1 /*
2 * Copyright (c) 1999, 2012, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "jvm_linux.h"
34 #include "memory/allocation.inline.hpp"
35 #include "memory/filemap.hpp"
36 #include "mutex_linux.inline.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "os_share_linux.hpp"
39 #include "prims/jniFastGetField.hpp"
40 #include "prims/jvm.h"
41 #include "prims/jvm_misc.hpp"
42 #include "runtime/arguments.hpp"
43 #include "runtime/extendedPC.hpp"
44 #include "runtime/globals.hpp"
45 #include "runtime/interfaceSupport.hpp"
46 #include "runtime/java.hpp"
47 #include "runtime/javaCalls.hpp"
48 #include "runtime/mutexLocker.hpp"
49 #include "runtime/objectMonitor.hpp"
50 #include "runtime/osThread.hpp"
51 #include "runtime/perfMemory.hpp"
52 #include "runtime/sharedRuntime.hpp"
53 #include "runtime/statSampler.hpp"
54 #include "runtime/stubRoutines.hpp"
55 #include "runtime/thread.inline.hpp"
56 #include "runtime/threadCritical.hpp"
57 #include "runtime/timer.hpp"
58 #include "services/attachListener.hpp"
59 #include "services/runtimeService.hpp"
60 #include "utilities/decoder.hpp"
61 #include "utilities/defaultStream.hpp"
62 #include "utilities/events.hpp"
63 #include "utilities/growableArray.hpp"
64 #include "utilities/vmError.hpp"
65 #ifdef TARGET_ARCH_x86
66 # include "assembler_x86.inline.hpp"
67 # include "nativeInst_x86.hpp"
68 #endif
69 #ifdef TARGET_ARCH_sparc
70 # include "assembler_sparc.inline.hpp"
71 # include "nativeInst_sparc.hpp"
72 #endif
73 #ifdef TARGET_ARCH_zero
74 # include "assembler_zero.inline.hpp"
75 # include "nativeInst_zero.hpp"
76 #endif
77 #ifdef TARGET_ARCH_arm
78 # include "assembler_arm.inline.hpp"
79 # include "nativeInst_arm.hpp"
80 #endif
81 #ifdef TARGET_ARCH_ppc
82 # include "assembler_ppc.inline.hpp"
83 # include "nativeInst_ppc.hpp"
84 #endif
85
86 // put OS-includes here
87 # include <sys/types.h>
88 # include <sys/mman.h>
89 # include <sys/stat.h>
90 # include <sys/select.h>
91 # include <pthread.h>
92 # include <signal.h>
93 # include <errno.h>
94 # include <dlfcn.h>
95 # include <stdio.h>
96 # include <unistd.h>
97 # include <sys/resource.h>
98 # include <pthread.h>
99 # include <sys/stat.h>
100 # include <sys/time.h>
101 # include <sys/times.h>
102 # include <sys/utsname.h>
103 # include <sys/socket.h>
104 # include <sys/wait.h>
105 # include <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 # include <sys/ioctl.h>
119
120 #define MAX_PATH (2 * K)
121
122 // for timer info max values which include all bits
123 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
124
125 #define LARGEPAGES_BIT (1 << 6)
126 ////////////////////////////////////////////////////////////////////////////////
127 // global variables
128 julong os::Linux::_physical_memory = 0;
129
130 address os::Linux::_initial_thread_stack_bottom = NULL;
131 uintptr_t os::Linux::_initial_thread_stack_size = 0;
132
133 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
134 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
135 Mutex* os::Linux::_createThread_lock = NULL;
136 pthread_t os::Linux::_main_thread;
137 int os::Linux::_page_size = -1;
138 bool os::Linux::_is_floating_stack = false;
139 bool os::Linux::_is_NPTL = false;
140 bool os::Linux::_supports_fast_thread_cpu_time = false;
141 const char * os::Linux::_glibc_version = NULL;
142 const char * os::Linux::_libpthread_version = NULL;
143
144 static jlong initial_time_count=0;
145
146 static int clock_tics_per_sec = 100;
147
148 // For diagnostics to print a message once. see run_periodic_checks
149 static sigset_t check_signal_done;
150 static bool check_signals = true;;
151
152 static pid_t _initial_pid = 0;
153
154 /* Signal number used to suspend/resume a thread */
155
156 /* do not use any signal number less than SIGSEGV, see 4355769 */
157 static int SR_signum = SIGUSR2;
158 sigset_t SR_sigset;
159
160 /* Used to protect dlsym() calls */
161 static pthread_mutex_t dl_mutex;
162
163 #ifdef JAVASE_EMBEDDED
164 class MemNotifyThread: public Thread {
165 friend class VMStructs;
166 public:
167 virtual void run();
168
169 private:
170 static MemNotifyThread* _memnotify_thread;
171 int _fd;
172
173 public:
174
175 // Constructor
176 MemNotifyThread(int fd);
177
178 // Tester
179 bool is_memnotify_thread() const { return true; }
180
181 // Printing
182 char* name() const { return (char*)"Linux MemNotify Thread"; }
183
184 // Returns the single instance of the MemNotifyThread
185 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
186
187 // Create and start the single instance of MemNotifyThread
188 static void start();
189 };
190 #endif // JAVASE_EMBEDDED
191
192 // utility functions
193
194 static int SR_initialize();
195 static int SR_finalize();
196
197 julong os::available_memory() {
198 return Linux::available_memory();
199 }
200
201 julong os::Linux::available_memory() {
202 // values in struct sysinfo are "unsigned long"
203 struct sysinfo si;
204 sysinfo(&si);
205
206 return (julong)si.freeram * si.mem_unit;
207 }
208
209 julong os::physical_memory() {
210 return Linux::physical_memory();
211 }
212
213 julong os::allocatable_physical_memory(julong size) {
214 #ifdef _LP64
215 return size;
216 #else
217 julong result = MIN2(size, (julong)3800*M);
218 if (!is_allocatable(result)) {
219 // See comments under solaris for alignment considerations
220 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
221 result = MIN2(size, reasonable_size);
222 }
223 return result;
224 #endif // _LP64
225 }
226
227 ////////////////////////////////////////////////////////////////////////////////
228 // environment support
229
230 bool os::getenv(const char* name, char* buf, int len) {
231 const char* val = ::getenv(name);
232 if (val != NULL && strlen(val) < (size_t)len) {
233 strcpy(buf, val);
234 return true;
235 }
236 if (len > 0) buf[0] = 0; // return a null string
237 return false;
238 }
239
240
241 // Return true if user is running as root.
242
243 bool os::have_special_privileges() {
244 static bool init = false;
245 static bool privileges = false;
246 if (!init) {
247 privileges = (getuid() != geteuid()) || (getgid() != getegid());
248 init = true;
249 }
250 return privileges;
251 }
252
253
254 #ifndef SYS_gettid
255 // i386: 224, ia64: 1105, amd64: 186, sparc 143
256 #ifdef __ia64__
257 #define SYS_gettid 1105
258 #elif __i386__
259 #define SYS_gettid 224
260 #elif __amd64__
261 #define SYS_gettid 186
262 #elif __sparc__
263 #define SYS_gettid 143
264 #else
265 #error define gettid for the arch
266 #endif
267 #endif
268
269 // Cpu architecture string
270 #if defined(ZERO)
271 static char cpu_arch[] = ZERO_LIBARCH;
272 #elif defined(IA64)
273 static char cpu_arch[] = "ia64";
274 #elif defined(IA32)
275 static char cpu_arch[] = "i386";
276 #elif defined(AMD64)
277 static char cpu_arch[] = "amd64";
278 #elif defined(ARM)
279 static char cpu_arch[] = "arm";
280 #elif defined(PPC)
281 static char cpu_arch[] = "ppc";
282 #elif defined(SPARC)
283 # ifdef _LP64
284 static char cpu_arch[] = "sparcv9";
285 # else
286 static char cpu_arch[] = "sparc";
287 # endif
288 #else
289 #error Add appropriate cpu_arch setting
290 #endif
291
292
293 // pid_t gettid()
294 //
295 // Returns the kernel thread id of the currently running thread. Kernel
296 // thread id is used to access /proc.
297 //
298 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
299 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
300 //
301 pid_t os::Linux::gettid() {
302 int rslt = syscall(SYS_gettid);
303 if (rslt == -1) {
304 // old kernel, no NPTL support
305 return getpid();
306 } else {
307 return (pid_t)rslt;
308 }
309 }
310
311 // Most versions of linux have a bug where the number of processors are
312 // determined by looking at the /proc file system. In a chroot environment,
313 // the system call returns 1. This causes the VM to act as if it is
314 // a single processor and elide locking (see is_MP() call).
315 static bool unsafe_chroot_detected = false;
316 static const char *unstable_chroot_error = "/proc file system not found.\n"
317 "Java may be unstable running multithreaded in a chroot "
318 "environment on Linux when /proc filesystem is not mounted.";
319
320 void os::Linux::initialize_system_info() {
321 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
322 if (processor_count() == 1) {
323 pid_t pid = os::Linux::gettid();
324 char fname[32];
325 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
326 FILE *fp = fopen(fname, "r");
327 if (fp == NULL) {
328 unsafe_chroot_detected = true;
329 } else {
330 fclose(fp);
331 }
332 }
333 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
334 assert(processor_count() > 0, "linux error");
335 }
336
337 void os::init_system_properties_values() {
338 // char arch[12];
339 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
340
341 // The next steps are taken in the product version:
342 //
343 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
344 // This library should be located at:
345 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
346 //
347 // If "/jre/lib/" appears at the right place in the path, then we
348 // assume libjvm[_g].so is installed in a JDK and we use this path.
349 //
350 // Otherwise exit with message: "Could not create the Java virtual machine."
351 //
352 // The following extra steps are taken in the debugging version:
353 //
354 // If "/jre/lib/" does NOT appear at the right place in the path
355 // instead of exit check for $JAVA_HOME environment variable.
356 //
357 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
358 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
359 // it looks like libjvm[_g].so is installed there
360 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
361 //
362 // Otherwise exit.
363 //
364 // Important note: if the location of libjvm.so changes this
365 // code needs to be changed accordingly.
366
367 // The next few definitions allow the code to be verbatim:
368 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
369 #define getenv(n) ::getenv(n)
370
371 /*
372 * See ld(1):
373 * The linker uses the following search paths to locate required
374 * shared libraries:
375 * 1: ...
376 * ...
377 * 7: The default directories, normally /lib and /usr/lib.
378 */
379 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
380 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
381 #else
382 #define DEFAULT_LIBPATH "/lib:/usr/lib"
383 #endif
384
385 #define EXTENSIONS_DIR "/lib/ext"
386 #define ENDORSED_DIR "/lib/endorsed"
387 #define REG_DIR "/usr/java/packages"
388
389 {
390 /* sysclasspath, java_home, dll_dir */
391 {
392 char *home_path;
393 char *dll_path;
394 char *pslash;
395 char buf[MAXPATHLEN];
396 os::jvm_path(buf, sizeof(buf));
397
398 // Found the full path to libjvm.so.
399 // Now cut the path to <java_home>/jre if we can.
400 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
401 pslash = strrchr(buf, '/');
402 if (pslash != NULL)
403 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
404 dll_path = malloc(strlen(buf) + 1);
405 if (dll_path == NULL)
406 return;
407 strcpy(dll_path, buf);
408 Arguments::set_dll_dir(dll_path);
409
410 if (pslash != NULL) {
411 pslash = strrchr(buf, '/');
412 if (pslash != NULL) {
413 *pslash = '\0'; /* get rid of /<arch> */
414 pslash = strrchr(buf, '/');
415 if (pslash != NULL)
416 *pslash = '\0'; /* get rid of /lib */
417 }
418 }
419
420 home_path = malloc(strlen(buf) + 1);
421 if (home_path == NULL)
422 return;
423 strcpy(home_path, buf);
424 Arguments::set_java_home(home_path);
425
426 if (!set_boot_path('/', ':'))
427 return;
428 }
429
430 /*
431 * Where to look for native libraries
432 *
433 * Note: Due to a legacy implementation, most of the library path
434 * is set in the launcher. This was to accomodate linking restrictions
435 * on legacy Linux implementations (which are no longer supported).
436 * Eventually, all the library path setting will be done here.
437 *
438 * However, to prevent the proliferation of improperly built native
439 * libraries, the new path component /usr/java/packages is added here.
440 * Eventually, all the library path setting will be done here.
441 */
442 {
443 char *ld_library_path;
444
445 /*
446 * Construct the invariant part of ld_library_path. Note that the
447 * space for the colon and the trailing null are provided by the
448 * nulls included by the sizeof operator (so actually we allocate
449 * a byte more than necessary).
450 */
451 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
452 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
453 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
454
455 /*
456 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
457 * should always exist (until the legacy problem cited above is
458 * addressed).
459 */
460 char *v = getenv("LD_LIBRARY_PATH");
461 if (v != NULL) {
462 char *t = ld_library_path;
463 /* That's +1 for the colon and +1 for the trailing '\0' */
464 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
465 sprintf(ld_library_path, "%s:%s", v, t);
466 }
467 Arguments::set_library_path(ld_library_path);
468 }
469
470 /*
471 * Extensions directories.
472 *
473 * Note that the space for the colon and the trailing null are provided
474 * by the nulls included by the sizeof operator (so actually one byte more
475 * than necessary is allocated).
476 */
477 {
478 char *buf = malloc(strlen(Arguments::get_java_home()) +
479 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
480 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
481 Arguments::get_java_home());
482 Arguments::set_ext_dirs(buf);
483 }
484
485 /* Endorsed standards default directory. */
486 {
487 char * buf;
488 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
489 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
490 Arguments::set_endorsed_dirs(buf);
491 }
492 }
493
494 #undef malloc
495 #undef getenv
496 #undef EXTENSIONS_DIR
497 #undef ENDORSED_DIR
498
499 // Done
500 return;
501 }
502
503 ////////////////////////////////////////////////////////////////////////////////
504 // breakpoint support
505
506 void os::breakpoint() {
507 BREAKPOINT;
508 }
509
510 extern "C" void breakpoint() {
511 // use debugger to set breakpoint here
512 }
513
514 ////////////////////////////////////////////////////////////////////////////////
515 // signal support
516
517 debug_only(static bool signal_sets_initialized = false);
518 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
519
520 bool os::Linux::is_sig_ignored(int sig) {
521 struct sigaction oact;
522 sigaction(sig, (struct sigaction*)NULL, &oact);
523 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
524 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
525 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
526 return true;
527 else
528 return false;
529 }
530
531 void os::Linux::signal_sets_init() {
532 // Should also have an assertion stating we are still single-threaded.
533 assert(!signal_sets_initialized, "Already initialized");
534 // Fill in signals that are necessarily unblocked for all threads in
535 // the VM. Currently, we unblock the following signals:
536 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
537 // by -Xrs (=ReduceSignalUsage));
538 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
539 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
540 // the dispositions or masks wrt these signals.
541 // Programs embedding the VM that want to use the above signals for their
542 // own purposes must, at this time, use the "-Xrs" option to prevent
543 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
544 // (See bug 4345157, and other related bugs).
545 // In reality, though, unblocking these signals is really a nop, since
546 // these signals are not blocked by default.
547 sigemptyset(&unblocked_sigs);
548 sigemptyset(&allowdebug_blocked_sigs);
549 sigaddset(&unblocked_sigs, SIGILL);
550 sigaddset(&unblocked_sigs, SIGSEGV);
551 sigaddset(&unblocked_sigs, SIGBUS);
552 sigaddset(&unblocked_sigs, SIGFPE);
553 sigaddset(&unblocked_sigs, SR_signum);
554
555 if (!ReduceSignalUsage) {
556 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
557 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
558 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
559 }
560 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
561 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
562 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
563 }
564 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
565 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
566 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
567 }
568 }
569 // Fill in signals that are blocked by all but the VM thread.
570 sigemptyset(&vm_sigs);
571 if (!ReduceSignalUsage)
572 sigaddset(&vm_sigs, BREAK_SIGNAL);
573 debug_only(signal_sets_initialized = true);
574
575 }
576
577 // These are signals that are unblocked while a thread is running Java.
578 // (For some reason, they get blocked by default.)
579 sigset_t* os::Linux::unblocked_signals() {
580 assert(signal_sets_initialized, "Not initialized");
581 return &unblocked_sigs;
582 }
583
584 // These are the signals that are blocked while a (non-VM) thread is
585 // running Java. Only the VM thread handles these signals.
586 sigset_t* os::Linux::vm_signals() {
587 assert(signal_sets_initialized, "Not initialized");
588 return &vm_sigs;
589 }
590
591 // These are signals that are blocked during cond_wait to allow debugger in
592 sigset_t* os::Linux::allowdebug_blocked_signals() {
593 assert(signal_sets_initialized, "Not initialized");
594 return &allowdebug_blocked_sigs;
595 }
596
597 void os::Linux::hotspot_sigmask(Thread* thread) {
598
599 //Save caller's signal mask before setting VM signal mask
600 sigset_t caller_sigmask;
601 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
602
603 OSThread* osthread = thread->osthread();
604 osthread->set_caller_sigmask(caller_sigmask);
605
606 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
607
608 if (!ReduceSignalUsage) {
609 if (thread->is_VM_thread()) {
610 // Only the VM thread handles BREAK_SIGNAL ...
611 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
612 } else {
613 // ... all other threads block BREAK_SIGNAL
614 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
615 }
616 }
617 }
618
619 //////////////////////////////////////////////////////////////////////////////
620 // detecting pthread library
621
622 void os::Linux::libpthread_init() {
623 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
624 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
625 // generic name for earlier versions.
626 // Define macros here so we can build HotSpot on old systems.
627 # ifndef _CS_GNU_LIBC_VERSION
628 # define _CS_GNU_LIBC_VERSION 2
629 # endif
630 # ifndef _CS_GNU_LIBPTHREAD_VERSION
631 # define _CS_GNU_LIBPTHREAD_VERSION 3
632 # endif
633
634 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
635 if (n > 0) {
636 char *str = (char *)malloc(n, mtInternal);
637 confstr(_CS_GNU_LIBC_VERSION, str, n);
638 os::Linux::set_glibc_version(str);
639 } else {
640 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
641 static char _gnu_libc_version[32];
642 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
643 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
644 os::Linux::set_glibc_version(_gnu_libc_version);
645 }
646
647 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
648 if (n > 0) {
649 char *str = (char *)malloc(n, mtInternal);
650 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
651 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
652 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
653 // is the case. LinuxThreads has a hard limit on max number of threads.
654 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
655 // On the other hand, NPTL does not have such a limit, sysconf()
656 // will return -1 and errno is not changed. Check if it is really NPTL.
657 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
658 strstr(str, "NPTL") &&
659 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
660 free(str);
661 os::Linux::set_libpthread_version("linuxthreads");
662 } else {
663 os::Linux::set_libpthread_version(str);
664 }
665 } else {
666 // glibc before 2.3.2 only has LinuxThreads.
667 os::Linux::set_libpthread_version("linuxthreads");
668 }
669
670 if (strstr(libpthread_version(), "NPTL")) {
671 os::Linux::set_is_NPTL();
672 } else {
673 os::Linux::set_is_LinuxThreads();
674 }
675
676 // LinuxThreads have two flavors: floating-stack mode, which allows variable
677 // stack size; and fixed-stack mode. NPTL is always floating-stack.
678 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
679 os::Linux::set_is_floating_stack();
680 }
681 }
682
683 /////////////////////////////////////////////////////////////////////////////
684 // thread stack
685
686 // Force Linux kernel to expand current thread stack. If "bottom" is close
687 // to the stack guard, caller should block all signals.
688 //
689 // MAP_GROWSDOWN:
690 // A special mmap() flag that is used to implement thread stacks. It tells
691 // kernel that the memory region should extend downwards when needed. This
692 // allows early versions of LinuxThreads to only mmap the first few pages
693 // when creating a new thread. Linux kernel will automatically expand thread
694 // stack as needed (on page faults).
695 //
696 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
697 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
698 // region, it's hard to tell if the fault is due to a legitimate stack
699 // access or because of reading/writing non-exist memory (e.g. buffer
700 // overrun). As a rule, if the fault happens below current stack pointer,
701 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
702 // application (see Linux kernel fault.c).
703 //
704 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
705 // stack overflow detection.
706 //
707 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
708 // not use this flag. However, the stack of initial thread is not created
709 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
710 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
711 // and then attach the thread to JVM.
712 //
713 // To get around the problem and allow stack banging on Linux, we need to
714 // manually expand thread stack after receiving the SIGSEGV.
715 //
716 // There are two ways to expand thread stack to address "bottom", we used
717 // both of them in JVM before 1.5:
718 // 1. adjust stack pointer first so that it is below "bottom", and then
719 // touch "bottom"
720 // 2. mmap() the page in question
721 //
722 // Now alternate signal stack is gone, it's harder to use 2. For instance,
723 // if current sp is already near the lower end of page 101, and we need to
724 // call mmap() to map page 100, it is possible that part of the mmap() frame
725 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
726 // That will destroy the mmap() frame and cause VM to crash.
