /<arch> (/lib on macosx) */
490 #ifndef __APPLE__
491 pslash = strrchr(buf, '/');
492 if (pslash != NULL)
493 *pslash = '\0'; /* get rid of /lib */
494 #endif
495 }
496 }
497
498 home_path = malloc(strlen(buf) + 1);
499 if (home_path == NULL)
500 return;
501 strcpy(home_path, buf);
502 Arguments::set_java_home(home_path);
503
504 if (!set_boot_path('/', ':'))
505 return;
506 }
507
508 /*
509 * Where to look for native libraries
510 *
511 * Note: Due to a legacy implementation, most of the library path
512 * is set in the launcher. This was to accomodate linking restrictions
513 * on legacy Bsd implementations (which are no longer supported).
514 * Eventually, all the library path setting will be done here.
515 *
516 * However, to prevent the proliferation of improperly built native
517 * libraries, the new path component /usr/java/packages is added here.
518 * Eventually, all the library path setting will be done here.
519 */
520 {
521 char *ld_library_path;
522
523 /*
524 * Construct the invariant part of ld_library_path. Note that the
525 * space for the colon and the trailing null are provided by the
526 * nulls included by the sizeof operator (so actually we allocate
527 * a byte more than necessary).
528 */
529 #ifdef __APPLE__
530 ld_library_path = (char *) malloc(system_ext_size);
531 sprintf(ld_library_path, "%s" SYS_EXTENSIONS_DIR ":" SYS_EXTENSIONS_DIRS, user_home_dir);
532 #else
533 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
534 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
535 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
536 #endif
537
538 /*
539 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
540 * should always exist (until the legacy problem cited above is
541 * addressed).
542 */
543 #ifdef __APPLE__
544 // Prepend the default path with the JAVA_LIBRARY_PATH so that the app launcher code can specify a directory inside an app wrapper
545 char *l = getenv("JAVA_LIBRARY_PATH");
546 if (l != NULL) {
547 char *t = ld_library_path;
548 /* That's +1 for the colon and +1 for the trailing '\0' */
549 ld_library_path = (char *) malloc(strlen(l) + 1 + strlen(t) + 1);
550 sprintf(ld_library_path, "%s:%s", l, t);
551 free(t);
552 }
553
554 char *v = getenv("DYLD_LIBRARY_PATH");
555 #else
556 char *v = getenv("LD_LIBRARY_PATH");
557 #endif
558 if (v != NULL) {
559 char *t = ld_library_path;
560 /* That's +1 for the colon and +1 for the trailing '\0' */
561 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
562 sprintf(ld_library_path, "%s:%s", v, t);
563 free(t);
564 }
565
566 #ifdef __APPLE__
567 // Apple's Java6 has "." at the beginning of java.library.path.
568 // OpenJDK on Windows has "." at the end of java.library.path.
569 // OpenJDK on Linux and Solaris don't have "." in java.library.path
570 // at all. To ease the transition from Apple's Java6 to OpenJDK7,
571 // "." is appended to the end of java.library.path. Yes, this
572 // could cause a change in behavior, but Apple's Java6 behavior
573 // can be achieved by putting "." at the beginning of the
574 // JAVA_LIBRARY_PATH environment variable.
575 {
576 char *t = ld_library_path;
577 // that's +3 for appending ":." and the trailing '\0'
578 ld_library_path = (char *) malloc(strlen(t) + 3);
579 sprintf(ld_library_path, "%s:%s", t, ".");
580 free(t);
581 }
582 #endif
583
584 Arguments::set_library_path(ld_library_path);
585 }
586
587 /*
588 * Extensions directories.
589 *
590 * Note that the space for the colon and the trailing null are provided
591 * by the nulls included by the sizeof operator (so actually one byte more
592 * than necessary is allocated).
593 */
594 {
595 #ifdef __APPLE__
596 char *buf = malloc(strlen(Arguments::get_java_home()) +
597 sizeof(EXTENSIONS_DIR) + system_ext_size);
598 sprintf(buf, "%s" SYS_EXTENSIONS_DIR ":%s" EXTENSIONS_DIR ":"
599 SYS_EXTENSIONS_DIRS, user_home_dir, Arguments::get_java_home());
600 #else
601 char *buf = malloc(strlen(Arguments::get_java_home()) +
602 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
603 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
604 Arguments::get_java_home());
605 #endif
606
607 Arguments::set_ext_dirs(buf);
608 }
609
610 /* Endorsed standards default directory. */
611 {
612 char * buf;
613 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
614 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
615 Arguments::set_endorsed_dirs(buf);
616 }
617 }
618
619 #ifdef __APPLE__
620 #undef SYS_EXTENSIONS_DIR
621 #endif
622 #undef malloc
623 #undef getenv
624 #undef EXTENSIONS_DIR
625 #undef ENDORSED_DIR
626
627 // Done
628 return;
629 }
630
631 ////////////////////////////////////////////////////////////////////////////////
632 // breakpoint support
633
634 void os::breakpoint() {
635 BREAKPOINT;
636 }
637
638 extern "C" void breakpoint() {
639 // use debugger to set breakpoint here
640 }
641
642 ////////////////////////////////////////////////////////////////////////////////
643 // signal support
644
645 debug_only(static bool signal_sets_initialized = false);
646 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
647
648 bool os::Bsd::is_sig_ignored(int sig) {
649 struct sigaction oact;
650 sigaction(sig, (struct sigaction*)NULL, &oact);
651 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
652 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
653 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
654 return true;
655 else
656 return false;
657 }
658
659 void os::Bsd::signal_sets_init() {
660 // Should also have an assertion stating we are still single-threaded.
661 assert(!signal_sets_initialized, "Already initialized");
662 // Fill in signals that are necessarily unblocked for all threads in
663 // the VM. Currently, we unblock the following signals:
664 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
665 // by -Xrs (=ReduceSignalUsage));
666 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
667 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
668 // the dispositions or masks wrt these signals.
669 // Programs embedding the VM that want to use the above signals for their
670 // own purposes must, at this time, use the "-Xrs" option to prevent
671 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
672 // (See bug 4345157, and other related bugs).
673 // In reality, though, unblocking these signals is really a nop, since
674 // these signals are not blocked by default.
675 sigemptyset(&unblocked_sigs);
676 sigemptyset(&allowdebug_blocked_sigs);
677 sigaddset(&unblocked_sigs, SIGILL);
678 sigaddset(&unblocked_sigs, SIGSEGV);
679 sigaddset(&unblocked_sigs, SIGBUS);
680 sigaddset(&unblocked_sigs, SIGFPE);
681 sigaddset(&unblocked_sigs, SR_signum);
682
683 if (!ReduceSignalUsage) {
684 if (!os::Bsd::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
685 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
686 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
687 }
688 if (!os::Bsd::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
689 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
690 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
691 }
692 if (!os::Bsd::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
693 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
694 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
695 }
696 }
697 // Fill in signals that are blocked by all but the VM thread.
698 sigemptyset(&vm_sigs);
699 if (!ReduceSignalUsage)
700 sigaddset(&vm_sigs, BREAK_SIGNAL);
701 debug_only(signal_sets_initialized = true);
702
703 }
704
705 // These are signals that are unblocked while a thread is running Java.
706 // (For some reason, they get blocked by default.)
707 sigset_t* os::Bsd::unblocked_signals() {
708 assert(signal_sets_initialized, "Not initialized");
709 return &unblocked_sigs;
710 }
711
712 // These are the signals that are blocked while a (non-VM) thread is
713 // running Java. Only the VM thread handles these signals.
714 sigset_t* os::Bsd::vm_signals() {
715 assert(signal_sets_initialized, "Not initialized");
716 return &vm_sigs;
717 }
718
719 // These are signals that are blocked during cond_wait to allow debugger in
720 sigset_t* os::Bsd::allowdebug_blocked_signals() {
721 assert(signal_sets_initialized, "Not initialized");
722 return &allowdebug_blocked_sigs;
723 }
724
725 void os::Bsd::hotspot_sigmask(Thread* thread) {
726
727 //Save caller's signal mask before setting VM signal mask
728 sigset_t caller_sigmask;
729 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
730
731 OSThread* osthread = thread->osthread();
732 osthread->set_caller_sigmask(caller_sigmask);
733
734 pthread_sigmask(SIG_UNBLOCK, os::Bsd::unblocked_signals(), NULL);
735
736 if (!ReduceSignalUsage) {
737 if (thread->is_VM_thread()) {
738 // Only the VM thread handles BREAK_SIGNAL ...
739 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
740 } else {
741 // ... all other threads block BREAK_SIGNAL
742 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
743 }
744 }
745 }
746
747 #ifndef _ALLBSD_SOURCE
748 //////////////////////////////////////////////////////////////////////////////
749 // detecting pthread library
750
751 void os::Bsd::libpthread_init() {
752 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
753 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
754 // generic name for earlier versions.
755 // Define macros here so we can build HotSpot on old systems.
756 # ifndef _CS_GNU_LIBC_VERSION
757 # define _CS_GNU_LIBC_VERSION 2
758 # endif
759 # ifndef _CS_GNU_LIBPTHREAD_VERSION
760 # define _CS_GNU_LIBPTHREAD_VERSION 3
761 # endif
762
763 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
764 if (n > 0) {
765 char *str = (char *)malloc(n);
766 confstr(_CS_GNU_LIBC_VERSION, str, n);
767 os::Bsd::set_glibc_version(str);
768 } else {
769 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
770 static char _gnu_libc_version[32];
771 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
772 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
773 os::Bsd::set_glibc_version(_gnu_libc_version);
774 }
775
776 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
777 if (n > 0) {
778 char *str = (char *)malloc(n);
779 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
780 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
781 // us "NPTL-0.29" even we are running with BsdThreads. Check if this
782 // is the case. BsdThreads has a hard limit on max number of threads.
783 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
784 // On the other hand, NPTL does not have such a limit, sysconf()
785 // will return -1 and errno is not changed. Check if it is really NPTL.
786 if (strcmp(os::Bsd::glibc_version(), "glibc 2.3.2") == 0 &&
787 strstr(str, "NPTL") &&
788 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
789 free(str);
790 os::Bsd::set_libpthread_version("bsdthreads");
791 } else {
792 os::Bsd::set_libpthread_version(str);
793 }
794 } else {
795 // glibc before 2.3.2 only has BsdThreads.
796 os::Bsd::set_libpthread_version("bsdthreads");
797 }
798
799 if (strstr(libpthread_version(), "NPTL")) {
800 os::Bsd::set_is_NPTL();
801 } else {
802 os::Bsd::set_is_BsdThreads();
803 }
804
805 // BsdThreads have two flavors: floating-stack mode, which allows variable
806 // stack size; and fixed-stack mode. NPTL is always floating-stack.
807 if (os::Bsd::is_NPTL() || os::Bsd::supports_variable_stack_size()) {
808 os::Bsd::set_is_floating_stack();
809 }
810 }
811
812 /////////////////////////////////////////////////////////////////////////////
813 // thread stack
814
815 // Force Bsd kernel to expand current thread stack. If "bottom" is close
816 // to the stack guard, caller should block all signals.
817 //
818 // MAP_GROWSDOWN:
819 // A special mmap() flag that is used to implement thread stacks. It tells
820 // kernel that the memory region should extend downwards when needed. This
821 // allows early versions of BsdThreads to only mmap the first few pages
822 // when creating a new thread. Bsd kernel will automatically expand thread
823 // stack as needed (on page faults).
824 //
825 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
826 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
827 // region, it's hard to tell if the fault is due to a legitimate stack
828 // access or because of reading/writing non-exist memory (e.g. buffer
829 // overrun). As a rule, if the fault happens below current stack pointer,
830 // Bsd kernel does not expand stack, instead a SIGSEGV is sent to the
831 // application (see Bsd kernel fault.c).
832 //
833 // This Bsd feature can cause SIGSEGV when VM bangs thread stack for
834 // stack overflow detection.
835 //
836 // Newer version of BsdThreads (since glibc-2.2, or, RH-7.x) and NPTL do
837 // not use this flag. However, the stack of initial thread is not created
838 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
839 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
840 // and then attach the thread to JVM.
841 //
842 // To get around the problem and allow stack banging on Bsd, we need to
843 // manually expand thread stack after receiving the SIGSEGV.
844 //
845 // There are two ways to expand thread stack to address "bottom", we used
846 // both of them in JVM before 1.5:
847 // 1. adjust stack pointer first so that it is below "bottom", and then
848 // touch "bottom"
849 // 2. mmap() the page in question
850 //
851 // Now alternate signal stack is gone, it's harder to use 2. For instance,
852 // if current sp is already near the lower end of page 101, and we need to
853 // call mmap() to map page 100, it is possible that part of the mmap() frame
854 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
855 // That will destroy the mmap() frame and cause VM to crash.
856 //
857 // The following code works by adjusting sp first, then accessing the "bottom"
858 // page to force a page fault. Bsd kernel will then automatically expand the
859 // stack mapping.
860 //
861 // _expand_stack_to() assumes its frame size is less than page size, which
862 // should always be true if the function is not inlined.
863
864 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
865 #define NOINLINE
866 #else
867 #define NOINLINE __attribute__ ((noinline))
868 #endif
869
870 static void _expand_stack_to(address bottom) NOINLINE;
871
872 static void _expand_stack_to(address bottom) {
873 address sp;
874 size_t size;
875 volatile char *p;
876
877 // Adjust bottom to point to the largest address within the same page, it
878 // gives us a one-page buffer if alloca() allocates slightly more memory.
879 bottom = (address)align_size_down((uintptr_t)bottom, os::Bsd::page_size());
880 bottom += os::Bsd::page_size() - 1;
881
882 // sp might be slightly above current stack pointer; if that's the case, we
883 // will alloca() a little more space than necessary, which is OK. Don't use
884 // os::current_stack_pointer(), as its result can be slightly below current
885 // stack pointer, causing us to not alloca enough to reach "bottom".
886 sp = (address)&sp;
887
888 if (sp > bottom) {
889 size = sp - bottom;
890 p = (volatile char *)alloca(size);
891 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
892 p[0] = '\0';
893 }
894 }
895
896 bool os::Bsd::manually_expand_stack(JavaThread * t, address addr) {
897 assert(t!=NULL, "just checking");
898 assert(t->osthread()->expanding_stack(), "expand should be set");
899 assert(t->stack_base() != NULL, "stack_base was not initialized");
900
901 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
902 sigset_t mask_all, old_sigset;
903 sigfillset(&mask_all);
904 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
905 _expand_stack_to(addr);
906 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
907 return true;
908 }
909 return false;
910 }
911 #endif
912
913 //////////////////////////////////////////////////////////////////////////////
914 // create new thread
915
916 static address highest_vm_reserved_address();
917
918 // check if it's safe to start a new thread
919 static bool _thread_safety_check(Thread* thread) {
920 #ifdef _ALLBSD_SOURCE
921 return true;
922 #else
923 if (os::Bsd::is_BsdThreads() && !os::Bsd::is_floating_stack()) {
924 // Fixed stack BsdThreads (SuSE Bsd/x86, and some versions of Redhat)
925 // Heap is mmap'ed at lower end of memory space. Thread stacks are
926 // allocated (MAP_FIXED) from high address space. Every thread stack
927 // occupies a fixed size slot (usually 2Mbytes, but user can change
928 // it to other values if they rebuild BsdThreads).
929 //
930 // Problem with MAP_FIXED is that mmap() can still succeed even part of
931 // the memory region has already been mmap'ed. That means if we have too
932 // many threads and/or very large heap, eventually thread stack will
933 // collide with heap.
934 //
935 // Here we try to prevent heap/stack collision by comparing current
936 // stack bottom with the highest address that has been mmap'ed by JVM
937 // plus a safety margin for memory maps created by native code.
938 //
939 // This feature can be disabled by setting ThreadSafetyMargin to 0
940 //
941 if (ThreadSafetyMargin > 0) {
942 address stack_bottom = os::current_stack_base() - os::current_stack_size();
943
944 // not safe if our stack extends below the safety margin
945 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
946 } else {
947 return true;
948 }
949 } else {
950 // Floating stack BsdThreads or NPTL:
951 // Unlike fixed stack BsdThreads, thread stacks are not MAP_FIXED. When
952 // there's not enough space left, pthread_create() will fail. If we come
953 // here, that means enough space has been reserved for stack.
954 return true;
955 }
956 #endif
957 }
958
959 #ifdef __APPLE__
960 // library handle for calling objc_registerThreadWithCollector()
961 // without static linking to the libobjc library
962 #define OBJC_LIB "/usr/lib/libobjc.dylib"
963 #define OBJC_GCREGISTER "objc_registerThreadWithCollector"
964 typedef void (*objc_registerThreadWithCollector_t)();
965 extern "C" objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction;
966 objc_registerThreadWithCollector_t objc_registerThreadWithCollectorFunction = NULL;
967 #endif
968
969 // Thread start routine for all newly created threads
970 static void *java_start(Thread *thread) {
971 // Try to randomize the cache line index of hot stack frames.
972 // This helps when threads of the same stack traces evict each other's
973 // cache lines. The threads can be either from the same JVM instance, or
974 // from different JVM instances. The benefit is especially true for
975 // processors with hyperthreading technology.
976 static int counter = 0;
977 int pid = os::current_process_id();
978 alloca(((pid ^ counter++) & 7) * 128);
979
980 ThreadLocalStorage::set_thread(thread);
981
982 OSThread* osthread = thread->osthread();
983 Monitor* sync = osthread->startThread_lock();
984
985 // non floating stack BsdThreads needs extra check, see above
986 if (!_thread_safety_check(thread)) {
987 // notify parent thread
988 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
989 osthread->set_state(ZOMBIE);
990 sync->notify_all();
991 return NULL;
992 }
993
994 #ifdef _ALLBSD_SOURCE
995 #ifdef __APPLE__
996 // thread_id is mach thread on macos
997 osthread->set_thread_id(::mach_thread_self());
998 #else
999 // thread_id is pthread_id on BSD
1000 osthread->set_thread_id(::pthread_self());
1001 #endif
1002 #else
1003 // thread_id is kernel thread id (similar to Solaris LWP id)
1004 osthread->set_thread_id(os::Bsd::gettid());
1005
1006 if (UseNUMA) {
1007 int lgrp_id = os::numa_get_group_id();
1008 if (lgrp_id != -1) {
1009 thread->set_lgrp_id(lgrp_id);
1010 }
1011 }
1012 #endif
1013 // initialize signal mask for this thread
1014 os::Bsd::hotspot_sigmask(thread);
1015
1016 // initialize floating point control register
1017 os::Bsd::init_thread_fpu_state();
1018
1019 #ifdef __APPLE__
1020 // register thread with objc gc
1021 if (objc_registerThreadWithCollectorFunction != NULL) {
1022 objc_registerThreadWithCollectorFunction();
1023 }
1024 #endif
1025
1026 // handshaking with parent thread
1027 {
1028 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
1029
1030 // notify parent thread
1031 osthread->set_state(INITIALIZED);
1032 sync->notify_all();
1033
1034 // wait until os::start_thread()
1035 while (osthread->get_state() == INITIALIZED) {
1036 sync->wait(Mutex::_no_safepoint_check_flag);
1037 }
1038 }
1039
1040 // call one more level start routine
1041 thread->run();
1042
1043 return 0;
1044 }
1045
1046 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1047 assert(thread->osthread() == NULL, "caller responsible");
1048
1049 // Allocate the OSThread object
1050 OSThread* osthread = new OSThread(NULL, NULL);
1051 if (osthread == NULL) {
1052 return false;
1053 }
1054
1055 // set the correct thread state
1056 osthread->set_thread_type(thr_type);
1057
1058 // Initial state is ALLOCATED but not INITIALIZED
1059 osthread->set_state(ALLOCATED);
1060
1061 thread->set_osthread(osthread);
1062
1063 // init thread attributes
1064 pthread_attr_t attr;
1065 pthread_attr_init(&attr);
1066 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1067
1068 // stack size
1069 if (os::Bsd::supports_variable_stack_size()) {
1070 // calculate stack size if it's not specified by caller
1071 if (stack_size == 0) {
1072 stack_size = os::Bsd::default_stack_size(thr_type);
1073
1074 switch (thr_type) {
1075 case os::java_thread:
1076 // Java threads use ThreadStackSize which default value can be
1077 // changed with the flag -Xss
1078 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
1079 stack_size = JavaThread::stack_size_at_create();
1080 break;
1081 case os::compiler_thread:
1082 if (CompilerThreadStackSize > 0) {
1083 stack_size = (size_t)(CompilerThreadStackSize * K);
1084 break;
1085 } // else fall through:
1086 // use VMThreadStackSize if CompilerThreadStackSize is not defined
1087 case os::vm_thread:
1088 case os::pgc_thread:
1089 case os::cgc_thread:
1090 case os::watcher_thread:
1091 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1092 break;
1093 }
1094 }
1095
1096 stack_size = MAX2(stack_size, os::Bsd::min_stack_allowed);
1097 pthread_attr_setstacksize(&attr, stack_size);
1098 } else {
1099 // let pthread_create() pick the default value.
