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 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2235 // but they're the same for all the linux arch that we support
2236 // and they're the same for solaris but there's no common place to put this.
2237 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2238 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2239 "ILL_COPROC", "ILL_BADSTK" };
2240
2241 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2242 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2243 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2244
2245 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2246
2247 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2248
2249 void os::print_siginfo(outputStream* st, void* siginfo) {
2250 st->print("siginfo:");
2251
2252 const int buflen = 100;
2253 char buf[buflen];
2254 siginfo_t *si = (siginfo_t*)siginfo;
2255 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2256 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2257 st->print("si_errno=%s", buf);
2258 } else {
2259 st->print("si_errno=%d", si->si_errno);
2260 }
2261 const int c = si->si_code;
2262 assert(c > 0, "unexpected si_code");
2263 switch (si->si_signo) {
2264 case SIGILL:
2265 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2266 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2267 break;
2268 case SIGFPE:
2269 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2270 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2271 break;
2272 case SIGSEGV:
2273 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2274 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2275 break;
2276 case SIGBUS:
2277 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2278 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2279 break;
2280 default:
2281 st->print(", si_code=%d", si->si_code);
2282 // no si_addr
2283 }
2284
2285 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2286 UseSharedSpaces) {
2287 FileMapInfo* mapinfo = FileMapInfo::current_info();
2288 if (mapinfo->is_in_shared_space(si->si_addr)) {
2289 st->print("\n\nError accessing class data sharing archive." \
2290 " Mapped file inaccessible during execution, " \
2291 " possible disk/network problem.");
2292 }
2293 }
2294 st->cr();
2295 }
2296
2297
2298 static void print_signal_handler(outputStream* st, int sig,
2299 char* buf, size_t buflen);
2300
2301 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2302 st->print_cr("Signal Handlers:");
2303 print_signal_handler(st, SIGSEGV, buf, buflen);
2304 print_signal_handler(st, SIGBUS , buf, buflen);
2305 print_signal_handler(st, SIGFPE , buf, buflen);
2306 print_signal_handler(st, SIGPIPE, buf, buflen);
2307 print_signal_handler(st, SIGXFSZ, buf, buflen);
2308 print_signal_handler(st, SIGILL , buf, buflen);
2309 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2310 print_signal_handler(st, SR_signum, buf, buflen);
2311 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2312 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2313 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2314 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2315 }
2316
2317 static char saved_jvm_path[MAXPATHLEN] = {0};
2318
2319 // Find the full path to the current module, libjvm.so
2320 void os::jvm_path(char *buf, jint buflen) {
2321 // Error checking.
2322 if (buflen < MAXPATHLEN) {
2323 assert(false, "must use a large-enough buffer");
2324 buf[0] = '\0';
2325 return;
2326 }
2327 // Lazy resolve the path to current module.
2328 if (saved_jvm_path[0] != 0) {
2329 strcpy(buf, saved_jvm_path);
2330 return;
2331 }
2332
2333 char dli_fname[MAXPATHLEN];
2334 bool ret = dll_address_to_library_name(
2335 CAST_FROM_FN_PTR(address, os::jvm_path),
2336 dli_fname, sizeof(dli_fname), NULL);
2337 assert(ret, "cannot locate libjvm");
2338 char *rp = NULL;
2339 if (ret && dli_fname[0] != '\0') {
2340 rp = realpath(dli_fname, buf);
2341 }
2342 if (rp == NULL)
2343 return;
2344
2345 if (Arguments::created_by_gamma_launcher()) {
2346 // Support for the gamma launcher. Typical value for buf is
2347 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2348 // the right place in the string, then assume we are installed in a JDK and
2349 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2350 // up the path so it looks like libjvm.so is installed there (append a
2351 // fake suffix hotspot/libjvm.so).
2352 const char *p = buf + strlen(buf) - 1;
2353 for (int count = 0; p > buf && count < 5; ++count) {
2354 for (--p; p > buf && *p != '/'; --p)
2355 /* empty */ ;
2356 }
2357
2358 if (strncmp(p, "/jre/lib/", 9) != 0) {
2359 // Look for JAVA_HOME in the environment.
2360 char* java_home_var = ::getenv("JAVA_HOME");
2361 if (java_home_var != NULL && java_home_var[0] != 0) {
2362 char* jrelib_p;
2363 int len;
2364
2365 // Check the current module name "libjvm.so".