727 //
728 // The following code works by adjusting sp first, then accessing the "bottom"
729 // page to force a page fault. Linux kernel will then automatically expand the
730 // stack mapping.
731 //
732 // _expand_stack_to() assumes its frame size is less than page size, which
733 // should always be true if the function is not inlined.
734
735 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
736 #define NOINLINE
737 #else
738 #define NOINLINE __attribute__ ((noinline))
739 #endif
740
741 static void _expand_stack_to(address bottom) NOINLINE;
742
743 static void _expand_stack_to(address bottom) {
744 address sp;
745 size_t size;
746 volatile char *p;
747
748 // Adjust bottom to point to the largest address within the same page, it
749 // gives us a one-page buffer if alloca() allocates slightly more memory.
750 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
751 bottom += os::Linux::page_size() - 1;
752
753 // sp might be slightly above current stack pointer; if that's the case, we
754 // will alloca() a little more space than necessary, which is OK. Don't use
755 // os::current_stack_pointer(), as its result can be slightly below current
756 // stack pointer, causing us to not alloca enough to reach "bottom".
757 sp = (address)&sp;
758
759 if (sp > bottom) {
760 size = sp - bottom;
761 p = (volatile char *)alloca(size);
762 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
763 p[0] = '\0';
764 }
765 }
766
767 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
768 assert(t!=NULL, "just checking");
769 assert(t->osthread()->expanding_stack(), "expand should be set");
770 assert(t->stack_base() != NULL, "stack_base was not initialized");
771
772 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
773 sigset_t mask_all, old_sigset;
774 sigfillset(&mask_all);
775 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
776 _expand_stack_to(addr);
777 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
778 return true;
779 }
780 return false;
781 }
782
783 //////////////////////////////////////////////////////////////////////////////
784 // create new thread
785
786 static address highest_vm_reserved_address();
787
788 // check if it's safe to start a new thread
789 static bool _thread_safety_check(Thread* thread) {
790 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
791 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
792 // Heap is mmap'ed at lower end of memory space. Thread stacks are
793 // allocated (MAP_FIXED) from high address space. Every thread stack
794 // occupies a fixed size slot (usually 2Mbytes, but user can change
795 // it to other values if they rebuild LinuxThreads).
796 //
797 // Problem with MAP_FIXED is that mmap() can still succeed even part of
798 // the memory region has already been mmap'ed. That means if we have too
799 // many threads and/or very large heap, eventually thread stack will
800 // collide with heap.
801 //
802 // Here we try to prevent heap/stack collision by comparing current
803 // stack bottom with the highest address that has been mmap'ed by JVM
804 // plus a safety margin for memory maps created by native code.
805 //
806 // This feature can be disabled by setting ThreadSafetyMargin to 0
807 //
808 if (ThreadSafetyMargin > 0) {
809 address stack_bottom = os::current_stack_base() - os::current_stack_size();
810
811 // not safe if our stack extends below the safety margin
812 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
813 } else {
814 return true;
815 }
816 } else {
817 // Floating stack LinuxThreads or NPTL:
818 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
819 // there's not enough space left, pthread_create() will fail. If we come
820 // here, that means enough space has been reserved for stack.
821 return true;
822 }
823 }
824
825 // Thread start routine for all newly created threads
826 static void *java_start(Thread *thread) {
827 // Try to randomize the cache line index of hot stack frames.
828 // This helps when threads of the same stack traces evict each other's
829 // cache lines. The threads can be either from the same JVM instance, or
830 // from different JVM instances. The benefit is especially true for
831 // processors with hyperthreading technology.
832 static int counter = 0;
833 int pid = os::current_process_id();
834 alloca(((pid ^ counter++) & 7) * 128);
835
836 ThreadLocalStorage::set_thread(thread);
837
838 OSThread* osthread = thread->osthread();
839 Monitor* sync = osthread->startThread_lock();
840
841 // non floating stack LinuxThreads needs extra check, see above
842 if (!_thread_safety_check(thread)) {
843 // notify parent thread
844 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
845 osthread->set_state(ZOMBIE);
846 sync->notify_all();
847 return NULL;
848 }
849
850 // thread_id is kernel thread id (similar to Solaris LWP id)
851 osthread->set_thread_id(os::Linux::gettid());
852
853 if (UseNUMA) {
854 int lgrp_id = os::numa_get_group_id();
855 if (lgrp_id != -1) {
856 thread->set_lgrp_id(lgrp_id);
857 }
858 }
859 // initialize signal mask for this thread
860 os::Linux::hotspot_sigmask(thread);
861
862 // initialize floating point control register
863 os::Linux::init_thread_fpu_state();
864
865 // handshaking with parent thread
866 {
867 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
868
869 // notify parent thread
870 osthread->set_state(INITIALIZED);
871 sync->notify_all();
872
873 // wait until os::start_thread()
874 while (osthread->get_state() == INITIALIZED) {
875 sync->wait(Mutex::_no_safepoint_check_flag);
876 }
877 }
878
879 // call one more level start routine
880 thread->run();
881
882 return 0;
883 }
884
885 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
886 assert(thread->osthread() == NULL, "caller responsible");
887
888 // Allocate the OSThread object
889 OSThread* osthread = new OSThread(NULL, NULL);
890 if (osthread == NULL) {
891 return false;
892 }
893
894 // set the correct thread state
895 osthread->set_thread_type(thr_type);
896
897 // Initial state is ALLOCATED but not INITIALIZED
898 osthread->set_state(ALLOCATED);
899
900 thread->set_osthread(osthread);
901
902 // init thread attributes
903 pthread_attr_t attr;
904 pthread_attr_init(&attr);
905 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
906
907 // stack size
908 if (os::Linux::supports_variable_stack_size()) {
909 // calculate stack size if it's not specified by caller
910 if (stack_size == 0) {
911 stack_size = os::Linux::default_stack_size(thr_type);
912
913 switch (thr_type) {
914 case os::java_thread:
915 // Java threads use ThreadStackSize which default value can be
916 // changed with the flag -Xss
917 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
918 stack_size = JavaThread::stack_size_at_create();
919 break;
920 case os::compiler_thread:
921 if (CompilerThreadStackSize > 0) {
922 stack_size = (size_t)(CompilerThreadStackSize * K);
923 break;
924 } // else fall through:
925 // use VMThreadStackSize if CompilerThreadStackSize is not defined
926 case os::vm_thread:
927 case os::pgc_thread:
928 case os::cgc_thread:
929 case os::watcher_thread:
930 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
931 break;
932 }
933 }
934
935 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
936 pthread_attr_setstacksize(&attr, stack_size);
937 } else {
938 // let pthread_create() pick the default value.
939 }
940
941 // glibc guard page
942 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
943
944 ThreadState state;
945
946 {
947 // Serialize thread creation if we are running with fixed stack LinuxThreads
948 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
949 if (lock) {
950 os::Linux::createThread_lock()->lock_without_safepoint_check();
951 }
952
953 pthread_t tid;
954 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
955
956 pthread_attr_destroy(&attr);
957
958 if (ret != 0) {
959 if (PrintMiscellaneous && (Verbose || WizardMode)) {
960 perror("pthread_create()");
961 }
962 // Need to clean up stuff we've allocated so far
963 thread->set_osthread(NULL);
964 delete osthread;
965 if (lock) os::Linux::createThread_lock()->unlock();
966 return false;
967 }
968
969 // Store pthread info into the OSThread
970 osthread->set_pthread_id(tid);
971
972 // Wait until child thread is either initialized or aborted
973 {
974 Monitor* sync_with_child = osthread->startThread_lock();
975 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
976 while ((state = osthread->get_state()) == ALLOCATED) {
977 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
978 }
979 }
980
981 if (lock) {
982 os::Linux::createThread_lock()->unlock();
983 }
984 }
985
986 // Aborted due to thread limit being reached
987 if (state == ZOMBIE) {
988 thread->set_osthread(NULL);
989 delete osthread;
990 return false;
991 }
992
993 // The thread is returned suspended (in state INITIALIZED),
994 // and is started higher up in the call chain
995 assert(state == INITIALIZED, "race condition");
996 return true;
997 }
998
999 /////////////////////////////////////////////////////////////////////////////
1000 // attach existing thread
1001
1002 // bootstrap the main thread
1003 bool os::create_main_thread(JavaThread* thread) {
1004 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1005 return create_attached_thread(thread);
1006 }
1007
1008 bool os::create_attached_thread(JavaThread* thread) {
1009 #ifdef ASSERT
1010 thread->verify_not_published();
1011 #endif
1012
1013 // Allocate the OSThread object
1014 OSThread* osthread = new OSThread(NULL, NULL);
1015
1016 if (osthread == NULL) {
1017 return false;
1018 }
1019
1020 // Store pthread info into the OSThread
1021 osthread->set_thread_id(os::Linux::gettid());
1022 osthread->set_pthread_id(::pthread_self());
1023
1024 // initialize floating point control register
1025 os::Linux::init_thread_fpu_state();
1026
1027 // Initial thread state is RUNNABLE
1028 osthread->set_state(RUNNABLE);
1029
1030 thread->set_osthread(osthread);
1031
1032 if (UseNUMA) {
1033 int lgrp_id = os::numa_get_group_id();
1034 if (lgrp_id != -1) {
1035 thread->set_lgrp_id(lgrp_id);
1036 }
1037 }
1038
1039 if (os::Linux::is_initial_thread()) {
1040 // If current thread is initial thread, its stack is mapped on demand,
1041 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1042 // the entire stack region to avoid SEGV in stack banging.
1043 // It is also useful to get around the heap-stack-gap problem on SuSE
1044 // kernel (see 4821821 for details). We first expand stack to the top
1045 // of yellow zone, then enable stack yellow zone (order is significant,
1046 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1047 // is no gap between the last two virtual memory regions.
1048
1049 JavaThread *jt = (JavaThread *)thread;
1050 address addr = jt->stack_yellow_zone_base();
1051 assert(addr != NULL, "initialization problem?");
1052 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1053
1054 osthread->set_expanding_stack();
1055 os::Linux::manually_expand_stack(jt, addr);
1056 osthread->clear_expanding_stack();
1057 }
1058
1059 // initialize signal mask for this thread
1060 // and save the caller's signal mask
1061 os::Linux::hotspot_sigmask(thread);
1062
1063 return true;
1064 }
1065
1066 void os::pd_start_thread(Thread* thread) {
1067 OSThread * osthread = thread->osthread();
1068 assert(osthread->get_state() != INITIALIZED, "just checking");
1069 Monitor* sync_with_child = osthread->startThread_lock();
1070 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1071 sync_with_child->notify();
1072 }
1073
1074 // Free Linux resources related to the OSThread
1075 void os::free_thread(OSThread* osthread) {
1076 assert(osthread != NULL, "osthread not set");
1077
1078 if (Thread::current()->osthread() == osthread) {
1079 // Restore caller's signal mask
1080 sigset_t sigmask = osthread->caller_sigmask();
1081 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1082 }
1083
1084 delete osthread;
1085 }
1086
1087 //////////////////////////////////////////////////////////////////////////////
1088 // thread local storage
1089
1090 int os::allocate_thread_local_storage() {
1091 pthread_key_t key;
1092 int rslt = pthread_key_create(&key, NULL);
1093 assert(rslt == 0, "cannot allocate thread local storage");
1094 return (int)key;
1095 }
1096
1097 // Note: This is currently not used by VM, as we don't destroy TLS key
1098 // on VM exit.
1099 void os::free_thread_local_storage(int index) {
1100 int rslt = pthread_key_delete((pthread_key_t)index);
1101 assert(rslt == 0, "invalid index");
1102 }
1103
1104 void os::thread_local_storage_at_put(int index, void* value) {
1105 int rslt = pthread_setspecific((pthread_key_t)index, value);
1106 assert(rslt == 0, "pthread_setspecific failed");
1107 }
1108
1109 extern "C" Thread* get_thread() {
1110 return ThreadLocalStorage::thread();
1111 }
1112
1113 //////////////////////////////////////////////////////////////////////////////
1114 // initial thread
1115
1116 // Check if current thread is the initial thread, similar to Solaris thr_main.
1117 bool os::Linux::is_initial_thread(void) {
1118 char dummy;
1119 // If called before init complete, thread stack bottom will be null.
1120 // Can be called if fatal error occurs before initialization.
1121 if (initial_thread_stack_bottom() == NULL) return false;
1122 assert(initial_thread_stack_bottom() != NULL &&
1123 initial_thread_stack_size() != 0,
1124 "os::init did not locate initial thread's stack region");
1125 if ((address)&dummy >= initial_thread_stack_bottom() &&
1126 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1127 return true;
1128 else return false;
1129 }
1130
1131 // Find the virtual memory area that contains addr
1132 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1133 FILE *fp = fopen("/proc/self/maps", "r");
1134 if (fp) {
1135 address low, high;
1136 while (!feof(fp)) {
1137 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1138 if (low <= addr && addr < high) {
1139 if (vma_low) *vma_low = low;
1140 if (vma_high) *vma_high = high;
1141 fclose (fp);
1142 return true;
1143 }
1144 }
1145 for (;;) {
1146 int ch = fgetc(fp);
1147 if (ch == EOF || ch == (int)'\n') break;
1148 }
1149 }
1150 fclose(fp);
1151 }
1152 return false;
1153 }
1154
1155 // Locate initial thread stack. This special handling of initial thread stack
1156 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1157 // bogus value for initial thread.
1158 void os::Linux::capture_initial_stack(size_t max_size) {
1159 // stack size is the easy part, get it from RLIMIT_STACK
1160 size_t stack_size;
1161 struct rlimit rlim;
1162 getrlimit(RLIMIT_STACK, &rlim);
1163 stack_size = rlim.rlim_cur;
1164
1165 // 6308388: a bug in ld.so will relocate its own .data section to the
1166 // lower end of primordial stack; reduce ulimit -s value a little bit
1167 // so we won't install guard page on ld.so's data section.
1168 stack_size -= 2 * page_size();
1169
1170 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1171 // 7.1, in both cases we will get 2G in return value.
1172 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1173 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1174 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1175 // in case other parts in glibc still assumes 2M max stack size.
1176 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1177 #ifndef IA64
1178 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1179 #else
1180 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1181 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1182 #endif
1183
1184 // Try to figure out where the stack base (top) is. This is harder.
1185 //
1186 // When an application is started, glibc saves the initial stack pointer in
1187 // a global variable "__libc_stack_end", which is then used by system
1188 // libraries. __libc_stack_end should be pretty close to stack top. The
1189 // variable is available since the very early days. However, because it is
1190 // a private interface, it could disappear in the future.
1191 //
1192 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1193 // to __libc_stack_end, it is very close to stack top, but isn't the real
1194 // stack top. Note that /proc may not exist if VM is running as a chroot
1195 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1196 // /proc/<pid>/stat could change in the future (though unlikely).
1197 //
1198 // We try __libc_stack_end first. If that doesn't work, look for
1199 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1200 // as a hint, which should work well in most cases.
1201
1202 uintptr_t stack_start;
1203
1204 // try __libc_stack_end first
1205 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1206 if (p && *p) {
1207 stack_start = *p;
1208 } else {
1209 // see if we can get the start_stack field from /proc/self/stat
1210 FILE *fp;
1211 int pid;
1212 char state;
1213 int ppid;
1214 int pgrp;
1215 int session;
1216 int nr;
1217 int tpgrp;
1218 unsigned long flags;
1219 unsigned long minflt;
1220 unsigned long cminflt;
1221 unsigned long majflt;
1222 unsigned long cmajflt;
1223 unsigned long utime;
1224 unsigned long stime;
1225 long cutime;
1226 long cstime;
1227 long prio;
1228 long nice;
1229 long junk;
1230 long it_real;
1231 uintptr_t start;
1232 uintptr_t vsize;
1233 intptr_t rss;
1234 uintptr_t rsslim;
1235 uintptr_t scodes;
1236 uintptr_t ecode;
1237 int i;
1238
1239 // Figure what the primordial thread stack base is. Code is inspired
1240 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1241 // followed by command name surrounded by parentheses, state, etc.
1242 char stat[2048];
1243 int statlen;
1244
1245 fp = fopen("/proc/self/stat", "r");
1246 if (fp) {
1247 statlen = fread(stat, 1, 2047, fp);
1248 stat[statlen] = '\0';
1249 fclose(fp);
1250
1251 // Skip pid and the command string. Note that we could be dealing with
1252 // weird command names, e.g. user could decide to rename java launcher
1253 // to "java 1.4.2 :)", then the stat file would look like
1254 // 1234 (java 1.4.2 :)) R ... ...
1255 // We don't really need to know the command string, just find the last
1256 // occurrence of ")" and then start parsing from there. See bug 4726580.
1257 char * s = strrchr(stat, ')');
1258
1259 i = 0;
1260 if (s) {
1261 // Skip blank chars
1262 do s++; while (isspace(*s));
1263
1264 #define _UFM UINTX_FORMAT
1265 #define _DFM INTX_FORMAT
1266
1267 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1268 /* 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 */
1269 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,
1270 &state, /* 3 %c */
1271 &ppid, /* 4 %d */
1272 &pgrp, /* 5 %d */
1273 &session, /* 6 %d */
1274 &nr, /* 7 %d */
1275 &tpgrp, /* 8 %d */
1276 &flags, /* 9 %lu */
1277 &minflt, /* 10 %lu */
1278 &cminflt, /* 11 %lu */
1279 &majflt, /* 12 %lu */
1280 &cmajflt, /* 13 %lu */
1281 &utime, /* 14 %lu */
1282 &stime, /* 15 %lu */
1283 &cutime, /* 16 %ld */
1284 &cstime, /* 17 %ld */
1285 &prio, /* 18 %ld */
1286 &nice, /* 19 %ld */
1287 &junk, /* 20 %ld */
1288 &it_real, /* 21 %ld */
1289 &start, /* 22 UINTX_FORMAT */
1290 &vsize, /* 23 UINTX_FORMAT */
1291 &rss, /* 24 INTX_FORMAT */
1292 &rsslim, /* 25 UINTX_FORMAT */
1293 &scodes, /* 26 UINTX_FORMAT */
1294 &ecode, /* 27 UINTX_FORMAT */
1295 &stack_start); /* 28 UINTX_FORMAT */
1296 }
1297
1298 #undef _UFM
1299 #undef _DFM
1300
1301 if (i != 28 - 2) {
1302 assert(false, "Bad conversion from /proc/self/stat");
1303 // product mode - assume we are the initial thread, good luck in the
1304 // embedded case.
1305 warning("Can't detect initial thread stack location - bad conversion");
1306 stack_start = (uintptr_t) &rlim;
1307 }
1308 } else {
1309 // For some reason we can't open /proc/self/stat (for example, running on
1310 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1311 // most cases, so don't abort:
1312 warning("Can't detect initial thread stack location - no /proc/self/stat");
1313 stack_start = (uintptr_t) &rlim;
1314 }
1315 }
1316
1317 // Now we have a pointer (stack_start) very close to the stack top, the
1318 // next thing to do is to figure out the exact location of stack top. We
1319 // can find out the virtual memory area that contains stack_start by
1320 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1321 // and its upper limit is the real stack top. (again, this would fail if
1322 // running inside chroot, because /proc may not exist.)
1323
1324 uintptr_t stack_top;
1325 address low, high;
1326 if (find_vma((address)stack_start, &low, &high)) {
1327 // success, "high" is the true stack top. (ignore "low", because initial
1328 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1329 stack_top = (uintptr_t)high;
1330 } else {
1331 // failed, likely because /proc/self/maps does not exist
1332 warning("Can't detect initial thread stack location - find_vma failed");
1333 // best effort: stack_start is normally within a few pages below the real
1334 // stack top, use it as stack top, and reduce stack size so we won't put
1335 // guard page outside stack.
1336 stack_top = stack_start;
1337 stack_size -= 16 * page_size();
1338 }
1339
1340 // stack_top could be partially down the page so align it
1341 stack_top = align_size_up(stack_top, page_size());
1342
1343 if (max_size && stack_size > max_size) {
1344 _initial_thread_stack_size = max_size;
1345 } else {
1346 _initial_thread_stack_size = stack_size;
1347 }
1348
1349 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1350 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1351 }
1352
1353 ////////////////////////////////////////////////////////////////////////////////
1354 // time support
1355
1356 // Time since start-up in seconds to a fine granularity.
1357 // Used by VMSelfDestructTimer and the MemProfiler.
1358 double os::elapsedTime() {
1359
1360 return (double)(os::elapsed_counter()) * 0.000001;
1361 }
1362
1363 jlong os::elapsed_counter() {
1364 timeval time;
1365 int status = gettimeofday(&time, NULL);
1366 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1367 }
1368
1369 jlong os::elapsed_frequency() {
1370 return (1000 * 1000);
1371 }
1372
1373 // For now, we say that linux does not support vtime. I have no idea
1374 // whether it can actually be made to (DLD, 9/13/05).
