1 /*
2 * Copyright (c) 1999, 2013, 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 "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_share_linux.hpp"
40 #include "prims/jniFastGetField.hpp"
41 #include "prims/jvm.h"
42 #include "prims/jvm_misc.hpp"
43 #include "runtime/arguments.hpp"
44 #include "runtime/extendedPC.hpp"
45 #include "runtime/globals.hpp"
46 #include "runtime/interfaceSupport.hpp"
47 #include "runtime/java.hpp"
48 #include "runtime/javaCalls.hpp"
49 #include "runtime/mutexLocker.hpp"
50 #include "runtime/objectMonitor.hpp"
51 #include "runtime/osThread.hpp"
52 #include "runtime/perfMemory.hpp"
53 #include "runtime/sharedRuntime.hpp"
54 #include "runtime/statSampler.hpp"
55 #include "runtime/stubRoutines.hpp"
56 #include "runtime/thread.inline.hpp"
57 #include "runtime/threadCritical.hpp"
58 #include "runtime/timer.hpp"
59 #include "services/attachListener.hpp"
60 #include "services/runtimeService.hpp"
61 #include "utilities/decoder.hpp"
62 #include "utilities/defaultStream.hpp"
63 #include "utilities/events.hpp"
64 #include "utilities/growableArray.hpp"
65 #include "utilities/vmError.hpp"
66
67 // put OS-includes here
68 # include <sys/types.h>
69 # include <sys/mman.h>
70 # include <sys/stat.h>
71 # include <sys/select.h>
72 # include <pthread.h>
73 # include <signal.h>
74 # include <errno.h>
75 # include <dlfcn.h>
76 # include <stdio.h>
77 # include <unistd.h>
78 # include <sys/resource.h>
79 # include <pthread.h>
80 # include <sys/stat.h>
81 # include <sys/time.h>
82 # include <sys/times.h>
83 # include <sys/utsname.h>
84 # include <sys/socket.h>
85 # include <sys/wait.h>
86 # include <pwd.h>
87 # include <poll.h>
88 # include <semaphore.h>
89 # include <fcntl.h>
90 # include <string.h>
91 # include <syscall.h>
92 # include <sys/sysinfo.h>
93 # include <gnu/libc-version.h>
94 # include <sys/ipc.h>
95 # include <sys/shm.h>
96 # include <link.h>
97 # include <stdint.h>
98 # include <inttypes.h>
99 # include <sys/ioctl.h>
100
101 #define MAX_PATH (2 * K)
102
103 // for timer info max values which include all bits
104 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
105
106 #define LARGEPAGES_BIT (1 << 6)
107 ////////////////////////////////////////////////////////////////////////////////
108 // global variables
109 julong os::Linux::_physical_memory = 0;
110
111 address os::Linux::_initial_thread_stack_bottom = NULL;
112 uintptr_t os::Linux::_initial_thread_stack_size = 0;
113
114 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
115 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
116 Mutex* os::Linux::_createThread_lock = NULL;
117 pthread_t os::Linux::_main_thread;
118 int os::Linux::_page_size = -1;
119 bool os::Linux::_is_floating_stack = false;
120 bool os::Linux::_is_NPTL = false;
121 bool os::Linux::_supports_fast_thread_cpu_time = false;
122 const char * os::Linux::_glibc_version = NULL;
123 const char * os::Linux::_libpthread_version = NULL;
124
125 static jlong initial_time_count=0;
126
127 static int clock_tics_per_sec = 100;
128
129 // For diagnostics to print a message once. see run_periodic_checks
130 static sigset_t check_signal_done;
131 static bool check_signals = true;;
132
133 static pid_t _initial_pid = 0;
134
135 /* Signal number used to suspend/resume a thread */
136
137 /* do not use any signal number less than SIGSEGV, see 4355769 */
138 static int SR_signum = SIGUSR2;
139 sigset_t SR_sigset;
140
141 /* Used to protect dlsym() calls */
142 static pthread_mutex_t dl_mutex;
143
144 #ifdef JAVASE_EMBEDDED
145 class MemNotifyThread: public Thread {
146 friend class VMStructs;
147 public:
148 virtual void run();
149
150 private:
151 static MemNotifyThread* _memnotify_thread;
152 int _fd;
153
154 public:
155
156 // Constructor
157 MemNotifyThread(int fd);
158
159 // Tester
160 bool is_memnotify_thread() const { return true; }
161
162 // Printing
163 char* name() const { return (char*)"Linux MemNotify Thread"; }
164
165 // Returns the single instance of the MemNotifyThread
166 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
167
168 // Create and start the single instance of MemNotifyThread
169 static void start();
170 };
171 #endif // JAVASE_EMBEDDED
172
173 // utility functions
174
175 static int SR_initialize();
176 static int SR_finalize();
177
178 julong os::available_memory() {
179 return Linux::available_memory();
180 }
181
182 julong os::Linux::available_memory() {
183 // values in struct sysinfo are "unsigned long"
184 struct sysinfo si;
185 sysinfo(&si);
186
187 return (julong)si.freeram * si.mem_unit;
188 }
189
190 julong os::physical_memory() {
191 return Linux::physical_memory();
192 }
193
194 bool os::has_allocatable_memory_limit(julong& limit) {
195 struct rlimit rlim;
196 int getrlimit_res = getrlimit(RLIMIT_AS, &rlim);
197 // if there was an error when calling getrlimit, assume that there is no limitation
198 // on virtual memory.
199 bool result;
200 if ((getrlimit_res != 0) || (rlim.rlim_cur == RLIM_INFINITY)) {
201 result = false;
202 } else {
203 limit = (julong)rlim.rlim_cur;
204 result = true;
205 }
206 #ifdef _LP64
207 return result;
208 #else
209 // arbitrary virtual space limit for Linux found by testing. If getrlimit
210 // above returned a limit, bound it with this limit. Otherwise directly use
211 // it as limit.
212 const julong max_virtual_limit = (julong)3800*M;
213 if (result) {
214 limit = MIN2(limit, max_virtual_limit);
215 } else {
216 limit = max_virtual_limit;
217 }
218 if (!is_allocatable(limit)) {
219 // See comments under solaris for alignment considerations
220 julong reasonable_limit = (julong)2*G - 2 * os::vm_page_size();
221 limit = MIN2(limit, reasonable_limit);
222 }
223 return true;
224 #endif
225 }
226
227 ////////////////////////////////////////////////////////////////////////////////
228 // environment support
229
230 bool os::getenv(const char* name, char* buf, int len) {
231 const char* val = ::getenv(name);
232 if (val != NULL && strlen(val) < (size_t)len) {
233 strcpy(buf, val);
234 return true;
235 }
236 if (len > 0) buf[0] = 0; // return a null string
237 return false;
238 }
239
240
241 // Return true if user is running as root.
242
243 bool os::have_special_privileges() {
244 static bool init = false;
245 static bool privileges = false;
246 if (!init) {
247 privileges = (getuid() != geteuid()) || (getgid() != getegid());
248 init = true;
249 }
250 return privileges;
251 }
252
253
254 #ifndef SYS_gettid
255 // i386: 224, ia64: 1105, amd64: 186, sparc 143
256 #ifdef __ia64__
257 #define SYS_gettid 1105
258 #elif __i386__
259 #define SYS_gettid 224
260 #elif __amd64__
261 #define SYS_gettid 186
262 #elif __sparc__
263 #define SYS_gettid 143
264 #else
265 #error define gettid for the arch
266 #endif
267 #endif
268
269 // Cpu architecture string
270 #if defined(ZERO)
271 static char cpu_arch[] = ZERO_LIBARCH;
272 #elif defined(IA64)
273 static char cpu_arch[] = "ia64";
274 #elif defined(IA32)
275 static char cpu_arch[] = "i386";
276 #elif defined(AMD64)
277 static char cpu_arch[] = "amd64";
278 #elif defined(ARM)
279 static char cpu_arch[] = "arm";
280 #elif defined(PPC)
281 static char cpu_arch[] = "ppc";
282 #elif defined(SPARC)
283 # ifdef _LP64
284 static char cpu_arch[] = "sparcv9";
285 # else
286 static char cpu_arch[] = "sparc";
287 # endif
288 #else
289 #error Add appropriate cpu_arch setting
290 #endif
291
292
293 // pid_t gettid()
294 //
295 // Returns the kernel thread id of the currently running thread. Kernel
296 // thread id is used to access /proc.
297 //
298 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
299 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
300 //
301 pid_t os::Linux::gettid() {
302 int rslt = syscall(SYS_gettid);
303 if (rslt == -1) {
304 // old kernel, no NPTL support
305 return getpid();
306 } else {
307 return (pid_t)rslt;
308 }
309 }
310
311 // Most versions of linux have a bug where the number of processors are
312 // determined by looking at the /proc file system. In a chroot environment,
313 // the system call returns 1. This causes the VM to act as if it is
314 // a single processor and elide locking (see is_MP() call).
315 static bool unsafe_chroot_detected = false;
316 static const char *unstable_chroot_error = "/proc file system not found.\n"
317 "Java may be unstable running multithreaded in a chroot "
318 "environment on Linux when /proc filesystem is not mounted.";
319
320 void os::Linux::initialize_system_info() {
321 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
322 if (processor_count() == 1) {
323 pid_t pid = os::Linux::gettid();
324 char fname[32];
325 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
326 FILE *fp = fopen(fname, "r");
327 if (fp == NULL) {
328 unsafe_chroot_detected = true;
329 } else {
330 fclose(fp);
331 }
332 }
333 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
334 assert(processor_count() > 0, "linux error");
335 }
336
337 void os::init_system_properties_values() {
338 // char arch[12];
339 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
340
341 // The next steps are taken in the product version:
342 //
343 // Obtain the JAVA_HOME value from the location of libjvm.so.
344 // This library should be located at:
345 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
346 //
347 // If "/jre/lib/" appears at the right place in the path, then we
348 // assume libjvm.so is installed in a JDK and we use this path.
349 //
350 // Otherwise exit with message: "Could not create the Java virtual machine."
351 //
352 // The following extra steps are taken in the debugging version:
353 //
354 // If "/jre/lib/" does NOT appear at the right place in the path
355 // instead of exit check for $JAVA_HOME environment variable.
356 //
357 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
358 // then we append a fake suffix "hotspot/libjvm.so" to this path so
359 // it looks like libjvm.so is installed there
360 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
361 //
362 // Otherwise exit.
363 //
364 // Important note: if the location of libjvm.so changes this
365 // code needs to be changed accordingly.
366
367 // The next few definitions allow the code to be verbatim:
368 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
369 #define getenv(n) ::getenv(n)
370
371 /*
372 * See ld(1):
373 * The linker uses the following search paths to locate required
374 * shared libraries:
375 * 1: ...
376 * ...
377 * 7: The default directories, normally /lib and /usr/lib.
378 */
379 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
380 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
381 #else
382 #define DEFAULT_LIBPATH "/lib:/usr/lib"
383 #endif
384
385 #define EXTENSIONS_DIR "/lib/ext"
386 #define ENDORSED_DIR "/lib/endorsed"
387 #define REG_DIR "/usr/java/packages"
388
389 {
390 /* sysclasspath, java_home, dll_dir */
391 {
392 char *home_path;
393 char *dll_path;
394 char *pslash;
395 char buf[MAXPATHLEN];
396 os::jvm_path(buf, sizeof(buf));
397
398 // Found the full path to libjvm.so.
399 // Now cut the path to <java_home>/jre if we can.
400 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
401 pslash = strrchr(buf, '/');
402 if (pslash != NULL)
403 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
404 dll_path = malloc(strlen(buf) + 1);
405 if (dll_path == NULL)
406 return;
407 strcpy(dll_path, buf);
408 Arguments::set_dll_dir(dll_path);
409
410 if (pslash != NULL) {
411 pslash = strrchr(buf, '/');
412 if (pslash != NULL) {
413 *pslash = '\0'; /* get rid of /<arch> */
414 pslash = strrchr(buf, '/');
415 if (pslash != NULL)
416 *pslash = '\0'; /* get rid of /lib */
417 }
418 }
419
420 home_path = malloc(strlen(buf) + 1);
421 if (home_path == NULL)
422 return;
423 strcpy(home_path, buf);
424 Arguments::set_java_home(home_path);
425
426 if (!set_boot_path('/', ':'))
427 return;
428 }
429
430 /*
431 * Where to look for native libraries
432 *
433 * Note: Due to a legacy implementation, most of the library path
434 * is set in the launcher. This was to accomodate linking restrictions
435 * on legacy Linux implementations (which are no longer supported).
436 * Eventually, all the library path setting will be done here.
437 *
438 * However, to prevent the proliferation of improperly built native
439 * libraries, the new path component /usr/java/packages is added here.
440 * Eventually, all the library path setting will be done here.
441 */
442 {
443 char *ld_library_path;
444
445 /*
446 * Construct the invariant part of ld_library_path. Note that the
447 * space for the colon and the trailing null are provided by the
448 * nulls included by the sizeof operator (so actually we allocate
449 * a byte more than necessary).
450 */
451 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
452 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
453 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
454
455 /*
456 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
457 * should always exist (until the legacy problem cited above is
458 * addressed).
459 */
460 char *v = getenv("LD_LIBRARY_PATH");
461 if (v != NULL) {
462 char *t = ld_library_path;
463 /* That's +1 for the colon and +1 for the trailing '\0' */
464 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
465 sprintf(ld_library_path, "%s:%s", v, t);
466 }
467 Arguments::set_library_path(ld_library_path);
468 }
469
470 /*
471 * Extensions directories.
472 *
473 * Note that the space for the colon and the trailing null are provided
474 * by the nulls included by the sizeof operator (so actually one byte more
475 * than necessary is allocated).
476 */
477 {
478 char *buf = malloc(strlen(Arguments::get_java_home()) +
479 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
480 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
481 Arguments::get_java_home());
482 Arguments::set_ext_dirs(buf);
483 }
484
485 /* Endorsed standards default directory. */
486 {
487 char * buf;
488 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
489 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
490 Arguments::set_endorsed_dirs(buf);
491 }
492 }
493
494 #undef malloc
495 #undef getenv
496 #undef EXTENSIONS_DIR
497 #undef ENDORSED_DIR
498
499 // Done
500 return;
501 }
502
503 ////////////////////////////////////////////////////////////////////////////////
504 // breakpoint support
505
506 void os::breakpoint() {
507 BREAKPOINT;
508 }
509
510 extern "C" void breakpoint() {
511 // use debugger to set breakpoint here
512 }
513
514 ////////////////////////////////////////////////////////////////////////////////
515 // signal support
516
517 debug_only(static bool signal_sets_initialized = false);
518 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
519
520 bool os::Linux::is_sig_ignored(int sig) {
521 struct sigaction oact;
522 sigaction(sig, (struct sigaction*)NULL, &oact);
523 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
524 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
525 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
526 return true;
527 else
528 return false;
529 }
530
531 void os::Linux::signal_sets_init() {
532 // Should also have an assertion stating we are still single-threaded.
533 assert(!signal_sets_initialized, "Already initialized");
534 // Fill in signals that are necessarily unblocked for all threads in
535 // the VM. Currently, we unblock the following signals:
536 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
537 // by -Xrs (=ReduceSignalUsage));
538 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
539 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
540 // the dispositions or masks wrt these signals.
541 // Programs embedding the VM that want to use the above signals for their
542 // own purposes must, at this time, use the "-Xrs" option to prevent
543 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
544 // (See bug 4345157, and other related bugs).
545 // In reality, though, unblocking these signals is really a nop, since
546 // these signals are not blocked by default.
547 sigemptyset(&unblocked_sigs);
548 sigemptyset(&allowdebug_blocked_sigs);
549 sigaddset(&unblocked_sigs, SIGILL);
550 sigaddset(&unblocked_sigs, SIGSEGV);
551 sigaddset(&unblocked_sigs, SIGBUS);
552 sigaddset(&unblocked_sigs, SIGFPE);
553 sigaddset(&unblocked_sigs, SR_signum);
554
555 if (!ReduceSignalUsage) {
556 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
557 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
558 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
559 }
560 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
561 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
562 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
563 }
564 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
565 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
566 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
567 }
568 }
569 // Fill in signals that are blocked by all but the VM thread.
570 sigemptyset(&vm_sigs);
571 if (!ReduceSignalUsage)
572 sigaddset(&vm_sigs, BREAK_SIGNAL);
573 debug_only(signal_sets_initialized = true);
574
575 }
576
577 // These are signals that are unblocked while a thread is running Java.
578 // (For some reason, they get blocked by default.)
579 sigset_t* os::Linux::unblocked_signals() {
580 assert(signal_sets_initialized, "Not initialized");
581 return &unblocked_sigs;
582 }
583
584 // These are the signals that are blocked while a (non-VM) thread is
585 // running Java. Only the VM thread handles these signals.
586 sigset_t* os::Linux::vm_signals() {
587 assert(signal_sets_initialized, "Not initialized");
588 return &vm_sigs;
589 }
590
591 // These are signals that are blocked during cond_wait to allow debugger in
592 sigset_t* os::Linux::allowdebug_blocked_signals() {
593 assert(signal_sets_initialized, "Not initialized");
594 return &allowdebug_blocked_sigs;
595 }
596
597 void os::Linux::hotspot_sigmask(Thread* thread) {
598
599 //Save caller's signal mask before setting VM signal mask
600 sigset_t caller_sigmask;
601 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
602
603 OSThread* osthread = thread->osthread();
604 osthread->set_caller_sigmask(caller_sigmask);
605
606 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
607
608 if (!ReduceSignalUsage) {
609 if (thread->is_VM_thread()) {
610 // Only the VM thread handles BREAK_SIGNAL ...
611 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
612 } else {
613 // ... all other threads block BREAK_SIGNAL
614 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
615 }
616 }
617 }
618
619 //////////////////////////////////////////////////////////////////////////////
620 // detecting pthread library
621
622 void os::Linux::libpthread_init() {
623 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
624 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
625 // generic name for earlier versions.
626 // Define macros here so we can build HotSpot on old systems.
627 # ifndef _CS_GNU_LIBC_VERSION
628 # define _CS_GNU_LIBC_VERSION 2
629 # endif
630 # ifndef _CS_GNU_LIBPTHREAD_VERSION
631 # define _CS_GNU_LIBPTHREAD_VERSION 3
632 # endif
633
634 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
635 if (n > 0) {
636 char *str = (char *)malloc(n, mtInternal);
637 confstr(_CS_GNU_LIBC_VERSION, str, n);
638 os::Linux::set_glibc_version(str);
639 } else {
640 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
641 static char _gnu_libc_version[32];
642 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
643 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
644 os::Linux::set_glibc_version(_gnu_libc_version);
645 }
646
647 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
648 if (n > 0) {
649 char *str = (char *)malloc(n, mtInternal);
650 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
651 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
652 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
653 // is the case. LinuxThreads has a hard limit on max number of threads.
654 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
655 // On the other hand, NPTL does not have such a limit, sysconf()
656 // will return -1 and errno is not changed. Check if it is really NPTL.
657 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
658 strstr(str, "NPTL") &&
659 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
660 free(str);
661 os::Linux::set_libpthread_version("linuxthreads");
662 } else {
663 os::Linux::set_libpthread_version(str);
664 }
665 } else {
666 // glibc before 2.3.2 only has LinuxThreads.
667 os::Linux::set_libpthread_version("linuxthreads");
668 }
669
670 if (strstr(libpthread_version(), "NPTL")) {
671 os::Linux::set_is_NPTL();
672 } else {
673 os::Linux::set_is_LinuxThreads();
674 }
675
676 // LinuxThreads have two flavors: floating-stack mode, which allows variable
677 // stack size; and fixed-stack mode. NPTL is always floating-stack.
678 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
679 os::Linux::set_is_floating_stack();
680 }
681 }
682
683 /////////////////////////////////////////////////////////////////////////////
684 // thread stack
685
686 // Force Linux kernel to expand current thread stack. If "bottom" is close
687 // to the stack guard, caller should block all signals.
688 //
689 // MAP_GROWSDOWN:
690 // A special mmap() flag that is used to implement thread stacks. It tells
691 // kernel that the memory region should extend downwards when needed. This
692 // allows early versions of LinuxThreads to only mmap the first few pages
693 // when creating a new thread. Linux kernel will automatically expand thread
694 // stack as needed (on page faults).
695 //
696 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
697 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
698 // region, it's hard to tell if the fault is due to a legitimate stack
699 // access or because of reading/writing non-exist memory (e.g. buffer
700 // overrun). As a rule, if the fault happens below current stack pointer,
701 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
702 // application (see Linux kernel fault.c).
703 //
704 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
705 // stack overflow detection.
706 //
707 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
708 // not use this flag. However, the stack of initial thread is not created
709 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
710 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
711 // and then attach the thread to JVM.
712 //
713 // To get around the problem and allow stack banging on Linux, we need to
714 // manually expand thread stack after receiving the SIGSEGV.
