1 /*
2 * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 # define __STDC_FORMAT_MACROS
26
27 // no precompiled headers
28 #include "classfile/classLoader.hpp"
29 #include "classfile/systemDictionary.hpp"
30 #include "classfile/vmSymbols.hpp"
31 #include "code/icBuffer.hpp"
32 #include "code/vtableStubs.hpp"
33 #include "compiler/compileBroker.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "jvm_linux.h"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "mutex_linux.inline.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/extendedPC.hpp"
46 #include "runtime/globals.hpp"
47 #include "runtime/interfaceSupport.hpp"
48 #include "runtime/java.hpp"
49 #include "runtime/javaCalls.hpp"
50 #include "runtime/mutexLocker.hpp"
51 #include "runtime/objectMonitor.hpp"
52 #include "runtime/osThread.hpp"
53 #include "runtime/perfMemory.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #include "runtime/statSampler.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "runtime/threadCritical.hpp"
58 #include "runtime/timer.hpp"
59 #include "services/attachListener.hpp"
60 #include "services/runtimeService.hpp"
61 #include "thread_linux.inline.hpp"
62 #include "utilities/decoder.hpp"
63 #include "utilities/defaultStream.hpp"
64 #include "utilities/events.hpp"
65 #include "utilities/growableArray.hpp"
66 #include "utilities/vmError.hpp"
67 #ifdef TARGET_ARCH_x86
68 # include "assembler_x86.inline.hpp"
69 # include "nativeInst_x86.hpp"
70 #endif
71 #ifdef TARGET_ARCH_sparc
72 # include "assembler_sparc.inline.hpp"
73 # include "nativeInst_sparc.hpp"
74 #endif
75 #ifdef TARGET_ARCH_zero
76 # include "assembler_zero.inline.hpp"
77 # include "nativeInst_zero.hpp"
78 #endif
79 #ifdef TARGET_ARCH_arm
80 # include "assembler_arm.inline.hpp"
81 # include "nativeInst_arm.hpp"
82 #endif
83 #ifdef TARGET_ARCH_ppc
84 # include "assembler_ppc.inline.hpp"
85 # include "nativeInst_ppc.hpp"
86 #endif
87 #ifdef COMPILER1
88 #include "c1/c1_Runtime1.hpp"
89 #endif
90 #ifdef COMPILER2
91 #include "opto/runtime.hpp"
92 #endif
93
94 // put OS-includes here
95 # include <sys/types.h>
96 # include <sys/mman.h>
97 # include <sys/stat.h>
98 # include <sys/select.h>
99 # include <pthread.h>
100 # include <signal.h>
101 # include <errno.h>
102 # include <dlfcn.h>
103 # include <stdio.h>
104 # include <unistd.h>
105 # include <sys/resource.h>
106 # include <pthread.h>
107 # include <sys/stat.h>
108 # include <sys/time.h>
109 # include <sys/times.h>
110 # include <sys/utsname.h>
111 # include <sys/socket.h>
112 # include <sys/wait.h>
113 # include <pwd.h>
114 # include <poll.h>
115 # include <semaphore.h>
116 # include <fcntl.h>
117 # include <string.h>
118 # include <syscall.h>
119 # include <sys/sysinfo.h>
120 # include <gnu/libc-version.h>
121 # include <sys/ipc.h>
122 # include <sys/shm.h>
123 # include <link.h>
124 # include <stdint.h>
125 # include <inttypes.h>
126 # include <sys/ioctl.h>
127
128 #ifdef AMD64
129 #include <asm/vsyscall.h>
130 #endif
131
132 #define MAX_PATH (2 * K)
133
134 // for timer info max values which include all bits
135 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
136 #define SEC_IN_NANOSECS 1000000000LL
137
138 #define LARGEPAGES_BIT (1 << 6)
139 ////////////////////////////////////////////////////////////////////////////////
140 // global variables
141 julong os::Linux::_physical_memory = 0;
142
143 address os::Linux::_initial_thread_stack_bottom = NULL;
144 uintptr_t os::Linux::_initial_thread_stack_size = 0;
145
146 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
147 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
148 Mutex* os::Linux::_createThread_lock = NULL;
149 pthread_t os::Linux::_main_thread;
150 int os::Linux::_page_size = -1;
151 bool os::Linux::_is_floating_stack = false;
152 bool os::Linux::_is_NPTL = false;
153 bool os::Linux::_supports_fast_thread_cpu_time = false;
154 const char * os::Linux::_glibc_version = NULL;
155 const char * os::Linux::_libpthread_version = NULL;
156
157 static jlong initial_time_count=0;
158
159 static int clock_tics_per_sec = 100;
160
161 // For diagnostics to print a message once. see run_periodic_checks
162 static sigset_t check_signal_done;
163 static bool check_signals = true;;
164
165 static pid_t _initial_pid = 0;
166
167 /* Signal number used to suspend/resume a thread */
168
169 /* do not use any signal number less than SIGSEGV, see 4355769 */
170 static int SR_signum = SIGUSR2;
171 sigset_t SR_sigset;
172
173 /* Used to protect dlsym() calls */
174 static pthread_mutex_t dl_mutex;
175
176 #ifdef JAVASE_EMBEDDED
177 class MemNotifyThread: public Thread {
178 friend class VMStructs;
179 public:
180 virtual void run();
181
182 private:
183 static MemNotifyThread* _memnotify_thread;
184 int _fd;
185
186 public:
187
188 // Constructor
189 MemNotifyThread(int fd);
190
191 // Tester
192 bool is_memnotify_thread() const { return true; }
193
194 // Printing
195 char* name() const { return (char*)"Linux MemNotify Thread"; }
196
197 // Returns the single instance of the MemNotifyThread
198 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
199
200 // Create and start the single instance of MemNotifyThread
201 static void start();
202 };
203 #endif // JAVASE_EMBEDDED
204
205 // utility functions
206
207 static int SR_initialize();
208 static int SR_finalize();
209
210 julong os::available_memory() {
211 return Linux::available_memory();
212 }
213
214 julong os::Linux::available_memory() {
215 // values in struct sysinfo are "unsigned long"
216 struct sysinfo si;
217 sysinfo(&si);
218
219 return (julong)si.freeram * si.mem_unit;
220 }
221
222 julong os::physical_memory() {
223 return Linux::physical_memory();
224 }
225
226 julong os::allocatable_physical_memory(julong size) {
227 #ifdef _LP64
228 return size;
229 #else
230 julong result = MIN2(size, (julong)3800*M);
231 if (!is_allocatable(result)) {
232 // See comments under solaris for alignment considerations
233 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
234 result = MIN2(size, reasonable_size);
235 }
236 return result;
237 #endif // _LP64
238 }
239
240 ////////////////////////////////////////////////////////////////////////////////
241 // environment support
242
243 bool os::getenv(const char* name, char* buf, int len) {
244 const char* val = ::getenv(name);
245 if (val != NULL && strlen(val) < (size_t)len) {
246 strcpy(buf, val);
247 return true;
248 }
249 if (len > 0) buf[0] = 0; // return a null string
250 return false;
251 }
252
253
254 // Return true if user is running as root.
255
256 bool os::have_special_privileges() {
257 static bool init = false;
258 static bool privileges = false;
259 if (!init) {
260 privileges = (getuid() != geteuid()) || (getgid() != getegid());
261 init = true;
262 }
263 return privileges;
264 }
265
266
267 #ifndef SYS_gettid
268 // i386: 224, ia64: 1105, amd64: 186, sparc 143
269 #ifdef __ia64__
270 #define SYS_gettid 1105
271 #elif __i386__
272 #define SYS_gettid 224
273 #elif __amd64__
274 #define SYS_gettid 186
275 #elif __sparc__
276 #define SYS_gettid 143
277 #else
278 #error define gettid for the arch
279 #endif
280 #endif
281
282 // Cpu architecture string
283 #if defined(ZERO)
284 static char cpu_arch[] = ZERO_LIBARCH;
285 #elif defined(IA64)
286 static char cpu_arch[] = "ia64";
287 #elif defined(IA32)
288 static char cpu_arch[] = "i386";
289 #elif defined(AMD64)
290 static char cpu_arch[] = "amd64";
291 #elif defined(ARM)
292 static char cpu_arch[] = "arm";
293 #elif defined(PPC)
294 static char cpu_arch[] = "ppc";
295 #elif defined(SPARC)
296 # ifdef _LP64
297 static char cpu_arch[] = "sparcv9";
298 # else
299 static char cpu_arch[] = "sparc";
300 # endif
301 #else
302 #error Add appropriate cpu_arch setting
303 #endif
304
305
306 // pid_t gettid()
307 //
308 // Returns the kernel thread id of the currently running thread. Kernel
309 // thread id is used to access /proc.
310 //
311 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
312 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
313 //
314 pid_t os::Linux::gettid() {
315 int rslt = syscall(SYS_gettid);
316 if (rslt == -1) {
317 // old kernel, no NPTL support
318 return getpid();
319 } else {
320 return (pid_t)rslt;
321 }
322 }
323
324 // Most versions of linux have a bug where the number of processors are
325 // determined by looking at the /proc file system. In a chroot environment,
326 // the system call returns 1. This causes the VM to act as if it is
327 // a single processor and elide locking (see is_MP() call).
328 static bool unsafe_chroot_detected = false;
329 static const char *unstable_chroot_error = "/proc file system not found.\n"
330 "Java may be unstable running multithreaded in a chroot "
331 "environment on Linux when /proc filesystem is not mounted.";
332
333 void os::Linux::initialize_system_info() {
334 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
335 if (processor_count() == 1) {
336 pid_t pid = os::Linux::gettid();
337 char fname[32];
338 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
339 FILE *fp = fopen(fname, "r");
340 if (fp == NULL) {
341 unsafe_chroot_detected = true;
342 } else {
343 fclose(fp);
344 }
345 }
346 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
347 assert(processor_count() > 0, "linux error");
348 }
349
350 void os::init_system_properties_values() {
351 // char arch[12];
352 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
353
354 // The next steps are taken in the product version:
355 //
356 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
357 // This library should be located at:
358 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
359 //
360 // If "/jre/lib/" appears at the right place in the path, then we
361 // assume libjvm[_g].so is installed in a JDK and we use this path.
362 //
363 // Otherwise exit with message: "Could not create the Java virtual machine."
364 //
365 // The following extra steps are taken in the debugging version:
366 //
367 // If "/jre/lib/" does NOT appear at the right place in the path
368 // instead of exit check for $JAVA_HOME environment variable.
369 //
370 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
371 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
372 // it looks like libjvm[_g].so is installed there
373 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
374 //
375 // Otherwise exit.
376 //
377 // Important note: if the location of libjvm.so changes this
378 // code needs to be changed accordingly.
379
380 // The next few definitions allow the code to be verbatim:
381 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
382 #define getenv(n) ::getenv(n)
383
384 /*
385 * See ld(1):
386 * The linker uses the following search paths to locate required
387 * shared libraries:
388 * 1: ...
389 * ...
390 * 7: The default directories, normally /lib and /usr/lib.
391 */
392 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
393 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
394 #else
395 #define DEFAULT_LIBPATH "/lib:/usr/lib"
396 #endif
397
398 #define EXTENSIONS_DIR "/lib/ext"
399 #define ENDORSED_DIR "/lib/endorsed"
400 #define REG_DIR "/usr/java/packages"
401
402 {
403 /* sysclasspath, java_home, dll_dir */
404 {
405 char *home_path;
406 char *dll_path;
407 char *pslash;
408 char buf[MAXPATHLEN];
409 os::jvm_path(buf, sizeof(buf));
410
411 // Found the full path to libjvm.so.
412 // Now cut the path to <java_home>/jre if we can.
413 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
414 pslash = strrchr(buf, '/');
415 if (pslash != NULL)
416 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
417 dll_path = malloc(strlen(buf) + 1);
418 if (dll_path == NULL)
419 return;
420 strcpy(dll_path, buf);
421 Arguments::set_dll_dir(dll_path);
422
423 if (pslash != NULL) {
424 pslash = strrchr(buf, '/');
425 if (pslash != NULL) {
426 *pslash = '\0'; /* get rid of /<arch> */
427 pslash = strrchr(buf, '/');
428 if (pslash != NULL)
429 *pslash = '\0'; /* get rid of /lib */
430 }
431 }
432
433 home_path = malloc(strlen(buf) + 1);
434 if (home_path == NULL)
435 return;
436 strcpy(home_path, buf);
437 Arguments::set_java_home(home_path);
438
439 if (!set_boot_path('/', ':'))
440 return;
441 }
442
443 /*
444 * Where to look for native libraries
445 *
446 * Note: Due to a legacy implementation, most of the library path
447 * is set in the launcher. This was to accomodate linking restrictions
448 * on legacy Linux implementations (which are no longer supported).
449 * Eventually, all the library path setting will be done here.
450 *
451 * However, to prevent the proliferation of improperly built native
452 * libraries, the new path component /usr/java/packages is added here.
453 * Eventually, all the library path setting will be done here.
454 */
455 {
456 char *ld_library_path;
457
458 /*
459 * Construct the invariant part of ld_library_path. Note that the
460 * space for the colon and the trailing null are provided by the
461 * nulls included by the sizeof operator (so actually we allocate
462 * a byte more than necessary).
463 */
464 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
465 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
466 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
467
468 /*
469 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
470 * should always exist (until the legacy problem cited above is
471 * addressed).
472 */
473 char *v = getenv("LD_LIBRARY_PATH");
474 if (v != NULL) {
475 char *t = ld_library_path;
476 /* That's +1 for the colon and +1 for the trailing '\0' */
477 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
478 sprintf(ld_library_path, "%s:%s", v, t);
479 }
480 Arguments::set_library_path(ld_library_path);
481 }
482
483 /*
484 * Extensions directories.
485 *
486 * Note that the space for the colon and the trailing null are provided
487 * by the nulls included by the sizeof operator (so actually one byte more
488 * than necessary is allocated).
489 */
490 {
491 char *buf = malloc(strlen(Arguments::get_java_home()) +
492 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
493 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
494 Arguments::get_java_home());
495 Arguments::set_ext_dirs(buf);
496 }
497
498 /* Endorsed standards default directory. */
499 {
500 char * buf;
501 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
502 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
503 Arguments::set_endorsed_dirs(buf);
504 }
505 }
506
507 #undef malloc
508 #undef getenv
509 #undef EXTENSIONS_DIR
510 #undef ENDORSED_DIR
511
512 // Done
513 return;
514 }
515
516 ////////////////////////////////////////////////////////////////////////////////
517 // breakpoint support
518
519 void os::breakpoint() {
520 BREAKPOINT;
521 }
522
523 extern "C" void breakpoint() {
524 // use debugger to set breakpoint here
525 }
526
527 ////////////////////////////////////////////////////////////////////////////////
528 // signal support
529
530 debug_only(static bool signal_sets_initialized = false);
531 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
532
533 bool os::Linux::is_sig_ignored(int sig) {
534 struct sigaction oact;
535 sigaction(sig, (struct sigaction*)NULL, &oact);
536 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
537 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
538 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
539 return true;
540 else
541 return false;
542 }
543
544 void os::Linux::signal_sets_init() {
545 // Should also have an assertion stating we are still single-threaded.
546 assert(!signal_sets_initialized, "Already initialized");
547 // Fill in signals that are necessarily unblocked for all threads in
548 // the VM. Currently, we unblock the following signals:
549 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
550 // by -Xrs (=ReduceSignalUsage));
551 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
552 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
553 // the dispositions or masks wrt these signals.
554 // Programs embedding the VM that want to use the above signals for their
555 // own purposes must, at this time, use the "-Xrs" option to prevent
556 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
557 // (See bug 4345157, and other related bugs).
558 // In reality, though, unblocking these signals is really a nop, since
559 // these signals are not blocked by default.
560 sigemptyset(&unblocked_sigs);
561 sigemptyset(&allowdebug_blocked_sigs);
562 sigaddset(&unblocked_sigs, SIGILL);
563 sigaddset(&unblocked_sigs, SIGSEGV);
564 sigaddset(&unblocked_sigs, SIGBUS);
565 sigaddset(&unblocked_sigs, SIGFPE);
566 sigaddset(&unblocked_sigs, SR_signum);
567
568 if (!ReduceSignalUsage) {
569 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
570 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
571 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
572 }
573 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
574 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
575 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
576 }
577 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
578 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
579 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
580 }
581 }
582 // Fill in signals that are blocked by all but the VM thread.
583 sigemptyset(&vm_sigs);
584 if (!ReduceSignalUsage)
585 sigaddset(&vm_sigs, BREAK_SIGNAL);
586 debug_only(signal_sets_initialized = true);
587
588 }
589
590 // These are signals that are unblocked while a thread is running Java.
591 // (For some reason, they get blocked by default.)
592 sigset_t* os::Linux::unblocked_signals() {
593 assert(signal_sets_initialized, "Not initialized");
594 return &unblocked_sigs;
595 }
596
597 // These are the signals that are blocked while a (non-VM) thread is
598 // running Java. Only the VM thread handles these signals.
599 sigset_t* os::Linux::vm_signals() {
600 assert(signal_sets_initialized, "Not initialized");
601 return &vm_sigs;
602 }
603
604 // These are signals that are blocked during cond_wait to allow debugger in
605 sigset_t* os::Linux::allowdebug_blocked_signals() {
606 assert(signal_sets_initialized, "Not initialized");
607 return &allowdebug_blocked_sigs;
608 }
609
610 void os::Linux::hotspot_sigmask(Thread* thread) {
611
612 //Save caller's signal mask before setting VM signal mask
613 sigset_t caller_sigmask;
614 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
615
616 OSThread* osthread = thread->osthread();
617 osthread->set_caller_sigmask(caller_sigmask);
618
619 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
620
621 if (!ReduceSignalUsage) {
622 if (thread->is_VM_thread()) {
623 // Only the VM thread handles BREAK_SIGNAL ...
624 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
625 } else {
626 // ... all other threads block BREAK_SIGNAL
627 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
628 }
629 }
630 }
631
632 //////////////////////////////////////////////////////////////////////////////
633 // detecting pthread library
634
635 void os::Linux::libpthread_init() {
636 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
637 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
638 // generic name for earlier versions.
639 // Define macros here so we can build HotSpot on old systems.
640 # ifndef _CS_GNU_LIBC_VERSION
641 # define _CS_GNU_LIBC_VERSION 2
642 # endif
643 # ifndef _CS_GNU_LIBPTHREAD_VERSION
644 # define _CS_GNU_LIBPTHREAD_VERSION 3
645 # endif
646
647 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
648 if (n > 0) {
649 char *str = (char *)malloc(n);
650 confstr(_CS_GNU_LIBC_VERSION, str, n);
651 os::Linux::set_glibc_version(str);
652 } else {
653 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
654 static char _gnu_libc_version[32];
655 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
656 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
657 os::Linux::set_glibc_version(_gnu_libc_version);
658 }
659
660 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
661 if (n > 0) {
662 char *str = (char *)malloc(n);
663 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
664 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
665 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
666 // is the case. LinuxThreads has a hard limit on max number of threads.
667 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
668 // On the other hand, NPTL does not have such a limit, sysconf()
669 // will return -1 and errno is not changed. Check if it is really NPTL.
670 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
671 strstr(str, "NPTL") &&
672 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
673 free(str);
674 os::Linux::set_libpthread_version("linuxthreads");
675 } else {
676 os::Linux::set_libpthread_version(str);
677 }
678 } else {
679 // glibc before 2.3.2 only has LinuxThreads.
680 os::Linux::set_libpthread_version("linuxthreads");
681 }
682
683 if (strstr(libpthread_version(), "NPTL")) {
684 os::Linux::set_is_NPTL();
685 } else {
686 os::Linux::set_is_LinuxThreads();
687 }
688
689 // LinuxThreads have two flavors: floating-stack mode, which allows variable
690 // stack size; and fixed-stack mode. NPTL is always floating-stack.
691 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
692 os::Linux::set_is_floating_stack();
693 }
694 }
695
696 /////////////////////////////////////////////////////////////////////////////
697 // thread stack
698
699 // Force Linux kernel to expand current thread stack. If "bottom" is close
700 // to the stack guard, caller should block all signals.
701 //
702 // MAP_GROWSDOWN:
703 // A special mmap() flag that is used to implement thread stacks. It tells
704 // kernel that the memory region should extend downwards when needed. This
705 // allows early versions of LinuxThreads to only mmap the first few pages
706 // when creating a new thread. Linux kernel will automatically expand thread
707 // stack as needed (on page faults).
708 //
709 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
710 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
711 // region, it's hard to tell if the fault is due to a legitimate stack
712 // access or because of reading/writing non-exist memory (e.g. buffer
713 // overrun). As a rule, if the fault happens below current stack pointer,
714 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
715 // application (see Linux kernel fault.c).
716 //
717 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
718 // stack overflow detection.
