1 /*
2 * Copyright (c) 1999, 2010, 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 # define __STDC_FORMAT_MACROS
26
27 // no precompiled headers
28 #include "classfile/classLoader.hpp"
29 #include "classfile/systemDictionary.hpp"
30 #include "classfile/vmSymbols.hpp"
31 #include "code/icBuffer.hpp"
32 #include "code/vtableStubs.hpp"
33 #include "compiler/compileBroker.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "jvm_linux.h"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "mutex_linux.inline.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/extendedPC.hpp"
46 #include "runtime/globals.hpp"
47 #include "runtime/hpi.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/java.hpp"
50 #include "runtime/javaCalls.hpp"
51 #include "runtime/mutexLocker.hpp"
52 #include "runtime/objectMonitor.hpp"
53 #include "runtime/osThread.hpp"
54 #include "runtime/perfMemory.hpp"
55 #include "runtime/sharedRuntime.hpp"
56 #include "runtime/statSampler.hpp"
57 #include "runtime/stubRoutines.hpp"
58 #include "runtime/threadCritical.hpp"
59 #include "runtime/timer.hpp"
60 #include "services/attachListener.hpp"
61 #include "services/runtimeService.hpp"
62 #include "thread_linux.inline.hpp"
63 #include "utilities/defaultStream.hpp"
64 #include "utilities/events.hpp"
65 #include "utilities/growableArray.hpp"
66 #include "utilities/vmError.hpp"
67 #ifdef TARGET_ARCH_x86
68 # include "assembler_x86.inline.hpp"
69 # include "nativeInst_x86.hpp"
70 #endif
71 #ifdef TARGET_ARCH_sparc
72 # include "assembler_sparc.inline.hpp"
73 # include "nativeInst_sparc.hpp"
74 #endif
75 #ifdef TARGET_ARCH_zero
76 # include "assembler_zero.inline.hpp"
77 # include "nativeInst_zero.hpp"
78 #endif
79 #ifdef COMPILER1
80 #include "c1/c1_Runtime1.hpp"
81 #endif
82 #ifdef COMPILER2
83 #include "opto/runtime.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 <pwd.h>
106 # include <poll.h>
107 # include <semaphore.h>
108 # include <fcntl.h>
109 # include <string.h>
110 # include <syscall.h>
111 # include <sys/sysinfo.h>
112 # include <gnu/libc-version.h>
113 # include <sys/ipc.h>
114 # include <sys/shm.h>
115 # include <link.h>
116 # include <stdint.h>
117 # include <inttypes.h>
118
119 #define MAX_PATH (2 * K)
120
121 // for timer info max values which include all bits
122 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
123 #define SEC_IN_NANOSECS 1000000000LL
124
125 ////////////////////////////////////////////////////////////////////////////////
126 // global variables
127 julong os::Linux::_physical_memory = 0;
128
129 address os::Linux::_initial_thread_stack_bottom = NULL;
130 uintptr_t os::Linux::_initial_thread_stack_size = 0;
131
132 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
133 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
134 Mutex* os::Linux::_createThread_lock = NULL;
135 pthread_t os::Linux::_main_thread;
136 int os::Linux::_page_size = -1;
137 bool os::Linux::_is_floating_stack = false;
138 bool os::Linux::_is_NPTL = false;
139 bool os::Linux::_supports_fast_thread_cpu_time = false;
140 const char * os::Linux::_glibc_version = NULL;
141 const char * os::Linux::_libpthread_version = NULL;
142
143 static jlong initial_time_count=0;
144
145 static int clock_tics_per_sec = 100;
146
147 // For diagnostics to print a message once. see run_periodic_checks
148 static sigset_t check_signal_done;
149 static bool check_signals = true;;
150
151 static pid_t _initial_pid = 0;
152
153 /* Signal number used to suspend/resume a thread */
154
155 /* do not use any signal number less than SIGSEGV, see 4355769 */
156 static int SR_signum = SIGUSR2;
157 sigset_t SR_sigset;
158
159 /* Used to protect dlsym() calls */
160 static pthread_mutex_t dl_mutex;
161
162 ////////////////////////////////////////////////////////////////////////////////
163 // utility functions
164
165 static int SR_initialize();
166 static int SR_finalize();
167
168 julong os::available_memory() {
169 return Linux::available_memory();
170 }
171
172 julong os::Linux::available_memory() {
173 // values in struct sysinfo are "unsigned long"
174 struct sysinfo si;
175 sysinfo(&si);
176
177 return (julong)si.freeram * si.mem_unit;
178 }
179
180 julong os::physical_memory() {
181 return Linux::physical_memory();
182 }
183
184 julong os::allocatable_physical_memory(julong size) {
185 #ifdef _LP64
186 return size;
187 #else
188 julong result = MIN2(size, (julong)3800*M);
189 if (!is_allocatable(result)) {
190 // See comments under solaris for alignment considerations
191 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
192 result = MIN2(size, reasonable_size);
193 }
194 return result;
195 #endif // _LP64
196 }
197
198 ////////////////////////////////////////////////////////////////////////////////
199 // environment support
200
201 bool os::getenv(const char* name, char* buf, int len) {
202 const char* val = ::getenv(name);
203 if (val != NULL && strlen(val) < (size_t)len) {
204 strcpy(buf, val);
205 return true;
206 }
207 if (len > 0) buf[0] = 0; // return a null string
208 return false;
209 }
210
211
212 // Return true if user is running as root.
213
214 bool os::have_special_privileges() {
215 static bool init = false;
216 static bool privileges = false;
217 if (!init) {
218 privileges = (getuid() != geteuid()) || (getgid() != getegid());
219 init = true;
220 }
221 return privileges;
222 }
223
224
225 #ifndef SYS_gettid
226 // i386: 224, ia64: 1105, amd64: 186, sparc 143
227 #ifdef __ia64__
228 #define SYS_gettid 1105
229 #elif __i386__
230 #define SYS_gettid 224
231 #elif __amd64__
232 #define SYS_gettid 186
233 #elif __sparc__
234 #define SYS_gettid 143
235 #else
236 #error define gettid for the arch
237 #endif
238 #endif
239
240 // Cpu architecture string
241 #if defined(ZERO)
242 static char cpu_arch[] = ZERO_LIBARCH;
243 #elif defined(IA64)
244 static char cpu_arch[] = "ia64";
245 #elif defined(IA32)
246 static char cpu_arch[] = "i386";
247 #elif defined(AMD64)
248 static char cpu_arch[] = "amd64";
249 #elif defined(ARM)
250 static char cpu_arch[] = "arm";
251 #elif defined(PPC)
252 static char cpu_arch[] = "ppc";
253 #elif defined(SPARC)
254 # ifdef _LP64
255 static char cpu_arch[] = "sparcv9";
256 # else
257 static char cpu_arch[] = "sparc";
258 # endif
259 #else
260 #error Add appropriate cpu_arch setting
261 #endif
262
263
264 // pid_t gettid()
265 //
266 // Returns the kernel thread id of the currently running thread. Kernel
267 // thread id is used to access /proc.
268 //
269 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
270 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
271 //
272 pid_t os::Linux::gettid() {
273 int rslt = syscall(SYS_gettid);
274 if (rslt == -1) {
275 // old kernel, no NPTL support
276 return getpid();
277 } else {
278 return (pid_t)rslt;
279 }
280 }
281
282 // Most versions of linux have a bug where the number of processors are
283 // determined by looking at the /proc file system. In a chroot environment,
284 // the system call returns 1. This causes the VM to act as if it is
285 // a single processor and elide locking (see is_MP() call).
286 static bool unsafe_chroot_detected = false;
287 static const char *unstable_chroot_error = "/proc file system not found.\n"
288 "Java may be unstable running multithreaded in a chroot "
289 "environment on Linux when /proc filesystem is not mounted.";
290
291 void os::Linux::initialize_system_info() {
292 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
293 if (processor_count() == 1) {
294 pid_t pid = os::Linux::gettid();
295 char fname[32];
296 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
297 FILE *fp = fopen(fname, "r");
298 if (fp == NULL) {
299 unsafe_chroot_detected = true;
300 } else {
301 fclose(fp);
302 }
303 }
304 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
305 assert(processor_count() > 0, "linux error");
306 }
307
308 void os::init_system_properties_values() {
309 // char arch[12];
310 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
311
312 // The next steps are taken in the product version:
313 //
314 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
315 // This library should be located at:
316 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
317 //
318 // If "/jre/lib/" appears at the right place in the path, then we
319 // assume libjvm[_g].so is installed in a JDK and we use this path.
320 //
321 // Otherwise exit with message: "Could not create the Java virtual machine."
322 //
323 // The following extra steps are taken in the debugging version:
324 //
325 // If "/jre/lib/" does NOT appear at the right place in the path
326 // instead of exit check for $JAVA_HOME environment variable.
327 //
328 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
329 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
330 // it looks like libjvm[_g].so is installed there
331 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
332 //
333 // Otherwise exit.
334 //
335 // Important note: if the location of libjvm.so changes this
336 // code needs to be changed accordingly.
337
338 // The next few definitions allow the code to be verbatim:
339 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
340 #define getenv(n) ::getenv(n)
341
342 /*
343 * See ld(1):
344 * The linker uses the following search paths to locate required
345 * shared libraries:
346 * 1: ...
347 * ...
348 * 7: The default directories, normally /lib and /usr/lib.
349 */
350 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
351 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
352 #else
353 #define DEFAULT_LIBPATH "/lib:/usr/lib"
354 #endif
355
356 #define EXTENSIONS_DIR "/lib/ext"
357 #define ENDORSED_DIR "/lib/endorsed"
358 #define REG_DIR "/usr/java/packages"
359
360 {
361 /* sysclasspath, java_home, dll_dir */
362 {
363 char *home_path;
364 char *dll_path;
365 char *pslash;
366 char buf[MAXPATHLEN];
367 os::jvm_path(buf, sizeof(buf));
368
369 // Found the full path to libjvm.so.
370 // Now cut the path to <java_home>/jre if we can.
371 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
372 pslash = strrchr(buf, '/');
373 if (pslash != NULL)
374 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
375 dll_path = malloc(strlen(buf) + 1);
376 if (dll_path == NULL)
377 return;
378 strcpy(dll_path, buf);
379 Arguments::set_dll_dir(dll_path);
380
381 if (pslash != NULL) {
382 pslash = strrchr(buf, '/');
383 if (pslash != NULL) {
384 *pslash = '\0'; /* get rid of /<arch> */
385 pslash = strrchr(buf, '/');
386 if (pslash != NULL)
387 *pslash = '\0'; /* get rid of /lib */
388 }
389 }
390
391 home_path = malloc(strlen(buf) + 1);
392 if (home_path == NULL)
393 return;
394 strcpy(home_path, buf);
395 Arguments::set_java_home(home_path);
396
397 if (!set_boot_path('/', ':'))
398 return;
399 }
400
401 /*
402 * Where to look for native libraries
403 *
404 * Note: Due to a legacy implementation, most of the library path
405 * is set in the launcher. This was to accomodate linking restrictions
406 * on legacy Linux implementations (which are no longer supported).
407 * Eventually, all the library path setting will be done here.
408 *
409 * However, to prevent the proliferation of improperly built native
410 * libraries, the new path component /usr/java/packages is added here.
411 * Eventually, all the library path setting will be done here.
412 */
413 {
414 char *ld_library_path;
415
416 /*
417 * Construct the invariant part of ld_library_path. Note that the
418 * space for the colon and the trailing null are provided by the
419 * nulls included by the sizeof operator (so actually we allocate
420 * a byte more than necessary).
421 */
422 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
423 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
424 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
425
426 /*
427 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
428 * should always exist (until the legacy problem cited above is
429 * addressed).
430 */
431 char *v = getenv("LD_LIBRARY_PATH");
432 if (v != NULL) {
433 char *t = ld_library_path;
434 /* That's +1 for the colon and +1 for the trailing '\0' */
435 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
436 sprintf(ld_library_path, "%s:%s", v, t);
437 }
438 Arguments::set_library_path(ld_library_path);
439 }
440
441 /*
442 * Extensions directories.
443 *
444 * Note that the space for the colon and the trailing null are provided
445 * by the nulls included by the sizeof operator (so actually one byte more
446 * than necessary is allocated).
447 */
448 {
449 char *buf = malloc(strlen(Arguments::get_java_home()) +
450 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
451 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
452 Arguments::get_java_home());
453 Arguments::set_ext_dirs(buf);
454 }
455
456 /* Endorsed standards default directory. */
457 {
458 char * buf;
459 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
460 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
461 Arguments::set_endorsed_dirs(buf);
462 }
463 }
464
465 #undef malloc
466 #undef getenv
467 #undef EXTENSIONS_DIR
468 #undef ENDORSED_DIR
469
470 // Done
471 return;
472 }
473
474 ////////////////////////////////////////////////////////////////////////////////
475 // breakpoint support
476
477 void os::breakpoint() {
478 BREAKPOINT;
479 }
480
481 extern "C" void breakpoint() {
482 // use debugger to set breakpoint here
483 }
484
485 ////////////////////////////////////////////////////////////////////////////////
486 // signal support
487
488 debug_only(static bool signal_sets_initialized = false);
489 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
490
491 bool os::Linux::is_sig_ignored(int sig) {
492 struct sigaction oact;
493 sigaction(sig, (struct sigaction*)NULL, &oact);
494 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
495 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
496 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
497 return true;
498 else
499 return false;
500 }
501
502 void os::Linux::signal_sets_init() {
503 // Should also have an assertion stating we are still single-threaded.
504 assert(!signal_sets_initialized, "Already initialized");
505 // Fill in signals that are necessarily unblocked for all threads in
506 // the VM. Currently, we unblock the following signals:
507 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
508 // by -Xrs (=ReduceSignalUsage));
509 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
510 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
511 // the dispositions or masks wrt these signals.
512 // Programs embedding the VM that want to use the above signals for their
513 // own purposes must, at this time, use the "-Xrs" option to prevent
514 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
515 // (See bug 4345157, and other related bugs).
516 // In reality, though, unblocking these signals is really a nop, since
517 // these signals are not blocked by default.
518 sigemptyset(&unblocked_sigs);
519 sigemptyset(&allowdebug_blocked_sigs);
520 sigaddset(&unblocked_sigs, SIGILL);
521 sigaddset(&unblocked_sigs, SIGSEGV);
522 sigaddset(&unblocked_sigs, SIGBUS);
523 sigaddset(&unblocked_sigs, SIGFPE);
524 sigaddset(&unblocked_sigs, SR_signum);
525
526 if (!ReduceSignalUsage) {
527 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
528 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
529 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
530 }
531 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
532 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
533 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
534 }
535 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
536 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
537 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
538 }
539 }
540 // Fill in signals that are blocked by all but the VM thread.
541 sigemptyset(&vm_sigs);
542 if (!ReduceSignalUsage)
543 sigaddset(&vm_sigs, BREAK_SIGNAL);
544 debug_only(signal_sets_initialized = true);
545
546 }
547
548 // These are signals that are unblocked while a thread is running Java.
549 // (For some reason, they get blocked by default.)
550 sigset_t* os::Linux::unblocked_signals() {
551 assert(signal_sets_initialized, "Not initialized");
552 return &unblocked_sigs;
553 }
554
555 // These are the signals that are blocked while a (non-VM) thread is
556 // running Java. Only the VM thread handles these signals.
557 sigset_t* os::Linux::vm_signals() {
558 assert(signal_sets_initialized, "Not initialized");
559 return &vm_sigs;
560 }
561
562 // These are signals that are blocked during cond_wait to allow debugger in
563 sigset_t* os::Linux::allowdebug_blocked_signals() {
564 assert(signal_sets_initialized, "Not initialized");
565 return &allowdebug_blocked_sigs;
566 }
567
568 void os::Linux::hotspot_sigmask(Thread* thread) {
569
570 //Save caller's signal mask before setting VM signal mask
571 sigset_t caller_sigmask;
572 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
573
574 OSThread* osthread = thread->osthread();
575 osthread->set_caller_sigmask(caller_sigmask);
576
577 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
578
579 if (!ReduceSignalUsage) {
580 if (thread->is_VM_thread()) {
581 // Only the VM thread handles BREAK_SIGNAL ...
582 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
583 } else {
584 // ... all other threads block BREAK_SIGNAL
585 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
586 }
587 }
588 }
589
590 //////////////////////////////////////////////////////////////////////////////
591 // detecting pthread library
592
593 void os::Linux::libpthread_init() {
594 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
595 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
596 // generic name for earlier versions.
597 // Define macros here so we can build HotSpot on old systems.
598 # ifndef _CS_GNU_LIBC_VERSION
599 # define _CS_GNU_LIBC_VERSION 2
600 # endif
601 # ifndef _CS_GNU_LIBPTHREAD_VERSION
602 # define _CS_GNU_LIBPTHREAD_VERSION 3
603 # endif
604
605 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
606 if (n > 0) {
607 char *str = (char *)malloc(n);
608 confstr(_CS_GNU_LIBC_VERSION, str, n);
609 os::Linux::set_glibc_version(str);
610 } else {
611 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
612 static char _gnu_libc_version[32];
613 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
614 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
615 os::Linux::set_glibc_version(_gnu_libc_version);
616 }
617
618 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
619 if (n > 0) {
620 char *str = (char *)malloc(n);
621 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
622 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
623 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
624 // is the case. LinuxThreads has a hard limit on max number of threads.
625 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
626 // On the other hand, NPTL does not have such a limit, sysconf()
627 // will return -1 and errno is not changed. Check if it is really NPTL.
628 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
629 strstr(str, "NPTL") &&
630 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
631 free(str);
632 os::Linux::set_libpthread_version("linuxthreads");
633 } else {
634 os::Linux::set_libpthread_version(str);
635 }
636 } else {
637 // glibc before 2.3.2 only has LinuxThreads.
638 os::Linux::set_libpthread_version("linuxthreads");
639 }
640
641 if (strstr(libpthread_version(), "NPTL")) {
642 os::Linux::set_is_NPTL();
643 } else {
644 os::Linux::set_is_LinuxThreads();
645 }
646
647 // LinuxThreads have two flavors: floating-stack mode, which allows variable
648 // stack size; and fixed-stack mode. NPTL is always floating-stack.
649 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
650 os::Linux::set_is_floating_stack();
651 }
652 }
653
654 /////////////////////////////////////////////////////////////////////////////
655 // thread stack
656
657 // Force Linux kernel to expand current thread stack. If "bottom" is close
658 // to the stack guard, caller should block all signals.
659 //
660 // MAP_GROWSDOWN:
661 // A special mmap() flag that is used to implement thread stacks. It tells
662 // kernel that the memory region should extend downwards when needed. This
663 // allows early versions of LinuxThreads to only mmap the first few pages
664 // when creating a new thread. Linux kernel will automatically expand thread
665 // stack as needed (on page faults).
666 //
667 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
668 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
669 // region, it's hard to tell if the fault is due to a legitimate stack
670 // access or because of reading/writing non-exist memory (e.g. buffer
671 // overrun). As a rule, if the fault happens below current stack pointer,
672 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
673 // application (see Linux kernel fault.c).
674 //
675 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
676 // stack overflow detection.
677 //
678 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
679 // not use this flag. However, the stack of initial thread is not created
680 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
681 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
682 // and then attach the thread to JVM.
683 //
684 // To get around the problem and allow stack banging on Linux, we need to
685 // manually expand thread stack after receiving the SIGSEGV.
