1 /*
2 * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_share_linux.hpp"
40 #include "prims/jniFastGetField.hpp"
41 #include "prims/jvm.h"
42 #include "prims/jvm_misc.hpp"
43 #include "runtime/arguments.hpp"
44 #include "runtime/extendedPC.hpp"
45 #include "runtime/globals.hpp"
46 #include "runtime/interfaceSupport.hpp"
47 #include "runtime/init.hpp"
48 #include "runtime/java.hpp"
49 #include "runtime/javaCalls.hpp"
50 #include "runtime/mutexLocker.hpp"
51 #include "runtime/objectMonitor.hpp"
52 #include "runtime/osThread.hpp"
53 #include "runtime/perfMemory.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #include "runtime/statSampler.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "runtime/thread.inline.hpp"
58 #include "runtime/threadCritical.hpp"
59 #include "runtime/timer.hpp"
60 #include "services/attachListener.hpp"
61 #include "services/memTracker.hpp"
62 #include "services/runtimeService.hpp"
63 #include "utilities/decoder.hpp"
64 #include "utilities/defaultStream.hpp"
65 #include "utilities/events.hpp"
66 #include "utilities/elfFile.hpp"
67 #include "utilities/growableArray.hpp"
68 #include "utilities/vmError.hpp"
69
70 // put OS-includes here
71 # include <sys/types.h>
72 # include <sys/mman.h>
73 # include <sys/stat.h>
74 # include <sys/select.h>
75 # include <pthread.h>
76 # include <signal.h>
77 # include <errno.h>
78 # include <dlfcn.h>
79 # include <stdio.h>
80 # include <unistd.h>
81 # include <sys/resource.h>
82 # include <pthread.h>
83 # include <sys/stat.h>
84 # include <sys/time.h>
85 # include <sys/times.h>
86 # include <sys/utsname.h>
87 # include <sys/socket.h>
88 # include <sys/wait.h>
89 # include <pwd.h>
90 # include <poll.h>
91 # include <semaphore.h>
92 # include <fcntl.h>
93 # include <string.h>
94 # include <syscall.h>
95 # include <sys/sysinfo.h>
96 # include <gnu/libc-version.h>
97 # include <sys/ipc.h>
98 # include <sys/shm.h>
99 # include <link.h>
100 # include <stdint.h>
101 # include <inttypes.h>
102 # include <sys/ioctl.h>
103
104 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
105 // getrusage() is prepared to handle the associated failure.
106 #ifndef RUSAGE_THREAD
107 #define RUSAGE_THREAD (1) /* only the calling thread */
108 #endif
109
110 #define MAX_PATH (2 * K)
111
112 // for timer info max values which include all bits
113 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
114
115 #define LARGEPAGES_BIT (1 << 6)
116 ////////////////////////////////////////////////////////////////////////////////
117 // global variables
118 julong os::Linux::_physical_memory = 0;
119
120 address os::Linux::_initial_thread_stack_bottom = NULL;
121 uintptr_t os::Linux::_initial_thread_stack_size = 0;
122
123 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
124 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
125 Mutex* os::Linux::_createThread_lock = NULL;
126 pthread_t os::Linux::_main_thread;
127 int os::Linux::_page_size = -1;
128 const int os::Linux::_vm_default_page_size = (8 * K);
129 bool os::Linux::_is_floating_stack = false;
130 bool os::Linux::_is_NPTL = false;
131 bool os::Linux::_supports_fast_thread_cpu_time = false;
132 const char * os::Linux::_glibc_version = NULL;
133 const char * os::Linux::_libpthread_version = NULL;
134
135 static jlong initial_time_count=0;
136
137 static int clock_tics_per_sec = 100;
138
139 // For diagnostics to print a message once. see run_periodic_checks
140 static sigset_t check_signal_done;
141 static bool check_signals = true;
142
143 static pid_t _initial_pid = 0;
144
145 /* Signal number used to suspend/resume a thread */
146
147 /* do not use any signal number less than SIGSEGV, see 4355769 */
148 static int SR_signum = SIGUSR2;
149 sigset_t SR_sigset;
150
151 /* Used to protect dlsym() calls */
152 static pthread_mutex_t dl_mutex;
153
154 // Declarations
155 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
156
157 #ifdef JAVASE_EMBEDDED
158 class MemNotifyThread: public Thread {
159 friend class VMStructs;
160 public:
161 virtual void run();
162
163 private:
164 static MemNotifyThread* _memnotify_thread;
165 int _fd;
166
167 public:
168
169 // Constructor
170 MemNotifyThread(int fd);
171
172 // Tester
173 bool is_memnotify_thread() const { return true; }
174
175 // Printing
176 char* name() const { return (char*)"Linux MemNotify Thread"; }
177
178 // Returns the single instance of the MemNotifyThread
179 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
180
181 // Create and start the single instance of MemNotifyThread
182 static void start();
183 };
184 #endif // JAVASE_EMBEDDED
185
186 // utility functions
187
188 static int SR_initialize();
189
190 julong os::available_memory() {
191 return Linux::available_memory();
192 }
193
194 julong os::Linux::available_memory() {
195 // values in struct sysinfo are "unsigned long"
196 struct sysinfo si;
197 sysinfo(&si);
198
199 return (julong)si.freeram * si.mem_unit;
200 }
201
202 julong os::physical_memory() {
203 return Linux::physical_memory();
204 }
205
206 ////////////////////////////////////////////////////////////////////////////////
207 // environment support
208
209 bool os::getenv(const char* name, char* buf, int len) {
210 const char* val = ::getenv(name);
211 if (val != NULL && strlen(val) < (size_t)len) {
212 strcpy(buf, val);
213 return true;
214 }
215 if (len > 0) buf[0] = 0; // return a null string
216 return false;
217 }
218
219
220 // Return true if user is running as root.
221
222 bool os::have_special_privileges() {
223 static bool init = false;
224 static bool privileges = false;
225 if (!init) {
226 privileges = (getuid() != geteuid()) || (getgid() != getegid());
227 init = true;
228 }
229 return privileges;
230 }
231
232
233 #ifndef SYS_gettid
234 // i386: 224, ia64: 1105, amd64: 186, sparc 143
235 #ifdef __ia64__
236 #define SYS_gettid 1105
237 #elif __i386__
238 #define SYS_gettid 224
239 #elif __amd64__
240 #define SYS_gettid 186
241 #elif __sparc__
242 #define SYS_gettid 143
243 #else
244 #error define gettid for the arch
245 #endif
246 #endif
247
248 // Cpu architecture string
249 #if defined(ZERO)
250 static char cpu_arch[] = ZERO_LIBARCH;
251 #elif defined(IA64)
252 static char cpu_arch[] = "ia64";
253 #elif defined(IA32)
254 static char cpu_arch[] = "i386";
255 #elif defined(AMD64)
256 static char cpu_arch[] = "amd64";
257 #elif defined(ARM)
258 static char cpu_arch[] = "arm";
259 #elif defined(PPC32)
260 static char cpu_arch[] = "ppc";
261 #elif defined(PPC64)
262 static char cpu_arch[] = "ppc64";
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 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1159 if (stack_size > 2 * K * K IA64_ONLY(*2))
1160 stack_size = 2 * K * K IA64_ONLY(*2);
1161 // Try to figure out where the stack base (top) is. This is harder.
1162 //
1163 // When an application is started, glibc saves the initial stack pointer in
1164 // a global variable "__libc_stack_end", which is then used by system
1165 // libraries. __libc_stack_end should be pretty close to stack top. The
1166 // variable is available since the very early days. However, because it is
1167 // a private interface, it could disappear in the future.
1168 //
1169 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1170 // to __libc_stack_end, it is very close to stack top, but isn't the real
1171 // stack top. Note that /proc may not exist if VM is running as a chroot
1172 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1173 // /proc/<pid>/stat could change in the future (though unlikely).
1174 //
1175 // We try __libc_stack_end first. If that doesn't work, look for
1176 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1177 // as a hint, which should work well in most cases.
1178
1179 uintptr_t stack_start;
1180
1181 // try __libc_stack_end first
1182 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1183 if (p && *p) {
1184 stack_start = *p;
1185 } else {
1186 // see if we can get the start_stack field from /proc/self/stat
1187 FILE *fp;
1188 int pid;
1189 char state;
1190 int ppid;
1191 int pgrp;
1192 int session;
1193 int nr;
1194 int tpgrp;
1195 unsigned long flags;
1196 unsigned long minflt;
1197 unsigned long cminflt;
1198 unsigned long majflt;
1199 unsigned long cmajflt;
1200 unsigned long utime;
1201 unsigned long stime;
1202 long cutime;
1203 long cstime;
1204 long prio;
1205 long nice;
1206 long junk;
1207 long it_real;
1208 uintptr_t start;
1209 uintptr_t vsize;
1210 intptr_t rss;
1211 uintptr_t rsslim;
1212 uintptr_t scodes;
1213 uintptr_t ecode;
1214 int i;
1215
1216 // Figure what the primordial thread stack base is. Code is inspired
1217 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1218 // followed by command name surrounded by parentheses, state, etc.
1219 char stat[2048];
1220 int statlen;
1221
1222 fp = fopen("/proc/self/stat", "r");
1223 if (fp) {
1224 statlen = fread(stat, 1, 2047, fp);
1225 stat[statlen] = '\0';
1226 fclose(fp);
1227
1228 // Skip pid and the command string. Note that we could be dealing with
1229 // weird command names, e.g. user could decide to rename java launcher
1230 // to "java 1.4.2 :)", then the stat file would look like
1231 // 1234 (java 1.4.2 :)) R ... ...
1232 // We don't really need to know the command string, just find the last
1233 // occurrence of ")" and then start parsing from there. See bug 4726580.
1234 char * s = strrchr(stat, ')');
1235
1236 i = 0;
1237 if (s) {
1238 // Skip blank chars
1239 do s++; while (isspace(*s));
1240
1241 #define _UFM UINTX_FORMAT
1242 #define _DFM INTX_FORMAT
1243
1244 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1245 /* 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 */
1246 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,
1247 &state, /* 3 %c */
1248 &ppid, /* 4 %d */
1249 &pgrp, /* 5 %d */
1250 &session, /* 6 %d */
1251 &nr, /* 7 %d */
1252 &tpgrp, /* 8 %d */
1253 &flags, /* 9 %lu */
1254 &minflt, /* 10 %lu */
1255 &cminflt, /* 11 %lu */
1256 &majflt, /* 12 %lu */
1257 &cmajflt, /* 13 %lu */
1258 &utime, /* 14 %lu */
1259 &stime, /* 15 %lu */
1260 &cutime, /* 16 %ld */
1261 &cstime, /* 17 %ld */
1262 &prio, /* 18 %ld */
1263 &nice, /* 19 %ld */
1264 &junk, /* 20 %ld */
1265 &it_real, /* 21 %ld */
1266 &start, /* 22 UINTX_FORMAT */
1267 &vsize, /* 23 UINTX_FORMAT */
1268 &rss, /* 24 INTX_FORMAT */
1269 &rsslim, /* 25 UINTX_FORMAT */
1270 &scodes, /* 26 UINTX_FORMAT */
1271 &ecode, /* 27 UINTX_FORMAT */
1272 &stack_start); /* 28 UINTX_FORMAT */
1273 }
1274
1275 #undef _UFM
1276 #undef _DFM
1277
1278 if (i != 28 - 2) {
1279 assert(false, "Bad conversion from /proc/self/stat");
1280 // product mode - assume we are the initial thread, good luck in the
1281 // embedded case.
1282 warning("Can't detect initial thread stack location - bad conversion");
1283 stack_start = (uintptr_t) &rlim;
1284 }
1285 } else {
1286 // For some reason we can't open /proc/self/stat (for example, running on
1287 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1288 // most cases, so don't abort:
1289 warning("Can't detect initial thread stack location - no /proc/self/stat");
1290 stack_start = (uintptr_t) &rlim;
1291 }
1292 }
1293
1294 // Now we have a pointer (stack_start) very close to the stack top, the
1295 // next thing to do is to figure out the exact location of stack top. We
1296 // can find out the virtual memory area that contains stack_start by
1297 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1298 // and its upper limit is the real stack top. (again, this would fail if
1299 // running inside chroot, because /proc may not exist.)
1300
1301 uintptr_t stack_top;
1302 address low, high;
1303 if (find_vma((address)stack_start, &low, &high)) {
1304 // success, "high" is the true stack top. (ignore "low", because initial
1305 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1306 stack_top = (uintptr_t)high;
1307 } else {
1308 // failed, likely because /proc/self/maps does not exist
1309 warning("Can't detect initial thread stack location - find_vma failed");
1310 // best effort: stack_start is normally within a few pages below the real
1311 // stack top, use it as stack top, and reduce stack size so we won't put
1312 // guard page outside stack.
1313 stack_top = stack_start;
1314 stack_size -= 16 * page_size();
1315 }
1316
1317 // stack_top could be partially down the page so align it
1318 stack_top = align_size_up(stack_top, page_size());
1319
1320 if (max_size && stack_size > max_size) {
1321 _initial_thread_stack_size = max_size;
1322 } else {
1323 _initial_thread_stack_size = stack_size;
1324 }
1325
1326 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1327 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1328 }
1329
1330 ////////////////////////////////////////////////////////////////////////////////
1331 // time support
1332
1333 // Time since start-up in seconds to a fine granularity.
1334 // Used by VMSelfDestructTimer and the MemProfiler.
1335 double os::elapsedTime() {
1336
1337 return (double)(os::elapsed_counter()) * 0.000001;
1338 }
1339
1340 jlong os::elapsed_counter() {
1341 timeval time;
1342 int status = gettimeofday(&time, NULL);
1343 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1344 }
1345
1346 jlong os::elapsed_frequency() {
1347 return (1000 * 1000);
1348 }
1349
1350 bool os::supports_vtime() { return true; }
1351 bool os::enable_vtime() { return false; }
1352 bool os::vtime_enabled() { return false; }
1353
1354 double os::elapsedVTime() {
1355 struct rusage usage;
1356 int retval = getrusage(RUSAGE_THREAD, &usage);
1357 if (retval == 0) {
1358 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1359 } else {
1360 // better than nothing, but not much
1361 return elapsedTime();
1362 }
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 if (pelements == NULL) {
1652 return false;
1653 }
1654 for (int i = 0 ; i < n ; i++) {
1655 // Really shouldn't be NULL, but check can't hurt
1656 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1657 continue; // skip the empty path values
1658 }
1659 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1660 if (file_exists(buffer)) {
1661 retval = true;
1662 break;
1663 }
1664 }
1665 // release the storage
1666 for (int i = 0 ; i < n ; i++) {
1667 if (pelements[i] != NULL) {
1668 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1669 }
1670 }
1671 if (pelements != NULL) {
1672 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1673 }
1674 } else {
1675 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1676 retval = true;
1677 }
1678 return retval;
1679 }
1680
1681 // check if addr is inside libjvm.so
1682 bool os::address_is_in_vm(address addr) {
1683 static address libjvm_base_addr;
1684 Dl_info dlinfo;
1685
1686 if (libjvm_base_addr == NULL) {
1687 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1688 libjvm_base_addr = (address)dlinfo.dli_fbase;
1689 }
1690 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1691 }
1692
1693 if (dladdr((void *)addr, &dlinfo) != 0) {
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 // buf is not optional, but offset is optional
1703 assert(buf != NULL, "sanity check");
1704
1705 Dl_info dlinfo;
1706
1707 if (dladdr((void*)addr, &dlinfo) != 0) {
1708 // see if we have a matching symbol
1709 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1710 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1711 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1712 }
1713 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1714 return true;
1715 }
1716 // no matching symbol so try for just file info
1717 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1718 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1719 buf, buflen, offset, dlinfo.dli_fname)) {
1720 return true;
1721 }
1722 }
1723 }
1724
1725 buf[0] = '\0';
1726 if (offset != NULL) *offset = -1;
1727 return false;
1728 }
1729
1730 struct _address_to_library_name {
1731 address addr; // input : memory address
1732 size_t buflen; // size of fname
1733 char* fname; // output: library name
1734 address base; // library base addr
1735 };
1736
1737 static int address_to_library_name_callback(struct dl_phdr_info *info,
1738 size_t size, void *data) {
1739 int i;
1740 bool found = false;
1741 address libbase = NULL;
1742 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1743
1744 // iterate through all loadable segments
1745 for (i = 0; i < info->dlpi_phnum; i++) {
1746 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1747 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1748 // base address of a library is the lowest address of its loaded
1749 // segments.
1750 if (libbase == NULL || libbase > segbase) {
1751 libbase = segbase;
1752 }
1753 // see if 'addr' is within current segment
1754 if (segbase <= d->addr &&
1755 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1756 found = true;
1757 }
1758 }
1759 }
1760
1761 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1762 // so dll_address_to_library_name() can fall through to use dladdr() which
1763 // can figure out executable name from argv[0].
1764 if (found && info->dlpi_name && info->dlpi_name[0]) {
1765 d->base = libbase;
1766 if (d->fname) {
1767 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1768 }
1769 return 1;
1770 }
1771 return 0;
1772 }
1773
1774 bool os::dll_address_to_library_name(address addr, char* buf,
1775 int buflen, int* offset) {
1776 // buf is not optional, but offset is optional
1777 assert(buf != NULL, "sanity check");
1778
1779 Dl_info dlinfo;
1780 struct _address_to_library_name data;
1781
1782 // There is a bug in old glibc dladdr() implementation that it could resolve
1783 // to wrong library name if the .so file has a base address != NULL. Here
1784 // we iterate through the program headers of all loaded libraries to find
1785 // out which library 'addr' really belongs to. This workaround can be
1786 // removed once the minimum requirement for glibc is moved to 2.3.x.