719 //
720 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
721 // not use this flag. However, the stack of initial thread is not created
722 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
723 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
724 // and then attach the thread to JVM.
725 //
726 // To get around the problem and allow stack banging on Linux, we need to
727 // manually expand thread stack after receiving the SIGSEGV.
728 //
729 // There are two ways to expand thread stack to address "bottom", we used
730 // both of them in JVM before 1.5:
731 // 1. adjust stack pointer first so that it is below "bottom", and then
732 // touch "bottom"
733 // 2. mmap() the page in question
734 //
735 // Now alternate signal stack is gone, it's harder to use 2. For instance,
736 // if current sp is already near the lower end of page 101, and we need to
737 // call mmap() to map page 100, it is possible that part of the mmap() frame
738 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
739 // That will destroy the mmap() frame and cause VM to crash.
740 //
741 // The following code works by adjusting sp first, then accessing the "bottom"
742 // page to force a page fault. Linux kernel will then automatically expand the
743 // stack mapping.
744 //
745 // _expand_stack_to() assumes its frame size is less than page size, which
746 // should always be true if the function is not inlined.
747
748 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
749 #define NOINLINE
750 #else
751 #define NOINLINE __attribute__ ((noinline))
752 #endif
753
754 static void _expand_stack_to(address bottom) NOINLINE;
755
756 static void _expand_stack_to(address bottom) {
757 address sp;
758 size_t size;
759 volatile char *p;
760
761 // Adjust bottom to point to the largest address within the same page, it
762 // gives us a one-page buffer if alloca() allocates slightly more memory.
763 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
764 bottom += os::Linux::page_size() - 1;
765
766 // sp might be slightly above current stack pointer; if that's the case, we
767 // will alloca() a little more space than necessary, which is OK. Don't use
768 // os::current_stack_pointer(), as its result can be slightly below current
769 // stack pointer, causing us to not alloca enough to reach "bottom".
770 sp = (address)&sp;
771
772 if (sp > bottom) {
773 size = sp - bottom;
774 p = (volatile char *)alloca(size);
775 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
776 p[0] = '\0';
777 }
778 }
779
780 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
781 assert(t!=NULL, "just checking");
782 assert(t->osthread()->expanding_stack(), "expand should be set");
783 assert(t->stack_base() != NULL, "stack_base was not initialized");
784
785 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
786 sigset_t mask_all, old_sigset;
787 sigfillset(&mask_all);
788 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
789 _expand_stack_to(addr);
790 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
791 return true;
792 }
793 return false;
794 }
795
796 //////////////////////////////////////////////////////////////////////////////
797 // create new thread
798
799 static address highest_vm_reserved_address();
800
801 // check if it's safe to start a new thread
802 static bool _thread_safety_check(Thread* thread) {
803 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
804 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
805 // Heap is mmap'ed at lower end of memory space. Thread stacks are
806 // allocated (MAP_FIXED) from high address space. Every thread stack
807 // occupies a fixed size slot (usually 2Mbytes, but user can change
808 // it to other values if they rebuild LinuxThreads).
809 //
810 // Problem with MAP_FIXED is that mmap() can still succeed even part of
811 // the memory region has already been mmap'ed. That means if we have too
812 // many threads and/or very large heap, eventually thread stack will
813 // collide with heap.
814 //
815 // Here we try to prevent heap/stack collision by comparing current
816 // stack bottom with the highest address that has been mmap'ed by JVM
817 // plus a safety margin for memory maps created by native code.
818 //
819 // This feature can be disabled by setting ThreadSafetyMargin to 0
820 //
821 if (ThreadSafetyMargin > 0) {
822 address stack_bottom = os::current_stack_base() - os::current_stack_size();
823
824 // not safe if our stack extends below the safety margin
825 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
826 } else {
827 return true;
828 }
829 } else {
830 // Floating stack LinuxThreads or NPTL:
831 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
832 // there's not enough space left, pthread_create() will fail. If we come
833 // here, that means enough space has been reserved for stack.
834 return true;
835 }
836 }
837
838 // Thread start routine for all newly created threads
839 static void *java_start(Thread *thread) {
840 // Try to randomize the cache line index of hot stack frames.
841 // This helps when threads of the same stack traces evict each other's
842 // cache lines. The threads can be either from the same JVM instance, or
843 // from different JVM instances. The benefit is especially true for
844 // processors with hyperthreading technology.
845 static int counter = 0;
846 int pid = os::current_process_id();
847 alloca(((pid ^ counter++) & 7) * 128);
848
849 ThreadLocalStorage::set_thread(thread);
850
851 OSThread* osthread = thread->osthread();
852 Monitor* sync = osthread->startThread_lock();
853
854 // non floating stack LinuxThreads needs extra check, see above
855 if (!_thread_safety_check(thread)) {
856 // notify parent thread
857 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
858 osthread->set_state(ZOMBIE);
859 sync->notify_all();
860 return NULL;
861 }
862
863 // thread_id is kernel thread id (similar to Solaris LWP id)
864 osthread->set_thread_id(os::Linux::gettid());
865
866 if (UseNUMA) {
867 int lgrp_id = os::numa_get_group_id();
868 if (lgrp_id != -1) {
869 thread->set_lgrp_id(lgrp_id);
870 }
871 }
872 // initialize signal mask for this thread
873 os::Linux::hotspot_sigmask(thread);
874
875 // initialize floating point control register
876 os::Linux::init_thread_fpu_state();
877
878 // handshaking with parent thread
879 {
880 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
881
882 // notify parent thread
883 osthread->set_state(INITIALIZED);
884 sync->notify_all();
885
886 // wait until os::start_thread()
887 while (osthread->get_state() == INITIALIZED) {
888 sync->wait(Mutex::_no_safepoint_check_flag);
889 }
890 }
891
892 // call one more level start routine
893 thread->run();
894
895 return 0;
896 }
897
898 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
899 assert(thread->osthread() == NULL, "caller responsible");
900
901 // Allocate the OSThread object
902 OSThread* osthread = new OSThread(NULL, NULL);
903 if (osthread == NULL) {
904 return false;
905 }
906
907 // set the correct thread state
908 osthread->set_thread_type(thr_type);
909
910 // Initial state is ALLOCATED but not INITIALIZED
911 osthread->set_state(ALLOCATED);
912
913 thread->set_osthread(osthread);
914
915 // init thread attributes
916 pthread_attr_t attr;
917 pthread_attr_init(&attr);
918 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
919
920 // stack size
921 if (os::Linux::supports_variable_stack_size()) {
922 // calculate stack size if it's not specified by caller
923 if (stack_size == 0) {
924 stack_size = os::Linux::default_stack_size(thr_type);
925
926 switch (thr_type) {
927 case os::java_thread:
928 // Java threads use ThreadStackSize which default value can be
929 // changed with the flag -Xss
930 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
931 stack_size = JavaThread::stack_size_at_create();
932 break;
933 case os::compiler_thread:
934 if (CompilerThreadStackSize > 0) {
935 stack_size = (size_t)(CompilerThreadStackSize * K);
936 break;
937 } // else fall through:
938 // use VMThreadStackSize if CompilerThreadStackSize is not defined
939 case os::vm_thread:
940 case os::pgc_thread:
941 case os::cgc_thread:
942 case os::watcher_thread:
943 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
944 break;
945 }
946 }
947
948 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
949 pthread_attr_setstacksize(&attr, stack_size);
950 } else {
951 // let pthread_create() pick the default value.
952 }
953
954 // glibc guard page
955 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
956
957 ThreadState state;
958
959 {
960 // Serialize thread creation if we are running with fixed stack LinuxThreads
961 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
962 if (lock) {
963 os::Linux::createThread_lock()->lock_without_safepoint_check();
964 }
965
966 pthread_t tid;
967 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
968
969 pthread_attr_destroy(&attr);
970
971 if (ret != 0) {
972 if (PrintMiscellaneous && (Verbose || WizardMode)) {
973 perror("pthread_create()");
974 }
975 // Need to clean up stuff we've allocated so far
976 thread->set_osthread(NULL);
977 delete osthread;
978 if (lock) os::Linux::createThread_lock()->unlock();
979 return false;
980 }
981
982 // Store pthread info into the OSThread
983 osthread->set_pthread_id(tid);
984
985 // Wait until child thread is either initialized or aborted
986 {
987 Monitor* sync_with_child = osthread->startThread_lock();
988 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
989 while ((state = osthread->get_state()) == ALLOCATED) {
990 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
991 }
992 }
993
994 if (lock) {
995 os::Linux::createThread_lock()->unlock();
996 }
997 }
998
999 // Aborted due to thread limit being reached
1000 if (state == ZOMBIE) {
1001 thread->set_osthread(NULL);
1002 delete osthread;
1003 return false;
1004 }
1005
1006 // The thread is returned suspended (in state INITIALIZED),
1007 // and is started higher up in the call chain
1008 assert(state == INITIALIZED, "race condition");
1009 return true;
1010 }
1011
1012 /////////////////////////////////////////////////////////////////////////////
1013 // attach existing thread
1014
1015 // bootstrap the main thread
1016 bool os::create_main_thread(JavaThread* thread) {
1017 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1018 return create_attached_thread(thread);
1019 }
1020
1021 bool os::create_attached_thread(JavaThread* thread) {
1022 #ifdef ASSERT
1023 thread->verify_not_published();
1024 #endif
1025
1026 // Allocate the OSThread object
1027 OSThread* osthread = new OSThread(NULL, NULL);
1028
1029 if (osthread == NULL) {
1030 return false;
1031 }
1032
1033 // Store pthread info into the OSThread
1034 osthread->set_thread_id(os::Linux::gettid());
1035 osthread->set_pthread_id(::pthread_self());
1036
1037 // initialize floating point control register
1038 os::Linux::init_thread_fpu_state();
1039
1040 // Initial thread state is RUNNABLE
1041 osthread->set_state(RUNNABLE);
1042
1043 thread->set_osthread(osthread);
1044
1045 if (UseNUMA) {
1046 int lgrp_id = os::numa_get_group_id();
1047 if (lgrp_id != -1) {
1048 thread->set_lgrp_id(lgrp_id);
1049 }
1050 }
1051
1052 if (os::Linux::is_initial_thread()) {
1053 // If current thread is initial thread, its stack is mapped on demand,
1054 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1055 // the entire stack region to avoid SEGV in stack banging.
1056 // It is also useful to get around the heap-stack-gap problem on SuSE
1057 // kernel (see 4821821 for details). We first expand stack to the top
1058 // of yellow zone, then enable stack yellow zone (order is significant,
1059 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1060 // is no gap between the last two virtual memory regions.
1061
1062 JavaThread *jt = (JavaThread *)thread;
1063 address addr = jt->stack_yellow_zone_base();
1064 assert(addr != NULL, "initialization problem?");
1065 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1066
1067 osthread->set_expanding_stack();
1068 os::Linux::manually_expand_stack(jt, addr);
1069 osthread->clear_expanding_stack();
1070 }
1071
1072 // initialize signal mask for this thread
1073 // and save the caller's signal mask
1074 os::Linux::hotspot_sigmask(thread);
1075
1076 return true;
1077 }
1078
1079 void os::pd_start_thread(Thread* thread) {
1080 OSThread * osthread = thread->osthread();
1081 assert(osthread->get_state() != INITIALIZED, "just checking");
1082 Monitor* sync_with_child = osthread->startThread_lock();
1083 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1084 sync_with_child->notify();
1085 }
1086
1087 // Free Linux resources related to the OSThread
1088 void os::free_thread(OSThread* osthread) {
1089 assert(osthread != NULL, "osthread not set");
1090
1091 if (Thread::current()->osthread() == osthread) {
1092 // Restore caller's signal mask
1093 sigset_t sigmask = osthread->caller_sigmask();
1094 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1095 }
1096
1097 delete osthread;
1098 }
1099
1100 //////////////////////////////////////////////////////////////////////////////
1101 // thread local storage
1102
1103 int os::allocate_thread_local_storage() {
1104 pthread_key_t key;
1105 int rslt = pthread_key_create(&key, NULL);
1106 assert(rslt == 0, "cannot allocate thread local storage");
1107 return (int)key;
1108 }
1109
1110 // Note: This is currently not used by VM, as we don't destroy TLS key
1111 // on VM exit.
1112 void os::free_thread_local_storage(int index) {
1113 int rslt = pthread_key_delete((pthread_key_t)index);
1114 assert(rslt == 0, "invalid index");
1115 }
1116
1117 void os::thread_local_storage_at_put(int index, void* value) {
1118 int rslt = pthread_setspecific((pthread_key_t)index, value);
1119 assert(rslt == 0, "pthread_setspecific failed");
1120 }
1121
1122 extern "C" Thread* get_thread() {
1123 return ThreadLocalStorage::thread();
1124 }
1125
1126 //////////////////////////////////////////////////////////////////////////////
1127 // initial thread
1128
1129 // Check if current thread is the initial thread, similar to Solaris thr_main.
1130 bool os::Linux::is_initial_thread(void) {
1131 char dummy;
1132 // If called before init complete, thread stack bottom will be null.
1133 // Can be called if fatal error occurs before initialization.
1134 if (initial_thread_stack_bottom() == NULL) return false;
1135 assert(initial_thread_stack_bottom() != NULL &&
1136 initial_thread_stack_size() != 0,
1137 "os::init did not locate initial thread's stack region");
1138 if ((address)&dummy >= initial_thread_stack_bottom() &&
1139 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1140 return true;
1141 else return false;
1142 }
1143
1144 // Find the virtual memory area that contains addr
1145 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1146 FILE *fp = fopen("/proc/self/maps", "r");
1147 if (fp) {
1148 address low, high;
1149 while (!feof(fp)) {
1150 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1151 if (low <= addr && addr < high) {
1152 if (vma_low) *vma_low = low;
1153 if (vma_high) *vma_high = high;
1154 fclose (fp);
1155 return true;
1156 }
1157 }
1158 for (;;) {
1159 int ch = fgetc(fp);
1160 if (ch == EOF || ch == (int)'\n') break;
1161 }
1162 }
1163 fclose(fp);
1164 }
1165 return false;
1166 }
1167
1168 // Locate initial thread stack. This special handling of initial thread stack
1169 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1170 // bogus value for initial thread.
1171 void os::Linux::capture_initial_stack(size_t max_size) {
1172 // stack size is the easy part, get it from RLIMIT_STACK
1173 size_t stack_size;
1174 struct rlimit rlim;
1175 getrlimit(RLIMIT_STACK, &rlim);
1176 stack_size = rlim.rlim_cur;
1177
1178 // 6308388: a bug in ld.so will relocate its own .data section to the
1179 // lower end of primordial stack; reduce ulimit -s value a little bit
1180 // so we won't install guard page on ld.so's data section.
1181 stack_size -= 2 * page_size();
1182
1183 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1184 // 7.1, in both cases we will get 2G in return value.
1185 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1186 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1187 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1188 // in case other parts in glibc still assumes 2M max stack size.
1189 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1190 #ifndef IA64
1191 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1192 #else
1193 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1194 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1195 #endif
1196
1197 // Try to figure out where the stack base (top) is. This is harder.
1198 //
1199 // When an application is started, glibc saves the initial stack pointer in
1200 // a global variable "__libc_stack_end", which is then used by system
1201 // libraries. __libc_stack_end should be pretty close to stack top. The
1202 // variable is available since the very early days. However, because it is
1203 // a private interface, it could disappear in the future.
1204 //
1205 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1206 // to __libc_stack_end, it is very close to stack top, but isn't the real
1207 // stack top. Note that /proc may not exist if VM is running as a chroot
1208 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1209 // /proc/<pid>/stat could change in the future (though unlikely).
1210 //
1211 // We try __libc_stack_end first. If that doesn't work, look for
1212 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1213 // as a hint, which should work well in most cases.
1214
1215 uintptr_t stack_start;
1216
1217 // try __libc_stack_end first
1218 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1219 if (p && *p) {
1220 stack_start = *p;
1221 } else {
1222 // see if we can get the start_stack field from /proc/self/stat
1223 FILE *fp;
1224 int pid;
1225 char state;
1226 int ppid;
1227 int pgrp;
1228 int session;
1229 int nr;
1230 int tpgrp;
1231 unsigned long flags;
1232 unsigned long minflt;
1233 unsigned long cminflt;
1234 unsigned long majflt;
1235 unsigned long cmajflt;
1236 unsigned long utime;
1237 unsigned long stime;
1238 long cutime;
1239 long cstime;
1240 long prio;
1241 long nice;
1242 long junk;
1243 long it_real;
1244 uintptr_t start;
1245 uintptr_t vsize;
1246 intptr_t rss;
1247 uintptr_t rsslim;
1248 uintptr_t scodes;
1249 uintptr_t ecode;
1250 int i;
1251
1252 // Figure what the primordial thread stack base is. Code is inspired
1253 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1254 // followed by command name surrounded by parentheses, state, etc.
1255 char stat[2048];
1256 int statlen;
1257
1258 fp = fopen("/proc/self/stat", "r");
1259 if (fp) {
1260 statlen = fread(stat, 1, 2047, fp);
1261 stat[statlen] = '\0';
1262 fclose(fp);
1263
1264 // Skip pid and the command string. Note that we could be dealing with
1265 // weird command names, e.g. user could decide to rename java launcher
1266 // to "java 1.4.2 :)", then the stat file would look like
1267 // 1234 (java 1.4.2 :)) R ... ...
1268 // We don't really need to know the command string, just find the last
1269 // occurrence of ")" and then start parsing from there. See bug 4726580.
1270 char * s = strrchr(stat, ')');
1271
1272 i = 0;
1273 if (s) {
1274 // Skip blank chars
1275 do s++; while (isspace(*s));
1276
1277 #define _UFM UINTX_FORMAT
1278 #define _DFM INTX_FORMAT
1279
1280 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1281 /* 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 */
1282 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,
1283 &state, /* 3 %c */
1284 &ppid, /* 4 %d */
1285 &pgrp, /* 5 %d */
1286 &session, /* 6 %d */
1287 &nr, /* 7 %d */
1288 &tpgrp, /* 8 %d */
1289 &flags, /* 9 %lu */
1290 &minflt, /* 10 %lu */
1291 &cminflt, /* 11 %lu */
1292 &majflt, /* 12 %lu */
1293 &cmajflt, /* 13 %lu */
1294 &utime, /* 14 %lu */
1295 &stime, /* 15 %lu */
1296 &cutime, /* 16 %ld */
1297 &cstime, /* 17 %ld */
1298 &prio, /* 18 %ld */
1299 &nice, /* 19 %ld */
1300 &junk, /* 20 %ld */
1301 &it_real, /* 21 %ld */
1302 &start, /* 22 UINTX_FORMAT */
1303 &vsize, /* 23 UINTX_FORMAT */
1304 &rss, /* 24 INTX_FORMAT */
1305 &rsslim, /* 25 UINTX_FORMAT */
1306 &scodes, /* 26 UINTX_FORMAT */
1307 &ecode, /* 27 UINTX_FORMAT */
1308 &stack_start); /* 28 UINTX_FORMAT */
1309 }
1310
1311 #undef _UFM
1312 #undef _DFM
1313
1314 if (i != 28 - 2) {
1315 assert(false, "Bad conversion from /proc/self/stat");
1316 // product mode - assume we are the initial thread, good luck in the
1317 // embedded case.
1318 warning("Can't detect initial thread stack location - bad conversion");
1319 stack_start = (uintptr_t) &rlim;
1320 }
1321 } else {
1322 // For some reason we can't open /proc/self/stat (for example, running on
1323 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1324 // most cases, so don't abort:
1325 warning("Can't detect initial thread stack location - no /proc/self/stat");
1326 stack_start = (uintptr_t) &rlim;
1327 }
1328 }
1329
1330 // Now we have a pointer (stack_start) very close to the stack top, the
1331 // next thing to do is to figure out the exact location of stack top. We
1332 // can find out the virtual memory area that contains stack_start by
1333 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1334 // and its upper limit is the real stack top. (again, this would fail if
1335 // running inside chroot, because /proc may not exist.)
1336
1337 uintptr_t stack_top;
1338 address low, high;
1339 if (find_vma((address)stack_start, &low, &high)) {
1340 // success, "high" is the true stack top. (ignore "low", because initial
1341 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1342 stack_top = (uintptr_t)high;
1343 } else {
1344 // failed, likely because /proc/self/maps does not exist
1345 warning("Can't detect initial thread stack location - find_vma failed");
1346 // best effort: stack_start is normally within a few pages below the real
1347 // stack top, use it as stack top, and reduce stack size so we won't put
1348 // guard page outside stack.
1349 stack_top = stack_start;
1350 stack_size -= 16 * page_size();
1351 }
1352
1353 // stack_top could be partially down the page so align it
1354 stack_top = align_size_up(stack_top, page_size());
1355
1356 if (max_size && stack_size > max_size) {
1357 _initial_thread_stack_size = max_size;
1358 } else {
1359 _initial_thread_stack_size = stack_size;
1360 }
1361
1362 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1363 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1364 }
1365
1366 ////////////////////////////////////////////////////////////////////////////////
1367 // time support
1368
1369 // Time since start-up in seconds to a fine granularity.