715 //
716 // There are two ways to expand thread stack to address "bottom", we used
717 // both of them in JVM before 1.5:
718 // 1. adjust stack pointer first so that it is below "bottom", and then
719 // touch "bottom"
720 // 2. mmap() the page in question
721 //
722 // Now alternate signal stack is gone, it's harder to use 2. For instance,
723 // if current sp is already near the lower end of page 101, and we need to
724 // call mmap() to map page 100, it is possible that part of the mmap() frame
725 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
726 // That will destroy the mmap() frame and cause VM to crash.
727 //
728 // The following code works by adjusting sp first, then accessing the "bottom"
729 // page to force a page fault. Linux kernel will then automatically expand the
730 // stack mapping.
731 //
732 // _expand_stack_to() assumes its frame size is less than page size, which
733 // should always be true if the function is not inlined.
734
735 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
736 #define NOINLINE
737 #else
738 #define NOINLINE __attribute__ ((noinline))
739 #endif
740
741 static void _expand_stack_to(address bottom) NOINLINE;
742
743 static void _expand_stack_to(address bottom) {
744 address sp;
745 size_t size;
746 volatile char *p;
747
748 // Adjust bottom to point to the largest address within the same page, it
749 // gives us a one-page buffer if alloca() allocates slightly more memory.
750 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
751 bottom += os::Linux::page_size() - 1;
752
753 // sp might be slightly above current stack pointer; if that's the case, we
754 // will alloca() a little more space than necessary, which is OK. Don't use
755 // os::current_stack_pointer(), as its result can be slightly below current
756 // stack pointer, causing us to not alloca enough to reach "bottom".
757 sp = (address)&sp;
758
759 if (sp > bottom) {
760 size = sp - bottom;
761 p = (volatile char *)alloca(size);
762 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
763 p[0] = '\0';
764 }
765 }
766
767 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
768 assert(t!=NULL, "just checking");
769 assert(t->osthread()->expanding_stack(), "expand should be set");
770 assert(t->stack_base() != NULL, "stack_base was not initialized");
771
772 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
773 sigset_t mask_all, old_sigset;
774 sigfillset(&mask_all);
775 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
776 _expand_stack_to(addr);
777 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
778 return true;
779 }
780 return false;
781 }
782
783 //////////////////////////////////////////////////////////////////////////////
784 // create new thread
785
786 static address highest_vm_reserved_address();
787
788 // check if it's safe to start a new thread
789 static bool _thread_safety_check(Thread* thread) {
790 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
791 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
792 // Heap is mmap'ed at lower end of memory space. Thread stacks are
793 // allocated (MAP_FIXED) from high address space. Every thread stack
794 // occupies a fixed size slot (usually 2Mbytes, but user can change
795 // it to other values if they rebuild LinuxThreads).
796 //
797 // Problem with MAP_FIXED is that mmap() can still succeed even part of
798 // the memory region has already been mmap'ed. That means if we have too
799 // many threads and/or very large heap, eventually thread stack will
800 // collide with heap.
801 //
802 // Here we try to prevent heap/stack collision by comparing current
803 // stack bottom with the highest address that has been mmap'ed by JVM
804 // plus a safety margin for memory maps created by native code.
805 //
806 // This feature can be disabled by setting ThreadSafetyMargin to 0
807 //
808 if (ThreadSafetyMargin > 0) {
809 address stack_bottom = os::current_stack_base() - os::current_stack_size();
810
811 // not safe if our stack extends below the safety margin
812 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
813 } else {
814 return true;
815 }
816 } else {
817 // Floating stack LinuxThreads or NPTL:
818 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
819 // there's not enough space left, pthread_create() will fail. If we come
820 // here, that means enough space has been reserved for stack.
821 return true;
822 }
823 }
824
825 // Thread start routine for all newly created threads
826 static void *java_start(Thread *thread) {
827 // Try to randomize the cache line index of hot stack frames.
828 // This helps when threads of the same stack traces evict each other's
829 // cache lines. The threads can be either from the same JVM instance, or
830 // from different JVM instances. The benefit is especially true for
831 // processors with hyperthreading technology.
832 static int counter = 0;
833 int pid = os::current_process_id();
834 alloca(((pid ^ counter++) & 7) * 128);
835
836 ThreadLocalStorage::set_thread(thread);
837
838 OSThread* osthread = thread->osthread();
839 Monitor* sync = osthread->startThread_lock();
840
841 // non floating stack LinuxThreads needs extra check, see above
842 if (!_thread_safety_check(thread)) {
843 // notify parent thread
844 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
845 osthread->set_state(ZOMBIE);
846 sync->notify_all();
847 return NULL;
848 }
849
850 // thread_id is kernel thread id (similar to Solaris LWP id)
851 osthread->set_thread_id(os::Linux::gettid());
852
853 if (UseNUMA) {
854 int lgrp_id = os::numa_get_group_id();
855 if (lgrp_id != -1) {
856 thread->set_lgrp_id(lgrp_id);
857 }
858 }
859 // initialize signal mask for this thread
860 os::Linux::hotspot_sigmask(thread);
861
862 // initialize floating point control register
863 os::Linux::init_thread_fpu_state();
864
865 // handshaking with parent thread
866 {
867 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
868
869 // notify parent thread
870 osthread->set_state(INITIALIZED);
871 sync->notify_all();
872
873 // wait until os::start_thread()
874 while (osthread->get_state() == INITIALIZED) {
875 sync->wait(Mutex::_no_safepoint_check_flag);
876 }
877 }
878
879 // call one more level start routine
880 thread->run();
881
882 return 0;
883 }
884
885 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
886 assert(thread->osthread() == NULL, "caller responsible");
887
888 // Allocate the OSThread object
889 OSThread* osthread = new OSThread(NULL, NULL);
890 if (osthread == NULL) {
891 return false;
892 }
893
894 // set the correct thread state
895 osthread->set_thread_type(thr_type);
896
897 // Initial state is ALLOCATED but not INITIALIZED
898 osthread->set_state(ALLOCATED);
899
900 thread->set_osthread(osthread);
901
902 // init thread attributes
903 pthread_attr_t attr;
904 pthread_attr_init(&attr);
905 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
906
907 // stack size
908 if (os::Linux::supports_variable_stack_size()) {
909 // calculate stack size if it's not specified by caller
910 if (stack_size == 0) {
911 stack_size = os::Linux::default_stack_size(thr_type);
912
913 switch (thr_type) {
914 case os::java_thread:
915 // Java threads use ThreadStackSize which default value can be
916 // changed with the flag -Xss
917 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
918 stack_size = JavaThread::stack_size_at_create();
919 break;
920 case os::compiler_thread:
921 if (CompilerThreadStackSize > 0) {
922 stack_size = (size_t)(CompilerThreadStackSize * K);
923 break;
924 } // else fall through:
925 // use VMThreadStackSize if CompilerThreadStackSize is not defined
926 case os::vm_thread:
927 case os::pgc_thread:
928 case os::cgc_thread:
929 case os::watcher_thread:
930 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
931 break;
932 }
933 }
934
935 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
936 pthread_attr_setstacksize(&attr, stack_size);
937 } else {
938 // let pthread_create() pick the default value.
939 }
940
941 // glibc guard page
942 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
943
944 ThreadState state;
945
946 {
947 // Serialize thread creation if we are running with fixed stack LinuxThreads
948 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
949 if (lock) {
950 os::Linux::createThread_lock()->lock_without_safepoint_check();
951 }
952
953 pthread_t tid;
954 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
955
956 pthread_attr_destroy(&attr);
957
958 if (ret != 0) {
959 if (PrintMiscellaneous && (Verbose || WizardMode)) {
960 perror("pthread_create()");
961 }
962 // Need to clean up stuff we've allocated so far
963 thread->set_osthread(NULL);
964 delete osthread;
965 if (lock) os::Linux::createThread_lock()->unlock();
966 return false;
967 }
968
969 // Store pthread info into the OSThread
970 osthread->set_pthread_id(tid);
971
972 // Wait until child thread is either initialized or aborted
973 {
974 Monitor* sync_with_child = osthread->startThread_lock();
975 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
976 while ((state = osthread->get_state()) == ALLOCATED) {
977 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
978 }
979 }
980
981 if (lock) {
982 os::Linux::createThread_lock()->unlock();
983 }
984 }
985
986 // Aborted due to thread limit being reached
987 if (state == ZOMBIE) {
988 thread->set_osthread(NULL);
989 delete osthread;
990 return false;
991 }
992
993 // The thread is returned suspended (in state INITIALIZED),
994 // and is started higher up in the call chain
995 assert(state == INITIALIZED, "race condition");
996 return true;
997 }
998
999 /////////////////////////////////////////////////////////////////////////////
1000 // attach existing thread
1001
1002 // bootstrap the main thread
1003 bool os::create_main_thread(JavaThread* thread) {
1004 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1005 return create_attached_thread(thread);
1006 }
1007
1008 bool os::create_attached_thread(JavaThread* thread) {
1009 #ifdef ASSERT
1010 thread->verify_not_published();
1011 #endif
1012
1013 // Allocate the OSThread object
1014 OSThread* osthread = new OSThread(NULL, NULL);
1015
1016 if (osthread == NULL) {
1017 return false;
1018 }
1019
1020 // Store pthread info into the OSThread
1021 osthread->set_thread_id(os::Linux::gettid());
1022 osthread->set_pthread_id(::pthread_self());
1023
1024 // initialize floating point control register
1025 os::Linux::init_thread_fpu_state();
1026
1027 // Initial thread state is RUNNABLE
1028 osthread->set_state(RUNNABLE);
1029
1030 thread->set_osthread(osthread);
1031
1032 if (UseNUMA) {
1033 int lgrp_id = os::numa_get_group_id();
1034 if (lgrp_id != -1) {
1035 thread->set_lgrp_id(lgrp_id);
1036 }
1037 }
1038
1039 if (os::Linux::is_initial_thread()) {
1040 // If current thread is initial thread, its stack is mapped on demand,
1041 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1042 // the entire stack region to avoid SEGV in stack banging.
1043 // It is also useful to get around the heap-stack-gap problem on SuSE
1044 // kernel (see 4821821 for details). We first expand stack to the top
1045 // of yellow zone, then enable stack yellow zone (order is significant,
1046 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1047 // is no gap between the last two virtual memory regions.
1048
1049 JavaThread *jt = (JavaThread *)thread;
1050 address addr = jt->stack_yellow_zone_base();
1051 assert(addr != NULL, "initialization problem?");
1052 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1053
1054 osthread->set_expanding_stack();
1055 os::Linux::manually_expand_stack(jt, addr);
1056 osthread->clear_expanding_stack();
1057 }
1058
1059 // initialize signal mask for this thread
1060 // and save the caller's signal mask
1061 os::Linux::hotspot_sigmask(thread);
1062
1063 return true;
1064 }
1065
1066 void os::pd_start_thread(Thread* thread) {
1067 OSThread * osthread = thread->osthread();
1068 assert(osthread->get_state() != INITIALIZED, "just checking");
1069 Monitor* sync_with_child = osthread->startThread_lock();
1070 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1071 sync_with_child->notify();
1072 }
1073
1074 // Free Linux resources related to the OSThread
1075 void os::free_thread(OSThread* osthread) {
1076 assert(osthread != NULL, "osthread not set");
1077
1078 if (Thread::current()->osthread() == osthread) {
1079 // Restore caller's signal mask
1080 sigset_t sigmask = osthread->caller_sigmask();
1081 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1082 }
1083
1084 delete osthread;
1085 }
1086
1087 //////////////////////////////////////////////////////////////////////////////
1088 // thread local storage
1089
1090 int os::allocate_thread_local_storage() {
1091 pthread_key_t key;
1092 int rslt = pthread_key_create(&key, NULL);
1093 assert(rslt == 0, "cannot allocate thread local storage");
1094 return (int)key;
1095 }
1096
1097 // Note: This is currently not used by VM, as we don't destroy TLS key
1098 // on VM exit.
1099 void os::free_thread_local_storage(int index) {
1100 int rslt = pthread_key_delete((pthread_key_t)index);
1101 assert(rslt == 0, "invalid index");
1102 }
1103
1104 void os::thread_local_storage_at_put(int index, void* value) {
1105 int rslt = pthread_setspecific((pthread_key_t)index, value);
1106 assert(rslt == 0, "pthread_setspecific failed");
1107 }
1108
1109 extern "C" Thread* get_thread() {
1110 return ThreadLocalStorage::thread();
1111 }
1112
1113 //////////////////////////////////////////////////////////////////////////////
1114 // initial thread
1115
1116 // Check if current thread is the initial thread, similar to Solaris thr_main.
1117 bool os::Linux::is_initial_thread(void) {
1118 char dummy;
1119 // If called before init complete, thread stack bottom will be null.
1120 // Can be called if fatal error occurs before initialization.
1121 if (initial_thread_stack_bottom() == NULL) return false;
1122 assert(initial_thread_stack_bottom() != NULL &&
1123 initial_thread_stack_size() != 0,
1124 "os::init did not locate initial thread's stack region");
1125 if ((address)&dummy >= initial_thread_stack_bottom() &&
1126 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1127 return true;
1128 else return false;
1129 }
1130
1131 // Find the virtual memory area that contains addr
1132 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1133 FILE *fp = fopen("/proc/self/maps", "r");
1134 if (fp) {
1135 address low, high;
1136 while (!feof(fp)) {
1137 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1138 if (low <= addr && addr < high) {
1139 if (vma_low) *vma_low = low;
1140 if (vma_high) *vma_high = high;
1141 fclose (fp);
1142 return true;
1143 }
1144 }
1145 for (;;) {
1146 int ch = fgetc(fp);
1147 if (ch == EOF || ch == (int)'\n') break;
1148 }
1149 }
1150 fclose(fp);
1151 }
1152 return false;
1153 }
1154
1155 // Locate initial thread stack. This special handling of initial thread stack
1156 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1157 // bogus value for initial thread.
1158 void os::Linux::capture_initial_stack(size_t max_size) {
1159 // stack size is the easy part, get it from RLIMIT_STACK
1160 size_t stack_size;
1161 struct rlimit rlim;
1162 getrlimit(RLIMIT_STACK, &rlim);
1163 stack_size = rlim.rlim_cur;
1164
1165 // 6308388: a bug in ld.so will relocate its own .data section to the
1166 // lower end of primordial stack; reduce ulimit -s value a little bit
1167 // so we won't install guard page on ld.so's data section.
1168 stack_size -= 2 * page_size();
1169
1170 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1171 // 7.1, in both cases we will get 2G in return value.
1172 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1173 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1174 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1175 // in case other parts in glibc still assumes 2M max stack size.
1176 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1177 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1178 if (stack_size > 2 * K * K IA64_ONLY(*2))
1179 stack_size = 2 * K * K IA64_ONLY(*2);
1180 // Try to figure out where the stack base (top) is. This is harder.
1181 //
1182 // When an application is started, glibc saves the initial stack pointer in
1183 // a global variable "__libc_stack_end", which is then used by system
1184 // libraries. __libc_stack_end should be pretty close to stack top. The
1185 // variable is available since the very early days. However, because it is
1186 // a private interface, it could disappear in the future.
1187 //
1188 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1189 // to __libc_stack_end, it is very close to stack top, but isn't the real
1190 // stack top. Note that /proc may not exist if VM is running as a chroot
1191 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1192 // /proc/<pid>/stat could change in the future (though unlikely).
1193 //
1194 // We try __libc_stack_end first. If that doesn't work, look for
1195 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1196 // as a hint, which should work well in most cases.
1197
1198 uintptr_t stack_start;
1199
1200 // try __libc_stack_end first
1201 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1202 if (p && *p) {
1203 stack_start = *p;
1204 } else {
1205 // see if we can get the start_stack field from /proc/self/stat
1206 FILE *fp;
1207 int pid;
1208 char state;
1209 int ppid;
1210 int pgrp;
1211 int session;
1212 int nr;
1213 int tpgrp;
1214 unsigned long flags;
1215 unsigned long minflt;
1216 unsigned long cminflt;
1217 unsigned long majflt;
1218 unsigned long cmajflt;
1219 unsigned long utime;
1220 unsigned long stime;
1221 long cutime;
1222 long cstime;
1223 long prio;
1224 long nice;
1225 long junk;
1226 long it_real;
1227 uintptr_t start;
1228 uintptr_t vsize;
1229 intptr_t rss;
1230 uintptr_t rsslim;
1231 uintptr_t scodes;
1232 uintptr_t ecode;
1233 int i;
1234
1235 // Figure what the primordial thread stack base is. Code is inspired
1236 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1237 // followed by command name surrounded by parentheses, state, etc.
1238 char stat[2048];
1239 int statlen;
1240
1241 fp = fopen("/proc/self/stat", "r");
1242 if (fp) {
1243 statlen = fread(stat, 1, 2047, fp);
1244 stat[statlen] = '\0';
1245 fclose(fp);
1246
1247 // Skip pid and the command string. Note that we could be dealing with
1248 // weird command names, e.g. user could decide to rename java launcher
1249 // to "java 1.4.2 :)", then the stat file would look like
1250 // 1234 (java 1.4.2 :)) R ... ...
1251 // We don't really need to know the command string, just find the last
1252 // occurrence of ")" and then start parsing from there. See bug 4726580.
1253 char * s = strrchr(stat, ')');
1254
1255 i = 0;
1256 if (s) {
1257 // Skip blank chars
1258 do s++; while (isspace(*s));
1259
1260 #define _UFM UINTX_FORMAT
1261 #define _DFM INTX_FORMAT
1262
1263 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1264 /* 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 */
1265 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,
1266 &state, /* 3 %c */
1267 &ppid, /* 4 %d */
1268 &pgrp, /* 5 %d */
1269 &session, /* 6 %d */
1270 &nr, /* 7 %d */
1271 &tpgrp, /* 8 %d */
1272 &flags, /* 9 %lu */
1273 &minflt, /* 10 %lu */
1274 &cminflt, /* 11 %lu */
1275 &majflt, /* 12 %lu */
1276 &cmajflt, /* 13 %lu */
1277 &utime, /* 14 %lu */
1278 &stime, /* 15 %lu */
1279 &cutime, /* 16 %ld */
1280 &cstime, /* 17 %ld */
1281 &prio, /* 18 %ld */
1282 &nice, /* 19 %ld */
1283 &junk, /* 20 %ld */
1284 &it_real, /* 21 %ld */
1285 &start, /* 22 UINTX_FORMAT */
1286 &vsize, /* 23 UINTX_FORMAT */
1287 &rss, /* 24 INTX_FORMAT */
1288 &rsslim, /* 25 UINTX_FORMAT */
1289 &scodes, /* 26 UINTX_FORMAT */
1290 &ecode, /* 27 UINTX_FORMAT */
1291 &stack_start); /* 28 UINTX_FORMAT */
1292 }
1293
1294 #undef _UFM
1295 #undef _DFM
1296
1297 if (i != 28 - 2) {
1298 assert(false, "Bad conversion from /proc/self/stat");
1299 // product mode - assume we are the initial thread, good luck in the
1300 // embedded case.
1301 warning("Can't detect initial thread stack location - bad conversion");
1302 stack_start = (uintptr_t) &rlim;
1303 }
1304 } else {
1305 // For some reason we can't open /proc/self/stat (for example, running on
1306 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1307 // most cases, so don't abort:
1308 warning("Can't detect initial thread stack location - no /proc/self/stat");
1309 stack_start = (uintptr_t) &rlim;
1310 }
1311 }
1312
1313 // Now we have a pointer (stack_start) very close to the stack top, the
1314 // next thing to do is to figure out the exact location of stack top. We
1315 // can find out the virtual memory area that contains stack_start by
1316 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1317 // and its upper limit is the real stack top. (again, this would fail if
1318 // running inside chroot, because /proc may not exist.)
1319
1320 uintptr_t stack_top;
1321 address low, high;
1322 if (find_vma((address)stack_start, &low, &high)) {
1323 // success, "high" is the true stack top. (ignore "low", because initial
1324 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1325 stack_top = (uintptr_t)high;
1326 } else {
1327 // failed, likely because /proc/self/maps does not exist
1328 warning("Can't detect initial thread stack location - find_vma failed");
1329 // best effort: stack_start is normally within a few pages below the real
1330 // stack top, use it as stack top, and reduce stack size so we won't put
1331 // guard page outside stack.
1332 stack_top = stack_start;
1333 stack_size -= 16 * page_size();
1334 }
1335
1336 // stack_top could be partially down the page so align it
1337 stack_top = align_size_up(stack_top, page_size());
1338
1339 if (max_size && stack_size > max_size) {
1340 _initial_thread_stack_size = max_size;
1341 } else {
1342 _initial_thread_stack_size = stack_size;
1343 }
1344
1345 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1346 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1347 }
1348
1349 ////////////////////////////////////////////////////////////////////////////////
1350 // time support
1351
1352 // Time since start-up in seconds to a fine granularity.