1787 data.addr = addr;
1788 data.fname = buf;
1789 data.buflen = buflen;
1790 data.base = NULL;
1791 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1792
1793 if (rslt) {
1794 // buf already contains library name
1795 if (offset) *offset = addr - data.base;
1796 return true;
1797 }
1798 if (dladdr((void*)addr, &dlinfo) != 0) {
1799 if (dlinfo.dli_fname != NULL) {
1800 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1801 }
1802 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1803 *offset = addr - (address)dlinfo.dli_fbase;
1804 }
1805 return true;
1806 }
1807
1808 buf[0] = '\0';
1809 if (offset) *offset = -1;
1810 return false;
1811 }
1812
1813 // Loads .dll/.so and
1814 // in case of error it checks if .dll/.so was built for the
1815 // same architecture as Hotspot is running on
1816
1817
1818 // Remember the stack's state. The Linux dynamic linker will change
1819 // the stack to 'executable' at most once, so we must safepoint only once.
1820 bool os::Linux::_stack_is_executable = false;
1821
1822 // VM operation that loads a library. This is necessary if stack protection
1823 // of the Java stacks can be lost during loading the library. If we
1824 // do not stop the Java threads, they can stack overflow before the stacks
1825 // are protected again.
1826 class VM_LinuxDllLoad: public VM_Operation {
1827 private:
1828 const char *_filename;
1829 char *_ebuf;
1830 int _ebuflen;
1831 void *_lib;
1832 public:
1833 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1834 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1835 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1836 void doit() {
1837 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1838 os::Linux::_stack_is_executable = true;
1839 }
1840 void* loaded_library() { return _lib; }
1841 };
1842
1843 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1844 {
1845 void * result = NULL;
1846 bool load_attempted = false;
1847
1848 // Check whether the library to load might change execution rights
1849 // of the stack. If they are changed, the protection of the stack
1850 // guard pages will be lost. We need a safepoint to fix this.
1851 //
1852 // See Linux man page execstack(8) for more info.
1853 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1854 ElfFile ef(filename);
1855 if (!ef.specifies_noexecstack()) {
1856 if (!is_init_completed()) {
1857 os::Linux::_stack_is_executable = true;
1858 // This is OK - No Java threads have been created yet, and hence no
1859 // stack guard pages to fix.
1860 //
1861 // This should happen only when you are building JDK7 using a very
1862 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1863 //
1864 // Dynamic loader will make all stacks executable after
1865 // this function returns, and will not do that again.
1866 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1867 } else {
1868 warning("You have loaded library %s which might have disabled stack guard. "
1869 "The VM will try to fix the stack guard now.\n"
1870 "It's highly recommended that you fix the library with "
1871 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1872 filename);
1873
1874 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1875 JavaThread *jt = JavaThread::current();
1876 if (jt->thread_state() != _thread_in_native) {
1877 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1878 // that requires ExecStack. Cannot enter safe point. Let's give up.
1879 warning("Unable to fix stack guard. Giving up.");
1880 } else {
1881 if (!LoadExecStackDllInVMThread) {
1882 // This is for the case where the DLL has an static
1883 // constructor function that executes JNI code. We cannot
1884 // load such DLLs in the VMThread.
1885 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1886 }
1887
1888 ThreadInVMfromNative tiv(jt);
1889 debug_only(VMNativeEntryWrapper vew;)
1890
1891 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1892 VMThread::execute(&op);
1893 if (LoadExecStackDllInVMThread) {
1894 result = op.loaded_library();
1895 }
1896 load_attempted = true;
1897 }
1898 }
1899 }
1900 }
1901
1902 if (!load_attempted) {
1903 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1904 }
1905
1906 if (result != NULL) {
1907 // Successful loading
1908 return result;
1909 }
1910
1911 Elf32_Ehdr elf_head;
1912 int diag_msg_max_length=ebuflen-strlen(ebuf);
1913 char* diag_msg_buf=ebuf+strlen(ebuf);
1914
1915 if (diag_msg_max_length==0) {
1916 // No more space in ebuf for additional diagnostics message
1917 return NULL;
1918 }
1919
1920
1921 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1922
1923 if (file_descriptor < 0) {
1924 // Can't open library, report dlerror() message
1925 return NULL;
1926 }
1927
1928 bool failed_to_read_elf_head=
1929 (sizeof(elf_head)!=
1930 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1931
1932 ::close(file_descriptor);
1933 if (failed_to_read_elf_head) {
1934 // file i/o error - report dlerror() msg
1935 return NULL;
1936 }
1937
1938 typedef struct {
1939 Elf32_Half code; // Actual value as defined in elf.h
1940 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1941 char elf_class; // 32 or 64 bit
1942 char endianess; // MSB or LSB
1943 char* name; // String representation
1944 } arch_t;
1945
1946 #ifndef EM_486
1947 #define EM_486 6 /* Intel 80486 */
1948 #endif
1949
1950 static const arch_t arch_array[]={
1951 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1952 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1953 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1954 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1955 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1956 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1957 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1958 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1959 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1960 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1961 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1962 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1963 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1964 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1965 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1966 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1967 };
1968
1969 #if (defined IA32)
1970 static Elf32_Half running_arch_code=EM_386;
1971 #elif (defined AMD64)
1972 static Elf32_Half running_arch_code=EM_X86_64;
1973 #elif (defined IA64)
1974 static Elf32_Half running_arch_code=EM_IA_64;
1975 #elif (defined __sparc) && (defined _LP64)
1976 static Elf32_Half running_arch_code=EM_SPARCV9;
1977 #elif (defined __sparc) && (!defined _LP64)
1978 static Elf32_Half running_arch_code=EM_SPARC;
1979 #elif (defined __powerpc64__)
1980 static Elf32_Half running_arch_code=EM_PPC64;
1981 #elif (defined __powerpc__)
1982 static Elf32_Half running_arch_code=EM_PPC;
1983 #elif (defined ARM)
1984 static Elf32_Half running_arch_code=EM_ARM;
1985 #elif (defined S390)
1986 static Elf32_Half running_arch_code=EM_S390;
1987 #elif (defined ALPHA)
1988 static Elf32_Half running_arch_code=EM_ALPHA;
1989 #elif (defined MIPSEL)
1990 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1991 #elif (defined PARISC)
1992 static Elf32_Half running_arch_code=EM_PARISC;
1993 #elif (defined MIPS)
1994 static Elf32_Half running_arch_code=EM_MIPS;
1995 #elif (defined M68K)
1996 static Elf32_Half running_arch_code=EM_68K;
1997 #else
1998 #error Method os::dll_load requires that one of following is defined:\
1999 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2000 #endif
2001
2002 // Identify compatability class for VM's architecture and library's architecture
2003 // Obtain string descriptions for architectures
2004
2005 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2006 int running_arch_index=-1;
2007
2008 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2009 if (running_arch_code == arch_array[i].code) {
2010 running_arch_index = i;
2011 }
2012 if (lib_arch.code == arch_array[i].code) {
2013 lib_arch.compat_class = arch_array[i].compat_class;
2014 lib_arch.name = arch_array[i].name;
2015 }
2016 }
2017
2018 assert(running_arch_index != -1,
2019 "Didn't find running architecture code (running_arch_code) in arch_array");
2020 if (running_arch_index == -1) {
2021 // Even though running architecture detection failed
2022 // we may still continue with reporting dlerror() message
2023 return NULL;
2024 }
2025
2026 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2027 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2028 return NULL;
2029 }
2030
2031 #ifndef S390
2032 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2033 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2034 return NULL;
2035 }
2036 #endif // !S390
2037
2038 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2039 if ( lib_arch.name!=NULL ) {
2040 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2041 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2042 lib_arch.name, arch_array[running_arch_index].name);
2043 } else {
2044 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2045 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2046 lib_arch.code,
2047 arch_array[running_arch_index].name);
2048 }
2049 }
2050
2051 return NULL;
2052 }
2053
2054 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2055 void * result = ::dlopen(filename, RTLD_LAZY);
2056 if (result == NULL) {
2057 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2058 ebuf[ebuflen-1] = '\0';
2059 }
2060 return result;
2061 }
2062
2063 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2064 void * result = NULL;
2065 if (LoadExecStackDllInVMThread) {
2066 result = dlopen_helper(filename, ebuf, ebuflen);
2067 }
2068
2069 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2070 // library that requires an executable stack, or which does not have this
2071 // stack attribute set, dlopen changes the stack attribute to executable. The
2072 // read protection of the guard pages gets lost.
2073 //
2074 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2075 // may have been queued at the same time.
2076
2077 if (!_stack_is_executable) {
2078 JavaThread *jt = Threads::first();
2079
2080 while (jt) {
2081 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2082 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2083 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2084 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2085 warning("Attempt to reguard stack yellow zone failed.");
2086 }
2087 }
2088 jt = jt->next();
2089 }
2090 }
2091
2092 return result;
2093 }
2094
2095 /*
2096 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2097 * chances are you might want to run the generated bits against glibc-2.0
2098 * libdl.so, so always use locking for any version of glibc.
2099 */
2100 void* os::dll_lookup(void* handle, const char* name) {
2101 pthread_mutex_lock(&dl_mutex);
2102 void* res = dlsym(handle, name);
2103 pthread_mutex_unlock(&dl_mutex);
2104 return res;
2105 }
2106
2107
2108 static bool _print_ascii_file(const char* filename, outputStream* st) {
2109 int fd = ::open(filename, O_RDONLY);
2110 if (fd == -1) {
2111 return false;
2112 }
2113
2114 char buf[32];
2115 int bytes;
2116 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2117 st->print_raw(buf, bytes);
2118 }
2119
2120 ::close(fd);
2121
2122 return true;
2123 }
2124
2125 void os::print_dll_info(outputStream *st) {
2126 st->print_cr("Dynamic libraries:");
2127
2128 char fname[32];
2129 pid_t pid = os::Linux::gettid();
2130
2131 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2132
2133 if (!_print_ascii_file(fname, st)) {
2134 st->print("Can not get library information for pid = %d\n", pid);
2135 }
2136 }
2137
2138 void os::print_os_info_brief(outputStream* st) {
2139 os::Linux::print_distro_info(st);
2140
2141 os::Posix::print_uname_info(st);
2142
2143 os::Linux::print_libversion_info(st);
2144
2145 }
2146
2147 void os::print_os_info(outputStream* st) {
2148 st->print("OS:");
2149
2150 os::Linux::print_distro_info(st);
2151
2152 os::Posix::print_uname_info(st);
2153
2154 // Print warning if unsafe chroot environment detected
2155 if (unsafe_chroot_detected) {
2156 st->print("WARNING!! ");
2157 st->print_cr(unstable_chroot_error);
2158 }
2159
2160 os::Linux::print_libversion_info(st);
2161
2162 os::Posix::print_rlimit_info(st);
2163
2164 os::Posix::print_load_average(st);
2165
2166 os::Linux::print_full_memory_info(st);
2167 }
2168
2169 // Try to identify popular distros.
2170 // Most Linux distributions have /etc/XXX-release file, which contains
2171 // the OS version string. Some have more than one /etc/XXX-release file
2172 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2173 // so the order is important.
2174 void os::Linux::print_distro_info(outputStream* st) {
2175 if (!_print_ascii_file("/etc/mandrake-release", st) &&
2176 !_print_ascii_file("/etc/sun-release", st) &&
2177 !_print_ascii_file("/etc/redhat-release", st) &&
2178 !_print_ascii_file("/etc/SuSE-release", st) &&
2179 !_print_ascii_file("/etc/turbolinux-release", st) &&
2180 !_print_ascii_file("/etc/gentoo-release", st) &&
2181 !_print_ascii_file("/etc/debian_version", st) &&
2182 !_print_ascii_file("/etc/ltib-release", st) &&
2183 !_print_ascii_file("/etc/angstrom-version", st)) {
2184 st->print("Linux");
2185 }
2186 st->cr();
2187 }
2188
2189 void os::Linux::print_libversion_info(outputStream* st) {
2190 // libc, pthread
2191 st->print("libc:");
2192 st->print(os::Linux::glibc_version()); st->print(" ");
2193 st->print(os::Linux::libpthread_version()); st->print(" ");
2194 if (os::Linux::is_LinuxThreads()) {
2195 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2196 }
2197 st->cr();
2198 }
2199
2200 void os::Linux::print_full_memory_info(outputStream* st) {
2201 st->print("\n/proc/meminfo:\n");
2202 _print_ascii_file("/proc/meminfo", st);
2203 st->cr();
2204 }
2205
2206 void os::print_memory_info(outputStream* st) {
2207
2208 st->print("Memory:");
2209 st->print(" %dk page", os::vm_page_size()>>10);
2210
2211 // values in struct sysinfo are "unsigned long"
2212 struct sysinfo si;
2213 sysinfo(&si);
2214
2215 st->print(", physical " UINT64_FORMAT "k",
2216 os::physical_memory() >> 10);
2217 st->print("(" UINT64_FORMAT "k free)",
2218 os::available_memory() >> 10);
2219 st->print(", swap " UINT64_FORMAT "k",
2220 ((jlong)si.totalswap * si.mem_unit) >> 10);
2221 st->print("(" UINT64_FORMAT "k free)",
2222 ((jlong)si.freeswap * si.mem_unit) >> 10);
2223 st->cr();
2224 }
2225
2226 void os::pd_print_cpu_info(outputStream* st) {
2227 st->print("\n/proc/cpuinfo:\n");
2228 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2229 st->print(" <Not Available>");
2230 }
2231 st->cr();
2232 }
2233
2234 void os::print_siginfo(outputStream* st, void* siginfo) {
2235 const siginfo_t* si = (const siginfo_t*)siginfo;
2236
2237 os::Posix::print_siginfo_brief(st, si);
2238
2239 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2240 UseSharedSpaces) {
2241 FileMapInfo* mapinfo = FileMapInfo::current_info();
2242 if (mapinfo->is_in_shared_space(si->si_addr)) {
2243 st->print("\n\nError accessing class data sharing archive." \
2244 " Mapped file inaccessible during execution, " \
2245 " possible disk/network problem.");
2246 }
2247 }
2248 st->cr();
2249 }
2250
2251
2252 static void print_signal_handler(outputStream* st, int sig,
2253 char* buf, size_t buflen);
2254
2255 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2256 st->print_cr("Signal Handlers:");
2257 print_signal_handler(st, SIGSEGV, buf, buflen);
2258 print_signal_handler(st, SIGBUS , buf, buflen);
2259 print_signal_handler(st, SIGFPE , buf, buflen);
2260 print_signal_handler(st, SIGPIPE, buf, buflen);
2261 print_signal_handler(st, SIGXFSZ, buf, buflen);
2262 print_signal_handler(st, SIGILL , buf, buflen);
2263 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2264 print_signal_handler(st, SR_signum, buf, buflen);
2265 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2266 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2267 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2268 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2269 }
2270
2271 static char saved_jvm_path[MAXPATHLEN] = {0};
2272
2273 // Find the full path to the current module, libjvm.so
2274 void os::jvm_path(char *buf, jint buflen) {
2275 // Error checking.
2276 if (buflen < MAXPATHLEN) {
2277 assert(false, "must use a large-enough buffer");
2278 buf[0] = '\0';
2279 return;
2280 }
2281 // Lazy resolve the path to current module.
2282 if (saved_jvm_path[0] != 0) {
2283 strcpy(buf, saved_jvm_path);
2284 return;
2285 }
2286
2287 char dli_fname[MAXPATHLEN];
2288 bool ret = dll_address_to_library_name(
2289 CAST_FROM_FN_PTR(address, os::jvm_path),
2290 dli_fname, sizeof(dli_fname), NULL);
2291 assert(ret, "cannot locate libjvm");
2292 char *rp = NULL;
2293 if (ret && dli_fname[0] != '\0') {
2294 rp = realpath(dli_fname, buf);
2295 }
2296 if (rp == NULL)
2297 return;
2298
2299 if (Arguments::created_by_gamma_launcher()) {
2300 // Support for the gamma launcher. Typical value for buf is
2301 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2302 // the right place in the string, then assume we are installed in a JDK and
2303 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2304 // up the path so it looks like libjvm.so is installed there (append a
2305 // fake suffix hotspot/libjvm.so).
2306 const char *p = buf + strlen(buf) - 1;
2307 for (int count = 0; p > buf && count < 5; ++count) {
2308 for (--p; p > buf && *p != '/'; --p)
2309 /* empty */ ;
2310 }
2311
2312 if (strncmp(p, "/jre/lib/", 9) != 0) {
2313 // Look for JAVA_HOME in the environment.
2314 char* java_home_var = ::getenv("JAVA_HOME");
2315 if (java_home_var != NULL && java_home_var[0] != 0) {
2316 char* jrelib_p;
2317 int len;
2318
2319 // Check the current module name "libjvm.so".