1370 // Used by VMSelfDestructTimer and the MemProfiler.
1371 double os::elapsedTime() {
1372
1373 return (double)(os::elapsed_counter()) * 0.000001;
1374 }
1375
1376 jlong os::elapsed_counter() {
1377 timeval time;
1378 int status = gettimeofday(&time, NULL);
1379 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1380 }
1381
1382 jlong os::elapsed_frequency() {
1383 return (1000 * 1000);
1384 }
1385
1386 // For now, we say that linux does not support vtime. I have no idea
1387 // whether it can actually be made to (DLD, 9/13/05).
1388
1389 bool os::supports_vtime() { return false; }
1390 bool os::enable_vtime() { return false; }
1391 bool os::vtime_enabled() { return false; }
1392 double os::elapsedVTime() {
1393 // better than nothing, but not much
1394 return elapsedTime();
1395 }
1396
1397 jlong os::javaTimeMillis() {
1398 timeval time;
1399 int status = gettimeofday(&time, NULL);
1400 assert(status != -1, "linux error");
1401 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1402 }
1403
1404 #ifndef CLOCK_MONOTONIC
1405 #define CLOCK_MONOTONIC (1)
1406 #endif
1407
1408 void os::Linux::clock_init() {
1409 // we do dlopen's in this particular order due to bug in linux
1410 // dynamical loader (see 6348968) leading to crash on exit
1411 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1412 if (handle == NULL) {
1413 handle = dlopen("librt.so", RTLD_LAZY);
1414 }
1415
1416 if (handle) {
1417 int (*clock_getres_func)(clockid_t, struct timespec*) =
1418 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1419 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1420 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1421 if (clock_getres_func && clock_gettime_func) {
1422 // See if monotonic clock is supported by the kernel. Note that some
1423 // early implementations simply return kernel jiffies (updated every
1424 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1425 // for nano time (though the monotonic property is still nice to have).
1426 // It's fixed in newer kernels, however clock_getres() still returns
1427 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1428 // resolution for now. Hopefully as people move to new kernels, this
1429 // won't be a problem.
1430 struct timespec res;
1431 struct timespec tp;
1432 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1433 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1434 // yes, monotonic clock is supported
1435 _clock_gettime = clock_gettime_func;
1436 } else {
1437 // close librt if there is no monotonic clock
1438 dlclose(handle);
1439 }
1440 }
1441 }
1442 }
1443
1444 #ifndef SYS_clock_getres
1445
1446 #if defined(IA32) || defined(AMD64)
1447 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1448 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1449 #else
1450 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1451 #define sys_clock_getres(x,y) -1
1452 #endif
1453
1454 #else
1455 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1456 #endif
1457
1458 void os::Linux::fast_thread_clock_init() {
1459 if (!UseLinuxPosixThreadCPUClocks) {
1460 return;
1461 }
1462 clockid_t clockid;
1463 struct timespec tp;
1464 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1465 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1466
1467 // Switch to using fast clocks for thread cpu time if
1468 // the sys_clock_getres() returns 0 error code.
1469 // Note, that some kernels may support the current thread
1470 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1471 // returned by the pthread_getcpuclockid().
1472 // If the fast Posix clocks are supported then the sys_clock_getres()
1473 // must return at least tp.tv_sec == 0 which means a resolution
1474 // better than 1 sec. This is extra check for reliability.
1475
1476 if(pthread_getcpuclockid_func &&
1477 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1478 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1479
1480 _supports_fast_thread_cpu_time = true;
1481 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1482 }
1483 }
1484
1485 jlong os::javaTimeNanos() {
1486 if (Linux::supports_monotonic_clock()) {
1487 struct timespec tp;
1488 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1489 assert(status == 0, "gettime error");
1490 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1491 return result;
1492 } else {
1493 timeval time;
1494 int status = gettimeofday(&time, NULL);
1495 assert(status != -1, "linux error");
1496 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1497 return 1000 * usecs;
1498 }
1499 }
1500
1501 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1502 if (Linux::supports_monotonic_clock()) {
1503 info_ptr->max_value = ALL_64_BITS;
1504
1505 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1506 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1507 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1508 } else {
1509 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1510 info_ptr->max_value = ALL_64_BITS;
1511
1512 // gettimeofday is a real time clock so it skips
1513 info_ptr->may_skip_backward = true;
1514 info_ptr->may_skip_forward = true;
1515 }
1516
1517 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1518 }
1519
1520 // Return the real, user, and system times in seconds from an
1521 // arbitrary fixed point in the past.
1522 bool os::getTimesSecs(double* process_real_time,
1523 double* process_user_time,
1524 double* process_system_time) {
1525 struct tms ticks;
1526 clock_t real_ticks = times(&ticks);
1527
1528 if (real_ticks == (clock_t) (-1)) {
1529 return false;
1530 } else {
1531 double ticks_per_second = (double) clock_tics_per_sec;
1532 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1533 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1534 *process_real_time = ((double) real_ticks) / ticks_per_second;
1535
1536 return true;
1537 }
1538 }
1539
1540
1541 char * os::local_time_string(char *buf, size_t buflen) {
1542 struct tm t;
1543 time_t long_time;
1544 time(&long_time);
1545 localtime_r(&long_time, &t);
1546 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1547 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1548 t.tm_hour, t.tm_min, t.tm_sec);
1549 return buf;
1550 }
1551
1552 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1553 return localtime_r(clock, res);
1554 }
1555
1556 ////////////////////////////////////////////////////////////////////////////////
1557 // runtime exit support
1558
1559 // Note: os::shutdown() might be called very early during initialization, or
1560 // called from signal handler. Before adding something to os::shutdown(), make
1561 // sure it is async-safe and can handle partially initialized VM.
1562 void os::shutdown() {
1563
1564 // allow PerfMemory to attempt cleanup of any persistent resources
1565 perfMemory_exit();
1566
1567 // needs to remove object in file system
1568 AttachListener::abort();
1569
1570 // flush buffered output, finish log files
1571 ostream_abort();
1572
1573 // Check for abort hook
1574 abort_hook_t abort_hook = Arguments::abort_hook();
1575 if (abort_hook != NULL) {
1576 abort_hook();
1577 }
1578
1579 }
1580
1581 // Note: os::abort() might be called very early during initialization, or
1582 // called from signal handler. Before adding something to os::abort(), make
1583 // sure it is async-safe and can handle partially initialized VM.
1584 void os::abort(bool dump_core) {
1585 os::shutdown();
1586 if (dump_core) {
1587 #ifndef PRODUCT
1588 fdStream out(defaultStream::output_fd());
1589 out.print_raw("Current thread is ");
1590 char buf[16];
1591 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1592 out.print_raw_cr(buf);
1593 out.print_raw_cr("Dumping core ...");
1594 #endif
1595 ::abort(); // dump core
1596 }
1597
1598 ::exit(1);
1599 }
1600
1601 // Die immediately, no exit hook, no abort hook, no cleanup.
1602 void os::die() {
1603 // _exit() on LinuxThreads only kills current thread
1604 ::abort();
1605 }
1606
1607 // unused on linux for now.
1608 void os::set_error_file(const char *logfile) {}
1609
1610
1611 // This method is a copy of JDK's sysGetLastErrorString
1612 // from src/solaris/hpi/src/system_md.c
1613
1614 size_t os::lasterror(char *buf, size_t len) {
1615
1616 if (errno == 0) return 0;
1617
1618 const char *s = ::strerror(errno);
1619 size_t n = ::strlen(s);
1620 if (n >= len) {
1621 n = len - 1;
1622 }
1623 ::strncpy(buf, s, n);
1624 buf[n] = '\0';
1625 return n;
1626 }
1627
1628 intx os::current_thread_id() { return (intx)pthread_self(); }
1629 int os::current_process_id() {
1630
1631 // Under the old linux thread library, linux gives each thread
1632 // its own process id. Because of this each thread will return
1633 // a different pid if this method were to return the result
1634 // of getpid(2). Linux provides no api that returns the pid
1635 // of the launcher thread for the vm. This implementation
1636 // returns a unique pid, the pid of the launcher thread
1637 // that starts the vm 'process'.
1638
1639 // Under the NPTL, getpid() returns the same pid as the
1640 // launcher thread rather than a unique pid per thread.
1641 // Use gettid() if you want the old pre NPTL behaviour.
1642
1643 // if you are looking for the result of a call to getpid() that
1644 // returns a unique pid for the calling thread, then look at the
1645 // OSThread::thread_id() method in osThread_linux.hpp file
1646
1647 return (int)(_initial_pid ? _initial_pid : getpid());
1648 }
1649
1650 // DLL functions
1651
1652 const char* os::dll_file_extension() { return ".so"; }
1653
1654 // This must be hard coded because it's the system's temporary
1655 // directory not the java application's temp directory, ala java.io.tmpdir.
1656 const char* os::get_temp_directory() { return "/tmp"; }
1657
1658 static bool file_exists(const char* filename) {
1659 struct stat statbuf;
1660 if (filename == NULL || strlen(filename) == 0) {
1661 return false;
1662 }
1663 return os::stat(filename, &statbuf) == 0;
1664 }
1665
1666 void os::dll_build_name(char* buffer, size_t buflen,
1667 const char* pname, const char* fname) {
1668 // Copied from libhpi
1669 const size_t pnamelen = pname ? strlen(pname) : 0;
1670
1671 // Quietly truncate on buffer overflow. Should be an error.
1672 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1673 *buffer = '\0';
1674 return;
1675 }
1676
1677 if (pnamelen == 0) {
1678 snprintf(buffer, buflen, "lib%s.so", fname);
1679 } else if (strchr(pname, *os::path_separator()) != NULL) {
1680 int n;
1681 char** pelements = split_path(pname, &n);
1682 for (int i = 0 ; i < n ; i++) {
1683 // Really shouldn't be NULL, but check can't hurt
1684 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1685 continue; // skip the empty path values
1686 }
1687 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1688 if (file_exists(buffer)) {
1689 break;
1690 }
1691 }
1692 // release the storage
1693 for (int i = 0 ; i < n ; i++) {
1694 if (pelements[i] != NULL) {
1695 FREE_C_HEAP_ARRAY(char, pelements[i]);
1696 }
1697 }
1698 if (pelements != NULL) {
1699 FREE_C_HEAP_ARRAY(char*, pelements);
1700 }
1701 } else {
1702 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1703 }
1704 }
1705
1706 const char* os::get_current_directory(char *buf, int buflen) {
1707 return getcwd(buf, buflen);
1708 }
1709
1710 // check if addr is inside libjvm[_g].so
1711 bool os::address_is_in_vm(address addr) {
1712 static address libjvm_base_addr;
1713 Dl_info dlinfo;
1714
1715 if (libjvm_base_addr == NULL) {
1716 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1717 libjvm_base_addr = (address)dlinfo.dli_fbase;
1718 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1719 }
1720
1721 if (dladdr((void *)addr, &dlinfo)) {
1722 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1723 }
1724
1725 return false;
1726 }
1727
1728 bool os::dll_address_to_function_name(address addr, char *buf,
1729 int buflen, int *offset) {
1730 Dl_info dlinfo;
1731
1732 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1733 if (buf != NULL) {
1734 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1735 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1736 }
1737 }
1738 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1739 return true;
1740 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1741 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1742 dlinfo.dli_fname, buf, buflen, offset) == Decoder::no_error) {
1743 return true;
1744 }
1745 }
1746
1747 if (buf != NULL) buf[0] = '\0';
1748 if (offset != NULL) *offset = -1;
1749 return false;
1750 }
1751
1752 struct _address_to_library_name {
1753 address addr; // input : memory address
1754 size_t buflen; // size of fname
1755 char* fname; // output: library name
1756 address base; // library base addr
1757 };
1758
1759 static int address_to_library_name_callback(struct dl_phdr_info *info,
1760 size_t size, void *data) {
1761 int i;
1762 bool found = false;
1763 address libbase = NULL;
1764 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1765
1766 // iterate through all loadable segments
1767 for (i = 0; i < info->dlpi_phnum; i++) {
1768 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1769 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1770 // base address of a library is the lowest address of its loaded
1771 // segments.
1772 if (libbase == NULL || libbase > segbase) {
1773 libbase = segbase;
1774 }
1775 // see if 'addr' is within current segment
1776 if (segbase <= d->addr &&
1777 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1778 found = true;
1779 }
1780 }
1781 }
1782
1783 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1784 // so dll_address_to_library_name() can fall through to use dladdr() which
1785 // can figure out executable name from argv[0].
1786 if (found && info->dlpi_name && info->dlpi_name[0]) {
1787 d->base = libbase;
1788 if (d->fname) {
1789 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1790 }
1791 return 1;
1792 }
1793 return 0;
1794 }
1795
1796 bool os::dll_address_to_library_name(address addr, char* buf,
1797 int buflen, int* offset) {
1798 Dl_info dlinfo;
1799 struct _address_to_library_name data;
1800
1801 // There is a bug in old glibc dladdr() implementation that it could resolve
1802 // to wrong library name if the .so file has a base address != NULL. Here
1803 // we iterate through the program headers of all loaded libraries to find
1804 // out which library 'addr' really belongs to. This workaround can be
1805 // removed once the minimum requirement for glibc is moved to 2.3.x.
1806 data.addr = addr;
1807 data.fname = buf;
1808 data.buflen = buflen;
1809 data.base = NULL;
1810 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1811
1812 if (rslt) {
1813 // buf already contains library name
1814 if (offset) *offset = addr - data.base;
1815 return true;
1816 } else if (dladdr((void*)addr, &dlinfo)){
1817 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1818 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1819 return true;
1820 } else {
1821 if (buf) buf[0] = '\0';
1822 if (offset) *offset = -1;
1823 return false;
1824 }
1825 }
1826
1827 // Loads .dll/.so and
1828 // in case of error it checks if .dll/.so was built for the
1829 // same architecture as Hotspot is running on
1830
1831 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1832 {
1833 void * result= ::dlopen(filename, RTLD_LAZY);
1834 if (result != NULL) {
1835 // Successful loading
1836 return result;
1837 }
1838
1839 Elf32_Ehdr elf_head;
1840
1841 // Read system error message into ebuf
1842 // It may or may not be overwritten below
1843 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1844 ebuf[ebuflen-1]='\0';
1845 int diag_msg_max_length=ebuflen-strlen(ebuf);
1846 char* diag_msg_buf=ebuf+strlen(ebuf);
1847
1848 if (diag_msg_max_length==0) {
1849 // No more space in ebuf for additional diagnostics message
1850 return NULL;
1851 }
1852
1853
1854 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1855
1856 if (file_descriptor < 0) {
1857 // Can't open library, report dlerror() message
1858 return NULL;
1859 }
1860
1861 bool failed_to_read_elf_head=
1862 (sizeof(elf_head)!=
1863 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1864
1865 ::close(file_descriptor);
1866 if (failed_to_read_elf_head) {
1867 // file i/o error - report dlerror() msg
1868 return NULL;
1869 }
1870
1871 typedef struct {
1872 Elf32_Half code; // Actual value as defined in elf.h
1873 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1874 char elf_class; // 32 or 64 bit
1875 char endianess; // MSB or LSB
1876 char* name; // String representation
1877 } arch_t;
1878
1879 #ifndef EM_486
1880 #define EM_486 6 /* Intel 80486 */
1881 #endif
1882
1883 static const arch_t arch_array[]={
1884 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1885 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1886 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1887 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1888 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1889 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1890 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1891 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1892 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1893 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1894 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1895 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1896 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1897 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1898 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1899 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1900 };
1901
1902 #if (defined IA32)
1903 static Elf32_Half running_arch_code=EM_386;
1904 #elif (defined AMD64)
1905 static Elf32_Half running_arch_code=EM_X86_64;
1906 #elif (defined IA64)
1907 static Elf32_Half running_arch_code=EM_IA_64;
1908 #elif (defined __sparc) && (defined _LP64)
1909 static Elf32_Half running_arch_code=EM_SPARCV9;
1910 #elif (defined __sparc) && (!defined _LP64)
1911 static Elf32_Half running_arch_code=EM_SPARC;
1912 #elif (defined __powerpc64__)
1913 static Elf32_Half running_arch_code=EM_PPC64;
1914 #elif (defined __powerpc__)
1915 static Elf32_Half running_arch_code=EM_PPC;
1916 #elif (defined ARM)
1917 static Elf32_Half running_arch_code=EM_ARM;
1918 #elif (defined S390)
1919 static Elf32_Half running_arch_code=EM_S390;
1920 #elif (defined ALPHA)
1921 static Elf32_Half running_arch_code=EM_ALPHA;
1922 #elif (defined MIPSEL)
1923 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1924 #elif (defined PARISC)
1925 static Elf32_Half running_arch_code=EM_PARISC;
1926 #elif (defined MIPS)
1927 static Elf32_Half running_arch_code=EM_MIPS;
1928 #elif (defined M68K)
1929 static Elf32_Half running_arch_code=EM_68K;
1930 #else
1931 #error Method os::dll_load requires that one of following is defined:\
1932 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1933 #endif
1934
1935 // Identify compatability class for VM's architecture and library's architecture
1936 // Obtain string descriptions for architectures
1937
1938 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1939 int running_arch_index=-1;
1940
1941 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1942 if (running_arch_code == arch_array[i].code) {
1943 running_arch_index = i;
1944 }
1945 if (lib_arch.code == arch_array[i].code) {
1946 lib_arch.compat_class = arch_array[i].compat_class;
1947 lib_arch.name = arch_array[i].name;
1948 }
1949 }
1950
1951 assert(running_arch_index != -1,
1952 "Didn't find running architecture code (running_arch_code) in arch_array");
1953 if (running_arch_index == -1) {
1954 // Even though running architecture detection failed
1955 // we may still continue with reporting dlerror() message
1956 return NULL;
1957 }
1958
1959 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1960 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1961 return NULL;
1962 }
1963
1964 #ifndef S390
1965 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1966 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1967 return NULL;
1968 }
1969 #endif // !S390
1970
1971 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1972 if ( lib_arch.name!=NULL ) {
1973 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1974 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1975 lib_arch.name, arch_array[running_arch_index].name);
1976 } else {
1977 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1978 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1979 lib_arch.code,
1980 arch_array[running_arch_index].name);
1981 }
1982 }
1983
1984 return NULL;
1985 }
1986
1987 /*
1988 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1989 * chances are you might want to run the generated bits against glibc-2.0
1990 * libdl.so, so always use locking for any version of glibc.
1991 */
1992 void* os::dll_lookup(void* handle, const char* name) {
1993 pthread_mutex_lock(&dl_mutex);
1994 void* res = dlsym(handle, name);
1995 pthread_mutex_unlock(&dl_mutex);
1996 return res;
1997 }
1998
1999
2000 static bool _print_ascii_file(const char* filename, outputStream* st) {
2001 int fd = ::open(filename, O_RDONLY);
2002 if (fd == -1) {
2003 return false;
2004 }
2005
2006 char buf[32];
2007 int bytes;
2008 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2009 st->print_raw(buf, bytes);
2010 }
2011
2012 ::close(fd);
2013
2014 return true;
2015 }
2016
2017 void os::print_dll_info(outputStream *st) {
2018 st->print_cr("Dynamic libraries:");
2019
2020 char fname[32];
2021 pid_t pid = os::Linux::gettid();
2022
2023 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2024
2025 if (!_print_ascii_file(fname, st)) {
2026 st->print("Can not get library information for pid = %d\n", pid);
2027 }
2028 }
2029
2030
2031 void os::print_os_info(outputStream* st) {
2032 st->print("OS:");
2033
2034 // Try to identify popular distros.
2035 // Most Linux distributions have /etc/XXX-release file, which contains
2036 // the OS version string. Some have more than one /etc/XXX-release file
2037 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2038 // so the order is important.