1375
1376 bool os::supports_vtime() { return false; }
1377 bool os::enable_vtime() { return false; }
1378 bool os::vtime_enabled() { return false; }
1379 double os::elapsedVTime() {
1380 // better than nothing, but not much
1381 return elapsedTime();
1382 }
1383
1384 jlong os::javaTimeMillis() {
1385 timeval time;
1386 int status = gettimeofday(&time, NULL);
1387 assert(status != -1, "linux error");
1388 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1389 }
1390
1391 #ifndef CLOCK_MONOTONIC
1392 #define CLOCK_MONOTONIC (1)
1393 #endif
1394
1395 void os::Linux::clock_init() {
1396 // we do dlopen's in this particular order due to bug in linux
1397 // dynamical loader (see 6348968) leading to crash on exit
1398 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1399 if (handle == NULL) {
1400 handle = dlopen("librt.so", RTLD_LAZY);
1401 }
1402
1403 if (handle) {
1404 int (*clock_getres_func)(clockid_t, struct timespec*) =
1405 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1406 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1407 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1408 if (clock_getres_func && clock_gettime_func) {
1409 // See if monotonic clock is supported by the kernel. Note that some
1410 // early implementations simply return kernel jiffies (updated every
1411 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1412 // for nano time (though the monotonic property is still nice to have).
1413 // It's fixed in newer kernels, however clock_getres() still returns
1414 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1415 // resolution for now. Hopefully as people move to new kernels, this
1416 // won't be a problem.
1417 struct timespec res;
1418 struct timespec tp;
1419 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1420 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1421 // yes, monotonic clock is supported
1422 _clock_gettime = clock_gettime_func;
1423 } else {
1424 // close librt if there is no monotonic clock
1425 dlclose(handle);
1426 }
1427 }
1428 }
1429 }
1430
1431 #ifndef SYS_clock_getres
1432
1433 #if defined(IA32) || defined(AMD64)
1434 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1435 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1436 #else
1437 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1438 #define sys_clock_getres(x,y) -1
1439 #endif
1440
1441 #else
1442 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1443 #endif
1444
1445 void os::Linux::fast_thread_clock_init() {
1446 if (!UseLinuxPosixThreadCPUClocks) {
1447 return;
1448 }
1449 clockid_t clockid;
1450 struct timespec tp;
1451 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1452 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1453
1454 // Switch to using fast clocks for thread cpu time if
1455 // the sys_clock_getres() returns 0 error code.
1456 // Note, that some kernels may support the current thread
1457 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1458 // returned by the pthread_getcpuclockid().
1459 // If the fast Posix clocks are supported then the sys_clock_getres()
1460 // must return at least tp.tv_sec == 0 which means a resolution
1461 // better than 1 sec. This is extra check for reliability.
1462
1463 if(pthread_getcpuclockid_func &&
1464 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1465 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1466
1467 _supports_fast_thread_cpu_time = true;
1468 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1469 }
1470 }
1471
1472 jlong os::javaTimeNanos() {
1473 if (Linux::supports_monotonic_clock()) {
1474 struct timespec tp;
1475 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1476 assert(status == 0, "gettime error");
1477 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1478 return result;
1479 } else {
1480 timeval time;
1481 int status = gettimeofday(&time, NULL);
1482 assert(status != -1, "linux error");
1483 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1484 return 1000 * usecs;
1485 }
1486 }
1487
1488 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1489 if (Linux::supports_monotonic_clock()) {
1490 info_ptr->max_value = ALL_64_BITS;
1491
1492 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1493 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1494 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1495 } else {
1496 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1497 info_ptr->max_value = ALL_64_BITS;
1498
1499 // gettimeofday is a real time clock so it skips
1500 info_ptr->may_skip_backward = true;
1501 info_ptr->may_skip_forward = true;
1502 }
1503
1504 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1505 }
1506
1507 // Return the real, user, and system times in seconds from an
1508 // arbitrary fixed point in the past.
1509 bool os::getTimesSecs(double* process_real_time,
1510 double* process_user_time,
1511 double* process_system_time) {
1512 struct tms ticks;
1513 clock_t real_ticks = times(&ticks);
1514
1515 if (real_ticks == (clock_t) (-1)) {
1516 return false;
1517 } else {
1518 double ticks_per_second = (double) clock_tics_per_sec;
1519 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1520 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1521 *process_real_time = ((double) real_ticks) / ticks_per_second;
1522
1523 return true;
1524 }
1525 }
1526
1527
1528 char * os::local_time_string(char *buf, size_t buflen) {
1529 struct tm t;
1530 time_t long_time;
1531 time(&long_time);
1532 localtime_r(&long_time, &t);
1533 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1534 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1535 t.tm_hour, t.tm_min, t.tm_sec);
1536 return buf;
1537 }
1538
1539 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1540 return localtime_r(clock, res);
1541 }
1542
1543 ////////////////////////////////////////////////////////////////////////////////
1544 // runtime exit support
1545
1546 // Note: os::shutdown() might be called very early during initialization, or
1547 // called from signal handler. Before adding something to os::shutdown(), make
1548 // sure it is async-safe and can handle partially initialized VM.
1549 void os::shutdown() {
1550
1551 // allow PerfMemory to attempt cleanup of any persistent resources
1552 perfMemory_exit();
1553
1554 // needs to remove object in file system
1555 AttachListener::abort();
1556
1557 // flush buffered output, finish log files
1558 ostream_abort();
1559
1560 // Check for abort hook
1561 abort_hook_t abort_hook = Arguments::abort_hook();
1562 if (abort_hook != NULL) {
1563 abort_hook();
1564 }
1565
1566 }
1567
1568 // Note: os::abort() might be called very early during initialization, or
1569 // called from signal handler. Before adding something to os::abort(), make
1570 // sure it is async-safe and can handle partially initialized VM.
1571 void os::abort(bool dump_core) {
1572 os::shutdown();
1573 if (dump_core) {
1574 #ifndef PRODUCT
1575 fdStream out(defaultStream::output_fd());
1576 out.print_raw("Current thread is ");
1577 char buf[16];
1578 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1579 out.print_raw_cr(buf);
1580 out.print_raw_cr("Dumping core ...");
1581 #endif
1582 ::abort(); // dump core
1583 }
1584
1585 ::exit(1);
1586 }
1587
1588 // Die immediately, no exit hook, no abort hook, no cleanup.
1589 void os::die() {
1590 // _exit() on LinuxThreads only kills current thread
1591 ::abort();
1592 }
1593
1594 // unused on linux for now.
1595 void os::set_error_file(const char *logfile) {}
1596
1597
1598 // This method is a copy of JDK's sysGetLastErrorString
1599 // from src/solaris/hpi/src/system_md.c
1600
1601 size_t os::lasterror(char *buf, size_t len) {
1602
1603 if (errno == 0) return 0;
1604
1605 const char *s = ::strerror(errno);
1606 size_t n = ::strlen(s);
1607 if (n >= len) {
1608 n = len - 1;
1609 }
1610 ::strncpy(buf, s, n);
1611 buf[n] = '\0';
1612 return n;
1613 }
1614
1615 intx os::current_thread_id() { return (intx)pthread_self(); }
1616 int os::current_process_id() {
1617
1618 // Under the old linux thread library, linux gives each thread
1619 // its own process id. Because of this each thread will return
1620 // a different pid if this method were to return the result
1621 // of getpid(2). Linux provides no api that returns the pid
1622 // of the launcher thread for the vm. This implementation
1623 // returns a unique pid, the pid of the launcher thread
1624 // that starts the vm 'process'.
1625
1626 // Under the NPTL, getpid() returns the same pid as the
1627 // launcher thread rather than a unique pid per thread.
1628 // Use gettid() if you want the old pre NPTL behaviour.
1629
1630 // if you are looking for the result of a call to getpid() that
1631 // returns a unique pid for the calling thread, then look at the
1632 // OSThread::thread_id() method in osThread_linux.hpp file
1633
1634 return (int)(_initial_pid ? _initial_pid : getpid());
1635 }
1636
1637 // DLL functions
1638
1639 const char* os::dll_file_extension() { return ".so"; }
1640
1641 // This must be hard coded because it's the system's temporary
1642 // directory not the java application's temp directory, ala java.io.tmpdir.
1643 const char* os::get_temp_directory() { return "/tmp"; }
1644
1645 static bool file_exists(const char* filename) {
1646 struct stat statbuf;
1647 if (filename == NULL || strlen(filename) == 0) {
1648 return false;
1649 }
1650 return os::stat(filename, &statbuf) == 0;
1651 }
1652
1653 bool os::dll_build_name(char* buffer, size_t buflen,
1654 const char* pname, const char* fname) {
1655 bool retval = false;
1656 // Copied from libhpi
1657 const size_t pnamelen = pname ? strlen(pname) : 0;
1658
1659 // Return error on buffer overflow.
1660 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1661 return retval;
1662 }
1663
1664 if (pnamelen == 0) {
1665 snprintf(buffer, buflen, "lib%s.so", fname);
1666 retval = true;
1667 } else if (strchr(pname, *os::path_separator()) != NULL) {
1668 int n;
1669 char** pelements = split_path(pname, &n);
1670 for (int i = 0 ; i < n ; i++) {
1671 // Really shouldn't be NULL, but check can't hurt
1672 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1673 continue; // skip the empty path values
1674 }
1675 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1676 if (file_exists(buffer)) {
1677 retval = true;
1678 break;
1679 }
1680 }
1681 // release the storage
1682 for (int i = 0 ; i < n ; i++) {
1683 if (pelements[i] != NULL) {
1684 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1685 }
1686 }
1687 if (pelements != NULL) {
1688 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1689 }
1690 } else {
1691 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1692 retval = true;
1693 }
1694 return retval;
1695 }
1696
1697 const char* os::get_current_directory(char *buf, int buflen) {
1698 return getcwd(buf, buflen);
1699 }
1700
1701 // check if addr is inside libjvm[_g].so
1702 bool os::address_is_in_vm(address addr) {
1703 static address libjvm_base_addr;
1704 Dl_info dlinfo;
1705
1706 if (libjvm_base_addr == NULL) {
1707 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1708 libjvm_base_addr = (address)dlinfo.dli_fbase;
1709 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1710 }
1711
1712 if (dladdr((void *)addr, &dlinfo)) {
1713 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1714 }
1715
1716 return false;
1717 }
1718
1719 bool os::dll_address_to_function_name(address addr, char *buf,
1720 int buflen, int *offset) {
1721 Dl_info dlinfo;
1722
1723 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1724 if (buf != NULL) {
1725 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1726 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1727 }
1728 }
1729 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1730 return true;
1731 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1732 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1733 buf, buflen, offset, dlinfo.dli_fname)) {
1734 return true;
1735 }
1736 }
1737
1738 if (buf != NULL) buf[0] = '\0';
1739 if (offset != NULL) *offset = -1;
1740 return false;
1741 }
1742
1743 struct _address_to_library_name {
1744 address addr; // input : memory address
1745 size_t buflen; // size of fname
1746 char* fname; // output: library name
1747 address base; // library base addr
1748 };
1749
1750 static int address_to_library_name_callback(struct dl_phdr_info *info,
1751 size_t size, void *data) {
1752 int i;
1753 bool found = false;
1754 address libbase = NULL;
1755 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1756
1757 // iterate through all loadable segments
1758 for (i = 0; i < info->dlpi_phnum; i++) {
1759 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1760 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1761 // base address of a library is the lowest address of its loaded
1762 // segments.
1763 if (libbase == NULL || libbase > segbase) {
1764 libbase = segbase;
1765 }
1766 // see if 'addr' is within current segment
1767 if (segbase <= d->addr &&
1768 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1769 found = true;
1770 }
1771 }
1772 }
1773
1774 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1775 // so dll_address_to_library_name() can fall through to use dladdr() which
1776 // can figure out executable name from argv[0].
1777 if (found && info->dlpi_name && info->dlpi_name[0]) {
1778 d->base = libbase;
1779 if (d->fname) {
1780 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1781 }
1782 return 1;
1783 }
1784 return 0;
1785 }
1786
1787 bool os::dll_address_to_library_name(address addr, char* buf,
1788 int buflen, int* offset) {
1789 Dl_info dlinfo;
1790 struct _address_to_library_name data;
1791
1792 // There is a bug in old glibc dladdr() implementation that it could resolve
1793 // to wrong library name if the .so file has a base address != NULL. Here
1794 // we iterate through the program headers of all loaded libraries to find
1795 // out which library 'addr' really belongs to. This workaround can be
1796 // removed once the minimum requirement for glibc is moved to 2.3.x.
1797 data.addr = addr;
1798 data.fname = buf;
1799 data.buflen = buflen;
1800 data.base = NULL;
1801 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1802
1803 if (rslt) {
1804 // buf already contains library name
1805 if (offset) *offset = addr - data.base;
1806 return true;
1807 } else if (dladdr((void*)addr, &dlinfo)){
1808 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1809 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1810 return true;
1811 } else {
1812 if (buf) buf[0] = '\0';
1813 if (offset) *offset = -1;
1814 return false;
1815 }
1816 }
1817
1818 // Loads .dll/.so and
1819 // in case of error it checks if .dll/.so was built for the
1820 // same architecture as Hotspot is running on
1821
1822 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1823 {
1824 void * result= ::dlopen(filename, RTLD_LAZY);
1825 if (result != NULL) {
1826 // Successful loading
1827 return result;
1828 }
1829
1830 Elf32_Ehdr elf_head;
1831
1832 // Read system error message into ebuf
1833 // It may or may not be overwritten below
1834 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1835 ebuf[ebuflen-1]='\0';
1836 int diag_msg_max_length=ebuflen-strlen(ebuf);
1837 char* diag_msg_buf=ebuf+strlen(ebuf);
1838
1839 if (diag_msg_max_length==0) {
1840 // No more space in ebuf for additional diagnostics message
1841 return NULL;
1842 }
1843
1844
1845 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1846
1847 if (file_descriptor < 0) {
1848 // Can't open library, report dlerror() message
1849 return NULL;
1850 }
1851
1852 bool failed_to_read_elf_head=
1853 (sizeof(elf_head)!=
1854 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1855
1856 ::close(file_descriptor);
1857 if (failed_to_read_elf_head) {
1858 // file i/o error - report dlerror() msg
1859 return NULL;
1860 }
1861
1862 typedef struct {
1863 Elf32_Half code; // Actual value as defined in elf.h
1864 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1865 char elf_class; // 32 or 64 bit
1866 char endianess; // MSB or LSB
1867 char* name; // String representation
1868 } arch_t;
1869
1870 #ifndef EM_486
1871 #define EM_486 6 /* Intel 80486 */
1872 #endif
1873
1874 static const arch_t arch_array[]={
1875 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1876 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1877 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1878 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1879 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1880 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1881 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1882 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1883 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1884 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1885 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1886 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1887 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1888 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1889 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1890 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1891 };
1892
1893 #if (defined IA32)
1894 static Elf32_Half running_arch_code=EM_386;
1895 #elif (defined AMD64)
1896 static Elf32_Half running_arch_code=EM_X86_64;
1897 #elif (defined IA64)
1898 static Elf32_Half running_arch_code=EM_IA_64;
1899 #elif (defined __sparc) && (defined _LP64)
1900 static Elf32_Half running_arch_code=EM_SPARCV9;
1901 #elif (defined __sparc) && (!defined _LP64)
1902 static Elf32_Half running_arch_code=EM_SPARC;
1903 #elif (defined __powerpc64__)
1904 static Elf32_Half running_arch_code=EM_PPC64;
1905 #elif (defined __powerpc__)
1906 static Elf32_Half running_arch_code=EM_PPC;
1907 #elif (defined ARM)
1908 static Elf32_Half running_arch_code=EM_ARM;
1909 #elif (defined S390)
1910 static Elf32_Half running_arch_code=EM_S390;
1911 #elif (defined ALPHA)
1912 static Elf32_Half running_arch_code=EM_ALPHA;
1913 #elif (defined MIPSEL)
1914 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1915 #elif (defined PARISC)
1916 static Elf32_Half running_arch_code=EM_PARISC;
1917 #elif (defined MIPS)
1918 static Elf32_Half running_arch_code=EM_MIPS;
1919 #elif (defined M68K)
1920 static Elf32_Half running_arch_code=EM_68K;
1921 #else
1922 #error Method os::dll_load requires that one of following is defined:\
1923 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1924 #endif
1925
1926 // Identify compatability class for VM's architecture and library's architecture
1927 // Obtain string descriptions for architectures
1928
1929 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1930 int running_arch_index=-1;
1931
1932 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1933 if (running_arch_code == arch_array[i].code) {
1934 running_arch_index = i;
1935 }
1936 if (lib_arch.code == arch_array[i].code) {
1937 lib_arch.compat_class = arch_array[i].compat_class;
1938 lib_arch.name = arch_array[i].name;
1939 }
1940 }
1941
1942 assert(running_arch_index != -1,
1943 "Didn't find running architecture code (running_arch_code) in arch_array");
1944 if (running_arch_index == -1) {
1945 // Even though running architecture detection failed
1946 // we may still continue with reporting dlerror() message
1947 return NULL;
1948 }
1949
1950 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1951 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1952 return NULL;
1953 }
1954
1955 #ifndef S390
1956 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1957 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1958 return NULL;
1959 }
1960 #endif // !S390
1961
1962 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1963 if ( lib_arch.name!=NULL ) {
1964 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1965 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1966 lib_arch.name, arch_array[running_arch_index].name);
1967 } else {
1968 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1969 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1970 lib_arch.code,
1971 arch_array[running_arch_index].name);
1972 }
1973 }
1974
1975 return NULL;
1976 }
1977
1978 /*
1979 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1980 * chances are you might want to run the generated bits against glibc-2.0
1981 * libdl.so, so always use locking for any version of glibc.
1982 */
1983 void* os::dll_lookup(void* handle, const char* name) {
1984 pthread_mutex_lock(&dl_mutex);
1985 void* res = dlsym(handle, name);
1986 pthread_mutex_unlock(&dl_mutex);
1987 return res;
1988 }
1989
1990
1991 static bool _print_ascii_file(const char* filename, outputStream* st) {
1992 int fd = ::open(filename, O_RDONLY);
1993 if (fd == -1) {
1994 return false;
1995 }
1996
1997 char buf[32];
1998 int bytes;
1999 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2000 st->print_raw(buf, bytes);
2001 }
2002
2003 ::close(fd);
2004
2005 return true;
2006 }
2007
2008 void os::print_dll_info(outputStream *st) {
2009 st->print_cr("Dynamic libraries:");
2010
2011 char fname[32];
2012 pid_t pid = os::Linux::gettid();
2013
2014 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2015
2016 if (!_print_ascii_file(fname, st)) {
2017 st->print("Can not get library information for pid = %d\n", pid);
2018 }
2019 }
2020
2021 void os::print_os_info_brief(outputStream* st) {
2022 os::Linux::print_distro_info(st);
2023
2024 os::Posix::print_uname_info(st);
2025
2026 os::Linux::print_libversion_info(st);
2027
2028 }
2029
2030 void os::print_os_info(outputStream* st) {
2031 st->print("OS:");
2032
2033 os::Linux::print_distro_info(st);
2034
2035 os::Posix::print_uname_info(st);
2036
2037 // Print warning if unsafe chroot environment detected
2038 if (unsafe_chroot_detected) {
2039 st->print("WARNING!! ");
2040 st->print_cr(unstable_chroot_error);
2041 }
2042
2043 os::Linux::print_libversion_info(st);
2044
2045 os::Posix::print_rlimit_info(st);
2046
2047 os::Posix::print_load_average(st);
2048
2049 os::Linux::print_full_memory_info(st);
2050 }
2051
2052 // Try to identify popular distros.
2053 // Most Linux distributions have /etc/XXX-release file, which contains
2054 // the OS version string. Some have more than one /etc/XXX-release file
2055 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2056 // so the order is important.