1353 // Used by VMSelfDestructTimer and the MemProfiler.
1354 double os::elapsedTime() {
1355
1356 return (double)(os::elapsed_counter()) * 0.000001;
1357 }
1358
1359 jlong os::elapsed_counter() {
1360 timeval time;
1361 int status = gettimeofday(&time, NULL);
1362 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1363 }
1364
1365 jlong os::elapsed_frequency() {
1366 return (1000 * 1000);
1367 }
1368
1369 // For now, we say that linux does not support vtime. I have no idea
1370 // whether it can actually be made to (DLD, 9/13/05).
1371
1372 bool os::supports_vtime() { return false; }
1373 bool os::enable_vtime() { return false; }
1374 bool os::vtime_enabled() { return false; }
1375 double os::elapsedVTime() {
1376 // better than nothing, but not much
1377 return elapsedTime();
1378 }
1379
1380 jlong os::javaTimeMillis() {
1381 timeval time;
1382 int status = gettimeofday(&time, NULL);
1383 assert(status != -1, "linux error");
1384 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1385 }
1386
1387 #ifndef CLOCK_MONOTONIC
1388 #define CLOCK_MONOTONIC (1)
1389 #endif
1390
1391 void os::Linux::clock_init() {
1392 // we do dlopen's in this particular order due to bug in linux
1393 // dynamical loader (see 6348968) leading to crash on exit
1394 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1395 if (handle == NULL) {
1396 handle = dlopen("librt.so", RTLD_LAZY);
1397 }
1398
1399 if (handle) {
1400 int (*clock_getres_func)(clockid_t, struct timespec*) =
1401 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1402 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1403 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1404 if (clock_getres_func && clock_gettime_func) {
1405 // See if monotonic clock is supported by the kernel. Note that some
1406 // early implementations simply return kernel jiffies (updated every
1407 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1408 // for nano time (though the monotonic property is still nice to have).
1409 // It's fixed in newer kernels, however clock_getres() still returns
1410 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1411 // resolution for now. Hopefully as people move to new kernels, this
1412 // won't be a problem.
1413 struct timespec res;
1414 struct timespec tp;
1415 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1416 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1417 // yes, monotonic clock is supported
1418 _clock_gettime = clock_gettime_func;
1419 } else {
1420 // close librt if there is no monotonic clock
1421 dlclose(handle);
1422 }
1423 }
1424 }
1425 }
1426
1427 #ifndef SYS_clock_getres
1428
1429 #if defined(IA32) || defined(AMD64)
1430 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1431 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1432 #else
1433 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1434 #define sys_clock_getres(x,y) -1
1435 #endif
1436
1437 #else
1438 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1439 #endif
1440
1441 void os::Linux::fast_thread_clock_init() {
1442 if (!UseLinuxPosixThreadCPUClocks) {
1443 return;
1444 }
1445 clockid_t clockid;
1446 struct timespec tp;
1447 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1448 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1449
1450 // Switch to using fast clocks for thread cpu time if
1451 // the sys_clock_getres() returns 0 error code.
1452 // Note, that some kernels may support the current thread
1453 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1454 // returned by the pthread_getcpuclockid().
1455 // If the fast Posix clocks are supported then the sys_clock_getres()
1456 // must return at least tp.tv_sec == 0 which means a resolution
1457 // better than 1 sec. This is extra check for reliability.
1458
1459 if(pthread_getcpuclockid_func &&
1460 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1461 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1462
1463 _supports_fast_thread_cpu_time = true;
1464 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1465 }
1466 }
1467
1468 jlong os::javaTimeNanos() {
1469 if (Linux::supports_monotonic_clock()) {
1470 struct timespec tp;
1471 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1472 assert(status == 0, "gettime error");
1473 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1474 return result;
1475 } else {
1476 timeval time;
1477 int status = gettimeofday(&time, NULL);
1478 assert(status != -1, "linux error");
1479 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1480 return 1000 * usecs;
1481 }
1482 }
1483
1484 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1485 if (Linux::supports_monotonic_clock()) {
1486 info_ptr->max_value = ALL_64_BITS;
1487
1488 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1489 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1490 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1491 } else {
1492 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1493 info_ptr->max_value = ALL_64_BITS;
1494
1495 // gettimeofday is a real time clock so it skips
1496 info_ptr->may_skip_backward = true;
1497 info_ptr->may_skip_forward = true;
1498 }
1499
1500 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1501 }
1502
1503 // Return the real, user, and system times in seconds from an
1504 // arbitrary fixed point in the past.
1505 bool os::getTimesSecs(double* process_real_time,
1506 double* process_user_time,
1507 double* process_system_time) {
1508 struct tms ticks;
1509 clock_t real_ticks = times(&ticks);
1510
1511 if (real_ticks == (clock_t) (-1)) {
1512 return false;
1513 } else {
1514 double ticks_per_second = (double) clock_tics_per_sec;
1515 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1516 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1517 *process_real_time = ((double) real_ticks) / ticks_per_second;
1518
1519 return true;
1520 }
1521 }
1522
1523
1524 char * os::local_time_string(char *buf, size_t buflen) {
1525 struct tm t;
1526 time_t long_time;
1527 time(&long_time);
1528 localtime_r(&long_time, &t);
1529 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1530 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1531 t.tm_hour, t.tm_min, t.tm_sec);
1532 return buf;
1533 }
1534
1535 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1536 return localtime_r(clock, res);
1537 }
1538
1539 ////////////////////////////////////////////////////////////////////////////////
1540 // runtime exit support
1541
1542 // Note: os::shutdown() might be called very early during initialization, or
1543 // called from signal handler. Before adding something to os::shutdown(), make
1544 // sure it is async-safe and can handle partially initialized VM.
1545 void os::shutdown() {
1546
1547 // allow PerfMemory to attempt cleanup of any persistent resources
1548 perfMemory_exit();
1549
1550 // needs to remove object in file system
1551 AttachListener::abort();
1552
1553 // flush buffered output, finish log files
1554 ostream_abort();
1555
1556 // Check for abort hook
1557 abort_hook_t abort_hook = Arguments::abort_hook();
1558 if (abort_hook != NULL) {
1559 abort_hook();
1560 }
1561
1562 }
1563
1564 // Note: os::abort() might be called very early during initialization, or
1565 // called from signal handler. Before adding something to os::abort(), make
1566 // sure it is async-safe and can handle partially initialized VM.
1567 void os::abort(bool dump_core) {
1568 os::shutdown();
1569 if (dump_core) {
1570 #ifndef PRODUCT
1571 fdStream out(defaultStream::output_fd());
1572 out.print_raw("Current thread is ");
1573 char buf[16];
1574 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1575 out.print_raw_cr(buf);
1576 out.print_raw_cr("Dumping core ...");
1577 #endif
1578 ::abort(); // dump core
1579 }
1580
1581 ::exit(1);
1582 }
1583
1584 // Die immediately, no exit hook, no abort hook, no cleanup.
1585 void os::die() {
1586 // _exit() on LinuxThreads only kills current thread
1587 ::abort();
1588 }
1589
1590 // unused on linux for now.
1591 void os::set_error_file(const char *logfile) {}
1592
1593
1594 // This method is a copy of JDK's sysGetLastErrorString
1595 // from src/solaris/hpi/src/system_md.c
1596
1597 size_t os::lasterror(char *buf, size_t len) {
1598
1599 if (errno == 0) return 0;
1600
1601 const char *s = ::strerror(errno);
1602 size_t n = ::strlen(s);
1603 if (n >= len) {
1604 n = len - 1;
1605 }
1606 ::strncpy(buf, s, n);
1607 buf[n] = '\0';
1608 return n;
1609 }
1610
1611 intx os::current_thread_id() { return (intx)pthread_self(); }
1612 int os::current_process_id() {
1613
1614 // Under the old linux thread library, linux gives each thread
1615 // its own process id. Because of this each thread will return
1616 // a different pid if this method were to return the result
1617 // of getpid(2). Linux provides no api that returns the pid
1618 // of the launcher thread for the vm. This implementation
1619 // returns a unique pid, the pid of the launcher thread
1620 // that starts the vm 'process'.
1621
1622 // Under the NPTL, getpid() returns the same pid as the
1623 // launcher thread rather than a unique pid per thread.
1624 // Use gettid() if you want the old pre NPTL behaviour.
1625
1626 // if you are looking for the result of a call to getpid() that
1627 // returns a unique pid for the calling thread, then look at the
1628 // OSThread::thread_id() method in osThread_linux.hpp file
1629
1630 return (int)(_initial_pid ? _initial_pid : getpid());
1631 }
1632
1633 // DLL functions
1634
1635 const char* os::dll_file_extension() { return ".so"; }
1636
1637 // This must be hard coded because it's the system's temporary
1638 // directory not the java application's temp directory, ala java.io.tmpdir.
1639 const char* os::get_temp_directory() { return "/tmp"; }
1640
1641 static bool file_exists(const char* filename) {
1642 struct stat statbuf;
1643 if (filename == NULL || strlen(filename) == 0) {
1644 return false;
1645 }
1646 return os::stat(filename, &statbuf) == 0;
1647 }
1648
1649 bool os::dll_build_name(char* buffer, size_t buflen,
1650 const char* pname, const char* fname) {
1651 bool retval = false;
1652 // Copied from libhpi
1653 const size_t pnamelen = pname ? strlen(pname) : 0;
1654
1655 // Return error on buffer overflow.
1656 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1657 return retval;
1658 }
1659
1660 if (pnamelen == 0) {
1661 snprintf(buffer, buflen, "lib%s.so", fname);
1662 retval = true;
1663 } else if (strchr(pname, *os::path_separator()) != NULL) {
1664 int n;
1665 char** pelements = split_path(pname, &n);
1666 for (int i = 0 ; i < n ; i++) {
1667 // Really shouldn't be NULL, but check can't hurt
1668 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1669 continue; // skip the empty path values
1670 }
1671 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1672 if (file_exists(buffer)) {
1673 retval = true;
1674 break;
1675 }
1676 }
1677 // release the storage
1678 for (int i = 0 ; i < n ; i++) {
1679 if (pelements[i] != NULL) {
1680 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1681 }
1682 }
1683 if (pelements != NULL) {
1684 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1685 }
1686 } else {
1687 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1688 retval = true;
1689 }
1690 return retval;
1691 }
1692
1693 const char* os::get_current_directory(char *buf, int buflen) {
1694 return getcwd(buf, buflen);
1695 }
1696
1697 // check if addr is inside libjvm.so
1698 bool os::address_is_in_vm(address addr) {
1699 static address libjvm_base_addr;
1700 Dl_info dlinfo;
1701
1702 if (libjvm_base_addr == NULL) {
1703 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1704 libjvm_base_addr = (address)dlinfo.dli_fbase;
1705 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1706 }
1707
1708 if (dladdr((void *)addr, &dlinfo)) {
1709 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1710 }
1711
1712 return false;
1713 }
1714
1715 bool os::dll_address_to_function_name(address addr, char *buf,
1716 int buflen, int *offset) {
1717 Dl_info dlinfo;
1718
1719 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1720 if (buf != NULL) {
1721 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1722 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1723 }
1724 }
1725 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1726 return true;
1727 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1728 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1729 buf, buflen, offset, dlinfo.dli_fname)) {
1730 return true;
1731 }
1732 }
1733
1734 if (buf != NULL) buf[0] = '\0';
1735 if (offset != NULL) *offset = -1;
1736 return false;
1737 }
1738
1739 struct _address_to_library_name {
1740 address addr; // input : memory address
1741 size_t buflen; // size of fname
1742 char* fname; // output: library name
1743 address base; // library base addr
1744 };
1745
1746 static int address_to_library_name_callback(struct dl_phdr_info *info,
1747 size_t size, void *data) {
1748 int i;
1749 bool found = false;
1750 address libbase = NULL;
1751 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1752
1753 // iterate through all loadable segments
1754 for (i = 0; i < info->dlpi_phnum; i++) {
1755 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1756 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1757 // base address of a library is the lowest address of its loaded
1758 // segments.
1759 if (libbase == NULL || libbase > segbase) {
1760 libbase = segbase;
1761 }
1762 // see if 'addr' is within current segment
1763 if (segbase <= d->addr &&
1764 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1765 found = true;
1766 }
1767 }
1768 }
1769
1770 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1771 // so dll_address_to_library_name() can fall through to use dladdr() which
1772 // can figure out executable name from argv[0].
1773 if (found && info->dlpi_name && info->dlpi_name[0]) {
1774 d->base = libbase;
1775 if (d->fname) {
1776 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1777 }
1778 return 1;
1779 }
1780 return 0;
1781 }
1782
1783 bool os::dll_address_to_library_name(address addr, char* buf,
1784 int buflen, int* offset) {
1785 Dl_info dlinfo;
1786 struct _address_to_library_name data;
1787
1788 // There is a bug in old glibc dladdr() implementation that it could resolve
1789 // to wrong library name if the .so file has a base address != NULL. Here
1790 // we iterate through the program headers of all loaded libraries to find
1791 // out which library 'addr' really belongs to. This workaround can be
1792 // removed once the minimum requirement for glibc is moved to 2.3.x.
1793 data.addr = addr;
1794 data.fname = buf;
1795 data.buflen = buflen;
1796 data.base = NULL;
1797 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1798
1799 if (rslt) {
1800 // buf already contains library name
1801 if (offset) *offset = addr - data.base;
1802 return true;
1803 } else if (dladdr((void*)addr, &dlinfo)){
1804 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1805 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1806 return true;
1807 } else {
1808 if (buf) buf[0] = '\0';
1809 if (offset) *offset = -1;
1810 return false;
1811 }
1812 }
1813
1814 // Loads .dll/.so and
1815 // in case of error it checks if .dll/.so was built for the
1816 // same architecture as Hotspot is running on
1817
1818 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1819 {
1820 void * result= ::dlopen(filename, RTLD_LAZY);
1821 if (result != NULL) {
1822 // Successful loading
1823 return result;
1824 }
1825
1826 Elf32_Ehdr elf_head;
1827
1828 // Read system error message into ebuf
1829 // It may or may not be overwritten below
1830 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1831 ebuf[ebuflen-1]='\0';
1832 int diag_msg_max_length=ebuflen-strlen(ebuf);
1833 char* diag_msg_buf=ebuf+strlen(ebuf);
1834
1835 if (diag_msg_max_length==0) {
1836 // No more space in ebuf for additional diagnostics message
1837 return NULL;
1838 }
1839
1840
1841 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1842
1843 if (file_descriptor < 0) {
1844 // Can't open library, report dlerror() message
1845 return NULL;
1846 }
1847
1848 bool failed_to_read_elf_head=
1849 (sizeof(elf_head)!=
1850 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1851
1852 ::close(file_descriptor);
1853 if (failed_to_read_elf_head) {
1854 // file i/o error - report dlerror() msg
1855 return NULL;
1856 }
1857
1858 typedef struct {
1859 Elf32_Half code; // Actual value as defined in elf.h
1860 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1861 char elf_class; // 32 or 64 bit
1862 char endianess; // MSB or LSB
1863 char* name; // String representation
1864 } arch_t;
1865
1866 #ifndef EM_486
1867 #define EM_486 6 /* Intel 80486 */
1868 #endif
1869
1870 static const arch_t arch_array[]={
1871 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1872 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1873 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1874 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1875 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1876 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1877 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1878 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1879 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1880 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1881 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1882 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1883 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1884 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1885 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1886 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1887 };
1888
1889 #if (defined IA32)
1890 static Elf32_Half running_arch_code=EM_386;
1891 #elif (defined AMD64)
1892 static Elf32_Half running_arch_code=EM_X86_64;
1893 #elif (defined IA64)
1894 static Elf32_Half running_arch_code=EM_IA_64;
1895 #elif (defined __sparc) && (defined _LP64)
1896 static Elf32_Half running_arch_code=EM_SPARCV9;
1897 #elif (defined __sparc) && (!defined _LP64)
1898 static Elf32_Half running_arch_code=EM_SPARC;
1899 #elif (defined __powerpc64__)
1900 static Elf32_Half running_arch_code=EM_PPC64;
1901 #elif (defined __powerpc__)
1902 static Elf32_Half running_arch_code=EM_PPC;
1903 #elif (defined ARM)
1904 static Elf32_Half running_arch_code=EM_ARM;
1905 #elif (defined S390)
1906 static Elf32_Half running_arch_code=EM_S390;
1907 #elif (defined ALPHA)
1908 static Elf32_Half running_arch_code=EM_ALPHA;
1909 #elif (defined MIPSEL)
1910 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1911 #elif (defined PARISC)
1912 static Elf32_Half running_arch_code=EM_PARISC;
1913 #elif (defined MIPS)
1914 static Elf32_Half running_arch_code=EM_MIPS;
1915 #elif (defined M68K)
1916 static Elf32_Half running_arch_code=EM_68K;
1917 #else
1918 #error Method os::dll_load requires that one of following is defined:\
1919 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1920 #endif
1921
1922 // Identify compatability class for VM's architecture and library's architecture
1923 // Obtain string descriptions for architectures
1924
1925 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1926 int running_arch_index=-1;
1927
1928 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1929 if (running_arch_code == arch_array[i].code) {
1930 running_arch_index = i;
1931 }
1932 if (lib_arch.code == arch_array[i].code) {
1933 lib_arch.compat_class = arch_array[i].compat_class;
1934 lib_arch.name = arch_array[i].name;
1935 }
1936 }
1937
1938 assert(running_arch_index != -1,
1939 "Didn't find running architecture code (running_arch_code) in arch_array");
1940 if (running_arch_index == -1) {
1941 // Even though running architecture detection failed
1942 // we may still continue with reporting dlerror() message
1943 return NULL;
1944 }
1945
1946 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1947 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1948 return NULL;
1949 }
1950
1951 #ifndef S390
1952 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1953 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1954 return NULL;
1955 }
1956 #endif // !S390
1957
1958 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1959 if ( lib_arch.name!=NULL ) {
1960 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1961 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1962 lib_arch.name, arch_array[running_arch_index].name);
1963 } else {
1964 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1965 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1966 lib_arch.code,
1967 arch_array[running_arch_index].name);
1968 }
1969 }
1970
1971 return NULL;
1972 }
1973
1974 /*
1975 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1976 * chances are you might want to run the generated bits against glibc-2.0
1977 * libdl.so, so always use locking for any version of glibc.
1978 */
1979 void* os::dll_lookup(void* handle, const char* name) {
1980 pthread_mutex_lock(&dl_mutex);
1981 void* res = dlsym(handle, name);
1982 pthread_mutex_unlock(&dl_mutex);
1983 return res;
1984 }
1985
1986
1987 static bool _print_ascii_file(const char* filename, outputStream* st) {
1988 int fd = ::open(filename, O_RDONLY);
1989 if (fd == -1) {
1990 return false;
1991 }
1992
1993 char buf[32];
1994 int bytes;
1995 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1996 st->print_raw(buf, bytes);
1997 }
1998
1999 ::close(fd);
2000
2001 return true;
2002 }
2003
2004 void os::print_dll_info(outputStream *st) {
2005 st->print_cr("Dynamic libraries:");
2006
2007 char fname[32];
2008 pid_t pid = os::Linux::gettid();
2009
2010 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2011
2012 if (!_print_ascii_file(fname, st)) {
2013 st->print("Can not get library information for pid = %d\n", pid);
2014 }
2015 }
2016
2017 void os::print_os_info_brief(outputStream* st) {
2018 os::Linux::print_distro_info(st);
2019
2020 os::Posix::print_uname_info(st);
2021
2022 os::Linux::print_libversion_info(st);
2023
2024 }
2025
2026 void os::print_os_info(outputStream* st) {
2027 st->print("OS:");
2028
2029 os::Linux::print_distro_info(st);
2030
2031 os::Posix::print_uname_info(st);
2032
2033 // Print warning if unsafe chroot environment detected
2034 if (unsafe_chroot_detected) {
2035 st->print("WARNING!! ");
2036 st->print_cr(unstable_chroot_error);
2037 }
2038
2039 os::Linux::print_libversion_info(st);
2040
2041 os::Posix::print_rlimit_info(st);
2042
2043 os::Posix::print_load_average(st);
2044
2045 os::Linux::print_full_memory_info(st);
2046 }
2047
2048 // Try to identify popular distros.
2049 // Most Linux distributions have /etc/XXX-release file, which contains
2050 // the OS version string. Some have more than one /etc/XXX-release file
2051 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2052 // so the order is important.