2320 p = strrchr(buf, '/');
2321 assert(strstr(p, "/libjvm") == p, "invalid library name");
2322
2323 rp = realpath(java_home_var, buf);
2324 if (rp == NULL)
2325 return;
2326
2327 // determine if this is a legacy image or modules image
2328 // modules image doesn't have "jre" subdirectory
2329 len = strlen(buf);
2330 jrelib_p = buf + len;
2331 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2332 if (0 != access(buf, F_OK)) {
2333 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2334 }
2335
2336 if (0 == access(buf, F_OK)) {
2337 // Use current module name "libjvm.so"
2338 len = strlen(buf);
2339 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2340 } else {
2341 // Go back to path of .so
2342 rp = realpath(dli_fname, buf);
2343 if (rp == NULL)
2344 return;
2345 }
2346 }
2347 }
2348 }
2349
2350 strcpy(saved_jvm_path, buf);
2351 }
2352
2353 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2354 // no prefix required, not even "_"
2355 }
2356
2357 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2358 // no suffix required
2359 }
2360
2361 ////////////////////////////////////////////////////////////////////////////////
2362 // sun.misc.Signal support
2363
2364 static volatile jint sigint_count = 0;
2365
2366 static void
2367 UserHandler(int sig, void *siginfo, void *context) {
2368 // 4511530 - sem_post is serialized and handled by the manager thread. When
2369 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2370 // don't want to flood the manager thread with sem_post requests.
2371 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2372 return;
2373
2374 // Ctrl-C is pressed during error reporting, likely because the error
2375 // handler fails to abort. Let VM die immediately.
2376 if (sig == SIGINT && is_error_reported()) {
2377 os::die();
2378 }
2379
2380 os::signal_notify(sig);
2381 }
2382
2383 void* os::user_handler() {
2384 return CAST_FROM_FN_PTR(void*, UserHandler);
2385 }
2386
2387 class Semaphore : public StackObj {
2388 public:
2389 Semaphore();
2390 ~Semaphore();
2391 void signal();
2392 void wait();
2393 bool trywait();
2394 bool timedwait(unsigned int sec, int nsec);
2395 private:
2396 sem_t _semaphore;
2397 };
2398
2399
2400 Semaphore::Semaphore() {
2401 sem_init(&_semaphore, 0, 0);
2402 }
2403
2404 Semaphore::~Semaphore() {
2405 sem_destroy(&_semaphore);
2406 }
2407
2408 void Semaphore::signal() {
2409 sem_post(&_semaphore);
2410 }
2411
2412 void Semaphore::wait() {
2413 sem_wait(&_semaphore);
2414 }
2415
2416 bool Semaphore::trywait() {
2417 return sem_trywait(&_semaphore) == 0;
2418 }
2419
2420 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2421 struct timespec ts;
2422 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2423
2424 while (1) {
2425 int result = sem_timedwait(&_semaphore, &ts);
2426 if (result == 0) {
2427 return true;
2428 } else if (errno == EINTR) {
2429 continue;
2430 } else if (errno == ETIMEDOUT) {
2431 return false;
2432 } else {
2433 return false;
2434 }
2435 }
2436 }
2437
2438 extern "C" {
2439 typedef void (*sa_handler_t)(int);
2440 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2441 }
2442
2443 void* os::signal(int signal_number, void* handler) {
2444 struct sigaction sigAct, oldSigAct;
2445
2446 sigfillset(&(sigAct.sa_mask));
2447 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2448 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2449
2450 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2451 // -1 means registration failed
2452 return (void *)-1;
2453 }
2454
2455 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2456 }
2457
2458 void os::signal_raise(int signal_number) {
2459 ::raise(signal_number);
2460 }
2461
2462 /*
2463 * The following code is moved from os.cpp for making this
2464 * code platform specific, which it is by its very nature.
2465 */
2466
2467 // Will be modified when max signal is changed to be dynamic
2468 int os::sigexitnum_pd() {
2469 return NSIG;
2470 }
2471
2472 // a counter for each possible signal value
2473 static volatile jint pending_signals[NSIG+1] = { 0 };
2474
2475 // Linux(POSIX) specific hand shaking semaphore.
2476 static sem_t sig_sem;
2477 static Semaphore sr_semaphore;
2478
2479 void os::signal_init_pd() {
2480 // Initialize signal structures
2481 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2482
2483 // Initialize signal semaphore
2484 ::sem_init(&sig_sem, 0, 0);
2485 }
2486
2487 void os::signal_notify(int sig) {
2488 Atomic::inc(&pending_signals[sig]);
2489 ::sem_post(&sig_sem);
2490 }
2491
2492 static int check_pending_signals(bool wait) {
2493 Atomic::store(0, &sigint_count);
2494 for (;;) {
2495 for (int i = 0; i < NSIG + 1; i++) {
2496 jint n = pending_signals[i];
2497 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2498 return i;
2499 }
2500 }
2501 if (!wait) {
2502 return -1;
2503 }
2504 JavaThread *thread = JavaThread::current();
2505 ThreadBlockInVM tbivm(thread);
2506
2507 bool threadIsSuspended;
2508 do {
2509 thread->set_suspend_equivalent();
2510 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2511 ::sem_wait(&sig_sem);
2512
2513 // were we externally suspended while we were waiting?
2514 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2515 if (threadIsSuspended) {
2516 //
2517 // The semaphore has been incremented, but while we were waiting
2518 // another thread suspended us. We don't want to continue running
2519 // while suspended because that would surprise the thread that
2520 // suspended us.
2521 //
2522 ::sem_post(&sig_sem);
2523
2524 thread->java_suspend_self();
2525 }
2526 } while (threadIsSuspended);
2527 }
2528 }
2529
2530 int os::signal_lookup() {
2531 return check_pending_signals(false);
2532 }
2533
2534 int os::signal_wait() {
2535 return check_pending_signals(true);
2536 }
2537
2538 ////////////////////////////////////////////////////////////////////////////////
2539 // Virtual Memory
2540
2541 int os::vm_page_size() {
2542 // Seems redundant as all get out
2543 assert(os::Linux::page_size() != -1, "must call os::init");
2544 return os::Linux::page_size();
2545 }
2546
2547 // Solaris allocates memory by pages.
2548 int os::vm_allocation_granularity() {
2549 assert(os::Linux::page_size() != -1, "must call os::init");
2550 return os::Linux::page_size();
2551 }
2552
2553 // Rationale behind this function:
2554 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2555 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2556 // samples for JITted code. Here we create private executable mapping over the code cache
2557 // and then we can use standard (well, almost, as mapping can change) way to provide
2558 // info for the reporting script by storing timestamp and location of symbol
2559 void linux_wrap_code(char* base, size_t size) {
2560 static volatile jint cnt = 0;
2561
2562 if (!UseOprofile) {
2563 return;
2564 }
2565
2566 char buf[PATH_MAX+1];
2567 int num = Atomic::add(1, &cnt);
2568
2569 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2570 os::get_temp_directory(), os::current_process_id(), num);
2571 unlink(buf);
2572
2573 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2574
2575 if (fd != -1) {
2576 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2577 if (rv != (off_t)-1) {
2578 if (::write(fd, "", 1) == 1) {
2579 mmap(base, size,
2580 PROT_READ|PROT_WRITE|PROT_EXEC,
2581 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2582 }
2583 }
2584 ::close(fd);
2585 unlink(buf);
2586 }
2587 }
2588
2589 static bool recoverable_mmap_error(int err) {
2590 // See if the error is one we can let the caller handle. This
2591 // list of errno values comes from JBS-6843484. I can't find a
2592 // Linux man page that documents this specific set of errno
2593 // values so while this list currently matches Solaris, it may
2594 // change as we gain experience with this failure mode.
2595 switch (err) {
2596 case EBADF:
2597 case EINVAL:
2598 case ENOTSUP:
2599 // let the caller deal with these errors
2600 return true;
2601
2602 default:
2603 // Any remaining errors on this OS can cause our reserved mapping
2604 // to be lost. That can cause confusion where different data
2605 // structures think they have the same memory mapped. The worst
2606 // scenario is if both the VM and a library think they have the
2607 // same memory mapped.
2608 return false;
2609 }
2610 }
2611
2612 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2613 int err) {
2614 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2615 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2616 strerror(err), err);
2617 }
2618
2619 static void warn_fail_commit_memory(char* addr, size_t size,
2620 size_t alignment_hint, bool exec,
2621 int err) {
2622 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2623 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2624 alignment_hint, exec, strerror(err), err);
2625 }
2626
2627 // NOTE: Linux kernel does not really reserve the pages for us.
2628 // All it does is to check if there are enough free pages
2629 // left at the time of mmap(). This could be a potential
2630 // problem.
2631 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2632 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2633 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2634 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2635 if (res != (uintptr_t) MAP_FAILED) {
2636 if (UseNUMAInterleaving) {
2637 numa_make_global(addr, size);
2638 }
2639 return 0;
2640 }
2641
2642 int err = errno; // save errno from mmap() call above
2643
2644 if (!recoverable_mmap_error(err)) {
2645 warn_fail_commit_memory(addr, size, exec, err);
2646 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2647 }
2648
2649 return err;
2650 }
2651
2652 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2653 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2654 }
2655
2656 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2657 const char* mesg) {
2658 assert(mesg != NULL, "mesg must be specified");
2659 int err = os::Linux::commit_memory_impl(addr, size, exec);
2660 if (err != 0) {
2661 // the caller wants all commit errors to exit with the specified mesg:
2662 warn_fail_commit_memory(addr, size, exec, err);
2663 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2664 }
2665 }
2666
2667 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2668 #ifndef MAP_HUGETLB
2669 #define MAP_HUGETLB 0x40000
2670 #endif
2671
2672 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2673 #ifndef MADV_HUGEPAGE
2674 #define MADV_HUGEPAGE 14
2675 #endif
2676
2677 int os::Linux::commit_memory_impl(char* addr, size_t size,
2678 size_t alignment_hint, bool exec) {
2679 int err;
2680 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2681 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2682 uintptr_t res =
2683 (uintptr_t) ::mmap(addr, size, prot,
2684 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2685 -1, 0);
2686 if (res != (uintptr_t) MAP_FAILED) {
2687 if (UseNUMAInterleaving) {
2688 numa_make_global(addr, size);
2689 }
2690 return 0;
2691 }
2692
2693 err = errno; // save errno from mmap() call above
2694
2695 if (!recoverable_mmap_error(err)) {
2696 // However, it is not clear that this loss of our reserved mapping
2697 // happens with large pages on Linux or that we cannot recover
2698 // from the loss. For now, we just issue a warning and we don't
2699 // call vm_exit_out_of_memory(). This issue is being tracked by
2700 // JBS-8007074.
2701 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2702 // vm_exit_out_of_memory(size, OOM_MMAP_ERROR,
2703 // "committing reserved memory.");
2704 }
2705 // Fall through and try to use small pages
2706 }
2707
2708 err = os::Linux::commit_memory_impl(addr, size, exec);
2709 if (err == 0) {
2710 realign_memory(addr, size, alignment_hint);
2711 }
2712 return err;
2713 }
2714
2715 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2716 bool exec) {
2717 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2718 }
2719
2720 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2721 size_t alignment_hint, bool exec,
2722 const char* mesg) {
2723 assert(mesg != NULL, "mesg must be specified");
2724 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2725 if (err != 0) {
2726 // the caller wants all commit errors to exit with the specified mesg:
2727 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2728 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2729 }
2730 }
2731
2732 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2733 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2734 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2735 // be supported or the memory may already be backed by huge pages.
2736 ::madvise(addr, bytes, MADV_HUGEPAGE);
2737 }
2738 }
2739
2740 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2741 // This method works by doing an mmap over an existing mmaping and effectively discarding
2742 // the existing pages. However it won't work for SHM-based large pages that cannot be
2743 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2744 // small pages on top of the SHM segment. This method always works for small pages, so we
2745 // allow that in any case.
2746 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2747 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2748 }
2749 }
2750
2751 void os::numa_make_global(char *addr, size_t bytes) {
2752 Linux::numa_interleave_memory(addr, bytes);
2753 }
2754
2755 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2756 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2757 }
2758
2759 bool os::numa_topology_changed() { return false; }
2760
2761 size_t os::numa_get_groups_num() {
2762 int max_node = Linux::numa_max_node();
2763 return max_node > 0 ? max_node + 1 : 1;
2764 }
2765
2766 int os::numa_get_group_id() {
2767 int cpu_id = Linux::sched_getcpu();
2768 if (cpu_id != -1) {
2769 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2770 if (lgrp_id != -1) {
2771 return lgrp_id;
2772 }
2773 }
2774 return 0;
2775 }
2776
2777 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2778 for (size_t i = 0; i < size; i++) {
2779 ids[i] = i;
2780 }
2781 return size;
2782 }
2783
2784 bool os::get_page_info(char *start, page_info* info) {
2785 return false;
2786 }
2787
2788 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2789 return end;
2790 }
2791
2792
2793 int os::Linux::sched_getcpu_syscall(void) {
2794 unsigned int cpu;
2795 int retval = -1;
2796
2797 #if defined(IA32)
2798 # ifndef SYS_getcpu
2799 # define SYS_getcpu 318
2800 # endif
2801 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2802 #elif defined(AMD64)
2803 // Unfortunately we have to bring all these macros here from vsyscall.h
2804 // to be able to compile on old linuxes.
2805 # define __NR_vgetcpu 2
2806 # define VSYSCALL_START (-10UL << 20)
2807 # define VSYSCALL_SIZE 1024
2808 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2809 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2810 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2811 retval = vgetcpu(&cpu, NULL, NULL);
2812 #endif
2813
2814 return (retval == -1) ? retval : cpu;
2815 }
2816
2817 // Something to do with the numa-aware allocator needs these symbols
2818 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2819 extern "C" JNIEXPORT void numa_error(char *where) { }
2820 extern "C" JNIEXPORT int fork1() { return fork(); }
2821
2822
2823 // If we are running with libnuma version > 2, then we should
2824 // be trying to use symbols with versions 1.1
2825 // If we are running with earlier version, which did not have symbol versions,
2826 // we should use the base version.
2827 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2828 void *f = dlvsym(handle, name, "libnuma_1.1");
2829 if (f == NULL) {
2830 f = dlsym(handle, name);
2831 }
2832 return f;
2833 }
2834
2835 bool os::Linux::libnuma_init() {
2836 // sched_getcpu() should be in libc.
2837 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2838 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2839
2840 // If it's not, try a direct syscall.
2841 if (sched_getcpu() == -1)
2842 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2843
2844 if (sched_getcpu() != -1) { // Does it work?
2845 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2846 if (handle != NULL) {
2847 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2848 libnuma_dlsym(handle, "numa_node_to_cpus")));
2849 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2850 libnuma_dlsym(handle, "numa_max_node")));
2851 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2852 libnuma_dlsym(handle, "numa_available")));
2853 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2854 libnuma_dlsym(handle, "numa_tonode_memory")));
2855 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2856 libnuma_dlsym(handle, "numa_interleave_memory")));
2857
2858
2859 if (numa_available() != -1) {
2860 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2861 // Create a cpu -> node mapping
2862 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2863 rebuild_cpu_to_node_map();
2864 return true;
2865 }
2866 }
2867 }
2868 return false;
2869 }
2870
2871 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2872 // The table is later used in get_node_by_cpu().
2873 void os::Linux::rebuild_cpu_to_node_map() {
2874 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2875 // in libnuma (possible values are starting from 16,
2876 // and continuing up with every other power of 2, but less
2877 // than the maximum number of CPUs supported by kernel), and
2878 // is a subject to change (in libnuma version 2 the requirements
2879 // are more reasonable) we'll just hardcode the number they use
2880 // in the library.
2881 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2882
2883 size_t cpu_num = os::active_processor_count();
2884 size_t cpu_map_size = NCPUS / BitsPerCLong;
2885 size_t cpu_map_valid_size =
2886 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2887
2888 cpu_to_node()->clear();
2889 cpu_to_node()->at_grow(cpu_num - 1);
2890 size_t node_num = numa_get_groups_num();
2891
2892 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2893 for (size_t i = 0; i < node_num; i++) {
2894 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2895 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2896 if (cpu_map[j] != 0) {
2897 for (size_t k = 0; k < BitsPerCLong; k++) {
2898 if (cpu_map[j] & (1UL << k)) {
2899 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2900 }
2901 }
2902 }
2903 }
2904 }
2905 }
2906 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2907 }
2908
2909 int os::Linux::get_node_by_cpu(int cpu_id) {
2910 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2911 return cpu_to_node()->at(cpu_id);
2912 }
2913 return -1;
2914 }
2915
2916 GrowableArray<int>* os::Linux::_cpu_to_node;
2917 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2918 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2919 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2920 os::Linux::numa_available_func_t os::Linux::_numa_available;
2921 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2922 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2923 unsigned long* os::Linux::_numa_all_nodes;
2924
2925 bool os::pd_uncommit_memory(char* addr, size_t size) {
2926 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2927 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2928 return res != (uintptr_t) MAP_FAILED;
2929 }
2930
2931 // Linux uses a growable mapping for the stack, and if the mapping for
2932 // the stack guard pages is not removed when we detach a thread the
2933 // stack cannot grow beyond the pages where the stack guard was
2934 // mapped. If at some point later in the process the stack expands to
2935 // that point, the Linux kernel cannot expand the stack any further
2936 // because the guard pages are in the way, and a segfault occurs.