2039 if (!_print_ascii_file("/etc/mandrake-release", st) &&
2040 !_print_ascii_file("/etc/sun-release", st) &&
2041 !_print_ascii_file("/etc/redhat-release", st) &&
2042 !_print_ascii_file("/etc/SuSE-release", st) &&
2043 !_print_ascii_file("/etc/turbolinux-release", st) &&
2044 !_print_ascii_file("/etc/gentoo-release", st) &&
2045 !_print_ascii_file("/etc/debian_version", st) &&
2046 !_print_ascii_file("/etc/ltib-release", st) &&
2047 !_print_ascii_file("/etc/angstrom-version", st)) {
2048 st->print("Linux");
2049 }
2050 st->cr();
2051
2052 // kernel
2053 st->print("uname:");
2054 struct utsname name;
2055 uname(&name);
2056 st->print(name.sysname); st->print(" ");
2057 st->print(name.release); st->print(" ");
2058 st->print(name.version); st->print(" ");
2059 st->print(name.machine);
2060 st->cr();
2061
2062 // Print warning if unsafe chroot environment detected
2063 if (unsafe_chroot_detected) {
2064 st->print("WARNING!! ");
2065 st->print_cr(unstable_chroot_error);
2066 }
2067
2068 // libc, pthread
2069 st->print("libc:");
2070 st->print(os::Linux::glibc_version()); st->print(" ");
2071 st->print(os::Linux::libpthread_version()); st->print(" ");
2072 if (os::Linux::is_LinuxThreads()) {
2073 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2074 }
2075 st->cr();
2076
2077 // rlimit
2078 st->print("rlimit:");
2079 struct rlimit rlim;
2080
2081 st->print(" STACK ");
2082 getrlimit(RLIMIT_STACK, &rlim);
2083 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2084 else st->print("%uk", rlim.rlim_cur >> 10);
2085
2086 st->print(", CORE ");
2087 getrlimit(RLIMIT_CORE, &rlim);
2088 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2089 else st->print("%uk", rlim.rlim_cur >> 10);
2090
2091 st->print(", NPROC ");
2092 getrlimit(RLIMIT_NPROC, &rlim);
2093 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2094 else st->print("%d", rlim.rlim_cur);
2095
2096 st->print(", NOFILE ");
2097 getrlimit(RLIMIT_NOFILE, &rlim);
2098 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2099 else st->print("%d", rlim.rlim_cur);
2100
2101 st->print(", AS ");
2102 getrlimit(RLIMIT_AS, &rlim);
2103 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2104 else st->print("%uk", rlim.rlim_cur >> 10);
2105 st->cr();
2106
2107 // load average
2108 st->print("load average:");
2109 double loadavg[3];
2110 os::loadavg(loadavg, 3);
2111 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2112 st->cr();
2113
2114 // meminfo
2115 st->print("\n/proc/meminfo:\n");
2116 _print_ascii_file("/proc/meminfo", st);
2117 st->cr();
2118 }
2119
2120 void os::pd_print_cpu_info(outputStream* st) {
2121 st->print("\n/proc/cpuinfo:\n");
2122 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2123 st->print(" <Not Available>");
2124 }
2125 st->cr();
2126 }
2127
2128 void os::print_memory_info(outputStream* st) {
2129
2130 st->print("Memory:");
2131 st->print(" %dk page", os::vm_page_size()>>10);
2132
2133 // values in struct sysinfo are "unsigned long"
2134 struct sysinfo si;
2135 sysinfo(&si);
2136
2137 st->print(", physical " UINT64_FORMAT "k",
2138 os::physical_memory() >> 10);
2139 st->print("(" UINT64_FORMAT "k free)",
2140 os::available_memory() >> 10);
2141 st->print(", swap " UINT64_FORMAT "k",
2142 ((jlong)si.totalswap * si.mem_unit) >> 10);
2143 st->print("(" UINT64_FORMAT "k free)",
2144 ((jlong)si.freeswap * si.mem_unit) >> 10);
2145 st->cr();
2146 }
2147
2148 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2149 // but they're the same for all the linux arch that we support
2150 // and they're the same for solaris but there's no common place to put this.
2151 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2152 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2153 "ILL_COPROC", "ILL_BADSTK" };
2154
2155 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2156 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2157 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2158
2159 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2160
2161 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2162
2163 void os::print_siginfo(outputStream* st, void* siginfo) {
2164 st->print("siginfo:");
2165
2166 const int buflen = 100;
2167 char buf[buflen];
2168 siginfo_t *si = (siginfo_t*)siginfo;
2169 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2170 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2171 st->print("si_errno=%s", buf);
2172 } else {
2173 st->print("si_errno=%d", si->si_errno);
2174 }
2175 const int c = si->si_code;
2176 assert(c > 0, "unexpected si_code");
2177 switch (si->si_signo) {
2178 case SIGILL:
2179 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2180 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2181 break;
2182 case SIGFPE:
2183 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2184 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2185 break;
2186 case SIGSEGV:
2187 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2188 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2189 break;
2190 case SIGBUS:
2191 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2192 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2193 break;
2194 default:
2195 st->print(", si_code=%d", si->si_code);
2196 // no si_addr
2197 }
2198
2199 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2200 UseSharedSpaces) {
2201 FileMapInfo* mapinfo = FileMapInfo::current_info();
2202 if (mapinfo->is_in_shared_space(si->si_addr)) {
2203 st->print("\n\nError accessing class data sharing archive." \
2204 " Mapped file inaccessible during execution, " \
2205 " possible disk/network problem.");
2206 }
2207 }
2208 st->cr();
2209 }
2210
2211
2212 static void print_signal_handler(outputStream* st, int sig,
2213 char* buf, size_t buflen);
2214
2215 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2216 st->print_cr("Signal Handlers:");
2217 print_signal_handler(st, SIGSEGV, buf, buflen);
2218 print_signal_handler(st, SIGBUS , buf, buflen);
2219 print_signal_handler(st, SIGFPE , buf, buflen);
2220 print_signal_handler(st, SIGPIPE, buf, buflen);
2221 print_signal_handler(st, SIGXFSZ, buf, buflen);
2222 print_signal_handler(st, SIGILL , buf, buflen);
2223 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2224 print_signal_handler(st, SR_signum, buf, buflen);
2225 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2226 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2227 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2228 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2229 }
2230
2231 static char saved_jvm_path[MAXPATHLEN] = {0};
2232
2233 // Find the full path to the current module, libjvm.so or libjvm_g.so
2234 void os::jvm_path(char *buf, jint buflen) {
2235 // Error checking.
2236 if (buflen < MAXPATHLEN) {
2237 assert(false, "must use a large-enough buffer");
2238 buf[0] = '\0';
2239 return;
2240 }
2241 // Lazy resolve the path to current module.
2242 if (saved_jvm_path[0] != 0) {
2243 strcpy(buf, saved_jvm_path);
2244 return;
2245 }
2246
2247 char dli_fname[MAXPATHLEN];
2248 bool ret = dll_address_to_library_name(
2249 CAST_FROM_FN_PTR(address, os::jvm_path),
2250 dli_fname, sizeof(dli_fname), NULL);
2251 assert(ret != 0, "cannot locate libjvm");
2252 char *rp = realpath(dli_fname, buf);
2253 if (rp == NULL)
2254 return;
2255
2256 if (Arguments::created_by_gamma_launcher()) {
2257 // Support for the gamma launcher. Typical value for buf is
2258 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2259 // the right place in the string, then assume we are installed in a JDK and
2260 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2261 // up the path so it looks like libjvm.so is installed there (append a
2262 // fake suffix hotspot/libjvm.so).
2263 const char *p = buf + strlen(buf) - 1;
2264 for (int count = 0; p > buf && count < 5; ++count) {
2265 for (--p; p > buf && *p != '/'; --p)
2266 /* empty */ ;
2267 }
2268
2269 if (strncmp(p, "/jre/lib/", 9) != 0) {
2270 // Look for JAVA_HOME in the environment.
2271 char* java_home_var = ::getenv("JAVA_HOME");
2272 if (java_home_var != NULL && java_home_var[0] != 0) {
2273 char* jrelib_p;
2274 int len;
2275
2276 // Check the current module name "libjvm.so" or "libjvm_g.so".
2277 p = strrchr(buf, '/');
2278 assert(strstr(p, "/libjvm") == p, "invalid library name");
2279 p = strstr(p, "_g") ? "_g" : "";
2280
2281 rp = realpath(java_home_var, buf);
2282 if (rp == NULL)
2283 return;
2284
2285 // determine if this is a legacy image or modules image
2286 // modules image doesn't have "jre" subdirectory
2287 len = strlen(buf);
2288 jrelib_p = buf + len;
2289 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2290 if (0 != access(buf, F_OK)) {
2291 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2292 }
2293
2294 if (0 == access(buf, F_OK)) {
2295 // Use current module name "libjvm[_g].so" instead of
2296 // "libjvm"debug_only("_g")".so" since for fastdebug version
2297 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2298 len = strlen(buf);
2299 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2300 } else {
2301 // Go back to path of .so
2302 rp = realpath(dli_fname, buf);
2303 if (rp == NULL)
2304 return;
2305 }
2306 }
2307 }
2308 }
2309
2310 strcpy(saved_jvm_path, buf);
2311 }
2312
2313 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2314 // no prefix required, not even "_"
2315 }
2316
2317 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2318 // no suffix required
2319 }
2320
2321 ////////////////////////////////////////////////////////////////////////////////
2322 // sun.misc.Signal support
2323
2324 static volatile jint sigint_count = 0;
2325
2326 static void
2327 UserHandler(int sig, void *siginfo, void *context) {
2328 // 4511530 - sem_post is serialized and handled by the manager thread. When
2329 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2330 // don't want to flood the manager thread with sem_post requests.
2331 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2332 return;
2333
2334 // Ctrl-C is pressed during error reporting, likely because the error
2335 // handler fails to abort. Let VM die immediately.
2336 if (sig == SIGINT && is_error_reported()) {
2337 os::die();
2338 }
2339
2340 os::signal_notify(sig);
2341 }
2342
2343 void* os::user_handler() {
2344 return CAST_FROM_FN_PTR(void*, UserHandler);
2345 }
2346
2347 extern "C" {
2348 typedef void (*sa_handler_t)(int);
2349 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2350 }
2351
2352 void* os::signal(int signal_number, void* handler) {
2353 struct sigaction sigAct, oldSigAct;
2354
2355 sigfillset(&(sigAct.sa_mask));
2356 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2357 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2358
2359 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2360 // -1 means registration failed
2361 return (void *)-1;
2362 }
2363
2364 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2365 }
2366
2367 void os::signal_raise(int signal_number) {
2368 ::raise(signal_number);
2369 }
2370
2371 /*
2372 * The following code is moved from os.cpp for making this
2373 * code platform specific, which it is by its very nature.
2374 */
2375
2376 // Will be modified when max signal is changed to be dynamic
2377 int os::sigexitnum_pd() {
2378 return NSIG;
2379 }
2380
2381 // a counter for each possible signal value
2382 static volatile jint pending_signals[NSIG+1] = { 0 };
2383
2384 // Linux(POSIX) specific hand shaking semaphore.
2385 static sem_t sig_sem;
2386
2387 void os::signal_init_pd() {
2388 // Initialize signal structures
2389 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2390
2391 // Initialize signal semaphore
2392 ::sem_init(&sig_sem, 0, 0);
2393 }
2394
2395 void os::signal_notify(int sig) {
2396 Atomic::inc(&pending_signals[sig]);
2397 ::sem_post(&sig_sem);
2398 }
2399
2400 static int check_pending_signals(bool wait) {
2401 Atomic::store(0, &sigint_count);
2402 for (;;) {
2403 for (int i = 0; i < NSIG + 1; i++) {
2404 jint n = pending_signals[i];
2405 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2406 return i;
2407 }
2408 }
2409 if (!wait) {
2410 return -1;
2411 }
2412 JavaThread *thread = JavaThread::current();
2413 ThreadBlockInVM tbivm(thread);
2414
2415 bool threadIsSuspended;
2416 do {
2417 thread->set_suspend_equivalent();
2418 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2419 ::sem_wait(&sig_sem);
2420
2421 // were we externally suspended while we were waiting?
2422 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2423 if (threadIsSuspended) {
2424 //
2425 // The semaphore has been incremented, but while we were waiting
2426 // another thread suspended us. We don't want to continue running
2427 // while suspended because that would surprise the thread that
2428 // suspended us.
2429 //
2430 ::sem_post(&sig_sem);
2431
2432 thread->java_suspend_self();
2433 }
2434 } while (threadIsSuspended);
2435 }
2436 }
2437
2438 int os::signal_lookup() {
2439 return check_pending_signals(false);
2440 }
2441
2442 int os::signal_wait() {
2443 return check_pending_signals(true);
2444 }
2445
2446 ////////////////////////////////////////////////////////////////////////////////
2447 // Virtual Memory
2448
2449 int os::vm_page_size() {
2450 // Seems redundant as all get out
2451 assert(os::Linux::page_size() != -1, "must call os::init");
2452 return os::Linux::page_size();
2453 }
2454
2455 // Solaris allocates memory by pages.
2456 int os::vm_allocation_granularity() {
2457 assert(os::Linux::page_size() != -1, "must call os::init");
2458 return os::Linux::page_size();
2459 }
2460
2461 // Rationale behind this function:
2462 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2463 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2464 // samples for JITted code. Here we create private executable mapping over the code cache
2465 // and then we can use standard (well, almost, as mapping can change) way to provide
2466 // info for the reporting script by storing timestamp and location of symbol
2467 void linux_wrap_code(char* base, size_t size) {
2468 static volatile jint cnt = 0;
2469
2470 if (!UseOprofile) {
2471 return;
2472 }
2473
2474 char buf[PATH_MAX+1];
2475 int num = Atomic::add(1, &cnt);
2476
2477 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2478 os::get_temp_directory(), os::current_process_id(), num);
2479 unlink(buf);
2480
2481 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2482
2483 if (fd != -1) {
2484 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2485 if (rv != (off_t)-1) {
2486 if (::write(fd, "", 1) == 1) {
2487 mmap(base, size,
2488 PROT_READ|PROT_WRITE|PROT_EXEC,
2489 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2490 }
2491 }
2492 ::close(fd);
2493 unlink(buf);
2494 }
2495 }
2496
2497 // NOTE: Linux kernel does not really reserve the pages for us.
2498 // All it does is to check if there are enough free pages
2499 // left at the time of mmap(). This could be a potential
2500 // problem.
2501 bool os::commit_memory(char* addr, size_t size, bool exec) {
2502 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2503 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2504 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2505 return res != (uintptr_t) MAP_FAILED;
2506 }
2507
2508 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2509 #ifndef MAP_HUGETLB
2510 #define MAP_HUGETLB 0x40000
2511 #endif
2512
2513 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2514 #ifndef MADV_HUGEPAGE
2515 #define MADV_HUGEPAGE 14
2516 #endif
2517
2518 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2519 bool exec) {
2520 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2521 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2522 uintptr_t res =
2523 (uintptr_t) ::mmap(addr, size, prot,
2524 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2525 -1, 0);
2526 return res != (uintptr_t) MAP_FAILED;
2527 }
2528
2529 return commit_memory(addr, size, exec);
2530 }
2531
2532 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2533 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2534 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2535 // be supported or the memory may already be backed by huge pages.
2536 ::madvise(addr, bytes, MADV_HUGEPAGE);
2537 }
2538 }
2539
2540 void os::free_memory(char *addr, size_t bytes) {
2541 commit_memory(addr, bytes, false);
2542 }
2543
2544 void os::numa_make_global(char *addr, size_t bytes) {
2545 Linux::numa_interleave_memory(addr, bytes);
2546 }
2547
2548 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2549 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2550 }
2551
2552 bool os::numa_topology_changed() { return false; }
2553
2554 size_t os::numa_get_groups_num() {
2555 int max_node = Linux::numa_max_node();
2556 return max_node > 0 ? max_node + 1 : 1;
2557 }
2558
2559 int os::numa_get_group_id() {
2560 int cpu_id = Linux::sched_getcpu();
2561 if (cpu_id != -1) {
2562 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2563 if (lgrp_id != -1) {
2564 return lgrp_id;
2565 }
2566 }
2567 return 0;
2568 }
2569
2570 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2571 for (size_t i = 0; i < size; i++) {
2572 ids[i] = i;
2573 }
2574 return size;
2575 }
2576
2577 bool os::get_page_info(char *start, page_info* info) {
2578 return false;
2579 }
2580
2581 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2582 return end;
2583 }
2584
2585
2586 int os::Linux::sched_getcpu_syscall(void) {
2587 unsigned int cpu;
2588 int retval = -1;
2589
2590 #if defined(IA32)
2591 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2592 #elif defined(AMD64)
2593 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2594 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2595 retval = vgetcpu(&cpu, NULL, NULL);
2596 #endif
2597
2598 return (retval == -1) ? retval : cpu;
2599 }
2600
2601 // Something to do with the numa-aware allocator needs these symbols
2602 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2603 extern "C" JNIEXPORT void numa_error(char *where) { }
2604 extern "C" JNIEXPORT int fork1() { return fork(); }
2605
2606
2607 // If we are running with libnuma version > 2, then we should
2608 // be trying to use symbols with versions 1.1
2609 // If we are running with earlier version, which did not have symbol versions,
2610 // we should use the base version.
2611 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2612 void *f = dlvsym(handle, name, "libnuma_1.1");
2613 if (f == NULL) {
2614 f = dlsym(handle, name);
2615 }
2616 return f;
2617 }
2618
2619 bool os::Linux::libnuma_init() {
2620 // sched_getcpu() should be in libc.
2621 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2622 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2623
2624 // If it's not, try a direct syscall.
2625 if (sched_getcpu() == -1)
2626 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2627
2628 if (sched_getcpu() != -1) { // Does it work?
2629 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2630 if (handle != NULL) {
2631 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2632 libnuma_dlsym(handle, "numa_node_to_cpus")));
2633 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2634 libnuma_dlsym(handle, "numa_max_node")));
2635 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2636 libnuma_dlsym(handle, "numa_available")));
2637 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2638 libnuma_dlsym(handle, "numa_tonode_memory")));
2639 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2640 libnuma_dlsym(handle, "numa_interleave_memory")));
2641
2642
2643 if (numa_available() != -1) {
2644 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2645 // Create a cpu -> node mapping
2646 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2647 rebuild_cpu_to_node_map();
2648 return true;
2649 }
2650 }
2651 }
2652 return false;
2653 }
2654
2655 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2656 // The table is later used in get_node_by_cpu().
2657 void os::Linux::rebuild_cpu_to_node_map() {
2658 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2659 // in libnuma (possible values are starting from 16,
2660 // and continuing up with every other power of 2, but less
2661 // than the maximum number of CPUs supported by kernel), and
2662 // is a subject to change (in libnuma version 2 the requirements
2663 // are more reasonable) we'll just hardcode the number they use
2664 // in the library.
2665 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2666
2667 size_t cpu_num = os::active_processor_count();
2668 size_t cpu_map_size = NCPUS / BitsPerCLong;
2669 size_t cpu_map_valid_size =
2670 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2671
2672 cpu_to_node()->clear();
2673 cpu_to_node()->at_grow(cpu_num - 1);
2674 size_t node_num = numa_get_groups_num();
2675
2676 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2677 for (size_t i = 0; i < node_num; i++) {
2678 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2679 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2680 if (cpu_map[j] != 0) {
2681 for (size_t k = 0; k < BitsPerCLong; k++) {
2682 if (cpu_map[j] & (1UL << k)) {
2683 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2684 }
2685 }
2686 }
2687 }
2688 }
2689 }
2690 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2691 }
2692
2693 int os::Linux::get_node_by_cpu(int cpu_id) {
2694 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2695 return cpu_to_node()->at(cpu_id);
2696 }
2697 return -1;
2698 }
2699
2700 GrowableArray<int>* os::Linux::_cpu_to_node;
2701 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2702 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2703 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2704 os::Linux::numa_available_func_t os::Linux::_numa_available;
2705 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2706 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2707 unsigned long* os::Linux::_numa_all_nodes;
2708
2709 bool os::uncommit_memory(char* addr, size_t size) {
2710 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2711 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2712 return res != (uintptr_t) MAP_FAILED;
2713 }
2714
2715 // Linux uses a growable mapping for the stack, and if the mapping for
2716 // the stack guard pages is not removed when we detach a thread the
2717 // stack cannot grow beyond the pages where the stack guard was
2718 // mapped. If at some point later in the process the stack expands to
2719 // that point, the Linux kernel cannot expand the stack any further
2720 // because the guard pages are in the way, and a segfault occurs.
2721 //
2722 // However, it's essential not to split the stack region by unmapping
2723 // a region (leaving a hole) that's already part of the stack mapping,
2724 // so if the stack mapping has already grown beyond the guard pages at
2725 // the time we create them, we have to truncate the stack mapping.
2726 // So, we need to know the extent of the stack mapping when
2727 // create_stack_guard_pages() is called.
2728
2729 // Find the bounds of the stack mapping. Return true for success.
2730 //
2731 // We only need this for stacks that are growable: at the time of
2732 // writing thread stacks don't use growable mappings (i.e. those
2733 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2734 // only applies to the main thread.