2053 void os::Linux::print_distro_info(outputStream* st) {
2054 if (!_print_ascii_file("/etc/mandrake-release", st) &&
2055 !_print_ascii_file("/etc/sun-release", st) &&
2056 !_print_ascii_file("/etc/redhat-release", st) &&
2057 !_print_ascii_file("/etc/SuSE-release", st) &&
2058 !_print_ascii_file("/etc/turbolinux-release", st) &&
2059 !_print_ascii_file("/etc/gentoo-release", st) &&
2060 !_print_ascii_file("/etc/debian_version", st) &&
2061 !_print_ascii_file("/etc/ltib-release", st) &&
2062 !_print_ascii_file("/etc/angstrom-version", st)) {
2063 st->print("Linux");
2064 }
2065 st->cr();
2066 }
2067
2068 void os::Linux::print_libversion_info(outputStream* st) {
2069 // libc, pthread
2070 st->print("libc:");
2071 st->print(os::Linux::glibc_version()); st->print(" ");
2072 st->print(os::Linux::libpthread_version()); st->print(" ");
2073 if (os::Linux::is_LinuxThreads()) {
2074 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2075 }
2076 st->cr();
2077 }
2078
2079 void os::Linux::print_full_memory_info(outputStream* st) {
2080 st->print("\n/proc/meminfo:\n");
2081 _print_ascii_file("/proc/meminfo", st);
2082 st->cr();
2083 }
2084
2085 void os::print_memory_info(outputStream* st) {
2086
2087 st->print("Memory:");
2088 st->print(" %dk page", os::vm_page_size()>>10);
2089
2090 // values in struct sysinfo are "unsigned long"
2091 struct sysinfo si;
2092 sysinfo(&si);
2093
2094 st->print(", physical " UINT64_FORMAT "k",
2095 os::physical_memory() >> 10);
2096 st->print("(" UINT64_FORMAT "k free)",
2097 os::available_memory() >> 10);
2098 st->print(", swap " UINT64_FORMAT "k",
2099 ((jlong)si.totalswap * si.mem_unit) >> 10);
2100 st->print("(" UINT64_FORMAT "k free)",
2101 ((jlong)si.freeswap * si.mem_unit) >> 10);
2102 st->cr();
2103 }
2104
2105 void os::pd_print_cpu_info(outputStream* st) {
2106 st->print("\n/proc/cpuinfo:\n");
2107 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2108 st->print(" <Not Available>");
2109 }
2110 st->cr();
2111 }
2112
2113 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2114 // but they're the same for all the linux arch that we support
2115 // and they're the same for solaris but there's no common place to put this.
2116 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2117 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2118 "ILL_COPROC", "ILL_BADSTK" };
2119
2120 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2121 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2122 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2123
2124 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2125
2126 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2127
2128 void os::print_siginfo(outputStream* st, void* siginfo) {
2129 st->print("siginfo:");
2130
2131 const int buflen = 100;
2132 char buf[buflen];
2133 siginfo_t *si = (siginfo_t*)siginfo;
2134 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2135 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2136 st->print("si_errno=%s", buf);
2137 } else {
2138 st->print("si_errno=%d", si->si_errno);
2139 }
2140 const int c = si->si_code;
2141 assert(c > 0, "unexpected si_code");
2142 switch (si->si_signo) {
2143 case SIGILL:
2144 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2145 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2146 break;
2147 case SIGFPE:
2148 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2149 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2150 break;
2151 case SIGSEGV:
2152 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2153 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2154 break;
2155 case SIGBUS:
2156 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2157 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2158 break;
2159 default:
2160 st->print(", si_code=%d", si->si_code);
2161 // no si_addr
2162 }
2163
2164 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2165 UseSharedSpaces) {
2166 FileMapInfo* mapinfo = FileMapInfo::current_info();
2167 if (mapinfo->is_in_shared_space(si->si_addr)) {
2168 st->print("\n\nError accessing class data sharing archive." \
2169 " Mapped file inaccessible during execution, " \
2170 " possible disk/network problem.");
2171 }
2172 }
2173 st->cr();
2174 }
2175
2176
2177 static void print_signal_handler(outputStream* st, int sig,
2178 char* buf, size_t buflen);
2179
2180 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2181 st->print_cr("Signal Handlers:");
2182 print_signal_handler(st, SIGSEGV, buf, buflen);
2183 print_signal_handler(st, SIGBUS , buf, buflen);
2184 print_signal_handler(st, SIGFPE , buf, buflen);
2185 print_signal_handler(st, SIGPIPE, buf, buflen);
2186 print_signal_handler(st, SIGXFSZ, buf, buflen);
2187 print_signal_handler(st, SIGILL , buf, buflen);
2188 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2189 print_signal_handler(st, SR_signum, buf, buflen);
2190 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2191 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2192 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2193 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2194 }
2195
2196 static char saved_jvm_path[MAXPATHLEN] = {0};
2197
2198 // Find the full path to the current module, libjvm.so
2199 void os::jvm_path(char *buf, jint buflen) {
2200 // Error checking.
2201 if (buflen < MAXPATHLEN) {
2202 assert(false, "must use a large-enough buffer");
2203 buf[0] = '\0';
2204 return;
2205 }
2206 // Lazy resolve the path to current module.
2207 if (saved_jvm_path[0] != 0) {
2208 strcpy(buf, saved_jvm_path);
2209 return;
2210 }
2211
2212 char dli_fname[MAXPATHLEN];
2213 bool ret = dll_address_to_library_name(
2214 CAST_FROM_FN_PTR(address, os::jvm_path),
2215 dli_fname, sizeof(dli_fname), NULL);
2216 assert(ret != 0, "cannot locate libjvm");
2217 char *rp = realpath(dli_fname, buf);
2218 if (rp == NULL)
2219 return;
2220
2221 if (Arguments::created_by_gamma_launcher()) {
2222 // Support for the gamma launcher. Typical value for buf is
2223 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2224 // the right place in the string, then assume we are installed in a JDK and
2225 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2226 // up the path so it looks like libjvm.so is installed there (append a
2227 // fake suffix hotspot/libjvm.so).
2228 const char *p = buf + strlen(buf) - 1;
2229 for (int count = 0; p > buf && count < 5; ++count) {
2230 for (--p; p > buf && *p != '/'; --p)
2231 /* empty */ ;
2232 }
2233
2234 if (strncmp(p, "/jre/lib/", 9) != 0) {
2235 // Look for JAVA_HOME in the environment.
2236 char* java_home_var = ::getenv("JAVA_HOME");
2237 if (java_home_var != NULL && java_home_var[0] != 0) {
2238 char* jrelib_p;
2239 int len;
2240
2241 // Check the current module name "libjvm.so".
2242 p = strrchr(buf, '/');
2243 assert(strstr(p, "/libjvm") == p, "invalid library name");
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.so"
2260 len = strlen(buf);
2261 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2262 } else {
2263 // Go back to path of .so
2264 rp = realpath(dli_fname, buf);
2265 if (rp == NULL)
2266 return;
2267 }
2268 }
2269 }
2270 }
2271
2272 strcpy(saved_jvm_path, buf);
2273 }
2274
2275 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2276 // no prefix required, not even "_"
2277 }
2278
2279 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2280 // no suffix required
2281 }
2282
2283 ////////////////////////////////////////////////////////////////////////////////
2284 // sun.misc.Signal support
2285
2286 static volatile jint sigint_count = 0;
2287
2288 static void
2289 UserHandler(int sig, void *siginfo, void *context) {
2290 // 4511530 - sem_post is serialized and handled by the manager thread. When
2291 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2292 // don't want to flood the manager thread with sem_post requests.
2293 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2294 return;
2295
2296 // Ctrl-C is pressed during error reporting, likely because the error
2297 // handler fails to abort. Let VM die immediately.
2298 if (sig == SIGINT && is_error_reported()) {
2299 os::die();
2300 }
2301
2302 os::signal_notify(sig);
2303 }
2304
2305 void* os::user_handler() {
2306 return CAST_FROM_FN_PTR(void*, UserHandler);
2307 }
2308
2309 extern "C" {
2310 typedef void (*sa_handler_t)(int);
2311 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2312 }
2313
2314 void* os::signal(int signal_number, void* handler) {
2315 struct sigaction sigAct, oldSigAct;
2316
2317 sigfillset(&(sigAct.sa_mask));
2318 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2319 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2320
2321 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2322 // -1 means registration failed
2323 return (void *)-1;
2324 }
2325
2326 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2327 }
2328
2329 void os::signal_raise(int signal_number) {
2330 ::raise(signal_number);
2331 }
2332
2333 /*
2334 * The following code is moved from os.cpp for making this
2335 * code platform specific, which it is by its very nature.
2336 */
2337
2338 // Will be modified when max signal is changed to be dynamic
2339 int os::sigexitnum_pd() {
2340 return NSIG;
2341 }
2342
2343 // a counter for each possible signal value
2344 static volatile jint pending_signals[NSIG+1] = { 0 };
2345
2346 // Linux(POSIX) specific hand shaking semaphore.
2347 static sem_t sig_sem;
2348
2349 void os::signal_init_pd() {
2350 // Initialize signal structures
2351 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2352
2353 // Initialize signal semaphore
2354 ::sem_init(&sig_sem, 0, 0);
2355 }
2356
2357 void os::signal_notify(int sig) {
2358 Atomic::inc(&pending_signals[sig]);
2359 ::sem_post(&sig_sem);
2360 }
2361
2362 static int check_pending_signals(bool wait) {
2363 Atomic::store(0, &sigint_count);
2364 for (;;) {
2365 for (int i = 0; i < NSIG + 1; i++) {
2366 jint n = pending_signals[i];
2367 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2368 return i;
2369 }
2370 }
2371 if (!wait) {
2372 return -1;
2373 }
2374 JavaThread *thread = JavaThread::current();
2375 ThreadBlockInVM tbivm(thread);
2376
2377 bool threadIsSuspended;
2378 do {
2379 thread->set_suspend_equivalent();
2380 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2381 ::sem_wait(&sig_sem);
2382
2383 // were we externally suspended while we were waiting?
2384 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2385 if (threadIsSuspended) {
2386 //
2387 // The semaphore has been incremented, but while we were waiting
2388 // another thread suspended us. We don't want to continue running
2389 // while suspended because that would surprise the thread that
2390 // suspended us.
2391 //
2392 ::sem_post(&sig_sem);
2393
2394 thread->java_suspend_self();
2395 }
2396 } while (threadIsSuspended);
2397 }
2398 }
2399
2400 int os::signal_lookup() {
2401 return check_pending_signals(false);
2402 }
2403
2404 int os::signal_wait() {
2405 return check_pending_signals(true);
2406 }
2407
2408 ////////////////////////////////////////////////////////////////////////////////
2409 // Virtual Memory
2410
2411 int os::vm_page_size() {
2412 // Seems redundant as all get out
2413 assert(os::Linux::page_size() != -1, "must call os::init");
2414 return os::Linux::page_size();
2415 }
2416
2417 // Solaris allocates memory by pages.
2418 int os::vm_allocation_granularity() {
2419 assert(os::Linux::page_size() != -1, "must call os::init");
2420 return os::Linux::page_size();
2421 }
2422
2423 // Rationale behind this function:
2424 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2425 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2426 // samples for JITted code. Here we create private executable mapping over the code cache
2427 // and then we can use standard (well, almost, as mapping can change) way to provide
2428 // info for the reporting script by storing timestamp and location of symbol
2429 void linux_wrap_code(char* base, size_t size) {
2430 static volatile jint cnt = 0;
2431
2432 if (!UseOprofile) {
2433 return;
2434 }
2435
2436 char buf[PATH_MAX+1];
2437 int num = Atomic::add(1, &cnt);
2438
2439 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2440 os::get_temp_directory(), os::current_process_id(), num);
2441 unlink(buf);
2442
2443 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2444
2445 if (fd != -1) {
2446 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2447 if (rv != (off_t)-1) {
2448 if (::write(fd, "", 1) == 1) {
2449 mmap(base, size,
2450 PROT_READ|PROT_WRITE|PROT_EXEC,
2451 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2452 }
2453 }
2454 ::close(fd);
2455 unlink(buf);
2456 }
2457 }
2458
2459 // NOTE: Linux kernel does not really reserve the pages for us.
2460 // All it does is to check if there are enough free pages
2461 // left at the time of mmap(). This could be a potential
2462 // problem.
2463 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2464 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2465 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2466 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2467 if (res != (uintptr_t) MAP_FAILED) {
2468 if (UseNUMAInterleaving) {
2469 numa_make_global(addr, size);
2470 }
2471 return true;
2472 }
2473 return false;
2474 }
2475
2476 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2477 #ifndef MAP_HUGETLB
2478 #define MAP_HUGETLB 0x40000
2479 #endif
2480
2481 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2482 #ifndef MADV_HUGEPAGE
2483 #define MADV_HUGEPAGE 14
2484 #endif
2485
2486 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2487 bool exec) {
2488 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2489 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2490 uintptr_t res =
2491 (uintptr_t) ::mmap(addr, size, prot,
2492 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2493 -1, 0);
2494 if (res != (uintptr_t) MAP_FAILED) {
2495 if (UseNUMAInterleaving) {
2496 numa_make_global(addr, size);
2497 }
2498 return true;
2499 }
2500 // Fall through and try to use small pages
2501 }
2502
2503 if (commit_memory(addr, size, exec)) {
2504 realign_memory(addr, size, alignment_hint);
2505 return true;
2506 }
2507 return false;
2508 }
2509
2510 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2511 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2512 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2513 // be supported or the memory may already be backed by huge pages.
2514 ::madvise(addr, bytes, MADV_HUGEPAGE);
2515 }
2516 }
2517
2518 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2519 // This method works by doing an mmap over an existing mmaping and effectively discarding
2520 // the existing pages. However it won't work for SHM-based large pages that cannot be
2521 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2522 // small pages on top of the SHM segment. This method always works for small pages, so we
2523 // allow that in any case.
2524 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2525 commit_memory(addr, bytes, alignment_hint, false);
2526 }
2527 }
2528
2529 void os::numa_make_global(char *addr, size_t bytes) {
2530 Linux::numa_interleave_memory(addr, bytes);
2531 }
2532
2533 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2534 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2535 }
2536
2537 bool os::numa_topology_changed() { return false; }
2538
2539 size_t os::numa_get_groups_num() {
2540 int max_node = Linux::numa_max_node();
2541 return max_node > 0 ? max_node + 1 : 1;
2542 }
2543
2544 int os::numa_get_group_id() {
2545 int cpu_id = Linux::sched_getcpu();
2546 if (cpu_id != -1) {
2547 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2548 if (lgrp_id != -1) {
2549 return lgrp_id;
2550 }
2551 }
2552 return 0;
2553 }
2554
2555 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2556 for (size_t i = 0; i < size; i++) {
2557 ids[i] = i;
2558 }
2559 return size;
2560 }
2561
2562 bool os::get_page_info(char *start, page_info* info) {
2563 return false;
2564 }
2565
2566 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2567 return end;
2568 }
2569
2570
2571 int os::Linux::sched_getcpu_syscall(void) {
2572 unsigned int cpu;
2573 int retval = -1;
2574
2575 #if defined(IA32)
2576 # ifndef SYS_getcpu
2577 # define SYS_getcpu 318
2578 # endif
2579 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2580 #elif defined(AMD64)
2581 // Unfortunately we have to bring all these macros here from vsyscall.h
2582 // to be able to compile on old linuxes.
2583 # define __NR_vgetcpu 2
2584 # define VSYSCALL_START (-10UL << 20)
2585 # define VSYSCALL_SIZE 1024
2586 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2587 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2588 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2589 retval = vgetcpu(&cpu, NULL, NULL);
2590 #endif
2591
2592 return (retval == -1) ? retval : cpu;
2593 }
2594
2595 // Something to do with the numa-aware allocator needs these symbols
2596 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2597 extern "C" JNIEXPORT void numa_error(char *where) { }
2598 extern "C" JNIEXPORT int fork1() { return fork(); }
2599
2600
2601 // If we are running with libnuma version > 2, then we should
2602 // be trying to use symbols with versions 1.1
2603 // If we are running with earlier version, which did not have symbol versions,
2604 // we should use the base version.
2605 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2606 void *f = dlvsym(handle, name, "libnuma_1.1");
2607 if (f == NULL) {
2608 f = dlsym(handle, name);
2609 }
2610 return f;
2611 }
2612
2613 bool os::Linux::libnuma_init() {
2614 // sched_getcpu() should be in libc.
2615 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2616 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2617
2618 // If it's not, try a direct syscall.
2619 if (sched_getcpu() == -1)
2620 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2621
2622 if (sched_getcpu() != -1) { // Does it work?
2623 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2624 if (handle != NULL) {
2625 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2626 libnuma_dlsym(handle, "numa_node_to_cpus")));
2627 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2628 libnuma_dlsym(handle, "numa_max_node")));
2629 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2630 libnuma_dlsym(handle, "numa_available")));
2631 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2632 libnuma_dlsym(handle, "numa_tonode_memory")));
2633 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2634 libnuma_dlsym(handle, "numa_interleave_memory")));
2635
2636
2637 if (numa_available() != -1) {
2638 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2639 // Create a cpu -> node mapping
2640 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2641 rebuild_cpu_to_node_map();
2642 return true;
2643 }
2644 }
2645 }
2646 return false;
2647 }
2648
2649 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2650 // The table is later used in get_node_by_cpu().
2651 void os::Linux::rebuild_cpu_to_node_map() {
2652 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2653 // in libnuma (possible values are starting from 16,
2654 // and continuing up with every other power of 2, but less
2655 // than the maximum number of CPUs supported by kernel), and
2656 // is a subject to change (in libnuma version 2 the requirements
2657 // are more reasonable) we'll just hardcode the number they use
2658 // in the library.
2659 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2660
2661 size_t cpu_num = os::active_processor_count();
2662 size_t cpu_map_size = NCPUS / BitsPerCLong;
2663 size_t cpu_map_valid_size =
2664 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2665
2666 cpu_to_node()->clear();
2667 cpu_to_node()->at_grow(cpu_num - 1);
2668 size_t node_num = numa_get_groups_num();
2669
2670 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2671 for (size_t i = 0; i < node_num; i++) {
2672 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2673 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2674 if (cpu_map[j] != 0) {
2675 for (size_t k = 0; k < BitsPerCLong; k++) {
2676 if (cpu_map[j] & (1UL << k)) {
2677 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2678 }
2679 }
2680 }
2681 }
2682 }
2683 }
2684 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2685 }
2686
2687 int os::Linux::get_node_by_cpu(int cpu_id) {
2688 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2689 return cpu_to_node()->at(cpu_id);
2690 }
2691 return -1;
2692 }
2693
2694 GrowableArray<int>* os::Linux::_cpu_to_node;
2695 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2696 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2697 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2698 os::Linux::numa_available_func_t os::Linux::_numa_available;
2699 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2700 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2701 unsigned long* os::Linux::_numa_all_nodes;
2702
2703 bool os::pd_uncommit_memory(char* addr, size_t size) {
2704 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2705 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2706 return res != (uintptr_t) MAP_FAILED;
2707 }
2708
2709 // Linux uses a growable mapping for the stack, and if the mapping for
2710 // the stack guard pages is not removed when we detach a thread the
2711 // stack cannot grow beyond the pages where the stack guard was
2712 // mapped. If at some point later in the process the stack expands to
2713 // that point, the Linux kernel cannot expand the stack any further
2714 // because the guard pages are in the way, and a segfault occurs.
2715 //
2716 // However, it's essential not to split the stack region by unmapping
2717 // a region (leaving a hole) that's already part of the stack mapping,
2718 // so if the stack mapping has already grown beyond the guard pages at
2719 // the time we create them, we have to truncate the stack mapping.
2720 // So, we need to know the extent of the stack mapping when
2721 // create_stack_guard_pages() is called.
2722
2723 // Find the bounds of the stack mapping. Return true for success.
2724 //
2725 // We only need this for stacks that are growable: at the time of
2726 // writing thread stacks don't use growable mappings (i.e. those
2727 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2728 // only applies to the main thread.