2937 //
2938 // However, it's essential not to split the stack region by unmapping
2939 // a region (leaving a hole) that's already part of the stack mapping,
2940 // so if the stack mapping has already grown beyond the guard pages at
2941 // the time we create them, we have to truncate the stack mapping.
2942 // So, we need to know the extent of the stack mapping when
2943 // create_stack_guard_pages() is called.
2944
2945 // Find the bounds of the stack mapping. Return true for success.
2946 //
2947 // We only need this for stacks that are growable: at the time of
2948 // writing thread stacks don't use growable mappings (i.e. those
2949 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2950 // only applies to the main thread.
2951
2952 static
2953 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2954
2955 char buf[128];
2956 int fd, sz;
2957
2958 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2959 return false;
2960 }
2961
2962 const char kw[] = "[stack]";
2963 const int kwlen = sizeof(kw)-1;
2964
2965 // Address part of /proc/self/maps couldn't be more than 128 bytes
2966 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2967 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2968 // Extract addresses
2969 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2970 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2971 if (sp >= *bottom && sp <= *top) {
2972 ::close(fd);
2973 return true;
2974 }
2975 }
2976 }
2977 }
2978
2979 ::close(fd);
2980 return false;
2981 }
2982
2983
2984 // If the (growable) stack mapping already extends beyond the point
2985 // where we're going to put our guard pages, truncate the mapping at
2986 // that point by munmap()ping it. This ensures that when we later
2987 // munmap() the guard pages we don't leave a hole in the stack
2988 // mapping. This only affects the main/initial thread, but guard
2989 // against future OS changes
2990 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2991 uintptr_t stack_extent, stack_base;
2992 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2993 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2994 assert(os::Linux::is_initial_thread(),
2995 "growable stack in non-initial thread");
2996 if (stack_extent < (uintptr_t)addr)
2997 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2998 }
2999
3000 return os::commit_memory(addr, size, !ExecMem);
3001 }
3002
3003 // If this is a growable mapping, remove the guard pages entirely by
3004 // munmap()ping them. If not, just call uncommit_memory(). This only
3005 // affects the main/initial thread, but guard against future OS changes
3006 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3007 uintptr_t stack_extent, stack_base;
3008 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
3009 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
3010 assert(os::Linux::is_initial_thread(),
3011 "growable stack in non-initial thread");
3012
3013 return ::munmap(addr, size) == 0;
3014 }
3015
3016 return os::uncommit_memory(addr, size);
3017 }
3018
3019 static address _highest_vm_reserved_address = NULL;
3020
3021 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3022 // at 'requested_addr'. If there are existing memory mappings at the same
3023 // location, however, they will be overwritten. If 'fixed' is false,
3024 // 'requested_addr' is only treated as a hint, the return value may or
3025 // may not start from the requested address. Unlike Linux mmap(), this
3026 // function returns NULL to indicate failure.
3027 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3028 char * addr;
3029 int flags;
3030
3031 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3032 if (fixed) {
3033 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3034 flags |= MAP_FIXED;
3035 }
3036
3037 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3038 // touch an uncommitted page. Otherwise, the read/write might
3039 // succeed if we have enough swap space to back the physical page.
3040 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3041 flags, -1, 0);
3042
3043 if (addr != MAP_FAILED) {
3044 // anon_mmap() should only get called during VM initialization,
3045 // don't need lock (actually we can skip locking even it can be called
3046 // from multiple threads, because _highest_vm_reserved_address is just a
3047 // hint about the upper limit of non-stack memory regions.)
3048 if ((address)addr + bytes > _highest_vm_reserved_address) {
3049 _highest_vm_reserved_address = (address)addr + bytes;
3050 }
3051 }
3052
3053 return addr == MAP_FAILED ? NULL : addr;
3054 }
3055
3056 // Don't update _highest_vm_reserved_address, because there might be memory
3057 // regions above addr + size. If so, releasing a memory region only creates
3058 // a hole in the address space, it doesn't help prevent heap-stack collision.
3059 //
3060 static int anon_munmap(char * addr, size_t size) {
3061 return ::munmap(addr, size) == 0;
3062 }
3063
3064 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3065 size_t alignment_hint) {
3066 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3067 }
3068
3069 bool os::pd_release_memory(char* addr, size_t size) {
3070 return anon_munmap(addr, size);
3071 }
3072
3073 static address highest_vm_reserved_address() {
3074 return _highest_vm_reserved_address;
3075 }
3076
3077 static bool linux_mprotect(char* addr, size_t size, int prot) {
3078 // Linux wants the mprotect address argument to be page aligned.
3079 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3080
3081 // According to SUSv3, mprotect() should only be used with mappings
3082 // established by mmap(), and mmap() always maps whole pages. Unaligned
3083 // 'addr' likely indicates problem in the VM (e.g. trying to change
3084 // protection of malloc'ed or statically allocated memory). Check the
3085 // caller if you hit this assert.
3086 assert(addr == bottom, "sanity check");
3087
3088 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3089 return ::mprotect(bottom, size, prot) == 0;
3090 }
3091
3092 // Set protections specified
3093 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3094 bool is_committed) {
3095 unsigned int p = 0;
3096 switch (prot) {
3097 case MEM_PROT_NONE: p = PROT_NONE; break;
3098 case MEM_PROT_READ: p = PROT_READ; break;
3099 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3100 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3101 default:
3102 ShouldNotReachHere();
3103 }
3104 // is_committed is unused.
3105 return linux_mprotect(addr, bytes, p);
3106 }
3107
3108 bool os::guard_memory(char* addr, size_t size) {
3109 return linux_mprotect(addr, size, PROT_NONE);
3110 }
3111
3112 bool os::unguard_memory(char* addr, size_t size) {
3113 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3114 }
3115
3116 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3117 bool result = false;
3118 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
3119 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3120 -1, 0);
3121
3122 if (p != MAP_FAILED) {
3123 // We don't know if this really is a huge page or not.
3124 FILE *fp = fopen("/proc/self/maps", "r");
3125 if (fp) {
3126 while (!feof(fp)) {
3127 char chars[257];
3128 long x = 0;
3129 if (fgets(chars, sizeof(chars), fp)) {
3130 if (sscanf(chars, "%lx-%*x", &x) == 1
3131 && x == (long)p) {
3132 if (strstr (chars, "hugepage")) {
3133 result = true;
3134 break;
3135 }
3136 }
3137 }
3138 }
3139 fclose(fp);
3140 }
3141 munmap (p, page_size);
3142 if (result)
3143 return true;
3144 }
3145
3146 if (warn) {
3147 warning("HugeTLBFS is not supported by the operating system.");
3148 }
3149
3150 return result;
3151 }
3152
3153 /*
3154 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3155 *
3156 * From the coredump_filter documentation:
3157 *
3158 * - (bit 0) anonymous private memory
3159 * - (bit 1) anonymous shared memory
3160 * - (bit 2) file-backed private memory
3161 * - (bit 3) file-backed shared memory
3162 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3163 * effective only if the bit 2 is cleared)
3164 * - (bit 5) hugetlb private memory
3165 * - (bit 6) hugetlb shared memory
3166 */
3167 static void set_coredump_filter(void) {
3168 FILE *f;
3169 long cdm;
3170
3171 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3172 return;
3173 }
3174
3175 if (fscanf(f, "%lx", &cdm) != 1) {
3176 fclose(f);
3177 return;
3178 }
3179
3180 rewind(f);
3181
3182 if ((cdm & LARGEPAGES_BIT) == 0) {
3183 cdm |= LARGEPAGES_BIT;
3184 fprintf(f, "%#lx", cdm);
3185 }
3186
3187 fclose(f);
3188 }
3189
3190 // Large page support
3191
3192 static size_t _large_page_size = 0;
3193
3194 void os::large_page_init() {
3195 if (!UseLargePages) {
3196 UseHugeTLBFS = false;
3197 UseSHM = false;
3198 return;
3199 }
3200
3201 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3202 // If UseLargePages is specified on the command line try both methods,
3203 // if it's default, then try only HugeTLBFS.
3204 if (FLAG_IS_DEFAULT(UseLargePages)) {
3205 UseHugeTLBFS = true;
3206 } else {
3207 UseHugeTLBFS = UseSHM = true;
3208 }
3209 }
3210
3211 if (LargePageSizeInBytes) {
3212 _large_page_size = LargePageSizeInBytes;
3213 } else {
3214 // large_page_size on Linux is used to round up heap size. x86 uses either
3215 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3216 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3217 // page as large as 256M.
3218 //
3219 // Here we try to figure out page size by parsing /proc/meminfo and looking
3220 // for a line with the following format:
3221 // Hugepagesize: 2048 kB
3222 //
3223 // If we can't determine the value (e.g. /proc is not mounted, or the text
3224 // format has been changed), we'll use the largest page size supported by
3225 // the processor.
3226
3227 #ifndef ZERO
3228 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3229 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3230 #endif // ZERO
3231
3232 FILE *fp = fopen("/proc/meminfo", "r");
3233 if (fp) {
3234 while (!feof(fp)) {
3235 int x = 0;
3236 char buf[16];
3237 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3238 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3239 _large_page_size = x * K;
3240 break;
3241 }
3242 } else {
3243 // skip to next line
3244 for (;;) {
3245 int ch = fgetc(fp);
3246 if (ch == EOF || ch == (int)'\n') break;
3247 }
3248 }
3249 }
3250 fclose(fp);
3251 }
3252 }
3253
3254 // print a warning if any large page related flag is specified on command line
3255 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3256
3257 const size_t default_page_size = (size_t)Linux::page_size();
3258 if (_large_page_size > default_page_size) {
3259 _page_sizes[0] = _large_page_size;
3260 _page_sizes[1] = default_page_size;
3261 _page_sizes[2] = 0;
3262 }
3263 UseHugeTLBFS = UseHugeTLBFS &&
3264 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3265
3266 if (UseHugeTLBFS)
3267 UseSHM = false;
3268
3269 UseLargePages = UseHugeTLBFS || UseSHM;
3270
3271 set_coredump_filter();
3272 }
3273
3274 #ifndef SHM_HUGETLB
3275 #define SHM_HUGETLB 04000
3276 #endif
3277
3278 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3279 // "exec" is passed in but not used. Creating the shared image for
3280 // the code cache doesn't have an SHM_X executable permission to check.
3281 assert(UseLargePages && UseSHM, "only for SHM large pages");
3282
3283 key_t key = IPC_PRIVATE;
3284 char *addr;
3285
3286 bool warn_on_failure = UseLargePages &&
3287 (!FLAG_IS_DEFAULT(UseLargePages) ||
3288 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3289 );
3290 char msg[128];
3291
3292 // Create a large shared memory region to attach to based on size.
3293 // Currently, size is the total size of the heap
3294 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3295 if (shmid == -1) {
3296 // Possible reasons for shmget failure:
3297 // 1. shmmax is too small for Java heap.
3298 // > check shmmax value: cat /proc/sys/kernel/shmmax
3299 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3300 // 2. not enough large page memory.
3301 // > check available large pages: cat /proc/meminfo
3302 // > increase amount of large pages:
3303 // echo new_value > /proc/sys/vm/nr_hugepages
3304 // Note 1: different Linux may use different name for this property,
3305 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3306 // Note 2: it's possible there's enough physical memory available but
3307 // they are so fragmented after a long run that they can't
3308 // coalesce into large pages. Try to reserve large pages when
3309 // the system is still "fresh".
3310 if (warn_on_failure) {
3311 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3312 warning(msg);
3313 }
3314 return NULL;
3315 }
3316
3317 // attach to the region
3318 addr = (char*)shmat(shmid, req_addr, 0);
3319 int err = errno;
3320
3321 // Remove shmid. If shmat() is successful, the actual shared memory segment
3322 // will be deleted when it's detached by shmdt() or when the process
3323 // terminates. If shmat() is not successful this will remove the shared
3324 // segment immediately.
3325 shmctl(shmid, IPC_RMID, NULL);
3326
3327 if ((intptr_t)addr == -1) {
3328 if (warn_on_failure) {
3329 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3330 warning(msg);
3331 }
3332 return NULL;
3333 }
3334
3335 if ((addr != NULL) && UseNUMAInterleaving) {
3336 numa_make_global(addr, bytes);
3337 }
3338
3339 // The memory is committed
3340 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3341
3342 return addr;
3343 }
3344
3345 bool os::release_memory_special(char* base, size_t bytes) {
3346 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3347 // detaching the SHM segment will also delete it, see reserve_memory_special()
3348 int rslt = shmdt(base);
3349 if (rslt == 0) {
3350 tkr.record((address)base, bytes);
3351 return true;
3352 } else {
3353 tkr.discard();
3354 return false;
3355 }
3356 }
3357
3358 size_t os::large_page_size() {
3359 return _large_page_size;
3360 }
3361
3362 // HugeTLBFS allows application to commit large page memory on demand;
3363 // with SysV SHM the entire memory region must be allocated as shared
3364 // memory.
3365 bool os::can_commit_large_page_memory() {
3366 return UseHugeTLBFS;
3367 }
3368
3369 bool os::can_execute_large_page_memory() {
3370 return UseHugeTLBFS;
3371 }
3372
3373 // Reserve memory at an arbitrary address, only if that area is
3374 // available (and not reserved for something else).
3375
3376 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3377 const int max_tries = 10;
3378 char* base[max_tries];
3379 size_t size[max_tries];
3380 const size_t gap = 0x000000;
3381
3382 // Assert only that the size is a multiple of the page size, since
3383 // that's all that mmap requires, and since that's all we really know
3384 // about at this low abstraction level. If we need higher alignment,
3385 // we can either pass an alignment to this method or verify alignment
3386 // in one of the methods further up the call chain. See bug 5044738.
3387 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3388
3389 // Repeatedly allocate blocks until the block is allocated at the
3390 // right spot. Give up after max_tries. Note that reserve_memory() will
3391 // automatically update _highest_vm_reserved_address if the call is
3392 // successful. The variable tracks the highest memory address every reserved
3393 // by JVM. It is used to detect heap-stack collision if running with
3394 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3395 // space than needed, it could confuse the collision detecting code. To
3396 // solve the problem, save current _highest_vm_reserved_address and
3397 // calculate the correct value before return.
3398 address old_highest = _highest_vm_reserved_address;
3399
3400 // Linux mmap allows caller to pass an address as hint; give it a try first,
3401 // if kernel honors the hint then we can return immediately.
3402 char * addr = anon_mmap(requested_addr, bytes, false);
3403 if (addr == requested_addr) {
3404 return requested_addr;
3405 }
3406
3407 if (addr != NULL) {
3408 // mmap() is successful but it fails to reserve at the requested address
3409 anon_munmap(addr, bytes);
3410 }
3411
3412 int i;
3413 for (i = 0; i < max_tries; ++i) {
3414 base[i] = reserve_memory(bytes);
3415
3416 if (base[i] != NULL) {
3417 // Is this the block we wanted?
3418 if (base[i] == requested_addr) {
3419 size[i] = bytes;
3420 break;
3421 }
3422
3423 // Does this overlap the block we wanted? Give back the overlapped
3424 // parts and try again.
3425
3426 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3427 if (top_overlap >= 0 && top_overlap < bytes) {
3428 unmap_memory(base[i], top_overlap);
3429 base[i] += top_overlap;
3430 size[i] = bytes - top_overlap;
3431 } else {
3432 size_t bottom_overlap = base[i] + bytes - requested_addr;
3433 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3434 unmap_memory(requested_addr, bottom_overlap);
3435 size[i] = bytes - bottom_overlap;
3436 } else {
3437 size[i] = bytes;
3438 }
3439 }
3440 }
3441 }
3442
3443 // Give back the unused reserved pieces.
3444
3445 for (int j = 0; j < i; ++j) {
3446 if (base[j] != NULL) {
3447 unmap_memory(base[j], size[j]);
3448 }
3449 }
3450
3451 if (i < max_tries) {
3452 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3453 return requested_addr;
3454 } else {
3455 _highest_vm_reserved_address = old_highest;
3456 return NULL;
3457 }
3458 }
3459
3460 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3461 return ::read(fd, buf, nBytes);
3462 }
3463
3464 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3465 // Solaris uses poll(), linux uses park().
3466 // Poll() is likely a better choice, assuming that Thread.interrupt()
3467 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3468 // SIGSEGV, see 4355769.
3469
3470 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3471 assert(thread == Thread::current(), "thread consistency check");
3472
3473 ParkEvent * const slp = thread->_SleepEvent ;
3474 slp->reset() ;
3475 OrderAccess::fence() ;
3476
3477 if (interruptible) {
3478 jlong prevtime = javaTimeNanos();
3479
3480 for (;;) {
3481 if (os::is_interrupted(thread, true)) {
3482 return OS_INTRPT;
3483 }
3484
3485 jlong newtime = javaTimeNanos();
3486
3487 if (newtime - prevtime < 0) {
3488 // time moving backwards, should only happen if no monotonic clock
3489 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3490 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3491 } else {
3492 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3493 }
3494
3495 if(millis <= 0) {
3496 return OS_OK;
3497 }
3498
3499 prevtime = newtime;
3500
3501 {
3502 assert(thread->is_Java_thread(), "sanity check");
3503 JavaThread *jt = (JavaThread *) thread;
3504 ThreadBlockInVM tbivm(jt);
3505 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3506
3507 jt->set_suspend_equivalent();
3508 // cleared by handle_special_suspend_equivalent_condition() or
3509 // java_suspend_self() via check_and_wait_while_suspended()
3510
3511 slp->park(millis);
3512
3513 // were we externally suspended while we were waiting?