2735
2736 static
2737 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2738
2739 char buf[128];
2740 int fd, sz;
2741
2742 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2743 return false;
2744 }
2745
2746 const char kw[] = "[stack]";
2747 const int kwlen = sizeof(kw)-1;
2748
2749 // Address part of /proc/self/maps couldn't be more than 128 bytes
2750 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2751 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2752 // Extract addresses
2753 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2754 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2755 if (sp >= *bottom && sp <= *top) {
2756 ::close(fd);
2757 return true;
2758 }
2759 }
2760 }
2761 }
2762
2763 ::close(fd);
2764 return false;
2765 }
2766
2767
2768 // If the (growable) stack mapping already extends beyond the point
2769 // where we're going to put our guard pages, truncate the mapping at
2770 // that point by munmap()ping it. This ensures that when we later
2771 // munmap() the guard pages we don't leave a hole in the stack
2772 // mapping. This only affects the main/initial thread, but guard
2773 // against future OS changes
2774 bool os::create_stack_guard_pages(char* addr, size_t size) {
2775 uintptr_t stack_extent, stack_base;
2776 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2777 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2778 assert(os::Linux::is_initial_thread(),
2779 "growable stack in non-initial thread");
2780 if (stack_extent < (uintptr_t)addr)
2781 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2782 }
2783
2784 return os::commit_memory(addr, size);
2785 }
2786
2787 // If this is a growable mapping, remove the guard pages entirely by
2788 // munmap()ping them. If not, just call uncommit_memory(). This only
2789 // affects the main/initial thread, but guard against future OS changes
2790 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2791 uintptr_t stack_extent, stack_base;
2792 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2793 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2794 assert(os::Linux::is_initial_thread(),
2795 "growable stack in non-initial thread");
2796
2797 return ::munmap(addr, size) == 0;
2798 }
2799
2800 return os::uncommit_memory(addr, size);
2801 }
2802
2803 static address _highest_vm_reserved_address = NULL;
2804
2805 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2806 // at 'requested_addr'. If there are existing memory mappings at the same
2807 // location, however, they will be overwritten. If 'fixed' is false,
2808 // 'requested_addr' is only treated as a hint, the return value may or
2809 // may not start from the requested address. Unlike Linux mmap(), this
2810 // function returns NULL to indicate failure.
2811 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2812 char * addr;
2813 int flags;
2814
2815 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2816 if (fixed) {
2817 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2818 flags |= MAP_FIXED;
2819 }
2820
2821 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2822 // to PROT_EXEC if executable when we commit the page.
2823 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2824 flags, -1, 0);
2825
2826 if (addr != MAP_FAILED) {
2827 // anon_mmap() should only get called during VM initialization,
2828 // don't need lock (actually we can skip locking even it can be called
2829 // from multiple threads, because _highest_vm_reserved_address is just a
2830 // hint about the upper limit of non-stack memory regions.)
2831 if ((address)addr + bytes > _highest_vm_reserved_address) {
2832 _highest_vm_reserved_address = (address)addr + bytes;
2833 }
2834 }
2835
2836 return addr == MAP_FAILED ? NULL : addr;
2837 }
2838
2839 // Don't update _highest_vm_reserved_address, because there might be memory
2840 // regions above addr + size. If so, releasing a memory region only creates
2841 // a hole in the address space, it doesn't help prevent heap-stack collision.
2842 //
2843 static int anon_munmap(char * addr, size_t size) {
2844 return ::munmap(addr, size) == 0;
2845 }
2846
2847 char* os::reserve_memory(size_t bytes, char* requested_addr,
2848 size_t alignment_hint) {
2849 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2850 }
2851
2852 bool os::release_memory(char* addr, size_t size) {
2853 return anon_munmap(addr, size);
2854 }
2855
2856 static address highest_vm_reserved_address() {
2857 return _highest_vm_reserved_address;
2858 }
2859
2860 static bool linux_mprotect(char* addr, size_t size, int prot) {
2861 // Linux wants the mprotect address argument to be page aligned.
2862 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2863
2864 // According to SUSv3, mprotect() should only be used with mappings
2865 // established by mmap(), and mmap() always maps whole pages. Unaligned
2866 // 'addr' likely indicates problem in the VM (e.g. trying to change
2867 // protection of malloc'ed or statically allocated memory). Check the
2868 // caller if you hit this assert.
2869 assert(addr == bottom, "sanity check");
2870
2871 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2872 return ::mprotect(bottom, size, prot) == 0;
2873 }
2874
2875 // Set protections specified
2876 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2877 bool is_committed) {
2878 unsigned int p = 0;
2879 switch (prot) {
2880 case MEM_PROT_NONE: p = PROT_NONE; break;
2881 case MEM_PROT_READ: p = PROT_READ; break;
2882 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2883 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2884 default:
2885 ShouldNotReachHere();
2886 }
2887 // is_committed is unused.
2888 return linux_mprotect(addr, bytes, p);
2889 }
2890
2891 bool os::guard_memory(char* addr, size_t size) {
2892 return linux_mprotect(addr, size, PROT_NONE);
2893 }
2894
2895 bool os::unguard_memory(char* addr, size_t size) {
2896 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2897 }
2898
2899 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2900 bool result = false;
2901 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2902 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2903 -1, 0);
2904
2905 if (p != (void *) -1) {
2906 // We don't know if this really is a huge page or not.
2907 FILE *fp = fopen("/proc/self/maps", "r");
2908 if (fp) {
2909 while (!feof(fp)) {
2910 char chars[257];
2911 long x = 0;
2912 if (fgets(chars, sizeof(chars), fp)) {
2913 if (sscanf(chars, "%lx-%*x", &x) == 1
2914 && x == (long)p) {
2915 if (strstr (chars, "hugepage")) {
2916 result = true;
2917 break;
2918 }
2919 }
2920 }
2921 }
2922 fclose(fp);
2923 }
2924 munmap (p, page_size);
2925 if (result)
2926 return true;
2927 }
2928
2929 if (warn) {
2930 warning("HugeTLBFS is not supported by the operating system.");
2931 }
2932
2933 return result;
2934 }
2935
2936 /*
2937 * Set the coredump_filter bits to include largepages in core dump (bit 6)
2938 *
2939 * From the coredump_filter documentation:
2940 *
2941 * - (bit 0) anonymous private memory
2942 * - (bit 1) anonymous shared memory
2943 * - (bit 2) file-backed private memory
2944 * - (bit 3) file-backed shared memory
2945 * - (bit 4) ELF header pages in file-backed private memory areas (it is
2946 * effective only if the bit 2 is cleared)
2947 * - (bit 5) hugetlb private memory
2948 * - (bit 6) hugetlb shared memory
2949 */
2950 static void set_coredump_filter(void) {
2951 FILE *f;
2952 long cdm;
2953
2954 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
2955 return;
2956 }
2957
2958 if (fscanf(f, "%lx", &cdm) != 1) {
2959 fclose(f);
2960 return;
2961 }
2962
2963 rewind(f);
2964
2965 if ((cdm & LARGEPAGES_BIT) == 0) {
2966 cdm |= LARGEPAGES_BIT;
2967 fprintf(f, "%#lx", cdm);
2968 }
2969
2970 fclose(f);
2971 }
2972
2973 // Large page support
2974
2975 static size_t _large_page_size = 0;
2976
2977 void os::large_page_init() {
2978 if (!UseLargePages) {
2979 UseHugeTLBFS = false;
2980 UseSHM = false;
2981 return;
2982 }
2983
2984 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
2985 // If UseLargePages is specified on the command line try both methods,
2986 // if it's default, then try only HugeTLBFS.
2987 if (FLAG_IS_DEFAULT(UseLargePages)) {
2988 UseHugeTLBFS = true;
2989 } else {
2990 UseHugeTLBFS = UseSHM = true;
2991 }
2992 }
2993
2994 if (LargePageSizeInBytes) {
2995 _large_page_size = LargePageSizeInBytes;
2996 } else {
2997 // large_page_size on Linux is used to round up heap size. x86 uses either
2998 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2999 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3000 // page as large as 256M.
3001 //
3002 // Here we try to figure out page size by parsing /proc/meminfo and looking
3003 // for a line with the following format:
3004 // Hugepagesize: 2048 kB
3005 //
3006 // If we can't determine the value (e.g. /proc is not mounted, or the text
3007 // format has been changed), we'll use the largest page size supported by
3008 // the processor.
3009
3010 #ifndef ZERO
3011 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3012 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3013 #endif // ZERO
3014
3015 FILE *fp = fopen("/proc/meminfo", "r");
3016 if (fp) {
3017 while (!feof(fp)) {
3018 int x = 0;
3019 char buf[16];
3020 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3021 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3022 _large_page_size = x * K;
3023 break;
3024 }
3025 } else {
3026 // skip to next line
3027 for (;;) {
3028 int ch = fgetc(fp);
3029 if (ch == EOF || ch == (int)'\n') break;
3030 }
3031 }
3032 }
3033 fclose(fp);
3034 }
3035 }
3036
3037 // print a warning if any large page related flag is specified on command line
3038 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3039
3040 const size_t default_page_size = (size_t)Linux::page_size();
3041 if (_large_page_size > default_page_size) {
3042 _page_sizes[0] = _large_page_size;
3043 _page_sizes[1] = default_page_size;
3044 _page_sizes[2] = 0;
3045 }
3046 UseHugeTLBFS = UseHugeTLBFS &&
3047 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3048
3049 if (UseHugeTLBFS)
3050 UseSHM = false;
3051
3052 UseLargePages = UseHugeTLBFS || UseSHM;
3053
3054 set_coredump_filter();
3055 }
3056
3057 #ifndef SHM_HUGETLB
3058 #define SHM_HUGETLB 04000
3059 #endif
3060
3061 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3062 // "exec" is passed in but not used. Creating the shared image for
3063 // the code cache doesn't have an SHM_X executable permission to check.
3064 assert(UseLargePages && UseSHM, "only for SHM large pages");
3065
3066 key_t key = IPC_PRIVATE;
3067 char *addr;
3068
3069 bool warn_on_failure = UseLargePages &&
3070 (!FLAG_IS_DEFAULT(UseLargePages) ||
3071 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3072 );
3073 char msg[128];
3074
3075 // Create a large shared memory region to attach to based on size.
3076 // Currently, size is the total size of the heap
3077 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3078 if (shmid == -1) {
3079 // Possible reasons for shmget failure:
3080 // 1. shmmax is too small for Java heap.
3081 // > check shmmax value: cat /proc/sys/kernel/shmmax
3082 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3083 // 2. not enough large page memory.
3084 // > check available large pages: cat /proc/meminfo
3085 // > increase amount of large pages:
3086 // echo new_value > /proc/sys/vm/nr_hugepages
3087 // Note 1: different Linux may use different name for this property,
3088 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3089 // Note 2: it's possible there's enough physical memory available but
3090 // they are so fragmented after a long run that they can't
3091 // coalesce into large pages. Try to reserve large pages when
3092 // the system is still "fresh".
3093 if (warn_on_failure) {
3094 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3095 warning(msg);
3096 }
3097 return NULL;
3098 }
3099
3100 // attach to the region
3101 addr = (char*)shmat(shmid, req_addr, 0);
3102 int err = errno;
3103
3104 // Remove shmid. If shmat() is successful, the actual shared memory segment
3105 // will be deleted when it's detached by shmdt() or when the process
3106 // terminates. If shmat() is not successful this will remove the shared
3107 // segment immediately.
3108 shmctl(shmid, IPC_RMID, NULL);
3109
3110 if ((intptr_t)addr == -1) {
3111 if (warn_on_failure) {
3112 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3113 warning(msg);
3114 }
3115 return NULL;
3116 }
3117
3118 return addr;
3119 }
3120
3121 bool os::release_memory_special(char* base, size_t bytes) {
3122 // detaching the SHM segment will also delete it, see reserve_memory_special()
3123 int rslt = shmdt(base);
3124 return rslt == 0;
3125 }
3126
3127 size_t os::large_page_size() {
3128 return _large_page_size;
3129 }
3130
3131 // HugeTLBFS allows application to commit large page memory on demand;
3132 // with SysV SHM the entire memory region must be allocated as shared
3133 // memory.
3134 bool os::can_commit_large_page_memory() {
3135 return UseHugeTLBFS;
3136 }
3137
3138 bool os::can_execute_large_page_memory() {
3139 return UseHugeTLBFS;
3140 }
3141
3142 // Reserve memory at an arbitrary address, only if that area is
3143 // available (and not reserved for something else).
3144
3145 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3146 const int max_tries = 10;
3147 char* base[max_tries];
3148 size_t size[max_tries];
3149 const size_t gap = 0x000000;
3150
3151 // Assert only that the size is a multiple of the page size, since
3152 // that's all that mmap requires, and since that's all we really know
3153 // about at this low abstraction level. If we need higher alignment,
3154 // we can either pass an alignment to this method or verify alignment
3155 // in one of the methods further up the call chain. See bug 5044738.
3156 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3157
3158 // Repeatedly allocate blocks until the block is allocated at the
3159 // right spot. Give up after max_tries. Note that reserve_memory() will
3160 // automatically update _highest_vm_reserved_address if the call is
3161 // successful. The variable tracks the highest memory address every reserved
3162 // by JVM. It is used to detect heap-stack collision if running with
3163 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3164 // space than needed, it could confuse the collision detecting code. To
3165 // solve the problem, save current _highest_vm_reserved_address and
3166 // calculate the correct value before return.
3167 address old_highest = _highest_vm_reserved_address;
3168
3169 // Linux mmap allows caller to pass an address as hint; give it a try first,
3170 // if kernel honors the hint then we can return immediately.
3171 char * addr = anon_mmap(requested_addr, bytes, false);
3172 if (addr == requested_addr) {
3173 return requested_addr;
3174 }
3175
3176 if (addr != NULL) {
3177 // mmap() is successful but it fails to reserve at the requested address
3178 anon_munmap(addr, bytes);
3179 }
3180
3181 int i;
3182 for (i = 0; i < max_tries; ++i) {
3183 base[i] = reserve_memory(bytes);
3184
3185 if (base[i] != NULL) {
3186 // Is this the block we wanted?
3187 if (base[i] == requested_addr) {
3188 size[i] = bytes;
3189 break;
3190 }
3191
3192 // Does this overlap the block we wanted? Give back the overlapped
3193 // parts and try again.
3194
3195 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3196 if (top_overlap >= 0 && top_overlap < bytes) {
3197 unmap_memory(base[i], top_overlap);
3198 base[i] += top_overlap;
3199 size[i] = bytes - top_overlap;
3200 } else {
3201 size_t bottom_overlap = base[i] + bytes - requested_addr;
3202 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3203 unmap_memory(requested_addr, bottom_overlap);
3204 size[i] = bytes - bottom_overlap;
3205 } else {
3206 size[i] = bytes;
3207 }
3208 }
3209 }
3210 }
3211
3212 // Give back the unused reserved pieces.
3213
3214 for (int j = 0; j < i; ++j) {
3215 if (base[j] != NULL) {
3216 unmap_memory(base[j], size[j]);
3217 }
3218 }
3219
3220 if (i < max_tries) {
3221 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3222 return requested_addr;
3223 } else {
3224 _highest_vm_reserved_address = old_highest;
3225 return NULL;
3226 }
3227 }
3228
3229 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3230 return ::read(fd, buf, nBytes);
3231 }
3232
3233 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3234 // Solaris uses poll(), linux uses park().
3235 // Poll() is likely a better choice, assuming that Thread.interrupt()
3236 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3237 // SIGSEGV, see 4355769.
3238
3239 const int NANOSECS_PER_MILLISECS = 1000000;
3240
3241 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3242 assert(thread == Thread::current(), "thread consistency check");
3243
3244 ParkEvent * const slp = thread->_SleepEvent ;
3245 slp->reset() ;
3246 OrderAccess::fence() ;
3247
3248 if (interruptible) {
3249 jlong prevtime = javaTimeNanos();
3250
3251 for (;;) {
3252 if (os::is_interrupted(thread, true)) {
3253 return OS_INTRPT;
3254 }
3255
3256 jlong newtime = javaTimeNanos();
3257
3258 if (newtime - prevtime < 0) {
3259 // time moving backwards, should only happen if no monotonic clock
3260 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3261 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3262 } else {
3263 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3264 }
3265
3266 if(millis <= 0) {
3267 return OS_OK;
3268 }
3269
3270 prevtime = newtime;
3271
3272 {
3273 assert(thread->is_Java_thread(), "sanity check");
3274 JavaThread *jt = (JavaThread *) thread;
3275 ThreadBlockInVM tbivm(jt);
3276 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3277
3278 jt->set_suspend_equivalent();
3279 // cleared by handle_special_suspend_equivalent_condition() or
3280 // java_suspend_self() via check_and_wait_while_suspended()
3281
3282 slp->park(millis);
3283
3284 // were we externally suspended while we were waiting?
3285 jt->check_and_wait_while_suspended();
3286 }
3287 }
3288 } else {
3289 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3290 jlong prevtime = javaTimeNanos();
3291
3292 for (;;) {
3293 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3294 // the 1st iteration ...
3295 jlong newtime = javaTimeNanos();
3296
3297 if (newtime - prevtime < 0) {
3298 // time moving backwards, should only happen if no monotonic clock
3299 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3300 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3301 } else {
3302 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3303 }
3304
3305 if(millis <= 0) break ;
3306
3307 prevtime = newtime;
3308 slp->park(millis);
3309 }
3310 return OS_OK ;
3311 }
3312 }
3313
3314 int os::naked_sleep() {
3315 // %% make the sleep time an integer flag. for now use 1 millisec.
3316 return os::sleep(Thread::current(), 1, false);
3317 }
3318
3319 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3320 void os::infinite_sleep() {
3321 while (true) { // sleep forever ...
3322 ::sleep(100); // ... 100 seconds at a time
3323 }
3324 }
3325
3326 // Used to convert frequent JVM_Yield() to nops
3327 bool os::dont_yield() {
3328 return DontYieldALot;
3329 }
3330
3331 void os::yield() {
3332 sched_yield();
3333 }
3334
3335 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3336
3337 void os::yield_all(int attempts) {
3338 // Yields to all threads, including threads with lower priorities
3339 // Threads on Linux are all with same priority. The Solaris style
3340 // os::yield_all() with nanosleep(1ms) is not necessary.
3341 sched_yield();
3342 }
3343
3344 // Called from the tight loops to possibly influence time-sharing heuristics
3345 void os::loop_breaker(int attempts) {
3346 os::yield_all(attempts);
3347 }
3348
3349 ////////////////////////////////////////////////////////////////////////////////
3350 // thread priority support
3351
3352 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3353 // only supports dynamic priority, static priority must be zero. For real-time
3354 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3355 // However, for large multi-threaded applications, SCHED_RR is not only slower
3356 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3357 // of 5 runs - Sep 2005).
3358 //
3359 // The following code actually changes the niceness of kernel-thread/LWP. It
3360 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3361 // not the entire user process, and user level threads are 1:1 mapped to kernel
3362 // threads. It has always been the case, but could change in the future. For
3363 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3364 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3365
3366 int os::java_to_os_priority[MaxPriority + 1] = {
3367 19, // 0 Entry should never be used
3368
3369 4, // 1 MinPriority
3370 3, // 2
3371 2, // 3
3372
3373 1, // 4
3374 0, // 5 NormPriority
3375 -1, // 6
3376
3377 -2, // 7
3378 -3, // 8
3379 -4, // 9 NearMaxPriority
3380
3381 -5 // 10 MaxPriority
3382 };
3383
3384 static int prio_init() {
3385 if (ThreadPriorityPolicy == 1) {
3386 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3387 // if effective uid is not root. Perhaps, a more elegant way of doing
3388 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3389 if (geteuid() != 0) {
3390 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3391 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3392 }
3393 ThreadPriorityPolicy = 0;
3394 }
3395 }
3396 return 0;
3397 }
3398
3399 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3400 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3401
3402 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3403 return (ret == 0) ? OS_OK : OS_ERR;
3404 }
3405
3406 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3407 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3408 *priority_ptr = java_to_os_priority[NormPriority];
3409 return OS_OK;
3410 }
3411
3412 errno = 0;
3413 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3414 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3415 }
3416
3417 // Hint to the underlying OS that a task switch would not be good.
3418 // Void return because it's a hint and can fail.
3419 void os::hint_no_preempt() {}
3420
3421 ////////////////////////////////////////////////////////////////////////////////
3422 // suspend/resume support
3423
3424 // the low-level signal-based suspend/resume support is a remnant from the
3425 // old VM-suspension that used to be for java-suspension, safepoints etc,
3426 // within hotspot. Now there is a single use-case for this:
3427 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3428 // that runs in the watcher thread.
3429 // The remaining code is greatly simplified from the more general suspension
3430 // code that used to be used.
3431 //
3432 // The protocol is quite simple:
3433 // - suspend:
3434 // - sends a signal to the target thread
3435 // - polls the suspend state of the osthread using a yield loop
3436 // - target thread signal handler (SR_handler) sets suspend state
3437 // and blocks in sigsuspend until continued
3438 // - resume:
3439 // - sets target osthread state to continue
3440 // - sends signal to end the sigsuspend loop in the SR_handler
3441 //
3442 // Note that the SR_lock plays no role in this suspend/resume protocol.