2729
2730 static
2731 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2732
2733 char buf[128];
2734 int fd, sz;
2735
2736 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2737 return false;
2738 }
2739
2740 const char kw[] = "[stack]";
2741 const int kwlen = sizeof(kw)-1;
2742
2743 // Address part of /proc/self/maps couldn't be more than 128 bytes
2744 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2745 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2746 // Extract addresses
2747 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2748 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2749 if (sp >= *bottom && sp <= *top) {
2750 ::close(fd);
2751 return true;
2752 }
2753 }
2754 }
2755 }
2756
2757 ::close(fd);
2758 return false;
2759 }
2760
2761
2762 // If the (growable) stack mapping already extends beyond the point
2763 // where we're going to put our guard pages, truncate the mapping at
2764 // that point by munmap()ping it. This ensures that when we later
2765 // munmap() the guard pages we don't leave a hole in the stack
2766 // mapping. This only affects the main/initial thread, but guard
2767 // against future OS changes
2768 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2769 uintptr_t stack_extent, stack_base;
2770 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2771 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2772 assert(os::Linux::is_initial_thread(),
2773 "growable stack in non-initial thread");
2774 if (stack_extent < (uintptr_t)addr)
2775 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2776 }
2777
2778 return os::commit_memory(addr, size);
2779 }
2780
2781 // If this is a growable mapping, remove the guard pages entirely by
2782 // munmap()ping them. If not, just call uncommit_memory(). This only
2783 // affects the main/initial thread, but guard against future OS changes
2784 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2785 uintptr_t stack_extent, stack_base;
2786 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2787 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2788 assert(os::Linux::is_initial_thread(),
2789 "growable stack in non-initial thread");
2790
2791 return ::munmap(addr, size) == 0;
2792 }
2793
2794 return os::uncommit_memory(addr, size);
2795 }
2796
2797 static address _highest_vm_reserved_address = NULL;
2798
2799 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2800 // at 'requested_addr'. If there are existing memory mappings at the same
2801 // location, however, they will be overwritten. If 'fixed' is false,
2802 // 'requested_addr' is only treated as a hint, the return value may or
2803 // may not start from the requested address. Unlike Linux mmap(), this
2804 // function returns NULL to indicate failure.
2805 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2806 char * addr;
2807 int flags;
2808
2809 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2810 if (fixed) {
2811 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2812 flags |= MAP_FIXED;
2813 }
2814
2815 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2816 // to PROT_EXEC if executable when we commit the page.
2817 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2818 flags, -1, 0);
2819
2820 if (addr != MAP_FAILED) {
2821 // anon_mmap() should only get called during VM initialization,
2822 // don't need lock (actually we can skip locking even it can be called
2823 // from multiple threads, because _highest_vm_reserved_address is just a
2824 // hint about the upper limit of non-stack memory regions.)
2825 if ((address)addr + bytes > _highest_vm_reserved_address) {
2826 _highest_vm_reserved_address = (address)addr + bytes;
2827 }
2828 }
2829
2830 return addr == MAP_FAILED ? NULL : addr;
2831 }
2832
2833 // Don't update _highest_vm_reserved_address, because there might be memory
2834 // regions above addr + size. If so, releasing a memory region only creates
2835 // a hole in the address space, it doesn't help prevent heap-stack collision.
2836 //
2837 static int anon_munmap(char * addr, size_t size) {
2838 return ::munmap(addr, size) == 0;
2839 }
2840
2841 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2842 size_t alignment_hint) {
2843 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2844 }
2845
2846 bool os::pd_release_memory(char* addr, size_t size) {
2847 return anon_munmap(addr, size);
2848 }
2849
2850 static address highest_vm_reserved_address() {
2851 return _highest_vm_reserved_address;
2852 }
2853
2854 static bool linux_mprotect(char* addr, size_t size, int prot) {
2855 // Linux wants the mprotect address argument to be page aligned.
2856 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2857
2858 // According to SUSv3, mprotect() should only be used with mappings
2859 // established by mmap(), and mmap() always maps whole pages. Unaligned
2860 // 'addr' likely indicates problem in the VM (e.g. trying to change
2861 // protection of malloc'ed or statically allocated memory). Check the
2862 // caller if you hit this assert.
2863 assert(addr == bottom, "sanity check");
2864
2865 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2866 return ::mprotect(bottom, size, prot) == 0;
2867 }
2868
2869 // Set protections specified
2870 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2871 bool is_committed) {
2872 unsigned int p = 0;
2873 switch (prot) {
2874 case MEM_PROT_NONE: p = PROT_NONE; break;
2875 case MEM_PROT_READ: p = PROT_READ; break;
2876 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2877 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2878 default:
2879 ShouldNotReachHere();
2880 }
2881 // is_committed is unused.
2882 return linux_mprotect(addr, bytes, p);
2883 }
2884
2885 bool os::guard_memory(char* addr, size_t size) {
2886 return linux_mprotect(addr, size, PROT_NONE);
2887 }
2888
2889 bool os::unguard_memory(char* addr, size_t size) {
2890 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2891 }
2892
2893 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2894 bool result = false;
2895 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2896 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2897 -1, 0);
2898
2899 if (p != (void *) -1) {
2900 // We don't know if this really is a huge page or not.
2901 FILE *fp = fopen("/proc/self/maps", "r");
2902 if (fp) {
2903 while (!feof(fp)) {
2904 char chars[257];
2905 long x = 0;
2906 if (fgets(chars, sizeof(chars), fp)) {
2907 if (sscanf(chars, "%lx-%*x", &x) == 1
2908 && x == (long)p) {
2909 if (strstr (chars, "hugepage")) {
2910 result = true;
2911 break;
2912 }
2913 }
2914 }
2915 }
2916 fclose(fp);
2917 }
2918 munmap (p, page_size);
2919 if (result)
2920 return true;
2921 }
2922
2923 if (warn) {
2924 warning("HugeTLBFS is not supported by the operating system.");
2925 }
2926
2927 return result;
2928 }
2929
2930 /*
2931 * Set the coredump_filter bits to include largepages in core dump (bit 6)
2932 *
2933 * From the coredump_filter documentation:
2934 *
2935 * - (bit 0) anonymous private memory
2936 * - (bit 1) anonymous shared memory
2937 * - (bit 2) file-backed private memory
2938 * - (bit 3) file-backed shared memory
2939 * - (bit 4) ELF header pages in file-backed private memory areas (it is
2940 * effective only if the bit 2 is cleared)
2941 * - (bit 5) hugetlb private memory
2942 * - (bit 6) hugetlb shared memory
2943 */
2944 static void set_coredump_filter(void) {
2945 FILE *f;
2946 long cdm;
2947
2948 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
2949 return;
2950 }
2951
2952 if (fscanf(f, "%lx", &cdm) != 1) {
2953 fclose(f);
2954 return;
2955 }
2956
2957 rewind(f);
2958
2959 if ((cdm & LARGEPAGES_BIT) == 0) {
2960 cdm |= LARGEPAGES_BIT;
2961 fprintf(f, "%#lx", cdm);
2962 }
2963
2964 fclose(f);
2965 }
2966
2967 // Large page support
2968
2969 static size_t _large_page_size = 0;
2970
2971 void os::large_page_init() {
2972 if (!UseLargePages) {
2973 UseHugeTLBFS = false;
2974 UseSHM = false;
2975 return;
2976 }
2977
2978 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
2979 // If UseLargePages is specified on the command line try both methods,
2980 // if it's default, then try only HugeTLBFS.
2981 if (FLAG_IS_DEFAULT(UseLargePages)) {
2982 UseHugeTLBFS = true;
2983 } else {
2984 UseHugeTLBFS = UseSHM = true;
2985 }
2986 }
2987
2988 if (LargePageSizeInBytes) {
2989 _large_page_size = LargePageSizeInBytes;
2990 } else {
2991 // large_page_size on Linux is used to round up heap size. x86 uses either
2992 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2993 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2994 // page as large as 256M.
2995 //
2996 // Here we try to figure out page size by parsing /proc/meminfo and looking
2997 // for a line with the following format:
2998 // Hugepagesize: 2048 kB
2999 //
3000 // If we can't determine the value (e.g. /proc is not mounted, or the text
3001 // format has been changed), we'll use the largest page size supported by
3002 // the processor.
3003
3004 #ifndef ZERO
3005 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3006 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3007 #endif // ZERO
3008
3009 FILE *fp = fopen("/proc/meminfo", "r");
3010 if (fp) {
3011 while (!feof(fp)) {
3012 int x = 0;
3013 char buf[16];
3014 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3015 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3016 _large_page_size = x * K;
3017 break;
3018 }
3019 } else {
3020 // skip to next line
3021 for (;;) {
3022 int ch = fgetc(fp);
3023 if (ch == EOF || ch == (int)'\n') break;
3024 }
3025 }
3026 }
3027 fclose(fp);
3028 }
3029 }
3030
3031 // print a warning if any large page related flag is specified on command line
3032 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3033
3034 const size_t default_page_size = (size_t)Linux::page_size();
3035 if (_large_page_size > default_page_size) {
3036 _page_sizes[0] = _large_page_size;
3037 _page_sizes[1] = default_page_size;
3038 _page_sizes[2] = 0;
3039 }
3040 UseHugeTLBFS = UseHugeTLBFS &&
3041 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3042
3043 if (UseHugeTLBFS)
3044 UseSHM = false;
3045
3046 UseLargePages = UseHugeTLBFS || UseSHM;
3047
3048 set_coredump_filter();
3049 }
3050
3051 #ifndef SHM_HUGETLB
3052 #define SHM_HUGETLB 04000
3053 #endif
3054
3055 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3056 // "exec" is passed in but not used. Creating the shared image for
3057 // the code cache doesn't have an SHM_X executable permission to check.
3058 assert(UseLargePages && UseSHM, "only for SHM large pages");
3059
3060 key_t key = IPC_PRIVATE;
3061 char *addr;
3062
3063 bool warn_on_failure = UseLargePages &&
3064 (!FLAG_IS_DEFAULT(UseLargePages) ||
3065 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3066 );
3067 char msg[128];
3068
3069 // Create a large shared memory region to attach to based on size.
3070 // Currently, size is the total size of the heap
3071 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3072 if (shmid == -1) {
3073 // Possible reasons for shmget failure:
3074 // 1. shmmax is too small for Java heap.
3075 // > check shmmax value: cat /proc/sys/kernel/shmmax
3076 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3077 // 2. not enough large page memory.
3078 // > check available large pages: cat /proc/meminfo
3079 // > increase amount of large pages:
3080 // echo new_value > /proc/sys/vm/nr_hugepages
3081 // Note 1: different Linux may use different name for this property,
3082 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3083 // Note 2: it's possible there's enough physical memory available but
3084 // they are so fragmented after a long run that they can't
3085 // coalesce into large pages. Try to reserve large pages when
3086 // the system is still "fresh".
3087 if (warn_on_failure) {
3088 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3089 warning(msg);
3090 }
3091 return NULL;
3092 }
3093
3094 // attach to the region
3095 addr = (char*)shmat(shmid, req_addr, 0);
3096 int err = errno;
3097
3098 // Remove shmid. If shmat() is successful, the actual shared memory segment
3099 // will be deleted when it's detached by shmdt() or when the process
3100 // terminates. If shmat() is not successful this will remove the shared
3101 // segment immediately.
3102 shmctl(shmid, IPC_RMID, NULL);
3103
3104 if ((intptr_t)addr == -1) {
3105 if (warn_on_failure) {
3106 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3107 warning(msg);
3108 }
3109 return NULL;
3110 }
3111
3112 if ((addr != NULL) && UseNUMAInterleaving) {
3113 numa_make_global(addr, bytes);
3114 }
3115
3116 return addr;
3117 }
3118
3119 bool os::release_memory_special(char* base, size_t bytes) {
3120 // detaching the SHM segment will also delete it, see reserve_memory_special()
3121 int rslt = shmdt(base);
3122 return rslt == 0;
3123 }
3124
3125 size_t os::large_page_size() {
3126 return _large_page_size;
3127 }
3128
3129 // HugeTLBFS allows application to commit large page memory on demand;
3130 // with SysV SHM the entire memory region must be allocated as shared
3131 // memory.
3132 bool os::can_commit_large_page_memory() {
3133 return UseHugeTLBFS;
3134 }
3135
3136 bool os::can_execute_large_page_memory() {
3137 return UseHugeTLBFS;
3138 }
3139
3140 // Reserve memory at an arbitrary address, only if that area is
3141 // available (and not reserved for something else).
3142
3143 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3144 const int max_tries = 10;
3145 char* base[max_tries];
3146 size_t size[max_tries];
3147 const size_t gap = 0x000000;
3148
3149 // Assert only that the size is a multiple of the page size, since
3150 // that's all that mmap requires, and since that's all we really know
3151 // about at this low abstraction level. If we need higher alignment,
3152 // we can either pass an alignment to this method or verify alignment
3153 // in one of the methods further up the call chain. See bug 5044738.
3154 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3155
3156 // Repeatedly allocate blocks until the block is allocated at the
3157 // right spot. Give up after max_tries. Note that reserve_memory() will
3158 // automatically update _highest_vm_reserved_address if the call is
3159 // successful. The variable tracks the highest memory address every reserved
3160 // by JVM. It is used to detect heap-stack collision if running with
3161 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3162 // space than needed, it could confuse the collision detecting code. To
3163 // solve the problem, save current _highest_vm_reserved_address and
3164 // calculate the correct value before return.
3165 address old_highest = _highest_vm_reserved_address;
3166
3167 // Linux mmap allows caller to pass an address as hint; give it a try first,
3168 // if kernel honors the hint then we can return immediately.
3169 char * addr = anon_mmap(requested_addr, bytes, false);
3170 if (addr == requested_addr) {
3171 return requested_addr;
3172 }
3173
3174 if (addr != NULL) {
3175 // mmap() is successful but it fails to reserve at the requested address
3176 anon_munmap(addr, bytes);
3177 }
3178
3179 int i;
3180 for (i = 0; i < max_tries; ++i) {
3181 base[i] = reserve_memory(bytes);
3182
3183 if (base[i] != NULL) {
3184 // Is this the block we wanted?
3185 if (base[i] == requested_addr) {
3186 size[i] = bytes;
3187 break;
3188 }
3189
3190 // Does this overlap the block we wanted? Give back the overlapped
3191 // parts and try again.
3192
3193 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3194 if (top_overlap >= 0 && top_overlap < bytes) {
3195 unmap_memory(base[i], top_overlap);
3196 base[i] += top_overlap;
3197 size[i] = bytes - top_overlap;
3198 } else {
3199 size_t bottom_overlap = base[i] + bytes - requested_addr;
3200 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3201 unmap_memory(requested_addr, bottom_overlap);
3202 size[i] = bytes - bottom_overlap;
3203 } else {
3204 size[i] = bytes;
3205 }
3206 }
3207 }
3208 }
3209
3210 // Give back the unused reserved pieces.
3211
3212 for (int j = 0; j < i; ++j) {
3213 if (base[j] != NULL) {
3214 unmap_memory(base[j], size[j]);
3215 }
3216 }
3217
3218 if (i < max_tries) {
3219 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3220 return requested_addr;
3221 } else {
3222 _highest_vm_reserved_address = old_highest;
3223 return NULL;
3224 }
3225 }
3226
3227 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3228 return ::read(fd, buf, nBytes);
3229 }
3230
3231 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3232 // Solaris uses poll(), linux uses park().
3233 // Poll() is likely a better choice, assuming that Thread.interrupt()
3234 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3235 // SIGSEGV, see 4355769.
3236
3237 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3238 assert(thread == Thread::current(), "thread consistency check");
3239
3240 ParkEvent * const slp = thread->_SleepEvent ;
3241 slp->reset() ;
3242 OrderAccess::fence() ;
3243
3244 if (interruptible) {
3245 jlong prevtime = javaTimeNanos();
3246
3247 for (;;) {
3248 if (os::is_interrupted(thread, true)) {
3249 return OS_INTRPT;
3250 }
3251
3252 jlong newtime = javaTimeNanos();
3253
3254 if (newtime - prevtime < 0) {
3255 // time moving backwards, should only happen if no monotonic clock
3256 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3257 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3258 } else {
3259 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3260 }
3261
3262 if(millis <= 0) {
3263 return OS_OK;
3264 }
3265
3266 prevtime = newtime;
3267
3268 {
3269 assert(thread->is_Java_thread(), "sanity check");
3270 JavaThread *jt = (JavaThread *) thread;
3271 ThreadBlockInVM tbivm(jt);
3272 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3273
3274 jt->set_suspend_equivalent();
3275 // cleared by handle_special_suspend_equivalent_condition() or
3276 // java_suspend_self() via check_and_wait_while_suspended()
3277
3278 slp->park(millis);
3279
3280 // were we externally suspended while we were waiting?
3281 jt->check_and_wait_while_suspended();
3282 }
3283 }
3284 } else {
3285 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3286 jlong prevtime = javaTimeNanos();
3287
3288 for (;;) {
3289 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3290 // the 1st iteration ...
3291 jlong newtime = javaTimeNanos();
3292
3293 if (newtime - prevtime < 0) {
3294 // time moving backwards, should only happen if no monotonic clock
3295 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3296 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3297 } else {
3298 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3299 }
3300
3301 if(millis <= 0) break ;
3302
3303 prevtime = newtime;
3304 slp->park(millis);
3305 }
3306 return OS_OK ;
3307 }
3308 }
3309
3310 int os::naked_sleep() {
3311 // %% make the sleep time an integer flag. for now use 1 millisec.
3312 return os::sleep(Thread::current(), 1, false);
3313 }
3314
3315 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3316 void os::infinite_sleep() {
3317 while (true) { // sleep forever ...
3318 ::sleep(100); // ... 100 seconds at a time
3319 }
3320 }
3321
3322 // Used to convert frequent JVM_Yield() to nops
3323 bool os::dont_yield() {
3324 return DontYieldALot;
3325 }
3326
3327 void os::yield() {
3328 sched_yield();
3329 }
3330
3331 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3332
3333 void os::yield_all(int attempts) {
3334 // Yields to all threads, including threads with lower priorities
3335 // Threads on Linux are all with same priority. The Solaris style
3336 // os::yield_all() with nanosleep(1ms) is not necessary.
3337 sched_yield();
3338 }
3339
3340 // Called from the tight loops to possibly influence time-sharing heuristics
3341 void os::loop_breaker(int attempts) {
3342 os::yield_all(attempts);
3343 }
3344
3345 ////////////////////////////////////////////////////////////////////////////////
3346 // thread priority support
3347
3348 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3349 // only supports dynamic priority, static priority must be zero. For real-time
3350 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3351 // However, for large multi-threaded applications, SCHED_RR is not only slower
3352 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3353 // of 5 runs - Sep 2005).
3354 //
3355 // The following code actually changes the niceness of kernel-thread/LWP. It
3356 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3357 // not the entire user process, and user level threads are 1:1 mapped to kernel
3358 // threads. It has always been the case, but could change in the future. For
3359 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3360 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3361
3362 int os::java_to_os_priority[CriticalPriority + 1] = {
3363 19, // 0 Entry should never be used
3364
3365 4, // 1 MinPriority
3366 3, // 2
3367 2, // 3
3368
3369 1, // 4
3370 0, // 5 NormPriority
3371 -1, // 6
3372
3373 -2, // 7
3374 -3, // 8
3375 -4, // 9 NearMaxPriority
3376
3377 -5, // 10 MaxPriority
3378
3379 -5 // 11 CriticalPriority
3380 };
3381
3382 static int prio_init() {
3383 if (ThreadPriorityPolicy == 1) {
3384 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3385 // if effective uid is not root. Perhaps, a more elegant way of doing
3386 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3387 if (geteuid() != 0) {
3388 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3389 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3390 }
3391 ThreadPriorityPolicy = 0;
3392 }
3393 }
3394 if (UseCriticalJavaThreadPriority) {
3395 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3396 }
3397 return 0;
3398 }
3399
3400 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3401 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3402
3403 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3404 return (ret == 0) ? OS_OK : OS_ERR;
3405 }
3406
3407 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3408 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3409 *priority_ptr = java_to_os_priority[NormPriority];
3410 return OS_OK;
3411 }
3412
3413 errno = 0;
3414 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3415 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3416 }
3417
3418 // Hint to the underlying OS that a task switch would not be good.
3419 // Void return because it's a hint and can fail.
3420 void os::hint_no_preempt() {}
3421
3422 ////////////////////////////////////////////////////////////////////////////////
3423 // suspend/resume support
3424
3425 // the low-level signal-based suspend/resume support is a remnant from the
3426 // old VM-suspension that used to be for java-suspension, safepoints etc,
3427 // within hotspot. Now there is a single use-case for this:
3428 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3429 // that runs in the watcher thread.
3430 // The remaining code is greatly simplified from the more general suspension
3431 // code that used to be used.
3432 //
3433 // The protocol is quite simple:
3434 // - suspend:
3435 // - sends a signal to the target thread
3436 // - polls the suspend state of the osthread using a yield loop
3437 // - target thread signal handler (SR_handler) sets suspend state
3438 // and blocks in sigsuspend until continued
3439 // - resume:
3440 // - sets target osthread state to continue
3441 // - sends signal to end the sigsuspend loop in the SR_handler
3442 //
3443 // Note that the SR_lock plays no role in this suspend/resume protocol.
3444 //
3445
3446 static void resume_clear_context(OSThread *osthread) {
3447 osthread->set_ucontext(NULL);
3448 osthread->set_siginfo(NULL);
3449
3450 // notify the suspend action is completed, we have now resumed
3451 osthread->sr.clear_suspended();
3452 }
3453
3454 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3455 osthread->set_ucontext(context);
3456 osthread->set_siginfo(siginfo);
3457 }
3458
3459 //
3460 // Handler function invoked when a thread's execution is suspended or
3461 // resumed. We have to be careful that only async-safe functions are
3462 // called here (Note: most pthread functions are not async safe and
3463 // should be avoided.)