2366 p = strrchr(buf, '/');
2367 assert(strstr(p, "/libjvm") == p, "invalid library name");
2368
2369 rp = realpath(java_home_var, buf);
2370 if (rp == NULL)
2371 return;
2372
2373 // determine if this is a legacy image or modules image
2374 // modules image doesn't have "jre" subdirectory
2375 len = strlen(buf);
2376 jrelib_p = buf + len;
2377 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2378 if (0 != access(buf, F_OK)) {
2379 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2380 }
2381
2382 if (0 == access(buf, F_OK)) {
2383 // Use current module name "libjvm.so"
2384 len = strlen(buf);
2385 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2386 } else {
2387 // Go back to path of .so
2388 rp = realpath(dli_fname, buf);
2389 if (rp == NULL)
2390 return;
2391 }
2392 }
2393 }
2394 }
2395
2396 strcpy(saved_jvm_path, buf);
2397 }
2398
2399 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2400 // no prefix required, not even "_"
2401 }
2402
2403 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2404 // no suffix required
2405 }
2406
2407 ////////////////////////////////////////////////////////////////////////////////
2408 // sun.misc.Signal support
2409
2410 static volatile jint sigint_count = 0;
2411
2412 static void
2413 UserHandler(int sig, void *siginfo, void *context) {
2414 // 4511530 - sem_post is serialized and handled by the manager thread. When
2415 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2416 // don't want to flood the manager thread with sem_post requests.
2417 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2418 return;
2419
2420 // Ctrl-C is pressed during error reporting, likely because the error
2421 // handler fails to abort. Let VM die immediately.
2422 if (sig == SIGINT && is_error_reported()) {
2423 os::die();
2424 }
2425
2426 os::signal_notify(sig);
2427 }
2428
2429 void* os::user_handler() {
2430 return CAST_FROM_FN_PTR(void*, UserHandler);
2431 }
2432
2433 class Semaphore : public StackObj {
2434 public:
2435 Semaphore();
2436 ~Semaphore();
2437 void signal();
2438 void wait();
2439 bool trywait();
2440 bool timedwait(unsigned int sec, int nsec);
2441 private:
2442 sem_t _semaphore;
2443 };
2444
2445
2446 Semaphore::Semaphore() {
2447 sem_init(&_semaphore, 0, 0);
2448 }
2449
2450 Semaphore::~Semaphore() {
2451 sem_destroy(&_semaphore);
2452 }
2453
2454 void Semaphore::signal() {
2455 sem_post(&_semaphore);
2456 }
2457
2458 void Semaphore::wait() {
2459 sem_wait(&_semaphore);
2460 }
2461
2462 bool Semaphore::trywait() {
2463 return sem_trywait(&_semaphore) == 0;
2464 }
2465
2466 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2467 struct timespec ts;
2468 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2469
2470 while (1) {
2471 int result = sem_timedwait(&_semaphore, &ts);
2472 if (result == 0) {
2473 return true;
2474 } else if (errno == EINTR) {
2475 continue;
2476 } else if (errno == ETIMEDOUT) {
2477 return false;
2478 } else {
2479 return false;
2480 }
2481 }
2482 }
2483
2484 extern "C" {
2485 typedef void (*sa_handler_t)(int);
2486 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2487 }
2488
2489 void* os::signal(int signal_number, void* handler) {
2490 struct sigaction sigAct, oldSigAct;
2491
2492 sigfillset(&(sigAct.sa_mask));
2493 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2494 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2495
2496 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2497 // -1 means registration failed
2498 return (void *)-1;
2499 }
2500
2501 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2502 }
2503
2504 void os::signal_raise(int signal_number) {
2505 ::raise(signal_number);
2506 }
2507
2508 /*
2509 * The following code is moved from os.cpp for making this
2510 * code platform specific, which it is by its very nature.
2511 */
2512
2513 // Will be modified when max signal is changed to be dynamic
2514 int os::sigexitnum_pd() {
2515 return NSIG;
2516 }
2517
2518 // a counter for each possible signal value
2519 static volatile jint pending_signals[NSIG+1] = { 0 };
2520
2521 // Linux(POSIX) specific hand shaking semaphore.
2522 static sem_t sig_sem;
2523 static Semaphore sr_semaphore;
2524
2525 void os::signal_init_pd() {
2526 // Initialize signal structures
2527 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2528
2529 // Initialize signal semaphore
2530 ::sem_init(&sig_sem, 0, 0);
2531 }
2532
2533 void os::signal_notify(int sig) {
2534 Atomic::inc(&pending_signals[sig]);
2535 ::sem_post(&sig_sem);
2536 }
2537
2538 static int check_pending_signals(bool wait) {
2539 Atomic::store(0, &sigint_count);
2540 for (;;) {
2541 for (int i = 0; i < NSIG + 1; i++) {
2542 jint n = pending_signals[i];
2543 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2544 return i;
2545 }
2546 }
2547 if (!wait) {
2548 return -1;
2549 }
2550 JavaThread *thread = JavaThread::current();
2551 ThreadBlockInVM tbivm(thread);
2552
2553 bool threadIsSuspended;
2554 do {
2555 thread->set_suspend_equivalent();
2556 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2557 ::sem_wait(&sig_sem);
2558
2559 // were we externally suspended while we were waiting?
