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