3443 //
3444
3445 static void resume_clear_context(OSThread *osthread) {
3446 osthread->set_ucontext(NULL);
3447 osthread->set_siginfo(NULL);
3448
3449 // notify the suspend action is completed, we have now resumed
3450 osthread->sr.clear_suspended();
3451 }
3452
3453 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3454 osthread->set_ucontext(context);
3455 osthread->set_siginfo(siginfo);
3456 }
3457
3458 //
3459 // Handler function invoked when a thread's execution is suspended or
3460 // resumed. We have to be careful that only async-safe functions are
3461 // called here (Note: most pthread functions are not async safe and
3462 // should be avoided.)
3463 //
3464 // Note: sigwait() is a more natural fit than sigsuspend() from an
3465 // interface point of view, but sigwait() prevents the signal hander
3466 // from being run. libpthread would get very confused by not having
3467 // its signal handlers run and prevents sigwait()'s use with the
3468 // mutex granting granting signal.
3469 //
3470 // Currently only ever called on the VMThread
3471 //
3472 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3473 // Save and restore errno to avoid confusing native code with EINTR
3474 // after sigsuspend.
3475 int old_errno = errno;
3476
3477 Thread* thread = Thread::current();
3478 OSThread* osthread = thread->osthread();
3479 assert(thread->is_VM_thread(), "Must be VMThread");
3480 // read current suspend action
3481 int action = osthread->sr.suspend_action();
3482 if (action == SR_SUSPEND) {
3483 suspend_save_context(osthread, siginfo, context);
3484
3485 // Notify the suspend action is about to be completed. do_suspend()
3486 // waits until SR_SUSPENDED is set and then returns. We will wait
3487 // here for a resume signal and that completes the suspend-other
3488 // action. do_suspend/do_resume is always called as a pair from
3489 // the same thread - so there are no races
3490
3491 // notify the caller
3492 osthread->sr.set_suspended();
3493
3494 sigset_t suspend_set; // signals for sigsuspend()
3495
3496 // get current set of blocked signals and unblock resume signal
3497 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3498 sigdelset(&suspend_set, SR_signum);
3499
3500 // wait here until we are resumed
3501 do {
3502 sigsuspend(&suspend_set);
3503 // ignore all returns until we get a resume signal
3504 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3505
3506 resume_clear_context(osthread);
3507
3508 } else {
3509 assert(action == SR_CONTINUE, "unexpected sr action");
3510 // nothing special to do - just leave the handler
3511 }
3512
3513 errno = old_errno;
3514 }
3515
3516
3517 static int SR_initialize() {
3518 struct sigaction act;
3519 char *s;
3520 /* Get signal number to use for suspend/resume */
3521 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3522 int sig = ::strtol(s, 0, 10);
3523 if (sig > 0 || sig < _NSIG) {
3524 SR_signum = sig;
3525 }
3526 }
3527
3528 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3529 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3530
3531 sigemptyset(&SR_sigset);
3532 sigaddset(&SR_sigset, SR_signum);
3533
3534 /* Set up signal handler for suspend/resume */
3535 act.sa_flags = SA_RESTART|SA_SIGINFO;
3536 act.sa_handler = (void (*)(int)) SR_handler;
3537
3538 // SR_signum is blocked by default.
3539 // 4528190 - We also need to block pthread restart signal (32 on all
3540 // supported Linux platforms). Note that LinuxThreads need to block
3541 // this signal for all threads to work properly. So we don't have
3542 // to use hard-coded signal number when setting up the mask.
3543 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3544
3545 if (sigaction(SR_signum, &act, 0) == -1) {
3546 return -1;
3547 }
3548
3549 // Save signal flag
3550 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3551 return 0;
3552 }
3553
3554 static int SR_finalize() {
3555 return 0;
3556 }
3557
3558
3559 // returns true on success and false on error - really an error is fatal
3560 // but this seems the normal response to library errors
3561 static bool do_suspend(OSThread* osthread) {
3562 // mark as suspended and send signal
3563 osthread->sr.set_suspend_action(SR_SUSPEND);
3564 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3565 assert_status(status == 0, status, "pthread_kill");
3566
3567 // check status and wait until notified of suspension
3568 if (status == 0) {
3569 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3570 os::yield_all(i);
3571 }
3572 osthread->sr.set_suspend_action(SR_NONE);
3573 return true;
3574 }
3575 else {
3576 osthread->sr.set_suspend_action(SR_NONE);
3577 return false;
3578 }
3579 }
3580
3581 static void do_resume(OSThread* osthread) {
3582 assert(osthread->sr.is_suspended(), "thread should be suspended");
3583 osthread->sr.set_suspend_action(SR_CONTINUE);
3584
3585 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3586 assert_status(status == 0, status, "pthread_kill");
3587 // check status and wait unit notified of resumption
3588 if (status == 0) {
3589 for (int i = 0; osthread->sr.is_suspended(); i++) {
3590 os::yield_all(i);
3591 }
3592 }
3593 osthread->sr.set_suspend_action(SR_NONE);
3594 }
3595
3596 ////////////////////////////////////////////////////////////////////////////////
3597 // interrupt support
3598
3599 void os::interrupt(Thread* thread) {
3600 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3601 "possibility of dangling Thread pointer");
3602
3603 OSThread* osthread = thread->osthread();
3604
3605 if (!osthread->interrupted()) {
3606 osthread->set_interrupted(true);
3607 // More than one thread can get here with the same value of osthread,
3608 // resulting in multiple notifications. We do, however, want the store
3609 // to interrupted() to be visible to other threads before we execute unpark().
3610 OrderAccess::fence();
3611 ParkEvent * const slp = thread->_SleepEvent ;
3612 if (slp != NULL) slp->unpark() ;
3613 }
3614
3615 // For JSR166. Unpark even if interrupt status already was set
3616 if (thread->is_Java_thread())
3617 ((JavaThread*)thread)->parker()->unpark();
3618
3619 ParkEvent * ev = thread->_ParkEvent ;
3620 if (ev != NULL) ev->unpark() ;
3621
3622 }
3623
3624 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3625 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3626 "possibility of dangling Thread pointer");
3627
3628 OSThread* osthread = thread->osthread();
3629
3630 bool interrupted = osthread->interrupted();
3631
3632 if (interrupted && clear_interrupted) {
3633 osthread->set_interrupted(false);
3634 // consider thread->_SleepEvent->reset() ... optional optimization
3635 }
3636
3637 return interrupted;
3638 }
3639
3640 ///////////////////////////////////////////////////////////////////////////////////
3641 // signal handling (except suspend/resume)
3642
3643 // This routine may be used by user applications as a "hook" to catch signals.
3644 // The user-defined signal handler must pass unrecognized signals to this
3645 // routine, and if it returns true (non-zero), then the signal handler must
3646 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3647 // routine will never retun false (zero), but instead will execute a VM panic
3648 // routine kill the process.
3649 //
3650 // If this routine returns false, it is OK to call it again. This allows
3651 // the user-defined signal handler to perform checks either before or after
3652 // the VM performs its own checks. Naturally, the user code would be making
3653 // a serious error if it tried to handle an exception (such as a null check
3654 // or breakpoint) that the VM was generating for its own correct operation.
3655 //
3656 // This routine may recognize any of the following kinds of signals:
3657 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3658 // It should be consulted by handlers for any of those signals.
3659 //
3660 // The caller of this routine must pass in the three arguments supplied
3661 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3662 // field of the structure passed to sigaction(). This routine assumes that
3663 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3664 //
3665 // Note that the VM will print warnings if it detects conflicting signal
3666 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3667 //
3668 extern "C" JNIEXPORT int
3669 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3670 void* ucontext, int abort_if_unrecognized);
3671
3672 void signalHandler(int sig, siginfo_t* info, void* uc) {
3673 assert(info != NULL && uc != NULL, "it must be old kernel");
3674 JVM_handle_linux_signal(sig, info, uc, true);
3675 }
3676
3677
3678 // This boolean allows users to forward their own non-matching signals
3679 // to JVM_handle_linux_signal, harmlessly.
3680 bool os::Linux::signal_handlers_are_installed = false;
3681
3682 // For signal-chaining
3683 struct sigaction os::Linux::sigact[MAXSIGNUM];
3684 unsigned int os::Linux::sigs = 0;
3685 bool os::Linux::libjsig_is_loaded = false;
3686 typedef struct sigaction *(*get_signal_t)(int);
3687 get_signal_t os::Linux::get_signal_action = NULL;
3688
3689 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3690 struct sigaction *actp = NULL;
3691
3692 if (libjsig_is_loaded) {
3693 // Retrieve the old signal handler from libjsig
3694 actp = (*get_signal_action)(sig);
3695 }
3696 if (actp == NULL) {
3697 // Retrieve the preinstalled signal handler from jvm
3698 actp = get_preinstalled_handler(sig);
3699 }
3700
3701 return actp;
3702 }
3703
3704 static bool call_chained_handler(struct sigaction *actp, int sig,
3705 siginfo_t *siginfo, void *context) {
3706 // Call the old signal handler
3707 if (actp->sa_handler == SIG_DFL) {
3708 // It's more reasonable to let jvm treat it as an unexpected exception
3709 // instead of taking the default action.
3710 return false;
3711 } else if (actp->sa_handler != SIG_IGN) {
3712 if ((actp->sa_flags & SA_NODEFER) == 0) {
3713 // automaticlly block the signal
3714 sigaddset(&(actp->sa_mask), sig);
3715 }
3716
3717 sa_handler_t hand;
3718 sa_sigaction_t sa;
3719 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3720 // retrieve the chained handler
3721 if (siginfo_flag_set) {
3722 sa = actp->sa_sigaction;
3723 } else {
3724 hand = actp->sa_handler;
3725 }
3726
3727 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3728 actp->sa_handler = SIG_DFL;
3729 }
3730
3731 // try to honor the signal mask
3732 sigset_t oset;
3733 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3734
3735 // call into the chained handler
3736 if (siginfo_flag_set) {
3737 (*sa)(sig, siginfo, context);
3738 } else {
3739 (*hand)(sig);
3740 }
3741
3742 // restore the signal mask
3743 pthread_sigmask(SIG_SETMASK, &oset, 0);
3744 }
3745 // Tell jvm's signal handler the signal is taken care of.
3746 return true;
3747 }
3748
3749 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3750 bool chained = false;
3751 // signal-chaining
3752 if (UseSignalChaining) {
3753 struct sigaction *actp = get_chained_signal_action(sig);
3754 if (actp != NULL) {
3755 chained = call_chained_handler(actp, sig, siginfo, context);
3756 }
3757 }
3758 return chained;
3759 }
3760
3761 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3762 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3763 return &sigact[sig];
3764 }
3765 return NULL;
3766 }
3767
3768 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3769 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3770 sigact[sig] = oldAct;
3771 sigs |= (unsigned int)1 << sig;
3772 }
3773
3774 // for diagnostic
3775 int os::Linux::sigflags[MAXSIGNUM];
3776
3777 int os::Linux::get_our_sigflags(int sig) {
3778 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3779 return sigflags[sig];
3780 }
3781
3782 void os::Linux::set_our_sigflags(int sig, int flags) {
3783 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3784 sigflags[sig] = flags;
3785 }
3786
3787 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3788 // Check for overwrite.
3789 struct sigaction oldAct;
3790 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3791
3792 void* oldhand = oldAct.sa_sigaction
3793 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3794 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3795 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3796 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3797 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3798 if (AllowUserSignalHandlers || !set_installed) {
3799 // Do not overwrite; user takes responsibility to forward to us.
3800 return;
3801 } else if (UseSignalChaining) {
3802 // save the old handler in jvm
3803 save_preinstalled_handler(sig, oldAct);
3804 // libjsig also interposes the sigaction() call below and saves the
3805 // old sigaction on it own.
3806 } else {
3807 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3808 "%#lx for signal %d.", (long)oldhand, sig));
3809 }
3810 }
3811
3812 struct sigaction sigAct;
3813 sigfillset(&(sigAct.sa_mask));
3814 sigAct.sa_handler = SIG_DFL;
3815 if (!set_installed) {
3816 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3817 } else {
3818 sigAct.sa_sigaction = signalHandler;
3819 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3820 }
3821 // Save flags, which are set by ours
3822 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3823 sigflags[sig] = sigAct.sa_flags;
3824
3825 int ret = sigaction(sig, &sigAct, &oldAct);
3826 assert(ret == 0, "check");
3827
3828 void* oldhand2 = oldAct.sa_sigaction
3829 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3830 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3831 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3832 }
3833
3834 // install signal handlers for signals that HotSpot needs to
3835 // handle in order to support Java-level exception handling.
3836
3837 void os::Linux::install_signal_handlers() {
3838 if (!signal_handlers_are_installed) {
3839 signal_handlers_are_installed = true;
3840
3841 // signal-chaining
3842 typedef void (*signal_setting_t)();
3843 signal_setting_t begin_signal_setting = NULL;
3844 signal_setting_t end_signal_setting = NULL;
3845 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3846 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3847 if (begin_signal_setting != NULL) {
3848 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3849 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3850 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3851 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3852 libjsig_is_loaded = true;
3853 assert(UseSignalChaining, "should enable signal-chaining");
3854 }
3855 if (libjsig_is_loaded) {
3856 // Tell libjsig jvm is setting signal handlers
3857 (*begin_signal_setting)();
3858 }
3859
3860 set_signal_handler(SIGSEGV, true);
3861 set_signal_handler(SIGPIPE, true);
3862 set_signal_handler(SIGBUS, true);
3863 set_signal_handler(SIGILL, true);
3864 set_signal_handler(SIGFPE, true);
3865 set_signal_handler(SIGXFSZ, true);
3866
3867 if (libjsig_is_loaded) {
3868 // Tell libjsig jvm finishes setting signal handlers
3869 (*end_signal_setting)();
3870 }
3871
3872 // We don't activate signal checker if libjsig is in place, we trust ourselves
3873 // and if UserSignalHandler is installed all bets are off
3874 if (CheckJNICalls) {
3875 if (libjsig_is_loaded) {
3876 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3877 check_signals = false;
3878 }
3879 if (AllowUserSignalHandlers) {
3880 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3881 check_signals = false;
3882 }
3883 }
3884 }
3885 }
3886
3887 // This is the fastest way to get thread cpu time on Linux.
3888 // Returns cpu time (user+sys) for any thread, not only for current.
3889 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3890 // It might work on 2.6.10+ with a special kernel/glibc patch.
3891 // For reference, please, see IEEE Std 1003.1-2004:
3892 // http://www.unix.org/single_unix_specification
3893
3894 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3895 struct timespec tp;
3896 int rc = os::Linux::clock_gettime(clockid, &tp);
3897 assert(rc == 0, "clock_gettime is expected to return 0 code");
3898
3899 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3900 }
3901
3902 /////
3903 // glibc on Linux platform uses non-documented flag
3904 // to indicate, that some special sort of signal
3905 // trampoline is used.
3906 // We will never set this flag, and we should
3907 // ignore this flag in our diagnostic
3908 #ifdef SIGNIFICANT_SIGNAL_MASK
3909 #undef SIGNIFICANT_SIGNAL_MASK
3910 #endif
3911 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3912
3913 static const char* get_signal_handler_name(address handler,
3914 char* buf, int buflen) {
3915 int offset;
3916 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3917 if (found) {
3918 // skip directory names
3919 const char *p1, *p2;
3920 p1 = buf;
3921 size_t len = strlen(os::file_separator());
3922 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3923 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3924 } else {
3925 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3926 }
3927 return buf;
3928 }
3929
3930 static void print_signal_handler(outputStream* st, int sig,
3931 char* buf, size_t buflen) {
3932 struct sigaction sa;
3933
3934 sigaction(sig, NULL, &sa);
3935
3936 // See comment for SIGNIFICANT_SIGNAL_MASK define
3937 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3938
3939 st->print("%s: ", os::exception_name(sig, buf, buflen));
3940
3941 address handler = (sa.sa_flags & SA_SIGINFO)
3942 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3943 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3944
3945 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3946 st->print("SIG_DFL");
3947 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3948 st->print("SIG_IGN");
3949 } else {
3950 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3951 }
3952
3953 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3954
3955 address rh = VMError::get_resetted_sighandler(sig);
3956 // May be, handler was resetted by VMError?
3957 if(rh != NULL) {
3958 handler = rh;
3959 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3960 }
3961
3962 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3963
3964 // Check: is it our handler?
3965 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3966 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3967 // It is our signal handler
3968 // check for flags, reset system-used one!
3969 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3970 st->print(
3971 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3972 os::Linux::get_our_sigflags(sig));
3973 }
3974 }
3975 st->cr();
3976 }
3977
3978
3979 #define DO_SIGNAL_CHECK(sig) \
3980 if (!sigismember(&check_signal_done, sig)) \
3981 os::Linux::check_signal_handler(sig)
3982
3983 // This method is a periodic task to check for misbehaving JNI applications
3984 // under CheckJNI, we can add any periodic checks here
3985
3986 void os::run_periodic_checks() {
3987
3988 if (check_signals == false) return;
3989
3990 // SEGV and BUS if overridden could potentially prevent
3991 // generation of hs*.log in the event of a crash, debugging
3992 // such a case can be very challenging, so we absolutely
3993 // check the following for a good measure:
3994 DO_SIGNAL_CHECK(SIGSEGV);
3995 DO_SIGNAL_CHECK(SIGILL);
3996 DO_SIGNAL_CHECK(SIGFPE);
3997 DO_SIGNAL_CHECK(SIGBUS);
3998 DO_SIGNAL_CHECK(SIGPIPE);
3999 DO_SIGNAL_CHECK(SIGXFSZ);
4000
4001
4002 // ReduceSignalUsage allows the user to override these handlers
4003 // see comments at the very top and jvm_solaris.h
4004 if (!ReduceSignalUsage) {
4005 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4006 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4007 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4008 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4009 }
4010
4011 DO_SIGNAL_CHECK(SR_signum);
4012 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4013 }
4014
4015 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4016
4017 static os_sigaction_t os_sigaction = NULL;
4018
4019 void os::Linux::check_signal_handler(int sig) {
4020 char buf[O_BUFLEN];
4021 address jvmHandler = NULL;
4022
4023
4024 struct sigaction act;
4025 if (os_sigaction == NULL) {
4026 // only trust the default sigaction, in case it has been interposed
4027 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4028 if (os_sigaction == NULL) return;
4029 }
4030
4031 os_sigaction(sig, (struct sigaction*)NULL, &act);
4032
4033
4034 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4035
4036 address thisHandler = (act.sa_flags & SA_SIGINFO)
4037 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4038 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4039
4040
4041 switch(sig) {
4042 case SIGSEGV:
4043 case SIGBUS:
4044 case SIGFPE:
4045 case SIGPIPE:
4046 case SIGILL:
4047 case SIGXFSZ:
4048 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4049 break;
4050
4051 case SHUTDOWN1_SIGNAL:
4052 case SHUTDOWN2_SIGNAL:
4053 case SHUTDOWN3_SIGNAL:
4054 case BREAK_SIGNAL:
4055 jvmHandler = (address)user_handler();
4056 break;
4057
4058 case INTERRUPT_SIGNAL:
4059 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4060 break;
4061
4062 default:
4063 if (sig == SR_signum) {
4064 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4065 } else {
4066 return;
4067 }
4068 break;
4069 }
4070
4071 if (thisHandler != jvmHandler) {
4072 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4073 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4074 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4075 // No need to check this sig any longer
4076 sigaddset(&check_signal_done, sig);
4077 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4078 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4079 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4080 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4081 // No need to check this sig any longer
4082 sigaddset(&check_signal_done, sig);
4083 }
4084
4085 // Dump all the signal
4086 if (sigismember(&check_signal_done, sig)) {
4087 print_signal_handlers(tty, buf, O_BUFLEN);
4088 }
4089 }
4090
4091 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4092
4093 extern bool signal_name(int signo, char* buf, size_t len);
4094
4095 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4096 if (0 < exception_code && exception_code <= SIGRTMAX) {
4097 // signal
4098 if (!signal_name(exception_code, buf, size)) {
4099 jio_snprintf(buf, size, "SIG%d", exception_code);
4100 }
4101 return buf;
4102 } else {
4103 return NULL;
4104 }
4105 }
4106
4107 // this is called _before_ the most of global arguments have been parsed
4108 void os::init(void) {
4109 char dummy; /* used to get a guess on initial stack address */
4110 // first_hrtime = gethrtime();
4111
4112 // With LinuxThreads the JavaMain thread pid (primordial thread)
4113 // is different than the pid of the java launcher thread.
4114 // So, on Linux, the launcher thread pid is passed to the VM
4115 // via the sun.java.launcher.pid property.
4116 // Use this property instead of getpid() if it was correctly passed.
4117 // See bug 6351349.