3464 //
3465 // Note: sigwait() is a more natural fit than sigsuspend() from an
3466 // interface point of view, but sigwait() prevents the signal hander
3467 // from being run. libpthread would get very confused by not having
3468 // its signal handlers run and prevents sigwait()'s use with the
3469 // mutex granting granting signal.
3470 //
3471 // Currently only ever called on the VMThread
3472 //
3473 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3474 // Save and restore errno to avoid confusing native code with EINTR
3475 // after sigsuspend.
3476 int old_errno = errno;
3477
3478 Thread* thread = Thread::current();
3479 OSThread* osthread = thread->osthread();
3480 assert(thread->is_VM_thread(), "Must be VMThread");
3481 // read current suspend action
3482 int action = osthread->sr.suspend_action();
3483 if (action == os::Linux::SuspendResume::SR_SUSPEND) {
3484 suspend_save_context(osthread, siginfo, context);
3485
3486 // Notify the suspend action is about to be completed. do_suspend()
3487 // waits until SR_SUSPENDED is set and then returns. We will wait
3488 // here for a resume signal and that completes the suspend-other
3489 // action. do_suspend/do_resume is always called as a pair from
3490 // the same thread - so there are no races
3491
3492 // notify the caller
3493 osthread->sr.set_suspended();
3494
3495 sigset_t suspend_set; // signals for sigsuspend()
3496
3497 // get current set of blocked signals and unblock resume signal
3498 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3499 sigdelset(&suspend_set, SR_signum);
3500
3501 // wait here until we are resumed
3502 do {
3503 sigsuspend(&suspend_set);
3504 // ignore all returns until we get a resume signal
3505 } while (osthread->sr.suspend_action() != os::Linux::SuspendResume::SR_CONTINUE);
3506
3507 resume_clear_context(osthread);
3508
3509 } else {
3510 assert(action == os::Linux::SuspendResume::SR_CONTINUE, "unexpected sr action");
3511 // nothing special to do - just leave the handler
3512 }
3513
3514 errno = old_errno;
3515 }
3516
3517
3518 static int SR_initialize() {
3519 struct sigaction act;
3520 char *s;
3521 /* Get signal number to use for suspend/resume */
3522 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3523 int sig = ::strtol(s, 0, 10);
3524 if (sig > 0 || sig < _NSIG) {
3525 SR_signum = sig;
3526 }
3527 }
3528
3529 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3530 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3531
3532 sigemptyset(&SR_sigset);
3533 sigaddset(&SR_sigset, SR_signum);
3534
3535 /* Set up signal handler for suspend/resume */
3536 act.sa_flags = SA_RESTART|SA_SIGINFO;
3537 act.sa_handler = (void (*)(int)) SR_handler;
3538
3539 // SR_signum is blocked by default.
3540 // 4528190 - We also need to block pthread restart signal (32 on all
3541 // supported Linux platforms). Note that LinuxThreads need to block
3542 // this signal for all threads to work properly. So we don't have
3543 // to use hard-coded signal number when setting up the mask.
3544 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3545
3546 if (sigaction(SR_signum, &act, 0) == -1) {
3547 return -1;
3548 }
3549
3550 // Save signal flag
3551 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3552 return 0;
3553 }
3554
3555 static int SR_finalize() {
3556 return 0;
3557 }
3558
3559
3560 // returns true on success and false on error - really an error is fatal
3561 // but this seems the normal response to library errors
3562 static bool do_suspend(OSThread* osthread) {
3563 // mark as suspended and send signal
3564 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_SUSPEND);
3565 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3566 assert_status(status == 0, status, "pthread_kill");
3567
3568 // check status and wait until notified of suspension
3569 if (status == 0) {
3570 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3571 os::yield_all(i);
3572 }
3573 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3574 return true;
3575 }
3576 else {
3577 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3578 return false;
3579 }
3580 }
3581
3582 static void do_resume(OSThread* osthread) {
3583 assert(osthread->sr.is_suspended(), "thread should be suspended");
3584 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_CONTINUE);
3585
3586 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3587 assert_status(status == 0, status, "pthread_kill");
3588 // check status and wait unit notified of resumption
3589 if (status == 0) {
3590 for (int i = 0; osthread->sr.is_suspended(); i++) {
3591 os::yield_all(i);
3592 }
3593 }
3594 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3595 }
3596
3597 ////////////////////////////////////////////////////////////////////////////////
3598 // interrupt support
3599
3600 void os::interrupt(Thread* thread) {
3601 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3602 "possibility of dangling Thread pointer");
3603
3604 OSThread* osthread = thread->osthread();
3605
3606 if (!osthread->interrupted()) {
3607 osthread->set_interrupted(true);
3608 // More than one thread can get here with the same value of osthread,
3609 // resulting in multiple notifications. We do, however, want the store
3610 // to interrupted() to be visible to other threads before we execute unpark().
3611 OrderAccess::fence();
3612 ParkEvent * const slp = thread->_SleepEvent ;
3613 if (slp != NULL) slp->unpark() ;
3614 }
3615
3616 // For JSR166. Unpark even if interrupt status already was set
3617 if (thread->is_Java_thread())
3618 ((JavaThread*)thread)->parker()->unpark();
3619
3620 ParkEvent * ev = thread->_ParkEvent ;
3621 if (ev != NULL) ev->unpark() ;
3622
3623 }
3624
3625 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3626 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3627 "possibility of dangling Thread pointer");
3628
3629 OSThread* osthread = thread->osthread();
3630
3631 bool interrupted = osthread->interrupted();
3632
3633 if (interrupted && clear_interrupted) {
3634 osthread->set_interrupted(false);
3635 // consider thread->_SleepEvent->reset() ... optional optimization
3636 }
3637
3638 return interrupted;
3639 }
3640
3641 ///////////////////////////////////////////////////////////////////////////////////
3642 // signal handling (except suspend/resume)
3643
3644 // This routine may be used by user applications as a "hook" to catch signals.
3645 // The user-defined signal handler must pass unrecognized signals to this
3646 // routine, and if it returns true (non-zero), then the signal handler must
3647 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3648 // routine will never retun false (zero), but instead will execute a VM panic
3649 // routine kill the process.
3650 //
3651 // If this routine returns false, it is OK to call it again. This allows
3652 // the user-defined signal handler to perform checks either before or after
3653 // the VM performs its own checks. Naturally, the user code would be making
3654 // a serious error if it tried to handle an exception (such as a null check
3655 // or breakpoint) that the VM was generating for its own correct operation.
3656 //
3657 // This routine may recognize any of the following kinds of signals:
3658 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3659 // It should be consulted by handlers for any of those signals.
3660 //
3661 // The caller of this routine must pass in the three arguments supplied
3662 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3663 // field of the structure passed to sigaction(). This routine assumes that
3664 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3665 //
3666 // Note that the VM will print warnings if it detects conflicting signal
3667 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3668 //
3669 extern "C" JNIEXPORT int
3670 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3671 void* ucontext, int abort_if_unrecognized);
3672
3673 void signalHandler(int sig, siginfo_t* info, void* uc) {
3674 assert(info != NULL && uc != NULL, "it must be old kernel");
3675 int orig_errno = errno; // Preserve errno value over signal handler.
3676 JVM_handle_linux_signal(sig, info, uc, true);
3677 errno = orig_errno;
3678 }
3679
3680
3681 // This boolean allows users to forward their own non-matching signals
3682 // to JVM_handle_linux_signal, harmlessly.
3683 bool os::Linux::signal_handlers_are_installed = false;
3684
3685 // For signal-chaining
3686 struct sigaction os::Linux::sigact[MAXSIGNUM];
3687 unsigned int os::Linux::sigs = 0;
3688 bool os::Linux::libjsig_is_loaded = false;
3689 typedef struct sigaction *(*get_signal_t)(int);
3690 get_signal_t os::Linux::get_signal_action = NULL;
3691
3692 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3693 struct sigaction *actp = NULL;
3694
3695 if (libjsig_is_loaded) {
3696 // Retrieve the old signal handler from libjsig
3697 actp = (*get_signal_action)(sig);
3698 }
3699 if (actp == NULL) {
3700 // Retrieve the preinstalled signal handler from jvm
3701 actp = get_preinstalled_handler(sig);
3702 }
3703
3704 return actp;
3705 }
3706
3707 static bool call_chained_handler(struct sigaction *actp, int sig,
3708 siginfo_t *siginfo, void *context) {
3709 // Call the old signal handler
3710 if (actp->sa_handler == SIG_DFL) {
3711 // It's more reasonable to let jvm treat it as an unexpected exception
3712 // instead of taking the default action.
3713 return false;
3714 } else if (actp->sa_handler != SIG_IGN) {
3715 if ((actp->sa_flags & SA_NODEFER) == 0) {
3716 // automaticlly block the signal
3717 sigaddset(&(actp->sa_mask), sig);
3718 }
3719
3720 sa_handler_t hand;
3721 sa_sigaction_t sa;
3722 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3723 // retrieve the chained handler
3724 if (siginfo_flag_set) {
3725 sa = actp->sa_sigaction;
3726 } else {
3727 hand = actp->sa_handler;
3728 }
3729
3730 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3731 actp->sa_handler = SIG_DFL;
3732 }
3733
3734 // try to honor the signal mask
3735 sigset_t oset;
3736 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3737
3738 // call into the chained handler
3739 if (siginfo_flag_set) {
3740 (*sa)(sig, siginfo, context);
3741 } else {
3742 (*hand)(sig);
3743 }
3744
3745 // restore the signal mask
3746 pthread_sigmask(SIG_SETMASK, &oset, 0);
3747 }
3748 // Tell jvm's signal handler the signal is taken care of.
3749 return true;
3750 }
3751
3752 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3753 bool chained = false;
3754 // signal-chaining
3755 if (UseSignalChaining) {
3756 struct sigaction *actp = get_chained_signal_action(sig);
3757 if (actp != NULL) {
3758 chained = call_chained_handler(actp, sig, siginfo, context);
3759 }
3760 }
3761 return chained;
3762 }
3763
3764 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3765 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3766 return &sigact[sig];
3767 }
3768 return NULL;
3769 }
3770
3771 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3772 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3773 sigact[sig] = oldAct;
3774 sigs |= (unsigned int)1 << sig;
3775 }
3776
3777 // for diagnostic
3778 int os::Linux::sigflags[MAXSIGNUM];
3779
3780 int os::Linux::get_our_sigflags(int sig) {
3781 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3782 return sigflags[sig];
3783 }
3784
3785 void os::Linux::set_our_sigflags(int sig, int flags) {
3786 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3787 sigflags[sig] = flags;
3788 }
3789
3790 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3791 // Check for overwrite.
3792 struct sigaction oldAct;
3793 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3794
3795 void* oldhand = oldAct.sa_sigaction
3796 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3797 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3798 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3799 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3800 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3801 if (AllowUserSignalHandlers || !set_installed) {
3802 // Do not overwrite; user takes responsibility to forward to us.
3803 return;
3804 } else if (UseSignalChaining) {
3805 // save the old handler in jvm
3806 save_preinstalled_handler(sig, oldAct);
3807 // libjsig also interposes the sigaction() call below and saves the
3808 // old sigaction on it own.
3809 } else {
3810 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3811 "%#lx for signal %d.", (long)oldhand, sig));
3812 }
3813 }
3814
3815 struct sigaction sigAct;
3816 sigfillset(&(sigAct.sa_mask));
3817 sigAct.sa_handler = SIG_DFL;
3818 if (!set_installed) {
3819 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3820 } else {
3821 sigAct.sa_sigaction = signalHandler;
3822 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3823 }
3824 // Save flags, which are set by ours
3825 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3826 sigflags[sig] = sigAct.sa_flags;
3827
3828 int ret = sigaction(sig, &sigAct, &oldAct);
3829 assert(ret == 0, "check");
3830
3831 void* oldhand2 = oldAct.sa_sigaction
3832 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3833 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3834 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3835 }
3836
3837 // install signal handlers for signals that HotSpot needs to
3838 // handle in order to support Java-level exception handling.
3839
3840 void os::Linux::install_signal_handlers() {
3841 if (!signal_handlers_are_installed) {
3842 signal_handlers_are_installed = true;
3843
3844 // signal-chaining
3845 typedef void (*signal_setting_t)();
3846 signal_setting_t begin_signal_setting = NULL;
3847 signal_setting_t end_signal_setting = NULL;
3848 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3849 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3850 if (begin_signal_setting != NULL) {
3851 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3852 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3853 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3854 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3855 libjsig_is_loaded = true;
3856 assert(UseSignalChaining, "should enable signal-chaining");
3857 }
3858 if (libjsig_is_loaded) {
3859 // Tell libjsig jvm is setting signal handlers
3860 (*begin_signal_setting)();
3861 }
3862
3863 set_signal_handler(SIGSEGV, true);
3864 set_signal_handler(SIGPIPE, true);
3865 set_signal_handler(SIGBUS, true);
3866 set_signal_handler(SIGILL, true);
3867 set_signal_handler(SIGFPE, true);
3868 set_signal_handler(SIGXFSZ, true);
3869
3870 if (libjsig_is_loaded) {
3871 // Tell libjsig jvm finishes setting signal handlers
3872 (*end_signal_setting)();
3873 }
3874
3875 // We don't activate signal checker if libjsig is in place, we trust ourselves
3876 // and if UserSignalHandler is installed all bets are off.
3877 // Log that signal checking is off only if -verbose:jni is specified.
3878 if (CheckJNICalls) {
3879 if (libjsig_is_loaded) {
3880 if (PrintJNIResolving) {
3881 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3882 }
3883 check_signals = false;
3884 }
3885 if (AllowUserSignalHandlers) {
3886 if (PrintJNIResolving) {
3887 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3888 }
3889 check_signals = false;
3890 }
3891 }
3892 }
3893 }
3894
3895 // This is the fastest way to get thread cpu time on Linux.
3896 // Returns cpu time (user+sys) for any thread, not only for current.
3897 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3898 // It might work on 2.6.10+ with a special kernel/glibc patch.
3899 // For reference, please, see IEEE Std 1003.1-2004:
3900 // http://www.unix.org/single_unix_specification
3901
3902 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3903 struct timespec tp;
3904 int rc = os::Linux::clock_gettime(clockid, &tp);
3905 assert(rc == 0, "clock_gettime is expected to return 0 code");
3906
3907 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
3908 }
3909
3910 /////
3911 // glibc on Linux platform uses non-documented flag
3912 // to indicate, that some special sort of signal
3913 // trampoline is used.
3914 // We will never set this flag, and we should
3915 // ignore this flag in our diagnostic
3916 #ifdef SIGNIFICANT_SIGNAL_MASK
3917 #undef SIGNIFICANT_SIGNAL_MASK
3918 #endif
3919 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3920
3921 static const char* get_signal_handler_name(address handler,
3922 char* buf, int buflen) {
3923 int offset;
3924 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3925 if (found) {
3926 // skip directory names
3927 const char *p1, *p2;
3928 p1 = buf;
3929 size_t len = strlen(os::file_separator());
3930 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3931 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3932 } else {
3933 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3934 }
3935 return buf;
3936 }
3937
3938 static void print_signal_handler(outputStream* st, int sig,
3939 char* buf, size_t buflen) {
3940 struct sigaction sa;
3941
3942 sigaction(sig, NULL, &sa);
3943
3944 // See comment for SIGNIFICANT_SIGNAL_MASK define
3945 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3946
3947 st->print("%s: ", os::exception_name(sig, buf, buflen));
3948
3949 address handler = (sa.sa_flags & SA_SIGINFO)
3950 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3951 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3952
3953 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3954 st->print("SIG_DFL");
3955 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3956 st->print("SIG_IGN");
3957 } else {
3958 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3959 }
3960
3961 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3962
3963 address rh = VMError::get_resetted_sighandler(sig);
3964 // May be, handler was resetted by VMError?
3965 if(rh != NULL) {
3966 handler = rh;
3967 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3968 }
3969
3970 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3971
3972 // Check: is it our handler?
3973 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3974 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3975 // It is our signal handler
3976 // check for flags, reset system-used one!
3977 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3978 st->print(
3979 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3980 os::Linux::get_our_sigflags(sig));
3981 }
3982 }
3983 st->cr();
3984 }
3985
3986
3987 #define DO_SIGNAL_CHECK(sig) \
3988 if (!sigismember(&check_signal_done, sig)) \
3989 os::Linux::check_signal_handler(sig)
3990
3991 // This method is a periodic task to check for misbehaving JNI applications
3992 // under CheckJNI, we can add any periodic checks here
3993
3994 void os::run_periodic_checks() {
3995
3996 if (check_signals == false) return;
3997
3998 // SEGV and BUS if overridden could potentially prevent
3999 // generation of hs*.log in the event of a crash, debugging
4000 // such a case can be very challenging, so we absolutely
4001 // check the following for a good measure:
4002 DO_SIGNAL_CHECK(SIGSEGV);
4003 DO_SIGNAL_CHECK(SIGILL);
4004 DO_SIGNAL_CHECK(SIGFPE);
4005 DO_SIGNAL_CHECK(SIGBUS);
4006 DO_SIGNAL_CHECK(SIGPIPE);
4007 DO_SIGNAL_CHECK(SIGXFSZ);
4008
4009
4010 // ReduceSignalUsage allows the user to override these handlers
4011 // see comments at the very top and jvm_solaris.h
4012 if (!ReduceSignalUsage) {
4013 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4014 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4015 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4016 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4017 }
4018
4019 DO_SIGNAL_CHECK(SR_signum);
4020 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4021 }
4022
4023 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4024
4025 static os_sigaction_t os_sigaction = NULL;
4026
4027 void os::Linux::check_signal_handler(int sig) {
4028 char buf[O_BUFLEN];
4029 address jvmHandler = NULL;
4030
4031
4032 struct sigaction act;
4033 if (os_sigaction == NULL) {
4034 // only trust the default sigaction, in case it has been interposed
4035 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4036 if (os_sigaction == NULL) return;
4037 }
4038
4039 os_sigaction(sig, (struct sigaction*)NULL, &act);
4040
4041
4042 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4043
4044 address thisHandler = (act.sa_flags & SA_SIGINFO)
4045 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4046 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4047
4048
4049 switch(sig) {
4050 case SIGSEGV:
4051 case SIGBUS:
4052 case SIGFPE:
4053 case SIGPIPE:
4054 case SIGILL:
4055 case SIGXFSZ:
4056 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4057 break;
4058
4059 case SHUTDOWN1_SIGNAL:
4060 case SHUTDOWN2_SIGNAL:
4061 case SHUTDOWN3_SIGNAL:
4062 case BREAK_SIGNAL:
4063 jvmHandler = (address)user_handler();
4064 break;
4065
4066 case INTERRUPT_SIGNAL:
4067 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4068 break;
4069
4070 default:
4071 if (sig == SR_signum) {
4072 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4073 } else {
4074 return;
4075 }
4076 break;
4077 }
4078
4079 if (thisHandler != jvmHandler) {
4080 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4081 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4082 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4083 // No need to check this sig any longer
4084 sigaddset(&check_signal_done, sig);
4085 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4086 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4087 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4088 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4089 // No need to check this sig any longer
4090 sigaddset(&check_signal_done, sig);
4091 }
4092
4093 // Dump all the signal
4094 if (sigismember(&check_signal_done, sig)) {
4095 print_signal_handlers(tty, buf, O_BUFLEN);
4096 }
4097 }
4098
4099 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4100
4101 extern bool signal_name(int signo, char* buf, size_t len);
4102
4103 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4104 if (0 < exception_code && exception_code <= SIGRTMAX) {
4105 // signal
4106 if (!signal_name(exception_code, buf, size)) {
4107 jio_snprintf(buf, size, "SIG%d", exception_code);
4108 }
4109 return buf;
4110 } else {
4111 return NULL;
4112 }
4113 }
4114
4115 // this is called _before_ the most of global arguments have been parsed
4116 void os::init(void) {
4117 char dummy; /* used to get a guess on initial stack address */
4118 // first_hrtime = gethrtime();
4119
4120 // With LinuxThreads the JavaMain thread pid (primordial thread)
4121 // is different than the pid of the java launcher thread.
4122 // So, on Linux, the launcher thread pid is passed to the VM
4123 // via the sun.java.launcher.pid property.
4124 // Use this property instead of getpid() if it was correctly passed.
4125 // See bug 6351349.