2057 void os::Linux::print_distro_info(outputStream* st) {
2058 if (!_print_ascii_file("/etc/mandrake-release", st) &&
2059 !_print_ascii_file("/etc/sun-release", st) &&
2060 !_print_ascii_file("/etc/redhat-release", st) &&
2061 !_print_ascii_file("/etc/SuSE-release", st) &&
2062 !_print_ascii_file("/etc/turbolinux-release", st) &&
2063 !_print_ascii_file("/etc/gentoo-release", st) &&
2064 !_print_ascii_file("/etc/debian_version", st) &&
2065 !_print_ascii_file("/etc/ltib-release", st) &&
2066 !_print_ascii_file("/etc/angstrom-version", st)) {
2067 st->print("Linux");
2068 }
2069 st->cr();
2070 }
2071
2072 void os::Linux::print_libversion_info(outputStream* st) {
2073 // libc, pthread
2074 st->print("libc:");
2075 st->print(os::Linux::glibc_version()); st->print(" ");
2076 st->print(os::Linux::libpthread_version()); st->print(" ");
2077 if (os::Linux::is_LinuxThreads()) {
2078 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2079 }
2080 st->cr();
2081 }
2082
2083 void os::Linux::print_full_memory_info(outputStream* st) {
2084 st->print("\n/proc/meminfo:\n");
2085 _print_ascii_file("/proc/meminfo", st);
2086 st->cr();
2087 }
2088
2089 void os::print_memory_info(outputStream* st) {
2090
2091 st->print("Memory:");
2092 st->print(" %dk page", os::vm_page_size()>>10);
2093
2094 // values in struct sysinfo are "unsigned long"
2095 struct sysinfo si;
2096 sysinfo(&si);
2097
2098 st->print(", physical " UINT64_FORMAT "k",
2099 os::physical_memory() >> 10);
2100 st->print("(" UINT64_FORMAT "k free)",
2101 os::available_memory() >> 10);
2102 st->print(", swap " UINT64_FORMAT "k",
2103 ((jlong)si.totalswap * si.mem_unit) >> 10);
2104 st->print("(" UINT64_FORMAT "k free)",
2105 ((jlong)si.freeswap * si.mem_unit) >> 10);
2106 st->cr();
2107 }
2108
2109 void os::pd_print_cpu_info(outputStream* st) {
2110 st->print("\n/proc/cpuinfo:\n");
2111 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2112 st->print(" <Not Available>");
2113 }
2114 st->cr();
2115 }
2116
2117 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2118 // but they're the same for all the linux arch that we support
2119 // and they're the same for solaris but there's no common place to put this.
2120 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2121 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2122 "ILL_COPROC", "ILL_BADSTK" };
2123
2124 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2125 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2126 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2127
2128 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2129
2130 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2131
2132 void os::print_siginfo(outputStream* st, void* siginfo) {
2133 st->print("siginfo:");
2134
2135 const int buflen = 100;
2136 char buf[buflen];
2137 siginfo_t *si = (siginfo_t*)siginfo;
2138 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2139 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2140 st->print("si_errno=%s", buf);
2141 } else {
2142 st->print("si_errno=%d", si->si_errno);
2143 }
2144 const int c = si->si_code;
2145 assert(c > 0, "unexpected si_code");
2146 switch (si->si_signo) {
2147 case SIGILL:
2148 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2149 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2150 break;
2151 case SIGFPE:
2152 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2153 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2154 break;
2155 case SIGSEGV:
2156 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2157 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2158 break;
2159 case SIGBUS:
2160 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2161 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2162 break;
2163 default:
2164 st->print(", si_code=%d", si->si_code);
2165 // no si_addr
2166 }
2167
2168 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2169 UseSharedSpaces) {
2170 FileMapInfo* mapinfo = FileMapInfo::current_info();
2171 if (mapinfo->is_in_shared_space(si->si_addr)) {
2172 st->print("\n\nError accessing class data sharing archive." \
2173 " Mapped file inaccessible during execution, " \
2174 " possible disk/network problem.");
2175 }
2176 }
2177 st->cr();
2178 }
2179
2180
2181 static void print_signal_handler(outputStream* st, int sig,
2182 char* buf, size_t buflen);
2183
2184 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2185 st->print_cr("Signal Handlers:");
2186 print_signal_handler(st, SIGSEGV, buf, buflen);
2187 print_signal_handler(st, SIGBUS , buf, buflen);
2188 print_signal_handler(st, SIGFPE , buf, buflen);
2189 print_signal_handler(st, SIGPIPE, buf, buflen);
2190 print_signal_handler(st, SIGXFSZ, buf, buflen);
2191 print_signal_handler(st, SIGILL , buf, buflen);
2192 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2193 print_signal_handler(st, SR_signum, buf, buflen);
2194 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2195 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2196 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2197 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2198 }
2199
2200 static char saved_jvm_path[MAXPATHLEN] = {0};
2201
2202 // Find the full path to the current module, libjvm.so or libjvm_g.so
2203 void os::jvm_path(char *buf, jint buflen) {
2204 // Error checking.
2205 if (buflen < MAXPATHLEN) {
2206 assert(false, "must use a large-enough buffer");
2207 buf[0] = '\0';
2208 return;
2209 }
2210 // Lazy resolve the path to current module.
2211 if (saved_jvm_path[0] != 0) {
2212 strcpy(buf, saved_jvm_path);
2213 return;
2214 }
2215
2216 char dli_fname[MAXPATHLEN];
2217 bool ret = dll_address_to_library_name(
2218 CAST_FROM_FN_PTR(address, os::jvm_path),
2219 dli_fname, sizeof(dli_fname), NULL);
2220 assert(ret != 0, "cannot locate libjvm");
2221 char *rp = realpath(dli_fname, buf);
2222 if (rp == NULL)
2223 return;
2224
2225 if (Arguments::created_by_gamma_launcher()) {
2226 // Support for the gamma launcher. Typical value for buf is
2227 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2228 // the right place in the string, then assume we are installed in a JDK and
2229 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2230 // up the path so it looks like libjvm.so is installed there (append a
2231 // fake suffix hotspot/libjvm.so).
2232 const char *p = buf + strlen(buf) - 1;
2233 for (int count = 0; p > buf && count < 5; ++count) {
2234 for (--p; p > buf && *p != '/'; --p)
2235 /* empty */ ;
2236 }
2237
2238 if (strncmp(p, "/jre/lib/", 9) != 0) {
2239 // Look for JAVA_HOME in the environment.
2240 char* java_home_var = ::getenv("JAVA_HOME");
2241 if (java_home_var != NULL && java_home_var[0] != 0) {
2242 char* jrelib_p;
2243 int len;
2244
2245 // Check the current module name "libjvm.so" or "libjvm_g.so".
2246 p = strrchr(buf, '/');
2247 assert(strstr(p, "/libjvm") == p, "invalid library name");
2248 p = strstr(p, "_g") ? "_g" : "";
2249
2250 rp = realpath(java_home_var, buf);
2251 if (rp == NULL)
2252 return;
2253
2254 // determine if this is a legacy image or modules image
2255 // modules image doesn't have "jre" subdirectory
2256 len = strlen(buf);
2257 jrelib_p = buf + len;
2258 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2259 if (0 != access(buf, F_OK)) {
2260 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2261 }
2262
2263 if (0 == access(buf, F_OK)) {
2264 // Use current module name "libjvm[_g].so" instead of
2265 // "libjvm"debug_only("_g")".so" since for fastdebug version
2266 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2267 len = strlen(buf);
2268 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2269 } else {
2270 // Go back to path of .so
2271 rp = realpath(dli_fname, buf);
2272 if (rp == NULL)
2273 return;
2274 }
2275 }
2276 }
2277 }
2278
2279 strcpy(saved_jvm_path, buf);
2280 }
2281
2282 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2283 // no prefix required, not even "_"
2284 }
2285
2286 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2287 // no suffix required
2288 }
2289
2290 ////////////////////////////////////////////////////////////////////////////////
2291 // sun.misc.Signal support
2292
2293 static volatile jint sigint_count = 0;
2294
2295 static void
2296 UserHandler(int sig, void *siginfo, void *context) {
2297 // 4511530 - sem_post is serialized and handled by the manager thread. When
2298 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2299 // don't want to flood the manager thread with sem_post requests.
2300 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2301 return;
2302
2303 // Ctrl-C is pressed during error reporting, likely because the error
2304 // handler fails to abort. Let VM die immediately.
2305 if (sig == SIGINT && is_error_reported()) {
2306 os::die();
2307 }
2308
2309 os::signal_notify(sig);
2310 }
2311
2312 void* os::user_handler() {
2313 return CAST_FROM_FN_PTR(void*, UserHandler);
2314 }
2315
2316 extern "C" {
2317 typedef void (*sa_handler_t)(int);
2318 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2319 }
2320
2321 void* os::signal(int signal_number, void* handler) {
2322 struct sigaction sigAct, oldSigAct;
2323
2324 sigfillset(&(sigAct.sa_mask));
2325 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2326 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2327
2328 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2329 // -1 means registration failed
2330 return (void *)-1;
2331 }
2332
2333 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2334 }
2335
2336 void os::signal_raise(int signal_number) {
2337 ::raise(signal_number);
2338 }
2339
2340 /*
2341 * The following code is moved from os.cpp for making this
2342 * code platform specific, which it is by its very nature.
2343 */
2344
2345 // Will be modified when max signal is changed to be dynamic
2346 int os::sigexitnum_pd() {
2347 return NSIG;
2348 }
2349
2350 // a counter for each possible signal value
2351 static volatile jint pending_signals[NSIG+1] = { 0 };
2352
2353 // Linux(POSIX) specific hand shaking semaphore.
2354 static sem_t sig_sem;
2355
2356 void os::signal_init_pd() {
2357 // Initialize signal structures
2358 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2359
2360 // Initialize signal semaphore
2361 ::sem_init(&sig_sem, 0, 0);
2362 }
2363
2364 void os::signal_notify(int sig) {
2365 Atomic::inc(&pending_signals[sig]);
2366 ::sem_post(&sig_sem);
2367 }
2368
2369 static int check_pending_signals(bool wait) {
2370 Atomic::store(0, &sigint_count);
2371 for (;;) {
2372 for (int i = 0; i < NSIG + 1; i++) {
2373 jint n = pending_signals[i];
2374 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2375 return i;
2376 }
2377 }
2378 if (!wait) {
2379 return -1;
2380 }
2381 JavaThread *thread = JavaThread::current();
2382 ThreadBlockInVM tbivm(thread);
2383
2384 bool threadIsSuspended;
2385 do {
2386 thread->set_suspend_equivalent();
2387 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2388 ::sem_wait(&sig_sem);
2389
2390 // were we externally suspended while we were waiting?
2391 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2392 if (threadIsSuspended) {
2393 //
2394 // The semaphore has been incremented, but while we were waiting
2395 // another thread suspended us. We don't want to continue running
2396 // while suspended because that would surprise the thread that
2397 // suspended us.
2398 //
2399 ::sem_post(&sig_sem);
2400
2401 thread->java_suspend_self();
2402 }
2403 } while (threadIsSuspended);
2404 }
2405 }
2406
2407 int os::signal_lookup() {
2408 return check_pending_signals(false);
2409 }
2410
2411 int os::signal_wait() {
2412 return check_pending_signals(true);
2413 }
2414
2415 ////////////////////////////////////////////////////////////////////////////////
2416 // Virtual Memory
2417
2418 int os::vm_page_size() {
2419 // Seems redundant as all get out
2420 assert(os::Linux::page_size() != -1, "must call os::init");
2421 return os::Linux::page_size();
2422 }
2423
2424 // Solaris allocates memory by pages.
2425 int os::vm_allocation_granularity() {
2426 assert(os::Linux::page_size() != -1, "must call os::init");
2427 return os::Linux::page_size();
2428 }
2429
2430 // Rationale behind this function:
2431 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2432 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2433 // samples for JITted code. Here we create private executable mapping over the code cache
2434 // and then we can use standard (well, almost, as mapping can change) way to provide
2435 // info for the reporting script by storing timestamp and location of symbol
2436 void linux_wrap_code(char* base, size_t size) {
2437 static volatile jint cnt = 0;
2438
2439 if (!UseOprofile) {
2440 return;
2441 }
2442
2443 char buf[PATH_MAX+1];
2444 int num = Atomic::add(1, &cnt);
2445
2446 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2447 os::get_temp_directory(), os::current_process_id(), num);
2448 unlink(buf);
2449
2450 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2451
2452 if (fd != -1) {
2453 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2454 if (rv != (off_t)-1) {
2455 if (::write(fd, "", 1) == 1) {
2456 mmap(base, size,
2457 PROT_READ|PROT_WRITE|PROT_EXEC,
2458 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2459 }
2460 }
2461 ::close(fd);
2462 unlink(buf);
2463 }
2464 }
2465
2466 // NOTE: Linux kernel does not really reserve the pages for us.
2467 // All it does is to check if there are enough free pages
2468 // left at the time of mmap(). This could be a potential
2469 // problem.
2470 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2471 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2472 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2473 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2474 if (res != (uintptr_t) MAP_FAILED) {
2475 if (UseNUMAInterleaving) {
2476 numa_make_global(addr, size);
2477 }
2478 return true;
2479 }
2480 return false;
2481 }
2482
2483 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2484 #ifndef MAP_HUGETLB
2485 #define MAP_HUGETLB 0x40000
2486 #endif
2487
2488 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2489 #ifndef MADV_HUGEPAGE
2490 #define MADV_HUGEPAGE 14
2491 #endif
2492
2493 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2494 bool exec) {
2495 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2496 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2497 uintptr_t res =
2498 (uintptr_t) ::mmap(addr, size, prot,
2499 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2500 -1, 0);
2501 if (res != (uintptr_t) MAP_FAILED) {
2502 if (UseNUMAInterleaving) {
2503 numa_make_global(addr, size);
2504 }
2505 return true;
2506 }
2507 // Fall through and try to use small pages
2508 }
2509
2510 if (commit_memory(addr, size, exec)) {
2511 realign_memory(addr, size, alignment_hint);
2512 return true;
2513 }
2514 return false;
2515 }
2516
2517 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2518 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2519 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2520 // be supported or the memory may already be backed by huge pages.
2521 ::madvise(addr, bytes, MADV_HUGEPAGE);
2522 }
2523 }
2524
2525 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2526 // This method works by doing an mmap over an existing mmaping and effectively discarding
2527 // the existing pages. However it won't work for SHM-based large pages that cannot be
2528 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2529 // small pages on top of the SHM segment. This method always works for small pages, so we
2530 // allow that in any case.
2531 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2532 commit_memory(addr, bytes, alignment_hint, false);
2533 }
2534 }
2535
2536 void os::numa_make_global(char *addr, size_t bytes) {
2537 Linux::numa_interleave_memory(addr, bytes);
2538 }
2539
2540 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2541 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2542 }
2543
2544 bool os::numa_topology_changed() { return false; }
2545
2546 size_t os::numa_get_groups_num() {
2547 int max_node = Linux::numa_max_node();
2548 return max_node > 0 ? max_node + 1 : 1;
2549 }
2550
2551 int os::numa_get_group_id() {
2552 int cpu_id = Linux::sched_getcpu();
2553 if (cpu_id != -1) {
2554 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2555 if (lgrp_id != -1) {
2556 return lgrp_id;
2557 }
2558 }
2559 return 0;
2560 }
2561
2562 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2563 for (size_t i = 0; i < size; i++) {
2564 ids[i] = i;
2565 }
2566 return size;
2567 }
2568
2569 bool os::get_page_info(char *start, page_info* info) {
2570 return false;
2571 }
2572
2573 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2574 return end;
2575 }
2576
2577
2578 int os::Linux::sched_getcpu_syscall(void) {
2579 unsigned int cpu;
2580 int retval = -1;
2581
2582 #if defined(IA32)
2583 # ifndef SYS_getcpu
2584 # define SYS_getcpu 318
2585 # endif
2586 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2587 #elif defined(AMD64)
2588 // Unfortunately we have to bring all these macros here from vsyscall.h
2589 // to be able to compile on old linuxes.
2590 # define __NR_vgetcpu 2
2591 # define VSYSCALL_START (-10UL << 20)
2592 # define VSYSCALL_SIZE 1024
2593 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2594 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2595 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2596 retval = vgetcpu(&cpu, NULL, NULL);
2597 #endif
2598
2599 return (retval == -1) ? retval : cpu;
2600 }
2601
2602 // Something to do with the numa-aware allocator needs these symbols
2603 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2604 extern "C" JNIEXPORT void numa_error(char *where) { }
2605 extern "C" JNIEXPORT int fork1() { return fork(); }
2606
2607
2608 // If we are running with libnuma version > 2, then we should
2609 // be trying to use symbols with versions 1.1
2610 // If we are running with earlier version, which did not have symbol versions,
2611 // we should use the base version.
2612 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2613 void *f = dlvsym(handle, name, "libnuma_1.1");
2614 if (f == NULL) {
2615 f = dlsym(handle, name);
2616 }
2617 return f;
2618 }
2619
2620 bool os::Linux::libnuma_init() {
2621 // sched_getcpu() should be in libc.
2622 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2623 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2624
2625 // If it's not, try a direct syscall.
2626 if (sched_getcpu() == -1)
2627 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2628
2629 if (sched_getcpu() != -1) { // Does it work?
2630 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2631 if (handle != NULL) {
2632 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2633 libnuma_dlsym(handle, "numa_node_to_cpus")));
2634 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2635 libnuma_dlsym(handle, "numa_max_node")));
2636 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2637 libnuma_dlsym(handle, "numa_available")));
2638 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2639 libnuma_dlsym(handle, "numa_tonode_memory")));
2640 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2641 libnuma_dlsym(handle, "numa_interleave_memory")));
2642
2643
2644 if (numa_available() != -1) {
2645 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2646 // Create a cpu -> node mapping
2647 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2648 rebuild_cpu_to_node_map();
2649 return true;
2650 }
2651 }
2652 }
2653 return false;
2654 }
2655
2656 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2657 // The table is later used in get_node_by_cpu().
2658 void os::Linux::rebuild_cpu_to_node_map() {
2659 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2660 // in libnuma (possible values are starting from 16,
2661 // and continuing up with every other power of 2, but less
2662 // than the maximum number of CPUs supported by kernel), and
2663 // is a subject to change (in libnuma version 2 the requirements
2664 // are more reasonable) we'll just hardcode the number they use
2665 // in the library.
2666 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2667
2668 size_t cpu_num = os::active_processor_count();
2669 size_t cpu_map_size = NCPUS / BitsPerCLong;
2670 size_t cpu_map_valid_size =
2671 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2672
2673 cpu_to_node()->clear();
2674 cpu_to_node()->at_grow(cpu_num - 1);
2675 size_t node_num = numa_get_groups_num();
2676
2677 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2678 for (size_t i = 0; i < node_num; i++) {
2679 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2680 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2681 if (cpu_map[j] != 0) {
2682 for (size_t k = 0; k < BitsPerCLong; k++) {
2683 if (cpu_map[j] & (1UL << k)) {
2684 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2685 }
2686 }
2687 }
2688 }
2689 }
2690 }
2691 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2692 }
2693
2694 int os::Linux::get_node_by_cpu(int cpu_id) {
2695 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2696 return cpu_to_node()->at(cpu_id);
2697 }
2698 return -1;
2699 }
2700
2701 GrowableArray<int>* os::Linux::_cpu_to_node;
2702 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2703 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2704 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2705 os::Linux::numa_available_func_t os::Linux::_numa_available;
2706 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2707 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2708 unsigned long* os::Linux::_numa_all_nodes;
2709
2710 bool os::pd_uncommit_memory(char* addr, size_t size) {
2711 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2712 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2713 return res != (uintptr_t) MAP_FAILED;
2714 }
2715
2716 // Linux uses a growable mapping for the stack, and if the mapping for
2717 // the stack guard pages is not removed when we detach a thread the
2718 // stack cannot grow beyond the pages where the stack guard was
2719 // mapped. If at some point later in the process the stack expands to
2720 // that point, the Linux kernel cannot expand the stack any further
2721 // because the guard pages are in the way, and a segfault occurs.
2722 //
2723 // However, it's essential not to split the stack region by unmapping
2724 // a region (leaving a hole) that's already part of the stack mapping,
2725 // so if the stack mapping has already grown beyond the guard pages at
2726 // the time we create them, we have to truncate the stack mapping.
2727 // So, we need to know the extent of the stack mapping when
2728 // create_stack_guard_pages() is called.
2729
2730 // Find the bounds of the stack mapping. Return true for success.
2731 //
2732 // We only need this for stacks that are growable: at the time of
2733 // writing thread stacks don't use growable mappings (i.e. those
2734 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2735 // only applies to the main thread.