4118 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4119
4120 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4121
4122 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4123
4124 init_random(1234567);
4125
4126 ThreadCritical::initialize();
4127
4128 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4129 if (Linux::page_size() == -1) {
4130 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4131 strerror(errno)));
4132 }
4133 init_page_sizes((size_t) Linux::page_size());
4134
4135 Linux::initialize_system_info();
4136
4137 // main_thread points to the aboriginal thread
4138 Linux::_main_thread = pthread_self();
4139
4140 Linux::clock_init();
4141 initial_time_count = os::elapsed_counter();
4142 pthread_mutex_init(&dl_mutex, NULL);
4143 }
4144
4145 // To install functions for atexit system call
4146 extern "C" {
4147 static void perfMemory_exit_helper() {
4148 perfMemory_exit();
4149 }
4150 }
4151
4152 // this is called _after_ the global arguments have been parsed
4153 jint os::init_2(void)
4154 {
4155 Linux::fast_thread_clock_init();
4156
4157 // Allocate a single page and mark it as readable for safepoint polling
4158 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4159 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4160
4161 os::set_polling_page( polling_page );
4162
4163 #ifndef PRODUCT
4164 if(Verbose && PrintMiscellaneous)
4165 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4166 #endif
4167
4168 if (!UseMembar) {
4169 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4170 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4171 os::set_memory_serialize_page( mem_serialize_page );
4172
4173 #ifndef PRODUCT
4174 if(Verbose && PrintMiscellaneous)
4175 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4176 #endif
4177 }
4178
4179 os::large_page_init();
4180
4181 // initialize suspend/resume support - must do this before signal_sets_init()
4182 if (SR_initialize() != 0) {
4183 perror("SR_initialize failed");
4184 return JNI_ERR;
4185 }
4186
4187 Linux::signal_sets_init();
4188 Linux::install_signal_handlers();
4189
4190 // Check minimum allowable stack size for thread creation and to initialize
4191 // the java system classes, including StackOverflowError - depends on page
4192 // size. Add a page for compiler2 recursion in main thread.
4193 // Add in 2*BytesPerWord times page size to account for VM stack during
4194 // class initialization depending on 32 or 64 bit VM.
4195 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4196 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4197 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4198
4199 size_t threadStackSizeInBytes = ThreadStackSize * K;
4200 if (threadStackSizeInBytes != 0 &&
4201 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4202 tty->print_cr("\nThe stack size specified is too small, "
4203 "Specify at least %dk",
4204 os::Linux::min_stack_allowed/ K);
4205 return JNI_ERR;
4206 }
4207
4208 // Make the stack size a multiple of the page size so that
4209 // the yellow/red zones can be guarded.
4210 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4211 vm_page_size()));
4212
4213 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4214
4215 Linux::libpthread_init();
4216 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4217 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4218 Linux::glibc_version(), Linux::libpthread_version(),
4219 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4220 }
4221
4222 if (UseNUMA) {
4223 if (!Linux::libnuma_init()) {
4224 UseNUMA = false;
4225 } else {
4226 if ((Linux::numa_max_node() < 1)) {
4227 // There's only one node(they start from 0), disable NUMA.
4228 UseNUMA = false;
4229 }
4230 }
4231 // With SHM large pages we cannot uncommit a page, so there's not way
4232 // we can make the adaptive lgrp chunk resizing work. If the user specified
4233 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4234 // disable adaptive resizing.
4235 if (UseNUMA && UseLargePages && UseSHM) {
4236 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4237 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4238 UseLargePages = false;
4239 } else {
4240 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4241 UseAdaptiveSizePolicy = false;
4242 UseAdaptiveNUMAChunkSizing = false;
4243 }
4244 } else {
4245 UseNUMA = false;
4246 }
4247 }
4248 if (!UseNUMA && ForceNUMA) {
4249 UseNUMA = true;
4250 }
4251 }
4252
4253 if (MaxFDLimit) {
4254 // set the number of file descriptors to max. print out error
4255 // if getrlimit/setrlimit fails but continue regardless.
4256 struct rlimit nbr_files;
4257 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4258 if (status != 0) {
4259 if (PrintMiscellaneous && (Verbose || WizardMode))
4260 perror("os::init_2 getrlimit failed");
4261 } else {
4262 nbr_files.rlim_cur = nbr_files.rlim_max;
4263 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4264 if (status != 0) {
4265 if (PrintMiscellaneous && (Verbose || WizardMode))
4266 perror("os::init_2 setrlimit failed");
4267 }
4268 }
4269 }
4270
4271 // Initialize lock used to serialize thread creation (see os::create_thread)
4272 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4273
4274 // at-exit methods are called in the reverse order of their registration.
4275 // atexit functions are called on return from main or as a result of a
4276 // call to exit(3C). There can be only 32 of these functions registered
4277 // and atexit() does not set errno.
4278
4279 if (PerfAllowAtExitRegistration) {
4280 // only register atexit functions if PerfAllowAtExitRegistration is set.
4281 // atexit functions can be delayed until process exit time, which
4282 // can be problematic for embedded VM situations. Embedded VMs should
4283 // call DestroyJavaVM() to assure that VM resources are released.
4284
4285 // note: perfMemory_exit_helper atexit function may be removed in
4286 // the future if the appropriate cleanup code can be added to the
4287 // VM_Exit VMOperation's doit method.
4288 if (atexit(perfMemory_exit_helper) != 0) {
4289 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4290 }
4291 }
4292
4293 // initialize thread priority policy
4294 prio_init();
4295
4296 return JNI_OK;
4297 }
4298
4299 // this is called at the end of vm_initialization
4300 void os::init_3(void)
4301 {
4302 #ifdef JAVASE_EMBEDDED
4303 // Start the MemNotifyThread
4304 if (LowMemoryProtection) {
4305 MemNotifyThread::start();
4306 }
4307 return;
4308 #endif
4309 }
4310
4311 // Mark the polling page as unreadable
4312 void os::make_polling_page_unreadable(void) {
4313 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4314 fatal("Could not disable polling page");
4315 };
4316
4317 // Mark the polling page as readable
4318 void os::make_polling_page_readable(void) {
4319 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4320 fatal("Could not enable polling page");
4321 }
4322 };
4323
4324 int os::active_processor_count() {
4325 // Linux doesn't yet have a (official) notion of processor sets,
4326 // so just return the number of online processors.
4327 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4328 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4329 return online_cpus;
4330 }
4331
4332 bool os::distribute_processes(uint length, uint* distribution) {
4333 // Not yet implemented.
4334 return false;
4335 }
4336
4337 bool os::bind_to_processor(uint processor_id) {
4338 // Not yet implemented.
4339 return false;
4340 }
4341
4342 ///
4343
4344 // Suspends the target using the signal mechanism and then grabs the PC before
4345 // resuming the target. Used by the flat-profiler only
4346 ExtendedPC os::get_thread_pc(Thread* thread) {
4347 // Make sure that it is called by the watcher for the VMThread
4348 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4349 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4350
4351 ExtendedPC epc;
4352
4353 OSThread* osthread = thread->osthread();
4354 if (do_suspend(osthread)) {
4355 if (osthread->ucontext() != NULL) {
4356 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4357 } else {
4358 // NULL context is unexpected, double-check this is the VMThread
4359 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4360 }
4361 do_resume(osthread);
4362 }
4363 // failure means pthread_kill failed for some reason - arguably this is
4364 // a fatal problem, but such problems are ignored elsewhere
4365
4366 return epc;
4367 }
4368
4369 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4370 {
4371 if (is_NPTL()) {
4372 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4373 } else {
4374 #ifndef IA64
4375 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4376 // word back to default 64bit precision if condvar is signaled. Java
4377 // wants 53bit precision. Save and restore current value.
4378 int fpu = get_fpu_control_word();
4379 #endif // IA64
4380 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4381 #ifndef IA64
4382 set_fpu_control_word(fpu);
4383 #endif // IA64
4384 return status;
4385 }
4386 }
4387
4388 ////////////////////////////////////////////////////////////////////////////////
4389 // debug support
4390
4391 static address same_page(address x, address y) {
4392 int page_bits = -os::vm_page_size();
4393 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4394 return x;
4395 else if (x > y)
4396 return (address)(intptr_t(y) | ~page_bits) + 1;
4397 else
4398 return (address)(intptr_t(y) & page_bits);
4399 }
4400
4401 bool os::find(address addr, outputStream* st) {
4402 Dl_info dlinfo;
4403 memset(&dlinfo, 0, sizeof(dlinfo));
4404 if (dladdr(addr, &dlinfo)) {
4405 st->print(PTR_FORMAT ": ", addr);
4406 if (dlinfo.dli_sname != NULL) {
4407 st->print("%s+%#x", dlinfo.dli_sname,
4408 addr - (intptr_t)dlinfo.dli_saddr);
4409 } else if (dlinfo.dli_fname) {
4410 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4411 } else {
4412 st->print("<absolute address>");
4413 }
4414 if (dlinfo.dli_fname) {
4415 st->print(" in %s", dlinfo.dli_fname);
4416 }
4417 if (dlinfo.dli_fbase) {
4418 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4419 }
4420 st->cr();
4421
4422 if (Verbose) {
4423 // decode some bytes around the PC
4424 address begin = same_page(addr-40, addr);
4425 address end = same_page(addr+40, addr);
4426 address lowest = (address) dlinfo.dli_sname;
4427 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4428 if (begin < lowest) begin = lowest;
4429 Dl_info dlinfo2;
4430 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4431 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4432 end = (address) dlinfo2.dli_saddr;
4433 Disassembler::decode(begin, end, st);
4434 }
4435 return true;
4436 }
4437 return false;
4438 }
4439
4440 ////////////////////////////////////////////////////////////////////////////////
4441 // misc
4442
4443 // This does not do anything on Linux. This is basically a hook for being
4444 // able to use structured exception handling (thread-local exception filters)
4445 // on, e.g., Win32.
4446 void
4447 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4448 JavaCallArguments* args, Thread* thread) {
4449 f(value, method, args, thread);
4450 }
4451
4452 void os::print_statistics() {
4453 }
4454
4455 int os::message_box(const char* title, const char* message) {
4456 int i;
4457 fdStream err(defaultStream::error_fd());
4458 for (i = 0; i < 78; i++) err.print_raw("=");
4459 err.cr();
4460 err.print_raw_cr(title);
4461 for (i = 0; i < 78; i++) err.print_raw("-");
4462 err.cr();
4463 err.print_raw_cr(message);
4464 for (i = 0; i < 78; i++) err.print_raw("=");
4465 err.cr();
4466
4467 char buf[16];
4468 // Prevent process from exiting upon "read error" without consuming all CPU
4469 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4470
4471 return buf[0] == 'y' || buf[0] == 'Y';
4472 }
4473
4474 int os::stat(const char *path, struct stat *sbuf) {
4475 char pathbuf[MAX_PATH];
4476 if (strlen(path) > MAX_PATH - 1) {
4477 errno = ENAMETOOLONG;
4478 return -1;
4479 }
4480 os::native_path(strcpy(pathbuf, path));
4481 return ::stat(pathbuf, sbuf);
4482 }
4483
4484 bool os::check_heap(bool force) {
4485 return true;
4486 }
4487
4488 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4489 return ::vsnprintf(buf, count, format, args);
4490 }
4491
4492 // Is a (classpath) directory empty?
4493 bool os::dir_is_empty(const char* path) {
4494 DIR *dir = NULL;
4495 struct dirent *ptr;
4496
4497 dir = opendir(path);
4498 if (dir == NULL) return true;
4499
4500 /* Scan the directory */
4501 bool result = true;
4502 char buf[sizeof(struct dirent) + MAX_PATH];
4503 while (result && (ptr = ::readdir(dir)) != NULL) {
4504 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4505 result = false;
4506 }
4507 }
4508 closedir(dir);
4509 return result;
4510 }
4511
4512 // This code originates from JDK's sysOpen and open64_w
4513 // from src/solaris/hpi/src/system_md.c
4514
4515 #ifndef O_DELETE
4516 #define O_DELETE 0x10000
4517 #endif
4518
4519 // Open a file. Unlink the file immediately after open returns
4520 // if the specified oflag has the O_DELETE flag set.
4521 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4522
4523 int os::open(const char *path, int oflag, int mode) {
4524
4525 if (strlen(path) > MAX_PATH - 1) {
4526 errno = ENAMETOOLONG;
4527 return -1;
4528 }
4529 int fd;
4530 int o_delete = (oflag & O_DELETE);
4531 oflag = oflag & ~O_DELETE;
4532
4533 fd = ::open64(path, oflag, mode);
4534 if (fd == -1) return -1;
4535
4536 //If the open succeeded, the file might still be a directory
4537 {
4538 struct stat64 buf64;
4539 int ret = ::fstat64(fd, &buf64);
4540 int st_mode = buf64.st_mode;
4541
4542 if (ret != -1) {
4543 if ((st_mode & S_IFMT) == S_IFDIR) {
4544 errno = EISDIR;
4545 ::close(fd);
4546 return -1;
4547 }
4548 } else {
4549 ::close(fd);
4550 return -1;
4551 }
4552 }
4553
4554 /*
4555 * All file descriptors that are opened in the JVM and not
4556 * specifically destined for a subprocess should have the
4557 * close-on-exec flag set. If we don't set it, then careless 3rd
4558 * party native code might fork and exec without closing all
4559 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4560 * UNIXProcess.c), and this in turn might:
4561 *
4562 * - cause end-of-file to fail to be detected on some file
4563 * descriptors, resulting in mysterious hangs, or
4564 *
4565 * - might cause an fopen in the subprocess to fail on a system
4566 * suffering from bug 1085341.
4567 *
4568 * (Yes, the default setting of the close-on-exec flag is a Unix
4569 * design flaw)
4570 *
4571 * See:
4572 * 1085341: 32-bit stdio routines should support file descriptors >255
4573 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4574 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4575 */
4576 #ifdef FD_CLOEXEC
4577 {
4578 int flags = ::fcntl(fd, F_GETFD);
4579 if (flags != -1)
4580 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4581 }
4582 #endif
4583
4584 if (o_delete != 0) {
4585 ::unlink(path);
4586 }
4587 return fd;
4588 }
4589
4590
4591 // create binary file, rewriting existing file if required
4592 int os::create_binary_file(const char* path, bool rewrite_existing) {
4593 int oflags = O_WRONLY | O_CREAT;
4594 if (!rewrite_existing) {
4595 oflags |= O_EXCL;
4596 }
4597 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4598 }
4599
4600 // return current position of file pointer
4601 jlong os::current_file_offset(int fd) {
4602 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4603 }
4604
4605 // move file pointer to the specified offset
4606 jlong os::seek_to_file_offset(int fd, jlong offset) {
4607 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4608 }
4609
4610 // This code originates from JDK's sysAvailable
4611 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4612
4613 int os::available(int fd, jlong *bytes) {
4614 jlong cur, end;
4615 int mode;
4616 struct stat64 buf64;
4617
4618 if (::fstat64(fd, &buf64) >= 0) {
4619 mode = buf64.st_mode;
4620 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4621 /*
4622 * XXX: is the following call interruptible? If so, this might
4623 * need to go through the INTERRUPT_IO() wrapper as for other
4624 * blocking, interruptible calls in this file.
4625 */
4626 int n;
4627 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4628 *bytes = n;
4629 return 1;
4630 }
4631 }
4632 }
4633 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4634 return 0;
4635 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4636 return 0;
4637 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4638 return 0;
4639 }
4640 *bytes = end - cur;
4641 return 1;
4642 }
4643
4644 int os::socket_available(int fd, jint *pbytes) {
4645 // Linux doc says EINTR not returned, unlike Solaris
4646 int ret = ::ioctl(fd, FIONREAD, pbytes);
4647
4648 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4649 // is expected to return 0 on failure and 1 on success to the jdk.
4650 return (ret < 0) ? 0 : 1;
4651 }
4652
4653 // Map a block of memory.
4654 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4655 char *addr, size_t bytes, bool read_only,
4656 bool allow_exec) {
4657 int prot;
4658 int flags;
4659
4660 if (read_only) {
4661 prot = PROT_READ;
4662 flags = MAP_SHARED;
4663 } else {
4664 prot = PROT_READ | PROT_WRITE;
4665 flags = MAP_PRIVATE;
4666 }
4667
4668 if (allow_exec) {
4669 prot |= PROT_EXEC;
4670 }
4671
4672 if (addr != NULL) {
4673 flags |= MAP_FIXED;
4674 }
4675
4676 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4677 fd, file_offset);
4678 if (mapped_address == MAP_FAILED) {
4679 return NULL;
4680 }
4681 return mapped_address;
4682 }
4683
4684
4685 // Remap a block of memory.
4686 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4687 char *addr, size_t bytes, bool read_only,
4688 bool allow_exec) {
4689 // same as map_memory() on this OS
4690 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4691 allow_exec);
4692 }
4693
4694
4695 // Unmap a block of memory.
4696 bool os::unmap_memory(char* addr, size_t bytes) {
4697 return munmap(addr, bytes) == 0;
4698 }
4699
4700 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4701
4702 static clockid_t thread_cpu_clockid(Thread* thread) {
4703 pthread_t tid = thread->osthread()->pthread_id();
4704 clockid_t clockid;
4705
4706 // Get thread clockid
4707 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4708 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4709 return clockid;
4710 }
4711
4712 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4713 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4714 // of a thread.
4715 //
4716 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4717 // the fast estimate available on the platform.
4718
4719 jlong os::current_thread_cpu_time() {
4720 if (os::Linux::supports_fast_thread_cpu_time()) {
4721 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4722 } else {
4723 // return user + sys since the cost is the same
4724 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4725 }
4726 }
4727
4728 jlong os::thread_cpu_time(Thread* thread) {
4729 // consistent with what current_thread_cpu_time() returns
4730 if (os::Linux::supports_fast_thread_cpu_time()) {
4731 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4732 } else {
4733 return slow_thread_cpu_time(thread, true /* user + sys */);
4734 }
4735 }
4736
4737 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4738 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4739 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4740 } else {
4741 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4742 }
4743 }
4744
4745 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4746 if (user_sys_cpu_time && 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, user_sys_cpu_time);
4750 }
4751 }
4752
4753 //
4754 // -1 on error.
4755 //
4756
4757 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4758 static bool proc_pid_cpu_avail = true;
4759 static bool proc_task_unchecked = true;
4760 static const char *proc_stat_path = "/proc/%d/stat";
4761 pid_t tid = thread->osthread()->thread_id();
4762 int i;
4763 char *s;
4764 char stat[2048];
4765 int statlen;
4766 char proc_name[64];
4767 int count;
4768 long sys_time, user_time;
4769 char string[64];
4770 char cdummy;
4771 int idummy;
4772 long ldummy;
4773 FILE *fp;
4774
4775 // We first try accessing /proc/<pid>/cpu since this is faster to
4776 // process. If this file is not present (linux kernels 2.5 and above)
4777 // then we open /proc/<pid>/stat.
4778 if ( proc_pid_cpu_avail ) {
4779 sprintf(proc_name, "/proc/%d/cpu", tid);
4780 fp = fopen(proc_name, "r");
4781 if ( fp != NULL ) {
4782 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4783 fclose(fp);
4784 if ( count != 3 ) return -1;
4785
4786 if (user_sys_cpu_time) {
4787 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4788 } else {
4789 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4790 }
4791 }
4792 else proc_pid_cpu_avail = false;
4793 }
4794
4795 // The /proc/<tid>/stat aggregates per-process usage on
4796 // new Linux kernels 2.6+ where NPTL is supported.
4797 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4798 // See bug 6328462.
4799 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4800 // and possibly in some other cases, so we check its availability.
4801 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4802 // This is executed only once
4803 proc_task_unchecked = false;
4804 fp = fopen("/proc/self/task", "r");
4805 if (fp != NULL) {
4806 proc_stat_path = "/proc/self/task/%d/stat";
4807 fclose(fp);
4808 }
4809 }
4810
4811 sprintf(proc_name, proc_stat_path, tid);
4812 fp = fopen(proc_name, "r");
4813 if ( fp == NULL ) return -1;
4814 statlen = fread(stat, 1, 2047, fp);
4815 stat[statlen] = '\0';
4816 fclose(fp);
4817
4818 // Skip pid and the command string. Note that we could be dealing with
4819 // weird command names, e.g. user could decide to rename java launcher
4820 // to "java 1.4.2 :)", then the stat file would look like
4821 // 1234 (java 1.4.2 :)) R ... ...
4822 // We don't really need to know the command string, just find the last
4823 // occurrence of ")" and then start parsing from there. See bug 4726580.
4824 s = strrchr(stat, ')');
4825 i = 0;
4826 if (s == NULL ) return -1;
4827
4828 // Skip blank chars
4829 do s++; while (isspace(*s));
4830
4831 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4832 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4833 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4834 &user_time, &sys_time);
4835 if ( count != 13 ) return -1;
4836 if (user_sys_cpu_time) {
4837 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4838 } else {
4839 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4840 }
4841 }
4842
4843 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4844 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4845 info_ptr->may_skip_backward = false; // elapsed time not wall time
4846 info_ptr->may_skip_forward = false; // elapsed time not wall time
4847 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4848 }
4849
4850 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4851 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4852 info_ptr->may_skip_backward = false; // elapsed time not wall time
4853 info_ptr->may_skip_forward = false; // elapsed time not wall time
4854 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4855 }
4856
4857 bool os::is_thread_cpu_time_supported() {
4858 return true;
4859 }
4860
4861 // System loadavg support. Returns -1 if load average cannot be obtained.