686 //
687 // There are two ways to expand thread stack to address "bottom", we used
688 // both of them in JVM before 1.5:
689 // 1. adjust stack pointer first so that it is below "bottom", and then
690 // touch "bottom"
691 // 2. mmap() the page in question
692 //
693 // Now alternate signal stack is gone, it's harder to use 2. For instance,
694 // if current sp is already near the lower end of page 101, and we need to
695 // call mmap() to map page 100, it is possible that part of the mmap() frame
696 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
697 // That will destroy the mmap() frame and cause VM to crash.
698 //
699 // The following code works by adjusting sp first, then accessing the "bottom"
700 // page to force a page fault. Linux kernel will then automatically expand the
701 // stack mapping.
702 //
703 // _expand_stack_to() assumes its frame size is less than page size, which
704 // should always be true if the function is not inlined.
705
706 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
707 #define NOINLINE
708 #else
709 #define NOINLINE __attribute__ ((noinline))
710 #endif
711
712 static void _expand_stack_to(address bottom) NOINLINE;
713
714 static void _expand_stack_to(address bottom) {
715 address sp;
716 size_t size;
717 volatile char *p;
718
719 // Adjust bottom to point to the largest address within the same page, it
720 // gives us a one-page buffer if alloca() allocates slightly more memory.
721 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
722 bottom += os::Linux::page_size() - 1;
723
724 // sp might be slightly above current stack pointer; if that's the case, we
725 // will alloca() a little more space than necessary, which is OK. Don't use
726 // os::current_stack_pointer(), as its result can be slightly below current
727 // stack pointer, causing us to not alloca enough to reach "bottom".
728 sp = (address)&sp;
729
730 if (sp > bottom) {
731 size = sp - bottom;
732 p = (volatile char *)alloca(size);
733 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
734 p[0] = '\0';
735 }
736 }
737
738 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
739 assert(t!=NULL, "just checking");
740 assert(t->osthread()->expanding_stack(), "expand should be set");
741 assert(t->stack_base() != NULL, "stack_base was not initialized");
742
743 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
744 sigset_t mask_all, old_sigset;
745 sigfillset(&mask_all);
746 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
747 _expand_stack_to(addr);
748 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
749 return true;
750 }
751 return false;
752 }
753
754 //////////////////////////////////////////////////////////////////////////////
755 // create new thread
756
757 static address highest_vm_reserved_address();
758
759 // check if it's safe to start a new thread
760 static bool _thread_safety_check(Thread* thread) {
761 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
762 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
763 // Heap is mmap'ed at lower end of memory space. Thread stacks are
764 // allocated (MAP_FIXED) from high address space. Every thread stack
765 // occupies a fixed size slot (usually 2Mbytes, but user can change
766 // it to other values if they rebuild LinuxThreads).
767 //
768 // Problem with MAP_FIXED is that mmap() can still succeed even part of
769 // the memory region has already been mmap'ed. That means if we have too
770 // many threads and/or very large heap, eventually thread stack will
771 // collide with heap.
772 //
773 // Here we try to prevent heap/stack collision by comparing current
774 // stack bottom with the highest address that has been mmap'ed by JVM
775 // plus a safety margin for memory maps created by native code.
776 //
777 // This feature can be disabled by setting ThreadSafetyMargin to 0
778 //
779 if (ThreadSafetyMargin > 0) {
780 address stack_bottom = os::current_stack_base() - os::current_stack_size();
781
782 // not safe if our stack extends below the safety margin
783 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
784 } else {
785 return true;
786 }
787 } else {
788 // Floating stack LinuxThreads or NPTL:
789 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
790 // there's not enough space left, pthread_create() will fail. If we come
791 // here, that means enough space has been reserved for stack.
792 return true;
793 }
794 }
795
796 // Thread start routine for all newly created threads
797 static void *java_start(Thread *thread) {
798 // Try to randomize the cache line index of hot stack frames.
799 // This helps when threads of the same stack traces evict each other's
800 // cache lines. The threads can be either from the same JVM instance, or
801 // from different JVM instances. The benefit is especially true for
802 // processors with hyperthreading technology.
803 static int counter = 0;
804 int pid = os::current_process_id();
805 alloca(((pid ^ counter++) & 7) * 128);
806
807 ThreadLocalStorage::set_thread(thread);
808
809 OSThread* osthread = thread->osthread();
810 Monitor* sync = osthread->startThread_lock();
811
812 // non floating stack LinuxThreads needs extra check, see above
813 if (!_thread_safety_check(thread)) {
814 // notify parent thread
815 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
816 osthread->set_state(ZOMBIE);
817 sync->notify_all();
818 return NULL;
819 }
820
821 // thread_id is kernel thread id (similar to Solaris LWP id)
822 osthread->set_thread_id(os::Linux::gettid());
823
824 if (UseNUMA) {
825 int lgrp_id = os::numa_get_group_id();
826 if (lgrp_id != -1) {
827 thread->set_lgrp_id(lgrp_id);
828 }
829 }
830 // initialize signal mask for this thread
831 os::Linux::hotspot_sigmask(thread);
832
833 // initialize floating point control register
834 os::Linux::init_thread_fpu_state();
835
836 // handshaking with parent thread
837 {
838 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
839
840 // notify parent thread
841 osthread->set_state(INITIALIZED);
842 sync->notify_all();
843
844 // wait until os::start_thread()
845 while (osthread->get_state() == INITIALIZED) {
846 sync->wait(Mutex::_no_safepoint_check_flag);
847 }
848 }
849
850 // call one more level start routine
851 thread->run();
852
853 return 0;
854 }
855
856 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
857 assert(thread->osthread() == NULL, "caller responsible");
858
859 // Allocate the OSThread object
860 OSThread* osthread = new OSThread(NULL, NULL);
861 if (osthread == NULL) {
862 return false;
863 }
864
865 // set the correct thread state
866 osthread->set_thread_type(thr_type);
867
868 // Initial state is ALLOCATED but not INITIALIZED
869 osthread->set_state(ALLOCATED);
870
871 thread->set_osthread(osthread);
872
873 // init thread attributes
874 pthread_attr_t attr;
875 pthread_attr_init(&attr);
876 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
877
878 // stack size
879 if (os::Linux::supports_variable_stack_size()) {
880 // calculate stack size if it's not specified by caller
881 if (stack_size == 0) {
882 stack_size = os::Linux::default_stack_size(thr_type);
883
884 switch (thr_type) {
885 case os::java_thread:
886 // Java threads use ThreadStackSize which default value can be
887 // changed with the flag -Xss
888 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
889 stack_size = JavaThread::stack_size_at_create();
890 break;
891 case os::compiler_thread:
892 if (CompilerThreadStackSize > 0) {
893 stack_size = (size_t)(CompilerThreadStackSize * K);
894 break;
895 } // else fall through:
896 // use VMThreadStackSize if CompilerThreadStackSize is not defined
897 case os::vm_thread:
898 case os::pgc_thread:
899 case os::cgc_thread:
900 case os::watcher_thread:
901 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
902 break;
903 }
904 }
905
906 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
907 pthread_attr_setstacksize(&attr, stack_size);
908 } else {
909 // let pthread_create() pick the default value.
910 }
911
912 // glibc guard page
913 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
914
915 ThreadState state;
916
917 {
918 // Serialize thread creation if we are running with fixed stack LinuxThreads
919 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
920 if (lock) {
921 os::Linux::createThread_lock()->lock_without_safepoint_check();
922 }
923
924 pthread_t tid;
925 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
926
927 pthread_attr_destroy(&attr);
928
929 if (ret != 0) {
930 if (PrintMiscellaneous && (Verbose || WizardMode)) {
931 perror("pthread_create()");
932 }
933 // Need to clean up stuff we've allocated so far
934 thread->set_osthread(NULL);
935 delete osthread;
936 if (lock) os::Linux::createThread_lock()->unlock();
937 return false;
938 }
939
940 // Store pthread info into the OSThread
941 osthread->set_pthread_id(tid);
942
943 // Wait until child thread is either initialized or aborted
944 {
945 Monitor* sync_with_child = osthread->startThread_lock();
946 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
947 while ((state = osthread->get_state()) == ALLOCATED) {
948 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
949 }
950 }
951
952 if (lock) {
953 os::Linux::createThread_lock()->unlock();
954 }
955 }
956
957 // Aborted due to thread limit being reached
958 if (state == ZOMBIE) {
959 thread->set_osthread(NULL);
960 delete osthread;
961 return false;
962 }
963
964 // The thread is returned suspended (in state INITIALIZED),
965 // and is started higher up in the call chain
966 assert(state == INITIALIZED, "race condition");
967 return true;
968 }
969
970 /////////////////////////////////////////////////////////////////////////////
971 // attach existing thread
972
973 // bootstrap the main thread
974 bool os::create_main_thread(JavaThread* thread) {
975 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
976 return create_attached_thread(thread);
977 }
978
979 bool os::create_attached_thread(JavaThread* thread) {
980 #ifdef ASSERT
981 thread->verify_not_published();
982 #endif
983
984 // Allocate the OSThread object
985 OSThread* osthread = new OSThread(NULL, NULL);
986
987 if (osthread == NULL) {
988 return false;
989 }
990
991 // Store pthread info into the OSThread
992 osthread->set_thread_id(os::Linux::gettid());
993 osthread->set_pthread_id(::pthread_self());
994
995 // initialize floating point control register
996 os::Linux::init_thread_fpu_state();
997
998 // Initial thread state is RUNNABLE
999 osthread->set_state(RUNNABLE);
1000
1001 thread->set_osthread(osthread);
1002
1003 if (UseNUMA) {
1004 int lgrp_id = os::numa_get_group_id();
1005 if (lgrp_id != -1) {
1006 thread->set_lgrp_id(lgrp_id);
1007 }
1008 }
1009
1010 if (os::Linux::is_initial_thread()) {
1011 // If current thread is initial thread, its stack is mapped on demand,
1012 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1013 // the entire stack region to avoid SEGV in stack banging.
1014 // It is also useful to get around the heap-stack-gap problem on SuSE
1015 // kernel (see 4821821 for details). We first expand stack to the top
1016 // of yellow zone, then enable stack yellow zone (order is significant,
1017 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1018 // is no gap between the last two virtual memory regions.
1019
1020 JavaThread *jt = (JavaThread *)thread;
1021 address addr = jt->stack_yellow_zone_base();
1022 assert(addr != NULL, "initialization problem?");
1023 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1024
1025 osthread->set_expanding_stack();
1026 os::Linux::manually_expand_stack(jt, addr);
1027 osthread->clear_expanding_stack();
1028 }
1029
1030 // initialize signal mask for this thread
1031 // and save the caller's signal mask
1032 os::Linux::hotspot_sigmask(thread);
1033
1034 return true;
1035 }
1036
1037 void os::pd_start_thread(Thread* thread) {
1038 OSThread * osthread = thread->osthread();
1039 assert(osthread->get_state() != INITIALIZED, "just checking");
1040 Monitor* sync_with_child = osthread->startThread_lock();
1041 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1042 sync_with_child->notify();
1043 }
1044
1045 // Free Linux resources related to the OSThread
1046 void os::free_thread(OSThread* osthread) {
1047 assert(osthread != NULL, "osthread not set");
1048
1049 if (Thread::current()->osthread() == osthread) {
1050 // Restore caller's signal mask
1051 sigset_t sigmask = osthread->caller_sigmask();
1052 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1053 }
1054
1055 delete osthread;
1056 }
1057
1058 //////////////////////////////////////////////////////////////////////////////
1059 // thread local storage
1060
1061 int os::allocate_thread_local_storage() {
1062 pthread_key_t key;
1063 int rslt = pthread_key_create(&key, NULL);
1064 assert(rslt == 0, "cannot allocate thread local storage");
1065 return (int)key;
1066 }
1067
1068 // Note: This is currently not used by VM, as we don't destroy TLS key
1069 // on VM exit.
1070 void os::free_thread_local_storage(int index) {
1071 int rslt = pthread_key_delete((pthread_key_t)index);
1072 assert(rslt == 0, "invalid index");
1073 }
1074
1075 void os::thread_local_storage_at_put(int index, void* value) {
1076 int rslt = pthread_setspecific((pthread_key_t)index, value);
1077 assert(rslt == 0, "pthread_setspecific failed");
1078 }
1079
1080 extern "C" Thread* get_thread() {
1081 return ThreadLocalStorage::thread();
1082 }
1083
1084 //////////////////////////////////////////////////////////////////////////////
1085 // initial thread
1086
1087 // Check if current thread is the initial thread, similar to Solaris thr_main.
1088 bool os::Linux::is_initial_thread(void) {
1089 char dummy;
1090 // If called before init complete, thread stack bottom will be null.
1091 // Can be called if fatal error occurs before initialization.
1092 if (initial_thread_stack_bottom() == NULL) return false;
1093 assert(initial_thread_stack_bottom() != NULL &&
1094 initial_thread_stack_size() != 0,
1095 "os::init did not locate initial thread's stack region");
1096 if ((address)&dummy >= initial_thread_stack_bottom() &&
1097 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1098 return true;
1099 else return false;
1100 }
1101
1102 // Find the virtual memory area that contains addr
1103 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1104 FILE *fp = fopen("/proc/self/maps", "r");
1105 if (fp) {
1106 address low, high;
1107 while (!feof(fp)) {
1108 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1109 if (low <= addr && addr < high) {
1110 if (vma_low) *vma_low = low;
1111 if (vma_high) *vma_high = high;
1112 fclose (fp);
1113 return true;
1114 }
1115 }
1116 for (;;) {
1117 int ch = fgetc(fp);
1118 if (ch == EOF || ch == (int)'\n') break;
1119 }
1120 }
1121 fclose(fp);
1122 }
1123 return false;
1124 }
1125
1126 // Locate initial thread stack. This special handling of initial thread stack
1127 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1128 // bogus value for initial thread.
1129 void os::Linux::capture_initial_stack(size_t max_size) {
1130 // stack size is the easy part, get it from RLIMIT_STACK
1131 size_t stack_size;
1132 struct rlimit rlim;
1133 getrlimit(RLIMIT_STACK, &rlim);
1134 stack_size = rlim.rlim_cur;
1135
1136 // 6308388: a bug in ld.so will relocate its own .data section to the
1137 // lower end of primordial stack; reduce ulimit -s value a little bit
1138 // so we won't install guard page on ld.so's data section.
1139 stack_size -= 2 * page_size();
1140
1141 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1142 // 7.1, in both cases we will get 2G in return value.
1143 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1144 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1145 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1146 // in case other parts in glibc still assumes 2M max stack size.
1147 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1148 #ifndef IA64
1149 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1150 #else
1151 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1152 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1153 #endif
1154
1155 // Try to figure out where the stack base (top) is. This is harder.
1156 //
1157 // When an application is started, glibc saves the initial stack pointer in
1158 // a global variable "__libc_stack_end", which is then used by system
1159 // libraries. __libc_stack_end should be pretty close to stack top. The
1160 // variable is available since the very early days. However, because it is
1161 // a private interface, it could disappear in the future.
1162 //
1163 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1164 // to __libc_stack_end, it is very close to stack top, but isn't the real
1165 // stack top. Note that /proc may not exist if VM is running as a chroot
1166 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1167 // /proc/<pid>/stat could change in the future (though unlikely).
1168 //
1169 // We try __libc_stack_end first. If that doesn't work, look for
1170 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1171 // as a hint, which should work well in most cases.
1172
1173 uintptr_t stack_start;
1174
1175 // try __libc_stack_end first
1176 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1177 if (p && *p) {
1178 stack_start = *p;
1179 } else {
1180 // see if we can get the start_stack field from /proc/self/stat
1181 FILE *fp;
1182 int pid;
1183 char state;
1184 int ppid;
1185 int pgrp;
1186 int session;
1187 int nr;
1188 int tpgrp;
1189 unsigned long flags;
1190 unsigned long minflt;
1191 unsigned long cminflt;
1192 unsigned long majflt;
1193 unsigned long cmajflt;
1194 unsigned long utime;
1195 unsigned long stime;
1196 long cutime;
1197 long cstime;
1198 long prio;
1199 long nice;
1200 long junk;
1201 long it_real;
1202 uintptr_t start;
1203 uintptr_t vsize;
1204 intptr_t rss;
1205 uintptr_t rsslim;
1206 uintptr_t scodes;
1207 uintptr_t ecode;
1208 int i;
1209
1210 // Figure what the primordial thread stack base is. Code is inspired
1211 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1212 // followed by command name surrounded by parentheses, state, etc.
1213 char stat[2048];
1214 int statlen;
1215
1216 fp = fopen("/proc/self/stat", "r");
1217 if (fp) {
1218 statlen = fread(stat, 1, 2047, fp);
1219 stat[statlen] = '\0';
1220 fclose(fp);
1221
1222 // Skip pid and the command string. Note that we could be dealing with
1223 // weird command names, e.g. user could decide to rename java launcher
1224 // to "java 1.4.2 :)", then the stat file would look like
1225 // 1234 (java 1.4.2 :)) R ... ...
1226 // We don't really need to know the command string, just find the last
1227 // occurrence of ")" and then start parsing from there. See bug 4726580.
1228 char * s = strrchr(stat, ')');
1229
1230 i = 0;
1231 if (s) {
1232 // Skip blank chars
1233 do s++; while (isspace(*s));
1234
1235 #define _UFM UINTX_FORMAT
1236 #define _DFM INTX_FORMAT
1237
1238 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1239 /* 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 */
1240 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,
1241 &state, /* 3 %c */
1242 &ppid, /* 4 %d */
1243 &pgrp, /* 5 %d */
1244 &session, /* 6 %d */
1245 &nr, /* 7 %d */
1246 &tpgrp, /* 8 %d */
1247 &flags, /* 9 %lu */
1248 &minflt, /* 10 %lu */
1249 &cminflt, /* 11 %lu */
1250 &majflt, /* 12 %lu */
1251 &cmajflt, /* 13 %lu */
1252 &utime, /* 14 %lu */
1253 &stime, /* 15 %lu */
1254 &cutime, /* 16 %ld */
1255 &cstime, /* 17 %ld */
1256 &prio, /* 18 %ld */
1257 &nice, /* 19 %ld */
1258 &junk, /* 20 %ld */
1259 &it_real, /* 21 %ld */
1260 &start, /* 22 UINTX_FORMAT */
1261 &vsize, /* 23 UINTX_FORMAT */
1262 &rss, /* 24 INTX_FORMAT */
1263 &rsslim, /* 25 UINTX_FORMAT */
1264 &scodes, /* 26 UINTX_FORMAT */
1265 &ecode, /* 27 UINTX_FORMAT */
1266 &stack_start); /* 28 UINTX_FORMAT */
1267 }
1268
1269 #undef _UFM
1270 #undef _DFM
1271
1272 if (i != 28 - 2) {
1273 assert(false, "Bad conversion from /proc/self/stat");
1274 // product mode - assume we are the initial thread, good luck in the
1275 // embedded case.
1276 warning("Can't detect initial thread stack location - bad conversion");
1277 stack_start = (uintptr_t) &rlim;
1278 }
1279 } else {
1280 // For some reason we can't open /proc/self/stat (for example, running on
1281 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1282 // most cases, so don't abort:
1283 warning("Can't detect initial thread stack location - no /proc/self/stat");
1284 stack_start = (uintptr_t) &rlim;
1285 }
1286 }
1287
1288 // Now we have a pointer (stack_start) very close to the stack top, the
1289 // next thing to do is to figure out the exact location of stack top. We
1290 // can find out the virtual memory area that contains stack_start by
1291 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1292 // and its upper limit is the real stack top. (again, this would fail if
1293 // running inside chroot, because /proc may not exist.)