2736
2737 static
2738 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2739
2740 char buf[128];
2741 int fd, sz;
2742
2743 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2744 return false;
2745 }
2746
2747 const char kw[] = "[stack]";
2748 const int kwlen = sizeof(kw)-1;
2749
2750 // Address part of /proc/self/maps couldn't be more than 128 bytes
2751 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2752 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2753 // Extract addresses
2754 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2755 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2756 if (sp >= *bottom && sp <= *top) {
2757 ::close(fd);
2758 return true;
2759 }
2760 }
2761 }
2762 }
2763
2764 ::close(fd);
2765 return false;
2766 }
2767
2768
2769 // If the (growable) stack mapping already extends beyond the point
2770 // where we're going to put our guard pages, truncate the mapping at
2771 // that point by munmap()ping it. This ensures that when we later
2772 // munmap() the guard pages we don't leave a hole in the stack
2773 // mapping. This only affects the main/initial thread, but guard
2774 // against future OS changes
2775 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2776 uintptr_t stack_extent, stack_base;
2777 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2778 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2779 assert(os::Linux::is_initial_thread(),
2780 "growable stack in non-initial thread");
2781 if (stack_extent < (uintptr_t)addr)
2782 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2783 }
2784
2785 return os::commit_memory(addr, size);
2786 }
2787
2788 // If this is a growable mapping, remove the guard pages entirely by
2789 // munmap()ping them. If not, just call uncommit_memory(). This only
2790 // affects the main/initial thread, but guard against future OS changes
2791 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2792 uintptr_t stack_extent, stack_base;
2793 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2794 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2795 assert(os::Linux::is_initial_thread(),
2796 "growable stack in non-initial thread");
2797
2798 return ::munmap(addr, size) == 0;
2799 }
2800
2801 return os::uncommit_memory(addr, size);
2802 }
2803
2804 static address _highest_vm_reserved_address = NULL;
2805
2806 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2807 // at 'requested_addr'. If there are existing memory mappings at the same
2808 // location, however, they will be overwritten. If 'fixed' is false,
2809 // 'requested_addr' is only treated as a hint, the return value may or
2810 // may not start from the requested address. Unlike Linux mmap(), this
2811 // function returns NULL to indicate failure.
2812 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2813 char * addr;
2814 int flags;
2815
2816 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2817 if (fixed) {
2818 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2819 flags |= MAP_FIXED;
2820 }
2821
2822 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2823 // to PROT_EXEC if executable when we commit the page.
2824 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2825 flags, -1, 0);
2826
2827 if (addr != MAP_FAILED) {
2828 // anon_mmap() should only get called during VM initialization,
2829 // don't need lock (actually we can skip locking even it can be called
2830 // from multiple threads, because _highest_vm_reserved_address is just a
2831 // hint about the upper limit of non-stack memory regions.)
2832 if ((address)addr + bytes > _highest_vm_reserved_address) {
2833 _highest_vm_reserved_address = (address)addr + bytes;
2834 }
2835 }
2836
2837 return addr == MAP_FAILED ? NULL : addr;
2838 }
2839
2840 // Don't update _highest_vm_reserved_address, because there might be memory
2841 // regions above addr + size. If so, releasing a memory region only creates
2842 // a hole in the address space, it doesn't help prevent heap-stack collision.
2843 //
2844 static int anon_munmap(char * addr, size_t size) {
2845 return ::munmap(addr, size) == 0;
2846 }
2847
2848 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2849 size_t alignment_hint) {
2850 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2851 }
2852
2853 bool os::pd_release_memory(char* addr, size_t size) {
2854 return anon_munmap(addr, size);
2855 }
2856
2857 static address highest_vm_reserved_address() {
2858 return _highest_vm_reserved_address;
2859 }
2860
2861 static bool linux_mprotect(char* addr, size_t size, int prot) {
2862 // Linux wants the mprotect address argument to be page aligned.
2863 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2864
2865 // According to SUSv3, mprotect() should only be used with mappings
2866 // established by mmap(), and mmap() always maps whole pages. Unaligned
2867 // 'addr' likely indicates problem in the VM (e.g. trying to change
2868 // protection of malloc'ed or statically allocated memory). Check the
2869 // caller if you hit this assert.
2870 assert(addr == bottom, "sanity check");
2871
2872 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2873 return ::mprotect(bottom, size, prot) == 0;
2874 }
2875
2876 // Set protections specified
2877 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2878 bool is_committed) {
2879 unsigned int p = 0;
2880 switch (prot) {
2881 case MEM_PROT_NONE: p = PROT_NONE; break;
2882 case MEM_PROT_READ: p = PROT_READ; break;
2883 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2884 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2885 default:
2886 ShouldNotReachHere();
2887 }
2888 // is_committed is unused.
2889 return linux_mprotect(addr, bytes, p);
2890 }
2891
2892 bool os::guard_memory(char* addr, size_t size) {
2893 return linux_mprotect(addr, size, PROT_NONE);
2894 }
2895
2896 bool os::unguard_memory(char* addr, size_t size) {
2897 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2898 }
2899
2900 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2901 bool result = false;
2902 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2903 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2904 -1, 0);
2905
2906 if (p != (void *) -1) {
2907 // We don't know if this really is a huge page or not.
2908 FILE *fp = fopen("/proc/self/maps", "r");
2909 if (fp) {
2910 while (!feof(fp)) {
2911 char chars[257];
2912 long x = 0;
2913 if (fgets(chars, sizeof(chars), fp)) {
2914 if (sscanf(chars, "%lx-%*x", &x) == 1
2915 && x == (long)p) {
2916 if (strstr (chars, "hugepage")) {
2917 result = true;
2918 break;
2919 }
2920 }
2921 }
2922 }
2923 fclose(fp);
2924 }
2925 munmap (p, page_size);
2926 if (result)
2927 return true;
2928 }
2929
2930 if (warn) {
2931 warning("HugeTLBFS is not supported by the operating system.");
2932 }
2933
2934 return result;
2935 }
2936
2937 /*
2938 * Set the coredump_filter bits to include largepages in core dump (bit 6)
2939 *
2940 * From the coredump_filter documentation:
2941 *
2942 * - (bit 0) anonymous private memory
2943 * - (bit 1) anonymous shared memory
2944 * - (bit 2) file-backed private memory
2945 * - (bit 3) file-backed shared memory
2946 * - (bit 4) ELF header pages in file-backed private memory areas (it is
2947 * effective only if the bit 2 is cleared)
2948 * - (bit 5) hugetlb private memory
2949 * - (bit 6) hugetlb shared memory
2950 */
2951 static void set_coredump_filter(void) {
2952 FILE *f;
2953 long cdm;
2954
2955 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
2956 return;
2957 }
2958
2959 if (fscanf(f, "%lx", &cdm) != 1) {
2960 fclose(f);
2961 return;
2962 }
2963
2964 rewind(f);
2965
2966 if ((cdm & LARGEPAGES_BIT) == 0) {
2967 cdm |= LARGEPAGES_BIT;
2968 fprintf(f, "%#lx", cdm);
2969 }
2970
2971 fclose(f);
2972 }
2973
2974 // Large page support
2975
2976 static size_t _large_page_size = 0;
2977
2978 void os::large_page_init() {
2979 if (!UseLargePages) {
2980 UseHugeTLBFS = false;
2981 UseSHM = false;
2982 return;
2983 }
2984
2985 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
2986 // If UseLargePages is specified on the command line try both methods,
2987 // if it's default, then try only HugeTLBFS.
2988 if (FLAG_IS_DEFAULT(UseLargePages)) {
2989 UseHugeTLBFS = true;
2990 } else {
2991 UseHugeTLBFS = UseSHM = true;
2992 }
2993 }
2994
2995 if (LargePageSizeInBytes) {
2996 _large_page_size = LargePageSizeInBytes;
2997 } else {
2998 // large_page_size on Linux is used to round up heap size. x86 uses either
2999 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3000 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3001 // page as large as 256M.
3002 //
3003 // Here we try to figure out page size by parsing /proc/meminfo and looking
3004 // for a line with the following format:
3005 // Hugepagesize: 2048 kB
3006 //
3007 // If we can't determine the value (e.g. /proc is not mounted, or the text
3008 // format has been changed), we'll use the largest page size supported by
3009 // the processor.
3010
3011 #ifndef ZERO
3012 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3013 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3014 #endif // ZERO
3015
3016 FILE *fp = fopen("/proc/meminfo", "r");
3017 if (fp) {
3018 while (!feof(fp)) {
3019 int x = 0;
3020 char buf[16];
3021 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3022 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3023 _large_page_size = x * K;
3024 break;
3025 }
3026 } else {
3027 // skip to next line
3028 for (;;) {
3029 int ch = fgetc(fp);
3030 if (ch == EOF || ch == (int)'\n') break;
3031 }
3032 }
3033 }
3034 fclose(fp);
3035 }
3036 }
3037
3038 // print a warning if any large page related flag is specified on command line
3039 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3040
3041 const size_t default_page_size = (size_t)Linux::page_size();
3042 if (_large_page_size > default_page_size) {
3043 _page_sizes[0] = _large_page_size;
3044 _page_sizes[1] = default_page_size;
3045 _page_sizes[2] = 0;
3046 }
3047 UseHugeTLBFS = UseHugeTLBFS &&
3048 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3049
3050 if (UseHugeTLBFS)
3051 UseSHM = false;
3052
3053 UseLargePages = UseHugeTLBFS || UseSHM;
3054
3055 set_coredump_filter();
3056 }
3057
3058 #ifndef SHM_HUGETLB
3059 #define SHM_HUGETLB 04000
3060 #endif
3061
3062 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3063 // "exec" is passed in but not used. Creating the shared image for
3064 // the code cache doesn't have an SHM_X executable permission to check.
3065 assert(UseLargePages && UseSHM, "only for SHM large pages");
3066
3067 key_t key = IPC_PRIVATE;
3068 char *addr;
3069
3070 bool warn_on_failure = UseLargePages &&
3071 (!FLAG_IS_DEFAULT(UseLargePages) ||
3072 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3073 );
3074 char msg[128];
3075
3076 // Create a large shared memory region to attach to based on size.
3077 // Currently, size is the total size of the heap
3078 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3079 if (shmid == -1) {
3080 // Possible reasons for shmget failure:
3081 // 1. shmmax is too small for Java heap.
3082 // > check shmmax value: cat /proc/sys/kernel/shmmax
3083 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3084 // 2. not enough large page memory.
3085 // > check available large pages: cat /proc/meminfo
3086 // > increase amount of large pages:
3087 // echo new_value > /proc/sys/vm/nr_hugepages
3088 // Note 1: different Linux may use different name for this property,
3089 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3090 // Note 2: it's possible there's enough physical memory available but
3091 // they are so fragmented after a long run that they can't
3092 // coalesce into large pages. Try to reserve large pages when
3093 // the system is still "fresh".
3094 if (warn_on_failure) {
3095 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3096 warning(msg);
3097 }
3098 return NULL;
3099 }
3100
3101 // attach to the region
3102 addr = (char*)shmat(shmid, req_addr, 0);
3103 int err = errno;
3104
3105 // Remove shmid. If shmat() is successful, the actual shared memory segment
3106 // will be deleted when it's detached by shmdt() or when the process
3107 // terminates. If shmat() is not successful this will remove the shared
3108 // segment immediately.
3109 shmctl(shmid, IPC_RMID, NULL);
3110
3111 if ((intptr_t)addr == -1) {
3112 if (warn_on_failure) {
3113 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3114 warning(msg);
3115 }
3116 return NULL;
3117 }
3118
3119 if ((addr != NULL) && UseNUMAInterleaving) {
3120 numa_make_global(addr, bytes);
3121 }
3122
3123 return addr;
3124 }
3125
3126 bool os::release_memory_special(char* base, size_t bytes) {
3127 // detaching the SHM segment will also delete it, see reserve_memory_special()
3128 int rslt = shmdt(base);
3129 return rslt == 0;
3130 }
3131
3132 size_t os::large_page_size() {
3133 return _large_page_size;
3134 }
3135
3136 // HugeTLBFS allows application to commit large page memory on demand;
3137 // with SysV SHM the entire memory region must be allocated as shared
3138 // memory.
3139 bool os::can_commit_large_page_memory() {
3140 return UseHugeTLBFS;
3141 }
3142
3143 bool os::can_execute_large_page_memory() {
3144 return UseHugeTLBFS;
3145 }
3146
3147 // Reserve memory at an arbitrary address, only if that area is
3148 // available (and not reserved for something else).
3149
3150 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3151 const int max_tries = 10;
3152 char* base[max_tries];
3153 size_t size[max_tries];
3154 const size_t gap = 0x000000;
3155
3156 // Assert only that the size is a multiple of the page size, since
3157 // that's all that mmap requires, and since that's all we really know
3158 // about at this low abstraction level. If we need higher alignment,
3159 // we can either pass an alignment to this method or verify alignment
3160 // in one of the methods further up the call chain. See bug 5044738.
3161 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3162
3163 // Repeatedly allocate blocks until the block is allocated at the
3164 // right spot. Give up after max_tries. Note that reserve_memory() will
3165 // automatically update _highest_vm_reserved_address if the call is
3166 // successful. The variable tracks the highest memory address every reserved
3167 // by JVM. It is used to detect heap-stack collision if running with
3168 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3169 // space than needed, it could confuse the collision detecting code. To
3170 // solve the problem, save current _highest_vm_reserved_address and
3171 // calculate the correct value before return.
3172 address old_highest = _highest_vm_reserved_address;
3173
3174 // Linux mmap allows caller to pass an address as hint; give it a try first,
3175 // if kernel honors the hint then we can return immediately.
3176 char * addr = anon_mmap(requested_addr, bytes, false);
3177 if (addr == requested_addr) {
3178 return requested_addr;
3179 }
3180
3181 if (addr != NULL) {
3182 // mmap() is successful but it fails to reserve at the requested address
3183 anon_munmap(addr, bytes);
3184 }
3185
3186 int i;
3187 for (i = 0; i < max_tries; ++i) {
3188 base[i] = reserve_memory(bytes);
3189
3190 if (base[i] != NULL) {
3191 // Is this the block we wanted?
3192 if (base[i] == requested_addr) {
3193 size[i] = bytes;
3194 break;
3195 }
3196
3197 // Does this overlap the block we wanted? Give back the overlapped
3198 // parts and try again.
3199
3200 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3201 if (top_overlap >= 0 && top_overlap < bytes) {
3202 unmap_memory(base[i], top_overlap);
3203 base[i] += top_overlap;
3204 size[i] = bytes - top_overlap;
3205 } else {
3206 size_t bottom_overlap = base[i] + bytes - requested_addr;
3207 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3208 unmap_memory(requested_addr, bottom_overlap);
3209 size[i] = bytes - bottom_overlap;
3210 } else {
3211 size[i] = bytes;
3212 }
3213 }
3214 }
3215 }
3216
3217 // Give back the unused reserved pieces.
3218
3219 for (int j = 0; j < i; ++j) {
3220 if (base[j] != NULL) {
3221 unmap_memory(base[j], size[j]);
3222 }
3223 }
3224
3225 if (i < max_tries) {
3226 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3227 return requested_addr;
3228 } else {
3229 _highest_vm_reserved_address = old_highest;
3230 return NULL;
3231 }
3232 }
3233
3234 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3235 return ::read(fd, buf, nBytes);
3236 }
3237
3238 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3239 // Solaris uses poll(), linux uses park().
3240 // Poll() is likely a better choice, assuming that Thread.interrupt()
3241 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3242 // SIGSEGV, see 4355769.
3243
3244 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3245 assert(thread == Thread::current(), "thread consistency check");
3246
3247 ParkEvent * const slp = thread->_SleepEvent ;
3248 slp->reset() ;
3249 OrderAccess::fence() ;
3250
3251 if (interruptible) {
3252 jlong prevtime = javaTimeNanos();
3253
3254 for (;;) {
3255 if (os::is_interrupted(thread, true)) {
3256 return OS_INTRPT;
3257 }
3258
3259 jlong newtime = javaTimeNanos();
3260
3261 if (newtime - prevtime < 0) {
3262 // time moving backwards, should only happen if no monotonic clock
3263 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3264 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3265 } else {
3266 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3267 }
3268
3269 if(millis <= 0) {
3270 return OS_OK;
3271 }
3272
3273 prevtime = newtime;
3274
3275 {
3276 assert(thread->is_Java_thread(), "sanity check");
3277 JavaThread *jt = (JavaThread *) thread;
3278 ThreadBlockInVM tbivm(jt);
3279 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3280
3281 jt->set_suspend_equivalent();
3282 // cleared by handle_special_suspend_equivalent_condition() or
3283 // java_suspend_self() via check_and_wait_while_suspended()
3284
3285 slp->park(millis);
3286
3287 // were we externally suspended while we were waiting?
3288 jt->check_and_wait_while_suspended();
3289 }
3290 }
3291 } else {
3292 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3293 jlong prevtime = javaTimeNanos();
3294
3295 for (;;) {
3296 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3297 // the 1st iteration ...
3298 jlong newtime = javaTimeNanos();
3299
3300 if (newtime - prevtime < 0) {
3301 // time moving backwards, should only happen if no monotonic clock
3302 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3303 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3304 } else {
3305 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3306 }
3307
3308 if(millis <= 0) break ;
3309
3310 prevtime = newtime;
3311 slp->park(millis);
3312 }
3313 return OS_OK ;
3314 }
3315 }
3316
3317 int os::naked_sleep() {
3318 // %% make the sleep time an integer flag. for now use 1 millisec.
3319 return os::sleep(Thread::current(), 1, false);
3320 }
3321
3322 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3323 void os::infinite_sleep() {
3324 while (true) { // sleep forever ...
3325 ::sleep(100); // ... 100 seconds at a time
3326 }
3327 }
3328
3329 // Used to convert frequent JVM_Yield() to nops
3330 bool os::dont_yield() {
3331 return DontYieldALot;
3332 }
3333
3334 void os::yield() {
3335 sched_yield();
3336 }
3337
3338 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3339
3340 void os::yield_all(int attempts) {
3341 // Yields to all threads, including threads with lower priorities
3342 // Threads on Linux are all with same priority. The Solaris style
3343 // os::yield_all() with nanosleep(1ms) is not necessary.
3344 sched_yield();
3345 }
3346
3347 // Called from the tight loops to possibly influence time-sharing heuristics
3348 void os::loop_breaker(int attempts) {
3349 os::yield_all(attempts);
3350 }
3351
3352 ////////////////////////////////////////////////////////////////////////////////
3353 // thread priority support
3354
3355 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3356 // only supports dynamic priority, static priority must be zero. For real-time
3357 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3358 // However, for large multi-threaded applications, SCHED_RR is not only slower
3359 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3360 // of 5 runs - Sep 2005).
3361 //
3362 // The following code actually changes the niceness of kernel-thread/LWP. It
3363 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3364 // not the entire user process, and user level threads are 1:1 mapped to kernel
3365 // threads. It has always been the case, but could change in the future. For
3366 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3367 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3368
3369 int os::java_to_os_priority[CriticalPriority + 1] = {
3370 19, // 0 Entry should never be used
3371
3372 4, // 1 MinPriority
3373 3, // 2
3374 2, // 3
3375
3376 1, // 4
3377 0, // 5 NormPriority
3378 -1, // 6
3379
3380 -2, // 7
3381 -3, // 8
3382 -4, // 9 NearMaxPriority
3383
3384 -5, // 10 MaxPriority
3385
3386 -5 // 11 CriticalPriority
3387 };
3388
3389 static int prio_init() {
3390 if (ThreadPriorityPolicy == 1) {
3391 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3392 // if effective uid is not root. Perhaps, a more elegant way of doing
3393 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3394 if (geteuid() != 0) {
3395 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3396 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3397 }
3398 ThreadPriorityPolicy = 0;
3399 }
3400 }
3401 if (UseCriticalJavaThreadPriority) {
3402 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3403 }
3404 return 0;
3405 }
3406
3407 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3408 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3409
3410 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3411 return (ret == 0) ? OS_OK : OS_ERR;
3412 }
3413
3414 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3415 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3416 *priority_ptr = java_to_os_priority[NormPriority];
3417 return OS_OK;
3418 }
3419
3420 errno = 0;
3421 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3422 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3423 }
3424
3425 // Hint to the underlying OS that a task switch would not be good.
3426 // Void return because it's a hint and can fail.
3427 void os::hint_no_preempt() {}
3428
3429 ////////////////////////////////////////////////////////////////////////////////
3430 // suspend/resume support
3431
3432 // the low-level signal-based suspend/resume support is a remnant from the
3433 // old VM-suspension that used to be for java-suspension, safepoints etc,
3434 // within hotspot. Now there is a single use-case for this:
3435 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3436 // that runs in the watcher thread.
3437 // The remaining code is greatly simplified from the more general suspension
3438 // code that used to be used.
3439 //
3440 // The protocol is quite simple:
3441 // - suspend:
3442 // - sends a signal to the target thread
3443 // - polls the suspend state of the osthread using a yield loop
3444 // - target thread signal handler (SR_handler) sets suspend state
3445 // and blocks in sigsuspend until continued
3446 // - resume:
3447 // - sets target osthread state to continue
3448 // - sends signal to end the sigsuspend loop in the SR_handler
3449 //
3450 // Note that the SR_lock plays no role in this suspend/resume protocol.
3451 //
3452
3453 static void resume_clear_context(OSThread *osthread) {
3454 osthread->set_ucontext(NULL);
3455 osthread->set_siginfo(NULL);
3456
3457 // notify the suspend action is completed, we have now resumed
3458 osthread->sr.clear_suspended();
3459 }
3460
3461 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3462 osthread->set_ucontext(context);
3463 osthread->set_siginfo(siginfo);
3464 }
3465
3466 //
3467 // Handler function invoked when a thread's execution is suspended or
3468 // resumed. We have to be careful that only async-safe functions are
3469 // called here (Note: most pthread functions are not async safe and
3470 // should be avoided.)