3514 jt->check_and_wait_while_suspended();
3515 }
3516 }
3517 } else {
3518 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3519 jlong prevtime = javaTimeNanos();
3520
3521 for (;;) {
3522 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3523 // the 1st iteration ...
3524 jlong newtime = javaTimeNanos();
3525
3526 if (newtime - prevtime < 0) {
3527 // time moving backwards, should only happen if no monotonic clock
3528 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3529 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3530 } else {
3531 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3532 }
3533
3534 if(millis <= 0) break ;
3535
3536 prevtime = newtime;
3537 slp->park(millis);
3538 }
3539 return OS_OK ;
3540 }
3541 }
3542
3543 int os::naked_sleep() {
3544 // %% make the sleep time an integer flag. for now use 1 millisec.
3545 return os::sleep(Thread::current(), 1, false);
3546 }
3547
3548 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3549 void os::infinite_sleep() {
3550 while (true) { // sleep forever ...
3551 ::sleep(100); // ... 100 seconds at a time
3552 }
3553 }
3554
3555 // Used to convert frequent JVM_Yield() to nops
3556 bool os::dont_yield() {
3557 return DontYieldALot;
3558 }
3559
3560 void os::yield() {
3561 sched_yield();
3562 }
3563
3564 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3565
3566 void os::yield_all(int attempts) {
3567 // Yields to all threads, including threads with lower priorities
3568 // Threads on Linux are all with same priority. The Solaris style
3569 // os::yield_all() with nanosleep(1ms) is not necessary.
3570 sched_yield();
3571 }
3572
3573 // Called from the tight loops to possibly influence time-sharing heuristics
3574 void os::loop_breaker(int attempts) {
3575 os::yield_all(attempts);
3576 }
3577
3578 ////////////////////////////////////////////////////////////////////////////////
3579 // thread priority support
3580
3581 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3582 // only supports dynamic priority, static priority must be zero. For real-time
3583 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3584 // However, for large multi-threaded applications, SCHED_RR is not only slower
3585 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3586 // of 5 runs - Sep 2005).
3587 //
3588 // The following code actually changes the niceness of kernel-thread/LWP. It
3589 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3590 // not the entire user process, and user level threads are 1:1 mapped to kernel
3591 // threads. It has always been the case, but could change in the future. For
3592 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3593 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3594
3595 int os::java_to_os_priority[CriticalPriority + 1] = {
3596 19, // 0 Entry should never be used
3597
3598 4, // 1 MinPriority
3599 3, // 2
3600 2, // 3
3601
3602 1, // 4
3603 0, // 5 NormPriority
3604 -1, // 6
3605
3606 -2, // 7
3607 -3, // 8
3608 -4, // 9 NearMaxPriority
3609
3610 -5, // 10 MaxPriority
3611
3612 -5 // 11 CriticalPriority
3613 };
3614
3615 static int prio_init() {
3616 if (ThreadPriorityPolicy == 1) {
3617 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3618 // if effective uid is not root. Perhaps, a more elegant way of doing
3619 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3620 if (geteuid() != 0) {
3621 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3622 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3623 }
3624 ThreadPriorityPolicy = 0;
3625 }
3626 }
3627 if (UseCriticalJavaThreadPriority) {
3628 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3629 }
3630 return 0;
3631 }
3632
3633 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3634 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3635
3636 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3637 return (ret == 0) ? OS_OK : OS_ERR;
3638 }
3639
3640 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3641 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3642 *priority_ptr = java_to_os_priority[NormPriority];
3643 return OS_OK;
3644 }
3645
3646 errno = 0;
3647 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3648 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3649 }
3650
3651 // Hint to the underlying OS that a task switch would not be good.
3652 // Void return because it's a hint and can fail.
3653 void os::hint_no_preempt() {}
3654
3655 ////////////////////////////////////////////////////////////////////////////////
3656 // suspend/resume support
3657
3658 // the low-level signal-based suspend/resume support is a remnant from the
3659 // old VM-suspension that used to be for java-suspension, safepoints etc,
3660 // within hotspot. Now there is a single use-case for this:
3661 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3662 // that runs in the watcher thread.
3663 // The remaining code is greatly simplified from the more general suspension
3664 // code that used to be used.
3665 //
3666 // The protocol is quite simple:
3667 // - suspend:
3668 // - sends a signal to the target thread
3669 // - polls the suspend state of the osthread using a yield loop
3670 // - target thread signal handler (SR_handler) sets suspend state
3671 // and blocks in sigsuspend until continued
3672 // - resume:
3673 // - sets target osthread state to continue
3674 // - sends signal to end the sigsuspend loop in the SR_handler
3675 //
3676 // Note that the SR_lock plays no role in this suspend/resume protocol.
3677 //
3678
3679 static void resume_clear_context(OSThread *osthread) {
3680 osthread->set_ucontext(NULL);
3681 osthread->set_siginfo(NULL);
3682 }
3683
3684 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3685 osthread->set_ucontext(context);
3686 osthread->set_siginfo(siginfo);
3687 }
3688
3689 //
3690 // Handler function invoked when a thread's execution is suspended or
3691 // resumed. We have to be careful that only async-safe functions are
3692 // called here (Note: most pthread functions are not async safe and
3693 // should be avoided.)
3694 //
3695 // Note: sigwait() is a more natural fit than sigsuspend() from an
3696 // interface point of view, but sigwait() prevents the signal hander
3697 // from being run. libpthread would get very confused by not having
3698 // its signal handlers run and prevents sigwait()'s use with the
3699 // mutex granting granting signal.
3700 //
3701 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3702 //
3703 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3704 // Save and restore errno to avoid confusing native code with EINTR
3705 // after sigsuspend.
3706 int old_errno = errno;
3707
3708 Thread* thread = Thread::current();
3709 OSThread* osthread = thread->osthread();
3710 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3711
3712 os::SuspendResume::State current = osthread->sr.state();
3713 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3714 suspend_save_context(osthread, siginfo, context);
3715
3716 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3717 os::SuspendResume::State state = osthread->sr.suspended();
3718 if (state == os::SuspendResume::SR_SUSPENDED) {
3719 sigset_t suspend_set; // signals for sigsuspend()
3720
3721 // get current set of blocked signals and unblock resume signal
3722 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3723 sigdelset(&suspend_set, SR_signum);
3724
3725 sr_semaphore.signal();
3726 // wait here until we are resumed
3727 while (1) {
3728 sigsuspend(&suspend_set);
3729
3730 os::SuspendResume::State result = osthread->sr.running();
3731 if (result == os::SuspendResume::SR_RUNNING) {
3732 sr_semaphore.signal();
3733 break;
3734 }
3735 }
3736
3737 } else if (state == os::SuspendResume::SR_RUNNING) {
3738 // request was cancelled, continue
3739 } else {
3740 ShouldNotReachHere();
3741 }
3742
3743 resume_clear_context(osthread);
3744 } else if (current == os::SuspendResume::SR_RUNNING) {
3745 // request was cancelled, continue
3746 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3747 // ignore
3748 } else {
3749 // ignore
3750 }
3751
3752 errno = old_errno;
3753 }
3754
3755
3756 static int SR_initialize() {
3757 struct sigaction act;
3758 char *s;
3759 /* Get signal number to use for suspend/resume */
3760 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3761 int sig = ::strtol(s, 0, 10);
3762 if (sig > 0 || sig < _NSIG) {
3763 SR_signum = sig;
3764 }
3765 }
3766
3767 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3768 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3769
3770 sigemptyset(&SR_sigset);
3771 sigaddset(&SR_sigset, SR_signum);
3772
3773 /* Set up signal handler for suspend/resume */
3774 act.sa_flags = SA_RESTART|SA_SIGINFO;
3775 act.sa_handler = (void (*)(int)) SR_handler;
3776
3777 // SR_signum is blocked by default.
3778 // 4528190 - We also need to block pthread restart signal (32 on all
3779 // supported Linux platforms). Note that LinuxThreads need to block
3780 // this signal for all threads to work properly. So we don't have
3781 // to use hard-coded signal number when setting up the mask.
3782 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3783
3784 if (sigaction(SR_signum, &act, 0) == -1) {
3785 return -1;
3786 }
3787
3788 // Save signal flag
3789 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3790 return 0;
3791 }
3792
3793 static int sr_notify(OSThread* osthread) {
3794 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3795 assert_status(status == 0, status, "pthread_kill");
3796 return status;
3797 }
3798
3799 // "Randomly" selected value for how long we want to spin
3800 // before bailing out on suspending a thread, also how often
3801 // we send a signal to a thread we want to resume
3802 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3803 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3804
3805 // returns true on success and false on error - really an error is fatal
3806 // but this seems the normal response to library errors
3807 static bool do_suspend(OSThread* osthread) {
3808 assert(osthread->sr.is_running(), "thread should be running");
3809 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3810
3811 // mark as suspended and send signal
3812 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3813 // failed to switch, state wasn't running?
3814 ShouldNotReachHere();
3815 return false;
3816 }
3817
3818 if (sr_notify(osthread) != 0) {
3819 ShouldNotReachHere();
3820 }
3821
3822 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3823 while (true) {
3824 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3825 break;
3826 } else {
3827 // timeout
3828 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3829 if (cancelled == os::SuspendResume::SR_RUNNING) {
3830 return false;
3831 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3832 // make sure that we consume the signal on the semaphore as well
3833 sr_semaphore.wait();
3834 break;
3835 } else {
3836 ShouldNotReachHere();
3837 return false;
3838 }
3839 }
3840 }
3841
3842 guarantee(osthread->sr.is_suspended(), "Must be suspended");
3843 return true;
3844 }
3845
3846 static void do_resume(OSThread* osthread) {
3847 assert(osthread->sr.is_suspended(), "thread should be suspended");
3848 assert(!sr_semaphore.trywait(), "invalid semaphore state");
3849
3850 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3851 // failed to switch to WAKEUP_REQUEST
3852 ShouldNotReachHere();
3853 return;
3854 }
3855
3856 while (true) {
3857 if (sr_notify(osthread) == 0) {
3858 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3859 if (osthread->sr.is_running()) {
3860 return;
3861 }
3862 }
3863 } else {
3864 ShouldNotReachHere();
3865 }
3866 }
3867
3868 guarantee(osthread->sr.is_running(), "Must be running!");
3869 }
3870
3871 ////////////////////////////////////////////////////////////////////////////////
3872 // interrupt support
3873
3874 void os::interrupt(Thread* thread) {
3875 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3876 "possibility of dangling Thread pointer");
3877
3878 OSThread* osthread = thread->osthread();
3879
3880 if (!osthread->interrupted()) {
3881 osthread->set_interrupted(true);
3882 // More than one thread can get here with the same value of osthread,
3883 // resulting in multiple notifications. We do, however, want the store
3884 // to interrupted() to be visible to other threads before we execute unpark().
3885 OrderAccess::fence();
3886 ParkEvent * const slp = thread->_SleepEvent ;
3887 if (slp != NULL) slp->unpark() ;
3888 }
3889
3890 // For JSR166. Unpark even if interrupt status already was set
3891 if (thread->is_Java_thread())
3892 ((JavaThread*)thread)->parker()->unpark();
3893
3894 ParkEvent * ev = thread->_ParkEvent ;
3895 if (ev != NULL) ev->unpark() ;
3896
3897 }
3898
3899 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3900 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3901 "possibility of dangling Thread pointer");
3902
3903 OSThread* osthread = thread->osthread();
3904
3905 bool interrupted = osthread->interrupted();
3906
3907 if (interrupted && clear_interrupted) {
3908 osthread->set_interrupted(false);
3909 // consider thread->_SleepEvent->reset() ... optional optimization
3910 }
3911
3912 return interrupted;
3913 }
3914
3915 ///////////////////////////////////////////////////////////////////////////////////
3916 // signal handling (except suspend/resume)
3917
3918 // This routine may be used by user applications as a "hook" to catch signals.
3919 // The user-defined signal handler must pass unrecognized signals to this
3920 // routine, and if it returns true (non-zero), then the signal handler must
3921 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3922 // routine will never retun false (zero), but instead will execute a VM panic
3923 // routine kill the process.
3924 //
3925 // If this routine returns false, it is OK to call it again. This allows
3926 // the user-defined signal handler to perform checks either before or after
3927 // the VM performs its own checks. Naturally, the user code would be making
3928 // a serious error if it tried to handle an exception (such as a null check
3929 // or breakpoint) that the VM was generating for its own correct operation.
3930 //
3931 // This routine may recognize any of the following kinds of signals:
3932 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3933 // It should be consulted by handlers for any of those signals.
3934 //
3935 // The caller of this routine must pass in the three arguments supplied
3936 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3937 // field of the structure passed to sigaction(). This routine assumes that
3938 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3939 //
3940 // Note that the VM will print warnings if it detects conflicting signal
3941 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3942 //
3943 extern "C" JNIEXPORT int
3944 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3945 void* ucontext, int abort_if_unrecognized);
3946
3947 void signalHandler(int sig, siginfo_t* info, void* uc) {
3948 assert(info != NULL && uc != NULL, "it must be old kernel");
3949 int orig_errno = errno; // Preserve errno value over signal handler.
3950 JVM_handle_linux_signal(sig, info, uc, true);
3951 errno = orig_errno;
3952 }
3953
3954
3955 // This boolean allows users to forward their own non-matching signals
3956 // to JVM_handle_linux_signal, harmlessly.
3957 bool os::Linux::signal_handlers_are_installed = false;
3958
3959 // For signal-chaining
3960 struct sigaction os::Linux::sigact[MAXSIGNUM];
3961 unsigned int os::Linux::sigs = 0;
3962 bool os::Linux::libjsig_is_loaded = false;
3963 typedef struct sigaction *(*get_signal_t)(int);
3964 get_signal_t os::Linux::get_signal_action = NULL;
3965
3966 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3967 struct sigaction *actp = NULL;
3968
3969 if (libjsig_is_loaded) {
3970 // Retrieve the old signal handler from libjsig
3971 actp = (*get_signal_action)(sig);
3972 }
3973 if (actp == NULL) {
3974 // Retrieve the preinstalled signal handler from jvm
3975 actp = get_preinstalled_handler(sig);
3976 }
3977
3978 return actp;
3979 }
3980
3981 static bool call_chained_handler(struct sigaction *actp, int sig,
3982 siginfo_t *siginfo, void *context) {
3983 // Call the old signal handler
3984 if (actp->sa_handler == SIG_DFL) {
3985 // It's more reasonable to let jvm treat it as an unexpected exception
3986 // instead of taking the default action.
3987 return false;
3988 } else if (actp->sa_handler != SIG_IGN) {
3989 if ((actp->sa_flags & SA_NODEFER) == 0) {
3990 // automaticlly block the signal
3991 sigaddset(&(actp->sa_mask), sig);
3992 }
3993
3994 sa_handler_t hand;
3995 sa_sigaction_t sa;
3996 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3997 // retrieve the chained handler
3998 if (siginfo_flag_set) {
3999 sa = actp->sa_sigaction;
4000 } else {
4001 hand = actp->sa_handler;
4002 }
4003
4004 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4005 actp->sa_handler = SIG_DFL;
4006 }
4007
4008 // try to honor the signal mask
4009 sigset_t oset;
4010 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4011
4012 // call into the chained handler
4013 if (siginfo_flag_set) {
4014 (*sa)(sig, siginfo, context);
4015 } else {
4016 (*hand)(sig);
4017 }
4018
4019 // restore the signal mask
4020 pthread_sigmask(SIG_SETMASK, &oset, 0);
4021 }
4022 // Tell jvm's signal handler the signal is taken care of.
4023 return true;
4024 }
4025
4026 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4027 bool chained = false;
4028 // signal-chaining
4029 if (UseSignalChaining) {
4030 struct sigaction *actp = get_chained_signal_action(sig);
4031 if (actp != NULL) {
4032 chained = call_chained_handler(actp, sig, siginfo, context);
4033 }
4034 }
4035 return chained;
4036 }
4037
4038 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4039 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4040 return &sigact[sig];
4041 }
4042 return NULL;
4043 }
4044
4045 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4046 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4047 sigact[sig] = oldAct;
4048 sigs |= (unsigned int)1 << sig;
4049 }
4050
4051 // for diagnostic
4052 int os::Linux::sigflags[MAXSIGNUM];
4053
4054 int os::Linux::get_our_sigflags(int sig) {
4055 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4056 return sigflags[sig];
4057 }
4058
4059 void os::Linux::set_our_sigflags(int sig, int flags) {
4060 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4061 sigflags[sig] = flags;
4062 }
4063
4064 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4065 // Check for overwrite.
4066 struct sigaction oldAct;
4067 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4068
4069 void* oldhand = oldAct.sa_sigaction
4070 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4071 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4072 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4073 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4074 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4075 if (AllowUserSignalHandlers || !set_installed) {
4076 // Do not overwrite; user takes responsibility to forward to us.
4077 return;
4078 } else if (UseSignalChaining) {
4079 // save the old handler in jvm
4080 save_preinstalled_handler(sig, oldAct);
4081 // libjsig also interposes the sigaction() call below and saves the
4082 // old sigaction on it own.