1100 }
1101
1102 #ifndef _ALLBSD_SOURCE
1103 // glibc guard page
1104 pthread_attr_setguardsize(&attr, os::Bsd::default_guard_size(thr_type));
1105 #endif
1106
1107 ThreadState state;
1108
1109 {
1110
1111 #ifndef _ALLBSD_SOURCE
1112 // Serialize thread creation if we are running with fixed stack BsdThreads
1113 bool lock = os::Bsd::is_BsdThreads() && !os::Bsd::is_floating_stack();
1114 if (lock) {
1115 os::Bsd::createThread_lock()->lock_without_safepoint_check();
1116 }
1117 #endif
1118
1119 pthread_t tid;
1120 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
1121
1122 pthread_attr_destroy(&attr);
1123
1124 if (ret != 0) {
1125 if (PrintMiscellaneous && (Verbose || WizardMode)) {
1126 perror("pthread_create()");
1127 }
1128 // Need to clean up stuff we've allocated so far
1129 thread->set_osthread(NULL);
1130 delete osthread;
1131 #ifndef _ALLBSD_SOURCE
1132 if (lock) os::Bsd::createThread_lock()->unlock();
1133 #endif
1134 return false;
1135 }
1136
1137 // Store pthread info into the OSThread
1138 osthread->set_pthread_id(tid);
1139
1140 // Wait until child thread is either initialized or aborted
1141 {
1142 Monitor* sync_with_child = osthread->startThread_lock();
1143 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1144 while ((state = osthread->get_state()) == ALLOCATED) {
1145 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
1146 }
1147 }
1148
1149 #ifndef _ALLBSD_SOURCE
1150 if (lock) {
1151 os::Bsd::createThread_lock()->unlock();
1152 }
1153 #endif
1154 }
1155
1156 // Aborted due to thread limit being reached
1157 if (state == ZOMBIE) {
1158 thread->set_osthread(NULL);
1159 delete osthread;
1160 return false;
1161 }
1162
1163 // The thread is returned suspended (in state INITIALIZED),
1164 // and is started higher up in the call chain
1165 assert(state == INITIALIZED, "race condition");
1166 return true;
1167 }
1168
1169 /////////////////////////////////////////////////////////////////////////////
1170 // attach existing thread
1171
1172 // bootstrap the main thread
1173 bool os::create_main_thread(JavaThread* thread) {
1174 assert(os::Bsd::_main_thread == pthread_self(), "should be called inside main thread");
1175 return create_attached_thread(thread);
1176 }
1177
1178 bool os::create_attached_thread(JavaThread* thread) {
1179 #ifdef ASSERT
1180 thread->verify_not_published();
1181 #endif
1182
1183 // Allocate the OSThread object
1184 OSThread* osthread = new OSThread(NULL, NULL);
1185
1186 if (osthread == NULL) {
1187 return false;
1188 }
1189
1190 // Store pthread info into the OSThread
1191 #ifdef _ALLBSD_SOURCE
1192 #ifdef __APPLE__
1193 osthread->set_thread_id(::mach_thread_self());
1194 #else
1195 osthread->set_thread_id(::pthread_self());
1196 #endif
1197 #else
1198 osthread->set_thread_id(os::Bsd::gettid());
1199 #endif
1200 osthread->set_pthread_id(::pthread_self());
1201
1202 // initialize floating point control register
1203 os::Bsd::init_thread_fpu_state();
1204
1205 // Initial thread state is RUNNABLE
1206 osthread->set_state(RUNNABLE);
1207
1208 thread->set_osthread(osthread);
1209
1210 #ifndef _ALLBSD_SOURCE
1211 if (UseNUMA) {
1212 int lgrp_id = os::numa_get_group_id();
1213 if (lgrp_id != -1) {
1214 thread->set_lgrp_id(lgrp_id);
1215 }
1216 }
1217
1218 if (os::Bsd::is_initial_thread()) {
1219 // If current thread is initial thread, its stack is mapped on demand,
1220 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1221 // the entire stack region to avoid SEGV in stack banging.
1222 // It is also useful to get around the heap-stack-gap problem on SuSE
1223 // kernel (see 4821821 for details). We first expand stack to the top
1224 // of yellow zone, then enable stack yellow zone (order is significant,
1225 // enabling yellow zone first will crash JVM on SuSE Bsd), so there
1226 // is no gap between the last two virtual memory regions.
1227
1228 JavaThread *jt = (JavaThread *)thread;
1229 address addr = jt->stack_yellow_zone_base();
1230 assert(addr != NULL, "initialization problem?");
1231 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1232
1233 osthread->set_expanding_stack();
1234 os::Bsd::manually_expand_stack(jt, addr);
1235 osthread->clear_expanding_stack();
1236 }
1237 #endif
1238
1239 // initialize signal mask for this thread
1240 // and save the caller's signal mask
1241 os::Bsd::hotspot_sigmask(thread);
1242
1243 return true;
1244 }
1245
1246 void os::pd_start_thread(Thread* thread) {
1247 OSThread * osthread = thread->osthread();
1248 assert(osthread->get_state() != INITIALIZED, "just checking");
1249 Monitor* sync_with_child = osthread->startThread_lock();
1250 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1251 sync_with_child->notify();
1252 }
1253
1254 // Free Bsd resources related to the OSThread
1255 void os::free_thread(OSThread* osthread) {
1256 assert(osthread != NULL, "osthread not set");
1257
1258 if (Thread::current()->osthread() == osthread) {
1259 // Restore caller's signal mask
1260 sigset_t sigmask = osthread->caller_sigmask();
1261 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1262 }
1263
1264 delete osthread;
1265 }
1266
1267 //////////////////////////////////////////////////////////////////////////////
1268 // thread local storage
1269
1270 int os::allocate_thread_local_storage() {
1271 pthread_key_t key;
1272 int rslt = pthread_key_create(&key, NULL);
1273 assert(rslt == 0, "cannot allocate thread local storage");
1274 return (int)key;
1275 }
1276
1277 // Note: This is currently not used by VM, as we don't destroy TLS key
1278 // on VM exit.
1279 void os::free_thread_local_storage(int index) {
1280 int rslt = pthread_key_delete((pthread_key_t)index);
1281 assert(rslt == 0, "invalid index");
1282 }
1283
1284 void os::thread_local_storage_at_put(int index, void* value) {
1285 int rslt = pthread_setspecific((pthread_key_t)index, value);
1286 assert(rslt == 0, "pthread_setspecific failed");
1287 }
1288
1289 extern "C" Thread* get_thread() {
1290 return ThreadLocalStorage::thread();
1291 }
1292
1293 //////////////////////////////////////////////////////////////////////////////
1294 // initial thread
1295
1296 #ifndef _ALLBSD_SOURCE
1297 // Check if current thread is the initial thread, similar to Solaris thr_main.
1298 bool os::Bsd::is_initial_thread(void) {
1299 char dummy;
1300 // If called before init complete, thread stack bottom will be null.
1301 // Can be called if fatal error occurs before initialization.
1302 if (initial_thread_stack_bottom() == NULL) return false;
1303 assert(initial_thread_stack_bottom() != NULL &&
1304 initial_thread_stack_size() != 0,
1305 "os::init did not locate initial thread's stack region");
1306 if ((address)&dummy >= initial_thread_stack_bottom() &&
1307 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1308 return true;
1309 else return false;
1310 }
1311
1312 // Find the virtual memory area that contains addr
1313 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1314 FILE *fp = fopen("/proc/self/maps", "r");
1315 if (fp) {
1316 address low, high;
1317 while (!feof(fp)) {
1318 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1319 if (low <= addr && addr < high) {
1320 if (vma_low) *vma_low = low;
1321 if (vma_high) *vma_high = high;
1322 fclose (fp);
1323 return true;
1324 }
1325 }
1326 for (;;) {
1327 int ch = fgetc(fp);
1328 if (ch == EOF || ch == (int)'\n') break;
1329 }
1330 }
1331 fclose(fp);
1332 }
1333 return false;
1334 }
1335
1336 // Locate initial thread stack. This special handling of initial thread stack
1337 // is needed because pthread_getattr_np() on most (all?) Bsd distros returns
1338 // bogus value for initial thread.
1339 void os::Bsd::capture_initial_stack(size_t max_size) {
1340 // stack size is the easy part, get it from RLIMIT_STACK
1341 size_t stack_size;
1342 struct rlimit rlim;
1343 getrlimit(RLIMIT_STACK, &rlim);
1344 stack_size = rlim.rlim_cur;
1345
1346 // 6308388: a bug in ld.so will relocate its own .data section to the
1347 // lower end of primordial stack; reduce ulimit -s value a little bit
1348 // so we won't install guard page on ld.so's data section.
1349 stack_size -= 2 * page_size();
1350
1351 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1352 // 7.1, in both cases we will get 2G in return value.
1353 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1354 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1355 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1356 // in case other parts in glibc still assumes 2M max stack size.
1357 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1358 #ifndef IA64
1359 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1360 #else
1361 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1362 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1363 #endif
1364
1365 // Try to figure out where the stack base (top) is. This is harder.
1366 //
1367 // When an application is started, glibc saves the initial stack pointer in
1368 // a global variable "__libc_stack_end", which is then used by system
1369 // libraries. __libc_stack_end should be pretty close to stack top. The
1370 // variable is available since the very early days. However, because it is
1371 // a private interface, it could disappear in the future.
1372 //
1373 // Bsd kernel saves start_stack information in /proc/<pid>/stat. Similar
1374 // to __libc_stack_end, it is very close to stack top, but isn't the real
1375 // stack top. Note that /proc may not exist if VM is running as a chroot
1376 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1377 // /proc/<pid>/stat could change in the future (though unlikely).
1378 //
1379 // We try __libc_stack_end first. If that doesn't work, look for
1380 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1381 // as a hint, which should work well in most cases.
1382
1383 uintptr_t stack_start;
1384
1385 // try __libc_stack_end first
1386 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1387 if (p && *p) {
1388 stack_start = *p;
1389 } else {
1390 // see if we can get the start_stack field from /proc/self/stat
1391 FILE *fp;
1392 int pid;
1393 char state;
1394 int ppid;
1395 int pgrp;
1396 int session;
1397 int nr;
1398 int tpgrp;
1399 unsigned long flags;
1400 unsigned long minflt;
1401 unsigned long cminflt;
1402 unsigned long majflt;
1403 unsigned long cmajflt;
1404 unsigned long utime;
1405 unsigned long stime;
1406 long cutime;
1407 long cstime;
1408 long prio;
1409 long nice;
1410 long junk;
1411 long it_real;
1412 uintptr_t start;
1413 uintptr_t vsize;
1414 intptr_t rss;
1415 uintptr_t rsslim;
1416 uintptr_t scodes;
1417 uintptr_t ecode;
1418 int i;
1419
1420 // Figure what the primordial thread stack base is. Code is inspired
1421 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1422 // followed by command name surrounded by parentheses, state, etc.
1423 char stat[2048];
1424 int statlen;
1425
1426 fp = fopen("/proc/self/stat", "r");
1427 if (fp) {
1428 statlen = fread(stat, 1, 2047, fp);
1429 stat[statlen] = '\0';
1430 fclose(fp);
1431
1432 // Skip pid and the command string. Note that we could be dealing with
1433 // weird command names, e.g. user could decide to rename java launcher
1434 // to "java 1.4.2 :)", then the stat file would look like
1435 // 1234 (java 1.4.2 :)) R ... ...
1436 // We don't really need to know the command string, just find the last
1437 // occurrence of ")" and then start parsing from there. See bug 4726580.
1438 char * s = strrchr(stat, ')');
1439
1440 i = 0;
1441 if (s) {
1442 // Skip blank chars
1443 do s++; while (isspace(*s));
1444
1445 #define _UFM UINTX_FORMAT
1446 #define _DFM INTX_FORMAT
1447
1448 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1449 /* 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 */
1450 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,
1451 &state, /* 3 %c */
1452 &ppid, /* 4 %d */
1453 &pgrp, /* 5 %d */
1454 &session, /* 6 %d */
1455 &nr, /* 7 %d */
1456 &tpgrp, /* 8 %d */
1457 &flags, /* 9 %lu */
1458 &minflt, /* 10 %lu */
1459 &cminflt, /* 11 %lu */
1460 &majflt, /* 12 %lu */
1461 &cmajflt, /* 13 %lu */
1462 &utime, /* 14 %lu */
1463 &stime, /* 15 %lu */
1464 &cutime, /* 16 %ld */
1465 &cstime, /* 17 %ld */
1466 &prio, /* 18 %ld */
1467 &nice, /* 19 %ld */
1468 &junk, /* 20 %ld */
1469 &it_real, /* 21 %ld */
1470 &start, /* 22 UINTX_FORMAT */
1471 &vsize, /* 23 UINTX_FORMAT */
1472 &rss, /* 24 INTX_FORMAT */
1473 &rsslim, /* 25 UINTX_FORMAT */
1474 &scodes, /* 26 UINTX_FORMAT */
1475 &ecode, /* 27 UINTX_FORMAT */
1476 &stack_start); /* 28 UINTX_FORMAT */
1477 }
1478
1479 #undef _UFM
1480 #undef _DFM
1481
1482 if (i != 28 - 2) {
1483 assert(false, "Bad conversion from /proc/self/stat");
1484 // product mode - assume we are the initial thread, good luck in the
1485 // embedded case.
1486 warning("Can't detect initial thread stack location - bad conversion");
1487 stack_start = (uintptr_t) &rlim;
1488 }
1489 } else {
1490 // For some reason we can't open /proc/self/stat (for example, running on
1491 // FreeBSD with a Bsd emulator, or inside chroot), this should work for
1492 // most cases, so don't abort:
1493 warning("Can't detect initial thread stack location - no /proc/self/stat");
1494 stack_start = (uintptr_t) &rlim;
1495 }
1496 }
1497
1498 // Now we have a pointer (stack_start) very close to the stack top, the
1499 // next thing to do is to figure out the exact location of stack top. We
1500 // can find out the virtual memory area that contains stack_start by
1501 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1502 // and its upper limit is the real stack top. (again, this would fail if
1503 // running inside chroot, because /proc may not exist.)
1504
1505 uintptr_t stack_top;
1506 address low, high;
1507 if (find_vma((address)stack_start, &low, &high)) {
1508 // success, "high" is the true stack top. (ignore "low", because initial
1509 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1510 stack_top = (uintptr_t)high;
1511 } else {
1512 // failed, likely because /proc/self/maps does not exist
1513 warning("Can't detect initial thread stack location - find_vma failed");
1514 // best effort: stack_start is normally within a few pages below the real
1515 // stack top, use it as stack top, and reduce stack size so we won't put
1516 // guard page outside stack.
1517 stack_top = stack_start;
1518 stack_size -= 16 * page_size();
1519 }
1520
1521 // stack_top could be partially down the page so align it
1522 stack_top = align_size_up(stack_top, page_size());
1523
1524 if (max_size && stack_size > max_size) {
1525 _initial_thread_stack_size = max_size;
1526 } else {
1527 _initial_thread_stack_size = stack_size;
1528 }
1529
1530 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1531 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1532 }
1533 #endif
1534
1535 ////////////////////////////////////////////////////////////////////////////////
1536 // time support
1537
1538 // Time since start-up in seconds to a fine granularity.
1539 // Used by VMSelfDestructTimer and the MemProfiler.
1540 double os::elapsedTime() {
1541
1542 return (double)(os::elapsed_counter()) * 0.000001;
1543 }
1544
1545 jlong os::elapsed_counter() {
1546 timeval time;
1547 int status = gettimeofday(&time, NULL);
1548 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1549 }
1550
1551 jlong os::elapsed_frequency() {
1552 return (1000 * 1000);
1553 }
1554
1555 // XXX: For now, code this as if BSD does not support vtime.
1556 bool os::supports_vtime() { return false; }
1557 bool os::enable_vtime() { return false; }
1558 bool os::vtime_enabled() { return false; }
1559 double os::elapsedVTime() {
1560 // better than nothing, but not much
1561 return elapsedTime();
1562 }
1563
1564 jlong os::javaTimeMillis() {
1565 timeval time;
1566 int status = gettimeofday(&time, NULL);
1567 assert(status != -1, "bsd error");
1568 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1569 }
1570
1571 #ifndef CLOCK_MONOTONIC
1572 #define CLOCK_MONOTONIC (1)
1573 #endif
1574
1575 #ifdef __APPLE__
1576 void os::Bsd::clock_init() {
1577 // XXXDARWIN: Investigate replacement monotonic clock
1578 }
1579 #elif defined(_ALLBSD_SOURCE)
1580 void os::Bsd::clock_init() {
1581 struct timespec res;
1582 struct timespec tp;
1583 if (::clock_getres(CLOCK_MONOTONIC, &res) == 0 &&
1584 ::clock_gettime(CLOCK_MONOTONIC, &tp) == 0) {
1585 // yes, monotonic clock is supported
1586 _clock_gettime = ::clock_gettime;
1587 }
1588 }
1589 #else
1590 void os::Bsd::clock_init() {
1591 // we do dlopen's in this particular order due to bug in bsd
1592 // dynamical loader (see 6348968) leading to crash on exit
1593 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1594 if (handle == NULL) {
1595 handle = dlopen("librt.so", RTLD_LAZY);
1596 }
1597
1598 if (handle) {
1599 int (*clock_getres_func)(clockid_t, struct timespec*) =
1600 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1601 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1602 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1603 if (clock_getres_func && clock_gettime_func) {
1604 // See if monotonic clock is supported by the kernel. Note that some
1605 // early implementations simply return kernel jiffies (updated every
1606 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1607 // for nano time (though the monotonic property is still nice to have).
1608 // It's fixed in newer kernels, however clock_getres() still returns
1609 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1610 // resolution for now. Hopefully as people move to new kernels, this
1611 // won't be a problem.
1612 struct timespec res;
1613 struct timespec tp;
1614 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1615 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1616 // yes, monotonic clock is supported
1617 _clock_gettime = clock_gettime_func;
1618 } else {
1619 // close librt if there is no monotonic clock
1620 dlclose(handle);
1621 }
1622 }
1623 }
1624 }
1625 #endif
1626
1627 #ifndef _ALLBSD_SOURCE
1628 #ifndef SYS_clock_getres
1629
1630 #if defined(IA32) || defined(AMD64)
1631 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1632 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1633 #else
1634 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1635 #define sys_clock_getres(x,y) -1
1636 #endif
1637
1638 #else
1639 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1640 #endif
1641
1642 void os::Bsd::fast_thread_clock_init() {
1643 if (!UseBsdPosixThreadCPUClocks) {
1644 return;
1645 }
1646 clockid_t clockid;
1647 struct timespec tp;
1648 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1649 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1650
1651 // Switch to using fast clocks for thread cpu time if
1652 // the sys_clock_getres() returns 0 error code.
1653 // Note, that some kernels may support the current thread
1654 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1655 // returned by the pthread_getcpuclockid().
1656 // If the fast Posix clocks are supported then the sys_clock_getres()
1657 // must return at least tp.tv_sec == 0 which means a resolution
1658 // better than 1 sec. This is extra check for reliability.
1659
1660 if(pthread_getcpuclockid_func &&
1661 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1662 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1663
1664 _supports_fast_thread_cpu_time = true;
1665 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1666 }
1667 }
1668 #endif
1669
1670 jlong os::javaTimeNanos() {
1671 if (Bsd::supports_monotonic_clock()) {
1672 struct timespec tp;
1673 int status = Bsd::clock_gettime(CLOCK_MONOTONIC, &tp);
1674 assert(status == 0, "gettime error");
1675 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1676 return result;
1677 } else {
1678 timeval time;
1679 int status = gettimeofday(&time, NULL);
1680 assert(status != -1, "bsd error");
1681 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1682 return 1000 * usecs;
1683 }
1684 }
1685
1686 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1687 if (Bsd::supports_monotonic_clock()) {
1688 info_ptr->max_value = ALL_64_BITS;
1689
1690 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1691 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1692 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1693 } else {
1694 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1695 info_ptr->max_value = ALL_64_BITS;
1696
1697 // gettimeofday is a real time clock so it skips
1698 info_ptr->may_skip_backward = true;
1699 info_ptr->may_skip_forward = true;
1700 }
1701
1702 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1703 }
1704
1705 // Return the real, user, and system times in seconds from an
1706 // arbitrary fixed point in the past.
1707 bool os::getTimesSecs(double* process_real_time,
1708 double* process_user_time,
1709 double* process_system_time) {
1710 struct tms ticks;
1711 clock_t real_ticks = times(&ticks);
1712
1713 if (real_ticks == (clock_t) (-1)) {
1714 return false;
1715 } else {
1716 double ticks_per_second = (double) clock_tics_per_sec;
1717 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1718 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1719 *process_real_time = ((double) real_ticks) / ticks_per_second;
1720
1721 return true;
1722 }
1723 }
1724
1725
1726 char * os::local_time_string(char *buf, size_t buflen) {
1727 struct tm t;
1728 time_t long_time;
1729 time(&long_time);
1730 localtime_r(&long_time, &t);
1731 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1732 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1733 t.tm_hour, t.tm_min, t.tm_sec);
1734 return buf;
1735 }
1736
1737 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1738 return localtime_r(clock, res);
1739 }
1740
1741 ////////////////////////////////////////////////////////////////////////////////
1742 // runtime exit support
1743
1744 // Note: os::shutdown() might be called very early during initialization, or
1745 // called from signal handler. Before adding something to os::shutdown(), make
1746 // sure it is async-safe and can handle partially initialized VM.
1747 void os::shutdown() {
1748
1749 // allow PerfMemory to attempt cleanup of any persistent resources
1750 perfMemory_exit();
1751
1752 // needs to remove object in file system
1753 AttachListener::abort();
1754
1755 // flush buffered output, finish log files
1756 ostream_abort();
1757
1758 // Check for abort hook
1759 abort_hook_t abort_hook = Arguments::abort_hook();
1760 if (abort_hook != NULL) {
1761 abort_hook();
1762 }
1763
1764 }
1765
1766 // Note: os::abort() might be called very early during initialization, or
1767 // called from signal handler. Before adding something to os::abort(), make
1768 // sure it is async-safe and can handle partially initialized VM.