4126 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4127
4128 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4129
4130 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4131
4132 init_random(1234567);
4133
4134 ThreadCritical::initialize();
4135
4136 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4137 if (Linux::page_size() == -1) {
4138 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4139 strerror(errno)));
4140 }
4141 init_page_sizes((size_t) Linux::page_size());
4142
4143 Linux::initialize_system_info();
4144
4145 // main_thread points to the aboriginal thread
4146 Linux::_main_thread = pthread_self();
4147
4148 Linux::clock_init();
4149 initial_time_count = os::elapsed_counter();
4150 pthread_mutex_init(&dl_mutex, NULL);
4151 }
4152
4153 // To install functions for atexit system call
4154 extern "C" {
4155 static void perfMemory_exit_helper() {
4156 perfMemory_exit();
4157 }
4158 }
4159
4160 // this is called _after_ the global arguments have been parsed
4161 jint os::init_2(void)
4162 {
4163 Linux::fast_thread_clock_init();
4164
4165 // Allocate a single page and mark it as readable for safepoint polling
4166 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4167 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4168
4169 os::set_polling_page( polling_page );
4170
4171 #ifndef PRODUCT
4172 if(Verbose && PrintMiscellaneous)
4173 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4174 #endif
4175
4176 if (!UseMembar) {
4177 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4178 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4179 os::set_memory_serialize_page( mem_serialize_page );
4180
4181 #ifndef PRODUCT
4182 if(Verbose && PrintMiscellaneous)
4183 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4184 #endif
4185 }
4186
4187 os::large_page_init();
4188
4189 // initialize suspend/resume support - must do this before signal_sets_init()
4190 if (SR_initialize() != 0) {
4191 perror("SR_initialize failed");
4192 return JNI_ERR;
4193 }
4194
4195 Linux::signal_sets_init();
4196 Linux::install_signal_handlers();
4197
4198 // Check minimum allowable stack size for thread creation and to initialize
4199 // the java system classes, including StackOverflowError - depends on page
4200 // size. Add a page for compiler2 recursion in main thread.
4201 // Add in 2*BytesPerWord times page size to account for VM stack during
4202 // class initialization depending on 32 or 64 bit VM.
4203 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4204 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4205 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4206
4207 size_t threadStackSizeInBytes = ThreadStackSize * K;
4208 if (threadStackSizeInBytes != 0 &&
4209 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4210 tty->print_cr("\nThe stack size specified is too small, "
4211 "Specify at least %dk",
4212 os::Linux::min_stack_allowed/ K);
4213 return JNI_ERR;
4214 }
4215
4216 // Make the stack size a multiple of the page size so that
4217 // the yellow/red zones can be guarded.
4218 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4219 vm_page_size()));
4220
4221 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4222
4223 Linux::libpthread_init();
4224 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4225 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4226 Linux::glibc_version(), Linux::libpthread_version(),
4227 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4228 }
4229
4230 if (UseNUMA) {
4231 if (!Linux::libnuma_init()) {
4232 UseNUMA = false;
4233 } else {
4234 if ((Linux::numa_max_node() < 1)) {
4235 // There's only one node(they start from 0), disable NUMA.
4236 UseNUMA = false;
4237 }
4238 }
4239 // With SHM large pages we cannot uncommit a page, so there's not way
4240 // we can make the adaptive lgrp chunk resizing work. If the user specified
4241 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4242 // disable adaptive resizing.
4243 if (UseNUMA && UseLargePages && UseSHM) {
4244 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4245 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4246 UseLargePages = false;
4247 } else {
4248 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4249 UseAdaptiveSizePolicy = false;
4250 UseAdaptiveNUMAChunkSizing = false;
4251 }
4252 } else {
4253 UseNUMA = false;
4254 }
4255 }
4256 if (!UseNUMA && ForceNUMA) {
4257 UseNUMA = true;
4258 }
4259 }
4260
4261 if (MaxFDLimit) {
4262 // set the number of file descriptors to max. print out error
4263 // if getrlimit/setrlimit fails but continue regardless.
4264 struct rlimit nbr_files;
4265 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4266 if (status != 0) {
4267 if (PrintMiscellaneous && (Verbose || WizardMode))
4268 perror("os::init_2 getrlimit failed");
4269 } else {
4270 nbr_files.rlim_cur = nbr_files.rlim_max;
4271 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4272 if (status != 0) {
4273 if (PrintMiscellaneous && (Verbose || WizardMode))
4274 perror("os::init_2 setrlimit failed");
4275 }
4276 }
4277 }
4278
4279 // Initialize lock used to serialize thread creation (see os::create_thread)
4280 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4281
4282 // at-exit methods are called in the reverse order of their registration.
4283 // atexit functions are called on return from main or as a result of a
4284 // call to exit(3C). There can be only 32 of these functions registered
4285 // and atexit() does not set errno.
4286
4287 if (PerfAllowAtExitRegistration) {
4288 // only register atexit functions if PerfAllowAtExitRegistration is set.
4289 // atexit functions can be delayed until process exit time, which
4290 // can be problematic for embedded VM situations. Embedded VMs should
4291 // call DestroyJavaVM() to assure that VM resources are released.
4292
4293 // note: perfMemory_exit_helper atexit function may be removed in
4294 // the future if the appropriate cleanup code can be added to the
4295 // VM_Exit VMOperation's doit method.
4296 if (atexit(perfMemory_exit_helper) != 0) {
4297 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4298 }
4299 }
4300
4301 // initialize thread priority policy
4302 prio_init();
4303
4304 return JNI_OK;
4305 }
4306
4307 // this is called at the end of vm_initialization
4308 void os::init_3(void)
4309 {
4310 #ifdef JAVASE_EMBEDDED
4311 // Start the MemNotifyThread
4312 if (LowMemoryProtection) {
4313 MemNotifyThread::start();
4314 }
4315 return;
4316 #endif
4317 }
4318
4319 // Mark the polling page as unreadable
4320 void os::make_polling_page_unreadable(void) {
4321 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4322 fatal("Could not disable polling page");
4323 };
4324
4325 // Mark the polling page as readable
4326 void os::make_polling_page_readable(void) {
4327 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4328 fatal("Could not enable polling page");
4329 }
4330 };
4331
4332 int os::active_processor_count() {
4333 // Linux doesn't yet have a (official) notion of processor sets,
4334 // so just return the number of online processors.
4335 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4336 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4337 return online_cpus;
4338 }
4339
4340 void os::set_native_thread_name(const char *name) {
4341 // Not yet implemented.
4342 return;
4343 }
4344
4345 bool os::distribute_processes(uint length, uint* distribution) {
4346 // Not yet implemented.
4347 return false;
4348 }
4349
4350 bool os::bind_to_processor(uint processor_id) {
4351 // Not yet implemented.
4352 return false;
4353 }
4354
4355 ///
4356
4357 // Suspends the target using the signal mechanism and then grabs the PC before
4358 // resuming the target. Used by the flat-profiler only
4359 ExtendedPC os::get_thread_pc(Thread* thread) {
4360 // Make sure that it is called by the watcher for the VMThread
4361 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4362 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4363
4364 ExtendedPC epc;
4365
4366 OSThread* osthread = thread->osthread();
4367 if (do_suspend(osthread)) {
4368 if (osthread->ucontext() != NULL) {
4369 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4370 } else {
4371 // NULL context is unexpected, double-check this is the VMThread
4372 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4373 }
4374 do_resume(osthread);
4375 }
4376 // failure means pthread_kill failed for some reason - arguably this is
4377 // a fatal problem, but such problems are ignored elsewhere
4378
4379 return epc;
4380 }
4381
4382 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4383 {
4384 if (is_NPTL()) {
4385 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4386 } else {
4387 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4388 // word back to default 64bit precision if condvar is signaled. Java
4389 // wants 53bit precision. Save and restore current value.
4390 int fpu = get_fpu_control_word();
4391 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4392 set_fpu_control_word(fpu);
4393 return status;
4394 }
4395 }
4396
4397 ////////////////////////////////////////////////////////////////////////////////
4398 // debug support
4399
4400 static address same_page(address x, address y) {
4401 int page_bits = -os::vm_page_size();
4402 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4403 return x;
4404 else if (x > y)
4405 return (address)(intptr_t(y) | ~page_bits) + 1;
4406 else
4407 return (address)(intptr_t(y) & page_bits);
4408 }
4409
4410 bool os::find(address addr, outputStream* st) {
4411 Dl_info dlinfo;
4412 memset(&dlinfo, 0, sizeof(dlinfo));
4413 if (dladdr(addr, &dlinfo)) {
4414 st->print(PTR_FORMAT ": ", addr);
4415 if (dlinfo.dli_sname != NULL) {
4416 st->print("%s+%#x", dlinfo.dli_sname,
4417 addr - (intptr_t)dlinfo.dli_saddr);
4418 } else if (dlinfo.dli_fname) {
4419 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4420 } else {
4421 st->print("<absolute address>");
4422 }
4423 if (dlinfo.dli_fname) {
4424 st->print(" in %s", dlinfo.dli_fname);
4425 }
4426 if (dlinfo.dli_fbase) {
4427 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4428 }
4429 st->cr();
4430
4431 if (Verbose) {
4432 // decode some bytes around the PC
4433 address begin = same_page(addr-40, addr);
4434 address end = same_page(addr+40, addr);
4435 address lowest = (address) dlinfo.dli_sname;
4436 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4437 if (begin < lowest) begin = lowest;
4438 Dl_info dlinfo2;
4439 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4440 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4441 end = (address) dlinfo2.dli_saddr;
4442 Disassembler::decode(begin, end, st);
4443 }
4444 return true;
4445 }
4446 return false;
4447 }
4448
4449 ////////////////////////////////////////////////////////////////////////////////
4450 // misc
4451
4452 // This does not do anything on Linux. This is basically a hook for being
4453 // able to use structured exception handling (thread-local exception filters)
4454 // on, e.g., Win32.
4455 void
4456 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4457 JavaCallArguments* args, Thread* thread) {
4458 f(value, method, args, thread);
4459 }
4460
4461 void os::print_statistics() {
4462 }
4463
4464 int os::message_box(const char* title, const char* message) {
4465 int i;
4466 fdStream err(defaultStream::error_fd());
4467 for (i = 0; i < 78; i++) err.print_raw("=");
4468 err.cr();
4469 err.print_raw_cr(title);
4470 for (i = 0; i < 78; i++) err.print_raw("-");
4471 err.cr();
4472 err.print_raw_cr(message);
4473 for (i = 0; i < 78; i++) err.print_raw("=");
4474 err.cr();
4475
4476 char buf[16];
4477 // Prevent process from exiting upon "read error" without consuming all CPU
4478 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4479
4480 return buf[0] == 'y' || buf[0] == 'Y';
4481 }
4482
4483 int os::stat(const char *path, struct stat *sbuf) {
4484 char pathbuf[MAX_PATH];
4485 if (strlen(path) > MAX_PATH - 1) {
4486 errno = ENAMETOOLONG;
4487 return -1;
4488 }
4489 os::native_path(strcpy(pathbuf, path));
4490 return ::stat(pathbuf, sbuf);
4491 }
4492
4493 bool os::check_heap(bool force) {
4494 return true;
4495 }
4496
4497 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4498 return ::vsnprintf(buf, count, format, args);
4499 }
4500
4501 // Is a (classpath) directory empty?
4502 bool os::dir_is_empty(const char* path) {
4503 DIR *dir = NULL;
4504 struct dirent *ptr;
4505
4506 dir = opendir(path);
4507 if (dir == NULL) return true;
4508
4509 /* Scan the directory */
4510 bool result = true;
4511 char buf[sizeof(struct dirent) + MAX_PATH];
4512 while (result && (ptr = ::readdir(dir)) != NULL) {
4513 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4514 result = false;
4515 }
4516 }
4517 closedir(dir);
4518 return result;
4519 }
4520
4521 // This code originates from JDK's sysOpen and open64_w
4522 // from src/solaris/hpi/src/system_md.c
4523
4524 #ifndef O_DELETE
4525 #define O_DELETE 0x10000
4526 #endif
4527
4528 // Open a file. Unlink the file immediately after open returns
4529 // if the specified oflag has the O_DELETE flag set.
4530 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4531
4532 int os::open(const char *path, int oflag, int mode) {
4533
4534 if (strlen(path) > MAX_PATH - 1) {
4535 errno = ENAMETOOLONG;
4536 return -1;
4537 }
4538 int fd;
4539 int o_delete = (oflag & O_DELETE);
4540 oflag = oflag & ~O_DELETE;
4541
4542 fd = ::open64(path, oflag, mode);
4543 if (fd == -1) return -1;
4544
4545 //If the open succeeded, the file might still be a directory
4546 {
4547 struct stat64 buf64;
4548 int ret = ::fstat64(fd, &buf64);
4549 int st_mode = buf64.st_mode;
4550
4551 if (ret != -1) {
4552 if ((st_mode & S_IFMT) == S_IFDIR) {
4553 errno = EISDIR;
4554 ::close(fd);
4555 return -1;
4556 }
4557 } else {
4558 ::close(fd);
4559 return -1;
4560 }
4561 }
4562
4563 /*
4564 * All file descriptors that are opened in the JVM and not
4565 * specifically destined for a subprocess should have the
4566 * close-on-exec flag set. If we don't set it, then careless 3rd
4567 * party native code might fork and exec without closing all
4568 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4569 * UNIXProcess.c), and this in turn might:
4570 *
4571 * - cause end-of-file to fail to be detected on some file
4572 * descriptors, resulting in mysterious hangs, or
4573 *
4574 * - might cause an fopen in the subprocess to fail on a system
4575 * suffering from bug 1085341.
4576 *
4577 * (Yes, the default setting of the close-on-exec flag is a Unix
4578 * design flaw)
4579 *
4580 * See:
4581 * 1085341: 32-bit stdio routines should support file descriptors >255
4582 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4583 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4584 */
4585 #ifdef FD_CLOEXEC
4586 {
4587 int flags = ::fcntl(fd, F_GETFD);
4588 if (flags != -1)
4589 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4590 }
4591 #endif
4592
4593 if (o_delete != 0) {
4594 ::unlink(path);
4595 }
4596 return fd;
4597 }
4598
4599
4600 // create binary file, rewriting existing file if required
4601 int os::create_binary_file(const char* path, bool rewrite_existing) {
4602 int oflags = O_WRONLY | O_CREAT;
4603 if (!rewrite_existing) {
4604 oflags |= O_EXCL;
4605 }
4606 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4607 }
4608
4609 // return current position of file pointer
4610 jlong os::current_file_offset(int fd) {
4611 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4612 }
4613
4614 // move file pointer to the specified offset
4615 jlong os::seek_to_file_offset(int fd, jlong offset) {
4616 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4617 }
4618
4619 // This code originates from JDK's sysAvailable
4620 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4621
4622 int os::available(int fd, jlong *bytes) {
4623 jlong cur, end;
4624 int mode;
4625 struct stat64 buf64;
4626
4627 if (::fstat64(fd, &buf64) >= 0) {
4628 mode = buf64.st_mode;
4629 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4630 /*
4631 * XXX: is the following call interruptible? If so, this might
4632 * need to go through the INTERRUPT_IO() wrapper as for other
4633 * blocking, interruptible calls in this file.
4634 */
4635 int n;
4636 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4637 *bytes = n;
4638 return 1;
4639 }
4640 }
4641 }
4642 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4643 return 0;
4644 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4645 return 0;
4646 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4647 return 0;
4648 }
4649 *bytes = end - cur;
4650 return 1;
4651 }
4652
4653 int os::socket_available(int fd, jint *pbytes) {
4654 // Linux doc says EINTR not returned, unlike Solaris
4655 int ret = ::ioctl(fd, FIONREAD, pbytes);
4656
4657 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4658 // is expected to return 0 on failure and 1 on success to the jdk.
4659 return (ret < 0) ? 0 : 1;
4660 }
4661
4662 // Map a block of memory.
4663 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4664 char *addr, size_t bytes, bool read_only,
4665 bool allow_exec) {
4666 int prot;
4667 int flags = MAP_PRIVATE;
4668
4669 if (read_only) {
4670 prot = PROT_READ;
4671 } else {
4672 prot = PROT_READ | PROT_WRITE;
4673 }
4674
4675 if (allow_exec) {
4676 prot |= PROT_EXEC;
4677 }
4678
4679 if (addr != NULL) {
4680 flags |= MAP_FIXED;
4681 }
4682
4683 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4684 fd, file_offset);
4685 if (mapped_address == MAP_FAILED) {
4686 return NULL;
4687 }
4688 return mapped_address;
4689 }
4690
4691
4692 // Remap a block of memory.
4693 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4694 char *addr, size_t bytes, bool read_only,
4695 bool allow_exec) {
4696 // same as map_memory() on this OS
4697 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4698 allow_exec);
4699 }
4700
4701
4702 // Unmap a block of memory.
4703 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4704 return munmap(addr, bytes) == 0;
4705 }
4706
4707 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4708
4709 static clockid_t thread_cpu_clockid(Thread* thread) {
4710 pthread_t tid = thread->osthread()->pthread_id();
4711 clockid_t clockid;
4712
4713 // Get thread clockid
4714 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4715 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4716 return clockid;
4717 }
4718
4719 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4720 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4721 // of a thread.
4722 //
4723 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4724 // the fast estimate available on the platform.
4725
4726 jlong os::current_thread_cpu_time() {
4727 if (os::Linux::supports_fast_thread_cpu_time()) {
4728 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4729 } else {
4730 // return user + sys since the cost is the same
4731 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4732 }
4733 }
4734
4735 jlong os::thread_cpu_time(Thread* thread) {
4736 // consistent with what current_thread_cpu_time() returns
4737 if (os::Linux::supports_fast_thread_cpu_time()) {
4738 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4739 } else {
4740 return slow_thread_cpu_time(thread, true /* user + sys */);
4741 }
4742 }
4743
4744 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4745 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4746 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4747 } else {
4748 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4749 }
4750 }
4751
4752 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4753 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4754 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4755 } else {
4756 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4757 }
4758 }
4759
4760 //
4761 // -1 on error.
4762 //
4763
4764 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4765 static bool proc_task_unchecked = true;
4766 static const char *proc_stat_path = "/proc/%d/stat";
4767 pid_t tid = thread->osthread()->thread_id();
4768 char *s;
4769 char stat[2048];
4770 int statlen;
4771 char proc_name[64];
4772 int count;
4773 long sys_time, user_time;
4774 char cdummy;
4775 int idummy;
4776 long ldummy;
4777 FILE *fp;
4778
4779 // The /proc/<tid>/stat aggregates per-process usage on
4780 // new Linux kernels 2.6+ where NPTL is supported.
4781 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4782 // See bug 6328462.
4783 // There possibly can be cases where there is no directory
4784 // /proc/self/task, so we check its availability.
4785 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4786 // This is executed only once
4787 proc_task_unchecked = false;
4788 fp = fopen("/proc/self/task", "r");
4789 if (fp != NULL) {
4790 proc_stat_path = "/proc/self/task/%d/stat";
4791 fclose(fp);
4792 }
4793 }
4794
4795 sprintf(proc_name, proc_stat_path, tid);
4796 fp = fopen(proc_name, "r");
4797 if ( fp == NULL ) return -1;
4798 statlen = fread(stat, 1, 2047, fp);
4799 stat[statlen] = '\0';
4800 fclose(fp);
4801
4802 // Skip pid and the command string. Note that we could be dealing with
4803 // weird command names, e.g. user could decide to rename java launcher
4804 // to "java 1.4.2 :)", then the stat file would look like
4805 // 1234 (java 1.4.2 :)) R ... ...
4806 // We don't really need to know the command string, just find the last
4807 // occurrence of ")" and then start parsing from there. See bug 4726580.
4808 s = strrchr(stat, ')');
4809 if (s == NULL ) return -1;
4810
4811 // Skip blank chars
4812 do s++; while (isspace(*s));
4813
4814 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4815 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4816 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4817 &user_time, &sys_time);
4818 if ( count != 13 ) return -1;
4819 if (user_sys_cpu_time) {
4820 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4821 } else {
4822 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4823 }
4824 }
4825
4826 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4827 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4828 info_ptr->may_skip_backward = false; // elapsed time not wall time
4829 info_ptr->may_skip_forward = false; // elapsed time not wall time
4830 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4831 }
4832
4833 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4834 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4835 info_ptr->may_skip_backward = false; // elapsed time not wall time
4836 info_ptr->may_skip_forward = false; // elapsed time not wall time
4837 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4838 }
4839
4840 bool os::is_thread_cpu_time_supported() {
4841 return true;
4842 }
4843
4844 // System loadavg support. Returns -1 if load average cannot be obtained.
4845 // Linux doesn't yet have a (official) notion of processor sets,
4846 // so just return the system wide load average.
4847 int os::loadavg(double loadavg[], int nelem) {
4848 return ::getloadavg(loadavg, nelem);
4849 }
4850
4851 void os::pause() {
4852 char filename[MAX_PATH];
4853 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4854 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4855 } else {
4856 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4857 }
4858
4859 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4860 if (fd != -1) {
4861 struct stat buf;
4862 ::close(fd);
4863 while (::stat(filename, &buf) == 0) {
4864 (void)::poll(NULL, 0, 100);
4865 }
4866 } else {
4867 jio_fprintf(stderr,
4868 "Could not open pause file '%s', continuing immediately.\n", filename);
4869 }
4870 }
4871
4872
4873 // Refer to the comments in os_solaris.cpp park-unpark.