1294
1295 uintptr_t stack_top;
1296 address low, high;
1297 if (find_vma((address)stack_start, &low, &high)) {
1298 // success, "high" is the true stack top. (ignore "low", because initial
1299 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1300 stack_top = (uintptr_t)high;
1301 } else {
1302 // failed, likely because /proc/self/maps does not exist
1303 warning("Can't detect initial thread stack location - find_vma failed");
1304 // best effort: stack_start is normally within a few pages below the real
1305 // stack top, use it as stack top, and reduce stack size so we won't put
1306 // guard page outside stack.
1307 stack_top = stack_start;
1308 stack_size -= 16 * page_size();
1309 }
1310
1311 // stack_top could be partially down the page so align it
1312 stack_top = align_size_up(stack_top, page_size());
1313
1314 if (max_size && stack_size > max_size) {
1315 _initial_thread_stack_size = max_size;
1316 } else {
1317 _initial_thread_stack_size = stack_size;
1318 }
1319
1320 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1321 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1322 }
1323
1324 ////////////////////////////////////////////////////////////////////////////////
1325 // time support
1326
1327 // Time since start-up in seconds to a fine granularity.
1328 // Used by VMSelfDestructTimer and the MemProfiler.
1329 double os::elapsedTime() {
1330
1331 return (double)(os::elapsed_counter()) * 0.000001;
1332 }
1333
1334 jlong os::elapsed_counter() {
1335 timeval time;
1336 int status = gettimeofday(&time, NULL);
1337 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1338 }
1339
1340 jlong os::elapsed_frequency() {
1341 return (1000 * 1000);
1342 }
1343
1344 // For now, we say that linux does not support vtime. I have no idea
1345 // whether it can actually be made to (DLD, 9/13/05).
1346
1347 bool os::supports_vtime() { return false; }
1348 bool os::enable_vtime() { return false; }
1349 bool os::vtime_enabled() { return false; }
1350 double os::elapsedVTime() {
1351 // better than nothing, but not much
1352 return elapsedTime();
1353 }
1354
1355 jlong os::javaTimeMillis() {
1356 timeval time;
1357 int status = gettimeofday(&time, NULL);
1358 assert(status != -1, "linux error");
1359 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1360 }
1361
1362 #ifndef CLOCK_MONOTONIC
1363 #define CLOCK_MONOTONIC (1)
1364 #endif
1365
1366 void os::Linux::clock_init() {
1367 // we do dlopen's in this particular order due to bug in linux
1368 // dynamical loader (see 6348968) leading to crash on exit
1369 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1370 if (handle == NULL) {
1371 handle = dlopen("librt.so", RTLD_LAZY);
1372 }
1373
1374 if (handle) {
1375 int (*clock_getres_func)(clockid_t, struct timespec*) =
1376 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1377 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1378 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1379 if (clock_getres_func && clock_gettime_func) {
1380 // See if monotonic clock is supported by the kernel. Note that some
1381 // early implementations simply return kernel jiffies (updated every
1382 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1383 // for nano time (though the monotonic property is still nice to have).
1384 // It's fixed in newer kernels, however clock_getres() still returns
1385 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1386 // resolution for now. Hopefully as people move to new kernels, this
1387 // won't be a problem.
1388 struct timespec res;
1389 struct timespec tp;
1390 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1391 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1392 // yes, monotonic clock is supported
1393 _clock_gettime = clock_gettime_func;
1394 } else {
1395 // close librt if there is no monotonic clock
1396 dlclose(handle);
1397 }
1398 }
1399 }
1400 }
1401
1402 #ifndef SYS_clock_getres
1403
1404 #if defined(IA32) || defined(AMD64)
1405 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1406 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1407 #else
1408 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1409 #define sys_clock_getres(x,y) -1
1410 #endif
1411
1412 #else
1413 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1414 #endif
1415
1416 void os::Linux::fast_thread_clock_init() {
1417 if (!UseLinuxPosixThreadCPUClocks) {
1418 return;
1419 }
1420 clockid_t clockid;
1421 struct timespec tp;
1422 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1423 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1424
1425 // Switch to using fast clocks for thread cpu time if
1426 // the sys_clock_getres() returns 0 error code.
1427 // Note, that some kernels may support the current thread
1428 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1429 // returned by the pthread_getcpuclockid().
1430 // If the fast Posix clocks are supported then the sys_clock_getres()
1431 // must return at least tp.tv_sec == 0 which means a resolution
1432 // better than 1 sec. This is extra check for reliability.
1433
1434 if(pthread_getcpuclockid_func &&
1435 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1436 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1437
1438 _supports_fast_thread_cpu_time = true;
1439 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1440 }
1441 }
1442
1443 jlong os::javaTimeNanos() {
1444 if (Linux::supports_monotonic_clock()) {
1445 struct timespec tp;
1446 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1447 assert(status == 0, "gettime error");
1448 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1449 return result;
1450 } else {
1451 timeval time;
1452 int status = gettimeofday(&time, NULL);
1453 assert(status != -1, "linux error");
1454 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1455 return 1000 * usecs;
1456 }
1457 }
1458
1459 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1460 if (Linux::supports_monotonic_clock()) {
1461 info_ptr->max_value = ALL_64_BITS;
1462
1463 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1464 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1465 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1466 } else {
1467 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1468 info_ptr->max_value = ALL_64_BITS;
1469
1470 // gettimeofday is a real time clock so it skips
1471 info_ptr->may_skip_backward = true;
1472 info_ptr->may_skip_forward = true;
1473 }
1474
1475 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1476 }
1477
1478 // Return the real, user, and system times in seconds from an
1479 // arbitrary fixed point in the past.
1480 bool os::getTimesSecs(double* process_real_time,
1481 double* process_user_time,
1482 double* process_system_time) {
1483 struct tms ticks;
1484 clock_t real_ticks = times(&ticks);
1485
1486 if (real_ticks == (clock_t) (-1)) {
1487 return false;
1488 } else {
1489 double ticks_per_second = (double) clock_tics_per_sec;
1490 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1491 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1492 *process_real_time = ((double) real_ticks) / ticks_per_second;
1493
1494 return true;
1495 }
1496 }
1497
1498
1499 char * os::local_time_string(char *buf, size_t buflen) {
1500 struct tm t;
1501 time_t long_time;
1502 time(&long_time);
1503 localtime_r(&long_time, &t);
1504 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1505 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1506 t.tm_hour, t.tm_min, t.tm_sec);
1507 return buf;
1508 }
1509
1510 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1511 return localtime_r(clock, res);
1512 }
1513
1514 ////////////////////////////////////////////////////////////////////////////////
1515 // runtime exit support
1516
1517 // Note: os::shutdown() might be called very early during initialization, or
1518 // called from signal handler. Before adding something to os::shutdown(), make
1519 // sure it is async-safe and can handle partially initialized VM.
1520 void os::shutdown() {
1521
1522 // allow PerfMemory to attempt cleanup of any persistent resources
1523 perfMemory_exit();
1524
1525 // needs to remove object in file system
1526 AttachListener::abort();
1527
1528 // flush buffered output, finish log files
1529 ostream_abort();
1530
1531 // Check for abort hook
1532 abort_hook_t abort_hook = Arguments::abort_hook();
1533 if (abort_hook != NULL) {
1534 abort_hook();
1535 }
1536
1537 }
1538
1539 // Note: os::abort() might be called very early during initialization, or
1540 // called from signal handler. Before adding something to os::abort(), make
1541 // sure it is async-safe and can handle partially initialized VM.
1542 void os::abort(bool dump_core) {
1543 os::shutdown();
1544 if (dump_core) {
1545 #ifndef PRODUCT
1546 fdStream out(defaultStream::output_fd());
1547 out.print_raw("Current thread is ");
1548 char buf[16];
1549 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1550 out.print_raw_cr(buf);
1551 out.print_raw_cr("Dumping core ...");
1552 #endif
1553 ::abort(); // dump core
1554 }
1555
1556 ::exit(1);
1557 }
1558
1559 // Die immediately, no exit hook, no abort hook, no cleanup.
1560 void os::die() {
1561 // _exit() on LinuxThreads only kills current thread
1562 ::abort();
1563 }
1564
1565 // unused on linux for now.
1566 void os::set_error_file(const char *logfile) {}
1567
1568 intx os::current_thread_id() { return (intx)pthread_self(); }
1569 int os::current_process_id() {
1570
1571 // Under the old linux thread library, linux gives each thread
1572 // its own process id. Because of this each thread will return
1573 // a different pid if this method were to return the result
1574 // of getpid(2). Linux provides no api that returns the pid
1575 // of the launcher thread for the vm. This implementation
1576 // returns a unique pid, the pid of the launcher thread
1577 // that starts the vm 'process'.
1578
1579 // Under the NPTL, getpid() returns the same pid as the
1580 // launcher thread rather than a unique pid per thread.
1581 // Use gettid() if you want the old pre NPTL behaviour.
1582
1583 // if you are looking for the result of a call to getpid() that
1584 // returns a unique pid for the calling thread, then look at the
1585 // OSThread::thread_id() method in osThread_linux.hpp file
1586
1587 return (int)(_initial_pid ? _initial_pid : getpid());
1588 }
1589
1590 // DLL functions
1591
1592 const char* os::dll_file_extension() { return ".so"; }
1593
1594 const char* os::get_temp_directory() {
1595 const char *prop = Arguments::get_property("java.io.tmpdir");
1596 return prop == NULL ? "/tmp" : prop;
1597 }
1598
1599 static bool file_exists(const char* filename) {
1600 struct stat statbuf;
1601 if (filename == NULL || strlen(filename) == 0) {
1602 return false;
1603 }
1604 return os::stat(filename, &statbuf) == 0;
1605 }
1606
1607 void os::dll_build_name(char* buffer, size_t buflen,
1608 const char* pname, const char* fname) {
1609 // Copied from libhpi
1610 const size_t pnamelen = pname ? strlen(pname) : 0;
1611
1612 // Quietly truncate on buffer overflow. Should be an error.
1613 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1614 *buffer = '\0';
1615 return;
1616 }
1617
1618 if (pnamelen == 0) {
1619 snprintf(buffer, buflen, "lib%s.so", fname);
1620 } else if (strchr(pname, *os::path_separator()) != NULL) {
1621 int n;
1622 char** pelements = split_path(pname, &n);
1623 for (int i = 0 ; i < n ; i++) {
1624 // Really shouldn't be NULL, but check can't hurt
1625 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1626 continue; // skip the empty path values
1627 }
1628 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1629 if (file_exists(buffer)) {
1630 break;
1631 }
1632 }
1633 // release the storage
1634 for (int i = 0 ; i < n ; i++) {
1635 if (pelements[i] != NULL) {
1636 FREE_C_HEAP_ARRAY(char, pelements[i]);
1637 }
1638 }
1639 if (pelements != NULL) {
1640 FREE_C_HEAP_ARRAY(char*, pelements);
1641 }
1642 } else {
1643 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1644 }
1645 }
1646
1647 const char* os::get_current_directory(char *buf, int buflen) {
1648 return getcwd(buf, buflen);
1649 }
1650
1651 // check if addr is inside libjvm[_g].so
1652 bool os::address_is_in_vm(address addr) {
1653 static address libjvm_base_addr;
1654 Dl_info dlinfo;
1655
1656 if (libjvm_base_addr == NULL) {
1657 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1658 libjvm_base_addr = (address)dlinfo.dli_fbase;
1659 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1660 }
1661
1662 if (dladdr((void *)addr, &dlinfo)) {
1663 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1664 }
1665
1666 return false;
1667 }
1668
1669 bool os::dll_address_to_function_name(address addr, char *buf,
1670 int buflen, int *offset) {
1671 Dl_info dlinfo;
1672
1673 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1674 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1675 if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1676 return true;
1677 } else {
1678 if (buf) buf[0] = '\0';
1679 if (offset) *offset = -1;
1680 return false;
1681 }
1682 }
1683
1684 struct _address_to_library_name {
1685 address addr; // input : memory address
1686 size_t buflen; // size of fname
1687 char* fname; // output: library name
1688 address base; // library base addr
1689 };
1690
1691 static int address_to_library_name_callback(struct dl_phdr_info *info,
1692 size_t size, void *data) {
1693 int i;
1694 bool found = false;
1695 address libbase = NULL;
1696 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1697
1698 // iterate through all loadable segments
1699 for (i = 0; i < info->dlpi_phnum; i++) {
1700 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1701 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1702 // base address of a library is the lowest address of its loaded
1703 // segments.
1704 if (libbase == NULL || libbase > segbase) {
1705 libbase = segbase;
1706 }
1707 // see if 'addr' is within current segment
1708 if (segbase <= d->addr &&
1709 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1710 found = true;
1711 }
1712 }
1713 }
1714
1715 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1716 // so dll_address_to_library_name() can fall through to use dladdr() which
1717 // can figure out executable name from argv[0].
1718 if (found && info->dlpi_name && info->dlpi_name[0]) {
1719 d->base = libbase;
1720 if (d->fname) {
1721 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1722 }
1723 return 1;
1724 }
1725 return 0;
1726 }
1727
1728 bool os::dll_address_to_library_name(address addr, char* buf,
1729 int buflen, int* offset) {
1730 Dl_info dlinfo;
1731 struct _address_to_library_name data;
1732
1733 // There is a bug in old glibc dladdr() implementation that it could resolve
1734 // to wrong library name if the .so file has a base address != NULL. Here
1735 // we iterate through the program headers of all loaded libraries to find
1736 // out which library 'addr' really belongs to. This workaround can be
1737 // removed once the minimum requirement for glibc is moved to 2.3.x.
1738 data.addr = addr;
1739 data.fname = buf;
1740 data.buflen = buflen;
1741 data.base = NULL;
1742 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1743
1744 if (rslt) {
1745 // buf already contains library name
1746 if (offset) *offset = addr - data.base;
1747 return true;
1748 } else if (dladdr((void*)addr, &dlinfo)){
1749 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1750 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1751 return true;
1752 } else {
1753 if (buf) buf[0] = '\0';
1754 if (offset) *offset = -1;
1755 return false;
1756 }
1757 }
1758
1759 // Loads .dll/.so and
1760 // in case of error it checks if .dll/.so was built for the
1761 // same architecture as Hotspot is running on
1762
1763 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1764 {
1765 void * result= ::dlopen(filename, RTLD_LAZY);
1766 if (result != NULL) {
1767 // Successful loading
1768 return result;
1769 }
1770
1771 Elf32_Ehdr elf_head;
1772
1773 // Read system error message into ebuf
1774 // It may or may not be overwritten below
1775 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1776 ebuf[ebuflen-1]='\0';
1777 int diag_msg_max_length=ebuflen-strlen(ebuf);
1778 char* diag_msg_buf=ebuf+strlen(ebuf);
1779
1780 if (diag_msg_max_length==0) {
1781 // No more space in ebuf for additional diagnostics message
1782 return NULL;
1783 }
1784
1785
1786 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1787
1788 if (file_descriptor < 0) {
1789 // Can't open library, report dlerror() message
1790 return NULL;
1791 }
1792
1793 bool failed_to_read_elf_head=
1794 (sizeof(elf_head)!=
1795 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1796
1797 ::close(file_descriptor);
1798 if (failed_to_read_elf_head) {
1799 // file i/o error - report dlerror() msg
1800 return NULL;
1801 }
1802
1803 typedef struct {
1804 Elf32_Half code; // Actual value as defined in elf.h
1805 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1806 char elf_class; // 32 or 64 bit
1807 char endianess; // MSB or LSB
1808 char* name; // String representation
1809 } arch_t;
1810
1811 #ifndef EM_486
1812 #define EM_486 6 /* Intel 80486 */
1813 #endif
1814
1815 static const arch_t arch_array[]={
1816 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1817 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1818 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1819 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1820 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1821 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1822 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1823 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1824 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1825 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1826 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1827 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1828 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1829 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1830 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1831 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1832 };
1833
1834 #if (defined IA32)
1835 static Elf32_Half running_arch_code=EM_386;
1836 #elif (defined AMD64)
1837 static Elf32_Half running_arch_code=EM_X86_64;
1838 #elif (defined IA64)
1839 static Elf32_Half running_arch_code=EM_IA_64;
1840 #elif (defined __sparc) && (defined _LP64)
1841 static Elf32_Half running_arch_code=EM_SPARCV9;
1842 #elif (defined __sparc) && (!defined _LP64)
1843 static Elf32_Half running_arch_code=EM_SPARC;
1844 #elif (defined __powerpc64__)
1845 static Elf32_Half running_arch_code=EM_PPC64;
1846 #elif (defined __powerpc__)
1847 static Elf32_Half running_arch_code=EM_PPC;
1848 #elif (defined ARM)
1849 static Elf32_Half running_arch_code=EM_ARM;
1850 #elif (defined S390)
1851 static Elf32_Half running_arch_code=EM_S390;
1852 #elif (defined ALPHA)
1853 static Elf32_Half running_arch_code=EM_ALPHA;
1854 #elif (defined MIPSEL)
1855 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1856 #elif (defined PARISC)
1857 static Elf32_Half running_arch_code=EM_PARISC;
1858 #elif (defined MIPS)
1859 static Elf32_Half running_arch_code=EM_MIPS;
1860 #elif (defined M68K)
1861 static Elf32_Half running_arch_code=EM_68K;
1862 #else
1863 #error Method os::dll_load requires that one of following is defined:\
1864 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1865 #endif
1866
1867 // Identify compatability class for VM's architecture and library's architecture
1868 // Obtain string descriptions for architectures
1869
1870 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1871 int running_arch_index=-1;
1872
1873 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1874 if (running_arch_code == arch_array[i].code) {
1875 running_arch_index = i;
1876 }
1877 if (lib_arch.code == arch_array[i].code) {
1878 lib_arch.compat_class = arch_array[i].compat_class;
1879 lib_arch.name = arch_array[i].name;
1880 }
1881 }
1882
1883 assert(running_arch_index != -1,
1884 "Didn't find running architecture code (running_arch_code) in arch_array");
1885 if (running_arch_index == -1) {
1886 // Even though running architecture detection failed
1887 // we may still continue with reporting dlerror() message
1888 return NULL;
1889 }
1890
1891 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1892 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1893 return NULL;
1894 }
1895
1896 #ifndef S390
1897 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1898 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1899 return NULL;
1900 }
1901 #endif // !S390
1902
1903 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1904 if ( lib_arch.name!=NULL ) {
1905 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1906 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1907 lib_arch.name, arch_array[running_arch_index].name);
1908 } else {
1909 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1910 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1911 lib_arch.code,
1912 arch_array[running_arch_index].name);
1913 }
1914 }
1915
1916 return NULL;
1917 }
1918
1919 /*
1920 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1921 * chances are you might want to run the generated bits against glibc-2.0
1922 * libdl.so, so always use locking for any version of glibc.
1923 */
1924 void* os::dll_lookup(void* handle, const char* name) {
1925 pthread_mutex_lock(&dl_mutex);
1926 void* res = dlsym(handle, name);
1927 pthread_mutex_unlock(&dl_mutex);
1928 return res;
1929 }
1930
1931
1932 bool _print_ascii_file(const char* filename, outputStream* st) {
1933 int fd = open(filename, O_RDONLY);
1934 if (fd == -1) {
1935 return false;
1936 }
1937
1938 char buf[32];
1939 int bytes;
1940 while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1941 st->print_raw(buf, bytes);
1942 }
1943
1944 close(fd);
1945
1946 return true;
1947 }
1948
1949 void os::print_dll_info(outputStream *st) {
1950 st->print_cr("Dynamic libraries:");
1951
1952 char fname[32];
1953 pid_t pid = os::Linux::gettid();
1954
1955 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1956
1957 if (!_print_ascii_file(fname, st)) {
1958 st->print("Can not get library information for pid = %d\n", pid);
1959 }
1960 }
1961
1962
1963 void os::print_os_info(outputStream* st) {
1964 st->print("OS:");
1965
1966 // Try to identify popular distros.