2560 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2561 if (threadIsSuspended) {
2562 //
2563 // The semaphore has been incremented, but while we were waiting
2564 // another thread suspended us. We don't want to continue running
2565 // while suspended because that would surprise the thread that
2566 // suspended us.
2567 //
2568 ::sem_post(&sig_sem);
2569
2570 thread->java_suspend_self();
2571 }
2572 } while (threadIsSuspended);
2573 }
2574 }
2575
2576 int os::signal_lookup() {
2577 return check_pending_signals(false);
2578 }
2579
2580 int os::signal_wait() {
2581 return check_pending_signals(true);
2582 }
2583
2584 ////////////////////////////////////////////////////////////////////////////////
2585 // Virtual Memory
2586
2587 int os::vm_page_size() {
2588 // Seems redundant as all get out
2589 assert(os::Linux::page_size() != -1, "must call os::init");
2590 return os::Linux::page_size();
2591 }
2592
2593 // Solaris allocates memory by pages.
2594 int os::vm_allocation_granularity() {
2595 assert(os::Linux::page_size() != -1, "must call os::init");
2596 return os::Linux::page_size();
2597 }
2598
2599 // Rationale behind this function:
2600 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2601 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2602 // samples for JITted code. Here we create private executable mapping over the code cache
2603 // and then we can use standard (well, almost, as mapping can change) way to provide
2604 // info for the reporting script by storing timestamp and location of symbol
2605 void linux_wrap_code(char* base, size_t size) {
2606 static volatile jint cnt = 0;
2607
2608 if (!UseOprofile) {
2609 return;
2610 }
2611
2612 char buf[PATH_MAX+1];
2613 int num = Atomic::add(1, &cnt);
2614
2615 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2616 os::get_temp_directory(), os::current_process_id(), num);
2617 unlink(buf);
2618
2619 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2620
2621 if (fd != -1) {
2622 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2623 if (rv != (off_t)-1) {
2624 if (::write(fd, "", 1) == 1) {
2625 mmap(base, size,
2626 PROT_READ|PROT_WRITE|PROT_EXEC,
2627 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2628 }
2629 }
2630 ::close(fd);
2631 unlink(buf);
2632 }
2633 }
2634
2635 static bool recoverable_mmap_error(int err) {
2636 // See if the error is one we can let the caller handle. This
2637 // list of errno values comes from JBS-6843484. I can't find a
2638 // Linux man page that documents this specific set of errno
2639 // values so while this list currently matches Solaris, it may
2640 // change as we gain experience with this failure mode.
2641 switch (err) {
2642 case EBADF:
2643 case EINVAL:
2644 case ENOTSUP:
2645 // let the caller deal with these errors
2646 return true;
2647
2648 default:
2649 // Any remaining errors on this OS can cause our reserved mapping
2650 // to be lost. That can cause confusion where different data
2651 // structures think they have the same memory mapped. The worst
2652 // scenario is if both the VM and a library think they have the
2653 // same memory mapped.
2654 return false;
2655 }
2656 }
2657
2658 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2659 int err) {
2660 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2661 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2662 strerror(err), err);
2663 }
2664
2665 static void warn_fail_commit_memory(char* addr, size_t size,
2666 size_t alignment_hint, bool exec,
2667 int err) {
2668 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2669 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2670 alignment_hint, exec, strerror(err), err);
2671 }
2672
2673 // NOTE: Linux kernel does not really reserve the pages for us.
2674 // All it does is to check if there are enough free pages
2675 // left at the time of mmap(). This could be a potential
2676 // problem.