1769 void os::abort(bool dump_core) {
1770 os::shutdown();
1771 if (dump_core) {
1772 #ifndef PRODUCT
1773 fdStream out(defaultStream::output_fd());
1774 out.print_raw("Current thread is ");
1775 char buf[16];
1776 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1777 out.print_raw_cr(buf);
1778 out.print_raw_cr("Dumping core ...");
1779 #endif
1780 ::abort(); // dump core
1781 }
1782
1783 ::exit(1);
1784 }
1785
1786 // Die immediately, no exit hook, no abort hook, no cleanup.
1787 void os::die() {
1788 // _exit() on BsdThreads only kills current thread
1789 ::abort();
1790 }
1791
1792 // unused on bsd for now.
1793 void os::set_error_file(const char *logfile) {}
1794
1795
1796 // This method is a copy of JDK's sysGetLastErrorString
1797 // from src/solaris/hpi/src/system_md.c
1798
1799 size_t os::lasterror(char *buf, size_t len) {
1800
1801 if (errno == 0) return 0;
1802
1803 const char *s = ::strerror(errno);
1804 size_t n = ::strlen(s);
1805 if (n >= len) {
1806 n = len - 1;
1807 }
1808 ::strncpy(buf, s, n);
1809 buf[n] = '\0';
1810 return n;
1811 }
1812
1813 intx os::current_thread_id() {
1814 #ifdef __APPLE__
1815 return (intx)::mach_thread_self();
1816 #else
1817 return (intx)::pthread_self();
1818 #endif
1819 }
1820 int os::current_process_id() {
1821
1822 // Under the old bsd thread library, bsd gives each thread
1823 // its own process id. Because of this each thread will return
1824 // a different pid if this method were to return the result
1825 // of getpid(2). Bsd provides no api that returns the pid
1826 // of the launcher thread for the vm. This implementation
1827 // returns a unique pid, the pid of the launcher thread
1828 // that starts the vm 'process'.
1829
1830 // Under the NPTL, getpid() returns the same pid as the
1831 // launcher thread rather than a unique pid per thread.
1832 // Use gettid() if you want the old pre NPTL behaviour.
1833
1834 // if you are looking for the result of a call to getpid() that
1835 // returns a unique pid for the calling thread, then look at the
1836 // OSThread::thread_id() method in osThread_bsd.hpp file
1837
1838 return (int)(_initial_pid ? _initial_pid : getpid());
1839 }
1840
1841 // DLL functions
1842
1843 #define JNI_LIB_PREFIX "lib"
1844 #ifdef __APPLE__
1845 #define JNI_LIB_SUFFIX ".dylib"
1846 #else
1847 #define JNI_LIB_SUFFIX ".so"
1848 #endif
1849
1850 const char* os::dll_file_extension() { return JNI_LIB_SUFFIX; }
1851
1852 // This must be hard coded because it's the system's temporary
1853 // directory not the java application's temp directory, ala java.io.tmpdir.
1854 #ifdef __APPLE__
1855 // macosx has a secure per-user temporary directory
1856 char temp_path_storage[PATH_MAX];
1857 const char* os::get_temp_directory() {
1858 static char *temp_path = NULL;
1859 if (temp_path == NULL) {
1860 int pathSize = confstr(_CS_DARWIN_USER_TEMP_DIR, temp_path_storage, PATH_MAX);
1861 if (pathSize == 0 || pathSize > PATH_MAX) {
1862 strlcpy(temp_path_storage, "/tmp/", sizeof(temp_path_storage));
1863 }
1864 temp_path = temp_path_storage;
1865 }
1866 return temp_path;
1867 }
1868 #else /* __APPLE__ */
1869 const char* os::get_temp_directory() { return "/tmp"; }
1870 #endif /* __APPLE__ */
1871
1872 static bool file_exists(const char* filename) {
1873 struct stat statbuf;
1874 if (filename == NULL || strlen(filename) == 0) {
1875 return false;
1876 }
1877 return os::stat(filename, &statbuf) == 0;
1878 }
1879
1880 void os::dll_build_name(char* buffer, size_t buflen,
1881 const char* pname, const char* fname) {
1882 // Copied from libhpi
1883 const size_t pnamelen = pname ? strlen(pname) : 0;
1884
1885 // Quietly truncate on buffer overflow. Should be an error.
1886 if (pnamelen + strlen(fname) + strlen(JNI_LIB_PREFIX) + strlen(JNI_LIB_SUFFIX) + 2 > buflen) {
1887 *buffer = '\0';
1888 return;
1889 }
1890
1891 if (pnamelen == 0) {
1892 snprintf(buffer, buflen, JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, fname);
1893 } else if (strchr(pname, *os::path_separator()) != NULL) {
1894 int n;
1895 char** pelements = split_path(pname, &n);
1896 for (int i = 0 ; i < n ; i++) {
1897 // Really shouldn't be NULL, but check can't hurt
1898 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1899 continue; // skip the empty path values
1900 }
1901 snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX,
1902 pelements[i], fname);
1903 if (file_exists(buffer)) {
1904 break;
1905 }
1906 }
1907 // release the storage
1908 for (int i = 0 ; i < n ; i++) {
1909 if (pelements[i] != NULL) {
1910 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1911 }
1912 }
1913 if (pelements != NULL) {
1914 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1915 }
1916 } else {
1917 snprintf(buffer, buflen, "%s/" JNI_LIB_PREFIX "%s" JNI_LIB_SUFFIX, pname, fname);
1918 }
1919 }
1920
1921 const char* os::get_current_directory(char *buf, int buflen) {
1922 return getcwd(buf, buflen);
1923 }
1924
1925 // check if addr is inside libjvm[_g].so
1926 bool os::address_is_in_vm(address addr) {
1927 static address libjvm_base_addr;
1928 Dl_info dlinfo;
1929
1930 if (libjvm_base_addr == NULL) {
1931 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1932 libjvm_base_addr = (address)dlinfo.dli_fbase;
1933 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1934 }
1935
1936 if (dladdr((void *)addr, &dlinfo)) {
1937 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1938 }
1939
1940 return false;
1941 }
1942
1943
1944 #define MACH_MAXSYMLEN 256
1945
1946 bool os::dll_address_to_function_name(address addr, char *buf,
1947 int buflen, int *offset) {
1948 Dl_info dlinfo;
1949 char localbuf[MACH_MAXSYMLEN];
1950
1951 // dladdr will find names of dynamic functions only, but does
1952 // it set dli_fbase with mach_header address when it "fails" ?
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 // Handle non-dymanic manually:
1969 if (dlinfo.dli_fbase != NULL &&
1970 Decoder::decode(addr, localbuf, MACH_MAXSYMLEN, offset, dlinfo.dli_fbase)) {
1971 if(!Decoder::demangle(localbuf, buf, buflen)) {
1972 jio_snprintf(buf, buflen, "%s", localbuf);
1973 }
1974 return true;
1975 }
1976 if (buf != NULL) buf[0] = '\0';
1977 if (offset != NULL) *offset = -1;
1978 return false;
1979 }
1980
1981 #ifdef _ALLBSD_SOURCE
1982 // ported from solaris version
1983 bool os::dll_address_to_library_name(address addr, char* buf,
1984 int buflen, int* offset) {
1985 Dl_info dlinfo;
1986
1987 if (dladdr((void*)addr, &dlinfo)){
1988 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1989 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1990 return true;
1991 } else {
1992 if (buf) buf[0] = '\0';
1993 if (offset) *offset = -1;
1994 return false;
1995 }
1996 }
1997 #else
1998 struct _address_to_library_name {
1999 address addr; // input : memory address
2000 size_t buflen; // size of fname
2001 char* fname; // output: library name
2002 address base; // library base addr
2003 };
2004
2005 static int address_to_library_name_callback(struct dl_phdr_info *info,
2006 size_t size, void *data) {
2007 int i;
2008 bool found = false;
2009 address libbase = NULL;
2010 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
2011
2012 // iterate through all loadable segments
2013 for (i = 0; i < info->dlpi_phnum; i++) {
2014 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
2015 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
2016 // base address of a library is the lowest address of its loaded
2017 // segments.
2018 if (libbase == NULL || libbase > segbase) {
2019 libbase = segbase;
2020 }
2021 // see if 'addr' is within current segment
2022 if (segbase <= d->addr &&
2023 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
2024 found = true;
2025 }
2026 }
2027 }
2028
2029 // dlpi_name is NULL or empty if the ELF file is executable, return 0
2030 // so dll_address_to_library_name() can fall through to use dladdr() which
2031 // can figure out executable name from argv[0].
2032 if (found && info->dlpi_name && info->dlpi_name[0]) {
2033 d->base = libbase;
2034 if (d->fname) {
2035 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
2036 }
2037 return 1;
2038 }
2039 return 0;
2040 }
2041
2042 bool os::dll_address_to_library_name(address addr, char* buf,
2043 int buflen, int* offset) {
2044 Dl_info dlinfo;
2045 struct _address_to_library_name data;
2046
2047 // There is a bug in old glibc dladdr() implementation that it could resolve
2048 // to wrong library name if the .so file has a base address != NULL. Here
2049 // we iterate through the program headers of all loaded libraries to find
2050 // out which library 'addr' really belongs to. This workaround can be
2051 // removed once the minimum requirement for glibc is moved to 2.3.x.
2052 data.addr = addr;
2053 data.fname = buf;
2054 data.buflen = buflen;
2055 data.base = NULL;
2056 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
2057
2058 if (rslt) {
2059 // buf already contains library name
2060 if (offset) *offset = addr - data.base;
2061 return true;
2062 } else if (dladdr((void*)addr, &dlinfo)){
2063 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
2064 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
2065 return true;
2066 } else {
2067 if (buf) buf[0] = '\0';
2068 if (offset) *offset = -1;
2069 return false;
2070 }
2071 }
2072 #endif
2073
2074 // Loads .dll/.so and
2075 // in case of error it checks if .dll/.so was built for the
2076 // same architecture as Hotspot is running on
2077
2078 #ifdef __APPLE__
2079 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
2080 void * result= ::dlopen(filename, RTLD_LAZY);
2081 if (result != NULL) {
2082 // Successful loading
2083 return result;
2084 }
2085
2086 // Read system error message into ebuf
2087 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2088 ebuf[ebuflen-1]='\0';
2089
2090 return NULL;
2091 }
2092 #else
2093 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
2094 {
2095 void * result= ::dlopen(filename, RTLD_LAZY);
2096 if (result != NULL) {
2097 // Successful loading
2098 return result;
2099 }
2100
2101 Elf32_Ehdr elf_head;
2102
2103 // Read system error message into ebuf
2104 // It may or may not be overwritten below
2105 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2106 ebuf[ebuflen-1]='\0';
2107 int diag_msg_max_length=ebuflen-strlen(ebuf);
2108 char* diag_msg_buf=ebuf+strlen(ebuf);
2109
2110 if (diag_msg_max_length==0) {
2111 // No more space in ebuf for additional diagnostics message
2112 return NULL;
2113 }
2114
2115
2116 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2117
2118 if (file_descriptor < 0) {
2119 // Can't open library, report dlerror() message
2120 return NULL;
2121 }
2122
2123 bool failed_to_read_elf_head=
2124 (sizeof(elf_head)!=
2125 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2126
2127 ::close(file_descriptor);
2128 if (failed_to_read_elf_head) {
2129 // file i/o error - report dlerror() msg
2130 return NULL;
2131 }
2132
2133 typedef struct {
2134 Elf32_Half code; // Actual value as defined in elf.h
2135 Elf32_Half compat_class; // Compatibility of archs at VM's sense
2136 char elf_class; // 32 or 64 bit
2137 char endianess; // MSB or LSB
2138 char* name; // String representation
2139 } arch_t;
2140
2141 #ifndef EM_486
2142 #define EM_486 6 /* Intel 80486 */
2143 #endif
2144
2145 #ifndef EM_MIPS_RS3_LE
2146 #define EM_MIPS_RS3_LE 10 /* MIPS */
2147 #endif
2148
2149 #ifndef EM_PPC64
2150 #define EM_PPC64 21 /* PowerPC64 */
2151 #endif
2152
2153 #ifndef EM_S390
2154 #define EM_S390 22 /* IBM System/390 */
2155 #endif
2156
2157 #ifndef EM_IA_64
2158 #define EM_IA_64 50 /* HP/Intel IA-64 */
2159 #endif
2160
2161 #ifndef EM_X86_64
2162 #define EM_X86_64 62 /* AMD x86-64 */
2163 #endif
2164
2165 static const arch_t arch_array[]={
2166 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2167 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2168 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2169 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2170 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2171 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2172 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2173 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2174 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
2175 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
2176 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
2177 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
2178 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
2179 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
2180 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
2181 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
2182 };
2183
2184 #if (defined IA32)
2185 static Elf32_Half running_arch_code=EM_386;
2186 #elif (defined AMD64)
2187 static Elf32_Half running_arch_code=EM_X86_64;
2188 #elif (defined IA64)
2189 static Elf32_Half running_arch_code=EM_IA_64;
2190 #elif (defined __sparc) && (defined _LP64)
2191 static Elf32_Half running_arch_code=EM_SPARCV9;
2192 #elif (defined __sparc) && (!defined _LP64)
2193 static Elf32_Half running_arch_code=EM_SPARC;
2194 #elif (defined __powerpc64__)
2195 static Elf32_Half running_arch_code=EM_PPC64;
2196 #elif (defined __powerpc__)
2197 static Elf32_Half running_arch_code=EM_PPC;
2198 #elif (defined ARM)
2199 static Elf32_Half running_arch_code=EM_ARM;
2200 #elif (defined S390)
2201 static Elf32_Half running_arch_code=EM_S390;
2202 #elif (defined ALPHA)
2203 static Elf32_Half running_arch_code=EM_ALPHA;
2204 #elif (defined MIPSEL)
2205 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
2206 #elif (defined PARISC)
2207 static Elf32_Half running_arch_code=EM_PARISC;
2208 #elif (defined MIPS)
2209 static Elf32_Half running_arch_code=EM_MIPS;
2210 #elif (defined M68K)
2211 static Elf32_Half running_arch_code=EM_68K;
2212 #else
2213 #error Method os::dll_load requires that one of following is defined:\
2214 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2215 #endif
2216
2217 // Identify compatability class for VM's architecture and library's architecture
2218 // Obtain string descriptions for architectures
2219
2220 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2221 int running_arch_index=-1;
2222
2223 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2224 if (running_arch_code == arch_array[i].code) {
2225 running_arch_index = i;
2226 }
2227 if (lib_arch.code == arch_array[i].code) {
2228 lib_arch.compat_class = arch_array[i].compat_class;
2229 lib_arch.name = arch_array[i].name;
2230 }
2231 }
2232
2233 assert(running_arch_index != -1,
2234 "Didn't find running architecture code (running_arch_code) in arch_array");
2235 if (running_arch_index == -1) {
2236 // Even though running architecture detection failed
2237 // we may still continue with reporting dlerror() message
2238 return NULL;
2239 }
2240
2241 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2242 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2243 return NULL;
2244 }
2245
2246 #ifndef S390
2247 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2248 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2249 return NULL;
2250 }
2251 #endif // !S390
2252
2253 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2254 if ( lib_arch.name!=NULL ) {
2255 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2256 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2257 lib_arch.name, arch_array[running_arch_index].name);
2258 } else {
2259 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2260 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2261 lib_arch.code,
2262 arch_array[running_arch_index].name);
2263 }
2264 }
2265
2266 return NULL;
2267 }
2268 #endif /* !__APPLE__ */
2269
2270 // XXX: Do we need a lock around this as per Linux?
2271 void* os::dll_lookup(void* handle, const char* name) {
2272 return dlsym(handle, name);
2273 }
2274
2275
2276 static bool _print_ascii_file(const char* filename, outputStream* st) {
2277 int fd = ::open(filename, O_RDONLY);
2278 if (fd == -1) {
2279 return false;
2280 }
2281
2282 char buf[32];
2283 int bytes;
2284 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2285 st->print_raw(buf, bytes);
2286 }
2287
2288 ::close(fd);
2289
2290 return true;
2291 }
2292
2293 void os::print_dll_info(outputStream *st) {
2294 st->print_cr("Dynamic libraries:");
2295 #ifdef _ALLBSD_SOURCE
2296 #ifdef RTLD_DI_LINKMAP
2297 Dl_info dli;
2298 void *handle;
2299 Link_map *map;
2300 Link_map *p;
2301
2302 if (!dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli)) {
2303 st->print_cr("Error: Cannot print dynamic libraries.");
2304 return;
2305 }
2306 handle = dlopen(dli.dli_fname, RTLD_LAZY);
2307 if (handle == NULL) {
2308 st->print_cr("Error: Cannot print dynamic libraries.");
2309 return;
2310 }
2311 dlinfo(handle, RTLD_DI_LINKMAP, &map);
2312 if (map == NULL) {
2313 st->print_cr("Error: Cannot print dynamic libraries.");
2314 return;
2315 }
2316
2317 while (map->l_prev != NULL)
2318 map = map->l_prev;
2319
2320 while (map != NULL) {
2321 st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
2322 map = map->l_next;
2323 }
2324
2325 dlclose(handle);
2326 #elif defined(__APPLE__)
2327 uint32_t count;
2328 uint32_t i;
2329
2330 count = _dyld_image_count();
2331 for (i = 1; i < count; i++) {
2332 const char *name = _dyld_get_image_name(i);
2333 intptr_t slide = _dyld_get_image_vmaddr_slide(i);
2334 st->print_cr(PTR_FORMAT " \t%s", slide, name);
2335 }
2336 #else
2337 st->print_cr("Error: Cannot print dynamic libraries.");
2338 #endif
2339 #else
2340 char fname[32];
2341 pid_t pid = os::Bsd::gettid();
2342
2343 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2344
2345 if (!_print_ascii_file(fname, st)) {
2346 st->print("Can not get library information for pid = %d\n", pid);
2347 }
2348 #endif
2349 }
2350
2351 void os::print_os_info_brief(outputStream* st) {
2352 st->print("Bsd");
2353
2354 os::Posix::print_uname_info(st);
2355 }
2356
2357 void os::print_os_info(outputStream* st) {
2358 st->print("OS:");
2359 st->print("Bsd");
2360
2361 os::Posix::print_uname_info(st);
2362
2363 os::Posix::print_rlimit_info(st);
2364
2365 os::Posix::print_load_average(st);
2366 }
2367
2368 void os::pd_print_cpu_info(outputStream* st) {
2369 // Nothing to do for now.
2370 }
2371
2372 void os::print_memory_info(outputStream* st) {
2373
2374 st->print("Memory:");
2375 st->print(" %dk page", os::vm_page_size()>>10);
2376
2377 #ifndef _ALLBSD_SOURCE
2378 // values in struct sysinfo are "unsigned long"
2379 struct sysinfo si;
2380 sysinfo(&si);
2381 #endif
2382
2383 st->print(", physical " UINT64_FORMAT "k",
2384 os::physical_memory() >> 10);
2385 st->print("(" UINT64_FORMAT "k free)",
2386 os::available_memory() >> 10);
2387 #ifndef _ALLBSD_SOURCE
2388 st->print(", swap " UINT64_FORMAT "k",
2389 ((jlong)si.totalswap * si.mem_unit) >> 10);
2390 st->print("(" UINT64_FORMAT "k free)",
2391 ((jlong)si.freeswap * si.mem_unit) >> 10);
2392 #endif
2393 st->cr();
2394
2395 // meminfo
2396 st->print("\n/proc/meminfo:\n");
2397 _print_ascii_file("/proc/meminfo", st);
2398 st->cr();
2399 }
2400
2401 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2402 // but they're the same for all the bsd arch that we support
2403 // and they're the same for solaris but there's no common place to put this.
2404 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2405 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2406 "ILL_COPROC", "ILL_BADSTK" };
2407
2408 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2409 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2410 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2411
2412 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2413
2414 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2415
2416 void os::print_siginfo(outputStream* st, void* siginfo) {
2417 st->print("siginfo:");
2418
2419 const int buflen = 100;
2420 char buf[buflen];
2421 siginfo_t *si = (siginfo_t*)siginfo;
2422 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2423 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2424 st->print("si_errno=%s", buf);
2425 } else {
2426 st->print("si_errno=%d", si->si_errno);
2427 }
2428 const int c = si->si_code;
2429 assert(c > 0, "unexpected si_code");
2430 switch (si->si_signo) {
2431 case SIGILL:
2432 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2433 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2434 break;
2435 case SIGFPE:
2436 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2437 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2438 break;
2439 case SIGSEGV:
2440 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2441 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2442 break;
2443 case SIGBUS:
2444 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2445 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2446 break;
2447 default:
2448 st->print(", si_code=%d", si->si_code);
2449 // no si_addr
2450 }
2451
2452 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2453 UseSharedSpaces) {
2454 FileMapInfo* mapinfo = FileMapInfo::current_info();
2455 if (mapinfo->is_in_shared_space(si->si_addr)) {
2456 st->print("\n\nError accessing class data sharing archive." \
2457 " Mapped file inaccessible during execution, " \
2458 " possible disk/network problem.");
2459 }
2460 }
2461 st->cr();
2462 }
2463
2464
2465 static void print_signal_handler(outputStream* st, int sig,
2466 char* buf, size_t buflen);
2467
2468 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2469 st->print_cr("Signal Handlers:");
2470 print_signal_handler(st, SIGSEGV, buf, buflen);
2471 print_signal_handler(st, SIGBUS , buf, buflen);
2472 print_signal_handler(st, SIGFPE , buf, buflen);
2473 print_signal_handler(st, SIGPIPE, buf, buflen);
2474 print_signal_handler(st, SIGXFSZ, buf, buflen);
2475 print_signal_handler(st, SIGILL , buf, buflen);
2476 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2477 print_signal_handler(st, SR_signum, buf, buflen);
2478 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2479 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2480 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2481 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2482 }
2483
2484 static char saved_jvm_path[MAXPATHLEN] = {0};
2485
2486 // Find the full path to the current module, libjvm or libjvm_g
2487 void os::jvm_path(char *buf, jint buflen) {
2488 // Error checking.