3471 //
3472 // Note: sigwait() is a more natural fit than sigsuspend() from an
3473 // interface point of view, but sigwait() prevents the signal hander
3474 // from being run. libpthread would get very confused by not having
3475 // its signal handlers run and prevents sigwait()'s use with the
3476 // mutex granting granting signal.
3477 //
3478 // Currently only ever called on the VMThread
3479 //
3480 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3481 // Save and restore errno to avoid confusing native code with EINTR
3482 // after sigsuspend.
3483 int old_errno = errno;
3484
3485 Thread* thread = Thread::current();
3486 OSThread* osthread = thread->osthread();
3487 assert(thread->is_VM_thread(), "Must be VMThread");
3488 // read current suspend action
3489 int action = osthread->sr.suspend_action();
3490 if (action == SR_SUSPEND) {
3491 suspend_save_context(osthread, siginfo, context);
3492
3493 // Notify the suspend action is about to be completed. do_suspend()
3494 // waits until SR_SUSPENDED is set and then returns. We will wait
3495 // here for a resume signal and that completes the suspend-other
3496 // action. do_suspend/do_resume is always called as a pair from
3497 // the same thread - so there are no races
3498
3499 // notify the caller
3500 osthread->sr.set_suspended();
3501
3502 sigset_t suspend_set; // signals for sigsuspend()
3503
3504 // get current set of blocked signals and unblock resume signal
3505 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3506 sigdelset(&suspend_set, SR_signum);
3507
3508 // wait here until we are resumed
3509 do {
3510 sigsuspend(&suspend_set);
3511 // ignore all returns until we get a resume signal
3512 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3513
3514 resume_clear_context(osthread);
3515
3516 } else {
3517 assert(action == SR_CONTINUE, "unexpected sr action");
3518 // nothing special to do - just leave the handler
3519 }
3520
3521 errno = old_errno;
3522 }
3523
3524
3525 static int SR_initialize() {
3526 struct sigaction act;
3527 char *s;
3528 /* Get signal number to use for suspend/resume */
3529 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3530 int sig = ::strtol(s, 0, 10);
3531 if (sig > 0 || sig < _NSIG) {
3532 SR_signum = sig;
3533 }
3534 }
3535
3536 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3537 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3538
3539 sigemptyset(&SR_sigset);
3540 sigaddset(&SR_sigset, SR_signum);
3541
3542 /* Set up signal handler for suspend/resume */
3543 act.sa_flags = SA_RESTART|SA_SIGINFO;
3544 act.sa_handler = (void (*)(int)) SR_handler;
3545
3546 // SR_signum is blocked by default.
3547 // 4528190 - We also need to block pthread restart signal (32 on all
3548 // supported Linux platforms). Note that LinuxThreads need to block
3549 // this signal for all threads to work properly. So we don't have
3550 // to use hard-coded signal number when setting up the mask.
3551 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3552
3553 if (sigaction(SR_signum, &act, 0) == -1) {
3554 return -1;
3555 }
3556
3557 // Save signal flag
3558 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3559 return 0;
3560 }
3561
3562 static int SR_finalize() {
3563 return 0;
3564 }
3565
3566
3567 // returns true on success and false on error - really an error is fatal
3568 // but this seems the normal response to library errors
3569 static bool do_suspend(OSThread* osthread) {
3570 // mark as suspended and send signal
3571 osthread->sr.set_suspend_action(SR_SUSPEND);
3572 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3573 assert_status(status == 0, status, "pthread_kill");
3574
3575 // check status and wait until notified of suspension
3576 if (status == 0) {
3577 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3578 os::yield_all(i);
3579 }
3580 osthread->sr.set_suspend_action(SR_NONE);
3581 return true;
3582 }
3583 else {
3584 osthread->sr.set_suspend_action(SR_NONE);
3585 return false;
3586 }
3587 }
3588
3589 static void do_resume(OSThread* osthread) {
3590 assert(osthread->sr.is_suspended(), "thread should be suspended");
3591 osthread->sr.set_suspend_action(SR_CONTINUE);
3592
3593 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3594 assert_status(status == 0, status, "pthread_kill");
3595 // check status and wait unit notified of resumption
3596 if (status == 0) {
3597 for (int i = 0; osthread->sr.is_suspended(); i++) {
3598 os::yield_all(i);
3599 }
3600 }
3601 osthread->sr.set_suspend_action(SR_NONE);
3602 }
3603
3604 ////////////////////////////////////////////////////////////////////////////////
3605 // interrupt support
3606
3607 void os::interrupt(Thread* thread) {
3608 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3609 "possibility of dangling Thread pointer");
3610
3611 OSThread* osthread = thread->osthread();
3612
3613 if (!osthread->interrupted()) {
3614 osthread->set_interrupted(true);
3615 // More than one thread can get here with the same value of osthread,
3616 // resulting in multiple notifications. We do, however, want the store
3617 // to interrupted() to be visible to other threads before we execute unpark().
3618 OrderAccess::fence();
3619 ParkEvent * const slp = thread->_SleepEvent ;
3620 if (slp != NULL) slp->unpark() ;
3621 }
3622
3623 // For JSR166. Unpark even if interrupt status already was set
3624 if (thread->is_Java_thread())
3625 ((JavaThread*)thread)->parker()->unpark();
3626
3627 ParkEvent * ev = thread->_ParkEvent ;
3628 if (ev != NULL) ev->unpark() ;
3629
3630 }
3631
3632 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3633 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3634 "possibility of dangling Thread pointer");
3635
3636 OSThread* osthread = thread->osthread();
3637
3638 bool interrupted = osthread->interrupted();
3639
3640 if (interrupted && clear_interrupted) {
3641 osthread->set_interrupted(false);
3642 // consider thread->_SleepEvent->reset() ... optional optimization
3643 }
3644
3645 return interrupted;
3646 }
3647
3648 ///////////////////////////////////////////////////////////////////////////////////
3649 // signal handling (except suspend/resume)
3650
3651 // This routine may be used by user applications as a "hook" to catch signals.
3652 // The user-defined signal handler must pass unrecognized signals to this
3653 // routine, and if it returns true (non-zero), then the signal handler must
3654 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3655 // routine will never retun false (zero), but instead will execute a VM panic
3656 // routine kill the process.
3657 //
3658 // If this routine returns false, it is OK to call it again. This allows
3659 // the user-defined signal handler to perform checks either before or after
3660 // the VM performs its own checks. Naturally, the user code would be making
3661 // a serious error if it tried to handle an exception (such as a null check
3662 // or breakpoint) that the VM was generating for its own correct operation.
3663 //
3664 // This routine may recognize any of the following kinds of signals:
3665 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3666 // It should be consulted by handlers for any of those signals.
3667 //
3668 // The caller of this routine must pass in the three arguments supplied
3669 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3670 // field of the structure passed to sigaction(). This routine assumes that
3671 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3672 //
3673 // Note that the VM will print warnings if it detects conflicting signal
3674 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3675 //
3676 extern "C" JNIEXPORT int
3677 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3678 void* ucontext, int abort_if_unrecognized);
3679
3680 void signalHandler(int sig, siginfo_t* info, void* uc) {
3681 assert(info != NULL && uc != NULL, "it must be old kernel");
3682 JVM_handle_linux_signal(sig, info, uc, true);
3683 }
3684
3685
3686 // This boolean allows users to forward their own non-matching signals
3687 // to JVM_handle_linux_signal, harmlessly.
3688 bool os::Linux::signal_handlers_are_installed = false;
3689
3690 // For signal-chaining
3691 struct sigaction os::Linux::sigact[MAXSIGNUM];
3692 unsigned int os::Linux::sigs = 0;
3693 bool os::Linux::libjsig_is_loaded = false;
3694 typedef struct sigaction *(*get_signal_t)(int);
3695 get_signal_t os::Linux::get_signal_action = NULL;
3696
3697 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3698 struct sigaction *actp = NULL;
3699
3700 if (libjsig_is_loaded) {
3701 // Retrieve the old signal handler from libjsig
3702 actp = (*get_signal_action)(sig);
3703 }
3704 if (actp == NULL) {
3705 // Retrieve the preinstalled signal handler from jvm
3706 actp = get_preinstalled_handler(sig);
3707 }
3708
3709 return actp;
3710 }
3711
3712 static bool call_chained_handler(struct sigaction *actp, int sig,
3713 siginfo_t *siginfo, void *context) {
3714 // Call the old signal handler
3715 if (actp->sa_handler == SIG_DFL) {
3716 // It's more reasonable to let jvm treat it as an unexpected exception
3717 // instead of taking the default action.
3718 return false;
3719 } else if (actp->sa_handler != SIG_IGN) {
3720 if ((actp->sa_flags & SA_NODEFER) == 0) {
3721 // automaticlly block the signal
3722 sigaddset(&(actp->sa_mask), sig);
3723 }
3724
3725 sa_handler_t hand;
3726 sa_sigaction_t sa;
3727 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3728 // retrieve the chained handler
3729 if (siginfo_flag_set) {
3730 sa = actp->sa_sigaction;
3731 } else {
3732 hand = actp->sa_handler;
3733 }
3734
3735 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3736 actp->sa_handler = SIG_DFL;
3737 }
3738
3739 // try to honor the signal mask
3740 sigset_t oset;
3741 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3742
3743 // call into the chained handler
3744 if (siginfo_flag_set) {
3745 (*sa)(sig, siginfo, context);
3746 } else {
3747 (*hand)(sig);
3748 }
3749
3750 // restore the signal mask
3751 pthread_sigmask(SIG_SETMASK, &oset, 0);
3752 }
3753 // Tell jvm's signal handler the signal is taken care of.
3754 return true;
3755 }
3756
3757 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3758 bool chained = false;
3759 // signal-chaining
3760 if (UseSignalChaining) {
3761 struct sigaction *actp = get_chained_signal_action(sig);
3762 if (actp != NULL) {
3763 chained = call_chained_handler(actp, sig, siginfo, context);
3764 }
3765 }
3766 return chained;
3767 }
3768
3769 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3770 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3771 return &sigact[sig];
3772 }
3773 return NULL;
3774 }
3775
3776 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3777 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3778 sigact[sig] = oldAct;
3779 sigs |= (unsigned int)1 << sig;
3780 }
3781
3782 // for diagnostic
3783 int os::Linux::sigflags[MAXSIGNUM];
3784
3785 int os::Linux::get_our_sigflags(int sig) {
3786 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3787 return sigflags[sig];
3788 }
3789
3790 void os::Linux::set_our_sigflags(int sig, int flags) {
3791 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3792 sigflags[sig] = flags;
3793 }
3794
3795 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3796 // Check for overwrite.
3797 struct sigaction oldAct;
3798 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3799
3800 void* oldhand = oldAct.sa_sigaction
3801 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3802 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3803 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3804 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3805 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3806 if (AllowUserSignalHandlers || !set_installed) {
3807 // Do not overwrite; user takes responsibility to forward to us.
3808 return;
3809 } else if (UseSignalChaining) {
3810 // save the old handler in jvm
3811 save_preinstalled_handler(sig, oldAct);
3812 // libjsig also interposes the sigaction() call below and saves the
3813 // old sigaction on it own.
3814 } else {
3815 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3816 "%#lx for signal %d.", (long)oldhand, sig));
3817 }
3818 }
3819
3820 struct sigaction sigAct;
3821 sigfillset(&(sigAct.sa_mask));
3822 sigAct.sa_handler = SIG_DFL;
3823 if (!set_installed) {
3824 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3825 } else {
3826 sigAct.sa_sigaction = signalHandler;
3827 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3828 }
3829 // Save flags, which are set by ours
3830 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3831 sigflags[sig] = sigAct.sa_flags;
3832
3833 int ret = sigaction(sig, &sigAct, &oldAct);
3834 assert(ret == 0, "check");
3835
3836 void* oldhand2 = oldAct.sa_sigaction
3837 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3838 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3839 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3840 }
3841
3842 // install signal handlers for signals that HotSpot needs to
3843 // handle in order to support Java-level exception handling.
3844
3845 void os::Linux::install_signal_handlers() {
3846 if (!signal_handlers_are_installed) {
3847 signal_handlers_are_installed = true;
3848
3849 // signal-chaining
3850 typedef void (*signal_setting_t)();
3851 signal_setting_t begin_signal_setting = NULL;
3852 signal_setting_t end_signal_setting = NULL;
3853 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3854 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3855 if (begin_signal_setting != NULL) {
3856 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3857 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3858 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3859 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3860 libjsig_is_loaded = true;
3861 assert(UseSignalChaining, "should enable signal-chaining");
3862 }
3863 if (libjsig_is_loaded) {
3864 // Tell libjsig jvm is setting signal handlers
3865 (*begin_signal_setting)();
3866 }
3867
3868 set_signal_handler(SIGSEGV, true);
3869 set_signal_handler(SIGPIPE, true);
3870 set_signal_handler(SIGBUS, true);
3871 set_signal_handler(SIGILL, true);
3872 set_signal_handler(SIGFPE, true);
3873 set_signal_handler(SIGXFSZ, true);
3874
3875 if (libjsig_is_loaded) {
3876 // Tell libjsig jvm finishes setting signal handlers
3877 (*end_signal_setting)();
3878 }
3879
3880 // We don't activate signal checker if libjsig is in place, we trust ourselves
3881 // and if UserSignalHandler is installed all bets are off.
3882 // Log that signal checking is off only if -verbose:jni is specified.
3883 if (CheckJNICalls) {
3884 if (libjsig_is_loaded) {
3885 if (PrintJNIResolving) {
3886 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3887 }
3888 check_signals = false;
3889 }
3890 if (AllowUserSignalHandlers) {
3891 if (PrintJNIResolving) {
3892 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3893 }
3894 check_signals = false;
3895 }
3896 }
3897 }
3898 }
3899
3900 // This is the fastest way to get thread cpu time on Linux.
3901 // Returns cpu time (user+sys) for any thread, not only for current.
3902 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3903 // It might work on 2.6.10+ with a special kernel/glibc patch.
3904 // For reference, please, see IEEE Std 1003.1-2004:
3905 // http://www.unix.org/single_unix_specification
3906
3907 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3908 struct timespec tp;
3909 int rc = os::Linux::clock_gettime(clockid, &tp);
3910 assert(rc == 0, "clock_gettime is expected to return 0 code");
3911
3912 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
3913 }
3914
3915 /////
3916 // glibc on Linux platform uses non-documented flag
3917 // to indicate, that some special sort of signal
3918 // trampoline is used.
3919 // We will never set this flag, and we should
3920 // ignore this flag in our diagnostic
3921 #ifdef SIGNIFICANT_SIGNAL_MASK
3922 #undef SIGNIFICANT_SIGNAL_MASK
3923 #endif
3924 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3925
3926 static const char* get_signal_handler_name(address handler,
3927 char* buf, int buflen) {
3928 int offset;
3929 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3930 if (found) {
3931 // skip directory names
3932 const char *p1, *p2;
3933 p1 = buf;
3934 size_t len = strlen(os::file_separator());
3935 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3936 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3937 } else {
3938 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3939 }
3940 return buf;
3941 }
3942
3943 static void print_signal_handler(outputStream* st, int sig,
3944 char* buf, size_t buflen) {
3945 struct sigaction sa;
3946
3947 sigaction(sig, NULL, &sa);
3948
3949 // See comment for SIGNIFICANT_SIGNAL_MASK define
3950 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3951
3952 st->print("%s: ", os::exception_name(sig, buf, buflen));
3953
3954 address handler = (sa.sa_flags & SA_SIGINFO)
3955 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3956 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3957
3958 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3959 st->print("SIG_DFL");
3960 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3961 st->print("SIG_IGN");
3962 } else {
3963 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3964 }
3965
3966 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3967
3968 address rh = VMError::get_resetted_sighandler(sig);
3969 // May be, handler was resetted by VMError?
3970 if(rh != NULL) {
3971 handler = rh;
3972 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3973 }
3974
3975 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3976
3977 // Check: is it our handler?
3978 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3979 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3980 // It is our signal handler
3981 // check for flags, reset system-used one!
3982 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3983 st->print(
3984 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3985 os::Linux::get_our_sigflags(sig));
3986 }
3987 }
3988 st->cr();
3989 }
3990
3991
3992 #define DO_SIGNAL_CHECK(sig) \
3993 if (!sigismember(&check_signal_done, sig)) \
3994 os::Linux::check_signal_handler(sig)
3995
3996 // This method is a periodic task to check for misbehaving JNI applications
3997 // under CheckJNI, we can add any periodic checks here
3998
3999 void os::run_periodic_checks() {
4000
4001 if (check_signals == false) return;
4002
4003 // SEGV and BUS if overridden could potentially prevent
4004 // generation of hs*.log in the event of a crash, debugging
4005 // such a case can be very challenging, so we absolutely
4006 // check the following for a good measure:
4007 DO_SIGNAL_CHECK(SIGSEGV);
4008 DO_SIGNAL_CHECK(SIGILL);
4009 DO_SIGNAL_CHECK(SIGFPE);
4010 DO_SIGNAL_CHECK(SIGBUS);
4011 DO_SIGNAL_CHECK(SIGPIPE);
4012 DO_SIGNAL_CHECK(SIGXFSZ);
4013
4014
4015 // ReduceSignalUsage allows the user to override these handlers
4016 // see comments at the very top and jvm_solaris.h
4017 if (!ReduceSignalUsage) {
4018 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4019 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4020 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4021 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4022 }
4023
4024 DO_SIGNAL_CHECK(SR_signum);
4025 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4026 }
4027
4028 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4029
4030 static os_sigaction_t os_sigaction = NULL;
4031
4032 void os::Linux::check_signal_handler(int sig) {
4033 char buf[O_BUFLEN];
4034 address jvmHandler = NULL;
4035
4036
4037 struct sigaction act;
4038 if (os_sigaction == NULL) {
4039 // only trust the default sigaction, in case it has been interposed
4040 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4041 if (os_sigaction == NULL) return;
4042 }
4043
4044 os_sigaction(sig, (struct sigaction*)NULL, &act);
4045
4046
4047 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4048
4049 address thisHandler = (act.sa_flags & SA_SIGINFO)
4050 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4051 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4052
4053
4054 switch(sig) {
4055 case SIGSEGV:
4056 case SIGBUS:
4057 case SIGFPE:
4058 case SIGPIPE:
4059 case SIGILL:
4060 case SIGXFSZ:
4061 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4062 break;
4063
4064 case SHUTDOWN1_SIGNAL:
4065 case SHUTDOWN2_SIGNAL:
4066 case SHUTDOWN3_SIGNAL:
4067 case BREAK_SIGNAL:
4068 jvmHandler = (address)user_handler();
4069 break;
4070
4071 case INTERRUPT_SIGNAL:
4072 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4073 break;
4074
4075 default:
4076 if (sig == SR_signum) {
4077 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4078 } else {
4079 return;
4080 }
4081 break;
4082 }
4083
4084 if (thisHandler != jvmHandler) {
4085 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4086 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4087 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4088 // No need to check this sig any longer
4089 sigaddset(&check_signal_done, sig);
4090 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4091 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4092 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4093 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4094 // No need to check this sig any longer
4095 sigaddset(&check_signal_done, sig);
4096 }
4097
4098 // Dump all the signal
4099 if (sigismember(&check_signal_done, sig)) {
4100 print_signal_handlers(tty, buf, O_BUFLEN);
4101 }
4102 }
4103
4104 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4105
4106 extern bool signal_name(int signo, char* buf, size_t len);
4107
4108 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4109 if (0 < exception_code && exception_code <= SIGRTMAX) {
4110 // signal
4111 if (!signal_name(exception_code, buf, size)) {
4112 jio_snprintf(buf, size, "SIG%d", exception_code);
4113 }
4114 return buf;
4115 } else {
4116 return NULL;
4117 }
4118 }
4119
4120 // this is called _before_ the most of global arguments have been parsed
4121 void os::init(void) {
4122 char dummy; /* used to get a guess on initial stack address */
4123 // first_hrtime = gethrtime();
4124
4125 // With LinuxThreads the JavaMain thread pid (primordial thread)
4126 // is different than the pid of the java launcher thread.
4127 // So, on Linux, the launcher thread pid is passed to the VM
4128 // via the sun.java.launcher.pid property.
4129 // Use this property instead of getpid() if it was correctly passed.
4130 // See bug 6351349.