1967 // Most Linux distributions have /etc/XXX-release file, which contains
1968 // the OS version string. Some have more than one /etc/XXX-release file
1969 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1970 // so the order is important.
1971 if (!_print_ascii_file("/etc/mandrake-release", st) &&
1972 !_print_ascii_file("/etc/sun-release", st) &&
1973 !_print_ascii_file("/etc/redhat-release", st) &&
1974 !_print_ascii_file("/etc/SuSE-release", st) &&
1975 !_print_ascii_file("/etc/turbolinux-release", st) &&
1976 !_print_ascii_file("/etc/gentoo-release", st) &&
1977 !_print_ascii_file("/etc/debian_version", st) &&
1978 !_print_ascii_file("/etc/ltib-release", st) &&
1979 !_print_ascii_file("/etc/angstrom-version", st)) {
1980 st->print("Linux");
1981 }
1982 st->cr();
1983
1984 // kernel
1985 st->print("uname:");
1986 struct utsname name;
1987 uname(&name);
1988 st->print(name.sysname); st->print(" ");
1989 st->print(name.release); st->print(" ");
1990 st->print(name.version); st->print(" ");
1991 st->print(name.machine);
1992 st->cr();
1993
1994 // Print warning if unsafe chroot environment detected
1995 if (unsafe_chroot_detected) {
1996 st->print("WARNING!! ");
1997 st->print_cr(unstable_chroot_error);
1998 }
1999
2000 // libc, pthread
2001 st->print("libc:");
2002 st->print(os::Linux::glibc_version()); st->print(" ");
2003 st->print(os::Linux::libpthread_version()); st->print(" ");
2004 if (os::Linux::is_LinuxThreads()) {
2005 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2006 }
2007 st->cr();
2008
2009 // rlimit
2010 st->print("rlimit:");
2011 struct rlimit rlim;
2012
2013 st->print(" STACK ");
2014 getrlimit(RLIMIT_STACK, &rlim);
2015 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2016 else st->print("%uk", rlim.rlim_cur >> 10);
2017
2018 st->print(", CORE ");
2019 getrlimit(RLIMIT_CORE, &rlim);
2020 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2021 else st->print("%uk", rlim.rlim_cur >> 10);
2022
2023 st->print(", NPROC ");
2024 getrlimit(RLIMIT_NPROC, &rlim);
2025 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2026 else st->print("%d", rlim.rlim_cur);
2027
2028 st->print(", NOFILE ");
2029 getrlimit(RLIMIT_NOFILE, &rlim);
2030 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2031 else st->print("%d", rlim.rlim_cur);
2032
2033 st->print(", AS ");
2034 getrlimit(RLIMIT_AS, &rlim);
2035 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2036 else st->print("%uk", rlim.rlim_cur >> 10);
2037 st->cr();
2038
2039 // load average
2040 st->print("load average:");
2041 double loadavg[3];
2042 os::loadavg(loadavg, 3);
2043 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2044 st->cr();
2045
2046 // meminfo
2047 st->print("\n/proc/meminfo:\n");
2048 _print_ascii_file("/proc/meminfo", st);
2049 st->cr();
2050 }
2051
2052 void os::print_memory_info(outputStream* st) {
2053
2054 st->print("Memory:");
2055 st->print(" %dk page", os::vm_page_size()>>10);
2056
2057 // values in struct sysinfo are "unsigned long"
2058 struct sysinfo si;
2059 sysinfo(&si);
2060
2061 st->print(", physical " UINT64_FORMAT "k",
2062 os::physical_memory() >> 10);
2063 st->print("(" UINT64_FORMAT "k free)",
2064 os::available_memory() >> 10);
2065 st->print(", swap " UINT64_FORMAT "k",
2066 ((jlong)si.totalswap * si.mem_unit) >> 10);
2067 st->print("(" UINT64_FORMAT "k free)",
2068 ((jlong)si.freeswap * si.mem_unit) >> 10);
2069 st->cr();
2070 }
2071
2072 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2073 // but they're the same for all the linux arch that we support
2074 // and they're the same for solaris but there's no common place to put this.
2075 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2076 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2077 "ILL_COPROC", "ILL_BADSTK" };
2078
2079 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2080 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2081 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2082
2083 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2084
2085 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2086
2087 void os::print_siginfo(outputStream* st, void* siginfo) {
2088 st->print("siginfo:");
2089
2090 const int buflen = 100;
2091 char buf[buflen];
2092 siginfo_t *si = (siginfo_t*)siginfo;
2093 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2094 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2095 st->print("si_errno=%s", buf);
2096 } else {
2097 st->print("si_errno=%d", si->si_errno);
2098 }
2099 const int c = si->si_code;
2100 assert(c > 0, "unexpected si_code");
2101 switch (si->si_signo) {
2102 case SIGILL:
2103 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2104 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2105 break;
2106 case SIGFPE:
2107 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2108 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2109 break;
2110 case SIGSEGV:
2111 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2112 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2113 break;
2114 case SIGBUS:
2115 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2116 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2117 break;
2118 default:
2119 st->print(", si_code=%d", si->si_code);
2120 // no si_addr
2121 }
2122
2123 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2124 UseSharedSpaces) {
2125 FileMapInfo* mapinfo = FileMapInfo::current_info();
2126 if (mapinfo->is_in_shared_space(si->si_addr)) {
2127 st->print("\n\nError accessing class data sharing archive." \
2128 " Mapped file inaccessible during execution, " \
2129 " possible disk/network problem.");
2130 }
2131 }
2132 st->cr();
2133 }
2134
2135
2136 static void print_signal_handler(outputStream* st, int sig,
2137 char* buf, size_t buflen);
2138
2139 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2140 st->print_cr("Signal Handlers:");
2141 print_signal_handler(st, SIGSEGV, buf, buflen);
2142 print_signal_handler(st, SIGBUS , buf, buflen);
2143 print_signal_handler(st, SIGFPE , buf, buflen);
2144 print_signal_handler(st, SIGPIPE, buf, buflen);
2145 print_signal_handler(st, SIGXFSZ, buf, buflen);
2146 print_signal_handler(st, SIGILL , buf, buflen);
2147 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2148 print_signal_handler(st, SR_signum, buf, buflen);
2149 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2150 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2151 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2152 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2153 }
2154
2155 static char saved_jvm_path[MAXPATHLEN] = {0};
2156
2157 // Find the full path to the current module, libjvm.so or libjvm_g.so
2158 void os::jvm_path(char *buf, jint buflen) {
2159 // Error checking.
2160 if (buflen < MAXPATHLEN) {
2161 assert(false, "must use a large-enough buffer");
2162 buf[0] = '\0';
2163 return;
2164 }
2165 // Lazy resolve the path to current module.
2166 if (saved_jvm_path[0] != 0) {
2167 strcpy(buf, saved_jvm_path);
2168 return;
2169 }
2170
2171 char dli_fname[MAXPATHLEN];
2172 bool ret = dll_address_to_library_name(
2173 CAST_FROM_FN_PTR(address, os::jvm_path),
2174 dli_fname, sizeof(dli_fname), NULL);
2175 assert(ret != 0, "cannot locate libjvm");
2176 char *rp = realpath(dli_fname, buf);
2177 if (rp == NULL)
2178 return;
2179
2180 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2181 // Support for the gamma launcher. Typical value for buf is
2182 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2183 // the right place in the string, then assume we are installed in a JDK and
2184 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2185 // up the path so it looks like libjvm.so is installed there (append a
2186 // fake suffix hotspot/libjvm.so).
2187 const char *p = buf + strlen(buf) - 1;
2188 for (int count = 0; p > buf && count < 5; ++count) {
2189 for (--p; p > buf && *p != '/'; --p)
2190 /* empty */ ;
2191 }
2192
2193 if (strncmp(p, "/jre/lib/", 9) != 0) {
2194 // Look for JAVA_HOME in the environment.
2195 char* java_home_var = ::getenv("JAVA_HOME");
2196 if (java_home_var != NULL && java_home_var[0] != 0) {
2197 char* jrelib_p;
2198 int len;
2199
2200 // Check the current module name "libjvm.so" or "libjvm_g.so".
2201 p = strrchr(buf, '/');
2202 assert(strstr(p, "/libjvm") == p, "invalid library name");
2203 p = strstr(p, "_g") ? "_g" : "";
2204
2205 rp = realpath(java_home_var, buf);
2206 if (rp == NULL)
2207 return;
2208
2209 // determine if this is a legacy image or modules image
2210 // modules image doesn't have "jre" subdirectory
2211 len = strlen(buf);
2212 jrelib_p = buf + len;
2213 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2214 if (0 != access(buf, F_OK)) {
2215 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2216 }
2217
2218 if (0 == access(buf, F_OK)) {
2219 // Use current module name "libjvm[_g].so" instead of
2220 // "libjvm"debug_only("_g")".so" since for fastdebug version
2221 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2222 // It is used when we are choosing the HPI library's name
2223 // "libhpi[_g].so" in hpi::initialize_get_interface().
2224 len = strlen(buf);
2225 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2226 } else {
2227 // Go back to path of .so
2228 rp = realpath(dli_fname, buf);
2229 if (rp == NULL)
2230 return;
2231 }
2232 }
2233 }
2234 }
2235
2236 strcpy(saved_jvm_path, buf);
2237 }
2238
2239 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2240 // no prefix required, not even "_"
2241 }
2242
2243 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2244 // no suffix required
2245 }
2246
2247 ////////////////////////////////////////////////////////////////////////////////
2248 // sun.misc.Signal support
2249
2250 static volatile jint sigint_count = 0;
2251
2252 static void
2253 UserHandler(int sig, void *siginfo, void *context) {
2254 // 4511530 - sem_post is serialized and handled by the manager thread. When
2255 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2256 // don't want to flood the manager thread with sem_post requests.
2257 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2258 return;
2259
2260 // Ctrl-C is pressed during error reporting, likely because the error
2261 // handler fails to abort. Let VM die immediately.
2262 if (sig == SIGINT && is_error_reported()) {
2263 os::die();
2264 }
2265
2266 os::signal_notify(sig);
2267 }
2268
2269 void* os::user_handler() {
2270 return CAST_FROM_FN_PTR(void*, UserHandler);
2271 }
2272
2273 extern "C" {
2274 typedef void (*sa_handler_t)(int);
2275 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2276 }
2277
2278 void* os::signal(int signal_number, void* handler) {
2279 struct sigaction sigAct, oldSigAct;
2280
2281 sigfillset(&(sigAct.sa_mask));
2282 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2283 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2284
2285 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2286 // -1 means registration failed
2287 return (void *)-1;
2288 }
2289
2290 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2291 }
2292
2293 void os::signal_raise(int signal_number) {
2294 ::raise(signal_number);
2295 }
2296
2297 /*
2298 * The following code is moved from os.cpp for making this
2299 * code platform specific, which it is by its very nature.
2300 */
2301
2302 // Will be modified when max signal is changed to be dynamic
2303 int os::sigexitnum_pd() {
2304 return NSIG;
2305 }
2306
2307 // a counter for each possible signal value
2308 static volatile jint pending_signals[NSIG+1] = { 0 };
2309
2310 // Linux(POSIX) specific hand shaking semaphore.
2311 static sem_t sig_sem;
2312
2313 void os::signal_init_pd() {
2314 // Initialize signal structures
2315 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2316
2317 // Initialize signal semaphore
2318 ::sem_init(&sig_sem, 0, 0);
2319 }
2320
2321 void os::signal_notify(int sig) {
2322 Atomic::inc(&pending_signals[sig]);
2323 ::sem_post(&sig_sem);
2324 }
2325
2326 static int check_pending_signals(bool wait) {
2327 Atomic::store(0, &sigint_count);
2328 for (;;) {
2329 for (int i = 0; i < NSIG + 1; i++) {
2330 jint n = pending_signals[i];
2331 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2332 return i;
2333 }
2334 }
2335 if (!wait) {
2336 return -1;
2337 }
2338 JavaThread *thread = JavaThread::current();
2339 ThreadBlockInVM tbivm(thread);
2340
2341 bool threadIsSuspended;
2342 do {
2343 thread->set_suspend_equivalent();
2344 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2345 ::sem_wait(&sig_sem);
2346
2347 // were we externally suspended while we were waiting?
2348 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2349 if (threadIsSuspended) {
2350 //
2351 // The semaphore has been incremented, but while we were waiting
2352 // another thread suspended us. We don't want to continue running
2353 // while suspended because that would surprise the thread that
2354 // suspended us.
2355 //
2356 ::sem_post(&sig_sem);
2357
2358 thread->java_suspend_self();
2359 }
2360 } while (threadIsSuspended);
2361 }
2362 }
2363
2364 int os::signal_lookup() {
2365 return check_pending_signals(false);
2366 }
2367
2368 int os::signal_wait() {
2369 return check_pending_signals(true);
2370 }
2371
2372 ////////////////////////////////////////////////////////////////////////////////
2373 // Virtual Memory
2374
2375 int os::vm_page_size() {
2376 // Seems redundant as all get out
2377 assert(os::Linux::page_size() != -1, "must call os::init");
2378 return os::Linux::page_size();
2379 }
2380
2381 // Solaris allocates memory by pages.
2382 int os::vm_allocation_granularity() {
2383 assert(os::Linux::page_size() != -1, "must call os::init");
2384 return os::Linux::page_size();
2385 }
2386
2387 // Rationale behind this function:
2388 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2389 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2390 // samples for JITted code. Here we create private executable mapping over the code cache
2391 // and then we can use standard (well, almost, as mapping can change) way to provide
2392 // info for the reporting script by storing timestamp and location of symbol
2393 void linux_wrap_code(char* base, size_t size) {
2394 static volatile jint cnt = 0;
2395
2396 if (!UseOprofile) {
2397 return;
2398 }
2399
2400 char buf[PATH_MAX+1];
2401 int num = Atomic::add(1, &cnt);
2402
2403 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2404 os::get_temp_directory(), os::current_process_id(), num);
2405 unlink(buf);
2406
2407 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2408
2409 if (fd != -1) {
2410 off_t rv = lseek(fd, size-2, SEEK_SET);
2411 if (rv != (off_t)-1) {
2412 if (write(fd, "", 1) == 1) {
2413 mmap(base, size,
2414 PROT_READ|PROT_WRITE|PROT_EXEC,
2415 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2416 }
2417 }
2418 close(fd);
2419 unlink(buf);
2420 }
2421 }
2422
2423 // NOTE: Linux kernel does not really reserve the pages for us.
2424 // All it does is to check if there are enough free pages
2425 // left at the time of mmap(). This could be a potential
2426 // problem.
2427 bool os::commit_memory(char* addr, size_t size, bool exec) {
2428 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2429 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2430 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2431 return res != (uintptr_t) MAP_FAILED;
2432 }
2433
2434 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2435 bool exec) {
2436 return commit_memory(addr, size, exec);
2437 }
2438
2439 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2440
2441 void os::free_memory(char *addr, size_t bytes) {
2442 ::mmap(addr, bytes, PROT_READ | PROT_WRITE,
2443 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2444 }
2445
2446 void os::numa_make_global(char *addr, size_t bytes) {
2447 Linux::numa_interleave_memory(addr, bytes);
2448 }
2449
2450 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2451 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2452 }
2453
2454 bool os::numa_topology_changed() { return false; }
2455
2456 size_t os::numa_get_groups_num() {
2457 int max_node = Linux::numa_max_node();
2458 return max_node > 0 ? max_node + 1 : 1;
2459 }
2460
2461 int os::numa_get_group_id() {
2462 int cpu_id = Linux::sched_getcpu();
2463 if (cpu_id != -1) {
2464 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2465 if (lgrp_id != -1) {
2466 return lgrp_id;
2467 }
2468 }
2469 return 0;
2470 }
2471
2472 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2473 for (size_t i = 0; i < size; i++) {
2474 ids[i] = i;
2475 }
2476 return size;
2477 }
2478
2479 bool os::get_page_info(char *start, page_info* info) {
2480 return false;
2481 }
2482
2483 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2484 return end;
2485 }
2486
2487 extern "C" void numa_warn(int number, char *where, ...) { }
2488 extern "C" void numa_error(char *where) { }
2489
2490
2491 // If we are running with libnuma version > 2, then we should
2492 // be trying to use symbols with versions 1.1
2493 // If we are running with earlier version, which did not have symbol versions,
2494 // we should use the base version.
2495 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2496 void *f = dlvsym(handle, name, "libnuma_1.1");
2497 if (f == NULL) {
2498 f = dlsym(handle, name);
2499 }
2500 return f;
2501 }
2502
2503 bool os::Linux::libnuma_init() {
2504 // sched_getcpu() should be in libc.
2505 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2506 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2507
2508 if (sched_getcpu() != -1) { // Does it work?
2509 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2510 if (handle != NULL) {
2511 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2512 libnuma_dlsym(handle, "numa_node_to_cpus")));
2513 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2514 libnuma_dlsym(handle, "numa_max_node")));
2515 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2516 libnuma_dlsym(handle, "numa_available")));
2517 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2518 libnuma_dlsym(handle, "numa_tonode_memory")));
2519 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2520 libnuma_dlsym(handle, "numa_interleave_memory")));
2521
2522
2523 if (numa_available() != -1) {
2524 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2525 // Create a cpu -> node mapping
2526 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2527 rebuild_cpu_to_node_map();
2528 return true;
2529 }
2530 }
2531 }
2532 return false;
2533 }
2534
2535 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2536 // The table is later used in get_node_by_cpu().
2537 void os::Linux::rebuild_cpu_to_node_map() {
2538 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2539 // in libnuma (possible values are starting from 16,
2540 // and continuing up with every other power of 2, but less
2541 // than the maximum number of CPUs supported by kernel), and
2542 // is a subject to change (in libnuma version 2 the requirements
2543 // are more reasonable) we'll just hardcode the number they use
2544 // in the library.
2545 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2546
2547 size_t cpu_num = os::active_processor_count();
2548 size_t cpu_map_size = NCPUS / BitsPerCLong;
2549 size_t cpu_map_valid_size =
2550 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2551
2552 cpu_to_node()->clear();
2553 cpu_to_node()->at_grow(cpu_num - 1);
2554 size_t node_num = numa_get_groups_num();
2555
2556 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2557 for (size_t i = 0; i < node_num; i++) {
2558 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2559 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2560 if (cpu_map[j] != 0) {
2561 for (size_t k = 0; k < BitsPerCLong; k++) {
2562 if (cpu_map[j] & (1UL << k)) {
2563 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2564 }
2565 }
2566 }
2567 }
2568 }
2569 }
2570 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2571 }
2572
2573 int os::Linux::get_node_by_cpu(int cpu_id) {
2574 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2575 return cpu_to_node()->at(cpu_id);
2576 }
2577 return -1;
2578 }
2579
2580 GrowableArray<int>* os::Linux::_cpu_to_node;
2581 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2582 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2583 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2584 os::Linux::numa_available_func_t os::Linux::_numa_available;
2585 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2586 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2587 unsigned long* os::Linux::_numa_all_nodes;
2588
2589 bool os::uncommit_memory(char* addr, size_t size) {
2590 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2591 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2592 return res != (uintptr_t) MAP_FAILED;
2593 }
2594
2595 // Linux uses a growable mapping for the stack, and if the mapping for
2596 // the stack guard pages is not removed when we detach a thread the
2597 // stack cannot grow beyond the pages where the stack guard was
2598 // mapped. If at some point later in the process the stack expands to
2599 // that point, the Linux kernel cannot expand the stack any further
2600 // because the guard pages are in the way, and a segfault occurs.