2489 if (buflen < MAXPATHLEN) {
2490 assert(false, "must use a large-enough buffer");
2491 buf[0] = '\0';
2492 return;
2493 }
2494 // Lazy resolve the path to current module.
2495 if (saved_jvm_path[0] != 0) {
2496 strcpy(buf, saved_jvm_path);
2497 return;
2498 }
2499
2500 char dli_fname[MAXPATHLEN];
2501 bool ret = dll_address_to_library_name(
2502 CAST_FROM_FN_PTR(address, os::jvm_path),
2503 dli_fname, sizeof(dli_fname), NULL);
2504 assert(ret != 0, "cannot locate libjvm");
2505 char *rp = realpath(dli_fname, buf);
2506 if (rp == NULL)
2507 return;
2508
2509 if (Arguments::created_by_gamma_launcher()) {
2510 // Support for the gamma launcher. Typical value for buf is
2511 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm". If "/jre/lib/" appears at
2512 // the right place in the string, then assume we are installed in a JDK and
2513 // we're done. Otherwise, check for a JAVA_HOME environment variable and
2514 // construct a path to the JVM being overridden.
2515
2516 const char *p = buf + strlen(buf) - 1;
2517 for (int count = 0; p > buf && count < 5; ++count) {
2518 for (--p; p > buf && *p != '/'; --p)
2519 /* empty */ ;
2520 }
2521
2522 if (strncmp(p, "/jre/lib/", 9) != 0) {
2523 // Look for JAVA_HOME in the environment.
2524 char* java_home_var = ::getenv("JAVA_HOME");
2525 if (java_home_var != NULL && java_home_var[0] != 0) {
2526 char* jrelib_p;
2527 int len;
2528
2529 // Check the current module name "libjvm" or "libjvm_g".
2530 p = strrchr(buf, '/');
2531 assert(strstr(p, "/libjvm") == p, "invalid library name");
2532 p = strstr(p, "_g") ? "_g" : "";
2533
2534 rp = realpath(java_home_var, buf);
2535 if (rp == NULL)
2536 return;
2537
2538 // determine if this is a legacy image or modules image
2539 // modules image doesn't have "jre" subdirectory
2540 len = strlen(buf);
2541 jrelib_p = buf + len;
2542
2543 // Add the appropriate library subdir
2544 snprintf(jrelib_p, buflen-len, "/jre/lib");
2545 if (0 != access(buf, F_OK)) {
2546 snprintf(jrelib_p, buflen-len, "/lib");
2547 }
2548
2549 // Add the appropriate client or server subdir
2550 len = strlen(buf);
2551 jrelib_p = buf + len;
2552 snprintf(jrelib_p, buflen-len, "/%s", COMPILER_VARIANT);
2553 if (0 != access(buf, F_OK)) {
2554 snprintf(jrelib_p, buflen-len, "");
2555 }
2556
2557 // If the path exists within JAVA_HOME, add the JVM library name
2558 // to complete the path to JVM being overridden. Otherwise fallback
2559 // to the path to the current library.
2560 if (0 == access(buf, F_OK)) {
2561 // Use current module name "libjvm[_g]" instead of
2562 // "libjvm"debug_only("_g")"" since for fastdebug version
2563 // we should have "libjvm" but debug_only("_g") adds "_g"!
2564 len = strlen(buf);
2565 snprintf(buf + len, buflen-len, "/libjvm%s%s", p, JNI_LIB_SUFFIX);
2566 } else {
2567 // Fall back to path of current library
2568 rp = realpath(dli_fname, buf);
2569 if (rp == NULL)
2570 return;
2571 }
2572 }
2573 }
2574 }
2575
2576 strcpy(saved_jvm_path, buf);
2577 }
2578
2579 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2580 // no prefix required, not even "_"
2581 }
2582
2583 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2584 // no suffix required
2585 }
2586
2587 ////////////////////////////////////////////////////////////////////////////////
2588 // sun.misc.Signal support
2589
2590 static volatile jint sigint_count = 0;
2591
2592 static void
2593 UserHandler(int sig, void *siginfo, void *context) {
2594 // 4511530 - sem_post is serialized and handled by the manager thread. When
2595 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2596 // don't want to flood the manager thread with sem_post requests.
2597 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2598 return;
2599
2600 // Ctrl-C is pressed during error reporting, likely because the error
2601 // handler fails to abort. Let VM die immediately.
2602 if (sig == SIGINT && is_error_reported()) {
2603 os::die();
2604 }
2605
2606 os::signal_notify(sig);
2607 }
2608
2609 void* os::user_handler() {
2610 return CAST_FROM_FN_PTR(void*, UserHandler);
2611 }
2612
2613 extern "C" {
2614 typedef void (*sa_handler_t)(int);
2615 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2616 }
2617
2618 void* os::signal(int signal_number, void* handler) {
2619 struct sigaction sigAct, oldSigAct;
2620
2621 sigfillset(&(sigAct.sa_mask));
2622 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2623 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2624
2625 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2626 // -1 means registration failed
2627 return (void *)-1;
2628 }
2629
2630 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2631 }
2632
2633 void os::signal_raise(int signal_number) {
2634 ::raise(signal_number);
2635 }
2636
2637 /*
2638 * The following code is moved from os.cpp for making this
2639 * code platform specific, which it is by its very nature.
2640 */
2641
2642 // Will be modified when max signal is changed to be dynamic
2643 int os::sigexitnum_pd() {
2644 return NSIG;
2645 }
2646
2647 // a counter for each possible signal value
2648 static volatile jint pending_signals[NSIG+1] = { 0 };
2649
2650 // Bsd(POSIX) specific hand shaking semaphore.
2651 #ifdef __APPLE__
2652 static semaphore_t sig_sem;
2653 #define SEM_INIT(sem, value) semaphore_create(mach_task_self(), &sem, SYNC_POLICY_FIFO, value)
2654 #define SEM_WAIT(sem) semaphore_wait(sem);
2655 #define SEM_POST(sem) semaphore_signal(sem);
2656 #else
2657 static sem_t sig_sem;
2658 #define SEM_INIT(sem, value) sem_init(&sem, 0, value)
2659 #define SEM_WAIT(sem) sem_wait(&sem);
2660 #define SEM_POST(sem) sem_post(&sem);
2661 #endif
2662
2663 void os::signal_init_pd() {
2664 // Initialize signal structures
2665 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2666
2667 // Initialize signal semaphore
2668 ::SEM_INIT(sig_sem, 0);
2669 }
2670
2671 void os::signal_notify(int sig) {
2672 Atomic::inc(&pending_signals[sig]);
2673 ::SEM_POST(sig_sem);
2674 }
2675
2676 static int check_pending_signals(bool wait) {
2677 Atomic::store(0, &sigint_count);
2678 for (;;) {
2679 for (int i = 0; i < NSIG + 1; i++) {
2680 jint n = pending_signals[i];
2681 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2682 return i;
2683 }
2684 }
2685 if (!wait) {
2686 return -1;
2687 }
2688 JavaThread *thread = JavaThread::current();
2689 ThreadBlockInVM tbivm(thread);
2690
2691 bool threadIsSuspended;
2692 do {
2693 thread->set_suspend_equivalent();
2694 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2695 ::SEM_WAIT(sig_sem);
2696
2697 // were we externally suspended while we were waiting?
2698 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2699 if (threadIsSuspended) {
2700 //
2701 // The semaphore has been incremented, but while we were waiting
2702 // another thread suspended us. We don't want to continue running
2703 // while suspended because that would surprise the thread that
2704 // suspended us.
2705 //
2706 ::SEM_POST(sig_sem);
2707
2708 thread->java_suspend_self();
2709 }
2710 } while (threadIsSuspended);
2711 }
2712 }
2713
2714 int os::signal_lookup() {
2715 return check_pending_signals(false);
2716 }
2717
2718 int os::signal_wait() {
2719 return check_pending_signals(true);
2720 }
2721
2722 ////////////////////////////////////////////////////////////////////////////////
2723 // Virtual Memory
2724
2725 int os::vm_page_size() {
2726 // Seems redundant as all get out
2727 assert(os::Bsd::page_size() != -1, "must call os::init");
2728 return os::Bsd::page_size();
2729 }
2730
2731 // Solaris allocates memory by pages.
2732 int os::vm_allocation_granularity() {
2733 assert(os::Bsd::page_size() != -1, "must call os::init");
2734 return os::Bsd::page_size();
2735 }
2736
2737 // Rationale behind this function:
2738 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2739 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2740 // samples for JITted code. Here we create private executable mapping over the code cache
2741 // and then we can use standard (well, almost, as mapping can change) way to provide
2742 // info for the reporting script by storing timestamp and location of symbol
2743 void bsd_wrap_code(char* base, size_t size) {
2744 static volatile jint cnt = 0;
2745
2746 if (!UseOprofile) {
2747 return;
2748 }
2749
2750 char buf[PATH_MAX + 1];
2751 int num = Atomic::add(1, &cnt);
2752
2753 snprintf(buf, PATH_MAX + 1, "%s/hs-vm-%d-%d",
2754 os::get_temp_directory(), os::current_process_id(), num);
2755 unlink(buf);
2756
2757 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2758
2759 if (fd != -1) {
2760 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2761 if (rv != (off_t)-1) {
2762 if (::write(fd, "", 1) == 1) {
2763 mmap(base, size,
2764 PROT_READ|PROT_WRITE|PROT_EXEC,
2765 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2766 }
2767 }
2768 ::close(fd);
2769 unlink(buf);
2770 }
2771 }
2772
2773 // NOTE: Bsd kernel does not really reserve the pages for us.
2774 // All it does is to check if there are enough free pages
2775 // left at the time of mmap(). This could be a potential
2776 // problem.
2777 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2778 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2779 #ifdef __OpenBSD__
2780 // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
2781 return ::mprotect(addr, size, prot) == 0;
2782 #else
2783 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2784 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2785 return res != (uintptr_t) MAP_FAILED;
2786 #endif
2787 }
2788
2789 #ifndef _ALLBSD_SOURCE
2790 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2791 #ifndef MAP_HUGETLB
2792 #define MAP_HUGETLB 0x40000
2793 #endif
2794
2795 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2796 #ifndef MADV_HUGEPAGE
2797 #define MADV_HUGEPAGE 14
2798 #endif
2799 #endif
2800
2801 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2802 bool exec) {
2803 #ifndef _ALLBSD_SOURCE
2804 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2805 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2806 uintptr_t res =
2807 (uintptr_t) ::mmap(addr, size, prot,
2808 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2809 -1, 0);
2810 return res != (uintptr_t) MAP_FAILED;
2811 }
2812 #endif
2813
2814 return commit_memory(addr, size, exec);
2815 }
2816
2817 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2818 #ifndef _ALLBSD_SOURCE
2819 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2820 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2821 // be supported or the memory may already be backed by huge pages.
2822 ::madvise(addr, bytes, MADV_HUGEPAGE);
2823 }
2824 #endif
2825 }
2826
2827 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2828 ::madvise(addr, bytes, MADV_DONTNEED);
2829 }
2830
2831 void os::numa_make_global(char *addr, size_t bytes) {
2832 }
2833
2834 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2835 }
2836
2837 bool os::numa_topology_changed() { return false; }
2838
2839 size_t os::numa_get_groups_num() {
2840 return 1;
2841 }
2842
2843 int os::numa_get_group_id() {
2844 return 0;
2845 }
2846
2847 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2848 if (size > 0) {
2849 ids[0] = 0;
2850 return 1;
2851 }
2852 return 0;
2853 }
2854
2855 bool os::get_page_info(char *start, page_info* info) {
2856 return false;
2857 }
2858
2859 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2860 return end;
2861 }
2862
2863 #ifndef _ALLBSD_SOURCE
2864 // Something to do with the numa-aware allocator needs these symbols
2865 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2866 extern "C" JNIEXPORT void numa_error(char *where) { }
2867 extern "C" JNIEXPORT int fork1() { return fork(); }
2868
2869
2870 // If we are running with libnuma version > 2, then we should
2871 // be trying to use symbols with versions 1.1
2872 // If we are running with earlier version, which did not have symbol versions,
2873 // we should use the base version.
2874 void* os::Bsd::libnuma_dlsym(void* handle, const char *name) {
2875 void *f = dlvsym(handle, name, "libnuma_1.1");
2876 if (f == NULL) {
2877 f = dlsym(handle, name);
2878 }
2879 return f;
2880 }
2881
2882 bool os::Bsd::libnuma_init() {
2883 // sched_getcpu() should be in libc.
2884 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2885 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2886
2887 if (sched_getcpu() != -1) { // Does it work?
2888 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2889 if (handle != NULL) {
2890 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2891 libnuma_dlsym(handle, "numa_node_to_cpus")));
2892 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2893 libnuma_dlsym(handle, "numa_max_node")));
2894 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2895 libnuma_dlsym(handle, "numa_available")));
2896 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2897 libnuma_dlsym(handle, "numa_tonode_memory")));
2898 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2899 libnuma_dlsym(handle, "numa_interleave_memory")));
2900
2901
2902 if (numa_available() != -1) {
2903 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2904 // Create a cpu -> node mapping
2905 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2906 rebuild_cpu_to_node_map();
2907 return true;
2908 }
2909 }
2910 }
2911 return false;
2912 }
2913
2914 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2915 // The table is later used in get_node_by_cpu().
2916 void os::Bsd::rebuild_cpu_to_node_map() {
2917 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2918 // in libnuma (possible values are starting from 16,
2919 // and continuing up with every other power of 2, but less
2920 // than the maximum number of CPUs supported by kernel), and
2921 // is a subject to change (in libnuma version 2 the requirements
2922 // are more reasonable) we'll just hardcode the number they use
2923 // in the library.
2924 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2925
2926 size_t cpu_num = os::active_processor_count();
2927 size_t cpu_map_size = NCPUS / BitsPerCLong;
2928 size_t cpu_map_valid_size =
2929 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2930
2931 cpu_to_node()->clear();
2932 cpu_to_node()->at_grow(cpu_num - 1);
2933 size_t node_num = numa_get_groups_num();
2934
2935 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2936 for (size_t i = 0; i < node_num; i++) {
2937 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2938 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2939 if (cpu_map[j] != 0) {
2940 for (size_t k = 0; k < BitsPerCLong; k++) {
2941 if (cpu_map[j] & (1UL << k)) {
2942 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2943 }
2944 }
2945 }
2946 }
2947 }
2948 }
2949 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2950 }
2951
2952 int os::Bsd::get_node_by_cpu(int cpu_id) {
2953 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2954 return cpu_to_node()->at(cpu_id);
2955 }
2956 return -1;
2957 }
2958
2959 GrowableArray<int>* os::Bsd::_cpu_to_node;
2960 os::Bsd::sched_getcpu_func_t os::Bsd::_sched_getcpu;
2961 os::Bsd::numa_node_to_cpus_func_t os::Bsd::_numa_node_to_cpus;
2962 os::Bsd::numa_max_node_func_t os::Bsd::_numa_max_node;
2963 os::Bsd::numa_available_func_t os::Bsd::_numa_available;
2964 os::Bsd::numa_tonode_memory_func_t os::Bsd::_numa_tonode_memory;
2965 os::Bsd::numa_interleave_memory_func_t os::Bsd::_numa_interleave_memory;
2966 unsigned long* os::Bsd::_numa_all_nodes;
2967 #endif
2968
2969 bool os::pd_uncommit_memory(char* addr, size_t size) {
2970 #ifdef __OpenBSD__
2971 // XXX: Work-around mmap/MAP_FIXED bug temporarily on OpenBSD
2972 return ::mprotect(addr, size, PROT_NONE) == 0;
2973 #else
2974 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2975 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2976 return res != (uintptr_t) MAP_FAILED;
2977 #endif
2978 }
2979
2980 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2981 return os::commit_memory(addr, size);
2982 }
2983
2984 // If this is a growable mapping, remove the guard pages entirely by
2985 // munmap()ping them. If not, just call uncommit_memory().
2986 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2987 return os::uncommit_memory(addr, size);
2988 }
2989
2990 static address _highest_vm_reserved_address = NULL;
2991
2992 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2993 // at 'requested_addr'. If there are existing memory mappings at the same
2994 // location, however, they will be overwritten. If 'fixed' is false,
2995 // 'requested_addr' is only treated as a hint, the return value may or
2996 // may not start from the requested address. Unlike Bsd mmap(), this
2997 // function returns NULL to indicate failure.
2998 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2999 char * addr;
3000 int flags;
3001
3002 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3003 if (fixed) {
3004 assert((uintptr_t)requested_addr % os::Bsd::page_size() == 0, "unaligned address");
3005 flags |= MAP_FIXED;
3006 }
3007
3008 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
3009 // to PROT_EXEC if executable when we commit the page.
3010 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
3011 flags, -1, 0);
3012
3013 if (addr != MAP_FAILED) {
3014 // anon_mmap() should only get called during VM initialization,
3015 // don't need lock (actually we can skip locking even it can be called
3016 // from multiple threads, because _highest_vm_reserved_address is just a
3017 // hint about the upper limit of non-stack memory regions.)
3018 if ((address)addr + bytes > _highest_vm_reserved_address) {
3019 _highest_vm_reserved_address = (address)addr + bytes;
3020 }
3021 }
3022
3023 return addr == MAP_FAILED ? NULL : addr;
3024 }
3025
3026 // Don't update _highest_vm_reserved_address, because there might be memory
3027 // regions above addr + size. If so, releasing a memory region only creates
3028 // a hole in the address space, it doesn't help prevent heap-stack collision.
3029 //
3030 static int anon_munmap(char * addr, size_t size) {
3031 return ::munmap(addr, size) == 0;
3032 }
3033
3034 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3035 size_t alignment_hint) {
3036 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3037 }
3038
3039 bool os::pd_release_memory(char* addr, size_t size) {
3040 return anon_munmap(addr, size);
3041 }
3042
3043 static address highest_vm_reserved_address() {
3044 return _highest_vm_reserved_address;
3045 }
3046
3047 static bool bsd_mprotect(char* addr, size_t size, int prot) {
3048 // Bsd wants the mprotect address argument to be page aligned.
3049 char* bottom = (char*)align_size_down((intptr_t)addr, os::Bsd::page_size());
3050
3051 // According to SUSv3, mprotect() should only be used with mappings
3052 // established by mmap(), and mmap() always maps whole pages. Unaligned
3053 // 'addr' likely indicates problem in the VM (e.g. trying to change
3054 // protection of malloc'ed or statically allocated memory). Check the
3055 // caller if you hit this assert.
3056 assert(addr == bottom, "sanity check");
3057
3058 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Bsd::page_size());
3059 return ::mprotect(bottom, size, prot) == 0;
3060 }
3061
3062 // Set protections specified
3063 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3064 bool is_committed) {
3065 unsigned int p = 0;
3066 switch (prot) {
3067 case MEM_PROT_NONE: p = PROT_NONE; break;
3068 case MEM_PROT_READ: p = PROT_READ; break;
3069 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3070 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3071 default:
3072 ShouldNotReachHere();
3073 }
3074 // is_committed is unused.
3075 return bsd_mprotect(addr, bytes, p);
3076 }
3077
3078 bool os::guard_memory(char* addr, size_t size) {
3079 return bsd_mprotect(addr, size, PROT_NONE);
3080 }
3081
3082 bool os::unguard_memory(char* addr, size_t size) {
3083 return bsd_mprotect(addr, size, PROT_READ|PROT_WRITE);
3084 }
3085
3086 bool os::Bsd::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3087 bool result = false;
3088 #ifndef _ALLBSD_SOURCE
3089 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
3090 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3091 -1, 0);
3092
3093 if (p != (void *) -1) {
3094 // We don't know if this really is a huge page or not.
3095 FILE *fp = fopen("/proc/self/maps", "r");
3096 if (fp) {
3097 while (!feof(fp)) {
3098 char chars[257];
3099 long x = 0;
3100 if (fgets(chars, sizeof(chars), fp)) {
3101 if (sscanf(chars, "%lx-%*x", &x) == 1
3102 && x == (long)p) {
3103 if (strstr (chars, "hugepage")) {
3104 result = true;
3105 break;
3106 }
3107 }
3108 }
3109 }
3110 fclose(fp);
3111 }
3112 munmap (p, page_size);
3113 if (result)
3114 return true;
3115 }
3116
3117 if (warn) {
3118 warning("HugeTLBFS is not supported by the operating system.");
3119 }
3120 #endif
3121
3122 return result;
3123 }
3124
3125 /*
3126 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3127 *
3128 * From the coredump_filter documentation:
3129 *
3130 * - (bit 0) anonymous private memory
3131 * - (bit 1) anonymous shared memory
3132 * - (bit 2) file-backed private memory
3133 * - (bit 3) file-backed shared memory
3134 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3135 * effective only if the bit 2 is cleared)
3136 * - (bit 5) hugetlb private memory
3137 * - (bit 6) hugetlb shared memory
3138 */
3139 static void set_coredump_filter(void) {
3140 FILE *f;
3141 long cdm;
3142
3143 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3144 return;
3145 }
3146
3147 if (fscanf(f, "%lx", &cdm) != 1) {
3148 fclose(f);
3149 return;
3150 }
3151
3152 rewind(f);
3153
3154 if ((cdm & LARGEPAGES_BIT) == 0) {
3155 cdm |= LARGEPAGES_BIT;
3156 fprintf(f, "%#lx", cdm);
3157 }
3158
3159 fclose(f);
3160 }
3161
3162 // Large page support
3163
3164 static size_t _large_page_size = 0;
3165
3166 void os::large_page_init() {
3167 #ifndef _ALLBSD_SOURCE
3168 if (!UseLargePages) {
3169 UseHugeTLBFS = false;
3170 UseSHM = false;
3171 return;
3172 }
3173
3174 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3175 // If UseLargePages is specified on the command line try both methods,
3176 // if it's default, then try only HugeTLBFS.