4083 } else {
4084 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4085 "%#lx for signal %d.", (long)oldhand, sig));
4086 }
4087 }
4088
4089 struct sigaction sigAct;
4090 sigfillset(&(sigAct.sa_mask));
4091 sigAct.sa_handler = SIG_DFL;
4092 if (!set_installed) {
4093 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4094 } else {
4095 sigAct.sa_sigaction = signalHandler;
4096 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4097 }
4098 // Save flags, which are set by ours
4099 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4100 sigflags[sig] = sigAct.sa_flags;
4101
4102 int ret = sigaction(sig, &sigAct, &oldAct);
4103 assert(ret == 0, "check");
4104
4105 void* oldhand2 = oldAct.sa_sigaction
4106 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4107 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4108 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4109 }
4110
4111 // install signal handlers for signals that HotSpot needs to
4112 // handle in order to support Java-level exception handling.
4113
4114 void os::Linux::install_signal_handlers() {
4115 if (!signal_handlers_are_installed) {
4116 signal_handlers_are_installed = true;
4117
4118 // signal-chaining
4119 typedef void (*signal_setting_t)();
4120 signal_setting_t begin_signal_setting = NULL;
4121 signal_setting_t end_signal_setting = NULL;
4122 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4123 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4124 if (begin_signal_setting != NULL) {
4125 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4126 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4127 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4128 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4129 libjsig_is_loaded = true;
4130 assert(UseSignalChaining, "should enable signal-chaining");
4131 }
4132 if (libjsig_is_loaded) {
4133 // Tell libjsig jvm is setting signal handlers
4134 (*begin_signal_setting)();
4135 }
4136
4137 set_signal_handler(SIGSEGV, true);
4138 set_signal_handler(SIGPIPE, true);
4139 set_signal_handler(SIGBUS, true);
4140 set_signal_handler(SIGILL, true);
4141 set_signal_handler(SIGFPE, true);
4142 set_signal_handler(SIGXFSZ, true);
4143
4144 if (libjsig_is_loaded) {
4145 // Tell libjsig jvm finishes setting signal handlers
4146 (*end_signal_setting)();
4147 }
4148
4149 // We don't activate signal checker if libjsig is in place, we trust ourselves
4150 // and if UserSignalHandler is installed all bets are off.
4151 // Log that signal checking is off only if -verbose:jni is specified.
4152 if (CheckJNICalls) {
4153 if (libjsig_is_loaded) {
4154 if (PrintJNIResolving) {
4155 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4156 }
4157 check_signals = false;
4158 }
4159 if (AllowUserSignalHandlers) {
4160 if (PrintJNIResolving) {
4161 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4162 }
4163 check_signals = false;
4164 }
4165 }
4166 }
4167 }
4168
4169 // This is the fastest way to get thread cpu time on Linux.
4170 // Returns cpu time (user+sys) for any thread, not only for current.
4171 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4172 // It might work on 2.6.10+ with a special kernel/glibc patch.
4173 // For reference, please, see IEEE Std 1003.1-2004:
4174 // http://www.unix.org/single_unix_specification
4175
4176 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4177 struct timespec tp;
4178 int rc = os::Linux::clock_gettime(clockid, &tp);
4179 assert(rc == 0, "clock_gettime is expected to return 0 code");
4180
4181 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4182 }
4183
4184 /////
4185 // glibc on Linux platform uses non-documented flag
4186 // to indicate, that some special sort of signal
4187 // trampoline is used.
4188 // We will never set this flag, and we should
4189 // ignore this flag in our diagnostic
4190 #ifdef SIGNIFICANT_SIGNAL_MASK
4191 #undef SIGNIFICANT_SIGNAL_MASK
4192 #endif
4193 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4194
4195 static const char* get_signal_handler_name(address handler,
4196 char* buf, int buflen) {
4197 int offset;
4198 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4199 if (found) {
4200 // skip directory names
4201 const char *p1, *p2;
4202 p1 = buf;
4203 size_t len = strlen(os::file_separator());
4204 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4205 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4206 } else {
4207 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4208 }
4209 return buf;
4210 }
4211
4212 static void print_signal_handler(outputStream* st, int sig,
4213 char* buf, size_t buflen) {
4214 struct sigaction sa;
4215
4216 sigaction(sig, NULL, &sa);
4217
4218 // See comment for SIGNIFICANT_SIGNAL_MASK define
4219 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4220
4221 st->print("%s: ", os::exception_name(sig, buf, buflen));
4222
4223 address handler = (sa.sa_flags & SA_SIGINFO)
4224 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4225 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4226
4227 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4228 st->print("SIG_DFL");
4229 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4230 st->print("SIG_IGN");
4231 } else {
4232 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4233 }
4234
4235 st->print(", sa_mask[0]=");
4236 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4237
4238 address rh = VMError::get_resetted_sighandler(sig);
4239 // May be, handler was resetted by VMError?
4240 if(rh != NULL) {
4241 handler = rh;
4242 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4243 }
4244
4245 st->print(", sa_flags=");
4246 os::Posix::print_sa_flags(st, sa.sa_flags);
4247
4248 // Check: is it our handler?
4249 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4250 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4251 // It is our signal handler
4252 // check for flags, reset system-used one!
4253 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4254 st->print(
4255 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4256 os::Linux::get_our_sigflags(sig));
4257 }
4258 }
4259 st->cr();
4260 }
4261
4262
4263 #define DO_SIGNAL_CHECK(sig) \
4264 if (!sigismember(&check_signal_done, sig)) \
4265 os::Linux::check_signal_handler(sig)
4266
4267 // This method is a periodic task to check for misbehaving JNI applications
4268 // under CheckJNI, we can add any periodic checks here
4269
4270 void os::run_periodic_checks() {
4271
4272 if (check_signals == false) return;
4273
4274 // SEGV and BUS if overridden could potentially prevent
4275 // generation of hs*.log in the event of a crash, debugging
4276 // such a case can be very challenging, so we absolutely
4277 // check the following for a good measure:
4278 DO_SIGNAL_CHECK(SIGSEGV);
4279 DO_SIGNAL_CHECK(SIGILL);
4280 DO_SIGNAL_CHECK(SIGFPE);
4281 DO_SIGNAL_CHECK(SIGBUS);
4282 DO_SIGNAL_CHECK(SIGPIPE);
4283 DO_SIGNAL_CHECK(SIGXFSZ);
4284
4285
4286 // ReduceSignalUsage allows the user to override these handlers
4287 // see comments at the very top and jvm_solaris.h
4288 if (!ReduceSignalUsage) {
4289 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4290 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4291 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4292 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4293 }
4294
4295 DO_SIGNAL_CHECK(SR_signum);
4296 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4297 }
4298
4299 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4300
4301 static os_sigaction_t os_sigaction = NULL;
4302
4303 void os::Linux::check_signal_handler(int sig) {
4304 char buf[O_BUFLEN];
4305 address jvmHandler = NULL;
4306
4307
4308 struct sigaction act;
4309 if (os_sigaction == NULL) {
4310 // only trust the default sigaction, in case it has been interposed
4311 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4312 if (os_sigaction == NULL) return;
4313 }
4314
4315 os_sigaction(sig, (struct sigaction*)NULL, &act);
4316
4317
4318 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4319
4320 address thisHandler = (act.sa_flags & SA_SIGINFO)
4321 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4322 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4323
4324
4325 switch(sig) {
4326 case SIGSEGV:
4327 case SIGBUS:
4328 case SIGFPE:
4329 case SIGPIPE:
4330 case SIGILL:
4331 case SIGXFSZ:
4332 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4333 break;
4334
4335 case SHUTDOWN1_SIGNAL:
4336 case SHUTDOWN2_SIGNAL:
4337 case SHUTDOWN3_SIGNAL:
4338 case BREAK_SIGNAL:
4339 jvmHandler = (address)user_handler();
4340 break;
4341
4342 case INTERRUPT_SIGNAL:
4343 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4344 break;
4345
4346 default:
4347 if (sig == SR_signum) {
4348 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4349 } else {
4350 return;
4351 }
4352 break;
4353 }
4354
4355 if (thisHandler != jvmHandler) {
4356 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4357 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4358 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4359 // No need to check this sig any longer
4360 sigaddset(&check_signal_done, sig);
4361 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4362 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4363 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4364 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4365 // No need to check this sig any longer
4366 sigaddset(&check_signal_done, sig);
4367 }
4368
4369 // Dump all the signal
4370 if (sigismember(&check_signal_done, sig)) {
4371 print_signal_handlers(tty, buf, O_BUFLEN);
4372 }
4373 }
4374
4375 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4376
4377 extern bool signal_name(int signo, char* buf, size_t len);
4378
4379 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4380 if (0 < exception_code && exception_code <= SIGRTMAX) {
4381 // signal
4382 if (!signal_name(exception_code, buf, size)) {
4383 jio_snprintf(buf, size, "SIG%d", exception_code);
4384 }
4385 return buf;
4386 } else {
4387 return NULL;
4388 }
4389 }
4390
4391 // this is called _before_ the most of global arguments have been parsed
4392 void os::init(void) {
4393 char dummy; /* used to get a guess on initial stack address */
4394 // first_hrtime = gethrtime();
4395
4396 // With LinuxThreads the JavaMain thread pid (primordial thread)
4397 // is different than the pid of the java launcher thread.
4398 // So, on Linux, the launcher thread pid is passed to the VM
4399 // via the sun.java.launcher.pid property.
4400 // Use this property instead of getpid() if it was correctly passed.
4401 // See bug 6351349.
4402 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4403
4404 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4405
4406 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4407
4408 init_random(1234567);
4409
4410 ThreadCritical::initialize();
4411
4412 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4413 if (Linux::page_size() == -1) {
4414 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4415 strerror(errno)));
4416 }
4417 init_page_sizes((size_t) Linux::page_size());
4418
4419 Linux::initialize_system_info();
4420
4421 // main_thread points to the aboriginal thread
4422 Linux::_main_thread = pthread_self();
4423
4424 Linux::clock_init();
4425 initial_time_count = os::elapsed_counter();
4426 pthread_mutex_init(&dl_mutex, NULL);
4427
4428 // If the pagesize of the VM is greater than 8K determine the appropriate
4429 // number of initial guard pages. The user can change this with the
4430 // command line arguments, if needed.
4431 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4432 StackYellowPages = 1;
4433 StackRedPages = 1;
4434 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4435 }
4436 }
4437
4438 // To install functions for atexit system call
4439 extern "C" {
4440 static void perfMemory_exit_helper() {
4441 perfMemory_exit();
4442 }
4443 }
4444
4445 // this is called _after_ the global arguments have been parsed
4446 jint os::init_2(void)
4447 {
4448 Linux::fast_thread_clock_init();
4449
4450 // Allocate a single page and mark it as readable for safepoint polling
4451 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4452 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4453
4454 os::set_polling_page( polling_page );
4455
4456 #ifndef PRODUCT
4457 if(Verbose && PrintMiscellaneous)
4458 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4459 #endif
4460
4461 if (!UseMembar) {
4462 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4463 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4464 os::set_memory_serialize_page( mem_serialize_page );
4465
4466 #ifndef PRODUCT
4467 if(Verbose && PrintMiscellaneous)
4468 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4469 #endif
4470 }
4471
4472 os::large_page_init();
4473
4474 // initialize suspend/resume support - must do this before signal_sets_init()
4475 if (SR_initialize() != 0) {
4476 perror("SR_initialize failed");
4477 return JNI_ERR;
4478 }
4479
4480 Linux::signal_sets_init();
4481 Linux::install_signal_handlers();
4482
4483 // Check minimum allowable stack size for thread creation and to initialize
4484 // the java system classes, including StackOverflowError - depends on page
4485 // size. Add a page for compiler2 recursion in main thread.
4486 // Add in 2*BytesPerWord times page size to account for VM stack during
4487 // class initialization depending on 32 or 64 bit VM.
4488 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4489 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4490 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4491
4492 size_t threadStackSizeInBytes = ThreadStackSize * K;
4493 if (threadStackSizeInBytes != 0 &&
4494 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4495 tty->print_cr("\nThe stack size specified is too small, "
4496 "Specify at least %dk",
4497 os::Linux::min_stack_allowed/ K);
4498 return JNI_ERR;
4499 }
4500
4501 // Make the stack size a multiple of the page size so that
4502 // the yellow/red zones can be guarded.
4503 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4504 vm_page_size()));
4505
4506 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4507
4508 Linux::libpthread_init();
4509 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4510 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4511 Linux::glibc_version(), Linux::libpthread_version(),
4512 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4513 }
4514
4515 if (UseNUMA) {
4516 if (!Linux::libnuma_init()) {
4517 UseNUMA = false;
4518 } else {
4519 if ((Linux::numa_max_node() < 1)) {
4520 // There's only one node(they start from 0), disable NUMA.
4521 UseNUMA = false;
4522 }
4523 }
4524 // With SHM large pages we cannot uncommit a page, so there's not way
4525 // we can make the adaptive lgrp chunk resizing work. If the user specified
4526 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4527 // disable adaptive resizing.
4528 if (UseNUMA && UseLargePages && UseSHM) {
4529 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4530 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4531 UseLargePages = false;
4532 } else {
4533 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4534 UseAdaptiveSizePolicy = false;
4535 UseAdaptiveNUMAChunkSizing = false;
4536 }
4537 } else {
4538 UseNUMA = false;
4539 }
4540 }
4541 if (!UseNUMA && ForceNUMA) {
4542 UseNUMA = true;
4543 }
4544 }
4545
4546 if (MaxFDLimit) {
4547 // set the number of file descriptors to max. print out error
4548 // if getrlimit/setrlimit fails but continue regardless.
4549 struct rlimit nbr_files;
4550 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4551 if (status != 0) {
4552 if (PrintMiscellaneous && (Verbose || WizardMode))
4553 perror("os::init_2 getrlimit failed");
4554 } else {
4555 nbr_files.rlim_cur = nbr_files.rlim_max;
4556 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4557 if (status != 0) {
4558 if (PrintMiscellaneous && (Verbose || WizardMode))
4559 perror("os::init_2 setrlimit failed");
4560 }
4561 }
4562 }
4563
4564 // Initialize lock used to serialize thread creation (see os::create_thread)
4565 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4566
4567 // at-exit methods are called in the reverse order of their registration.
4568 // atexit functions are called on return from main or as a result of a
4569 // call to exit(3C). There can be only 32 of these functions registered
4570 // and atexit() does not set errno.
4571
4572 if (PerfAllowAtExitRegistration) {
4573 // only register atexit functions if PerfAllowAtExitRegistration is set.
4574 // atexit functions can be delayed until process exit time, which
4575 // can be problematic for embedded VM situations. Embedded VMs should
4576 // call DestroyJavaVM() to assure that VM resources are released.
4577
4578 // note: perfMemory_exit_helper atexit function may be removed in
4579 // the future if the appropriate cleanup code can be added to the
4580 // VM_Exit VMOperation's doit method.
4581 if (atexit(perfMemory_exit_helper) != 0) {
4582 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4583 }
4584 }
4585
4586 // initialize thread priority policy
4587 prio_init();
4588
4589 return JNI_OK;
4590 }
4591
4592 // this is called at the end of vm_initialization
4593 void os::init_3(void) {
4594 #ifdef JAVASE_EMBEDDED
4595 // Start the MemNotifyThread
4596 if (LowMemoryProtection) {
4597 MemNotifyThread::start();
4598 }
4599 return;
4600 #endif
4601 }
4602
4603 // Mark the polling page as unreadable
4604 void os::make_polling_page_unreadable(void) {
4605 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4606 fatal("Could not disable polling page");
4607 };
4608
4609 // Mark the polling page as readable
4610 void os::make_polling_page_readable(void) {
4611 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4612 fatal("Could not enable polling page");
4613 }
4614 };
4615
4616 int os::active_processor_count() {
4617 // Linux doesn't yet have a (official) notion of processor sets,
4618 // so just return the number of online processors.
4619 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4620 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4621 return online_cpus;
4622 }
4623
4624 void os::set_native_thread_name(const char *name) {
4625 // Not yet implemented.
4626 return;
4627 }
4628
4629 bool os::distribute_processes(uint length, uint* distribution) {
4630 // Not yet implemented.
4631 return false;
4632 }
4633
4634 bool os::bind_to_processor(uint processor_id) {
4635 // Not yet implemented.
4636 return false;
4637 }
4638
4639 ///
4640
4641 void os::SuspendedThreadTask::internal_do_task() {
4642 if (do_suspend(_thread->osthread())) {
4643 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4644 do_task(context);
4645 do_resume(_thread->osthread());
4646 }
4647 }
4648
4649 class PcFetcher : public os::SuspendedThreadTask {
4650 public:
4651 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4652 ExtendedPC result();
4653 protected:
4654 void do_task(const os::SuspendedThreadTaskContext& context);
4655 private:
4656 ExtendedPC _epc;
4657 };
4658
4659 ExtendedPC PcFetcher::result() {
4660 guarantee(is_done(), "task is not done yet.");
4661 return _epc;
4662 }
4663
4664 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4665 Thread* thread = context.thread();
4666 OSThread* osthread = thread->osthread();
4667 if (osthread->ucontext() != NULL) {
4668 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4669 } else {
4670 // NULL context is unexpected, double-check this is the VMThread
4671 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4672 }
4673 }
4674
4675 // Suspends the target using the signal mechanism and then grabs the PC before
4676 // resuming the target. Used by the flat-profiler only
4677 ExtendedPC os::get_thread_pc(Thread* thread) {
4678 // Make sure that it is called by the watcher for the VMThread
4679 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4680 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4681
4682 PcFetcher fetcher(thread);
4683 fetcher.run();
4684 return fetcher.result();
4685 }
4686
4687 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4688 {
4689 if (is_NPTL()) {
4690 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4691 } else {
4692 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4693 // word back to default 64bit precision if condvar is signaled. Java
4694 // wants 53bit precision. Save and restore current value.