2601 //
2602 // However, it's essential not to split the stack region by unmapping
2603 // a region (leaving a hole) that's already part of the stack mapping,
2604 // so if the stack mapping has already grown beyond the guard pages at
2605 // the time we create them, we have to truncate the stack mapping.
2606 // So, we need to know the extent of the stack mapping when
2607 // create_stack_guard_pages() is called.
2608
2609 // Find the bounds of the stack mapping. Return true for success.
2610 //
2611 // We only need this for stacks that are growable: at the time of
2612 // writing thread stacks don't use growable mappings (i.e. those
2613 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2614 // only applies to the main thread.
2615 static bool
2616 get_stack_bounds(uintptr_t *bottom, uintptr_t *top)
2617 {
2618 FILE *f = fopen("/proc/self/maps", "r");
2619 if (f == NULL)
2620 return false;
2621
2622 while (!feof(f)) {
2623 size_t dummy;
2624 char *str = NULL;
2625 ssize_t len = getline(&str, &dummy, f);
2626 if (len == -1) {
2627 fclose(f);
2628 return false;
2629 }
2630
2631 if (len > 0 && str[len-1] == '\n') {
2632 str[len-1] = 0;
2633 len--;
2634 }
2635
2636 static const char *stack_str = "[stack]";
2637 if (len > (ssize_t)strlen(stack_str)
2638 && (strcmp(str + len - strlen(stack_str), stack_str) == 0)) {
2639 if (sscanf(str, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2640 uintptr_t sp = (uintptr_t)__builtin_frame_address(0);
2641 if (sp >= *bottom && sp <= *top) {
2642 free(str);
2643 fclose(f);
2644 return true;
2645 }
2646 }
2647 }
2648 free(str);
2649 }
2650 fclose(f);
2651 return false;
2652 }
2653
2654 // If the (growable) stack mapping already extends beyond the point
2655 // where we're going to put our guard pages, truncate the mapping at
2656 // that point by munmap()ping it. This ensures that when we later
2657 // munmap() the guard pages we don't leave a hole in the stack
2658 // mapping. This only affects the main/initial thread, but guard
2659 // against future OS changes
2660 bool os::create_stack_guard_pages(char* addr, size_t size) {
2661 uintptr_t stack_extent, stack_base;
2662 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2663 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2664 assert(os::Linux::is_initial_thread(),
2665 "growable stack in non-initial thread");
2666 if (stack_extent < (uintptr_t)addr)
2667 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2668 }
2669
2670 return os::commit_memory(addr, size);
2671 }
2672
2673 // If this is a growable mapping, remove the guard pages entirely by
2674 // munmap()ping them. If not, just call uncommit_memory(). This only
2675 // affects the main/initial thread, but guard against future OS changes
2676 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2677 uintptr_t stack_extent, stack_base;
2678 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2679 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2680 assert(os::Linux::is_initial_thread(),
2681 "growable stack in non-initial thread");
2682
2683 return ::munmap(addr, size) == 0;
2684 }
2685
2686 return os::uncommit_memory(addr, size);
2687 }
2688
2689 static address _highest_vm_reserved_address = NULL;
2690
2691 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2692 // at 'requested_addr'. If there are existing memory mappings at the same
2693 // location, however, they will be overwritten. If 'fixed' is false,
2694 // 'requested_addr' is only treated as a hint, the return value may or
2695 // may not start from the requested address. Unlike Linux mmap(), this
2696 // function returns NULL to indicate failure.
2697 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2698 char * addr;
2699 int flags;
2700
2701 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2702 if (fixed) {
2703 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2704 flags |= MAP_FIXED;
2705 }
2706
2707 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2708 // to PROT_EXEC if executable when we commit the page.
2709 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2710 flags, -1, 0);
2711
2712 if (addr != MAP_FAILED) {
2713 // anon_mmap() should only get called during VM initialization,
2714 // don't need lock (actually we can skip locking even it can be called
2715 // from multiple threads, because _highest_vm_reserved_address is just a
2716 // hint about the upper limit of non-stack memory regions.)
2717 if ((address)addr + bytes > _highest_vm_reserved_address) {
2718 _highest_vm_reserved_address = (address)addr + bytes;
2719 }
2720 }
2721
2722 return addr == MAP_FAILED ? NULL : addr;
2723 }
2724
2725 // Don't update _highest_vm_reserved_address, because there might be memory
2726 // regions above addr + size. If so, releasing a memory region only creates
2727 // a hole in the address space, it doesn't help prevent heap-stack collision.
2728 //
2729 static int anon_munmap(char * addr, size_t size) {
2730 return ::munmap(addr, size) == 0;
2731 }
2732
2733 char* os::reserve_memory(size_t bytes, char* requested_addr,
2734 size_t alignment_hint) {
2735 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2736 }
2737
2738 bool os::release_memory(char* addr, size_t size) {
2739 return anon_munmap(addr, size);
2740 }
2741
2742 static address highest_vm_reserved_address() {
2743 return _highest_vm_reserved_address;
2744 }
2745
2746 static bool linux_mprotect(char* addr, size_t size, int prot) {
2747 // Linux wants the mprotect address argument to be page aligned.
2748 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2749
2750 // According to SUSv3, mprotect() should only be used with mappings
2751 // established by mmap(), and mmap() always maps whole pages. Unaligned
2752 // 'addr' likely indicates problem in the VM (e.g. trying to change
2753 // protection of malloc'ed or statically allocated memory). Check the
2754 // caller if you hit this assert.
2755 assert(addr == bottom, "sanity check");
2756
2757 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2758 return ::mprotect(bottom, size, prot) == 0;
2759 }
2760
2761 // Set protections specified
2762 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2763 bool is_committed) {
2764 unsigned int p = 0;
2765 switch (prot) {
2766 case MEM_PROT_NONE: p = PROT_NONE; break;
2767 case MEM_PROT_READ: p = PROT_READ; break;
2768 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2769 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2770 default:
2771 ShouldNotReachHere();
2772 }
2773 // is_committed is unused.
2774 return linux_mprotect(addr, bytes, p);
2775 }
2776
2777 bool os::guard_memory(char* addr, size_t size) {
2778 return linux_mprotect(addr, size, PROT_NONE);
2779 }
2780
2781 bool os::unguard_memory(char* addr, size_t size) {
2782 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2783 }
2784
2785 // Large page support
2786
2787 static size_t _large_page_size = 0;
2788
2789 bool os::large_page_init() {
2790 if (!UseLargePages) return false;
2791
2792 if (LargePageSizeInBytes) {
2793 _large_page_size = LargePageSizeInBytes;
2794 } else {
2795 // large_page_size on Linux is used to round up heap size. x86 uses either
2796 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2797 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2798 // page as large as 256M.
2799 //
2800 // Here we try to figure out page size by parsing /proc/meminfo and looking
2801 // for a line with the following format:
2802 // Hugepagesize: 2048 kB
2803 //
2804 // If we can't determine the value (e.g. /proc is not mounted, or the text
2805 // format has been changed), we'll use the largest page size supported by
2806 // the processor.
2807
2808 #ifndef ZERO
2809 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
2810 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
2811 #endif // ZERO
2812
2813 FILE *fp = fopen("/proc/meminfo", "r");
2814 if (fp) {
2815 while (!feof(fp)) {
2816 int x = 0;
2817 char buf[16];
2818 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2819 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2820 _large_page_size = x * K;
2821 break;
2822 }
2823 } else {
2824 // skip to next line
2825 for (;;) {
2826 int ch = fgetc(fp);
2827 if (ch == EOF || ch == (int)'\n') break;
2828 }
2829 }
2830 }
2831 fclose(fp);
2832 }
2833 }
2834
2835 const size_t default_page_size = (size_t)Linux::page_size();
2836 if (_large_page_size > default_page_size) {
2837 _page_sizes[0] = _large_page_size;
2838 _page_sizes[1] = default_page_size;
2839 _page_sizes[2] = 0;
2840 }
2841
2842 // Large page support is available on 2.6 or newer kernel, some vendors
2843 // (e.g. Redhat) have backported it to their 2.4 based distributions.
2844 // We optimistically assume the support is available. If later it turns out
2845 // not true, VM will automatically switch to use regular page size.
2846 return true;
2847 }
2848
2849 #ifndef SHM_HUGETLB
2850 #define SHM_HUGETLB 04000
2851 #endif
2852
2853 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
2854 // "exec" is passed in but not used. Creating the shared image for
2855 // the code cache doesn't have an SHM_X executable permission to check.
2856 assert(UseLargePages, "only for large pages");
2857
2858 key_t key = IPC_PRIVATE;
2859 char *addr;
2860
2861 bool warn_on_failure = UseLargePages &&
2862 (!FLAG_IS_DEFAULT(UseLargePages) ||
2863 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2864 );
2865 char msg[128];
2866
2867 // Create a large shared memory region to attach to based on size.
2868 // Currently, size is the total size of the heap
2869 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2870 if (shmid == -1) {
2871 // Possible reasons for shmget failure:
2872 // 1. shmmax is too small for Java heap.
2873 // > check shmmax value: cat /proc/sys/kernel/shmmax
2874 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2875 // 2. not enough large page memory.
2876 // > check available large pages: cat /proc/meminfo
2877 // > increase amount of large pages:
2878 // echo new_value > /proc/sys/vm/nr_hugepages
2879 // Note 1: different Linux may use different name for this property,
2880 // e.g. on Redhat AS-3 it is "hugetlb_pool".
2881 // Note 2: it's possible there's enough physical memory available but
2882 // they are so fragmented after a long run that they can't
2883 // coalesce into large pages. Try to reserve large pages when
2884 // the system is still "fresh".
2885 if (warn_on_failure) {
2886 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2887 warning(msg);
2888 }
2889 return NULL;
2890 }
2891
2892 // attach to the region
2893 addr = (char*)shmat(shmid, req_addr, 0);
2894 int err = errno;
2895
2896 // Remove shmid. If shmat() is successful, the actual shared memory segment
2897 // will be deleted when it's detached by shmdt() or when the process
2898 // terminates. If shmat() is not successful this will remove the shared
2899 // segment immediately.
2900 shmctl(shmid, IPC_RMID, NULL);
2901
2902 if ((intptr_t)addr == -1) {
2903 if (warn_on_failure) {
2904 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2905 warning(msg);
2906 }
2907 return NULL;
2908 }
2909
2910 return addr;
2911 }
2912
2913 bool os::release_memory_special(char* base, size_t bytes) {
2914 // detaching the SHM segment will also delete it, see reserve_memory_special()
2915 int rslt = shmdt(base);
2916 return rslt == 0;
2917 }
2918
2919 size_t os::large_page_size() {
2920 return _large_page_size;
2921 }
2922
2923 // Linux does not support anonymous mmap with large page memory. The only way
2924 // to reserve large page memory without file backing is through SysV shared
2925 // memory API. The entire memory region is committed and pinned upfront.
2926 // Hopefully this will change in the future...
2927 bool os::can_commit_large_page_memory() {
2928 return false;
2929 }
2930
2931 bool os::can_execute_large_page_memory() {
2932 return false;
2933 }
2934
2935 // Reserve memory at an arbitrary address, only if that area is
2936 // available (and not reserved for something else).
2937
2938 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2939 const int max_tries = 10;
2940 char* base[max_tries];
2941 size_t size[max_tries];
2942 const size_t gap = 0x000000;
2943
2944 // Assert only that the size is a multiple of the page size, since
2945 // that's all that mmap requires, and since that's all we really know
2946 // about at this low abstraction level. If we need higher alignment,
2947 // we can either pass an alignment to this method or verify alignment
2948 // in one of the methods further up the call chain. See bug 5044738.
2949 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2950
2951 // Repeatedly allocate blocks until the block is allocated at the
2952 // right spot. Give up after max_tries. Note that reserve_memory() will
2953 // automatically update _highest_vm_reserved_address if the call is
2954 // successful. The variable tracks the highest memory address every reserved
2955 // by JVM. It is used to detect heap-stack collision if running with
2956 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2957 // space than needed, it could confuse the collision detecting code. To
2958 // solve the problem, save current _highest_vm_reserved_address and
2959 // calculate the correct value before return.
2960 address old_highest = _highest_vm_reserved_address;
2961
2962 // Linux mmap allows caller to pass an address as hint; give it a try first,
2963 // if kernel honors the hint then we can return immediately.
2964 char * addr = anon_mmap(requested_addr, bytes, false);
2965 if (addr == requested_addr) {
2966 return requested_addr;
2967 }
2968
2969 if (addr != NULL) {
2970 // mmap() is successful but it fails to reserve at the requested address
2971 anon_munmap(addr, bytes);
2972 }
2973
2974 int i;
2975 for (i = 0; i < max_tries; ++i) {
2976 base[i] = reserve_memory(bytes);
2977
2978 if (base[i] != NULL) {
2979 // Is this the block we wanted?
2980 if (base[i] == requested_addr) {
2981 size[i] = bytes;
2982 break;
2983 }
2984
2985 // Does this overlap the block we wanted? Give back the overlapped
2986 // parts and try again.
2987
2988 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2989 if (top_overlap >= 0 && top_overlap < bytes) {
2990 unmap_memory(base[i], top_overlap);
2991 base[i] += top_overlap;
2992 size[i] = bytes - top_overlap;
2993 } else {
2994 size_t bottom_overlap = base[i] + bytes - requested_addr;
2995 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2996 unmap_memory(requested_addr, bottom_overlap);
2997 size[i] = bytes - bottom_overlap;
2998 } else {
2999 size[i] = bytes;
3000 }
3001 }
3002 }
3003 }
3004
3005 // Give back the unused reserved pieces.
3006
3007 for (int j = 0; j < i; ++j) {
3008 if (base[j] != NULL) {
3009 unmap_memory(base[j], size[j]);
3010 }
3011 }
3012
3013 if (i < max_tries) {
3014 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3015 return requested_addr;
3016 } else {
3017 _highest_vm_reserved_address = old_highest;
3018 return NULL;
3019 }
3020 }
3021
3022 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3023 return ::read(fd, buf, nBytes);
3024 }
3025
3026 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3027 // Solaris uses poll(), linux uses park().
3028 // Poll() is likely a better choice, assuming that Thread.interrupt()
3029 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3030 // SIGSEGV, see 4355769.
3031
3032 const int NANOSECS_PER_MILLISECS = 1000000;
3033
3034 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3035 assert(thread == Thread::current(), "thread consistency check");
3036
3037 ParkEvent * const slp = thread->_SleepEvent ;
3038 slp->reset() ;
3039 OrderAccess::fence() ;
3040
3041 if (interruptible) {
3042 jlong prevtime = javaTimeNanos();
3043
3044 for (;;) {
3045 if (os::is_interrupted(thread, true)) {
3046 return OS_INTRPT;
3047 }
3048
3049 jlong newtime = javaTimeNanos();
3050
3051 if (newtime - prevtime < 0) {
3052 // time moving backwards, should only happen if no monotonic clock
3053 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3054 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3055 } else {
3056 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3057 }
3058
3059 if(millis <= 0) {
3060 return OS_OK;
3061 }
3062
3063 prevtime = newtime;
3064
3065 {
3066 assert(thread->is_Java_thread(), "sanity check");
3067 JavaThread *jt = (JavaThread *) thread;
3068 ThreadBlockInVM tbivm(jt);
3069 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3070
3071 jt->set_suspend_equivalent();
3072 // cleared by handle_special_suspend_equivalent_condition() or
3073 // java_suspend_self() via check_and_wait_while_suspended()
3074
3075 slp->park(millis);
3076
3077 // were we externally suspended while we were waiting?
3078 jt->check_and_wait_while_suspended();
3079 }
3080 }
3081 } else {
3082 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3083 jlong prevtime = javaTimeNanos();
3084
3085 for (;;) {
3086 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3087 // the 1st iteration ...
3088 jlong newtime = javaTimeNanos();
3089
3090 if (newtime - prevtime < 0) {
3091 // time moving backwards, should only happen if no monotonic clock
3092 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3093 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3094 } else {
3095 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3096 }
3097
3098 if(millis <= 0) break ;
3099
3100 prevtime = newtime;
3101 slp->park(millis);
3102 }
3103 return OS_OK ;
3104 }
3105 }
3106
3107 int os::naked_sleep() {
3108 // %% make the sleep time an integer flag. for now use 1 millisec.
3109 return os::sleep(Thread::current(), 1, false);
3110 }
3111
3112 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3113 void os::infinite_sleep() {
3114 while (true) { // sleep forever ...
3115 ::sleep(100); // ... 100 seconds at a time
3116 }
3117 }
3118
3119 // Used to convert frequent JVM_Yield() to nops
3120 bool os::dont_yield() {
3121 return DontYieldALot;
3122 }
3123
3124 void os::yield() {
3125 sched_yield();
3126 }
3127
3128 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3129
3130 void os::yield_all(int attempts) {
3131 // Yields to all threads, including threads with lower priorities
3132 // Threads on Linux are all with same priority. The Solaris style
3133 // os::yield_all() with nanosleep(1ms) is not necessary.
3134 sched_yield();
3135 }
3136
3137 // Called from the tight loops to possibly influence time-sharing heuristics
3138 void os::loop_breaker(int attempts) {
3139 os::yield_all(attempts);
3140 }
3141
3142 ////////////////////////////////////////////////////////////////////////////////
3143 // thread priority support
3144
3145 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3146 // only supports dynamic priority, static priority must be zero. For real-time
3147 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3148 // However, for large multi-threaded applications, SCHED_RR is not only slower
3149 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3150 // of 5 runs - Sep 2005).
3151 //
3152 // The following code actually changes the niceness of kernel-thread/LWP. It
3153 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3154 // not the entire user process, and user level threads are 1:1 mapped to kernel
3155 // threads. It has always been the case, but could change in the future. For
3156 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3157 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3158
3159 int os::java_to_os_priority[MaxPriority + 1] = {
3160 19, // 0 Entry should never be used
3161
3162 4, // 1 MinPriority
3163 3, // 2
3164 2, // 3
3165
3166 1, // 4
3167 0, // 5 NormPriority
3168 -1, // 6
3169
3170 -2, // 7
3171 -3, // 8
3172 -4, // 9 NearMaxPriority
3173
3174 -5 // 10 MaxPriority
3175 };
3176
3177 static int prio_init() {
3178 if (ThreadPriorityPolicy == 1) {
3179 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3180 // if effective uid is not root. Perhaps, a more elegant way of doing
3181 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3182 if (geteuid() != 0) {
3183 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3184 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3185 }
3186 ThreadPriorityPolicy = 0;
3187 }
3188 }
3189 return 0;
3190 }
3191
3192 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3193 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3194
3195 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3196 return (ret == 0) ? OS_OK : OS_ERR;
3197 }
3198
3199 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3200 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3201 *priority_ptr = java_to_os_priority[NormPriority];
3202 return OS_OK;
3203 }
3204
3205 errno = 0;
3206 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3207 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3208 }
3209
3210 // Hint to the underlying OS that a task switch would not be good.
3211 // Void return because it's a hint and can fail.
3212 void os::hint_no_preempt() {}
3213
3214 ////////////////////////////////////////////////////////////////////////////////
3215 // suspend/resume support
3216
3217 // the low-level signal-based suspend/resume support is a remnant from the
3218 // old VM-suspension that used to be for java-suspension, safepoints etc,
3219 // within hotspot. Now there is a single use-case for this:
3220 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3221 // that runs in the watcher thread.