4862 // Linux doesn't yet have a (official) notion of processor sets,
4863 // so just return the system wide load average.
4864 int os::loadavg(double loadavg[], int nelem) {
4865 return ::getloadavg(loadavg, nelem);
4866 }
4867
4868 void os::pause() {
4869 char filename[MAX_PATH];
4870 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4871 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4872 } else {
4873 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4874 }
4875
4876 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4877 if (fd != -1) {
4878 struct stat buf;
4879 ::close(fd);
4880 while (::stat(filename, &buf) == 0) {
4881 (void)::poll(NULL, 0, 100);
4882 }
4883 } else {
4884 jio_fprintf(stderr,
4885 "Could not open pause file '%s', continuing immediately.\n", filename);
4886 }
4887 }
4888
4889
4890 // Refer to the comments in os_solaris.cpp park-unpark.
4891 //
4892 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4893 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4894 // For specifics regarding the bug see GLIBC BUGID 261237 :
4895 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4896 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4897 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4898 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4899 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4900 // and monitorenter when we're using 1-0 locking. All those operations may result in
4901 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4902 // of libpthread avoids the problem, but isn't practical.
4903 //
4904 // Possible remedies:
4905 //
4906 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4907 // This is palliative and probabilistic, however. If the thread is preempted
4908 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4909 // than the minimum period may have passed, and the abstime may be stale (in the
4910 // past) resultin in a hang. Using this technique reduces the odds of a hang
4911 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4912 //
4913 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4914 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4915 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4916 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4917 // thread.
4918 //
4919 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4920 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4921 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4922 // This also works well. In fact it avoids kernel-level scalability impediments
4923 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4924 // timers in a graceful fashion.
4925 //
4926 // 4. When the abstime value is in the past it appears that control returns
4927 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4928 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4929 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4930 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4931 // It may be possible to avoid reinitialization by checking the return
4932 // value from pthread_cond_timedwait(). In addition to reinitializing the
4933 // condvar we must establish the invariant that cond_signal() is only called
4934 // within critical sections protected by the adjunct mutex. This prevents
4935 // cond_signal() from "seeing" a condvar that's in the midst of being
4936 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4937 // desirable signal-after-unlock optimization that avoids futile context switching.
4938 //
4939 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4940 // structure when a condvar is used or initialized. cond_destroy() would
4941 // release the helper structure. Our reinitialize-after-timedwait fix
4942 // put excessive stress on malloc/free and locks protecting the c-heap.
4943 //
4944 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4945 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4946 // and only enabling the work-around for vulnerable environments.
4947
4948 // utility to compute the abstime argument to timedwait:
4949 // millis is the relative timeout time
4950 // abstime will be the absolute timeout time
4951 // TODO: replace compute_abstime() with unpackTime()
4952
4953 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4954 if (millis < 0) millis = 0;
4955 struct timeval now;
4956 int status = gettimeofday(&now, NULL);
4957 assert(status == 0, "gettimeofday");
4958 jlong seconds = millis / 1000;
4959 millis %= 1000;
4960 if (seconds > 50000000) { // see man cond_timedwait(3T)
4961 seconds = 50000000;
4962 }
4963 abstime->tv_sec = now.tv_sec + seconds;
4964 long usec = now.tv_usec + millis * 1000;
4965 if (usec >= 1000000) {
4966 abstime->tv_sec += 1;
4967 usec -= 1000000;
4968 }
4969 abstime->tv_nsec = usec * 1000;
4970 return abstime;
4971 }
4972
4973
4974 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4975 // Conceptually TryPark() should be equivalent to park(0).
4976
4977 int os::PlatformEvent::TryPark() {
4978 for (;;) {
4979 const int v = _Event ;
4980 guarantee ((v == 0) || (v == 1), "invariant") ;
4981 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4982 }
4983 }
4984
4985 void os::PlatformEvent::park() { // AKA "down()"
4986 // Invariant: Only the thread associated with the Event/PlatformEvent
4987 // may call park().
4988 // TODO: assert that _Assoc != NULL or _Assoc == Self
4989 int v ;
4990 for (;;) {
4991 v = _Event ;
4992 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4993 }
4994 guarantee (v >= 0, "invariant") ;
4995 if (v == 0) {
4996 // Do this the hard way by blocking ...
4997 int status = pthread_mutex_lock(_mutex);
4998 assert_status(status == 0, status, "mutex_lock");
4999 guarantee (_nParked == 0, "invariant") ;
5000 ++ _nParked ;
5001 while (_Event < 0) {
5002 status = pthread_cond_wait(_cond, _mutex);
5003 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5004 // Treat this the same as if the wait was interrupted
5005 if (status == ETIME) { status = EINTR; }
5006 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5007 }
5008 -- _nParked ;
5009
5010 // In theory we could move the ST of 0 into _Event past the unlock(),
5011 // but then we'd need a MEMBAR after the ST.
5012 _Event = 0 ;
5013 status = pthread_mutex_unlock(_mutex);
5014 assert_status(status == 0, status, "mutex_unlock");
5015 }
5016 guarantee (_Event >= 0, "invariant") ;
5017 }
5018
5019 int os::PlatformEvent::park(jlong millis) {
5020 guarantee (_nParked == 0, "invariant") ;
5021
5022 int v ;
5023 for (;;) {
5024 v = _Event ;
5025 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5026 }
5027 guarantee (v >= 0, "invariant") ;
5028 if (v != 0) return OS_OK ;
5029
5030 // We do this the hard way, by blocking the thread.
5031 // Consider enforcing a minimum timeout value.
5032 struct timespec abst;
5033 compute_abstime(&abst, millis);
5034
5035 int ret = OS_TIMEOUT;
5036 int status = pthread_mutex_lock(_mutex);
5037 assert_status(status == 0, status, "mutex_lock");
5038 guarantee (_nParked == 0, "invariant") ;
5039 ++_nParked ;
5040
5041 // Object.wait(timo) will return because of
5042 // (a) notification
5043 // (b) timeout
5044 // (c) thread.interrupt
5045 //
5046 // Thread.interrupt and object.notify{All} both call Event::set.
5047 // That is, we treat thread.interrupt as a special case of notification.
5048 // The underlying Solaris implementation, cond_timedwait, admits
5049 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5050 // JVM from making those visible to Java code. As such, we must
5051 // filter out spurious wakeups. We assume all ETIME returns are valid.
5052 //
5053 // TODO: properly differentiate simultaneous notify+interrupt.
5054 // In that case, we should propagate the notify to another waiter.
5055
5056 while (_Event < 0) {
5057 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5058 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5059 pthread_cond_destroy (_cond);
5060 pthread_cond_init (_cond, NULL) ;
5061 }
5062 assert_status(status == 0 || status == EINTR ||
5063 status == ETIME || status == ETIMEDOUT,
5064 status, "cond_timedwait");
5065 if (!FilterSpuriousWakeups) break ; // previous semantics
5066 if (status == ETIME || status == ETIMEDOUT) break ;
5067 // We consume and ignore EINTR and spurious wakeups.
5068 }
5069 --_nParked ;
5070 if (_Event >= 0) {
5071 ret = OS_OK;
5072 }
5073 _Event = 0 ;
5074 status = pthread_mutex_unlock(_mutex);
5075 assert_status(status == 0, status, "mutex_unlock");
5076 assert (_nParked == 0, "invariant") ;
5077 return ret;
5078 }
5079
5080 void os::PlatformEvent::unpark() {
5081 int v, AnyWaiters ;
5082 for (;;) {
5083 v = _Event ;
5084 if (v > 0) {
5085 // The LD of _Event could have reordered or be satisfied
5086 // by a read-aside from this processor's write buffer.
5087 // To avoid problems execute a barrier and then
5088 // ratify the value.
5089 OrderAccess::fence() ;
5090 if (_Event == v) return ;
5091 continue ;
5092 }
5093 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5094 }
5095 if (v < 0) {
5096 // Wait for the thread associated with the event to vacate
5097 int status = pthread_mutex_lock(_mutex);
5098 assert_status(status == 0, status, "mutex_lock");
5099 AnyWaiters = _nParked ;
5100 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5101 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5102 AnyWaiters = 0 ;
5103 pthread_cond_signal (_cond);
5104 }
5105 status = pthread_mutex_unlock(_mutex);
5106 assert_status(status == 0, status, "mutex_unlock");
5107 if (AnyWaiters != 0) {
5108 status = pthread_cond_signal(_cond);
5109 assert_status(status == 0, status, "cond_signal");
5110 }
5111 }
5112
5113 // Note that we signal() _after dropping the lock for "immortal" Events.
5114 // This is safe and avoids a common class of futile wakeups. In rare
5115 // circumstances this can cause a thread to return prematurely from
5116 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5117 // simply re-test the condition and re-park itself.
5118 }
5119
5120
5121 // JSR166
5122 // -------------------------------------------------------
5123
5124 /*
5125 * The solaris and linux implementations of park/unpark are fairly
5126 * conservative for now, but can be improved. They currently use a
5127 * mutex/condvar pair, plus a a count.
5128 * Park decrements count if > 0, else does a condvar wait. Unpark
5129 * sets count to 1 and signals condvar. Only one thread ever waits
5130 * on the condvar. Contention seen when trying to park implies that someone
5131 * is unparking you, so don't wait. And spurious returns are fine, so there
5132 * is no need to track notifications.
5133 */
5134
5135
5136 #define NANOSECS_PER_SEC 1000000000
5137 #define NANOSECS_PER_MILLISEC 1000000
5138 #define MAX_SECS 100000000
5139 /*
5140 * This code is common to linux and solaris and will be moved to a
5141 * common place in dolphin.
5142 *
5143 * The passed in time value is either a relative time in nanoseconds
5144 * or an absolute time in milliseconds. Either way it has to be unpacked
5145 * into suitable seconds and nanoseconds components and stored in the
5146 * given timespec structure.
5147 * Given time is a 64-bit value and the time_t used in the timespec is only
5148 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5149 * overflow if times way in the future are given. Further on Solaris versions
5150 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5151 * number of seconds, in abstime, is less than current_time + 100,000,000.
5152 * As it will be 28 years before "now + 100000000" will overflow we can
5153 * ignore overflow and just impose a hard-limit on seconds using the value
5154 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5155 * years from "now".
5156 */
5157
5158 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5159 assert (time > 0, "convertTime");
5160
5161 struct timeval now;
5162 int status = gettimeofday(&now, NULL);
5163 assert(status == 0, "gettimeofday");
5164
5165 time_t max_secs = now.tv_sec + MAX_SECS;
5166
5167 if (isAbsolute) {
5168 jlong secs = time / 1000;
5169 if (secs > max_secs) {
5170 absTime->tv_sec = max_secs;
5171 }
5172 else {
5173 absTime->tv_sec = secs;
5174 }
5175 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5176 }
5177 else {
5178 jlong secs = time / NANOSECS_PER_SEC;
5179 if (secs >= MAX_SECS) {
5180 absTime->tv_sec = max_secs;
5181 absTime->tv_nsec = 0;
5182 }
5183 else {
5184 absTime->tv_sec = now.tv_sec + secs;
5185 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5186 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5187 absTime->tv_nsec -= NANOSECS_PER_SEC;
5188 ++absTime->tv_sec; // note: this must be <= max_secs
5189 }
5190 }
5191 }
5192 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5193 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5194 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5195 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5196 }
5197
5198 void Parker::park(bool isAbsolute, jlong time) {
5199 // Optional fast-path check:
5200 // Return immediately if a permit is available.
5201 if (_counter > 0) {
5202 _counter = 0 ;
5203 OrderAccess::fence();
5204 return ;
5205 }
5206
5207 Thread* thread = Thread::current();
5208 assert(thread->is_Java_thread(), "Must be JavaThread");
5209 JavaThread *jt = (JavaThread *)thread;
5210
5211 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5212 // Check interrupt before trying to wait
5213 if (Thread::is_interrupted(thread, false)) {
5214 return;
5215 }
5216
5217 // Next, demultiplex/decode time arguments
5218 timespec absTime;
5219 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5220 return;
5221 }
5222 if (time > 0) {
5223 unpackTime(&absTime, isAbsolute, time);
5224 }
5225
5226
5227 // Enter safepoint region
5228 // Beware of deadlocks such as 6317397.
5229 // The per-thread Parker:: mutex is a classic leaf-lock.
5230 // In particular a thread must never block on the Threads_lock while
5231 // holding the Parker:: mutex. If safepoints are pending both the
5232 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5233 ThreadBlockInVM tbivm(jt);
5234
5235 // Don't wait if cannot get lock since interference arises from
5236 // unblocking. Also. check interrupt before trying wait
5237 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5238 return;
5239 }
5240
5241 int status ;
5242 if (_counter > 0) { // no wait needed
5243 _counter = 0;
5244 status = pthread_mutex_unlock(_mutex);
5245 assert (status == 0, "invariant") ;
5246 OrderAccess::fence();
5247 return;
5248 }
5249
5250 #ifdef ASSERT
5251 // Don't catch signals while blocked; let the running threads have the signals.
5252 // (This allows a debugger to break into the running thread.)
5253 sigset_t oldsigs;
5254 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5255 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5256 #endif
5257
5258 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5259 jt->set_suspend_equivalent();
5260 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5261
5262 if (time == 0) {
5263 status = pthread_cond_wait (_cond, _mutex) ;
5264 } else {
5265 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5266 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5267 pthread_cond_destroy (_cond) ;
5268 pthread_cond_init (_cond, NULL);
5269 }
5270 }
5271 assert_status(status == 0 || status == EINTR ||
5272 status == ETIME || status == ETIMEDOUT,
5273 status, "cond_timedwait");
5274
5275 #ifdef ASSERT
5276 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5277 #endif
5278
5279 _counter = 0 ;
5280 status = pthread_mutex_unlock(_mutex) ;
5281 assert_status(status == 0, status, "invariant") ;
5282 // If externally suspended while waiting, re-suspend
5283 if (jt->handle_special_suspend_equivalent_condition()) {
5284 jt->java_suspend_self();
5285 }
5286
5287 OrderAccess::fence();
5288 }
5289
5290 void Parker::unpark() {
5291 int s, status ;
5292 status = pthread_mutex_lock(_mutex);
5293 assert (status == 0, "invariant") ;
5294 s = _counter;
5295 _counter = 1;
5296 if (s < 1) {
5297 if (WorkAroundNPTLTimedWaitHang) {
5298 status = pthread_cond_signal (_cond) ;
5299 assert (status == 0, "invariant") ;
5300 status = pthread_mutex_unlock(_mutex);
5301 assert (status == 0, "invariant") ;
5302 } else {
5303 status = pthread_mutex_unlock(_mutex);
5304 assert (status == 0, "invariant") ;
5305 status = pthread_cond_signal (_cond) ;
5306 assert (status == 0, "invariant") ;
5307 }
5308 } else {
5309 pthread_mutex_unlock(_mutex);
5310 assert (status == 0, "invariant") ;
5311 }
5312 }
5313
5314
5315 extern char** environ;
5316
5317 #ifndef __NR_fork
5318 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5319 #endif
5320
5321 #ifndef __NR_execve
5322 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5323 #endif
5324
5325 // Run the specified command in a separate process. Return its exit value,
5326 // or -1 on failure (e.g. can't fork a new process).
5327 // Unlike system(), this function can be called from signal handler. It
5328 // doesn't block SIGINT et al.
5329 int os::fork_and_exec(char* cmd) {
5330 const char * argv[4] = {"sh", "-c", cmd, NULL};
5331
5332 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5333 // pthread_atfork handlers and reset pthread library. All we need is a
5334 // separate process to execve. Make a direct syscall to fork process.
5335 // On IA64 there's no fork syscall, we have to use fork() and hope for
5336 // the best...
5337 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5338 IA64_ONLY(fork();)
5339
5340 if (pid < 0) {
5341 // fork failed
5342 return -1;
5343
5344 } else if (pid == 0) {
5345 // child process
5346
5347 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5348 // first to kill every thread on the thread list. Because this list is
5349 // not reset by fork() (see notes above), execve() will instead kill
5350 // every thread in the parent process. We know this is the only thread
5351 // in the new process, so make a system call directly.
5352 // IA64 should use normal execve() from glibc to match the glibc fork()
5353 // above.
5354 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5355 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5356
5357 // execve failed
5358 _exit(-1);
5359
5360 } else {
5361 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5362 // care about the actual exit code, for now.
5363
5364 int status;
5365
5366 // Wait for the child process to exit. This returns immediately if
5367 // the child has already exited. */
5368 while (waitpid(pid, &status, 0) < 0) {
5369 switch (errno) {
5370 case ECHILD: return 0;
5371 case EINTR: break;
5372 default: return -1;
5373 }
5374 }
5375
5376 if (WIFEXITED(status)) {
5377 // The child exited normally; get its exit code.
5378 return WEXITSTATUS(status);
5379 } else if (WIFSIGNALED(status)) {
5380 // The child exited because of a signal
5381 // The best value to return is 0x80 + signal number,
5382 // because that is what all Unix shells do, and because
5383 // it allows callers to distinguish between process exit and
5384 // process death by signal.
5385 return 0x80 + WTERMSIG(status);
5386 } else {
5387 // Unknown exit code; pass it through
5388 return status;
5389 }
5390 }
5391 }
5392
5393 // is_headless_jre()
5394 //
5395 // Test for the existence of libmawt in motif21 or xawt directories
5396 // in order to report if we are running in a headless jre
5397 //
5398 bool os::is_headless_jre() {
5399 struct stat statbuf;
5400 char buf[MAXPATHLEN];
5401 char libmawtpath[MAXPATHLEN];
5402 const char *xawtstr = "/xawt/libmawt.so";
5403 const char *motifstr = "/motif21/libmawt.so";
5404 char *p;
5405
5406 // Get path to libjvm.so
5407 os::jvm_path(buf, sizeof(buf));
5408
5409 // Get rid of libjvm.so
5410 p = strrchr(buf, '/');
5411 if (p == NULL) return false;
5412 else *p = '\0';
5413
5414 // Get rid of client or server
5415 p = strrchr(buf, '/');
5416 if (p == NULL) return false;
5417 else *p = '\0';
5418
5419 // check xawt/libmawt.so
5420 strcpy(libmawtpath, buf);
5421 strcat(libmawtpath, xawtstr);
5422 if (::stat(libmawtpath, &statbuf) == 0) return false;
5423
5424 // check motif21/libmawt.so
5425 strcpy(libmawtpath, buf);
5426 strcat(libmawtpath, motifstr);
5427 if (::stat(libmawtpath, &statbuf) == 0) return false;
5428
5429 return true;
5430 }
5431
5432
5433 #ifdef JAVASE_EMBEDDED
5434 //
5435 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5436 //
5437 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5438
5439 // ctor
5440 //
5441 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5442 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5443 _fd = fd;
5444
5445 if (os::create_thread(this, os::os_thread)) {
5446 _memnotify_thread = this;
5447 os::set_priority(this, NearMaxPriority);
5448 os::start_thread(this);
5449 }
5450 }
5451
5452 // Where all the work gets done
5453 //
5454 void MemNotifyThread::run() {
5455 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5456
5457 // Set up the select arguments
5458 fd_set rfds;
5459 if (_fd != -1) {
5460 FD_ZERO(&rfds);
5461 FD_SET(_fd, &rfds);
5462 }
5463
5464 // Now wait for the mem_notify device to wake up
5465 while (1) {
5466 // Wait for the mem_notify device to signal us..
5467 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5468 if (rc == -1) {
5469 perror("select!\n");
5470 break;
5471 } else if (rc) {
5472 //ssize_t free_before = os::available_memory();
5473 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5474
5475 // The kernel is telling us there is not much memory left...
5476 // try to do something about that
5477
5478 // If we are not already in a GC, try one.
5479 if (!Universe::heap()->is_gc_active()) {
5480 Universe::heap()->collect(GCCause::_allocation_failure);
5481
5482 //ssize_t free_after = os::available_memory();
5483 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5484 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5485 }
5486 // We might want to do something like the following if we find the GC's are not helping...
5487 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5488 }
5489 }
5490 }
5491
5492 //
5493 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5494 //
5495 void MemNotifyThread::start() {
5496 int fd;
5497 fd = open ("/dev/mem_notify", O_RDONLY, 0);
5498 if (fd < 0) {
5499 return;
5500 }
5501
5502 if (memnotify_thread() == NULL) {
5503 new MemNotifyThread(fd);
5504 }
5505 }
5506 #endif // JAVASE_EMBEDDED