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