3177 if (FLAG_IS_DEFAULT(UseLargePages)) {
3178 UseHugeTLBFS = true;
3179 } else {
3180 UseHugeTLBFS = UseSHM = true;
3181 }
3182 }
3183
3184 if (LargePageSizeInBytes) {
3185 _large_page_size = LargePageSizeInBytes;
3186 } else {
3187 // large_page_size on Bsd is used to round up heap size. x86 uses either
3188 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3189 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3190 // page as large as 256M.
3191 //
3192 // Here we try to figure out page size by parsing /proc/meminfo and looking
3193 // for a line with the following format:
3194 // Hugepagesize: 2048 kB
3195 //
3196 // If we can't determine the value (e.g. /proc is not mounted, or the text
3197 // format has been changed), we'll use the largest page size supported by
3198 // the processor.
3199
3200 #ifndef ZERO
3201 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3202 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3203 #endif // ZERO
3204
3205 FILE *fp = fopen("/proc/meminfo", "r");
3206 if (fp) {
3207 while (!feof(fp)) {
3208 int x = 0;
3209 char buf[16];
3210 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3211 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3212 _large_page_size = x * K;
3213 break;
3214 }
3215 } else {
3216 // skip to next line
3217 for (;;) {
3218 int ch = fgetc(fp);
3219 if (ch == EOF || ch == (int)'\n') break;
3220 }
3221 }
3222 }
3223 fclose(fp);
3224 }
3225 }
3226
3227 // print a warning if any large page related flag is specified on command line
3228 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3229
3230 const size_t default_page_size = (size_t)Bsd::page_size();
3231 if (_large_page_size > default_page_size) {
3232 _page_sizes[0] = _large_page_size;
3233 _page_sizes[1] = default_page_size;
3234 _page_sizes[2] = 0;
3235 }
3236 UseHugeTLBFS = UseHugeTLBFS &&
3237 Bsd::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3238
3239 if (UseHugeTLBFS)
3240 UseSHM = false;
3241
3242 UseLargePages = UseHugeTLBFS || UseSHM;
3243
3244 set_coredump_filter();
3245 #endif
3246 }
3247
3248 #ifndef _ALLBSD_SOURCE
3249 #ifndef SHM_HUGETLB
3250 #define SHM_HUGETLB 04000
3251 #endif
3252 #endif
3253
3254 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3255 // "exec" is passed in but not used. Creating the shared image for
3256 // the code cache doesn't have an SHM_X executable permission to check.
3257 assert(UseLargePages && UseSHM, "only for SHM large pages");
3258
3259 key_t key = IPC_PRIVATE;
3260 char *addr;
3261
3262 bool warn_on_failure = UseLargePages &&
3263 (!FLAG_IS_DEFAULT(UseLargePages) ||
3264 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3265 );
3266 char msg[128];
3267
3268 // Create a large shared memory region to attach to based on size.
3269 // Currently, size is the total size of the heap
3270 #ifndef _ALLBSD_SOURCE
3271 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3272 #else
3273 int shmid = shmget(key, bytes, IPC_CREAT|SHM_R|SHM_W);
3274 #endif
3275 if (shmid == -1) {
3276 // Possible reasons for shmget failure:
3277 // 1. shmmax is too small for Java heap.
3278 // > check shmmax value: cat /proc/sys/kernel/shmmax
3279 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3280 // 2. not enough large page memory.
3281 // > check available large pages: cat /proc/meminfo
3282 // > increase amount of large pages:
3283 // echo new_value > /proc/sys/vm/nr_hugepages
3284 // Note 1: different Bsd may use different name for this property,
3285 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3286 // Note 2: it's possible there's enough physical memory available but
3287 // they are so fragmented after a long run that they can't
3288 // coalesce into large pages. Try to reserve large pages when
3289 // the system is still "fresh".
3290 if (warn_on_failure) {
3291 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3292 warning(msg);
3293 }
3294 return NULL;
3295 }
3296
3297 // attach to the region
3298 addr = (char*)shmat(shmid, req_addr, 0);
3299 int err = errno;
3300
3301 // Remove shmid. If shmat() is successful, the actual shared memory segment
3302 // will be deleted when it's detached by shmdt() or when the process
3303 // terminates. If shmat() is not successful this will remove the shared
3304 // segment immediately.
3305 shmctl(shmid, IPC_RMID, NULL);
3306
3307 if ((intptr_t)addr == -1) {
3308 if (warn_on_failure) {
3309 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3310 warning(msg);
3311 }
3312 return NULL;
3313 }
3314
3315 return addr;
3316 }
3317
3318 bool os::release_memory_special(char* base, size_t bytes) {
3319 // detaching the SHM segment will also delete it, see reserve_memory_special()
3320 int rslt = shmdt(base);
3321 return rslt == 0;
3322 }
3323
3324 size_t os::large_page_size() {
3325 return _large_page_size;
3326 }
3327
3328 // HugeTLBFS allows application to commit large page memory on demand;
3329 // with SysV SHM the entire memory region must be allocated as shared
3330 // memory.
3331 bool os::can_commit_large_page_memory() {
3332 return UseHugeTLBFS;
3333 }
3334
3335 bool os::can_execute_large_page_memory() {
3336 return UseHugeTLBFS;
3337 }
3338
3339 // Reserve memory at an arbitrary address, only if that area is
3340 // available (and not reserved for something else).
3341
3342 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3343 const int max_tries = 10;
3344 char* base[max_tries];
3345 size_t size[max_tries];
3346 const size_t gap = 0x000000;
3347
3348 // Assert only that the size is a multiple of the page size, since
3349 // that's all that mmap requires, and since that's all we really know
3350 // about at this low abstraction level. If we need higher alignment,
3351 // we can either pass an alignment to this method or verify alignment
3352 // in one of the methods further up the call chain. See bug 5044738.
3353 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3354
3355 // Repeatedly allocate blocks until the block is allocated at the
3356 // right spot. Give up after max_tries. Note that reserve_memory() will
3357 // automatically update _highest_vm_reserved_address if the call is
3358 // successful. The variable tracks the highest memory address every reserved
3359 // by JVM. It is used to detect heap-stack collision if running with
3360 // fixed-stack BsdThreads. Because here we may attempt to reserve more
3361 // space than needed, it could confuse the collision detecting code. To
3362 // solve the problem, save current _highest_vm_reserved_address and
3363 // calculate the correct value before return.
3364 address old_highest = _highest_vm_reserved_address;
3365
3366 // Bsd mmap allows caller to pass an address as hint; give it a try first,
3367 // if kernel honors the hint then we can return immediately.
3368 char * addr = anon_mmap(requested_addr, bytes, false);
3369 if (addr == requested_addr) {
3370 return requested_addr;
3371 }
3372
3373 if (addr != NULL) {
3374 // mmap() is successful but it fails to reserve at the requested address
3375 anon_munmap(addr, bytes);
3376 }
3377
3378 int i;
3379 for (i = 0; i < max_tries; ++i) {
3380 base[i] = reserve_memory(bytes);
3381
3382 if (base[i] != NULL) {
3383 // Is this the block we wanted?
3384 if (base[i] == requested_addr) {
3385 size[i] = bytes;
3386 break;
3387 }
3388
3389 // Does this overlap the block we wanted? Give back the overlapped
3390 // parts and try again.
3391
3392 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3393 if (top_overlap >= 0 && top_overlap < bytes) {
3394 unmap_memory(base[i], top_overlap);
3395 base[i] += top_overlap;
3396 size[i] = bytes - top_overlap;
3397 } else {
3398 size_t bottom_overlap = base[i] + bytes - requested_addr;
3399 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3400 unmap_memory(requested_addr, bottom_overlap);
3401 size[i] = bytes - bottom_overlap;
3402 } else {
3403 size[i] = bytes;
3404 }
3405 }
3406 }
3407 }
3408
3409 // Give back the unused reserved pieces.
3410
3411 for (int j = 0; j < i; ++j) {
3412 if (base[j] != NULL) {
3413 unmap_memory(base[j], size[j]);
3414 }
3415 }
3416
3417 if (i < max_tries) {
3418 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3419 return requested_addr;
3420 } else {
3421 _highest_vm_reserved_address = old_highest;
3422 return NULL;
3423 }
3424 }
3425
3426 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3427 RESTARTABLE_RETURN_INT(::read(fd, buf, nBytes));
3428 }
3429
3430 // TODO-FIXME: reconcile Solaris' os::sleep with the bsd variation.
3431 // Solaris uses poll(), bsd uses park().
3432 // Poll() is likely a better choice, assuming that Thread.interrupt()
3433 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3434 // SIGSEGV, see 4355769.
3435
3436 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3437 assert(thread == Thread::current(), "thread consistency check");
3438
3439 ParkEvent * const slp = thread->_SleepEvent ;
3440 slp->reset() ;
3441 OrderAccess::fence() ;
3442
3443 if (interruptible) {
3444 jlong prevtime = javaTimeNanos();
3445
3446 for (;;) {
3447 if (os::is_interrupted(thread, true)) {
3448 return OS_INTRPT;
3449 }
3450
3451 jlong newtime = javaTimeNanos();
3452
3453 if (newtime - prevtime < 0) {
3454 // time moving backwards, should only happen if no monotonic clock
3455 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3456 assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
3457 } else {
3458 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3459 }
3460
3461 if(millis <= 0) {
3462 return OS_OK;
3463 }
3464
3465 prevtime = newtime;
3466
3467 {
3468 assert(thread->is_Java_thread(), "sanity check");
3469 JavaThread *jt = (JavaThread *) thread;
3470 ThreadBlockInVM tbivm(jt);
3471 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3472
3473 jt->set_suspend_equivalent();
3474 // cleared by handle_special_suspend_equivalent_condition() or
3475 // java_suspend_self() via check_and_wait_while_suspended()
3476
3477 slp->park(millis);
3478
3479 // were we externally suspended while we were waiting?
3480 jt->check_and_wait_while_suspended();
3481 }
3482 }
3483 } else {
3484 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3485 jlong prevtime = javaTimeNanos();
3486
3487 for (;;) {
3488 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3489 // the 1st iteration ...
3490 jlong newtime = javaTimeNanos();
3491
3492 if (newtime - prevtime < 0) {
3493 // time moving backwards, should only happen if no monotonic clock
3494 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3495 assert(!Bsd::supports_monotonic_clock(), "time moving backwards");
3496 } else {
3497 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3498 }
3499
3500 if(millis <= 0) break ;
3501
3502 prevtime = newtime;
3503 slp->park(millis);
3504 }
3505 return OS_OK ;
3506 }
3507 }
3508
3509 int os::naked_sleep() {
3510 // %% make the sleep time an integer flag. for now use 1 millisec.
3511 return os::sleep(Thread::current(), 1, false);
3512 }
3513
3514 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3515 void os::infinite_sleep() {
3516 while (true) { // sleep forever ...
3517 ::sleep(100); // ... 100 seconds at a time
3518 }
3519 }
3520
3521 // Used to convert frequent JVM_Yield() to nops
3522 bool os::dont_yield() {
3523 return DontYieldALot;
3524 }
3525
3526 void os::yield() {
3527 sched_yield();
3528 }
3529
3530 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3531
3532 void os::yield_all(int attempts) {
3533 // Yields to all threads, including threads with lower priorities
3534 // Threads on Bsd are all with same priority. The Solaris style
3535 // os::yield_all() with nanosleep(1ms) is not necessary.
3536 sched_yield();
3537 }
3538
3539 // Called from the tight loops to possibly influence time-sharing heuristics
3540 void os::loop_breaker(int attempts) {
3541 os::yield_all(attempts);
3542 }
3543
3544 ////////////////////////////////////////////////////////////////////////////////
3545 // thread priority support
3546
3547 // Note: Normal Bsd applications are run with SCHED_OTHER policy. SCHED_OTHER
3548 // only supports dynamic priority, static priority must be zero. For real-time
3549 // applications, Bsd supports SCHED_RR which allows static priority (1-99).
3550 // However, for large multi-threaded applications, SCHED_RR is not only slower
3551 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3552 // of 5 runs - Sep 2005).
3553 //
3554 // The following code actually changes the niceness of kernel-thread/LWP. It
3555 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3556 // not the entire user process, and user level threads are 1:1 mapped to kernel
3557 // threads. It has always been the case, but could change in the future. For
3558 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3559 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3560
3561 #if defined(_ALLBSD_SOURCE) && !defined(__APPLE__)
3562 int os::java_to_os_priority[CriticalPriority + 1] = {
3563 19, // 0 Entry should never be used
3564
3565 0, // 1 MinPriority
3566 3, // 2
3567 6, // 3
3568
3569 10, // 4
3570 15, // 5 NormPriority
3571 18, // 6
3572
3573 21, // 7
3574 25, // 8
3575 28, // 9 NearMaxPriority
3576
3577 31, // 10 MaxPriority
3578
3579 31 // 11 CriticalPriority
3580 };
3581 #elif defined(__APPLE__)
3582 /* Using Mach high-level priority assignments */
3583 int os::java_to_os_priority[CriticalPriority + 1] = {
3584 0, // 0 Entry should never be used (MINPRI_USER)
3585
3586 27, // 1 MinPriority
3587 28, // 2
3588 29, // 3
3589
3590 30, // 4
3591 31, // 5 NormPriority (BASEPRI_DEFAULT)
3592 32, // 6
3593
3594 33, // 7
3595 34, // 8
3596 35, // 9 NearMaxPriority
3597
3598 36, // 10 MaxPriority
3599
3600 36 // 11 CriticalPriority
3601 };
3602 #else
3603 int os::java_to_os_priority[CriticalPriority + 1] = {
3604 19, // 0 Entry should never be used
3605
3606 4, // 1 MinPriority
3607 3, // 2
3608 2, // 3
3609
3610 1, // 4
3611 0, // 5 NormPriority
3612 -1, // 6
3613
3614 -2, // 7
3615 -3, // 8
3616 -4, // 9 NearMaxPriority
3617
3618 -5, // 10 MaxPriority
3619
3620 -5 // 11 CriticalPriority
3621 };
3622 #endif
3623
3624 static int prio_init() {
3625 if (ThreadPriorityPolicy == 1) {
3626 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3627 // if effective uid is not root. Perhaps, a more elegant way of doing
3628 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3629 if (geteuid() != 0) {
3630 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3631 warning("-XX:ThreadPriorityPolicy requires root privilege on Bsd");
3632 }
3633 ThreadPriorityPolicy = 0;
3634 }
3635 }
3636 if (UseCriticalJavaThreadPriority) {
3637 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3638 }
3639 return 0;
3640 }
3641
3642 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3643 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3644
3645 #ifdef __OpenBSD__
3646 // OpenBSD pthread_setprio starves low priority threads
3647 return OS_OK;
3648 #elif defined(__FreeBSD__)
3649 int ret = pthread_setprio(thread->osthread()->pthread_id(), newpri);
3650 #elif defined(__APPLE__) || defined(__NetBSD__)
3651 struct sched_param sp;
3652 int policy;
3653 pthread_t self = pthread_self();
3654
3655 if (pthread_getschedparam(self, &policy, &sp) != 0)
3656 return OS_ERR;
3657
3658 sp.sched_priority = newpri;
3659 if (pthread_setschedparam(self, policy, &sp) != 0)
3660 return OS_ERR;
3661
3662 return OS_OK;
3663 #else
3664 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3665 return (ret == 0) ? OS_OK : OS_ERR;
3666 #endif
3667 }
3668
3669 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3670 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3671 *priority_ptr = java_to_os_priority[NormPriority];
3672 return OS_OK;
3673 }
3674
3675 errno = 0;
3676 #if defined(__OpenBSD__) || defined(__FreeBSD__)
3677 *priority_ptr = pthread_getprio(thread->osthread()->pthread_id());
3678 #elif defined(__APPLE__) || defined(__NetBSD__)
3679 int policy;
3680 struct sched_param sp;
3681
3682 pthread_getschedparam(pthread_self(), &policy, &sp);
3683 *priority_ptr = sp.sched_priority;
3684 #else
3685 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3686 #endif
3687 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3688 }
3689
3690 // Hint to the underlying OS that a task switch would not be good.
3691 // Void return because it's a hint and can fail.
3692 void os::hint_no_preempt() {}
3693
3694 ////////////////////////////////////////////////////////////////////////////////
3695 // suspend/resume support
3696
3697 // the low-level signal-based suspend/resume support is a remnant from the
3698 // old VM-suspension that used to be for java-suspension, safepoints etc,
3699 // within hotspot. Now there is a single use-case for this:
3700 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3701 // that runs in the watcher thread.
3702 // The remaining code is greatly simplified from the more general suspension
3703 // code that used to be used.
3704 //
3705 // The protocol is quite simple:
3706 // - suspend:
3707 // - sends a signal to the target thread
3708 // - polls the suspend state of the osthread using a yield loop
3709 // - target thread signal handler (SR_handler) sets suspend state
3710 // and blocks in sigsuspend until continued
3711 // - resume:
3712 // - sets target osthread state to continue
3713 // - sends signal to end the sigsuspend loop in the SR_handler
3714 //
3715 // Note that the SR_lock plays no role in this suspend/resume protocol.
3716 //
3717
3718 static void resume_clear_context(OSThread *osthread) {
3719 osthread->set_ucontext(NULL);
3720 osthread->set_siginfo(NULL);
3721
3722 // notify the suspend action is completed, we have now resumed
3723 osthread->sr.clear_suspended();
3724 }
3725
3726 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3727 osthread->set_ucontext(context);
3728 osthread->set_siginfo(siginfo);
3729 }
3730
3731 //
3732 // Handler function invoked when a thread's execution is suspended or
3733 // resumed. We have to be careful that only async-safe functions are
3734 // called here (Note: most pthread functions are not async safe and
3735 // should be avoided.)
3736 //
3737 // Note: sigwait() is a more natural fit than sigsuspend() from an
3738 // interface point of view, but sigwait() prevents the signal hander
3739 // from being run. libpthread would get very confused by not having
3740 // its signal handlers run and prevents sigwait()'s use with the
3741 // mutex granting granting signal.
3742 //
3743 // Currently only ever called on the VMThread
3744 //
3745 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3746 // Save and restore errno to avoid confusing native code with EINTR
3747 // after sigsuspend.
3748 int old_errno = errno;
3749
3750 Thread* thread = Thread::current();
3751 OSThread* osthread = thread->osthread();
3752 assert(thread->is_VM_thread(), "Must be VMThread");
3753 // read current suspend action
3754 int action = osthread->sr.suspend_action();
3755 if (action == SR_SUSPEND) {
3756 suspend_save_context(osthread, siginfo, context);
3757
3758 // Notify the suspend action is about to be completed. do_suspend()
3759 // waits until SR_SUSPENDED is set and then returns. We will wait
3760 // here for a resume signal and that completes the suspend-other
3761 // action. do_suspend/do_resume is always called as a pair from
3762 // the same thread - so there are no races
3763
3764 // notify the caller
3765 osthread->sr.set_suspended();
3766
3767 sigset_t suspend_set; // signals for sigsuspend()
3768
3769 // get current set of blocked signals and unblock resume signal
3770 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3771 sigdelset(&suspend_set, SR_signum);
3772
3773 // wait here until we are resumed
3774 do {
3775 sigsuspend(&suspend_set);
3776 // ignore all returns until we get a resume signal
3777 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3778
3779 resume_clear_context(osthread);
3780
3781 } else {
3782 assert(action == SR_CONTINUE, "unexpected sr action");
3783 // nothing special to do - just leave the handler
3784 }
3785
3786 errno = old_errno;
3787 }
3788
3789
3790 static int SR_initialize() {
3791 struct sigaction act;
3792 char *s;
3793 /* Get signal number to use for suspend/resume */
3794 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3795 int sig = ::strtol(s, 0, 10);
3796 if (sig > 0 || sig < NSIG) {
3797 SR_signum = sig;
3798 }
3799 }
3800
3801 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3802 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3803
3804 sigemptyset(&SR_sigset);
3805 sigaddset(&SR_sigset, SR_signum);
3806
3807 /* Set up signal handler for suspend/resume */
3808 act.sa_flags = SA_RESTART|SA_SIGINFO;
3809 act.sa_handler = (void (*)(int)) SR_handler;
3810
3811 // SR_signum is blocked by default.
3812 // 4528190 - We also need to block pthread restart signal (32 on all
3813 // supported Bsd platforms). Note that BsdThreads need to block
3814 // this signal for all threads to work properly. So we don't have
3815 // to use hard-coded signal number when setting up the mask.