4874 //
4875 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4876 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4877 // For specifics regarding the bug see GLIBC BUGID 261237 :
4878 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4879 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4880 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4881 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4882 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4883 // and monitorenter when we're using 1-0 locking. All those operations may result in
4884 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4885 // of libpthread avoids the problem, but isn't practical.
4886 //
4887 // Possible remedies:
4888 //
4889 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4890 // This is palliative and probabilistic, however. If the thread is preempted
4891 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4892 // than the minimum period may have passed, and the abstime may be stale (in the
4893 // past) resultin in a hang. Using this technique reduces the odds of a hang
4894 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4895 //
4896 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4897 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4898 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4899 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4900 // thread.
4901 //
4902 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4903 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4904 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4905 // This also works well. In fact it avoids kernel-level scalability impediments
4906 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4907 // timers in a graceful fashion.
4908 //
4909 // 4. When the abstime value is in the past it appears that control returns
4910 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4911 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4912 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4913 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4914 // It may be possible to avoid reinitialization by checking the return
4915 // value from pthread_cond_timedwait(). In addition to reinitializing the
4916 // condvar we must establish the invariant that cond_signal() is only called
4917 // within critical sections protected by the adjunct mutex. This prevents
4918 // cond_signal() from "seeing" a condvar that's in the midst of being
4919 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4920 // desirable signal-after-unlock optimization that avoids futile context switching.
4921 //
4922 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4923 // structure when a condvar is used or initialized. cond_destroy() would
4924 // release the helper structure. Our reinitialize-after-timedwait fix
4925 // put excessive stress on malloc/free and locks protecting the c-heap.
4926 //
4927 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4928 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4929 // and only enabling the work-around for vulnerable environments.
4930
4931 // utility to compute the abstime argument to timedwait:
4932 // millis is the relative timeout time
4933 // abstime will be the absolute timeout time
4934 // TODO: replace compute_abstime() with unpackTime()
4935
4936 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4937 if (millis < 0) millis = 0;
4938 struct timeval now;
4939 int status = gettimeofday(&now, NULL);
4940 assert(status == 0, "gettimeofday");
4941 jlong seconds = millis / 1000;
4942 millis %= 1000;
4943 if (seconds > 50000000) { // see man cond_timedwait(3T)
4944 seconds = 50000000;
4945 }
4946 abstime->tv_sec = now.tv_sec + seconds;
4947 long usec = now.tv_usec + millis * 1000;
4948 if (usec >= 1000000) {
4949 abstime->tv_sec += 1;
4950 usec -= 1000000;
4951 }
4952 abstime->tv_nsec = usec * 1000;
4953 return abstime;
4954 }
4955
4956
4957 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4958 // Conceptually TryPark() should be equivalent to park(0).
4959
4960 int os::PlatformEvent::TryPark() {
4961 for (;;) {
4962 const int v = _Event ;
4963 guarantee ((v == 0) || (v == 1), "invariant") ;
4964 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4965 }
4966 }
4967
4968 void os::PlatformEvent::park() { // AKA "down()"
4969 // Invariant: Only the thread associated with the Event/PlatformEvent
4970 // may call park().
4971 // TODO: assert that _Assoc != NULL or _Assoc == Self
4972 int v ;
4973 for (;;) {
4974 v = _Event ;
4975 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4976 }
4977 guarantee (v >= 0, "invariant") ;
4978 if (v == 0) {
4979 // Do this the hard way by blocking ...
4980 int status = pthread_mutex_lock(_mutex);
4981 assert_status(status == 0, status, "mutex_lock");
4982 guarantee (_nParked == 0, "invariant") ;
4983 ++ _nParked ;
4984 while (_Event < 0) {
4985 status = pthread_cond_wait(_cond, _mutex);
4986 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4987 // Treat this the same as if the wait was interrupted
4988 if (status == ETIME) { status = EINTR; }
4989 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4990 }
4991 -- _nParked ;
4992
4993 _Event = 0 ;
4994 status = pthread_mutex_unlock(_mutex);
4995 assert_status(status == 0, status, "mutex_unlock");
4996 // Paranoia to ensure our locked and lock-free paths interact
4997 // correctly with each other.
4998 OrderAccess::fence();
4999 }
5000 guarantee (_Event >= 0, "invariant") ;
5001 }
5002
5003 int os::PlatformEvent::park(jlong millis) {
5004 guarantee (_nParked == 0, "invariant") ;
5005
5006 int v ;
5007 for (;;) {
5008 v = _Event ;
5009 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5010 }
5011 guarantee (v >= 0, "invariant") ;
5012 if (v != 0) return OS_OK ;
5013
5014 // We do this the hard way, by blocking the thread.
5015 // Consider enforcing a minimum timeout value.
5016 struct timespec abst;
5017 compute_abstime(&abst, millis);
5018
5019 int ret = OS_TIMEOUT;
5020 int status = pthread_mutex_lock(_mutex);
5021 assert_status(status == 0, status, "mutex_lock");
5022 guarantee (_nParked == 0, "invariant") ;
5023 ++_nParked ;
5024
5025 // Object.wait(timo) will return because of
5026 // (a) notification
5027 // (b) timeout
5028 // (c) thread.interrupt
5029 //
5030 // Thread.interrupt and object.notify{All} both call Event::set.
5031 // That is, we treat thread.interrupt as a special case of notification.
5032 // The underlying Solaris implementation, cond_timedwait, admits
5033 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5034 // JVM from making those visible to Java code. As such, we must
5035 // filter out spurious wakeups. We assume all ETIME returns are valid.
5036 //
5037 // TODO: properly differentiate simultaneous notify+interrupt.
5038 // In that case, we should propagate the notify to another waiter.
5039
5040 while (_Event < 0) {
5041 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5042 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5043 pthread_cond_destroy (_cond);
5044 pthread_cond_init (_cond, NULL) ;
5045 }
5046 assert_status(status == 0 || status == EINTR ||
5047 status == ETIME || status == ETIMEDOUT,
5048 status, "cond_timedwait");
5049 if (!FilterSpuriousWakeups) break ; // previous semantics
5050 if (status == ETIME || status == ETIMEDOUT) break ;
5051 // We consume and ignore EINTR and spurious wakeups.
5052 }
5053 --_nParked ;
5054 if (_Event >= 0) {
5055 ret = OS_OK;
5056 }
5057 _Event = 0 ;
5058 status = pthread_mutex_unlock(_mutex);
5059 assert_status(status == 0, status, "mutex_unlock");
5060 assert (_nParked == 0, "invariant") ;
5061 // Paranoia to ensure our locked and lock-free paths interact
5062 // correctly with each other.
5063 OrderAccess::fence();
5064 return ret;
5065 }
5066
5067 void os::PlatformEvent::unpark() {
5068 // Transitions for _Event:
5069 // 0 :=> 1
5070 // 1 :=> 1
5071 // -1 :=> either 0 or 1; must signal target thread
5072 // That is, we can safely transition _Event from -1 to either
5073 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5074 // unpark() calls.
5075 // See also: "Semaphores in Plan 9" by Mullender & Cox
5076 //
5077 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5078 // that it will take two back-to-back park() calls for the owning
5079 // thread to block. This has the benefit of forcing a spurious return
5080 // from the first park() call after an unpark() call which will help
5081 // shake out uses of park() and unpark() without condition variables.
5082
5083 if (Atomic::xchg(1, &_Event) >= 0) return;
5084
5085 // Wait for the thread associated with the event to vacate
5086 int status = pthread_mutex_lock(_mutex);
5087 assert_status(status == 0, status, "mutex_lock");
5088 int AnyWaiters = _nParked;
5089 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5090 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5091 AnyWaiters = 0;
5092 pthread_cond_signal(_cond);
5093 }
5094 status = pthread_mutex_unlock(_mutex);
5095 assert_status(status == 0, status, "mutex_unlock");
5096 if (AnyWaiters != 0) {
5097 status = pthread_cond_signal(_cond);
5098 assert_status(status == 0, status, "cond_signal");
5099 }
5100
5101 // Note that we signal() _after dropping the lock for "immortal" Events.
5102 // This is safe and avoids a common class of futile wakeups. In rare
5103 // circumstances this can cause a thread to return prematurely from
5104 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5105 // simply re-test the condition and re-park itself.
5106 }
5107
5108
5109 // JSR166
5110 // -------------------------------------------------------
5111
5112 /*
5113 * The solaris and linux implementations of park/unpark are fairly
5114 * conservative for now, but can be improved. They currently use a
5115 * mutex/condvar pair, plus a a count.
5116 * Park decrements count if > 0, else does a condvar wait. Unpark
5117 * sets count to 1 and signals condvar. Only one thread ever waits
5118 * on the condvar. Contention seen when trying to park implies that someone
5119 * is unparking you, so don't wait. And spurious returns are fine, so there
5120 * is no need to track notifications.
5121 */
5122
5123 #define MAX_SECS 100000000
5124 /*
5125 * This code is common to linux and solaris and will be moved to a
5126 * common place in dolphin.
5127 *
5128 * The passed in time value is either a relative time in nanoseconds
5129 * or an absolute time in milliseconds. Either way it has to be unpacked
5130 * into suitable seconds and nanoseconds components and stored in the
5131 * given timespec structure.
5132 * Given time is a 64-bit value and the time_t used in the timespec is only
5133 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5134 * overflow if times way in the future are given. Further on Solaris versions
5135 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5136 * number of seconds, in abstime, is less than current_time + 100,000,000.
5137 * As it will be 28 years before "now + 100000000" will overflow we can
5138 * ignore overflow and just impose a hard-limit on seconds using the value
5139 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5140 * years from "now".
5141 */
5142
5143 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5144 assert (time > 0, "convertTime");
5145
5146 struct timeval now;
5147 int status = gettimeofday(&now, NULL);
5148 assert(status == 0, "gettimeofday");
5149
5150 time_t max_secs = now.tv_sec + MAX_SECS;
5151
5152 if (isAbsolute) {
5153 jlong secs = time / 1000;
5154 if (secs > max_secs) {
5155 absTime->tv_sec = max_secs;
5156 }
5157 else {
5158 absTime->tv_sec = secs;
5159 }
5160 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5161 }
5162 else {
5163 jlong secs = time / NANOSECS_PER_SEC;
5164 if (secs >= MAX_SECS) {
5165 absTime->tv_sec = max_secs;
5166 absTime->tv_nsec = 0;
5167 }
5168 else {
5169 absTime->tv_sec = now.tv_sec + secs;
5170 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5171 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5172 absTime->tv_nsec -= NANOSECS_PER_SEC;
5173 ++absTime->tv_sec; // note: this must be <= max_secs
5174 }
5175 }
5176 }
5177 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5178 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5179 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5180 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5181 }
5182
5183 void Parker::park(bool isAbsolute, jlong time) {
5184 // Ideally we'd do something useful while spinning, such
5185 // as calling unpackTime().
5186
5187 // Optional fast-path check:
5188 // Return immediately if a permit is available.
5189 // We depend on Atomic::xchg() having full barrier semantics
5190 // since we are doing a lock-free update to _counter.
5191 if (Atomic::xchg(0, &_counter) > 0) return;
5192
5193 Thread* thread = Thread::current();
5194 assert(thread->is_Java_thread(), "Must be JavaThread");
5195 JavaThread *jt = (JavaThread *)thread;
5196
5197 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5198 // Check interrupt before trying to wait
5199 if (Thread::is_interrupted(thread, false)) {
5200 return;
5201 }
5202
5203 // Next, demultiplex/decode time arguments
5204 timespec absTime;
5205 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5206 return;
5207 }
5208 if (time > 0) {
5209 unpackTime(&absTime, isAbsolute, time);
5210 }
5211
5212
5213 // Enter safepoint region
5214 // Beware of deadlocks such as 6317397.
5215 // The per-thread Parker:: mutex is a classic leaf-lock.
5216 // In particular a thread must never block on the Threads_lock while
5217 // holding the Parker:: mutex. If safepoints are pending both the
5218 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5219 ThreadBlockInVM tbivm(jt);
5220
5221 // Don't wait if cannot get lock since interference arises from
5222 // unblocking. Also. check interrupt before trying wait
5223 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5224 return;
5225 }
5226
5227 int status ;
5228 if (_counter > 0) { // no wait needed
5229 _counter = 0;
5230 status = pthread_mutex_unlock(_mutex);
5231 assert (status == 0, "invariant") ;
5232 // Paranoia to ensure our locked and lock-free paths interact
5233 // correctly with each other and Java-level accesses.
5234 OrderAccess::fence();
5235 return;
5236 }
5237
5238 #ifdef ASSERT
5239 // Don't catch signals while blocked; let the running threads have the signals.
5240 // (This allows a debugger to break into the running thread.)
5241 sigset_t oldsigs;
5242 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5243 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5244 #endif
5245
5246 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5247 jt->set_suspend_equivalent();
5248 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5249
5250 if (time == 0) {
5251 status = pthread_cond_wait (_cond, _mutex) ;
5252 } else {
5253 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5254 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5255 pthread_cond_destroy (_cond) ;
5256 pthread_cond_init (_cond, NULL);
5257 }
5258 }
5259 assert_status(status == 0 || status == EINTR ||
5260 status == ETIME || status == ETIMEDOUT,
5261 status, "cond_timedwait");
5262
5263 #ifdef ASSERT
5264 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5265 #endif
5266
5267 _counter = 0 ;
5268 status = pthread_mutex_unlock(_mutex) ;
5269 assert_status(status == 0, status, "invariant") ;
5270 // Paranoia to ensure our locked and lock-free paths interact
5271 // correctly with each other and Java-level accesses.
5272 OrderAccess::fence();
5273
5274 // If externally suspended while waiting, re-suspend
5275 if (jt->handle_special_suspend_equivalent_condition()) {
5276 jt->java_suspend_self();
5277 }
5278 }
5279
5280 void Parker::unpark() {
5281 int s, status ;
5282 status = pthread_mutex_lock(_mutex);
5283 assert (status == 0, "invariant") ;
5284 s = _counter;
5285 _counter = 1;
5286 if (s < 1) {
5287 if (WorkAroundNPTLTimedWaitHang) {
5288 status = pthread_cond_signal (_cond) ;
5289 assert (status == 0, "invariant") ;
5290 status = pthread_mutex_unlock(_mutex);
5291 assert (status == 0, "invariant") ;
5292 } else {
5293 status = pthread_mutex_unlock(_mutex);
5294 assert (status == 0, "invariant") ;
5295 status = pthread_cond_signal (_cond) ;
5296 assert (status == 0, "invariant") ;
5297 }
5298 } else {
5299 pthread_mutex_unlock(_mutex);
5300 assert (status == 0, "invariant") ;
5301 }
5302 }
5303
5304
5305 extern char** environ;
5306
5307 #ifndef __NR_fork
5308 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5309 #endif
5310
5311 #ifndef __NR_execve
5312 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5313 #endif
5314
5315 // Run the specified command in a separate process. Return its exit value,
5316 // or -1 on failure (e.g. can't fork a new process).
5317 // Unlike system(), this function can be called from signal handler. It
5318 // doesn't block SIGINT et al.
5319 int os::fork_and_exec(char* cmd) {
5320 const char * argv[4] = {"sh", "-c", cmd, NULL};
5321
5322 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5323 // pthread_atfork handlers and reset pthread library. All we need is a
5324 // separate process to execve. Make a direct syscall to fork process.
5325 // On IA64 there's no fork syscall, we have to use fork() and hope for
5326 // the best...
5327 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5328 IA64_ONLY(fork();)
5329
5330 if (pid < 0) {
5331 // fork failed
5332 return -1;
5333
5334 } else if (pid == 0) {
5335 // child process
5336
5337 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5338 // first to kill every thread on the thread list. Because this list is
5339 // not reset by fork() (see notes above), execve() will instead kill
5340 // every thread in the parent process. We know this is the only thread
5341 // in the new process, so make a system call directly.
5342 // IA64 should use normal execve() from glibc to match the glibc fork()
5343 // above.
5344 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5345 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5346
5347 // execve failed
5348 _exit(-1);
5349
5350 } else {
5351 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5352 // care about the actual exit code, for now.
5353
5354 int status;
5355
5356 // Wait for the child process to exit. This returns immediately if
5357 // the child has already exited. */
5358 while (waitpid(pid, &status, 0) < 0) {
5359 switch (errno) {
5360 case ECHILD: return 0;
5361 case EINTR: break;
5362 default: return -1;
5363 }
5364 }
5365
5366 if (WIFEXITED(status)) {
5367 // The child exited normally; get its exit code.
5368 return WEXITSTATUS(status);
5369 } else if (WIFSIGNALED(status)) {
5370 // The child exited because of a signal
5371 // The best value to return is 0x80 + signal number,
5372 // because that is what all Unix shells do, and because
5373 // it allows callers to distinguish between process exit and
5374 // process death by signal.
5375 return 0x80 + WTERMSIG(status);
5376 } else {
5377 // Unknown exit code; pass it through
5378 return status;
5379 }
5380 }
5381 }
5382
5383 // is_headless_jre()
5384 //
5385 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5386 // in order to report if we are running in a headless jre
5387 //
5388 // Since JDK8 xawt/libmawt.so was moved into the same directory
5389 // as libawt.so, and renamed libawt_xawt.so
5390 //
5391 bool os::is_headless_jre() {
5392 struct stat statbuf;
5393 char buf[MAXPATHLEN];
5394 char libmawtpath[MAXPATHLEN];
5395 const char *xawtstr = "/xawt/libmawt.so";
5396 const char *new_xawtstr = "/libawt_xawt.so";
5397 char *p;
5398
5399 // Get path to libjvm.so
5400 os::jvm_path(buf, sizeof(buf));
5401
5402 // Get rid of libjvm.so
5403 p = strrchr(buf, '/');
5404 if (p == NULL) return false;
5405 else *p = '\0';
5406
5407 // Get rid of client or server
5408 p = strrchr(buf, '/');
5409 if (p == NULL) return false;
5410 else *p = '\0';
5411
5412 // check xawt/libmawt.so
5413 strcpy(libmawtpath, buf);
5414 strcat(libmawtpath, xawtstr);
5415 if (::stat(libmawtpath, &statbuf) == 0) return false;
5416
5417 // check libawt_xawt.so
5418 strcpy(libmawtpath, buf);
5419 strcat(libmawtpath, new_xawtstr);
5420 if (::stat(libmawtpath, &statbuf) == 0) return false;
5421
5422 return true;
5423 }
5424
5425 // Get the default path to the core file
5426 // Returns the length of the string
5427 int os::get_core_path(char* buffer, size_t bufferSize) {
5428 const char* p = get_current_directory(buffer, bufferSize);
5429
5430 if (p == NULL) {
5431 assert(p != NULL, "failed to get current directory");
5432 return 0;
5433 }
5434
5435 return strlen(buffer);
5436 }
5437
5438 #ifdef JAVASE_EMBEDDED
5439 //
5440 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5441 //
5442 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5443
5444 // ctor
5445 //
5446 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5447 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5448 _fd = fd;
5449
5450 if (os::create_thread(this, os::os_thread)) {
5451 _memnotify_thread = this;
5452 os::set_priority(this, NearMaxPriority);
5453 os::start_thread(this);
5454 }
5455 }
5456
5457 // Where all the work gets done
5458 //
5459 void MemNotifyThread::run() {
5460 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5461
5462 // Set up the select arguments
5463 fd_set rfds;
5464 if (_fd != -1) {
5465 FD_ZERO(&rfds);
5466 FD_SET(_fd, &rfds);
5467 }
5468
5469 // Now wait for the mem_notify device to wake up
5470 while (1) {
5471 // Wait for the mem_notify device to signal us..
5472 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5473 if (rc == -1) {
5474 perror("select!\n");
5475 break;
5476 } else if (rc) {
5477 //ssize_t free_before = os::available_memory();
5478 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5479
5480 // The kernel is telling us there is not much memory left...
5481 // try to do something about that
5482
5483 // If we are not already in a GC, try one.
5484 if (!Universe::heap()->is_gc_active()) {
5485 Universe::heap()->collect(GCCause::_allocation_failure);
5486
5487 //ssize_t free_after = os::available_memory();
5488 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5489 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5490 }
5491 // We might want to do something like the following if we find the GC's are not helping...
5492 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5493 }
5494 }
5495 }
5496
5497 //
5498 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5499 //
5500 void MemNotifyThread::start() {
5501 int fd;
5502 fd = open ("/dev/mem_notify", O_RDONLY, 0);
5503 if (fd < 0) {
5504 return;
5505 }
5506
5507 if (memnotify_thread() == NULL) {
5508 new MemNotifyThread(fd);
5509 }
5510 }
5511 #endif // JAVASE_EMBEDDED