4695 int fpu = get_fpu_control_word();
4696 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4697 set_fpu_control_word(fpu);
4698 return status;
4699 }
4700 }
4701
4702 ////////////////////////////////////////////////////////////////////////////////
4703 // debug support
4704
4705 bool os::find(address addr, outputStream* st) {
4706 Dl_info dlinfo;
4707 memset(&dlinfo, 0, sizeof(dlinfo));
4708 if (dladdr(addr, &dlinfo) != 0) {
4709 st->print(PTR_FORMAT ": ", addr);
4710 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4711 st->print("%s+%#x", dlinfo.dli_sname,
4712 addr - (intptr_t)dlinfo.dli_saddr);
4713 } else if (dlinfo.dli_fbase != NULL) {
4714 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4715 } else {
4716 st->print("<absolute address>");
4717 }
4718 if (dlinfo.dli_fname != NULL) {
4719 st->print(" in %s", dlinfo.dli_fname);
4720 }
4721 if (dlinfo.dli_fbase != NULL) {
4722 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4723 }
4724 st->cr();
4725
4726 if (Verbose) {
4727 // decode some bytes around the PC
4728 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4729 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4730 address lowest = (address) dlinfo.dli_sname;
4731 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4732 if (begin < lowest) begin = lowest;
4733 Dl_info dlinfo2;
4734 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4735 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4736 end = (address) dlinfo2.dli_saddr;
4737 Disassembler::decode(begin, end, st);
4738 }
4739 return true;
4740 }
4741 return false;
4742 }
4743
4744 ////////////////////////////////////////////////////////////////////////////////
4745 // misc
4746
4747 // This does not do anything on Linux. This is basically a hook for being
4748 // able to use structured exception handling (thread-local exception filters)
4749 // on, e.g., Win32.
4750 void
4751 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4752 JavaCallArguments* args, Thread* thread) {
4753 f(value, method, args, thread);
4754 }
4755
4756 void os::print_statistics() {
4757 }
4758
4759 int os::message_box(const char* title, const char* message) {
4760 int i;
4761 fdStream err(defaultStream::error_fd());
4762 for (i = 0; i < 78; i++) err.print_raw("=");
4763 err.cr();
4764 err.print_raw_cr(title);
4765 for (i = 0; i < 78; i++) err.print_raw("-");
4766 err.cr();
4767 err.print_raw_cr(message);
4768 for (i = 0; i < 78; i++) err.print_raw("=");
4769 err.cr();
4770
4771 char buf[16];
4772 // Prevent process from exiting upon "read error" without consuming all CPU
4773 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4774
4775 return buf[0] == 'y' || buf[0] == 'Y';
4776 }
4777
4778 int os::stat(const char *path, struct stat *sbuf) {
4779 char pathbuf[MAX_PATH];
4780 if (strlen(path) > MAX_PATH - 1) {
4781 errno = ENAMETOOLONG;
4782 return -1;
4783 }
4784 os::native_path(strcpy(pathbuf, path));
4785 return ::stat(pathbuf, sbuf);
4786 }
4787
4788 bool os::check_heap(bool force) {
4789 return true;
4790 }
4791
4792 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4793 return ::vsnprintf(buf, count, format, args);
4794 }
4795
4796 // Is a (classpath) directory empty?
4797 bool os::dir_is_empty(const char* path) {
4798 DIR *dir = NULL;
4799 struct dirent *ptr;
4800
4801 dir = opendir(path);
4802 if (dir == NULL) return true;
4803
4804 /* Scan the directory */
4805 bool result = true;
4806 char buf[sizeof(struct dirent) + MAX_PATH];
4807 while (result && (ptr = ::readdir(dir)) != NULL) {
4808 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4809 result = false;
4810 }
4811 }
4812 closedir(dir);
4813 return result;
4814 }
4815
4816 // This code originates from JDK's sysOpen and open64_w
4817 // from src/solaris/hpi/src/system_md.c
4818
4819 #ifndef O_DELETE
4820 #define O_DELETE 0x10000
4821 #endif
4822
4823 // Open a file. Unlink the file immediately after open returns
4824 // if the specified oflag has the O_DELETE flag set.
4825 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4826
4827 int os::open(const char *path, int oflag, int mode) {
4828
4829 if (strlen(path) > MAX_PATH - 1) {
4830 errno = ENAMETOOLONG;
4831 return -1;
4832 }
4833 int fd;
4834 int o_delete = (oflag & O_DELETE);
4835 oflag = oflag & ~O_DELETE;
4836
4837 fd = ::open64(path, oflag, mode);
4838 if (fd == -1) return -1;
4839
4840 //If the open succeeded, the file might still be a directory
4841 {
4842 struct stat64 buf64;
4843 int ret = ::fstat64(fd, &buf64);
4844 int st_mode = buf64.st_mode;
4845
4846 if (ret != -1) {
4847 if ((st_mode & S_IFMT) == S_IFDIR) {
4848 errno = EISDIR;
4849 ::close(fd);
4850 return -1;
4851 }
4852 } else {
4853 ::close(fd);
4854 return -1;
4855 }
4856 }
4857
4858 /*
4859 * All file descriptors that are opened in the JVM and not
4860 * specifically destined for a subprocess should have the
4861 * close-on-exec flag set. If we don't set it, then careless 3rd
4862 * party native code might fork and exec without closing all
4863 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4864 * UNIXProcess.c), and this in turn might:
4865 *
4866 * - cause end-of-file to fail to be detected on some file
4867 * descriptors, resulting in mysterious hangs, or
4868 *
4869 * - might cause an fopen in the subprocess to fail on a system
4870 * suffering from bug 1085341.
4871 *
4872 * (Yes, the default setting of the close-on-exec flag is a Unix
4873 * design flaw)
4874 *
4875 * See:
4876 * 1085341: 32-bit stdio routines should support file descriptors >255
4877 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4878 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4879 */
4880 #ifdef FD_CLOEXEC
4881 {
4882 int flags = ::fcntl(fd, F_GETFD);
4883 if (flags != -1)
4884 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4885 }
4886 #endif
4887
4888 if (o_delete != 0) {
4889 ::unlink(path);
4890 }
4891 return fd;
4892 }
4893
4894
4895 // create binary file, rewriting existing file if required
4896 int os::create_binary_file(const char* path, bool rewrite_existing) {
4897 int oflags = O_WRONLY | O_CREAT;
4898 if (!rewrite_existing) {
4899 oflags |= O_EXCL;
4900 }
4901 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4902 }
4903
4904 // return current position of file pointer
4905 jlong os::current_file_offset(int fd) {
4906 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4907 }
4908
4909 // move file pointer to the specified offset
4910 jlong os::seek_to_file_offset(int fd, jlong offset) {
4911 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4912 }
4913
4914 // This code originates from JDK's sysAvailable
4915 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4916
4917 int os::available(int fd, jlong *bytes) {
4918 jlong cur, end;
4919 int mode;
4920 struct stat64 buf64;
4921
4922 if (::fstat64(fd, &buf64) >= 0) {
4923 mode = buf64.st_mode;
4924 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4925 /*
4926 * XXX: is the following call interruptible? If so, this might
4927 * need to go through the INTERRUPT_IO() wrapper as for other
4928 * blocking, interruptible calls in this file.
4929 */
4930 int n;
4931 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4932 *bytes = n;
4933 return 1;
4934 }
4935 }
4936 }
4937 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4938 return 0;
4939 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4940 return 0;
4941 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4942 return 0;
4943 }
4944 *bytes = end - cur;
4945 return 1;
4946 }
4947
4948 int os::socket_available(int fd, jint *pbytes) {
4949 // Linux doc says EINTR not returned, unlike Solaris
4950 int ret = ::ioctl(fd, FIONREAD, pbytes);
4951
4952 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4953 // is expected to return 0 on failure and 1 on success to the jdk.
4954 return (ret < 0) ? 0 : 1;
4955 }
4956
4957 // Map a block of memory.
4958 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4959 char *addr, size_t bytes, bool read_only,
4960 bool allow_exec) {
4961 int prot;
4962 int flags = MAP_PRIVATE;
4963
4964 if (read_only) {
4965 prot = PROT_READ;
4966 } else {
4967 prot = PROT_READ | PROT_WRITE;
4968 }
4969
4970 if (allow_exec) {
4971 prot |= PROT_EXEC;
4972 }
4973
4974 if (addr != NULL) {
4975 flags |= MAP_FIXED;
4976 }
4977
4978 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4979 fd, file_offset);
4980 if (mapped_address == MAP_FAILED) {
4981 return NULL;
4982 }
4983 return mapped_address;
4984 }
4985
4986
4987 // Remap a block of memory.
4988 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4989 char *addr, size_t bytes, bool read_only,
4990 bool allow_exec) {
4991 // same as map_memory() on this OS
4992 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4993 allow_exec);
4994 }
4995
4996
4997 // Unmap a block of memory.
4998 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4999 return munmap(addr, bytes) == 0;
5000 }
5001
5002 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5003
5004 static clockid_t thread_cpu_clockid(Thread* thread) {
5005 pthread_t tid = thread->osthread()->pthread_id();
5006 clockid_t clockid;
5007
5008 // Get thread clockid
5009 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5010 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5011 return clockid;
5012 }
5013
5014 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5015 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5016 // of a thread.
5017 //
5018 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5019 // the fast estimate available on the platform.
5020
5021 jlong os::current_thread_cpu_time() {
5022 if (os::Linux::supports_fast_thread_cpu_time()) {
5023 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5024 } else {
5025 // return user + sys since the cost is the same
5026 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5027 }
5028 }
5029
5030 jlong os::thread_cpu_time(Thread* thread) {
5031 // consistent with what current_thread_cpu_time() returns
5032 if (os::Linux::supports_fast_thread_cpu_time()) {
5033 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5034 } else {
5035 return slow_thread_cpu_time(thread, true /* user + sys */);
5036 }
5037 }
5038
5039 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5040 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5041 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5042 } else {
5043 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5044 }
5045 }
5046
5047 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5048 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5049 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5050 } else {
5051 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5052 }
5053 }
5054
5055 //
5056 // -1 on error.
5057 //
5058
5059 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5060 static bool proc_task_unchecked = true;
5061 static const char *proc_stat_path = "/proc/%d/stat";
5062 pid_t tid = thread->osthread()->thread_id();
5063 char *s;
5064 char stat[2048];
5065 int statlen;
5066 char proc_name[64];
5067 int count;
5068 long sys_time, user_time;
5069 char cdummy;
5070 int idummy;
5071 long ldummy;
5072 FILE *fp;
5073
5074 // The /proc/<tid>/stat aggregates per-process usage on
5075 // new Linux kernels 2.6+ where NPTL is supported.
5076 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5077 // See bug 6328462.
5078 // There possibly can be cases where there is no directory
5079 // /proc/self/task, so we check its availability.
5080 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5081 // This is executed only once
5082 proc_task_unchecked = false;
5083 fp = fopen("/proc/self/task", "r");
5084 if (fp != NULL) {
5085 proc_stat_path = "/proc/self/task/%d/stat";
5086 fclose(fp);
5087 }
5088 }
5089
5090 sprintf(proc_name, proc_stat_path, tid);
5091 fp = fopen(proc_name, "r");
5092 if ( fp == NULL ) return -1;
5093 statlen = fread(stat, 1, 2047, fp);
5094 stat[statlen] = '\0';
5095 fclose(fp);
5096
5097 // Skip pid and the command string. Note that we could be dealing with
5098 // weird command names, e.g. user could decide to rename java launcher
5099 // to "java 1.4.2 :)", then the stat file would look like
5100 // 1234 (java 1.4.2 :)) R ... ...
5101 // We don't really need to know the command string, just find the last
5102 // occurrence of ")" and then start parsing from there. See bug 4726580.
5103 s = strrchr(stat, ')');
5104 if (s == NULL ) return -1;
5105
5106 // Skip blank chars
5107 do s++; while (isspace(*s));
5108
5109 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5110 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5111 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5112 &user_time, &sys_time);
5113 if ( count != 13 ) return -1;
5114 if (user_sys_cpu_time) {
5115 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5116 } else {
5117 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5118 }
5119 }
5120
5121 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5122 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5123 info_ptr->may_skip_backward = false; // elapsed time not wall time
5124 info_ptr->may_skip_forward = false; // elapsed time not wall time
5125 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5126 }
5127
5128 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5129 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5130 info_ptr->may_skip_backward = false; // elapsed time not wall time
5131 info_ptr->may_skip_forward = false; // elapsed time not wall time
5132 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5133 }
5134
5135 bool os::is_thread_cpu_time_supported() {
5136 return true;
5137 }
5138
5139 // System loadavg support. Returns -1 if load average cannot be obtained.
5140 // Linux doesn't yet have a (official) notion of processor sets,
5141 // so just return the system wide load average.
5142 int os::loadavg(double loadavg[], int nelem) {
5143 return ::getloadavg(loadavg, nelem);
5144 }
5145
5146 void os::pause() {
5147 char filename[MAX_PATH];
5148 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5149 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5150 } else {
5151 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5152 }
5153
5154 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5155 if (fd != -1) {
5156 struct stat buf;
5157 ::close(fd);
5158 while (::stat(filename, &buf) == 0) {
5159 (void)::poll(NULL, 0, 100);
5160 }
5161 } else {
5162 jio_fprintf(stderr,
5163 "Could not open pause file '%s', continuing immediately.\n", filename);
5164 }
5165 }
5166
5167
5168 // Refer to the comments in os_solaris.cpp park-unpark.
5169 //
5170 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5171 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5172 // For specifics regarding the bug see GLIBC BUGID 261237 :
5173 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5174 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5175 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5176 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5177 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5178 // and monitorenter when we're using 1-0 locking. All those operations may result in
5179 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5180 // of libpthread avoids the problem, but isn't practical.
5181 //
5182 // Possible remedies:
5183 //
5184 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5185 // This is palliative and probabilistic, however. If the thread is preempted
5186 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5187 // than the minimum period may have passed, and the abstime may be stale (in the
5188 // past) resultin in a hang. Using this technique reduces the odds of a hang
5189 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5190 //
5191 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5192 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5193 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5194 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5195 // thread.
5196 //
5197 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5198 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5199 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5200 // This also works well. In fact it avoids kernel-level scalability impediments
5201 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5202 // timers in a graceful fashion.
5203 //
5204 // 4. When the abstime value is in the past it appears that control returns
5205 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5206 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5207 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5208 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5209 // It may be possible to avoid reinitialization by checking the return
5210 // value from pthread_cond_timedwait(). In addition to reinitializing the
5211 // condvar we must establish the invariant that cond_signal() is only called
5212 // within critical sections protected by the adjunct mutex. This prevents
5213 // cond_signal() from "seeing" a condvar that's in the midst of being
5214 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5215 // desirable signal-after-unlock optimization that avoids futile context switching.
5216 //
5217 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5218 // structure when a condvar is used or initialized. cond_destroy() would
5219 // release the helper structure. Our reinitialize-after-timedwait fix
5220 // put excessive stress on malloc/free and locks protecting the c-heap.
5221 //
5222 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5223 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5224 // and only enabling the work-around for vulnerable environments.
5225
5226 // utility to compute the abstime argument to timedwait:
5227 // millis is the relative timeout time
5228 // abstime will be the absolute timeout time
5229 // TODO: replace compute_abstime() with unpackTime()
5230
5231 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5232 if (millis < 0) millis = 0;
5233 struct timeval now;
5234 int status = gettimeofday(&now, NULL);
5235 assert(status == 0, "gettimeofday");
5236 jlong seconds = millis / 1000;
5237 millis %= 1000;
5238 if (seconds > 50000000) { // see man cond_timedwait(3T)
5239 seconds = 50000000;
5240 }
5241 abstime->tv_sec = now.tv_sec + seconds;
5242 long usec = now.tv_usec + millis * 1000;
5243 if (usec >= 1000000) {
5244 abstime->tv_sec += 1;
5245 usec -= 1000000;
5246 }
5247 abstime->tv_nsec = usec * 1000;
5248 return abstime;
5249 }
5250
5251
5252 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5253 // Conceptually TryPark() should be equivalent to park(0).
5254
5255 int os::PlatformEvent::TryPark() {
5256 for (;;) {
5257 const int v = _Event ;
5258 guarantee ((v == 0) || (v == 1), "invariant") ;
5259 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5260 }
5261 }
5262
5263 void os::PlatformEvent::park() { // AKA "down()"
5264 // Invariant: Only the thread associated with the Event/PlatformEvent
5265 // may call park().
5266 // TODO: assert that _Assoc != NULL or _Assoc == Self
5267 int v ;
5268 for (;;) {
5269 v = _Event ;
5270 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5271 }
5272 guarantee (v >= 0, "invariant") ;
5273 if (v == 0) {
5274 // Do this the hard way by blocking ...