4131 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4132
4133 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4134
4135 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4136
4137 init_random(1234567);
4138
4139 ThreadCritical::initialize();
4140
4141 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4142 if (Linux::page_size() == -1) {
4143 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4144 strerror(errno)));
4145 }
4146 init_page_sizes((size_t) Linux::page_size());
4147
4148 Linux::initialize_system_info();
4149
4150 // main_thread points to the aboriginal thread
4151 Linux::_main_thread = pthread_self();
4152
4153 Linux::clock_init();
4154 initial_time_count = os::elapsed_counter();
4155 pthread_mutex_init(&dl_mutex, NULL);
4156 }
4157
4158 // To install functions for atexit system call
4159 extern "C" {
4160 static void perfMemory_exit_helper() {
4161 perfMemory_exit();
4162 }
4163 }
4164
4165 // this is called _after_ the global arguments have been parsed
4166 jint os::init_2(void)
4167 {
4168 Linux::fast_thread_clock_init();
4169
4170 // Allocate a single page and mark it as readable for safepoint polling
4171 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4172 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4173
4174 os::set_polling_page( polling_page );
4175
4176 #ifndef PRODUCT
4177 if(Verbose && PrintMiscellaneous)
4178 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4179 #endif
4180
4181 if (!UseMembar) {
4182 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4183 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4184 os::set_memory_serialize_page( mem_serialize_page );
4185
4186 #ifndef PRODUCT
4187 if(Verbose && PrintMiscellaneous)
4188 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4189 #endif
4190 }
4191
4192 os::large_page_init();
4193
4194 // initialize suspend/resume support - must do this before signal_sets_init()
4195 if (SR_initialize() != 0) {
4196 perror("SR_initialize failed");
4197 return JNI_ERR;
4198 }
4199
4200 Linux::signal_sets_init();
4201 Linux::install_signal_handlers();
4202
4203 // Check minimum allowable stack size for thread creation and to initialize
4204 // the java system classes, including StackOverflowError - depends on page
4205 // size. Add a page for compiler2 recursion in main thread.
4206 // Add in 2*BytesPerWord times page size to account for VM stack during
4207 // class initialization depending on 32 or 64 bit VM.
4208 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4209 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4210 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4211
4212 size_t threadStackSizeInBytes = ThreadStackSize * K;
4213 if (threadStackSizeInBytes != 0 &&
4214 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4215 tty->print_cr("\nThe stack size specified is too small, "
4216 "Specify at least %dk",
4217 os::Linux::min_stack_allowed/ K);
4218 return JNI_ERR;
4219 }
4220
4221 // Make the stack size a multiple of the page size so that
4222 // the yellow/red zones can be guarded.
4223 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4224 vm_page_size()));
4225
4226 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4227
4228 Linux::libpthread_init();
4229 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4230 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4231 Linux::glibc_version(), Linux::libpthread_version(),
4232 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4233 }
4234
4235 if (UseNUMA) {
4236 if (!Linux::libnuma_init()) {
4237 UseNUMA = false;
4238 } else {
4239 if ((Linux::numa_max_node() < 1)) {
4240 // There's only one node(they start from 0), disable NUMA.
4241 UseNUMA = false;
4242 }
4243 }
4244 // With SHM large pages we cannot uncommit a page, so there's not way
4245 // we can make the adaptive lgrp chunk resizing work. If the user specified
4246 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4247 // disable adaptive resizing.
4248 if (UseNUMA && UseLargePages && UseSHM) {
4249 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4250 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4251 UseLargePages = false;
4252 } else {
4253 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4254 UseAdaptiveSizePolicy = false;
4255 UseAdaptiveNUMAChunkSizing = false;
4256 }
4257 } else {
4258 UseNUMA = false;
4259 }
4260 }
4261 if (!UseNUMA && ForceNUMA) {
4262 UseNUMA = true;
4263 }
4264 }
4265
4266 if (MaxFDLimit) {
4267 // set the number of file descriptors to max. print out error
4268 // if getrlimit/setrlimit fails but continue regardless.
4269 struct rlimit nbr_files;
4270 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4271 if (status != 0) {
4272 if (PrintMiscellaneous && (Verbose || WizardMode))
4273 perror("os::init_2 getrlimit failed");
4274 } else {
4275 nbr_files.rlim_cur = nbr_files.rlim_max;
4276 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4277 if (status != 0) {
4278 if (PrintMiscellaneous && (Verbose || WizardMode))
4279 perror("os::init_2 setrlimit failed");
4280 }
4281 }
4282 }
4283
4284 // Initialize lock used to serialize thread creation (see os::create_thread)
4285 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4286
4287 // at-exit methods are called in the reverse order of their registration.
4288 // atexit functions are called on return from main or as a result of a
4289 // call to exit(3C). There can be only 32 of these functions registered
4290 // and atexit() does not set errno.
4291
4292 if (PerfAllowAtExitRegistration) {
4293 // only register atexit functions if PerfAllowAtExitRegistration is set.
4294 // atexit functions can be delayed until process exit time, which
4295 // can be problematic for embedded VM situations. Embedded VMs should
4296 // call DestroyJavaVM() to assure that VM resources are released.
4297
4298 // note: perfMemory_exit_helper atexit function may be removed in
4299 // the future if the appropriate cleanup code can be added to the
4300 // VM_Exit VMOperation's doit method.
4301 if (atexit(perfMemory_exit_helper) != 0) {
4302 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4303 }
4304 }
4305
4306 // initialize thread priority policy
4307 prio_init();
4308
4309 return JNI_OK;
4310 }
4311
4312 // this is called at the end of vm_initialization
4313 void os::init_3(void)
4314 {
4315 #ifdef JAVASE_EMBEDDED
4316 // Start the MemNotifyThread
4317 if (LowMemoryProtection) {
4318 MemNotifyThread::start();
4319 }
4320 return;
4321 #endif
4322 }
4323
4324 // Mark the polling page as unreadable
4325 void os::make_polling_page_unreadable(void) {
4326 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4327 fatal("Could not disable polling page");
4328 };
4329
4330 // Mark the polling page as readable
4331 void os::make_polling_page_readable(void) {
4332 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4333 fatal("Could not enable polling page");
4334 }
4335 };
4336
4337 int os::active_processor_count() {
4338 // Linux doesn't yet have a (official) notion of processor sets,
4339 // so just return the number of online processors.
4340 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4341 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4342 return online_cpus;
4343 }
4344
4345 void os::set_native_thread_name(const char *name) {
4346 // Not yet implemented.
4347 return;
4348 }
4349
4350 bool os::distribute_processes(uint length, uint* distribution) {
4351 // Not yet implemented.
4352 return false;
4353 }
4354
4355 bool os::bind_to_processor(uint processor_id) {
4356 // Not yet implemented.
4357 return false;
4358 }
4359
4360 ///
4361
4362 // Suspends the target using the signal mechanism and then grabs the PC before
4363 // resuming the target. Used by the flat-profiler only
4364 ExtendedPC os::get_thread_pc(Thread* thread) {
4365 // Make sure that it is called by the watcher for the VMThread
4366 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4367 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4368
4369 ExtendedPC epc;
4370
4371 OSThread* osthread = thread->osthread();
4372 if (do_suspend(osthread)) {
4373 if (osthread->ucontext() != NULL) {
4374 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4375 } else {
4376 // NULL context is unexpected, double-check this is the VMThread
4377 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4378 }
4379 do_resume(osthread);
4380 }
4381 // failure means pthread_kill failed for some reason - arguably this is
4382 // a fatal problem, but such problems are ignored elsewhere
4383
4384 return epc;
4385 }
4386
4387 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4388 {
4389 if (is_NPTL()) {
4390 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4391 } else {
4392 #ifndef IA64
4393 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4394 // word back to default 64bit precision if condvar is signaled. Java
4395 // wants 53bit precision. Save and restore current value.
4396 int fpu = get_fpu_control_word();
4397 #endif // IA64
4398 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4399 #ifndef IA64
4400 set_fpu_control_word(fpu);
4401 #endif // IA64
4402 return status;
4403 }
4404 }
4405
4406 ////////////////////////////////////////////////////////////////////////////////
4407 // debug support
4408
4409 static address same_page(address x, address y) {
4410 int page_bits = -os::vm_page_size();
4411 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4412 return x;
4413 else if (x > y)
4414 return (address)(intptr_t(y) | ~page_bits) + 1;
4415 else
4416 return (address)(intptr_t(y) & page_bits);
4417 }
4418
4419 bool os::find(address addr, outputStream* st) {
4420 Dl_info dlinfo;
4421 memset(&dlinfo, 0, sizeof(dlinfo));
4422 if (dladdr(addr, &dlinfo)) {
4423 st->print(PTR_FORMAT ": ", addr);
4424 if (dlinfo.dli_sname != NULL) {
4425 st->print("%s+%#x", dlinfo.dli_sname,
4426 addr - (intptr_t)dlinfo.dli_saddr);
4427 } else if (dlinfo.dli_fname) {
4428 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4429 } else {
4430 st->print("<absolute address>");
4431 }
4432 if (dlinfo.dli_fname) {
4433 st->print(" in %s", dlinfo.dli_fname);
4434 }
4435 if (dlinfo.dli_fbase) {
4436 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4437 }
4438 st->cr();
4439
4440 if (Verbose) {
4441 // decode some bytes around the PC
4442 address begin = same_page(addr-40, addr);
4443 address end = same_page(addr+40, addr);
4444 address lowest = (address) dlinfo.dli_sname;
4445 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4446 if (begin < lowest) begin = lowest;
4447 Dl_info dlinfo2;
4448 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4449 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4450 end = (address) dlinfo2.dli_saddr;
4451 Disassembler::decode(begin, end, st);
4452 }
4453 return true;
4454 }
4455 return false;
4456 }
4457
4458 ////////////////////////////////////////////////////////////////////////////////
4459 // misc
4460
4461 // This does not do anything on Linux. This is basically a hook for being
4462 // able to use structured exception handling (thread-local exception filters)
4463 // on, e.g., Win32.
4464 void
4465 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4466 JavaCallArguments* args, Thread* thread) {
4467 f(value, method, args, thread);
4468 }
4469
4470 void os::print_statistics() {
4471 }
4472
4473 int os::message_box(const char* title, const char* message) {
4474 int i;
4475 fdStream err(defaultStream::error_fd());
4476 for (i = 0; i < 78; i++) err.print_raw("=");
4477 err.cr();
4478 err.print_raw_cr(title);
4479 for (i = 0; i < 78; i++) err.print_raw("-");
4480 err.cr();
4481 err.print_raw_cr(message);
4482 for (i = 0; i < 78; i++) err.print_raw("=");
4483 err.cr();
4484
4485 char buf[16];
4486 // Prevent process from exiting upon "read error" without consuming all CPU
4487 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4488
4489 return buf[0] == 'y' || buf[0] == 'Y';
4490 }
4491
4492 int os::stat(const char *path, struct stat *sbuf) {
4493 char pathbuf[MAX_PATH];
4494 if (strlen(path) > MAX_PATH - 1) {
4495 errno = ENAMETOOLONG;
4496 return -1;
4497 }
4498 os::native_path(strcpy(pathbuf, path));
4499 return ::stat(pathbuf, sbuf);
4500 }
4501
4502 bool os::check_heap(bool force) {
4503 return true;
4504 }
4505
4506 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4507 return ::vsnprintf(buf, count, format, args);
4508 }
4509
4510 // Is a (classpath) directory empty?
4511 bool os::dir_is_empty(const char* path) {
4512 DIR *dir = NULL;
4513 struct dirent *ptr;
4514
4515 dir = opendir(path);
4516 if (dir == NULL) return true;
4517
4518 /* Scan the directory */
4519 bool result = true;
4520 char buf[sizeof(struct dirent) + MAX_PATH];
4521 while (result && (ptr = ::readdir(dir)) != NULL) {
4522 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4523 result = false;
4524 }
4525 }
4526 closedir(dir);
4527 return result;
4528 }
4529
4530 // This code originates from JDK's sysOpen and open64_w
4531 // from src/solaris/hpi/src/system_md.c
4532
4533 #ifndef O_DELETE
4534 #define O_DELETE 0x10000
4535 #endif
4536
4537 // Open a file. Unlink the file immediately after open returns
4538 // if the specified oflag has the O_DELETE flag set.
4539 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4540
4541 int os::open(const char *path, int oflag, int mode) {
4542
4543 if (strlen(path) > MAX_PATH - 1) {
4544 errno = ENAMETOOLONG;
4545 return -1;
4546 }
4547 int fd;
4548 int o_delete = (oflag & O_DELETE);
4549 oflag = oflag & ~O_DELETE;
4550
4551 fd = ::open64(path, oflag, mode);
4552 if (fd == -1) return -1;
4553
4554 //If the open succeeded, the file might still be a directory
4555 {
4556 struct stat64 buf64;
4557 int ret = ::fstat64(fd, &buf64);
4558 int st_mode = buf64.st_mode;
4559
4560 if (ret != -1) {
4561 if ((st_mode & S_IFMT) == S_IFDIR) {
4562 errno = EISDIR;
4563 ::close(fd);
4564 return -1;
4565 }
4566 } else {
4567 ::close(fd);
4568 return -1;
4569 }
4570 }
4571
4572 /*
4573 * All file descriptors that are opened in the JVM and not
4574 * specifically destined for a subprocess should have the
4575 * close-on-exec flag set. If we don't set it, then careless 3rd
4576 * party native code might fork and exec without closing all
4577 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4578 * UNIXProcess.c), and this in turn might:
4579 *
4580 * - cause end-of-file to fail to be detected on some file
4581 * descriptors, resulting in mysterious hangs, or
4582 *
4583 * - might cause an fopen in the subprocess to fail on a system
4584 * suffering from bug 1085341.
4585 *
4586 * (Yes, the default setting of the close-on-exec flag is a Unix
4587 * design flaw)
4588 *
4589 * See:
4590 * 1085341: 32-bit stdio routines should support file descriptors >255
4591 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4592 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4593 */
4594 #ifdef FD_CLOEXEC
4595 {
4596 int flags = ::fcntl(fd, F_GETFD);
4597 if (flags != -1)
4598 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4599 }
4600 #endif
4601
4602 if (o_delete != 0) {
4603 ::unlink(path);
4604 }
4605 return fd;
4606 }
4607
4608
4609 // create binary file, rewriting existing file if required
4610 int os::create_binary_file(const char* path, bool rewrite_existing) {
4611 int oflags = O_WRONLY | O_CREAT;
4612 if (!rewrite_existing) {
4613 oflags |= O_EXCL;
4614 }
4615 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4616 }
4617
4618 // return current position of file pointer
4619 jlong os::current_file_offset(int fd) {
4620 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4621 }
4622
4623 // move file pointer to the specified offset
4624 jlong os::seek_to_file_offset(int fd, jlong offset) {
4625 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4626 }
4627
4628 // This code originates from JDK's sysAvailable
4629 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4630
4631 int os::available(int fd, jlong *bytes) {
4632 jlong cur, end;
4633 int mode;
4634 struct stat64 buf64;
4635
4636 if (::fstat64(fd, &buf64) >= 0) {
4637 mode = buf64.st_mode;
4638 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4639 /*
4640 * XXX: is the following call interruptible? If so, this might
4641 * need to go through the INTERRUPT_IO() wrapper as for other
4642 * blocking, interruptible calls in this file.
4643 */
4644 int n;
4645 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4646 *bytes = n;
4647 return 1;
4648 }
4649 }
4650 }
4651 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4652 return 0;
4653 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4654 return 0;
4655 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4656 return 0;
4657 }
4658 *bytes = end - cur;
4659 return 1;
4660 }
4661
4662 int os::socket_available(int fd, jint *pbytes) {
4663 // Linux doc says EINTR not returned, unlike Solaris
4664 int ret = ::ioctl(fd, FIONREAD, pbytes);
4665
4666 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4667 // is expected to return 0 on failure and 1 on success to the jdk.
4668 return (ret < 0) ? 0 : 1;
4669 }
4670
4671 // Map a block of memory.
4672 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4673 char *addr, size_t bytes, bool read_only,
4674 bool allow_exec) {
4675 int prot;
4676 int flags = MAP_PRIVATE;
4677
4678 if (read_only) {
4679 prot = PROT_READ;
4680 } else {
4681 prot = PROT_READ | PROT_WRITE;
4682 }
4683
4684 if (allow_exec) {
4685 prot |= PROT_EXEC;
4686 }
4687
4688 if (addr != NULL) {
4689 flags |= MAP_FIXED;
4690 }
4691
4692 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4693 fd, file_offset);
4694 if (mapped_address == MAP_FAILED) {
4695 return NULL;
4696 }
4697 return mapped_address;
4698 }
4699
4700
4701 // Remap a block of memory.
4702 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4703 char *addr, size_t bytes, bool read_only,
4704 bool allow_exec) {
4705 // same as map_memory() on this OS
4706 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4707 allow_exec);
4708 }
4709
4710
4711 // Unmap a block of memory.
4712 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4713 return munmap(addr, bytes) == 0;
4714 }
4715
4716 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4717
4718 static clockid_t thread_cpu_clockid(Thread* thread) {
4719 pthread_t tid = thread->osthread()->pthread_id();
4720 clockid_t clockid;
4721
4722 // Get thread clockid
4723 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4724 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4725 return clockid;
4726 }
4727
4728 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4729 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4730 // of a thread.
4731 //
4732 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4733 // the fast estimate available on the platform.
4734
4735 jlong os::current_thread_cpu_time() {
4736 if (os::Linux::supports_fast_thread_cpu_time()) {
4737 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4738 } else {
4739 // return user + sys since the cost is the same
4740 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4741 }
4742 }
4743
4744 jlong os::thread_cpu_time(Thread* thread) {
4745 // consistent with what current_thread_cpu_time() returns
4746 if (os::Linux::supports_fast_thread_cpu_time()) {
4747 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4748 } else {
4749 return slow_thread_cpu_time(thread, true /* user + sys */);
4750 }
4751 }
4752
4753 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4754 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4755 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4756 } else {
4757 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4758 }
4759 }
4760
4761 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4762 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4763 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4764 } else {
4765 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4766 }
4767 }
4768
4769 //
4770 // -1 on error.
4771 //
4772
4773 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4774 static bool proc_pid_cpu_avail = true;
4775 static bool proc_task_unchecked = true;
4776 static const char *proc_stat_path = "/proc/%d/stat";
4777 pid_t tid = thread->osthread()->thread_id();
4778 int i;
4779 char *s;
4780 char stat[2048];
4781 int statlen;
4782 char proc_name[64];
4783 int count;
4784 long sys_time, user_time;
4785 char string[64];
4786 char cdummy;
4787 int idummy;
4788 long ldummy;
4789 FILE *fp;
4790
4791 // We first try accessing /proc/<pid>/cpu since this is faster to
4792 // process. If this file is not present (linux kernels 2.5 and above)
4793 // then we open /proc/<pid>/stat.
4794 if ( proc_pid_cpu_avail ) {
4795 sprintf(proc_name, "/proc/%d/cpu", tid);
4796 fp = fopen(proc_name, "r");
4797 if ( fp != NULL ) {
4798 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4799 fclose(fp);
4800 if ( count != 3 ) return -1;
4801
4802 if (user_sys_cpu_time) {
4803 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4804 } else {
4805 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4806 }
4807 }
4808 else proc_pid_cpu_avail = false;
4809 }
4810
4811 // The /proc/<tid>/stat aggregates per-process usage on
4812 // new Linux kernels 2.6+ where NPTL is supported.
4813 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4814 // See bug 6328462.
4815 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4816 // and possibly in some other cases, so we check its availability.
4817 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4818 // This is executed only once
4819 proc_task_unchecked = false;
4820 fp = fopen("/proc/self/task", "r");
4821 if (fp != NULL) {
4822 proc_stat_path = "/proc/self/task/%d/stat";
4823 fclose(fp);
4824 }
4825 }
4826
4827 sprintf(proc_name, proc_stat_path, tid);
4828 fp = fopen(proc_name, "r");
4829 if ( fp == NULL ) return -1;
4830 statlen = fread(stat, 1, 2047, fp);
4831 stat[statlen] = '\0';
4832 fclose(fp);
4833
4834 // Skip pid and the command string. Note that we could be dealing with
4835 // weird command names, e.g. user could decide to rename java launcher
4836 // to "java 1.4.2 :)", then the stat file would look like
4837 // 1234 (java 1.4.2 :)) R ... ...
4838 // We don't really need to know the command string, just find the last
4839 // occurrence of ")" and then start parsing from there. See bug 4726580.
4840 s = strrchr(stat, ')');
4841 i = 0;
4842 if (s == NULL ) return -1;
4843
4844 // Skip blank chars
4845 do s++; while (isspace(*s));
4846
4847 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4848 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4849 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4850 &user_time, &sys_time);
4851 if ( count != 13 ) return -1;
4852 if (user_sys_cpu_time) {
4853 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4854 } else {
4855 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4856 }
4857 }
4858
4859 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4860 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4861 info_ptr->may_skip_backward = false; // elapsed time not wall time
4862 info_ptr->may_skip_forward = false; // elapsed time not wall time
4863 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4864 }
4865
4866 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4867 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4868 info_ptr->may_skip_backward = false; // elapsed time not wall time
4869 info_ptr->may_skip_forward = false; // elapsed time not wall time
4870 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4871 }
4872
4873 bool os::is_thread_cpu_time_supported() {
4874 return true;
4875 }
4876
4877 // System loadavg support. Returns -1 if load average cannot be obtained.