3816 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3817
3818 if (sigaction(SR_signum, &act, 0) == -1) {
3819 return -1;
3820 }
3821
3822 // Save signal flag
3823 os::Bsd::set_our_sigflags(SR_signum, act.sa_flags);
3824 return 0;
3825 }
3826
3827 static int SR_finalize() {
3828 return 0;
3829 }
3830
3831
3832 // returns true on success and false on error - really an error is fatal
3833 // but this seems the normal response to library errors
3834 static bool do_suspend(OSThread* osthread) {
3835 // mark as suspended and send signal
3836 osthread->sr.set_suspend_action(SR_SUSPEND);
3837 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3838 assert_status(status == 0, status, "pthread_kill");
3839
3840 // check status and wait until notified of suspension
3841 if (status == 0) {
3842 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3843 os::yield_all(i);
3844 }
3845 osthread->sr.set_suspend_action(SR_NONE);
3846 return true;
3847 }
3848 else {
3849 osthread->sr.set_suspend_action(SR_NONE);
3850 return false;
3851 }
3852 }
3853
3854 static void do_resume(OSThread* osthread) {
3855 assert(osthread->sr.is_suspended(), "thread should be suspended");
3856 osthread->sr.set_suspend_action(SR_CONTINUE);
3857
3858 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3859 assert_status(status == 0, status, "pthread_kill");
3860 // check status and wait unit notified of resumption
3861 if (status == 0) {
3862 for (int i = 0; osthread->sr.is_suspended(); i++) {
3863 os::yield_all(i);
3864 }
3865 }
3866 osthread->sr.set_suspend_action(SR_NONE);
3867 }
3868
3869 ////////////////////////////////////////////////////////////////////////////////
3870 // interrupt support
3871
3872 void os::interrupt(Thread* thread) {
3873 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3874 "possibility of dangling Thread pointer");
3875
3876 OSThread* osthread = thread->osthread();
3877
3878 if (!osthread->interrupted()) {
3879 osthread->set_interrupted(true);
3880 // More than one thread can get here with the same value of osthread,
3881 // resulting in multiple notifications. We do, however, want the store
3882 // to interrupted() to be visible to other threads before we execute unpark().
3883 OrderAccess::fence();
3884 ParkEvent * const slp = thread->_SleepEvent ;
3885 if (slp != NULL) slp->unpark() ;
3886 }
3887
3888 // For JSR166. Unpark even if interrupt status already was set
3889 if (thread->is_Java_thread())
3890 ((JavaThread*)thread)->parker()->unpark();
3891
3892 ParkEvent * ev = thread->_ParkEvent ;
3893 if (ev != NULL) ev->unpark() ;
3894
3895 }
3896
3897 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3898 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3899 "possibility of dangling Thread pointer");
3900
3901 OSThread* osthread = thread->osthread();
3902
3903 bool interrupted = osthread->interrupted();
3904
3905 if (interrupted && clear_interrupted) {
3906 osthread->set_interrupted(false);
3907 // consider thread->_SleepEvent->reset() ... optional optimization
3908 }
3909
3910 return interrupted;
3911 }
3912
3913 ///////////////////////////////////////////////////////////////////////////////////
3914 // signal handling (except suspend/resume)
3915
3916 // This routine may be used by user applications as a "hook" to catch signals.
3917 // The user-defined signal handler must pass unrecognized signals to this
3918 // routine, and if it returns true (non-zero), then the signal handler must
3919 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3920 // routine will never retun false (zero), but instead will execute a VM panic
3921 // routine kill the process.
3922 //
3923 // If this routine returns false, it is OK to call it again. This allows
3924 // the user-defined signal handler to perform checks either before or after
3925 // the VM performs its own checks. Naturally, the user code would be making
3926 // a serious error if it tried to handle an exception (such as a null check
3927 // or breakpoint) that the VM was generating for its own correct operation.
3928 //
3929 // This routine may recognize any of the following kinds of signals:
3930 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3931 // It should be consulted by handlers for any of those signals.
3932 //
3933 // The caller of this routine must pass in the three arguments supplied
3934 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3935 // field of the structure passed to sigaction(). This routine assumes that
3936 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3937 //
3938 // Note that the VM will print warnings if it detects conflicting signal
3939 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3940 //
3941 extern "C" JNIEXPORT int
3942 JVM_handle_bsd_signal(int signo, siginfo_t* siginfo,
3943 void* ucontext, int abort_if_unrecognized);
3944
3945 void signalHandler(int sig, siginfo_t* info, void* uc) {
3946 assert(info != NULL && uc != NULL, "it must be old kernel");
3947 JVM_handle_bsd_signal(sig, info, uc, true);
3948 }
3949
3950
3951 // This boolean allows users to forward their own non-matching signals
3952 // to JVM_handle_bsd_signal, harmlessly.
3953 bool os::Bsd::signal_handlers_are_installed = false;
3954
3955 // For signal-chaining
3956 struct sigaction os::Bsd::sigact[MAXSIGNUM];
3957 unsigned int os::Bsd::sigs = 0;
3958 bool os::Bsd::libjsig_is_loaded = false;
3959 typedef struct sigaction *(*get_signal_t)(int);
3960 get_signal_t os::Bsd::get_signal_action = NULL;
3961
3962 struct sigaction* os::Bsd::get_chained_signal_action(int sig) {
3963 struct sigaction *actp = NULL;
3964
3965 if (libjsig_is_loaded) {
3966 // Retrieve the old signal handler from libjsig
3967 actp = (*get_signal_action)(sig);
3968 }
3969 if (actp == NULL) {
3970 // Retrieve the preinstalled signal handler from jvm
3971 actp = get_preinstalled_handler(sig);
3972 }
3973
3974 return actp;
3975 }
3976
3977 static bool call_chained_handler(struct sigaction *actp, int sig,
3978 siginfo_t *siginfo, void *context) {
3979 // Call the old signal handler
3980 if (actp->sa_handler == SIG_DFL) {
3981 // It's more reasonable to let jvm treat it as an unexpected exception
3982 // instead of taking the default action.
3983 return false;
3984 } else if (actp->sa_handler != SIG_IGN) {
3985 if ((actp->sa_flags & SA_NODEFER) == 0) {
3986 // automaticlly block the signal
3987 sigaddset(&(actp->sa_mask), sig);
3988 }
3989
3990 sa_handler_t hand;
3991 sa_sigaction_t sa;
3992 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3993 // retrieve the chained handler
3994 if (siginfo_flag_set) {
3995 sa = actp->sa_sigaction;
3996 } else {
3997 hand = actp->sa_handler;
3998 }
3999
4000 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4001 actp->sa_handler = SIG_DFL;
4002 }
4003
4004 // try to honor the signal mask
4005 sigset_t oset;
4006 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4007
4008 // call into the chained handler
4009 if (siginfo_flag_set) {
4010 (*sa)(sig, siginfo, context);
4011 } else {
4012 (*hand)(sig);
4013 }
4014
4015 // restore the signal mask
4016 pthread_sigmask(SIG_SETMASK, &oset, 0);
4017 }
4018 // Tell jvm's signal handler the signal is taken care of.
4019 return true;
4020 }
4021
4022 bool os::Bsd::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4023 bool chained = false;
4024 // signal-chaining
4025 if (UseSignalChaining) {
4026 struct sigaction *actp = get_chained_signal_action(sig);
4027 if (actp != NULL) {
4028 chained = call_chained_handler(actp, sig, siginfo, context);
4029 }
4030 }
4031 return chained;
4032 }
4033
4034 struct sigaction* os::Bsd::get_preinstalled_handler(int sig) {
4035 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4036 return &sigact[sig];
4037 }
4038 return NULL;
4039 }
4040
4041 void os::Bsd::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4042 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4043 sigact[sig] = oldAct;
4044 sigs |= (unsigned int)1 << sig;
4045 }
4046
4047 // for diagnostic
4048 int os::Bsd::sigflags[MAXSIGNUM];
4049
4050 int os::Bsd::get_our_sigflags(int sig) {
4051 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4052 return sigflags[sig];
4053 }
4054
4055 void os::Bsd::set_our_sigflags(int sig, int flags) {
4056 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4057 sigflags[sig] = flags;
4058 }
4059
4060 void os::Bsd::set_signal_handler(int sig, bool set_installed) {
4061 // Check for overwrite.
4062 struct sigaction oldAct;
4063 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4064
4065 void* oldhand = oldAct.sa_sigaction
4066 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4067 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4068 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4069 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4070 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4071 if (AllowUserSignalHandlers || !set_installed) {
4072 // Do not overwrite; user takes responsibility to forward to us.
4073 return;
4074 } else if (UseSignalChaining) {
4075 // save the old handler in jvm
4076 save_preinstalled_handler(sig, oldAct);
4077 // libjsig also interposes the sigaction() call below and saves the
4078 // old sigaction on it own.
4079 } else {
4080 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4081 "%#lx for signal %d.", (long)oldhand, sig));
4082 }
4083 }
4084
4085 struct sigaction sigAct;
4086 sigfillset(&(sigAct.sa_mask));
4087 sigAct.sa_handler = SIG_DFL;
4088 if (!set_installed) {
4089 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4090 } else {
4091 sigAct.sa_sigaction = signalHandler;
4092 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4093 }
4094 // Save flags, which are set by ours
4095 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4096 sigflags[sig] = sigAct.sa_flags;
4097
4098 int ret = sigaction(sig, &sigAct, &oldAct);
4099 assert(ret == 0, "check");
4100
4101 void* oldhand2 = oldAct.sa_sigaction
4102 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4103 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4104 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4105 }
4106
4107 // install signal handlers for signals that HotSpot needs to
4108 // handle in order to support Java-level exception handling.
4109
4110 void os::Bsd::install_signal_handlers() {
4111 if (!signal_handlers_are_installed) {
4112 signal_handlers_are_installed = true;
4113
4114 // signal-chaining
4115 typedef void (*signal_setting_t)();
4116 signal_setting_t begin_signal_setting = NULL;
4117 signal_setting_t end_signal_setting = NULL;
4118 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4119 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4120 if (begin_signal_setting != NULL) {
4121 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4122 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4123 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4124 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4125 libjsig_is_loaded = true;
4126 assert(UseSignalChaining, "should enable signal-chaining");
4127 }
4128 if (libjsig_is_loaded) {
4129 // Tell libjsig jvm is setting signal handlers
4130 (*begin_signal_setting)();
4131 }
4132
4133 set_signal_handler(SIGSEGV, true);
4134 set_signal_handler(SIGPIPE, true);
4135 set_signal_handler(SIGBUS, true);
4136 set_signal_handler(SIGILL, true);
4137 set_signal_handler(SIGFPE, true);
4138 set_signal_handler(SIGXFSZ, true);
4139
4140 #if defined(__APPLE__)
4141 // In Mac OS X 10.4, CrashReporter will write a crash log for all 'fatal' signals, including
4142 // signals caught and handled by the JVM. To work around this, we reset the mach task
4143 // signal handler that's placed on our process by CrashReporter. This disables
4144 // CrashReporter-based reporting.
4145 //
4146 // This work-around is not necessary for 10.5+, as CrashReporter no longer intercedes
4147 // on caught fatal signals.
4148 //
4149 // Additionally, gdb installs both standard BSD signal handlers, and mach exception
4150 // handlers. By replacing the existing task exception handler, we disable gdb's mach
4151 // exception handling, while leaving the standard BSD signal handlers functional.
4152 kern_return_t kr;
4153 kr = task_set_exception_ports(mach_task_self(),
4154 EXC_MASK_BAD_ACCESS | EXC_MASK_ARITHMETIC,
4155 MACH_PORT_NULL,
4156 EXCEPTION_STATE_IDENTITY,
4157 MACHINE_THREAD_STATE);
4158
4159 assert(kr == KERN_SUCCESS, "could not set mach task signal handler");
4160 #endif
4161
4162 if (libjsig_is_loaded) {
4163 // Tell libjsig jvm finishes setting signal handlers
4164 (*end_signal_setting)();
4165 }
4166
4167 // We don't activate signal checker if libjsig is in place, we trust ourselves
4168 // and if UserSignalHandler is installed all bets are off
4169 if (CheckJNICalls) {
4170 if (libjsig_is_loaded) {
4171 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4172 check_signals = false;
4173 }
4174 if (AllowUserSignalHandlers) {
4175 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4176 check_signals = false;
4177 }
4178 }
4179 }
4180 }
4181
4182 #ifndef _ALLBSD_SOURCE
4183 // This is the fastest way to get thread cpu time on Bsd.
4184 // Returns cpu time (user+sys) for any thread, not only for current.
4185 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4186 // It might work on 2.6.10+ with a special kernel/glibc patch.
4187 // For reference, please, see IEEE Std 1003.1-2004:
4188 // http://www.unix.org/single_unix_specification
4189
4190 jlong os::Bsd::fast_thread_cpu_time(clockid_t clockid) {
4191 struct timespec tp;
4192 int rc = os::Bsd::clock_gettime(clockid, &tp);
4193 assert(rc == 0, "clock_gettime is expected to return 0 code");
4194
4195 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4196 }
4197 #endif
4198
4199 /////
4200 // glibc on Bsd platform uses non-documented flag
4201 // to indicate, that some special sort of signal
4202 // trampoline is used.
4203 // We will never set this flag, and we should
4204 // ignore this flag in our diagnostic
4205 #ifdef SIGNIFICANT_SIGNAL_MASK
4206 #undef SIGNIFICANT_SIGNAL_MASK
4207 #endif
4208 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4209
4210 static const char* get_signal_handler_name(address handler,
4211 char* buf, int buflen) {
4212 int offset;
4213 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4214 if (found) {
4215 // skip directory names
4216 const char *p1, *p2;
4217 p1 = buf;
4218 size_t len = strlen(os::file_separator());
4219 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4220 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4221 } else {
4222 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4223 }
4224 return buf;
4225 }
4226
4227 static void print_signal_handler(outputStream* st, int sig,
4228 char* buf, size_t buflen) {
4229 struct sigaction sa;
4230
4231 sigaction(sig, NULL, &sa);
4232
4233 // See comment for SIGNIFICANT_SIGNAL_MASK define
4234 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4235
4236 st->print("%s: ", os::exception_name(sig, buf, buflen));
4237
4238 address handler = (sa.sa_flags & SA_SIGINFO)
4239 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4240 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4241
4242 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4243 st->print("SIG_DFL");
4244 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4245 st->print("SIG_IGN");
4246 } else {
4247 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4248 }
4249
4250 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4251
4252 address rh = VMError::get_resetted_sighandler(sig);
4253 // May be, handler was resetted by VMError?
4254 if(rh != NULL) {
4255 handler = rh;
4256 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4257 }
4258
4259 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
4260
4261 // Check: is it our handler?
4262 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4263 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4264 // It is our signal handler
4265 // check for flags, reset system-used one!
4266 if((int)sa.sa_flags != os::Bsd::get_our_sigflags(sig)) {
4267 st->print(
4268 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4269 os::Bsd::get_our_sigflags(sig));
4270 }
4271 }
4272 st->cr();
4273 }
4274
4275
4276 #define DO_SIGNAL_CHECK(sig) \
4277 if (!sigismember(&check_signal_done, sig)) \
4278 os::Bsd::check_signal_handler(sig)
4279
4280 // This method is a periodic task to check for misbehaving JNI applications
4281 // under CheckJNI, we can add any periodic checks here
4282
4283 void os::run_periodic_checks() {
4284
4285 if (check_signals == false) return;
4286
4287 // SEGV and BUS if overridden could potentially prevent
4288 // generation of hs*.log in the event of a crash, debugging
4289 // such a case can be very challenging, so we absolutely
4290 // check the following for a good measure:
4291 DO_SIGNAL_CHECK(SIGSEGV);
4292 DO_SIGNAL_CHECK(SIGILL);
4293 DO_SIGNAL_CHECK(SIGFPE);
4294 DO_SIGNAL_CHECK(SIGBUS);
4295 DO_SIGNAL_CHECK(SIGPIPE);
4296 DO_SIGNAL_CHECK(SIGXFSZ);
4297
4298
4299 // ReduceSignalUsage allows the user to override these handlers
4300 // see comments at the very top and jvm_solaris.h
4301 if (!ReduceSignalUsage) {
4302 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4303 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4304 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4305 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4306 }
4307
4308 DO_SIGNAL_CHECK(SR_signum);
4309 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4310 }
4311
4312 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4313
4314 static os_sigaction_t os_sigaction = NULL;
4315
4316 void os::Bsd::check_signal_handler(int sig) {
4317 char buf[O_BUFLEN];
4318 address jvmHandler = NULL;
4319
4320
4321 struct sigaction act;
4322 if (os_sigaction == NULL) {
4323 // only trust the default sigaction, in case it has been interposed
4324 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4325 if (os_sigaction == NULL) return;
4326 }
4327
4328 os_sigaction(sig, (struct sigaction*)NULL, &act);
4329
4330
4331 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4332
4333 address thisHandler = (act.sa_flags & SA_SIGINFO)
4334 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4335 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4336
4337
4338 switch(sig) {
4339 case SIGSEGV:
4340 case SIGBUS:
4341 case SIGFPE:
4342 case SIGPIPE:
4343 case SIGILL:
4344 case SIGXFSZ:
4345 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4346 break;
4347
4348 case SHUTDOWN1_SIGNAL:
4349 case SHUTDOWN2_SIGNAL:
4350 case SHUTDOWN3_SIGNAL:
4351 case BREAK_SIGNAL:
4352 jvmHandler = (address)user_handler();
4353 break;
4354
4355 case INTERRUPT_SIGNAL:
4356 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4357 break;
4358
4359 default:
4360 if (sig == SR_signum) {
4361 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4362 } else {
4363 return;
4364 }
4365 break;
4366 }
4367
4368 if (thisHandler != jvmHandler) {
4369 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4370 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4371 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4372 // No need to check this sig any longer
4373 sigaddset(&check_signal_done, sig);
4374 } else if(os::Bsd::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Bsd::get_our_sigflags(sig)) {
4375 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4376 tty->print("expected:" PTR32_FORMAT, os::Bsd::get_our_sigflags(sig));
4377 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4378 // No need to check this sig any longer
4379 sigaddset(&check_signal_done, sig);
4380 }
4381
4382 // Dump all the signal
4383 if (sigismember(&check_signal_done, sig)) {
4384 print_signal_handlers(tty, buf, O_BUFLEN);
4385 }
4386 }
4387
4388 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4389
4390 extern bool signal_name(int signo, char* buf, size_t len);
4391
4392 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4393 if (0 < exception_code && exception_code <= SIGRTMAX) {
4394 // signal
4395 if (!signal_name(exception_code, buf, size)) {
4396 jio_snprintf(buf, size, "SIG%d", exception_code);
4397 }
4398 return buf;
4399 } else {
4400 return NULL;
4401 }
4402 }
4403
4404 // this is called _before_ the most of global arguments have been parsed
4405 void os::init(void) {
4406 char dummy; /* used to get a guess on initial stack address */
4407 // first_hrtime = gethrtime();
4408
4409 // With BsdThreads the JavaMain thread pid (primordial thread)
4410 // is different than the pid of the java launcher thread.
4411 // So, on Bsd, the launcher thread pid is passed to the VM
4412 // via the sun.java.launcher.pid property.
4413 // Use this property instead of getpid() if it was correctly passed.
4414 // See bug 6351349.
4415 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4416
4417 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4418
4419 clock_tics_per_sec = CLK_TCK;
4420
4421 init_random(1234567);
4422
4423 ThreadCritical::initialize();
4424
4425 Bsd::set_page_size(getpagesize());
4426 if (Bsd::page_size() == -1) {
4427 fatal(err_msg("os_bsd.cpp: os::init: sysconf failed (%s)",
4428 strerror(errno)));
4429 }
4430 init_page_sizes((size_t) Bsd::page_size());
4431
4432 Bsd::initialize_system_info();
4433
4434 // main_thread points to the aboriginal thread
4435 Bsd::_main_thread = pthread_self();
4436
4437 Bsd::clock_init();
4438 initial_time_count = os::elapsed_counter();
4439
4440 #ifdef __APPLE__
4441 // XXXDARWIN
4442 // Work around the unaligned VM callbacks in hotspot's
4443 // sharedRuntime. The callbacks don't use SSE2 instructions, and work on
4444 // Linux, Solaris, and FreeBSD. On Mac OS X, dyld (rightly so) enforces
4445 // alignment when doing symbol lookup. To work around this, we force early
4446 // binding of all symbols now, thus binding when alignment is known-good.
4447 _dyld_bind_fully_image_containing_address((const void *) &os::init);
4448 #endif
4449 }
4450
4451 // To install functions for atexit system call
4452 extern "C" {
4453 static void perfMemory_exit_helper() {
4454 perfMemory_exit();
4455 }
4456 }
4457
4458 // this is called _after_ the global arguments have been parsed
4459 jint os::init_2(void)
4460 {
4461 #ifndef _ALLBSD_SOURCE
4462 Bsd::fast_thread_clock_init();
4463 #endif
4464
4465 // Allocate a single page and mark it as readable for safepoint polling
4466 address polling_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4467 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4468
4469 os::set_polling_page( polling_page );
4470
4471 #ifndef PRODUCT
4472 if(Verbose && PrintMiscellaneous)
4473 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4474 #endif
4475
4476 if (!UseMembar) {
4477 address mem_serialize_page = (address) ::mmap(NULL, Bsd::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4478 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4479 os::set_memory_serialize_page( mem_serialize_page );
4480
4481 #ifndef PRODUCT
4482 if(Verbose && PrintMiscellaneous)
4483 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4484 #endif
4485 }
4486
4487 os::large_page_init();
4488
4489 // initialize suspend/resume support - must do this before signal_sets_init()
4490 if (SR_initialize() != 0) {
4491 perror("SR_initialize failed");
4492 return JNI_ERR;
4493 }
4494
4495 Bsd::signal_sets_init();
4496 Bsd::install_signal_handlers();
4497
4498 // Check minimum allowable stack size for thread creation and to initialize
4499 // the java system classes, including StackOverflowError - depends on page
4500 // size. Add a page for compiler2 recursion in main thread.
4501 // Add in 2*BytesPerWord times page size to account for VM stack during
4502 // class initialization depending on 32 or 64 bit VM.