3222 // The remaining code is greatly simplified from the more general suspension
3223 // code that used to be used.
3224 //
3225 // The protocol is quite simple:
3226 // - suspend:
3227 // - sends a signal to the target thread
3228 // - polls the suspend state of the osthread using a yield loop
3229 // - target thread signal handler (SR_handler) sets suspend state
3230 // and blocks in sigsuspend until continued
3231 // - resume:
3232 // - sets target osthread state to continue
3233 // - sends signal to end the sigsuspend loop in the SR_handler
3234 //
3235 // Note that the SR_lock plays no role in this suspend/resume protocol.
3236 //
3237
3238 static void resume_clear_context(OSThread *osthread) {
3239 osthread->set_ucontext(NULL);
3240 osthread->set_siginfo(NULL);
3241
3242 // notify the suspend action is completed, we have now resumed
3243 osthread->sr.clear_suspended();
3244 }
3245
3246 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3247 osthread->set_ucontext(context);
3248 osthread->set_siginfo(siginfo);
3249 }
3250
3251 //
3252 // Handler function invoked when a thread's execution is suspended or
3253 // resumed. We have to be careful that only async-safe functions are
3254 // called here (Note: most pthread functions are not async safe and
3255 // should be avoided.)
3256 //
3257 // Note: sigwait() is a more natural fit than sigsuspend() from an
3258 // interface point of view, but sigwait() prevents the signal hander
3259 // from being run. libpthread would get very confused by not having
3260 // its signal handlers run and prevents sigwait()'s use with the
3261 // mutex granting granting signal.
3262 //
3263 // Currently only ever called on the VMThread
3264 //
3265 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3266 // Save and restore errno to avoid confusing native code with EINTR
3267 // after sigsuspend.
3268 int old_errno = errno;
3269
3270 Thread* thread = Thread::current();
3271 OSThread* osthread = thread->osthread();
3272 assert(thread->is_VM_thread(), "Must be VMThread");
3273 // read current suspend action
3274 int action = osthread->sr.suspend_action();
3275 if (action == SR_SUSPEND) {
3276 suspend_save_context(osthread, siginfo, context);
3277
3278 // Notify the suspend action is about to be completed. do_suspend()
3279 // waits until SR_SUSPENDED is set and then returns. We will wait
3280 // here for a resume signal and that completes the suspend-other
3281 // action. do_suspend/do_resume is always called as a pair from
3282 // the same thread - so there are no races
3283
3284 // notify the caller
3285 osthread->sr.set_suspended();
3286
3287 sigset_t suspend_set; // signals for sigsuspend()
3288
3289 // get current set of blocked signals and unblock resume signal
3290 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3291 sigdelset(&suspend_set, SR_signum);
3292
3293 // wait here until we are resumed
3294 do {
3295 sigsuspend(&suspend_set);
3296 // ignore all returns until we get a resume signal
3297 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3298
3299 resume_clear_context(osthread);
3300
3301 } else {
3302 assert(action == SR_CONTINUE, "unexpected sr action");
3303 // nothing special to do - just leave the handler
3304 }
3305
3306 errno = old_errno;
3307 }
3308
3309
3310 static int SR_initialize() {
3311 struct sigaction act;
3312 char *s;
3313 /* Get signal number to use for suspend/resume */
3314 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3315 int sig = ::strtol(s, 0, 10);
3316 if (sig > 0 || sig < _NSIG) {
3317 SR_signum = sig;
3318 }
3319 }
3320
3321 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3322 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3323
3324 sigemptyset(&SR_sigset);
3325 sigaddset(&SR_sigset, SR_signum);
3326
3327 /* Set up signal handler for suspend/resume */
3328 act.sa_flags = SA_RESTART|SA_SIGINFO;
3329 act.sa_handler = (void (*)(int)) SR_handler;
3330
3331 // SR_signum is blocked by default.
3332 // 4528190 - We also need to block pthread restart signal (32 on all
3333 // supported Linux platforms). Note that LinuxThreads need to block
3334 // this signal for all threads to work properly. So we don't have
3335 // to use hard-coded signal number when setting up the mask.
3336 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3337
3338 if (sigaction(SR_signum, &act, 0) == -1) {
3339 return -1;
3340 }
3341
3342 // Save signal flag
3343 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3344 return 0;
3345 }
3346
3347 static int SR_finalize() {
3348 return 0;
3349 }
3350
3351
3352 // returns true on success and false on error - really an error is fatal
3353 // but this seems the normal response to library errors
3354 static bool do_suspend(OSThread* osthread) {
3355 // mark as suspended and send signal
3356 osthread->sr.set_suspend_action(SR_SUSPEND);
3357 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3358 assert_status(status == 0, status, "pthread_kill");
3359
3360 // check status and wait until notified of suspension
3361 if (status == 0) {
3362 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3363 os::yield_all(i);
3364 }
3365 osthread->sr.set_suspend_action(SR_NONE);
3366 return true;
3367 }
3368 else {
3369 osthread->sr.set_suspend_action(SR_NONE);
3370 return false;
3371 }
3372 }
3373
3374 static void do_resume(OSThread* osthread) {
3375 assert(osthread->sr.is_suspended(), "thread should be suspended");
3376 osthread->sr.set_suspend_action(SR_CONTINUE);
3377
3378 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3379 assert_status(status == 0, status, "pthread_kill");
3380 // check status and wait unit notified of resumption
3381 if (status == 0) {
3382 for (int i = 0; osthread->sr.is_suspended(); i++) {
3383 os::yield_all(i);
3384 }
3385 }
3386 osthread->sr.set_suspend_action(SR_NONE);
3387 }
3388
3389 ////////////////////////////////////////////////////////////////////////////////
3390 // interrupt support
3391
3392 void os::interrupt(Thread* thread) {
3393 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3394 "possibility of dangling Thread pointer");
3395
3396 OSThread* osthread = thread->osthread();
3397
3398 if (!osthread->interrupted()) {
3399 osthread->set_interrupted(true);
3400 // More than one thread can get here with the same value of osthread,
3401 // resulting in multiple notifications. We do, however, want the store
3402 // to interrupted() to be visible to other threads before we execute unpark().
3403 OrderAccess::fence();
3404 ParkEvent * const slp = thread->_SleepEvent ;
3405 if (slp != NULL) slp->unpark() ;
3406 }
3407
3408 // For JSR166. Unpark even if interrupt status already was set
3409 if (thread->is_Java_thread())
3410 ((JavaThread*)thread)->parker()->unpark();
3411
3412 ParkEvent * ev = thread->_ParkEvent ;
3413 if (ev != NULL) ev->unpark() ;
3414
3415 }
3416
3417 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3418 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3419 "possibility of dangling Thread pointer");
3420
3421 OSThread* osthread = thread->osthread();
3422
3423 bool interrupted = osthread->interrupted();
3424
3425 if (interrupted && clear_interrupted) {
3426 osthread->set_interrupted(false);
3427 // consider thread->_SleepEvent->reset() ... optional optimization
3428 }
3429
3430 return interrupted;
3431 }
3432
3433 ///////////////////////////////////////////////////////////////////////////////////
3434 // signal handling (except suspend/resume)
3435
3436 // This routine may be used by user applications as a "hook" to catch signals.
3437 // The user-defined signal handler must pass unrecognized signals to this
3438 // routine, and if it returns true (non-zero), then the signal handler must
3439 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3440 // routine will never retun false (zero), but instead will execute a VM panic
3441 // routine kill the process.
3442 //
3443 // If this routine returns false, it is OK to call it again. This allows
3444 // the user-defined signal handler to perform checks either before or after
3445 // the VM performs its own checks. Naturally, the user code would be making
3446 // a serious error if it tried to handle an exception (such as a null check
3447 // or breakpoint) that the VM was generating for its own correct operation.
3448 //
3449 // This routine may recognize any of the following kinds of signals:
3450 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3451 // It should be consulted by handlers for any of those signals.
3452 //
3453 // The caller of this routine must pass in the three arguments supplied
3454 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3455 // field of the structure passed to sigaction(). This routine assumes that
3456 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3457 //
3458 // Note that the VM will print warnings if it detects conflicting signal
3459 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3460 //
3461 extern "C" int
3462 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3463 void* ucontext, int abort_if_unrecognized);
3464
3465 void signalHandler(int sig, siginfo_t* info, void* uc) {
3466 assert(info != NULL && uc != NULL, "it must be old kernel");
3467 JVM_handle_linux_signal(sig, info, uc, true);
3468 }
3469
3470
3471 // This boolean allows users to forward their own non-matching signals
3472 // to JVM_handle_linux_signal, harmlessly.
3473 bool os::Linux::signal_handlers_are_installed = false;
3474
3475 // For signal-chaining
3476 struct sigaction os::Linux::sigact[MAXSIGNUM];
3477 unsigned int os::Linux::sigs = 0;
3478 bool os::Linux::libjsig_is_loaded = false;
3479 typedef struct sigaction *(*get_signal_t)(int);
3480 get_signal_t os::Linux::get_signal_action = NULL;
3481
3482 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3483 struct sigaction *actp = NULL;
3484
3485 if (libjsig_is_loaded) {
3486 // Retrieve the old signal handler from libjsig
3487 actp = (*get_signal_action)(sig);
3488 }
3489 if (actp == NULL) {
3490 // Retrieve the preinstalled signal handler from jvm
3491 actp = get_preinstalled_handler(sig);
3492 }
3493
3494 return actp;
3495 }
3496
3497 static bool call_chained_handler(struct sigaction *actp, int sig,
3498 siginfo_t *siginfo, void *context) {
3499 // Call the old signal handler
3500 if (actp->sa_handler == SIG_DFL) {
3501 // It's more reasonable to let jvm treat it as an unexpected exception
3502 // instead of taking the default action.
3503 return false;
3504 } else if (actp->sa_handler != SIG_IGN) {
3505 if ((actp->sa_flags & SA_NODEFER) == 0) {
3506 // automaticlly block the signal
3507 sigaddset(&(actp->sa_mask), sig);
3508 }
3509
3510 sa_handler_t hand;
3511 sa_sigaction_t sa;
3512 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3513 // retrieve the chained handler
3514 if (siginfo_flag_set) {
3515 sa = actp->sa_sigaction;
3516 } else {
3517 hand = actp->sa_handler;
3518 }
3519
3520 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3521 actp->sa_handler = SIG_DFL;
3522 }
3523
3524 // try to honor the signal mask
3525 sigset_t oset;
3526 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3527
3528 // call into the chained handler
3529 if (siginfo_flag_set) {
3530 (*sa)(sig, siginfo, context);
3531 } else {
3532 (*hand)(sig);
3533 }
3534
3535 // restore the signal mask
3536 pthread_sigmask(SIG_SETMASK, &oset, 0);
3537 }
3538 // Tell jvm's signal handler the signal is taken care of.
3539 return true;
3540 }
3541
3542 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3543 bool chained = false;
3544 // signal-chaining
3545 if (UseSignalChaining) {
3546 struct sigaction *actp = get_chained_signal_action(sig);
3547 if (actp != NULL) {
3548 chained = call_chained_handler(actp, sig, siginfo, context);
3549 }
3550 }
3551 return chained;
3552 }
3553
3554 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3555 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3556 return &sigact[sig];
3557 }
3558 return NULL;
3559 }
3560
3561 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3562 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3563 sigact[sig] = oldAct;
3564 sigs |= (unsigned int)1 << sig;
3565 }
3566
3567 // for diagnostic
3568 int os::Linux::sigflags[MAXSIGNUM];
3569
3570 int os::Linux::get_our_sigflags(int sig) {
3571 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3572 return sigflags[sig];
3573 }
3574
3575 void os::Linux::set_our_sigflags(int sig, int flags) {
3576 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3577 sigflags[sig] = flags;
3578 }
3579
3580 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3581 // Check for overwrite.
3582 struct sigaction oldAct;
3583 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3584
3585 void* oldhand = oldAct.sa_sigaction
3586 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3587 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3588 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3589 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3590 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3591 if (AllowUserSignalHandlers || !set_installed) {
3592 // Do not overwrite; user takes responsibility to forward to us.
3593 return;
3594 } else if (UseSignalChaining) {
3595 // save the old handler in jvm
3596 save_preinstalled_handler(sig, oldAct);
3597 // libjsig also interposes the sigaction() call below and saves the
3598 // old sigaction on it own.
3599 } else {
3600 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3601 "%#lx for signal %d.", (long)oldhand, sig));
3602 }
3603 }
3604
3605 struct sigaction sigAct;
3606 sigfillset(&(sigAct.sa_mask));
3607 sigAct.sa_handler = SIG_DFL;
3608 if (!set_installed) {
3609 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3610 } else {
3611 sigAct.sa_sigaction = signalHandler;
3612 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3613 }
3614 // Save flags, which are set by ours
3615 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3616 sigflags[sig] = sigAct.sa_flags;
3617
3618 int ret = sigaction(sig, &sigAct, &oldAct);
3619 assert(ret == 0, "check");
3620
3621 void* oldhand2 = oldAct.sa_sigaction
3622 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3623 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3624 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3625 }
3626
3627 // install signal handlers for signals that HotSpot needs to
3628 // handle in order to support Java-level exception handling.
3629
3630 void os::Linux::install_signal_handlers() {
3631 if (!signal_handlers_are_installed) {
3632 signal_handlers_are_installed = true;
3633
3634 // signal-chaining
3635 typedef void (*signal_setting_t)();
3636 signal_setting_t begin_signal_setting = NULL;
3637 signal_setting_t end_signal_setting = NULL;
3638 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3639 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3640 if (begin_signal_setting != NULL) {
3641 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3642 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3643 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3644 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3645 libjsig_is_loaded = true;
3646 assert(UseSignalChaining, "should enable signal-chaining");
3647 }
3648 if (libjsig_is_loaded) {
3649 // Tell libjsig jvm is setting signal handlers
3650 (*begin_signal_setting)();
3651 }
3652
3653 set_signal_handler(SIGSEGV, true);
3654 set_signal_handler(SIGPIPE, true);
3655 set_signal_handler(SIGBUS, true);
3656 set_signal_handler(SIGILL, true);
3657 set_signal_handler(SIGFPE, true);
3658 set_signal_handler(SIGXFSZ, true);
3659
3660 if (libjsig_is_loaded) {
3661 // Tell libjsig jvm finishes setting signal handlers
3662 (*end_signal_setting)();
3663 }
3664
3665 // We don't activate signal checker if libjsig is in place, we trust ourselves
3666 // and if UserSignalHandler is installed all bets are off
3667 if (CheckJNICalls) {
3668 if (libjsig_is_loaded) {
3669 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3670 check_signals = false;
3671 }
3672 if (AllowUserSignalHandlers) {
3673 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3674 check_signals = false;
3675 }
3676 }
3677 }
3678 }
3679
3680 // This is the fastest way to get thread cpu time on Linux.
3681 // Returns cpu time (user+sys) for any thread, not only for current.
3682 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3683 // It might work on 2.6.10+ with a special kernel/glibc patch.
3684 // For reference, please, see IEEE Std 1003.1-2004:
3685 // http://www.unix.org/single_unix_specification
3686
3687 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3688 struct timespec tp;
3689 int rc = os::Linux::clock_gettime(clockid, &tp);
3690 assert(rc == 0, "clock_gettime is expected to return 0 code");
3691
3692 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3693 }
3694
3695 /////
3696 // glibc on Linux platform uses non-documented flag
3697 // to indicate, that some special sort of signal
3698 // trampoline is used.
3699 // We will never set this flag, and we should
3700 // ignore this flag in our diagnostic
3701 #ifdef SIGNIFICANT_SIGNAL_MASK
3702 #undef SIGNIFICANT_SIGNAL_MASK
3703 #endif
3704 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3705
3706 static const char* get_signal_handler_name(address handler,
3707 char* buf, int buflen) {
3708 int offset;
3709 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3710 if (found) {
3711 // skip directory names
3712 const char *p1, *p2;
3713 p1 = buf;
3714 size_t len = strlen(os::file_separator());
3715 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3716 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3717 } else {
3718 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3719 }
3720 return buf;
3721 }
3722
3723 static void print_signal_handler(outputStream* st, int sig,
3724 char* buf, size_t buflen) {
3725 struct sigaction sa;
3726
3727 sigaction(sig, NULL, &sa);
3728
3729 // See comment for SIGNIFICANT_SIGNAL_MASK define
3730 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3731
3732 st->print("%s: ", os::exception_name(sig, buf, buflen));
3733
3734 address handler = (sa.sa_flags & SA_SIGINFO)
3735 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3736 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3737
3738 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3739 st->print("SIG_DFL");
3740 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3741 st->print("SIG_IGN");
3742 } else {
3743 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3744 }
3745
3746 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3747
3748 address rh = VMError::get_resetted_sighandler(sig);
3749 // May be, handler was resetted by VMError?
3750 if(rh != NULL) {
3751 handler = rh;
3752 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3753 }
3754
3755 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3756
3757 // Check: is it our handler?
3758 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3759 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3760 // It is our signal handler
3761 // check for flags, reset system-used one!
3762 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3763 st->print(
3764 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3765 os::Linux::get_our_sigflags(sig));
3766 }
3767 }
3768 st->cr();
3769 }
3770
3771
3772 #define DO_SIGNAL_CHECK(sig) \
3773 if (!sigismember(&check_signal_done, sig)) \
3774 os::Linux::check_signal_handler(sig)
3775
3776 // This method is a periodic task to check for misbehaving JNI applications
3777 // under CheckJNI, we can add any periodic checks here
3778
3779 void os::run_periodic_checks() {
3780
3781 if (check_signals == false) return;
3782
3783 // SEGV and BUS if overridden could potentially prevent
3784 // generation of hs*.log in the event of a crash, debugging
3785 // such a case can be very challenging, so we absolutely
3786 // check the following for a good measure:
3787 DO_SIGNAL_CHECK(SIGSEGV);
3788 DO_SIGNAL_CHECK(SIGILL);
3789 DO_SIGNAL_CHECK(SIGFPE);
3790 DO_SIGNAL_CHECK(SIGBUS);
3791 DO_SIGNAL_CHECK(SIGPIPE);
3792 DO_SIGNAL_CHECK(SIGXFSZ);
3793
3794
3795 // ReduceSignalUsage allows the user to override these handlers
3796 // see comments at the very top and jvm_solaris.h
3797 if (!ReduceSignalUsage) {
3798 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3799 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3800 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3801 DO_SIGNAL_CHECK(BREAK_SIGNAL);
3802 }
3803
3804 DO_SIGNAL_CHECK(SR_signum);
3805 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3806 }
3807
3808 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3809
3810 static os_sigaction_t os_sigaction = NULL;
3811
3812 void os::Linux::check_signal_handler(int sig) {
3813 char buf[O_BUFLEN];
3814 address jvmHandler = NULL;
3815
3816
3817 struct sigaction act;
3818 if (os_sigaction == NULL) {
3819 // only trust the default sigaction, in case it has been interposed
3820 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3821 if (os_sigaction == NULL) return;
3822 }
3823
3824 os_sigaction(sig, (struct sigaction*)NULL, &act);
3825
3826
3827 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3828
3829 address thisHandler = (act.sa_flags & SA_SIGINFO)
3830 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3831 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3832
3833
3834 switch(sig) {
3835 case SIGSEGV:
3836 case SIGBUS:
3837 case SIGFPE:
3838 case SIGPIPE:
3839 case SIGILL:
3840 case SIGXFSZ:
3841 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3842 break;
3843
3844 case SHUTDOWN1_SIGNAL:
3845 case SHUTDOWN2_SIGNAL:
3846 case SHUTDOWN3_SIGNAL:
3847 case BREAK_SIGNAL:
3848 jvmHandler = (address)user_handler();
3849 break;
3850
3851 case INTERRUPT_SIGNAL:
3852 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3853 break;
3854
3855 default:
3856 if (sig == SR_signum) {
3857 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3858 } else {
3859 return;
3860 }
3861 break;
3862 }
3863
3864 if (thisHandler != jvmHandler) {
3865 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3866 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3867 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3868 // No need to check this sig any longer
3869 sigaddset(&check_signal_done, sig);
3870 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3871 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3872 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3873 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
3874 // No need to check this sig any longer
3875 sigaddset(&check_signal_done, sig);
3876 }
3877
3878 // Dump all the signal
3879 if (sigismember(&check_signal_done, sig)) {
3880 print_signal_handlers(tty, buf, O_BUFLEN);
3881 }
3882 }
3883
3884 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3885
3886 extern bool signal_name(int signo, char* buf, size_t len);
3887
3888 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3889 if (0 < exception_code && exception_code <= SIGRTMAX) {
3890 // signal
3891 if (!signal_name(exception_code, buf, size)) {
3892 jio_snprintf(buf, size, "SIG%d", exception_code);
3893 }
3894 return buf;
3895 } else {
3896 return NULL;
3897 }
3898 }
3899
3900 // this is called _before_ the most of global arguments have been parsed
3901 void os::init(void) {
3902 char dummy; /* used to get a guess on initial stack address */
3903 // first_hrtime = gethrtime();
3904
3905 // With LinuxThreads the JavaMain thread pid (primordial thread)
3906 // is different than the pid of the java launcher thread.