4878 // Linux doesn't yet have a (official) notion of processor sets,
4879 // so just return the system wide load average.
4880 int os::loadavg(double loadavg[], int nelem) {
4881 return ::getloadavg(loadavg, nelem);
4882 }
4883
4884 void os::pause() {
4885 char filename[MAX_PATH];
4886 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4887 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4888 } else {
4889 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4890 }
4891
4892 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4893 if (fd != -1) {
4894 struct stat buf;
4895 ::close(fd);
4896 while (::stat(filename, &buf) == 0) {
4897 (void)::poll(NULL, 0, 100);
4898 }
4899 } else {
4900 jio_fprintf(stderr,
4901 "Could not open pause file '%s', continuing immediately.\n", filename);
4902 }
4903 }
4904
4905
4906 // Refer to the comments in os_solaris.cpp park-unpark.
4907 //
4908 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4909 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4910 // For specifics regarding the bug see GLIBC BUGID 261237 :
4911 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4912 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4913 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4914 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4915 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4916 // and monitorenter when we're using 1-0 locking. All those operations may result in
4917 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4918 // of libpthread avoids the problem, but isn't practical.
4919 //
4920 // Possible remedies:
4921 //
4922 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4923 // This is palliative and probabilistic, however. If the thread is preempted
4924 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4925 // than the minimum period may have passed, and the abstime may be stale (in the
4926 // past) resultin in a hang. Using this technique reduces the odds of a hang
4927 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4928 //
4929 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4930 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4931 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4932 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4933 // thread.
4934 //
4935 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4936 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4937 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4938 // This also works well. In fact it avoids kernel-level scalability impediments
4939 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4940 // timers in a graceful fashion.
4941 //
4942 // 4. When the abstime value is in the past it appears that control returns
4943 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4944 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4945 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4946 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4947 // It may be possible to avoid reinitialization by checking the return
4948 // value from pthread_cond_timedwait(). In addition to reinitializing the
4949 // condvar we must establish the invariant that cond_signal() is only called
4950 // within critical sections protected by the adjunct mutex. This prevents
4951 // cond_signal() from "seeing" a condvar that's in the midst of being
4952 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4953 // desirable signal-after-unlock optimization that avoids futile context switching.
4954 //
4955 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4956 // structure when a condvar is used or initialized. cond_destroy() would
4957 // release the helper structure. Our reinitialize-after-timedwait fix
4958 // put excessive stress on malloc/free and locks protecting the c-heap.
4959 //
4960 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4961 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4962 // and only enabling the work-around for vulnerable environments.
4963
4964 // utility to compute the abstime argument to timedwait:
4965 // millis is the relative timeout time
4966 // abstime will be the absolute timeout time
4967 // TODO: replace compute_abstime() with unpackTime()
4968
4969 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4970 if (millis < 0) millis = 0;
4971 struct timeval now;
4972 int status = gettimeofday(&now, NULL);
4973 assert(status == 0, "gettimeofday");
4974 jlong seconds = millis / 1000;
4975 millis %= 1000;
4976 if (seconds > 50000000) { // see man cond_timedwait(3T)
4977 seconds = 50000000;
4978 }
4979 abstime->tv_sec = now.tv_sec + seconds;
4980 long usec = now.tv_usec + millis * 1000;
4981 if (usec >= 1000000) {
4982 abstime->tv_sec += 1;
4983 usec -= 1000000;
4984 }
4985 abstime->tv_nsec = usec * 1000;
4986 return abstime;
4987 }
4988
4989
4990 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4991 // Conceptually TryPark() should be equivalent to park(0).
4992
4993 int os::PlatformEvent::TryPark() {
4994 for (;;) {
4995 const int v = _Event ;
4996 guarantee ((v == 0) || (v == 1), "invariant") ;
4997 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4998 }
4999 }
5000
5001 void os::PlatformEvent::park() { // AKA "down()"
5002 // Invariant: Only the thread associated with the Event/PlatformEvent
5003 // may call park().
5004 // TODO: assert that _Assoc != NULL or _Assoc == Self
5005 int v ;
5006 for (;;) {
5007 v = _Event ;
5008 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5009 }
5010 guarantee (v >= 0, "invariant") ;
5011 if (v == 0) {
5012 // Do this the hard way by blocking ...
5013 int status = pthread_mutex_lock(_mutex);
5014 assert_status(status == 0, status, "mutex_lock");
5015 guarantee (_nParked == 0, "invariant") ;
5016 ++ _nParked ;
5017 while (_Event < 0) {
5018 status = pthread_cond_wait(_cond, _mutex);
5019 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5020 // Treat this the same as if the wait was interrupted
5021 if (status == ETIME) { status = EINTR; }
5022 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5023 }
5024 -- _nParked ;
5025
5026 // In theory we could move the ST of 0 into _Event past the unlock(),
5027 // but then we'd need a MEMBAR after the ST.
5028 _Event = 0 ;
5029 status = pthread_mutex_unlock(_mutex);
5030 assert_status(status == 0, status, "mutex_unlock");
5031 }
5032 guarantee (_Event >= 0, "invariant") ;
5033 }
5034
5035 int os::PlatformEvent::park(jlong millis) {
5036 guarantee (_nParked == 0, "invariant") ;
5037
5038 int v ;
5039 for (;;) {
5040 v = _Event ;
5041 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5042 }
5043 guarantee (v >= 0, "invariant") ;
5044 if (v != 0) return OS_OK ;
5045
5046 // We do this the hard way, by blocking the thread.
5047 // Consider enforcing a minimum timeout value.
5048 struct timespec abst;
5049 compute_abstime(&abst, millis);
5050
5051 int ret = OS_TIMEOUT;
5052 int status = pthread_mutex_lock(_mutex);
5053 assert_status(status == 0, status, "mutex_lock");
5054 guarantee (_nParked == 0, "invariant") ;
5055 ++_nParked ;
5056
5057 // Object.wait(timo) will return because of
5058 // (a) notification
5059 // (b) timeout
5060 // (c) thread.interrupt
5061 //
5062 // Thread.interrupt and object.notify{All} both call Event::set.
5063 // That is, we treat thread.interrupt as a special case of notification.
5064 // The underlying Solaris implementation, cond_timedwait, admits
5065 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5066 // JVM from making those visible to Java code. As such, we must
5067 // filter out spurious wakeups. We assume all ETIME returns are valid.
5068 //
5069 // TODO: properly differentiate simultaneous notify+interrupt.
5070 // In that case, we should propagate the notify to another waiter.
5071
5072 while (_Event < 0) {
5073 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5074 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5075 pthread_cond_destroy (_cond);
5076 pthread_cond_init (_cond, NULL) ;
5077 }
5078 assert_status(status == 0 || status == EINTR ||
5079 status == ETIME || status == ETIMEDOUT,
5080 status, "cond_timedwait");
5081 if (!FilterSpuriousWakeups) break ; // previous semantics
5082 if (status == ETIME || status == ETIMEDOUT) break ;
5083 // We consume and ignore EINTR and spurious wakeups.
5084 }
5085 --_nParked ;
5086 if (_Event >= 0) {
5087 ret = OS_OK;
5088 }
5089 _Event = 0 ;
5090 status = pthread_mutex_unlock(_mutex);
5091 assert_status(status == 0, status, "mutex_unlock");
5092 assert (_nParked == 0, "invariant") ;
5093 return ret;
5094 }
5095
5096 void os::PlatformEvent::unpark() {
5097 int v, AnyWaiters ;
5098 for (;;) {
5099 v = _Event ;
5100 if (v > 0) {
5101 // The LD of _Event could have reordered or be satisfied
5102 // by a read-aside from this processor's write buffer.
5103 // To avoid problems execute a barrier and then
5104 // ratify the value.
5105 OrderAccess::fence() ;
5106 if (_Event == v) return ;
5107 continue ;
5108 }
5109 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5110 }
5111 if (v < 0) {
5112 // Wait for the thread associated with the event to vacate
5113 int status = pthread_mutex_lock(_mutex);
5114 assert_status(status == 0, status, "mutex_lock");
5115 AnyWaiters = _nParked ;
5116 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5117 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5118 AnyWaiters = 0 ;
5119 pthread_cond_signal (_cond);
5120 }
5121 status = pthread_mutex_unlock(_mutex);
5122 assert_status(status == 0, status, "mutex_unlock");
5123 if (AnyWaiters != 0) {
5124 status = pthread_cond_signal(_cond);
5125 assert_status(status == 0, status, "cond_signal");
5126 }
5127 }
5128
5129 // Note that we signal() _after dropping the lock for "immortal" Events.
5130 // This is safe and avoids a common class of futile wakeups. In rare
5131 // circumstances this can cause a thread to return prematurely from
5132 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5133 // simply re-test the condition and re-park itself.
5134 }
5135
5136
5137 // JSR166
5138 // -------------------------------------------------------
5139
5140 /*
5141 * The solaris and linux implementations of park/unpark are fairly
5142 * conservative for now, but can be improved. They currently use a
5143 * mutex/condvar pair, plus a a count.
5144 * Park decrements count if > 0, else does a condvar wait. Unpark
5145 * sets count to 1 and signals condvar. Only one thread ever waits
5146 * on the condvar. Contention seen when trying to park implies that someone
5147 * is unparking you, so don't wait. And spurious returns are fine, so there
5148 * is no need to track notifications.
5149 */
5150
5151 #define MAX_SECS 100000000
5152 /*
5153 * This code is common to linux and solaris and will be moved to a
5154 * common place in dolphin.
5155 *
5156 * The passed in time value is either a relative time in nanoseconds
5157 * or an absolute time in milliseconds. Either way it has to be unpacked
5158 * into suitable seconds and nanoseconds components and stored in the
5159 * given timespec structure.
5160 * Given time is a 64-bit value and the time_t used in the timespec is only
5161 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5162 * overflow if times way in the future are given. Further on Solaris versions
5163 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5164 * number of seconds, in abstime, is less than current_time + 100,000,000.
5165 * As it will be 28 years before "now + 100000000" will overflow we can
5166 * ignore overflow and just impose a hard-limit on seconds using the value
5167 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5168 * years from "now".
5169 */
5170
5171 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5172 assert (time > 0, "convertTime");
5173
5174 struct timeval now;
5175 int status = gettimeofday(&now, NULL);
5176 assert(status == 0, "gettimeofday");
5177
5178 time_t max_secs = now.tv_sec + MAX_SECS;
5179
5180 if (isAbsolute) {
5181 jlong secs = time / 1000;
5182 if (secs > max_secs) {
5183 absTime->tv_sec = max_secs;
5184 }
5185 else {
5186 absTime->tv_sec = secs;
5187 }
5188 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5189 }
5190 else {
5191 jlong secs = time / NANOSECS_PER_SEC;
5192 if (secs >= MAX_SECS) {
5193 absTime->tv_sec = max_secs;
5194 absTime->tv_nsec = 0;
5195 }
5196 else {
5197 absTime->tv_sec = now.tv_sec + secs;
5198 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5199 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5200 absTime->tv_nsec -= NANOSECS_PER_SEC;
5201 ++absTime->tv_sec; // note: this must be <= max_secs
5202 }
5203 }
5204 }
5205 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5206 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5207 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5208 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5209 }
5210
5211 void Parker::park(bool isAbsolute, jlong time) {
5212 // Optional fast-path check:
5213 // Return immediately if a permit is available.
5214 if (_counter > 0) {
5215 _counter = 0 ;
5216 OrderAccess::fence();
5217 return ;
5218 }
5219
5220 Thread* thread = Thread::current();
5221 assert(thread->is_Java_thread(), "Must be JavaThread");
5222 JavaThread *jt = (JavaThread *)thread;
5223
5224 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5225 // Check interrupt before trying to wait
5226 if (Thread::is_interrupted(thread, false)) {
5227 return;
5228 }
5229
5230 // Next, demultiplex/decode time arguments
5231 timespec absTime;
5232 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5233 return;
5234 }
5235 if (time > 0) {
5236 unpackTime(&absTime, isAbsolute, time);
5237 }
5238
5239
5240 // Enter safepoint region
5241 // Beware of deadlocks such as 6317397.
5242 // The per-thread Parker:: mutex is a classic leaf-lock.
5243 // In particular a thread must never block on the Threads_lock while
5244 // holding the Parker:: mutex. If safepoints are pending both the
5245 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5246 ThreadBlockInVM tbivm(jt);
5247
5248 // Don't wait if cannot get lock since interference arises from
5249 // unblocking. Also. check interrupt before trying wait
5250 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5251 return;
5252 }
5253
5254 int status ;
5255 if (_counter > 0) { // no wait needed
5256 _counter = 0;
5257 status = pthread_mutex_unlock(_mutex);
5258 assert (status == 0, "invariant") ;
5259 OrderAccess::fence();
5260 return;
5261 }
5262
5263 #ifdef ASSERT
5264 // Don't catch signals while blocked; let the running threads have the signals.
5265 // (This allows a debugger to break into the running thread.)
5266 sigset_t oldsigs;
5267 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5268 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5269 #endif
5270
5271 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5272 jt->set_suspend_equivalent();
5273 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5274
5275 if (time == 0) {
5276 status = pthread_cond_wait (_cond, _mutex) ;
5277 } else {
5278 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5279 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5280 pthread_cond_destroy (_cond) ;
5281 pthread_cond_init (_cond, NULL);
5282 }
5283 }
5284 assert_status(status == 0 || status == EINTR ||
5285 status == ETIME || status == ETIMEDOUT,
5286 status, "cond_timedwait");
5287
5288 #ifdef ASSERT
5289 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5290 #endif
5291
5292 _counter = 0 ;
5293 status = pthread_mutex_unlock(_mutex) ;
5294 assert_status(status == 0, status, "invariant") ;
5295 // If externally suspended while waiting, re-suspend
5296 if (jt->handle_special_suspend_equivalent_condition()) {
5297 jt->java_suspend_self();
5298 }
5299
5300 OrderAccess::fence();
5301 }
5302
5303 void Parker::unpark() {
5304 int s, status ;
5305 status = pthread_mutex_lock(_mutex);
5306 assert (status == 0, "invariant") ;
5307 s = _counter;
5308 _counter = 1;
5309 if (s < 1) {
5310 if (WorkAroundNPTLTimedWaitHang) {
5311 status = pthread_cond_signal (_cond) ;
5312 assert (status == 0, "invariant") ;
5313 status = pthread_mutex_unlock(_mutex);
5314 assert (status == 0, "invariant") ;
5315 } else {
5316 status = pthread_mutex_unlock(_mutex);
5317 assert (status == 0, "invariant") ;
5318 status = pthread_cond_signal (_cond) ;
5319 assert (status == 0, "invariant") ;
5320 }
5321 } else {
5322 pthread_mutex_unlock(_mutex);
5323 assert (status == 0, "invariant") ;
5324 }
5325 }
5326
5327
5328 extern char** environ;
5329
5330 #ifndef __NR_fork
5331 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5332 #endif
5333
5334 #ifndef __NR_execve
5335 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5336 #endif
5337
5338 // Run the specified command in a separate process. Return its exit value,
5339 // or -1 on failure (e.g. can't fork a new process).
5340 // Unlike system(), this function can be called from signal handler. It
5341 // doesn't block SIGINT et al.
5342 int os::fork_and_exec(char* cmd) {
5343 const char * argv[4] = {"sh", "-c", cmd, NULL};
5344
5345 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5346 // pthread_atfork handlers and reset pthread library. All we need is a
5347 // separate process to execve. Make a direct syscall to fork process.
5348 // On IA64 there's no fork syscall, we have to use fork() and hope for
5349 // the best...
5350 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5351 IA64_ONLY(fork();)
5352
5353 if (pid < 0) {
5354 // fork failed
5355 return -1;
5356
5357 } else if (pid == 0) {
5358 // child process
5359
5360 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5361 // first to kill every thread on the thread list. Because this list is
5362 // not reset by fork() (see notes above), execve() will instead kill
5363 // every thread in the parent process. We know this is the only thread
5364 // in the new process, so make a system call directly.
5365 // IA64 should use normal execve() from glibc to match the glibc fork()
5366 // above.
5367 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5368 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5369
5370 // execve failed
5371 _exit(-1);
5372
5373 } else {
5374 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5375 // care about the actual exit code, for now.
5376
5377 int status;
5378
5379 // Wait for the child process to exit. This returns immediately if
5380 // the child has already exited. */
5381 while (waitpid(pid, &status, 0) < 0) {
5382 switch (errno) {
5383 case ECHILD: return 0;
5384 case EINTR: break;
5385 default: return -1;
5386 }
5387 }
5388
5389 if (WIFEXITED(status)) {
5390 // The child exited normally; get its exit code.
5391 return WEXITSTATUS(status);
5392 } else if (WIFSIGNALED(status)) {
5393 // The child exited because of a signal
5394 // The best value to return is 0x80 + signal number,
5395 // because that is what all Unix shells do, and because
5396 // it allows callers to distinguish between process exit and
5397 // process death by signal.
5398 return 0x80 + WTERMSIG(status);
5399 } else {
5400 // Unknown exit code; pass it through
5401 return status;
5402 }
5403 }
5404 }
5405
5406 // is_headless_jre()
5407 //
5408 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5409 // in order to report if we are running in a headless jre
5410 //
5411 // Since JDK8 xawt/libmawt.so was moved into the same directory
5412 // as libawt.so, and renamed libawt_xawt.so
5413 //
5414 bool os::is_headless_jre() {
5415 struct stat statbuf;
5416 char buf[MAXPATHLEN];
5417 char libmawtpath[MAXPATHLEN];
5418 const char *xawtstr = "/xawt/libmawt.so";
5419 const char *new_xawtstr = "/libawt_xawt.so";
5420 char *p;
5421
5422 // Get path to libjvm.so
5423 os::jvm_path(buf, sizeof(buf));
5424
5425 // Get rid of libjvm.so
5426 p = strrchr(buf, '/');
5427 if (p == NULL) return false;
5428 else *p = '\0';
5429
5430 // Get rid of client or server
5431 p = strrchr(buf, '/');
5432 if (p == NULL) return false;
5433 else *p = '\0';
5434
5435 // check xawt/libmawt.so
5436 strcpy(libmawtpath, buf);
5437 strcat(libmawtpath, xawtstr);
5438 if (::stat(libmawtpath, &statbuf) == 0) return false;
5439
5440 // check libawt_xawt.so
5441 strcpy(libmawtpath, buf);
5442 strcat(libmawtpath, new_xawtstr);
5443 if (::stat(libmawtpath, &statbuf) == 0) return false;
5444
5445 return true;
5446 }
5447
5448 // Get the default path to the core file
5449 // Returns the length of the string
5450 int os::get_core_path(char* buffer, size_t bufferSize) {
5451 const char* p = get_current_directory(buffer, bufferSize);
5452
5453 if (p == NULL) {
5454 assert(p != NULL, "failed to get current directory");
5455 return 0;
5456 }
5457
5458 return strlen(buffer);
5459 }
5460
5461 #ifdef JAVASE_EMBEDDED
5462 //
5463 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5464 //
5465 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5466
5467 // ctor
5468 //
5469 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5470 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5471 _fd = fd;
5472
5473 if (os::create_thread(this, os::os_thread)) {
5474 _memnotify_thread = this;
5475 os::set_priority(this, NearMaxPriority);
5476 os::start_thread(this);
5477 }
5478 }
5479
5480 // Where all the work gets done
5481 //
5482 void MemNotifyThread::run() {
5483 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5484
5485 // Set up the select arguments
5486 fd_set rfds;
5487 if (_fd != -1) {
5488 FD_ZERO(&rfds);
5489 FD_SET(_fd, &rfds);
5490 }
5491
5492 // Now wait for the mem_notify device to wake up
5493 while (1) {
5494 // Wait for the mem_notify device to signal us..
5495 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5496 if (rc == -1) {
5497 perror("select!\n");
5498 break;
5499 } else if (rc) {
5500 //ssize_t free_before = os::available_memory();
5501 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5502
5503 // The kernel is telling us there is not much memory left...
5504 // try to do something about that
5505
5506 // If we are not already in a GC, try one.
5507 if (!Universe::heap()->is_gc_active()) {
5508 Universe::heap()->collect(GCCause::_allocation_failure);
5509
5510 //ssize_t free_after = os::available_memory();
5511 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5512 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5513 }
5514 // We might want to do something like the following if we find the GC's are not helping...
5515 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5516 }
5517 }
5518 }
5519
5520 //
5521 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5522 //
5523 void MemNotifyThread::start() {
5524 int fd;
5525 fd = open ("/dev/mem_notify", O_RDONLY, 0);
5526 if (fd < 0) {
5527 return;
5528 }
5529
5530 if (memnotify_thread() == NULL) {
5531 new MemNotifyThread(fd);
5532 }
5533 }
5534 #endif // JAVASE_EMBEDDED