4503 os::Bsd::min_stack_allowed = MAX2(os::Bsd::min_stack_allowed,
4504 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4505 2*BytesPerWord COMPILER2_PRESENT(+1)) * Bsd::page_size());
4506
4507 size_t threadStackSizeInBytes = ThreadStackSize * K;
4508 if (threadStackSizeInBytes != 0 &&
4509 threadStackSizeInBytes < os::Bsd::min_stack_allowed) {
4510 tty->print_cr("\nThe stack size specified is too small, "
4511 "Specify at least %dk",
4512 os::Bsd::min_stack_allowed/ K);
4513 return JNI_ERR;
4514 }
4515
4516 // Make the stack size a multiple of the page size so that
4517 // the yellow/red zones can be guarded.
4518 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4519 vm_page_size()));
4520
4521 #ifndef _ALLBSD_SOURCE
4522 Bsd::capture_initial_stack(JavaThread::stack_size_at_create());
4523
4524 Bsd::libpthread_init();
4525 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4526 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4527 Bsd::glibc_version(), Bsd::libpthread_version(),
4528 Bsd::is_floating_stack() ? "floating stack" : "fixed stack");
4529 }
4530
4531 if (UseNUMA) {
4532 if (!Bsd::libnuma_init()) {
4533 UseNUMA = false;
4534 } else {
4535 if ((Bsd::numa_max_node() < 1)) {
4536 // There's only one node(they start from 0), disable NUMA.
4537 UseNUMA = false;
4538 }
4539 }
4540 // With SHM large pages we cannot uncommit a page, so there's not way
4541 // we can make the adaptive lgrp chunk resizing work. If the user specified
4542 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4543 // disable adaptive resizing.
4544 if (UseNUMA && UseLargePages && UseSHM) {
4545 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4546 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4547 UseLargePages = false;
4548 } else {
4549 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4550 UseAdaptiveSizePolicy = false;
4551 UseAdaptiveNUMAChunkSizing = false;
4552 }
4553 } else {
4554 UseNUMA = false;
4555 }
4556 }
4557 if (!UseNUMA && ForceNUMA) {
4558 UseNUMA = true;
4559 }
4560 }
4561 #endif
4562
4563 if (MaxFDLimit) {
4564 // set the number of file descriptors to max. print out error
4565 // if getrlimit/setrlimit fails but continue regardless.
4566 struct rlimit nbr_files;
4567 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4568 if (status != 0) {
4569 if (PrintMiscellaneous && (Verbose || WizardMode))
4570 perror("os::init_2 getrlimit failed");
4571 } else {
4572 nbr_files.rlim_cur = nbr_files.rlim_max;
4573
4574 #ifdef __APPLE__
4575 // Darwin returns RLIM_INFINITY for rlim_max, but fails with EINVAL if
4576 // you attempt to use RLIM_INFINITY. As per setrlimit(2), OPEN_MAX must
4577 // be used instead
4578 nbr_files.rlim_cur = MIN(OPEN_MAX, nbr_files.rlim_cur);
4579 #endif
4580
4581 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4582 if (status != 0) {
4583 if (PrintMiscellaneous && (Verbose || WizardMode))
4584 perror("os::init_2 setrlimit failed");
4585 }
4586 }
4587 }
4588
4589 #ifndef _ALLBSD_SOURCE
4590 // Initialize lock used to serialize thread creation (see os::create_thread)
4591 Bsd::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4592 #endif
4593
4594 // at-exit methods are called in the reverse order of their registration.
4595 // atexit functions are called on return from main or as a result of a
4596 // call to exit(3C). There can be only 32 of these functions registered
4597 // and atexit() does not set errno.
4598
4599 if (PerfAllowAtExitRegistration) {
4600 // only register atexit functions if PerfAllowAtExitRegistration is set.
4601 // atexit functions can be delayed until process exit time, which
4602 // can be problematic for embedded VM situations. Embedded VMs should
4603 // call DestroyJavaVM() to assure that VM resources are released.
4604
4605 // note: perfMemory_exit_helper atexit function may be removed in
4606 // the future if the appropriate cleanup code can be added to the
4607 // VM_Exit VMOperation's doit method.
4608 if (atexit(perfMemory_exit_helper) != 0) {
4609 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4610 }
4611 }
4612
4613 // initialize thread priority policy
4614 prio_init();
4615
4616 #ifdef __APPLE__
4617 // dynamically link to objective c gc registration
4618 void *handleLibObjc = dlopen(OBJC_LIB, RTLD_LAZY);
4619 if (handleLibObjc != NULL) {
4620 objc_registerThreadWithCollectorFunction = (objc_registerThreadWithCollector_t) dlsym(handleLibObjc, OBJC_GCREGISTER);
4621 }
4622 #endif
4623
4624 return JNI_OK;
4625 }
4626
4627 // this is called at the end of vm_initialization
4628 void os::init_3(void) { }
4629
4630 // Mark the polling page as unreadable
4631 void os::make_polling_page_unreadable(void) {
4632 if( !guard_memory((char*)_polling_page, Bsd::page_size()) )
4633 fatal("Could not disable polling page");
4634 };
4635
4636 // Mark the polling page as readable
4637 void os::make_polling_page_readable(void) {
4638 if( !bsd_mprotect((char *)_polling_page, Bsd::page_size(), PROT_READ)) {
4639 fatal("Could not enable polling page");
4640 }
4641 };
4642
4643 int os::active_processor_count() {
4644 #ifdef _ALLBSD_SOURCE
4645 return _processor_count;
4646 #else
4647 // Bsd doesn't yet have a (official) notion of processor sets,
4648 // so just return the number of online processors.
4649 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4650 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4651 return online_cpus;
4652 #endif
4653 }
4654
4655 void os::set_native_thread_name(const char *name) {
4656 #if defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED > MAC_OS_X_VERSION_10_5
4657 // This is only supported in Snow Leopard and beyond
4658 if (name != NULL) {
4659 // Add a "Java: " prefix to the name
4660 char buf[MAXTHREADNAMESIZE];
4661 snprintf(buf, sizeof(buf), "Java: %s", name);
4662 pthread_setname_np(buf);
4663 }
4664 #endif
4665 }
4666
4667 bool os::distribute_processes(uint length, uint* distribution) {
4668 // Not yet implemented.
4669 return false;
4670 }
4671
4672 bool os::bind_to_processor(uint processor_id) {
4673 // Not yet implemented.
4674 return false;
4675 }
4676
4677 ///
4678
4679 // Suspends the target using the signal mechanism and then grabs the PC before
4680 // resuming the target. Used by the flat-profiler only
4681 ExtendedPC os::get_thread_pc(Thread* thread) {
4682 // Make sure that it is called by the watcher for the VMThread
4683 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4684 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4685
4686 ExtendedPC epc;
4687
4688 OSThread* osthread = thread->osthread();
4689 if (do_suspend(osthread)) {
4690 if (osthread->ucontext() != NULL) {
4691 epc = os::Bsd::ucontext_get_pc(osthread->ucontext());
4692 } else {
4693 // NULL context is unexpected, double-check this is the VMThread
4694 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4695 }
4696 do_resume(osthread);
4697 }
4698 // failure means pthread_kill failed for some reason - arguably this is
4699 // a fatal problem, but such problems are ignored elsewhere
4700
4701 return epc;
4702 }
4703
4704 int os::Bsd::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4705 {
4706 #ifdef _ALLBSD_SOURCE
4707 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4708 #else
4709 if (is_NPTL()) {
4710 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4711 } else {
4712 #ifndef IA64
4713 // 6292965: BsdThreads pthread_cond_timedwait() resets FPU control
4714 // word back to default 64bit precision if condvar is signaled. Java
4715 // wants 53bit precision. Save and restore current value.
4716 int fpu = get_fpu_control_word();
4717 #endif // IA64
4718 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4719 #ifndef IA64
4720 set_fpu_control_word(fpu);
4721 #endif // IA64
4722 return status;
4723 }
4724 #endif
4725 }
4726
4727 ////////////////////////////////////////////////////////////////////////////////
4728 // debug support
4729
4730 static address same_page(address x, address y) {
4731 int page_bits = -os::vm_page_size();
4732 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4733 return x;
4734 else if (x > y)
4735 return (address)(intptr_t(y) | ~page_bits) + 1;
4736 else
4737 return (address)(intptr_t(y) & page_bits);
4738 }
4739
4740 bool os::find(address addr, outputStream* st) {
4741 Dl_info dlinfo;
4742 memset(&dlinfo, 0, sizeof(dlinfo));
4743 if (dladdr(addr, &dlinfo)) {
4744 st->print(PTR_FORMAT ": ", addr);
4745 if (dlinfo.dli_sname != NULL) {
4746 st->print("%s+%#x", dlinfo.dli_sname,
4747 addr - (intptr_t)dlinfo.dli_saddr);
4748 } else if (dlinfo.dli_fname) {
4749 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4750 } else {
4751 st->print("<absolute address>");
4752 }
4753 if (dlinfo.dli_fname) {
4754 st->print(" in %s", dlinfo.dli_fname);
4755 }
4756 if (dlinfo.dli_fbase) {
4757 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4758 }
4759 st->cr();
4760
4761 if (Verbose) {
4762 // decode some bytes around the PC
4763 address begin = same_page(addr-40, addr);
4764 address end = same_page(addr+40, addr);
4765 address lowest = (address) dlinfo.dli_sname;
4766 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4767 if (begin < lowest) begin = lowest;
4768 Dl_info dlinfo2;
4769 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4770 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4771 end = (address) dlinfo2.dli_saddr;
4772 Disassembler::decode(begin, end, st);
4773 }
4774 return true;
4775 }
4776 return false;
4777 }
4778
4779 ////////////////////////////////////////////////////////////////////////////////
4780 // misc
4781
4782 // This does not do anything on Bsd. This is basically a hook for being
4783 // able to use structured exception handling (thread-local exception filters)
4784 // on, e.g., Win32.
4785 void
4786 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4787 JavaCallArguments* args, Thread* thread) {
4788 f(value, method, args, thread);
4789 }
4790
4791 void os::print_statistics() {
4792 }
4793
4794 int os::message_box(const char* title, const char* message) {
4795 int i;
4796 fdStream err(defaultStream::error_fd());
4797 for (i = 0; i < 78; i++) err.print_raw("=");
4798 err.cr();
4799 err.print_raw_cr(title);
4800 for (i = 0; i < 78; i++) err.print_raw("-");
4801 err.cr();
4802 err.print_raw_cr(message);
4803 for (i = 0; i < 78; i++) err.print_raw("=");
4804 err.cr();
4805
4806 char buf[16];
4807 // Prevent process from exiting upon "read error" without consuming all CPU
4808 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4809
4810 return buf[0] == 'y' || buf[0] == 'Y';
4811 }
4812
4813 int os::stat(const char *path, struct stat *sbuf) {
4814 char pathbuf[MAX_PATH];
4815 if (strlen(path) > MAX_PATH - 1) {
4816 errno = ENAMETOOLONG;
4817 return -1;
4818 }
4819 os::native_path(strcpy(pathbuf, path));
4820 return ::stat(pathbuf, sbuf);
4821 }
4822
4823 bool os::check_heap(bool force) {
4824 return true;
4825 }
4826
4827 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4828 return ::vsnprintf(buf, count, format, args);
4829 }
4830
4831 // Is a (classpath) directory empty?
4832 bool os::dir_is_empty(const char* path) {
4833 DIR *dir = NULL;
4834 struct dirent *ptr;
4835
4836 dir = opendir(path);
4837 if (dir == NULL) return true;
4838
4839 /* Scan the directory */
4840 bool result = true;
4841 char buf[sizeof(struct dirent) + MAX_PATH];
4842 while (result && (ptr = ::readdir(dir)) != NULL) {
4843 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4844 result = false;
4845 }
4846 }
4847 closedir(dir);
4848 return result;
4849 }
4850
4851 // This code originates from JDK's sysOpen and open64_w
4852 // from src/solaris/hpi/src/system_md.c
4853
4854 #ifndef O_DELETE
4855 #define O_DELETE 0x10000
4856 #endif
4857
4858 // Open a file. Unlink the file immediately after open returns
4859 // if the specified oflag has the O_DELETE flag set.
4860 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4861
4862 int os::open(const char *path, int oflag, int mode) {
4863
4864 if (strlen(path) > MAX_PATH - 1) {
4865 errno = ENAMETOOLONG;
4866 return -1;
4867 }
4868 int fd;
4869 int o_delete = (oflag & O_DELETE);
4870 oflag = oflag & ~O_DELETE;
4871
4872 fd = ::open(path, oflag, mode);
4873 if (fd == -1) return -1;
4874
4875 //If the open succeeded, the file might still be a directory
4876 {
4877 struct stat buf;
4878 int ret = ::fstat(fd, &buf);
4879 int st_mode = buf.st_mode;
4880
4881 if (ret != -1) {
4882 if ((st_mode & S_IFMT) == S_IFDIR) {
4883 errno = EISDIR;
4884 ::close(fd);
4885 return -1;
4886 }
4887 } else {
4888 ::close(fd);
4889 return -1;
4890 }
4891 }
4892
4893 /*
4894 * All file descriptors that are opened in the JVM and not
4895 * specifically destined for a subprocess should have the
4896 * close-on-exec flag set. If we don't set it, then careless 3rd
4897 * party native code might fork and exec without closing all
4898 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4899 * UNIXProcess.c), and this in turn might:
4900 *
4901 * - cause end-of-file to fail to be detected on some file
4902 * descriptors, resulting in mysterious hangs, or
4903 *
4904 * - might cause an fopen in the subprocess to fail on a system
4905 * suffering from bug 1085341.
4906 *
4907 * (Yes, the default setting of the close-on-exec flag is a Unix
4908 * design flaw)
4909 *
4910 * See:
4911 * 1085341: 32-bit stdio routines should support file descriptors >255
4912 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4913 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4914 */
4915 #ifdef FD_CLOEXEC
4916 {
4917 int flags = ::fcntl(fd, F_GETFD);
4918 if (flags != -1)
4919 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4920 }
4921 #endif
4922
4923 if (o_delete != 0) {
4924 ::unlink(path);
4925 }
4926 return fd;
4927 }
4928
4929
4930 // create binary file, rewriting existing file if required
4931 int os::create_binary_file(const char* path, bool rewrite_existing) {
4932 int oflags = O_WRONLY | O_CREAT;
4933 if (!rewrite_existing) {
4934 oflags |= O_EXCL;
4935 }
4936 return ::open(path, oflags, S_IREAD | S_IWRITE);
4937 }
4938
4939 // return current position of file pointer
4940 jlong os::current_file_offset(int fd) {
4941 return (jlong)::lseek(fd, (off_t)0, SEEK_CUR);
4942 }
4943
4944 // move file pointer to the specified offset
4945 jlong os::seek_to_file_offset(int fd, jlong offset) {
4946 return (jlong)::lseek(fd, (off_t)offset, SEEK_SET);
4947 }
4948
4949 // This code originates from JDK's sysAvailable
4950 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4951
4952 int os::available(int fd, jlong *bytes) {
4953 jlong cur, end;
4954 int mode;
4955 struct stat buf;
4956
4957 if (::fstat(fd, &buf) >= 0) {
4958 mode = buf.st_mode;
4959 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4960 /*
4961 * XXX: is the following call interruptible? If so, this might
4962 * need to go through the INTERRUPT_IO() wrapper as for other
4963 * blocking, interruptible calls in this file.
4964 */
4965 int n;
4966 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4967 *bytes = n;
4968 return 1;
4969 }
4970 }
4971 }
4972 if ((cur = ::lseek(fd, 0L, SEEK_CUR)) == -1) {
4973 return 0;
4974 } else if ((end = ::lseek(fd, 0L, SEEK_END)) == -1) {
4975 return 0;
4976 } else if (::lseek(fd, cur, SEEK_SET) == -1) {
4977 return 0;
4978 }
4979 *bytes = end - cur;
4980 return 1;
4981 }
4982
4983 int os::socket_available(int fd, jint *pbytes) {
4984 if (fd < 0)
4985 return OS_OK;
4986
4987 int ret;
4988
4989 RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
4990
4991 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4992 // is expected to return 0 on failure and 1 on success to the jdk.
4993
4994 return (ret == OS_ERR) ? 0 : 1;
4995 }
4996
4997 // Map a block of memory.
4998 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4999 char *addr, size_t bytes, bool read_only,
5000 bool allow_exec) {
5001 int prot;
5002 int flags;
5003
5004 if (read_only) {
5005 prot = PROT_READ;
5006 flags = MAP_SHARED;
5007 } else {
5008 prot = PROT_READ | PROT_WRITE;
5009 flags = MAP_PRIVATE;
5010 }
5011
5012 if (allow_exec) {
5013 prot |= PROT_EXEC;
5014 }
5015
5016 if (addr != NULL) {
5017 flags |= MAP_FIXED;
5018 }
5019
5020 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5021 fd, file_offset);
5022 if (mapped_address == MAP_FAILED) {
5023 return NULL;
5024 }
5025 return mapped_address;
5026 }
5027
5028
5029 // Remap a block of memory.
5030 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5031 char *addr, size_t bytes, bool read_only,
5032 bool allow_exec) {
5033 // same as map_memory() on this OS
5034 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5035 allow_exec);
5036 }
5037
5038
5039 // Unmap a block of memory.
5040 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5041 return munmap(addr, bytes) == 0;
5042 }
5043
5044 #ifndef _ALLBSD_SOURCE
5045 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5046
5047 static clockid_t thread_cpu_clockid(Thread* thread) {
5048 pthread_t tid = thread->osthread()->pthread_id();
5049 clockid_t clockid;
5050
5051 // Get thread clockid
5052 int rc = os::Bsd::pthread_getcpuclockid(tid, &clockid);
5053 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5054 return clockid;
5055 }
5056 #endif
5057
5058 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5059 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5060 // of a thread.
5061 //
5062 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5063 // the fast estimate available on the platform.
5064
5065 jlong os::current_thread_cpu_time() {
5066 #ifdef __APPLE__
5067 return os::thread_cpu_time(Thread::current(), true /* user + sys */);
5068 #elif !defined(_ALLBSD_SOURCE)
5069 if (os::Bsd::supports_fast_thread_cpu_time()) {
5070 return os::Bsd::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5071 } else {
5072 // return user + sys since the cost is the same
5073 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5074 }
5075 #endif
5076 }
5077
5078 jlong os::thread_cpu_time(Thread* thread) {
5079 #ifndef _ALLBSD_SOURCE
5080 // consistent with what current_thread_cpu_time() returns
5081 if (os::Bsd::supports_fast_thread_cpu_time()) {
5082 return os::Bsd::fast_thread_cpu_time(thread_cpu_clockid(thread));
5083 } else {
5084 return slow_thread_cpu_time(thread, true /* user + sys */);
5085 }
5086 #endif
5087 }
5088
5089 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5090 #ifdef __APPLE__
5091 return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5092 #elif !defined(_ALLBSD_SOURCE)
5093 if (user_sys_cpu_time && os::Bsd::supports_fast_thread_cpu_time()) {
5094 return os::Bsd::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5095 } else {
5096 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5097 }
5098 #endif
5099 }
5100
5101 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5102 #ifdef __APPLE__
5103 struct thread_basic_info tinfo;
5104 mach_msg_type_number_t tcount = THREAD_INFO_MAX;
5105 kern_return_t kr;
5106 thread_t mach_thread;
5107
5108 mach_thread = thread->osthread()->thread_id();
5109 kr = thread_info(mach_thread, THREAD_BASIC_INFO, (thread_info_t)&tinfo, &tcount);
5110 if (kr != KERN_SUCCESS)
5111 return -1;
5112
5113 if (user_sys_cpu_time) {
5114 jlong nanos;
5115 nanos = ((jlong) tinfo.system_time.seconds + tinfo.user_time.seconds) * (jlong)1000000000;
5116 nanos += ((jlong) tinfo.system_time.microseconds + (jlong) tinfo.user_time.microseconds) * (jlong)1000;
5117 return nanos;
5118 } else {
5119 return ((jlong)tinfo.user_time.seconds * 1000000000) + ((jlong)tinfo.user_time.microseconds * (jlong)1000);
5120 }
5121 #elif !defined(_ALLBSD_SOURCE)
5122 if (user_sys_cpu_time && os::Bsd::supports_fast_thread_cpu_time()) {
5123 return os::Bsd::fast_thread_cpu_time(thread_cpu_clockid(thread));
5124 } else {
5125 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5126 }
5127 #endif
5128 }
5129
5130 #ifndef _ALLBSD_SOURCE
5131 //
5132 // -1 on error.
5133 //
5134
5135 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5136 static bool proc_pid_cpu_avail = true;
5137 static bool proc_task_unchecked = true;
5138 static const char *proc_stat_path = "/proc/%d/stat";
5139 pid_t tid = thread->osthread()->thread_id();
5140 int i;
5141 char *s;
5142 char stat[2048];
5143 int statlen;
5144 char proc_name[64];
5145 int count;
5146 long sys_time, user_time;
5147 char string[64];
5148 char cdummy;
5149 int idummy;
5150 long ldummy;
5151 FILE *fp;
5152
5153 // We first try accessing /proc/<pid>/cpu since this is faster to
5154 // process. If this file is not present (bsd kernels 2.5 and above)
5155 // then we open /proc/<pid>/stat.
5156 if ( proc_pid_cpu_avail ) {
5157 sprintf(proc_name, "/proc/%d/cpu", tid);
5158 fp = fopen(proc_name, "r");
5159 if ( fp != NULL ) {
5160 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
5161 fclose(fp);
5162 if ( count != 3 ) return -1;
5163
5164 if (user_sys_cpu_time) {
5165 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5166 } else {
5167 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5168 }
5169 }
5170 else proc_pid_cpu_avail = false;
5171 }
5172
5173 // The /proc/<tid>/stat aggregates per-process usage on
5174 // new Bsd kernels 2.6+ where NPTL is supported.
5175 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5176 // See bug 6328462.
5177 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
5178 // and possibly in some other cases, so we check its availability.