3907 // So, on Linux, the launcher thread pid is passed to the VM
3908 // via the sun.java.launcher.pid property.
3909 // Use this property instead of getpid() if it was correctly passed.
3910 // See bug 6351349.
3911 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3912
3913 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3914
3915 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3916
3917 init_random(1234567);
3918
3919 ThreadCritical::initialize();
3920
3921 Linux::set_page_size(sysconf(_SC_PAGESIZE));
3922 if (Linux::page_size() == -1) {
3923 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
3924 strerror(errno)));
3925 }
3926 init_page_sizes((size_t) Linux::page_size());
3927
3928 Linux::initialize_system_info();
3929
3930 // main_thread points to the aboriginal thread
3931 Linux::_main_thread = pthread_self();
3932
3933 Linux::clock_init();
3934 initial_time_count = os::elapsed_counter();
3935 pthread_mutex_init(&dl_mutex, NULL);
3936 }
3937
3938 // To install functions for atexit system call
3939 extern "C" {
3940 static void perfMemory_exit_helper() {
3941 perfMemory_exit();
3942 }
3943 }
3944
3945 // this is called _after_ the global arguments have been parsed
3946 jint os::init_2(void)
3947 {
3948 Linux::fast_thread_clock_init();
3949
3950 // Allocate a single page and mark it as readable for safepoint polling
3951 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3952 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3953
3954 os::set_polling_page( polling_page );
3955
3956 #ifndef PRODUCT
3957 if(Verbose && PrintMiscellaneous)
3958 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3959 #endif
3960
3961 if (!UseMembar) {
3962 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3963 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3964 os::set_memory_serialize_page( mem_serialize_page );
3965
3966 #ifndef PRODUCT
3967 if(Verbose && PrintMiscellaneous)
3968 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3969 #endif
3970 }
3971
3972 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3973
3974 // initialize suspend/resume support - must do this before signal_sets_init()
3975 if (SR_initialize() != 0) {
3976 perror("SR_initialize failed");
3977 return JNI_ERR;
3978 }
3979
3980 Linux::signal_sets_init();
3981 Linux::install_signal_handlers();
3982
3983 // Check minimum allowable stack size for thread creation and to initialize
3984 // the java system classes, including StackOverflowError - depends on page
3985 // size. Add a page for compiler2 recursion in main thread.
3986 // Add in 2*BytesPerWord times page size to account for VM stack during
3987 // class initialization depending on 32 or 64 bit VM.
3988 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
3989 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
3990 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
3991
3992 size_t threadStackSizeInBytes = ThreadStackSize * K;
3993 if (threadStackSizeInBytes != 0 &&
3994 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
3995 tty->print_cr("\nThe stack size specified is too small, "
3996 "Specify at least %dk",
3997 os::Linux::min_stack_allowed/ K);
3998 return JNI_ERR;
3999 }
4000
4001 // Make the stack size a multiple of the page size so that
4002 // the yellow/red zones can be guarded.
4003 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4004 vm_page_size()));
4005
4006 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4007
4008 Linux::libpthread_init();
4009 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4010 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4011 Linux::glibc_version(), Linux::libpthread_version(),
4012 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4013 }
4014
4015 if (UseNUMA) {
4016 if (!Linux::libnuma_init()) {
4017 UseNUMA = false;
4018 } else {
4019 if ((Linux::numa_max_node() < 1)) {
4020 // There's only one node(they start from 0), disable NUMA.
4021 UseNUMA = false;
4022 }
4023 }
4024 if (!UseNUMA && ForceNUMA) {
4025 UseNUMA = true;
4026 }
4027 }
4028
4029 if (MaxFDLimit) {
4030 // set the number of file descriptors to max. print out error
4031 // if getrlimit/setrlimit fails but continue regardless.
4032 struct rlimit nbr_files;
4033 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4034 if (status != 0) {
4035 if (PrintMiscellaneous && (Verbose || WizardMode))
4036 perror("os::init_2 getrlimit failed");
4037 } else {
4038 nbr_files.rlim_cur = nbr_files.rlim_max;
4039 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4040 if (status != 0) {
4041 if (PrintMiscellaneous && (Verbose || WizardMode))
4042 perror("os::init_2 setrlimit failed");
4043 }
4044 }
4045 }
4046
4047 // Initialize lock used to serialize thread creation (see os::create_thread)
4048 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4049
4050 // Initialize HPI.
4051 jint hpi_result = hpi::initialize();
4052 if (hpi_result != JNI_OK) {
4053 tty->print_cr("There was an error trying to initialize the HPI library.");
4054 return hpi_result;
4055 }
4056
4057 // at-exit methods are called in the reverse order of their registration.
4058 // atexit functions are called on return from main or as a result of a
4059 // call to exit(3C). There can be only 32 of these functions registered
4060 // and atexit() does not set errno.
4061
4062 if (PerfAllowAtExitRegistration) {
4063 // only register atexit functions if PerfAllowAtExitRegistration is set.
4064 // atexit functions can be delayed until process exit time, which
4065 // can be problematic for embedded VM situations. Embedded VMs should
4066 // call DestroyJavaVM() to assure that VM resources are released.
4067
4068 // note: perfMemory_exit_helper atexit function may be removed in
4069 // the future if the appropriate cleanup code can be added to the
4070 // VM_Exit VMOperation's doit method.
4071 if (atexit(perfMemory_exit_helper) != 0) {
4072 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4073 }
4074 }
4075
4076 // initialize thread priority policy
4077 prio_init();
4078
4079 return JNI_OK;
4080 }
4081
4082 // this is called at the end of vm_initialization
4083 void os::init_3(void) { }
4084
4085 // Mark the polling page as unreadable
4086 void os::make_polling_page_unreadable(void) {
4087 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4088 fatal("Could not disable polling page");
4089 };
4090
4091 // Mark the polling page as readable
4092 void os::make_polling_page_readable(void) {
4093 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4094 fatal("Could not enable polling page");
4095 }
4096 };
4097
4098 int os::active_processor_count() {
4099 // Linux doesn't yet have a (official) notion of processor sets,
4100 // so just return the number of online processors.
4101 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4102 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4103 return online_cpus;
4104 }
4105
4106 bool os::distribute_processes(uint length, uint* distribution) {
4107 // Not yet implemented.
4108 return false;
4109 }
4110
4111 bool os::bind_to_processor(uint processor_id) {
4112 // Not yet implemented.
4113 return false;
4114 }
4115
4116 ///
4117
4118 // Suspends the target using the signal mechanism and then grabs the PC before
4119 // resuming the target. Used by the flat-profiler only
4120 ExtendedPC os::get_thread_pc(Thread* thread) {
4121 // Make sure that it is called by the watcher for the VMThread
4122 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4123 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4124
4125 ExtendedPC epc;
4126
4127 OSThread* osthread = thread->osthread();
4128 if (do_suspend(osthread)) {
4129 if (osthread->ucontext() != NULL) {
4130 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4131 } else {
4132 // NULL context is unexpected, double-check this is the VMThread
4133 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4134 }
4135 do_resume(osthread);
4136 }
4137 // failure means pthread_kill failed for some reason - arguably this is
4138 // a fatal problem, but such problems are ignored elsewhere
4139
4140 return epc;
4141 }
4142
4143 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4144 {
4145 if (is_NPTL()) {
4146 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4147 } else {
4148 #ifndef IA64
4149 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4150 // word back to default 64bit precision if condvar is signaled. Java
4151 // wants 53bit precision. Save and restore current value.
4152 int fpu = get_fpu_control_word();
4153 #endif // IA64
4154 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4155 #ifndef IA64
4156 set_fpu_control_word(fpu);
4157 #endif // IA64
4158 return status;
4159 }
4160 }
4161
4162 ////////////////////////////////////////////////////////////////////////////////
4163 // debug support
4164
4165 static address same_page(address x, address y) {
4166 int page_bits = -os::vm_page_size();
4167 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4168 return x;
4169 else if (x > y)
4170 return (address)(intptr_t(y) | ~page_bits) + 1;
4171 else
4172 return (address)(intptr_t(y) & page_bits);
4173 }
4174
4175 bool os::find(address addr, outputStream* st) {
4176 Dl_info dlinfo;
4177 memset(&dlinfo, 0, sizeof(dlinfo));
4178 if (dladdr(addr, &dlinfo)) {
4179 st->print(PTR_FORMAT ": ", addr);
4180 if (dlinfo.dli_sname != NULL) {
4181 st->print("%s+%#x", dlinfo.dli_sname,
4182 addr - (intptr_t)dlinfo.dli_saddr);
4183 } else if (dlinfo.dli_fname) {
4184 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4185 } else {
4186 st->print("<absolute address>");
4187 }
4188 if (dlinfo.dli_fname) {
4189 st->print(" in %s", dlinfo.dli_fname);
4190 }
4191 if (dlinfo.dli_fbase) {
4192 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4193 }
4194 st->cr();
4195
4196 if (Verbose) {
4197 // decode some bytes around the PC
4198 address begin = same_page(addr-40, addr);
4199 address end = same_page(addr+40, addr);
4200 address lowest = (address) dlinfo.dli_sname;
4201 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4202 if (begin < lowest) begin = lowest;
4203 Dl_info dlinfo2;
4204 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4205 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4206 end = (address) dlinfo2.dli_saddr;
4207 Disassembler::decode(begin, end, st);
4208 }
4209 return true;
4210 }
4211 return false;
4212 }
4213
4214 ////////////////////////////////////////////////////////////////////////////////
4215 // misc
4216
4217 // This does not do anything on Linux. This is basically a hook for being
4218 // able to use structured exception handling (thread-local exception filters)
4219 // on, e.g., Win32.
4220 void
4221 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4222 JavaCallArguments* args, Thread* thread) {
4223 f(value, method, args, thread);
4224 }
4225
4226 void os::print_statistics() {
4227 }
4228
4229 int os::message_box(const char* title, const char* message) {
4230 int i;
4231 fdStream err(defaultStream::error_fd());
4232 for (i = 0; i < 78; i++) err.print_raw("=");
4233 err.cr();
4234 err.print_raw_cr(title);
4235 for (i = 0; i < 78; i++) err.print_raw("-");
4236 err.cr();
4237 err.print_raw_cr(message);
4238 for (i = 0; i < 78; i++) err.print_raw("=");
4239 err.cr();
4240
4241 char buf[16];
4242 // Prevent process from exiting upon "read error" without consuming all CPU
4243 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4244
4245 return buf[0] == 'y' || buf[0] == 'Y';
4246 }
4247
4248 int os::stat(const char *path, struct stat *sbuf) {
4249 char pathbuf[MAX_PATH];
4250 if (strlen(path) > MAX_PATH - 1) {
4251 errno = ENAMETOOLONG;
4252 return -1;
4253 }
4254 hpi::native_path(strcpy(pathbuf, path));
4255 return ::stat(pathbuf, sbuf);
4256 }
4257
4258 bool os::check_heap(bool force) {
4259 return true;
4260 }
4261
4262 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4263 return ::vsnprintf(buf, count, format, args);
4264 }
4265
4266 // Is a (classpath) directory empty?
4267 bool os::dir_is_empty(const char* path) {
4268 DIR *dir = NULL;
4269 struct dirent *ptr;
4270
4271 dir = opendir(path);
4272 if (dir == NULL) return true;
4273
4274 /* Scan the directory */
4275 bool result = true;
4276 char buf[sizeof(struct dirent) + MAX_PATH];
4277 while (result && (ptr = ::readdir(dir)) != NULL) {
4278 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4279 result = false;
4280 }
4281 }
4282 closedir(dir);
4283 return result;
4284 }
4285
4286 // create binary file, rewriting existing file if required
4287 int os::create_binary_file(const char* path, bool rewrite_existing) {
4288 int oflags = O_WRONLY | O_CREAT;
4289 if (!rewrite_existing) {
4290 oflags |= O_EXCL;
4291 }
4292 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4293 }
4294
4295 // return current position of file pointer
4296 jlong os::current_file_offset(int fd) {
4297 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4298 }
4299
4300 // move file pointer to the specified offset
4301 jlong os::seek_to_file_offset(int fd, jlong offset) {
4302 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4303 }
4304
4305 // Map a block of memory.
4306 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4307 char *addr, size_t bytes, bool read_only,
4308 bool allow_exec) {
4309 int prot;
4310 int flags;
4311
4312 if (read_only) {
4313 prot = PROT_READ;
4314 flags = MAP_SHARED;
4315 } else {
4316 prot = PROT_READ | PROT_WRITE;
4317 flags = MAP_PRIVATE;
4318 }
4319
4320 if (allow_exec) {
4321 prot |= PROT_EXEC;
4322 }
4323
4324 if (addr != NULL) {
4325 flags |= MAP_FIXED;
4326 }
4327
4328 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4329 fd, file_offset);
4330 if (mapped_address == MAP_FAILED) {
4331 return NULL;
4332 }
4333 return mapped_address;
4334 }
4335
4336
4337 // Remap a block of memory.
4338 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4339 char *addr, size_t bytes, bool read_only,
4340 bool allow_exec) {
4341 // same as map_memory() on this OS
4342 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4343 allow_exec);
4344 }
4345
4346
4347 // Unmap a block of memory.
4348 bool os::unmap_memory(char* addr, size_t bytes) {
4349 return munmap(addr, bytes) == 0;
4350 }
4351
4352 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4353
4354 static clockid_t thread_cpu_clockid(Thread* thread) {
4355 pthread_t tid = thread->osthread()->pthread_id();
4356 clockid_t clockid;
4357
4358 // Get thread clockid
4359 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4360 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4361 return clockid;
4362 }
4363
4364 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4365 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4366 // of a thread.
4367 //
4368 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4369 // the fast estimate available on the platform.
4370
4371 jlong os::current_thread_cpu_time() {
4372 if (os::Linux::supports_fast_thread_cpu_time()) {
4373 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4374 } else {
4375 // return user + sys since the cost is the same
4376 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4377 }
4378 }
4379
4380 jlong os::thread_cpu_time(Thread* thread) {
4381 // consistent with what current_thread_cpu_time() returns
4382 if (os::Linux::supports_fast_thread_cpu_time()) {
4383 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4384 } else {
4385 return slow_thread_cpu_time(thread, true /* user + sys */);
4386 }
4387 }
4388
4389 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4390 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4391 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4392 } else {
4393 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4394 }
4395 }
4396
4397 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4398 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4399 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4400 } else {
4401 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4402 }
4403 }
4404
4405 //
4406 // -1 on error.
4407 //
4408
4409 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4410 static bool proc_pid_cpu_avail = true;
4411 static bool proc_task_unchecked = true;
4412 static const char *proc_stat_path = "/proc/%d/stat";
4413 pid_t tid = thread->osthread()->thread_id();
4414 int i;
4415 char *s;
4416 char stat[2048];
4417 int statlen;
4418 char proc_name[64];
4419 int count;
4420 long sys_time, user_time;
4421 char string[64];
4422 char cdummy;
4423 int idummy;
4424 long ldummy;
4425 FILE *fp;
4426
4427 // We first try accessing /proc/<pid>/cpu since this is faster to
4428 // process. If this file is not present (linux kernels 2.5 and above)
4429 // then we open /proc/<pid>/stat.
4430 if ( proc_pid_cpu_avail ) {
4431 sprintf(proc_name, "/proc/%d/cpu", tid);
4432 fp = fopen(proc_name, "r");
4433 if ( fp != NULL ) {
4434 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4435 fclose(fp);
4436 if ( count != 3 ) return -1;
4437
4438 if (user_sys_cpu_time) {
4439 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4440 } else {
4441 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4442 }
4443 }
4444 else proc_pid_cpu_avail = false;
4445 }
4446
4447 // The /proc/<tid>/stat aggregates per-process usage on
4448 // new Linux kernels 2.6+ where NPTL is supported.
4449 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4450 // See bug 6328462.
4451 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4452 // and possibly in some other cases, so we check its availability.
4453 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4454 // This is executed only once
4455 proc_task_unchecked = false;
4456 fp = fopen("/proc/self/task", "r");
4457 if (fp != NULL) {
4458 proc_stat_path = "/proc/self/task/%d/stat";
4459 fclose(fp);
4460 }
4461 }
4462
4463 sprintf(proc_name, proc_stat_path, tid);
4464 fp = fopen(proc_name, "r");
4465 if ( fp == NULL ) return -1;
4466 statlen = fread(stat, 1, 2047, fp);
4467 stat[statlen] = '\0';
4468 fclose(fp);
4469
4470 // Skip pid and the command string. Note that we could be dealing with
4471 // weird command names, e.g. user could decide to rename java launcher
4472 // to "java 1.4.2 :)", then the stat file would look like
4473 // 1234 (java 1.4.2 :)) R ... ...
4474 // We don't really need to know the command string, just find the last
4475 // occurrence of ")" and then start parsing from there. See bug 4726580.
4476 s = strrchr(stat, ')');
4477 i = 0;
4478 if (s == NULL ) return -1;
4479
4480 // Skip blank chars
4481 do s++; while (isspace(*s));
4482
4483 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4484 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4485 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4486 &user_time, &sys_time);
4487 if ( count != 13 ) return -1;
4488 if (user_sys_cpu_time) {
4489 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4490 } else {
4491 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4492 }
4493 }
4494
4495 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4496 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4497 info_ptr->may_skip_backward = false; // elapsed time not wall time
4498 info_ptr->may_skip_forward = false; // elapsed time not wall time
4499 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4500 }
4501
4502 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4503 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4504 info_ptr->may_skip_backward = false; // elapsed time not wall time
4505 info_ptr->may_skip_forward = false; // elapsed time not wall time
4506 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4507 }
4508
4509 bool os::is_thread_cpu_time_supported() {
4510 return true;
4511 }
4512
4513 // System loadavg support. Returns -1 if load average cannot be obtained.
4514 // Linux doesn't yet have a (official) notion of processor sets,
4515 // so just return the system wide load average.