5275 int status = pthread_mutex_lock(_mutex);
5276 assert_status(status == 0, status, "mutex_lock");
5277 guarantee (_nParked == 0, "invariant") ;
5278 ++ _nParked ;
5279 while (_Event < 0) {
5280 status = pthread_cond_wait(_cond, _mutex);
5281 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5282 // Treat this the same as if the wait was interrupted
5283 if (status == ETIME) { status = EINTR; }
5284 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5285 }
5286 -- _nParked ;
5287
5288 _Event = 0 ;
5289 status = pthread_mutex_unlock(_mutex);
5290 assert_status(status == 0, status, "mutex_unlock");
5291 // Paranoia to ensure our locked and lock-free paths interact
5292 // correctly with each other.
5293 OrderAccess::fence();
5294 }
5295 guarantee (_Event >= 0, "invariant") ;
5296 }
5297
5298 int os::PlatformEvent::park(jlong millis) {
5299 guarantee (_nParked == 0, "invariant") ;
5300
5301 int v ;
5302 for (;;) {
5303 v = _Event ;
5304 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5305 }
5306 guarantee (v >= 0, "invariant") ;
5307 if (v != 0) return OS_OK ;
5308
5309 // We do this the hard way, by blocking the thread.
5310 // Consider enforcing a minimum timeout value.
5311 struct timespec abst;
5312 compute_abstime(&abst, millis);
5313
5314 int ret = OS_TIMEOUT;
5315 int status = pthread_mutex_lock(_mutex);
5316 assert_status(status == 0, status, "mutex_lock");
5317 guarantee (_nParked == 0, "invariant") ;
5318 ++_nParked ;
5319
5320 // Object.wait(timo) will return because of
5321 // (a) notification
5322 // (b) timeout
5323 // (c) thread.interrupt
5324 //
5325 // Thread.interrupt and object.notify{All} both call Event::set.
5326 // That is, we treat thread.interrupt as a special case of notification.
5327 // The underlying Solaris implementation, cond_timedwait, admits
5328 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5329 // JVM from making those visible to Java code. As such, we must
5330 // filter out spurious wakeups. We assume all ETIME returns are valid.
5331 //
5332 // TODO: properly differentiate simultaneous notify+interrupt.
5333 // In that case, we should propagate the notify to another waiter.
5334
5335 while (_Event < 0) {
5336 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5337 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5338 pthread_cond_destroy (_cond);
5339 pthread_cond_init (_cond, NULL) ;
5340 }
5341 assert_status(status == 0 || status == EINTR ||
5342 status == ETIME || status == ETIMEDOUT,
5343 status, "cond_timedwait");
5344 if (!FilterSpuriousWakeups) break ; // previous semantics
5345 if (status == ETIME || status == ETIMEDOUT) break ;
5346 // We consume and ignore EINTR and spurious wakeups.
5347 }
5348 --_nParked ;
5349 if (_Event >= 0) {
5350 ret = OS_OK;
5351 }
5352 _Event = 0 ;
5353 status = pthread_mutex_unlock(_mutex);
5354 assert_status(status == 0, status, "mutex_unlock");
5355 assert (_nParked == 0, "invariant") ;
5356 // Paranoia to ensure our locked and lock-free paths interact
5357 // correctly with each other.
5358 OrderAccess::fence();
5359 return ret;
5360 }
5361
5362 void os::PlatformEvent::unpark() {
5363 // Transitions for _Event:
5364 // 0 :=> 1
5365 // 1 :=> 1
5366 // -1 :=> either 0 or 1; must signal target thread
5367 // That is, we can safely transition _Event from -1 to either
5368 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5369 // unpark() calls.
5370 // See also: "Semaphores in Plan 9" by Mullender & Cox
5371 //
5372 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5373 // that it will take two back-to-back park() calls for the owning
5374 // thread to block. This has the benefit of forcing a spurious return
5375 // from the first park() call after an unpark() call which will help
5376 // shake out uses of park() and unpark() without condition variables.
5377
5378 if (Atomic::xchg(1, &_Event) >= 0) return;
5379
5380 // Wait for the thread associated with the event to vacate
5381 int status = pthread_mutex_lock(_mutex);
5382 assert_status(status == 0, status, "mutex_lock");
5383 int AnyWaiters = _nParked;
5384 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5385 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5386 AnyWaiters = 0;
5387 pthread_cond_signal(_cond);
5388 }
5389 status = pthread_mutex_unlock(_mutex);
5390 assert_status(status == 0, status, "mutex_unlock");
5391 if (AnyWaiters != 0) {
5392 status = pthread_cond_signal(_cond);
5393 assert_status(status == 0, status, "cond_signal");
5394 }
5395
5396 // Note that we signal() _after dropping the lock for "immortal" Events.
5397 // This is safe and avoids a common class of futile wakeups. In rare
5398 // circumstances this can cause a thread to return prematurely from
5399 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5400 // simply re-test the condition and re-park itself.
5401 }
5402
5403
5404 // JSR166
5405 // -------------------------------------------------------
5406
5407 /*
5408 * The solaris and linux implementations of park/unpark are fairly
5409 * conservative for now, but can be improved. They currently use a
5410 * mutex/condvar pair, plus a a count.
5411 * Park decrements count if > 0, else does a condvar wait. Unpark
5412 * sets count to 1 and signals condvar. Only one thread ever waits
5413 * on the condvar. Contention seen when trying to park implies that someone
5414 * is unparking you, so don't wait. And spurious returns are fine, so there
5415 * is no need to track notifications.
5416 */
5417
5418 #define MAX_SECS 100000000
5419 /*
5420 * This code is common to linux and solaris and will be moved to a
5421 * common place in dolphin.
5422 *
5423 * The passed in time value is either a relative time in nanoseconds
5424 * or an absolute time in milliseconds. Either way it has to be unpacked
5425 * into suitable seconds and nanoseconds components and stored in the
5426 * given timespec structure.
5427 * Given time is a 64-bit value and the time_t used in the timespec is only
5428 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5429 * overflow if times way in the future are given. Further on Solaris versions
5430 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5431 * number of seconds, in abstime, is less than current_time + 100,000,000.
5432 * As it will be 28 years before "now + 100000000" will overflow we can
5433 * ignore overflow and just impose a hard-limit on seconds using the value
5434 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5435 * years from "now".
5436 */
5437
5438 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5439 assert (time > 0, "convertTime");
5440
5441 struct timeval now;
5442 int status = gettimeofday(&now, NULL);
5443 assert(status == 0, "gettimeofday");
5444
5445 time_t max_secs = now.tv_sec + MAX_SECS;
5446
5447 if (isAbsolute) {
5448 jlong secs = time / 1000;
5449 if (secs > max_secs) {
5450 absTime->tv_sec = max_secs;
5451 }
5452 else {
5453 absTime->tv_sec = secs;
5454 }
5455 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5456 }
5457 else {
5458 jlong secs = time / NANOSECS_PER_SEC;
5459 if (secs >= MAX_SECS) {
5460 absTime->tv_sec = max_secs;
5461 absTime->tv_nsec = 0;
5462 }
5463 else {
5464 absTime->tv_sec = now.tv_sec + secs;
5465 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5466 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5467 absTime->tv_nsec -= NANOSECS_PER_SEC;
5468 ++absTime->tv_sec; // note: this must be <= max_secs
5469 }
5470 }
5471 }
5472 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5473 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5474 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5475 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5476 }
5477
5478 void Parker::park(bool isAbsolute, jlong time) {
5479 // Ideally we'd do something useful while spinning, such
5480 // as calling unpackTime().
5481
5482 // Optional fast-path check:
5483 // Return immediately if a permit is available.
5484 // We depend on Atomic::xchg() having full barrier semantics
5485 // since we are doing a lock-free update to _counter.
5486 if (Atomic::xchg(0, &_counter) > 0) return;
5487
5488 Thread* thread = Thread::current();
5489 assert(thread->is_Java_thread(), "Must be JavaThread");
5490 JavaThread *jt = (JavaThread *)thread;
5491
5492 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5493 // Check interrupt before trying to wait
5494 if (Thread::is_interrupted(thread, false)) {
5495 return;
5496 }
5497
5498 // Next, demultiplex/decode time arguments
5499 timespec absTime;
5500 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5501 return;
5502 }
5503 if (time > 0) {
5504 unpackTime(&absTime, isAbsolute, time);
5505 }
5506
5507
5508 // Enter safepoint region
5509 // Beware of deadlocks such as 6317397.
5510 // The per-thread Parker:: mutex is a classic leaf-lock.
5511 // In particular a thread must never block on the Threads_lock while
5512 // holding the Parker:: mutex. If safepoints are pending both the
5513 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5514 ThreadBlockInVM tbivm(jt);
5515
5516 // Don't wait if cannot get lock since interference arises from
5517 // unblocking. Also. check interrupt before trying wait
5518 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5519 return;
5520 }
5521
5522 int status ;
5523 if (_counter > 0) { // no wait needed
5524 _counter = 0;
5525 status = pthread_mutex_unlock(_mutex);
5526 assert (status == 0, "invariant") ;
5527 // Paranoia to ensure our locked and lock-free paths interact
5528 // correctly with each other and Java-level accesses.
5529 OrderAccess::fence();
5530 return;
5531 }
5532
5533 #ifdef ASSERT
5534 // Don't catch signals while blocked; let the running threads have the signals.
5535 // (This allows a debugger to break into the running thread.)
5536 sigset_t oldsigs;
5537 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5538 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5539 #endif
5540
5541 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5542 jt->set_suspend_equivalent();
5543 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5544
5545 if (time == 0) {
5546 status = pthread_cond_wait (_cond, _mutex) ;
5547 } else {
5548 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5549 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5550 pthread_cond_destroy (_cond) ;
5551 pthread_cond_init (_cond, NULL);
5552 }
5553 }
5554 assert_status(status == 0 || status == EINTR ||
5555 status == ETIME || status == ETIMEDOUT,
5556 status, "cond_timedwait");
5557
5558 #ifdef ASSERT
5559 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5560 #endif
5561
5562 _counter = 0 ;
5563 status = pthread_mutex_unlock(_mutex) ;
5564 assert_status(status == 0, status, "invariant") ;
5565 // Paranoia to ensure our locked and lock-free paths interact
5566 // correctly with each other and Java-level accesses.
5567 OrderAccess::fence();
5568
5569 // If externally suspended while waiting, re-suspend
5570 if (jt->handle_special_suspend_equivalent_condition()) {
5571 jt->java_suspend_self();
5572 }
5573 }
5574
5575 void Parker::unpark() {
5576 int s, status ;
5577 status = pthread_mutex_lock(_mutex);
5578 assert (status == 0, "invariant") ;
5579 s = _counter;
5580 _counter = 1;
5581 if (s < 1) {
5582 if (WorkAroundNPTLTimedWaitHang) {
5583 status = pthread_cond_signal (_cond) ;
5584 assert (status == 0, "invariant") ;
5585 status = pthread_mutex_unlock(_mutex);
5586 assert (status == 0, "invariant") ;
5587 } else {
5588 status = pthread_mutex_unlock(_mutex);
5589 assert (status == 0, "invariant") ;
5590 status = pthread_cond_signal (_cond) ;
5591 assert (status == 0, "invariant") ;
5592 }
5593 } else {
5594 pthread_mutex_unlock(_mutex);
5595 assert (status == 0, "invariant") ;
5596 }
5597 }
5598
5599
5600 extern char** environ;
5601
5602 #ifndef __NR_fork
5603 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5604 #endif
5605
5606 #ifndef __NR_execve
5607 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5608 #endif
5609
5610 // Run the specified command in a separate process. Return its exit value,
5611 // or -1 on failure (e.g. can't fork a new process).
5612 // Unlike system(), this function can be called from signal handler. It
5613 // doesn't block SIGINT et al.
5614 int os::fork_and_exec(char* cmd) {
5615 const char * argv[4] = {"sh", "-c", cmd, NULL};
5616
5617 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5618 // pthread_atfork handlers and reset pthread library. All we need is a
5619 // separate process to execve. Make a direct syscall to fork process.
5620 // On IA64 there's no fork syscall, we have to use fork() and hope for
5621 // the best...
5622 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5623 IA64_ONLY(fork();)
5624
5625 if (pid < 0) {
5626 // fork failed
5627 return -1;
5628
5629 } else if (pid == 0) {
5630 // child process
5631
5632 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5633 // first to kill every thread on the thread list. Because this list is
5634 // not reset by fork() (see notes above), execve() will instead kill
5635 // every thread in the parent process. We know this is the only thread
5636 // in the new process, so make a system call directly.
5637 // IA64 should use normal execve() from glibc to match the glibc fork()
5638 // above.
5639 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5640 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5641
5642 // execve failed
5643 _exit(-1);
5644
5645 } else {
5646 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5647 // care about the actual exit code, for now.
5648
5649 int status;
5650
5651 // Wait for the child process to exit. This returns immediately if
5652 // the child has already exited. */
5653 while (waitpid(pid, &status, 0) < 0) {
5654 switch (errno) {
5655 case ECHILD: return 0;
5656 case EINTR: break;
5657 default: return -1;
5658 }
5659 }
5660
5661 if (WIFEXITED(status)) {
5662 // The child exited normally; get its exit code.
5663 return WEXITSTATUS(status);
5664 } else if (WIFSIGNALED(status)) {
5665 // The child exited because of a signal
5666 // The best value to return is 0x80 + signal number,
5667 // because that is what all Unix shells do, and because
5668 // it allows callers to distinguish between process exit and
5669 // process death by signal.
5670 return 0x80 + WTERMSIG(status);
5671 } else {
5672 // Unknown exit code; pass it through
5673 return status;
5674 }
5675 }
5676 }
5677
5678 // is_headless_jre()
5679 //
5680 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5681 // in order to report if we are running in a headless jre
5682 //
5683 // Since JDK8 xawt/libmawt.so was moved into the same directory
5684 // as libawt.so, and renamed libawt_xawt.so
5685 //
5686 bool os::is_headless_jre() {
5687 struct stat statbuf;
5688 char buf[MAXPATHLEN];
5689 char libmawtpath[MAXPATHLEN];
5690 const char *xawtstr = "/xawt/libmawt.so";
5691 const char *new_xawtstr = "/libawt_xawt.so";
5692 char *p;
5693
5694 // Get path to libjvm.so
5695 os::jvm_path(buf, sizeof(buf));
5696
5697 // Get rid of libjvm.so
5698 p = strrchr(buf, '/');
5699 if (p == NULL) return false;
5700 else *p = '\0';
5701
5702 // Get rid of client or server
5703 p = strrchr(buf, '/');
5704 if (p == NULL) return false;
5705 else *p = '\0';
5706
5707 // check xawt/libmawt.so
5708 strcpy(libmawtpath, buf);
5709 strcat(libmawtpath, xawtstr);
5710 if (::stat(libmawtpath, &statbuf) == 0) return false;
5711
5712 // check libawt_xawt.so
5713 strcpy(libmawtpath, buf);
5714 strcat(libmawtpath, new_xawtstr);
5715 if (::stat(libmawtpath, &statbuf) == 0) return false;
5716
5717 return true;
5718 }
5719
5720 // Get the default path to the core file
5721 // Returns the length of the string
5722 int os::get_core_path(char* buffer, size_t bufferSize) {
5723 const char* p = get_current_directory(buffer, bufferSize);
5724
5725 if (p == NULL) {
5726 assert(p != NULL, "failed to get current directory");
5727 return 0;
5728 }
5729
5730 return strlen(buffer);
5731 }
5732
5733 #ifdef JAVASE_EMBEDDED
5734 //
5735 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5736 //
5737 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5738
5739 // ctor
5740 //
5741 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5742 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5743 _fd = fd;
5744
5745 if (os::create_thread(this, os::os_thread)) {
5746 _memnotify_thread = this;
5747 os::set_priority(this, NearMaxPriority);
5748 os::start_thread(this);
5749 }
5750 }
5751
5752 // Where all the work gets done
5753 //
5754 void MemNotifyThread::run() {
5755 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5756
5757 // Set up the select arguments
5758 fd_set rfds;
5759 if (_fd != -1) {
5760 FD_ZERO(&rfds);
5761 FD_SET(_fd, &rfds);
5762 }
5763
5764 // Now wait for the mem_notify device to wake up
5765 while (1) {
5766 // Wait for the mem_notify device to signal us..
5767 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5768 if (rc == -1) {
5769 perror("select!\n");
5770 break;
5771 } else if (rc) {
5772 //ssize_t free_before = os::available_memory();
5773 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5774
5775 // The kernel is telling us there is not much memory left...
5776 // try to do something about that
5777
5778 // If we are not already in a GC, try one.
5779 if (!Universe::heap()->is_gc_active()) {
5780 Universe::heap()->collect(GCCause::_allocation_failure);
5781
5782 //ssize_t free_after = os::available_memory();
5783 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5784 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5785 }
5786 // We might want to do something like the following if we find the GC's are not helping...
5787 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5788 }
5789 }
5790 }
5791
5792 //
5793 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5794 //
5795 void MemNotifyThread::start() {
5796 int fd;
5797 fd = open ("/dev/mem_notify", O_RDONLY, 0);
5798 if (fd < 0) {
5799 return;
5800 }
5801
5802 if (memnotify_thread() == NULL) {
5803 new MemNotifyThread(fd);
5804 }
5805 }
5806
5807 #endif // JAVASE_EMBEDDED