5179 if (proc_task_unchecked && os::Bsd::is_NPTL()) {
5180 // This is executed only once
5181 proc_task_unchecked = false;
5182 fp = fopen("/proc/self/task", "r");
5183 if (fp != NULL) {
5184 proc_stat_path = "/proc/self/task/%d/stat";
5185 fclose(fp);
5186 }
5187 }
5188
5189 sprintf(proc_name, proc_stat_path, tid);
5190 fp = fopen(proc_name, "r");
5191 if ( fp == NULL ) return -1;
5192 statlen = fread(stat, 1, 2047, fp);
5193 stat[statlen] = '\0';
5194 fclose(fp);
5195
5196 // Skip pid and the command string. Note that we could be dealing with
5197 // weird command names, e.g. user could decide to rename java launcher
5198 // to "java 1.4.2 :)", then the stat file would look like
5199 // 1234 (java 1.4.2 :)) R ... ...
5200 // We don't really need to know the command string, just find the last
5201 // occurrence of ")" and then start parsing from there. See bug 4726580.
5202 s = strrchr(stat, ')');
5203 i = 0;
5204 if (s == NULL ) return -1;
5205
5206 // Skip blank chars
5207 do s++; while (isspace(*s));
5208
5209 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5210 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5211 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5212 &user_time, &sys_time);
5213 if ( count != 13 ) return -1;
5214 if (user_sys_cpu_time) {
5215 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5216 } else {
5217 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5218 }
5219 }
5220 #endif
5221
5222 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5223 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5224 info_ptr->may_skip_backward = false; // elapsed time not wall time
5225 info_ptr->may_skip_forward = false; // elapsed time not wall time
5226 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5227 }
5228
5229 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5230 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5231 info_ptr->may_skip_backward = false; // elapsed time not wall time
5232 info_ptr->may_skip_forward = false; // elapsed time not wall time
5233 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5234 }
5235
5236 bool os::is_thread_cpu_time_supported() {
5237 #ifdef __APPLE__
5238 return true;
5239 #elif defined(_ALLBSD_SOURCE)
5240 return false;
5241 #else
5242 return true;
5243 #endif
5244 }
5245
5246 // System loadavg support. Returns -1 if load average cannot be obtained.
5247 // Bsd doesn't yet have a (official) notion of processor sets,
5248 // so just return the system wide load average.
5249 int os::loadavg(double loadavg[], int nelem) {
5250 return ::getloadavg(loadavg, nelem);
5251 }
5252
5253 void os::pause() {
5254 char filename[MAX_PATH];
5255 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5256 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5257 } else {
5258 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5259 }
5260
5261 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5262 if (fd != -1) {
5263 struct stat buf;
5264 ::close(fd);
5265 while (::stat(filename, &buf) == 0) {
5266 (void)::poll(NULL, 0, 100);
5267 }
5268 } else {
5269 jio_fprintf(stderr,
5270 "Could not open pause file '%s', continuing immediately.\n", filename);
5271 }
5272 }
5273
5274
5275 // Refer to the comments in os_solaris.cpp park-unpark.
5276 //
5277 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5278 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5279 // For specifics regarding the bug see GLIBC BUGID 261237 :
5280 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5281 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5282 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5283 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5284 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5285 // and monitorenter when we're using 1-0 locking. All those operations may result in
5286 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5287 // of libpthread avoids the problem, but isn't practical.
5288 //
5289 // Possible remedies:
5290 //
5291 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5292 // This is palliative and probabilistic, however. If the thread is preempted
5293 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5294 // than the minimum period may have passed, and the abstime may be stale (in the
5295 // past) resultin in a hang. Using this technique reduces the odds of a hang
5296 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5297 //
5298 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5299 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5300 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5301 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5302 // thread.
5303 //
5304 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5305 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5306 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5307 // This also works well. In fact it avoids kernel-level scalability impediments
5308 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5309 // timers in a graceful fashion.
5310 //
5311 // 4. When the abstime value is in the past it appears that control returns
5312 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5313 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5314 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5315 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5316 // It may be possible to avoid reinitialization by checking the return
5317 // value from pthread_cond_timedwait(). In addition to reinitializing the
5318 // condvar we must establish the invariant that cond_signal() is only called
5319 // within critical sections protected by the adjunct mutex. This prevents
5320 // cond_signal() from "seeing" a condvar that's in the midst of being
5321 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5322 // desirable signal-after-unlock optimization that avoids futile context switching.
5323 //
5324 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5325 // structure when a condvar is used or initialized. cond_destroy() would
5326 // release the helper structure. Our reinitialize-after-timedwait fix
5327 // put excessive stress on malloc/free and locks protecting the c-heap.
5328 //
5329 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5330 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5331 // and only enabling the work-around for vulnerable environments.
5332
5333 // utility to compute the abstime argument to timedwait:
5334 // millis is the relative timeout time
5335 // abstime will be the absolute timeout time
5336 // TODO: replace compute_abstime() with unpackTime()
5337
5338 static struct timespec* compute_abstime(struct timespec* abstime, jlong millis) {
5339 if (millis < 0) millis = 0;
5340 struct timeval now;
5341 int status = gettimeofday(&now, NULL);
5342 assert(status == 0, "gettimeofday");
5343 jlong seconds = millis / 1000;
5344 millis %= 1000;
5345 if (seconds > 50000000) { // see man cond_timedwait(3T)
5346 seconds = 50000000;
5347 }
5348 abstime->tv_sec = now.tv_sec + seconds;
5349 long usec = now.tv_usec + millis * 1000;
5350 if (usec >= 1000000) {
5351 abstime->tv_sec += 1;
5352 usec -= 1000000;
5353 }
5354 abstime->tv_nsec = usec * 1000;
5355 return abstime;
5356 }
5357
5358
5359 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5360 // Conceptually TryPark() should be equivalent to park(0).
5361
5362 int os::PlatformEvent::TryPark() {
5363 for (;;) {
5364 const int v = _Event ;
5365 guarantee ((v == 0) || (v == 1), "invariant") ;
5366 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5367 }
5368 }
5369
5370 void os::PlatformEvent::park() { // AKA "down()"
5371 // Invariant: Only the thread associated with the Event/PlatformEvent
5372 // may call park().
5373 // TODO: assert that _Assoc != NULL or _Assoc == Self
5374 int v ;
5375 for (;;) {
5376 v = _Event ;
5377 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5378 }
5379 guarantee (v >= 0, "invariant") ;
5380 if (v == 0) {
5381 // Do this the hard way by blocking ...
5382 int status = pthread_mutex_lock(_mutex);
5383 assert_status(status == 0, status, "mutex_lock");
5384 guarantee (_nParked == 0, "invariant") ;
5385 ++ _nParked ;
5386 while (_Event < 0) {
5387 status = pthread_cond_wait(_cond, _mutex);
5388 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5389 // Treat this the same as if the wait was interrupted
5390 if (status == ETIMEDOUT) { status = EINTR; }
5391 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5392 }
5393 -- _nParked ;
5394
5395 // In theory we could move the ST of 0 into _Event past the unlock(),
5396 // but then we'd need a MEMBAR after the ST.
5397 _Event = 0 ;
5398 status = pthread_mutex_unlock(_mutex);
5399 assert_status(status == 0, status, "mutex_unlock");
5400 }
5401 guarantee (_Event >= 0, "invariant") ;
5402 }
5403
5404 int os::PlatformEvent::park(jlong millis) {
5405 guarantee (_nParked == 0, "invariant") ;
5406
5407 int v ;
5408 for (;;) {
5409 v = _Event ;
5410 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5411 }
5412 guarantee (v >= 0, "invariant") ;
5413 if (v != 0) return OS_OK ;
5414
5415 // We do this the hard way, by blocking the thread.
5416 // Consider enforcing a minimum timeout value.
5417 struct timespec abst;
5418 compute_abstime(&abst, millis);
5419
5420 int ret = OS_TIMEOUT;
5421 int status = pthread_mutex_lock(_mutex);
5422 assert_status(status == 0, status, "mutex_lock");
5423 guarantee (_nParked == 0, "invariant") ;
5424 ++_nParked ;
5425
5426 // Object.wait(timo) will return because of
5427 // (a) notification
5428 // (b) timeout
5429 // (c) thread.interrupt
5430 //
5431 // Thread.interrupt and object.notify{All} both call Event::set.
5432 // That is, we treat thread.interrupt as a special case of notification.
5433 // The underlying Solaris implementation, cond_timedwait, admits
5434 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5435 // JVM from making those visible to Java code. As such, we must
5436 // filter out spurious wakeups. We assume all ETIME returns are valid.
5437 //
5438 // TODO: properly differentiate simultaneous notify+interrupt.
5439 // In that case, we should propagate the notify to another waiter.
5440
5441 while (_Event < 0) {
5442 status = os::Bsd::safe_cond_timedwait(_cond, _mutex, &abst);
5443 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5444 pthread_cond_destroy (_cond);
5445 pthread_cond_init (_cond, NULL) ;
5446 }
5447 assert_status(status == 0 || status == EINTR ||
5448 status == ETIMEDOUT,
5449 status, "cond_timedwait");
5450 if (!FilterSpuriousWakeups) break ; // previous semantics
5451 if (status == ETIMEDOUT) break ;
5452 // We consume and ignore EINTR and spurious wakeups.
5453 }
5454 --_nParked ;
5455 if (_Event >= 0) {
5456 ret = OS_OK;
5457 }
5458 _Event = 0 ;
5459 status = pthread_mutex_unlock(_mutex);
5460 assert_status(status == 0, status, "mutex_unlock");
5461 assert (_nParked == 0, "invariant") ;
5462 return ret;
5463 }
5464
5465 void os::PlatformEvent::unpark() {
5466 int v, AnyWaiters ;
5467 for (;;) {
5468 v = _Event ;
5469 if (v > 0) {
5470 // The LD of _Event could have reordered or be satisfied
5471 // by a read-aside from this processor's write buffer.
5472 // To avoid problems execute a barrier and then
5473 // ratify the value.
5474 OrderAccess::fence() ;
5475 if (_Event == v) return ;
5476 continue ;
5477 }
5478 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5479 }
5480 if (v < 0) {
5481 // Wait for the thread associated with the event to vacate
5482 int status = pthread_mutex_lock(_mutex);
5483 assert_status(status == 0, status, "mutex_lock");
5484 AnyWaiters = _nParked ;
5485 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5486 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5487 AnyWaiters = 0 ;
5488 pthread_cond_signal (_cond);
5489 }
5490 status = pthread_mutex_unlock(_mutex);
5491 assert_status(status == 0, status, "mutex_unlock");
5492 if (AnyWaiters != 0) {
5493 status = pthread_cond_signal(_cond);
5494 assert_status(status == 0, status, "cond_signal");
5495 }
5496 }
5497
5498 // Note that we signal() _after dropping the lock for "immortal" Events.
5499 // This is safe and avoids a common class of futile wakeups. In rare
5500 // circumstances this can cause a thread to return prematurely from
5501 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5502 // simply re-test the condition and re-park itself.
5503 }
5504
5505
5506 // JSR166
5507 // -------------------------------------------------------
5508
5509 /*
5510 * The solaris and bsd implementations of park/unpark are fairly
5511 * conservative for now, but can be improved. They currently use a
5512 * mutex/condvar pair, plus a a count.
5513 * Park decrements count if > 0, else does a condvar wait. Unpark
5514 * sets count to 1 and signals condvar. Only one thread ever waits
5515 * on the condvar. Contention seen when trying to park implies that someone
5516 * is unparking you, so don't wait. And spurious returns are fine, so there
5517 * is no need to track notifications.
5518 */
5519
5520 #define MAX_SECS 100000000
5521 /*
5522 * This code is common to bsd and solaris and will be moved to a
5523 * common place in dolphin.
5524 *
5525 * The passed in time value is either a relative time in nanoseconds
5526 * or an absolute time in milliseconds. Either way it has to be unpacked
5527 * into suitable seconds and nanoseconds components and stored in the
5528 * given timespec structure.
5529 * Given time is a 64-bit value and the time_t used in the timespec is only
5530 * a signed-32-bit value (except on 64-bit Bsd) we have to watch for
5531 * overflow if times way in the future are given. Further on Solaris versions
5532 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5533 * number of seconds, in abstime, is less than current_time + 100,000,000.
5534 * As it will be 28 years before "now + 100000000" will overflow we can
5535 * ignore overflow and just impose a hard-limit on seconds using the value
5536 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5537 * years from "now".
5538 */
5539
5540 static void unpackTime(struct timespec* absTime, bool isAbsolute, jlong time) {
5541 assert (time > 0, "convertTime");
5542
5543 struct timeval now;
5544 int status = gettimeofday(&now, NULL);
5545 assert(status == 0, "gettimeofday");
5546
5547 time_t max_secs = now.tv_sec + MAX_SECS;
5548
5549 if (isAbsolute) {
5550 jlong secs = time / 1000;
5551 if (secs > max_secs) {
5552 absTime->tv_sec = max_secs;
5553 }
5554 else {
5555 absTime->tv_sec = secs;
5556 }
5557 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5558 }
5559 else {
5560 jlong secs = time / NANOSECS_PER_SEC;
5561 if (secs >= MAX_SECS) {
5562 absTime->tv_sec = max_secs;
5563 absTime->tv_nsec = 0;
5564 }
5565 else {
5566 absTime->tv_sec = now.tv_sec + secs;
5567 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5568 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5569 absTime->tv_nsec -= NANOSECS_PER_SEC;
5570 ++absTime->tv_sec; // note: this must be <= max_secs
5571 }
5572 }
5573 }
5574 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5575 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5576 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5577 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5578 }
5579
5580 void Parker::park(bool isAbsolute, jlong time) {
5581 // Optional fast-path check:
5582 // Return immediately if a permit is available.
5583 if (_counter > 0) {
5584 _counter = 0 ;
5585 OrderAccess::fence();
5586 return ;
5587 }
5588
5589 Thread* thread = Thread::current();
5590 assert(thread->is_Java_thread(), "Must be JavaThread");
5591 JavaThread *jt = (JavaThread *)thread;
5592
5593 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5594 // Check interrupt before trying to wait
5595 if (Thread::is_interrupted(thread, false)) {
5596 return;
5597 }
5598
5599 // Next, demultiplex/decode time arguments
5600 struct timespec absTime;
5601 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5602 return;
5603 }
5604 if (time > 0) {
5605 unpackTime(&absTime, isAbsolute, time);
5606 }
5607
5608
5609 // Enter safepoint region
5610 // Beware of deadlocks such as 6317397.
5611 // The per-thread Parker:: mutex is a classic leaf-lock.
5612 // In particular a thread must never block on the Threads_lock while
5613 // holding the Parker:: mutex. If safepoints are pending both the
5614 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5615 ThreadBlockInVM tbivm(jt);
5616
5617 // Don't wait if cannot get lock since interference arises from
5618 // unblocking. Also. check interrupt before trying wait
5619 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5620 return;
5621 }
5622
5623 int status ;
5624 if (_counter > 0) { // no wait needed
5625 _counter = 0;
5626 status = pthread_mutex_unlock(_mutex);
5627 assert (status == 0, "invariant") ;
5628 OrderAccess::fence();
5629 return;
5630 }
5631
5632 #ifdef ASSERT
5633 // Don't catch signals while blocked; let the running threads have the signals.
5634 // (This allows a debugger to break into the running thread.)
5635 sigset_t oldsigs;
5636 sigset_t* allowdebug_blocked = os::Bsd::allowdebug_blocked_signals();
5637 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5638 #endif
5639
5640 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5641 jt->set_suspend_equivalent();
5642 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5643
5644 if (time == 0) {
5645 status = pthread_cond_wait (_cond, _mutex) ;
5646 } else {
5647 status = os::Bsd::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5648 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5649 pthread_cond_destroy (_cond) ;
5650 pthread_cond_init (_cond, NULL);
5651 }
5652 }
5653 assert_status(status == 0 || status == EINTR ||
5654 status == ETIMEDOUT,
5655 status, "cond_timedwait");
5656
5657 #ifdef ASSERT
5658 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5659 #endif
5660
5661 _counter = 0 ;
5662 status = pthread_mutex_unlock(_mutex) ;
5663 assert_status(status == 0, status, "invariant") ;
5664 // If externally suspended while waiting, re-suspend
5665 if (jt->handle_special_suspend_equivalent_condition()) {
5666 jt->java_suspend_self();
5667 }
5668
5669 OrderAccess::fence();
5670 }
5671
5672 void Parker::unpark() {
5673 int s, status ;
5674 status = pthread_mutex_lock(_mutex);
5675 assert (status == 0, "invariant") ;
5676 s = _counter;
5677 _counter = 1;
5678 if (s < 1) {
5679 if (WorkAroundNPTLTimedWaitHang) {
5680 status = pthread_cond_signal (_cond) ;
5681 assert (status == 0, "invariant") ;
5682 status = pthread_mutex_unlock(_mutex);
5683 assert (status == 0, "invariant") ;
5684 } else {
5685 status = pthread_mutex_unlock(_mutex);
5686 assert (status == 0, "invariant") ;
5687 status = pthread_cond_signal (_cond) ;
5688 assert (status == 0, "invariant") ;
5689 }
5690 } else {
5691 pthread_mutex_unlock(_mutex);
5692 assert (status == 0, "invariant") ;
5693 }
5694 }
5695
5696
5697 /* Darwin has no "environ" in a dynamic library. */
5698 #ifdef __APPLE__
5699 #include <crt_externs.h>
5700 #define environ (*_NSGetEnviron())
5701 #else
5702 extern char** environ;
5703 #endif
5704
5705 // Run the specified command in a separate process. Return its exit value,
5706 // or -1 on failure (e.g. can't fork a new process).
5707 // Unlike system(), this function can be called from signal handler. It
5708 // doesn't block SIGINT et al.
5709 int os::fork_and_exec(char* cmd) {
5710 const char * argv[4] = {"sh", "-c", cmd, NULL};
5711
5712 // fork() in BsdThreads/NPTL is not async-safe. It needs to run
5713 // pthread_atfork handlers and reset pthread library. All we need is a
5714 // separate process to execve. Make a direct syscall to fork process.
5715 // On IA64 there's no fork syscall, we have to use fork() and hope for
5716 // the best...
5717 pid_t pid = fork();
5718
5719 if (pid < 0) {
5720 // fork failed
5721 return -1;
5722
5723 } else if (pid == 0) {
5724 // child process
5725
5726 // execve() in BsdThreads will call pthread_kill_other_threads_np()
5727 // first to kill every thread on the thread list. Because this list is
5728 // not reset by fork() (see notes above), execve() will instead kill
5729 // every thread in the parent process. We know this is the only thread
5730 // in the new process, so make a system call directly.
5731 // IA64 should use normal execve() from glibc to match the glibc fork()
5732 // above.
5733 execve("/bin/sh", (char* const*)argv, environ);
5734
5735 // execve failed
5736 _exit(-1);
5737
5738 } else {
5739 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5740 // care about the actual exit code, for now.
5741
5742 int status;
5743
5744 // Wait for the child process to exit. This returns immediately if
5745 // the child has already exited. */
5746 while (waitpid(pid, &status, 0) < 0) {
5747 switch (errno) {
5748 case ECHILD: return 0;
5749 case EINTR: break;
5750 default: return -1;
5751 }
5752 }
5753
5754 if (WIFEXITED(status)) {
5755 // The child exited normally; get its exit code.
5756 return WEXITSTATUS(status);
5757 } else if (WIFSIGNALED(status)) {
5758 // The child exited because of a signal
5759 // The best value to return is 0x80 + signal number,
5760 // because that is what all Unix shells do, and because
5761 // it allows callers to distinguish between process exit and
5762 // process death by signal.
5763 return 0x80 + WTERMSIG(status);
5764 } else {
5765 // Unknown exit code; pass it through
5766 return status;
5767 }
5768 }
5769 }
5770
5771 // is_headless_jre()
5772 //
5773 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5774 // in order to report if we are running in a headless jre
5775 //
5776 // Since JDK8 xawt/libmawt.so was moved into the same directory
5777 // as libawt.so, and renamed libawt_xawt.so
5778 //
5779 bool os::is_headless_jre() {
5780 struct stat statbuf;
5781 char buf[MAXPATHLEN];
5782 char libmawtpath[MAXPATHLEN];
5783 const char *xawtstr = "/xawt/libmawt" JNI_LIB_SUFFIX;
5784 const char *new_xawtstr = "/libawt_xawt" JNI_LIB_SUFFIX;
5785 char *p;
5786
5787 // Get path to libjvm.so
5788 os::jvm_path(buf, sizeof(buf));
5789
5790 // Get rid of libjvm.so
5791 p = strrchr(buf, '/');
5792 if (p == NULL) return false;
5793 else *p = '\0';
5794
5795 // Get rid of client or server
5796 p = strrchr(buf, '/');
5797 if (p == NULL) return false;
5798 else *p = '\0';
5799
5800 // check xawt/libmawt.so
5801 strcpy(libmawtpath, buf);
5802 strcat(libmawtpath, xawtstr);
5803 if (::stat(libmawtpath, &statbuf) == 0) return false;
5804
5805 // check libawt_xawt.so
5806 strcpy(libmawtpath, buf);
5807 strcat(libmawtpath, new_xawtstr);
5808 if (::stat(libmawtpath, &statbuf) == 0) return false;
5809
5810 return true;
5811 }
5812
5813 // Get the default path to the core file
5814 // Returns the length of the string
5815 int os::get_core_path(char* buffer, size_t bufferSize) {
5816 int n = jio_snprintf(buffer, bufferSize, "/cores");
5817
5818 // Truncate if theoretical string was longer than bufferSize
5819 n = MIN2(n, (int)bufferSize);
5820
5821 return n;
5822 }