4516 int os::loadavg(double loadavg[], int nelem) {
4517 return ::getloadavg(loadavg, nelem);
4518 }
4519
4520 void os::pause() {
4521 char filename[MAX_PATH];
4522 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4523 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4524 } else {
4525 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4526 }
4527
4528 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4529 if (fd != -1) {
4530 struct stat buf;
4531 close(fd);
4532 while (::stat(filename, &buf) == 0) {
4533 (void)::poll(NULL, 0, 100);
4534 }
4535 } else {
4536 jio_fprintf(stderr,
4537 "Could not open pause file '%s', continuing immediately.\n", filename);
4538 }
4539 }
4540
4541 extern "C" {
4542
4543 /**
4544 * NOTE: the following code is to keep the green threads code
4545 * in the libjava.so happy. Once the green threads is removed,
4546 * these code will no longer be needed.
4547 */
4548 int
4549 jdk_waitpid(pid_t pid, int* status, int options) {
4550 return waitpid(pid, status, options);
4551 }
4552
4553 int
4554 fork1() {
4555 return fork();
4556 }
4557
4558 int
4559 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4560 return sem_init(sem, pshared, value);
4561 }
4562
4563 int
4564 jdk_sem_post(sem_t *sem) {
4565 return sem_post(sem);
4566 }
4567
4568 int
4569 jdk_sem_wait(sem_t *sem) {
4570 return sem_wait(sem);
4571 }
4572
4573 int
4574 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4575 return pthread_sigmask(how , newmask, oldmask);
4576 }
4577
4578 }
4579
4580 // Refer to the comments in os_solaris.cpp park-unpark.
4581 //
4582 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4583 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4584 // For specifics regarding the bug see GLIBC BUGID 261237 :
4585 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4586 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4587 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4588 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4589 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4590 // and monitorenter when we're using 1-0 locking. All those operations may result in
4591 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4592 // of libpthread avoids the problem, but isn't practical.
4593 //
4594 // Possible remedies:
4595 //
4596 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4597 // This is palliative and probabilistic, however. If the thread is preempted
4598 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4599 // than the minimum period may have passed, and the abstime may be stale (in the
4600 // past) resultin in a hang. Using this technique reduces the odds of a hang
4601 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4602 //
4603 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4604 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4605 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4606 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4607 // thread.
4608 //
4609 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4610 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4611 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4612 // This also works well. In fact it avoids kernel-level scalability impediments
4613 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4614 // timers in a graceful fashion.
4615 //
4616 // 4. When the abstime value is in the past it appears that control returns
4617 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4618 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4619 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4620 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4621 // It may be possible to avoid reinitialization by checking the return
4622 // value from pthread_cond_timedwait(). In addition to reinitializing the
4623 // condvar we must establish the invariant that cond_signal() is only called
4624 // within critical sections protected by the adjunct mutex. This prevents
4625 // cond_signal() from "seeing" a condvar that's in the midst of being
4626 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4627 // desirable signal-after-unlock optimization that avoids futile context switching.
4628 //
4629 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4630 // structure when a condvar is used or initialized. cond_destroy() would
4631 // release the helper structure. Our reinitialize-after-timedwait fix
4632 // put excessive stress on malloc/free and locks protecting the c-heap.
4633 //
4634 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4635 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4636 // and only enabling the work-around for vulnerable environments.
4637
4638 // utility to compute the abstime argument to timedwait:
4639 // millis is the relative timeout time
4640 // abstime will be the absolute timeout time
4641 // TODO: replace compute_abstime() with unpackTime()
4642
4643 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4644 if (millis < 0) millis = 0;
4645 struct timeval now;
4646 int status = gettimeofday(&now, NULL);
4647 assert(status == 0, "gettimeofday");
4648 jlong seconds = millis / 1000;
4649 millis %= 1000;
4650 if (seconds > 50000000) { // see man cond_timedwait(3T)
4651 seconds = 50000000;
4652 }
4653 abstime->tv_sec = now.tv_sec + seconds;
4654 long usec = now.tv_usec + millis * 1000;
4655 if (usec >= 1000000) {
4656 abstime->tv_sec += 1;
4657 usec -= 1000000;
4658 }
4659 abstime->tv_nsec = usec * 1000;
4660 return abstime;
4661 }
4662
4663
4664 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4665 // Conceptually TryPark() should be equivalent to park(0).
4666
4667 int os::PlatformEvent::TryPark() {
4668 for (;;) {
4669 const int v = _Event ;
4670 guarantee ((v == 0) || (v == 1), "invariant") ;
4671 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4672 }
4673 }
4674
4675 void os::PlatformEvent::park() { // AKA "down()"
4676 // Invariant: Only the thread associated with the Event/PlatformEvent
4677 // may call park().
4678 // TODO: assert that _Assoc != NULL or _Assoc == Self
4679 int v ;
4680 for (;;) {
4681 v = _Event ;
4682 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4683 }
4684 guarantee (v >= 0, "invariant") ;
4685 if (v == 0) {
4686 // Do this the hard way by blocking ...
4687 int status = pthread_mutex_lock(_mutex);
4688 assert_status(status == 0, status, "mutex_lock");
4689 guarantee (_nParked == 0, "invariant") ;
4690 ++ _nParked ;
4691 while (_Event < 0) {
4692 status = pthread_cond_wait(_cond, _mutex);
4693 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4694 // Treat this the same as if the wait was interrupted
4695 if (status == ETIME) { status = EINTR; }
4696 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4697 }
4698 -- _nParked ;
4699
4700 // In theory we could move the ST of 0 into _Event past the unlock(),
4701 // but then we'd need a MEMBAR after the ST.
4702 _Event = 0 ;
4703 status = pthread_mutex_unlock(_mutex);
4704 assert_status(status == 0, status, "mutex_unlock");
4705 }
4706 guarantee (_Event >= 0, "invariant") ;
4707 }
4708
4709 int os::PlatformEvent::park(jlong millis) {
4710 guarantee (_nParked == 0, "invariant") ;
4711
4712 int v ;
4713 for (;;) {
4714 v = _Event ;
4715 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4716 }
4717 guarantee (v >= 0, "invariant") ;
4718 if (v != 0) return OS_OK ;
4719
4720 // We do this the hard way, by blocking the thread.
4721 // Consider enforcing a minimum timeout value.
4722 struct timespec abst;
4723 compute_abstime(&abst, millis);
4724
4725 int ret = OS_TIMEOUT;
4726 int status = pthread_mutex_lock(_mutex);
4727 assert_status(status == 0, status, "mutex_lock");
4728 guarantee (_nParked == 0, "invariant") ;
4729 ++_nParked ;
4730
4731 // Object.wait(timo) will return because of
4732 // (a) notification
4733 // (b) timeout
4734 // (c) thread.interrupt
4735 //
4736 // Thread.interrupt and object.notify{All} both call Event::set.
4737 // That is, we treat thread.interrupt as a special case of notification.
4738 // The underlying Solaris implementation, cond_timedwait, admits
4739 // spurious/premature wakeups, but the JLS/JVM spec prevents the
4740 // JVM from making those visible to Java code. As such, we must
4741 // filter out spurious wakeups. We assume all ETIME returns are valid.
4742 //
4743 // TODO: properly differentiate simultaneous notify+interrupt.
4744 // In that case, we should propagate the notify to another waiter.
4745
4746 while (_Event < 0) {
4747 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4748 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4749 pthread_cond_destroy (_cond);
4750 pthread_cond_init (_cond, NULL) ;
4751 }
4752 assert_status(status == 0 || status == EINTR ||
4753 status == ETIME || status == ETIMEDOUT,
4754 status, "cond_timedwait");
4755 if (!FilterSpuriousWakeups) break ; // previous semantics
4756 if (status == ETIME || status == ETIMEDOUT) break ;
4757 // We consume and ignore EINTR and spurious wakeups.
4758 }
4759 --_nParked ;
4760 if (_Event >= 0) {
4761 ret = OS_OK;
4762 }
4763 _Event = 0 ;
4764 status = pthread_mutex_unlock(_mutex);
4765 assert_status(status == 0, status, "mutex_unlock");
4766 assert (_nParked == 0, "invariant") ;
4767 return ret;
4768 }
4769
4770 void os::PlatformEvent::unpark() {
4771 int v, AnyWaiters ;
4772 for (;;) {
4773 v = _Event ;
4774 if (v > 0) {
4775 // The LD of _Event could have reordered or be satisfied
4776 // by a read-aside from this processor's write buffer.
4777 // To avoid problems execute a barrier and then
4778 // ratify the value.
4779 OrderAccess::fence() ;
4780 if (_Event == v) return ;
4781 continue ;
4782 }
4783 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4784 }
4785 if (v < 0) {
4786 // Wait for the thread associated with the event to vacate
4787 int status = pthread_mutex_lock(_mutex);
4788 assert_status(status == 0, status, "mutex_lock");
4789 AnyWaiters = _nParked ;
4790 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4791 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4792 AnyWaiters = 0 ;
4793 pthread_cond_signal (_cond);
4794 }
4795 status = pthread_mutex_unlock(_mutex);
4796 assert_status(status == 0, status, "mutex_unlock");
4797 if (AnyWaiters != 0) {
4798 status = pthread_cond_signal(_cond);
4799 assert_status(status == 0, status, "cond_signal");
4800 }
4801 }
4802
4803 // Note that we signal() _after dropping the lock for "immortal" Events.
4804 // This is safe and avoids a common class of futile wakeups. In rare
4805 // circumstances this can cause a thread to return prematurely from
4806 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4807 // simply re-test the condition and re-park itself.
4808 }
4809
4810
4811 // JSR166
4812 // -------------------------------------------------------
4813
4814 /*
4815 * The solaris and linux implementations of park/unpark are fairly
4816 * conservative for now, but can be improved. They currently use a
4817 * mutex/condvar pair, plus a a count.
4818 * Park decrements count if > 0, else does a condvar wait. Unpark
4819 * sets count to 1 and signals condvar. Only one thread ever waits
4820 * on the condvar. Contention seen when trying to park implies that someone
4821 * is unparking you, so don't wait. And spurious returns are fine, so there
4822 * is no need to track notifications.
4823 */
4824
4825
4826 #define NANOSECS_PER_SEC 1000000000
4827 #define NANOSECS_PER_MILLISEC 1000000
4828 #define MAX_SECS 100000000
4829 /*
4830 * This code is common to linux and solaris and will be moved to a
4831 * common place in dolphin.
4832 *
4833 * The passed in time value is either a relative time in nanoseconds
4834 * or an absolute time in milliseconds. Either way it has to be unpacked
4835 * into suitable seconds and nanoseconds components and stored in the
4836 * given timespec structure.
4837 * Given time is a 64-bit value and the time_t used in the timespec is only
4838 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4839 * overflow if times way in the future are given. Further on Solaris versions
4840 * prior to 10 there is a restriction (see cond_timedwait) that the specified
4841 * number of seconds, in abstime, is less than current_time + 100,000,000.
4842 * As it will be 28 years before "now + 100000000" will overflow we can
4843 * ignore overflow and just impose a hard-limit on seconds using the value
4844 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4845 * years from "now".
4846 */
4847
4848 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4849 assert (time > 0, "convertTime");
4850
4851 struct timeval now;
4852 int status = gettimeofday(&now, NULL);
4853 assert(status == 0, "gettimeofday");
4854
4855 time_t max_secs = now.tv_sec + MAX_SECS;
4856
4857 if (isAbsolute) {
4858 jlong secs = time / 1000;
4859 if (secs > max_secs) {
4860 absTime->tv_sec = max_secs;
4861 }
4862 else {
4863 absTime->tv_sec = secs;
4864 }
4865 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4866 }
4867 else {
4868 jlong secs = time / NANOSECS_PER_SEC;
4869 if (secs >= MAX_SECS) {
4870 absTime->tv_sec = max_secs;
4871 absTime->tv_nsec = 0;
4872 }
4873 else {
4874 absTime->tv_sec = now.tv_sec + secs;
4875 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4876 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4877 absTime->tv_nsec -= NANOSECS_PER_SEC;
4878 ++absTime->tv_sec; // note: this must be <= max_secs
4879 }
4880 }
4881 }
4882 assert(absTime->tv_sec >= 0, "tv_sec < 0");
4883 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4884 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4885 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4886 }
4887
4888 void Parker::park(bool isAbsolute, jlong time) {
4889 // Optional fast-path check:
4890 // Return immediately if a permit is available.
4891 if (_counter > 0) {
4892 _counter = 0 ;
4893 OrderAccess::fence();
4894 return ;
4895 }
4896
4897 Thread* thread = Thread::current();
4898 assert(thread->is_Java_thread(), "Must be JavaThread");
4899 JavaThread *jt = (JavaThread *)thread;
4900
4901 // Optional optimization -- avoid state transitions if there's an interrupt pending.
4902 // Check interrupt before trying to wait
4903 if (Thread::is_interrupted(thread, false)) {
4904 return;
4905 }
4906
4907 // Next, demultiplex/decode time arguments
4908 timespec absTime;
4909 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
4910 return;
4911 }
4912 if (time > 0) {
4913 unpackTime(&absTime, isAbsolute, time);
4914 }
4915
4916
4917 // Enter safepoint region
4918 // Beware of deadlocks such as 6317397.
4919 // The per-thread Parker:: mutex is a classic leaf-lock.
4920 // In particular a thread must never block on the Threads_lock while
4921 // holding the Parker:: mutex. If safepoints are pending both the
4922 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4923 ThreadBlockInVM tbivm(jt);
4924
4925 // Don't wait if cannot get lock since interference arises from
4926 // unblocking. Also. check interrupt before trying wait
4927 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4928 return;
4929 }
4930
4931 int status ;
4932 if (_counter > 0) { // no wait needed
4933 _counter = 0;
4934 status = pthread_mutex_unlock(_mutex);
4935 assert (status == 0, "invariant") ;
4936 OrderAccess::fence();
4937 return;
4938 }
4939
4940 #ifdef ASSERT
4941 // Don't catch signals while blocked; let the running threads have the signals.
4942 // (This allows a debugger to break into the running thread.)
4943 sigset_t oldsigs;
4944 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4945 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4946 #endif
4947
4948 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4949 jt->set_suspend_equivalent();
4950 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4951
4952 if (time == 0) {
4953 status = pthread_cond_wait (_cond, _mutex) ;
4954 } else {
4955 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4956 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4957 pthread_cond_destroy (_cond) ;
4958 pthread_cond_init (_cond, NULL);
4959 }
4960 }
4961 assert_status(status == 0 || status == EINTR ||
4962 status == ETIME || status == ETIMEDOUT,
4963 status, "cond_timedwait");
4964
4965 #ifdef ASSERT
4966 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4967 #endif
4968
4969 _counter = 0 ;
4970 status = pthread_mutex_unlock(_mutex) ;
4971 assert_status(status == 0, status, "invariant") ;
4972 // If externally suspended while waiting, re-suspend
4973 if (jt->handle_special_suspend_equivalent_condition()) {
4974 jt->java_suspend_self();
4975 }
4976
4977 OrderAccess::fence();
4978 }
4979
4980 void Parker::unpark() {
4981 int s, status ;
4982 status = pthread_mutex_lock(_mutex);
4983 assert (status == 0, "invariant") ;
4984 s = _counter;
4985 _counter = 1;
4986 if (s < 1) {
4987 if (WorkAroundNPTLTimedWaitHang) {
4988 status = pthread_cond_signal (_cond) ;
4989 assert (status == 0, "invariant") ;
4990 status = pthread_mutex_unlock(_mutex);
4991 assert (status == 0, "invariant") ;
4992 } else {
4993 status = pthread_mutex_unlock(_mutex);
4994 assert (status == 0, "invariant") ;
4995 status = pthread_cond_signal (_cond) ;
4996 assert (status == 0, "invariant") ;
4997 }
4998 } else {
4999 pthread_mutex_unlock(_mutex);
5000 assert (status == 0, "invariant") ;
5001 }
5002 }
5003
5004
5005 extern char** environ;
5006
5007 #ifndef __NR_fork
5008 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5009 #endif
5010
5011 #ifndef __NR_execve
5012 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5013 #endif
5014
5015 // Run the specified command in a separate process. Return its exit value,
5016 // or -1 on failure (e.g. can't fork a new process).
5017 // Unlike system(), this function can be called from signal handler. It
5018 // doesn't block SIGINT et al.
5019 int os::fork_and_exec(char* cmd) {
5020 const char * argv[4] = {"sh", "-c", cmd, NULL};
5021
5022 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5023 // pthread_atfork handlers and reset pthread library. All we need is a
5024 // separate process to execve. Make a direct syscall to fork process.
5025 // On IA64 there's no fork syscall, we have to use fork() and hope for
5026 // the best...
5027 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5028 IA64_ONLY(fork();)
5029
5030 if (pid < 0) {
5031 // fork failed
5032 return -1;
5033
5034 } else if (pid == 0) {
5035 // child process
5036
5037 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5038 // first to kill every thread on the thread list. Because this list is
5039 // not reset by fork() (see notes above), execve() will instead kill
5040 // every thread in the parent process. We know this is the only thread
5041 // in the new process, so make a system call directly.
5042 // IA64 should use normal execve() from glibc to match the glibc fork()
5043 // above.
5044 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5045 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5046
5047 // execve failed
5048 _exit(-1);
5049
5050 } else {
5051 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5052 // care about the actual exit code, for now.
5053
5054 int status;
5055
5056 // Wait for the child process to exit. This returns immediately if
5057 // the child has already exited. */
5058 while (waitpid(pid, &status, 0) < 0) {
5059 switch (errno) {
5060 case ECHILD: return 0;
5061 case EINTR: break;
5062 default: return -1;
5063 }
5064 }
5065
5066 if (WIFEXITED(status)) {
5067 // The child exited normally; get its exit code.
5068 return WEXITSTATUS(status);
5069 } else if (WIFSIGNALED(status)) {
5070 // The child exited because of a signal
5071 // The best value to return is 0x80 + signal number,
5072 // because that is what all Unix shells do, and because
5073 // it allows callers to distinguish between process exit and
5074 // process death by signal.
5075 return 0x80 + WTERMSIG(status);
5076 } else {
5077 // Unknown exit code; pass it through
5078 return status;
5079 }
5080 }
5081 }
5082
5083 // is_headless_jre()
5084 //
5085 // Test for the existence of libmawt in motif21 or xawt directories
5086 // in order to report if we are running in a headless jre
5087 //
5088 bool os::is_headless_jre() {
5089 struct stat statbuf;
5090 char buf[MAXPATHLEN];
5091 char libmawtpath[MAXPATHLEN];
5092 const char *xawtstr = "/xawt/libmawt.so";
5093 const char *motifstr = "/motif21/libmawt.so";
5094 char *p;
5095
5096 // Get path to libjvm.so
5097 os::jvm_path(buf, sizeof(buf));
5098
5099 // Get rid of libjvm.so
5100 p = strrchr(buf, '/');
5101 if (p == NULL) return false;
5102 else *p = '\0';
5103
5104 // Get rid of client or server
5105 p = strrchr(buf, '/');
5106 if (p == NULL) return false;
5107 else *p = '\0';
5108
5109 // check xawt/libmawt.so
5110 strcpy(libmawtpath, buf);
5111 strcat(libmawtpath, xawtstr);
5112 if (::stat(libmawtpath, &statbuf) == 0) return false;
5113
5114 // check motif21/libmawt.so
5115 strcpy(libmawtpath, buf);
5116 strcat(libmawtpath, motifstr);
5117 if (::stat(libmawtpath, &statbuf) == 0) return false;
5118
5119 return true;
5120 }
5121