2677 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2678 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2679 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2680 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2681 if (res != (uintptr_t) MAP_FAILED) {
2682 if (UseNUMAInterleaving) {
2683 numa_make_global(addr, size);
2684 }
2685 return 0;
2686 }
2687
2688 int err = errno; // save errno from mmap() call above
2689
2690 if (!recoverable_mmap_error(err)) {
2691 warn_fail_commit_memory(addr, size, exec, err);
2692 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2693 }
2694
2695 return err;
2696 }
2697
2698 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2699 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2700 }
2701
2702 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2703 const char* mesg) {
2704 assert(mesg != NULL, "mesg must be specified");
2705 int err = os::Linux::commit_memory_impl(addr, size, exec);
2706 if (err != 0) {
2707 // the caller wants all commit errors to exit with the specified mesg:
2708 warn_fail_commit_memory(addr, size, exec, err);
2709 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2710 }
2711 }
2712
2713 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2714 #ifndef MAP_HUGETLB
2715 #define MAP_HUGETLB 0x40000
2716 #endif
2717
2718 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2719 #ifndef MADV_HUGEPAGE
2720 #define MADV_HUGEPAGE 14
2721 #endif
2722
2723 int os::Linux::commit_memory_impl(char* addr, size_t size,
2724 size_t alignment_hint, bool exec) {
2725 int err;
2726 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2727 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2728 uintptr_t res =
2729 (uintptr_t) ::mmap(addr, size, prot,
2730 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2731 -1, 0);
2732 if (res != (uintptr_t) MAP_FAILED) {
2733 if (UseNUMAInterleaving) {
2734 numa_make_global(addr, size);
2735 }
2736 return 0;
2737 }
2738
2739 err = errno; // save errno from mmap() call above
2740
2741 if (!recoverable_mmap_error(err)) {
2742 // However, it is not clear that this loss of our reserved mapping
2743 // happens with large pages on Linux or that we cannot recover
2744 // from the loss. For now, we just issue a warning and we don't
2745 // call vm_exit_out_of_memory(). This issue is being tracked by
2746 // JBS-8007074.
2747 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2748 // vm_exit_out_of_memory(size, OOM_MMAP_ERROR,
2749 // "committing reserved memory.");
2750 }
2751 // Fall through and try to use small pages
2752 }
2753
2754 err = os::Linux::commit_memory_impl(addr, size, exec);
2755 if (err == 0) {
2756 realign_memory(addr, size, alignment_hint);
2757 }
2758 return err;
2759 }
2760
2761 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2762 bool exec) {
2763 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2764 }
2765
2766 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2767 size_t alignment_hint, bool exec,
2768 const char* mesg) {
2769 assert(mesg != NULL, "mesg must be specified");
2770 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2771 if (err != 0) {
2772 // the caller wants all commit errors to exit with the specified mesg:
2773 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2774 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2775 }
2776 }
2777
2778 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2779 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2780 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2781 // be supported or the memory may already be backed by huge pages.
2782 ::madvise(addr, bytes, MADV_HUGEPAGE);
2783 }
2784 }
2785
2786 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2787 // This method works by doing an mmap over an existing mmaping and effectively discarding
2788 // the existing pages. However it won't work for SHM-based large pages that cannot be
2789 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2790 // small pages on top of the SHM segment. This method always works for small pages, so we
2791 // allow that in any case.
2792 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2793 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2794 }
2795 }
2796
2797 void os::numa_make_global(char *addr, size_t bytes) {
2798 Linux::numa_interleave_memory(addr, bytes);
2799 }
2800
2801 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2802 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2803 }
2804
2805 bool os::numa_topology_changed() { return false; }
2806
2807 size_t os::numa_get_groups_num() {
2808 int max_node = Linux::numa_max_node();
2809 return max_node > 0 ? max_node + 1 : 1;
2810 }
2811
2812 int os::numa_get_group_id() {
2813 int cpu_id = Linux::sched_getcpu();
2814 if (cpu_id != -1) {
2815 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2816 if (lgrp_id != -1) {
2817 return lgrp_id;
2818 }
2819 }
2820 return 0;
2821 }
2822
2823 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2824 for (size_t i = 0; i < size; i++) {
2825 ids[i] = i;
2826 }
2827 return size;
2828 }
2829
2830 bool os::get_page_info(char *start, page_info* info) {
2831 return false;
2832 }
2833
2834 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2835 return end;
2836 }
2837
2838
2839 int os::Linux::sched_getcpu_syscall(void) {
2840 unsigned int cpu;
2841 int retval = -1;
2842
2843 #if defined(IA32)
2844 # ifndef SYS_getcpu
2845 # define SYS_getcpu 318
2846 # endif
2847 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2848 #elif defined(AMD64)
2849 // Unfortunately we have to bring all these macros here from vsyscall.h
2850 // to be able to compile on old linuxes.
2851 # define __NR_vgetcpu 2
2852 # define VSYSCALL_START (-10UL << 20)
2853 # define VSYSCALL_SIZE 1024
2854 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2855 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2856 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2857 retval = vgetcpu(&cpu, NULL, NULL);
2858 #endif
2859
2860 return (retval == -1) ? retval : cpu;
2861 }
2862
2863 // Something to do with the numa-aware allocator needs these symbols
2864 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2865 extern "C" JNIEXPORT void numa_error(char *where) { }
2866 extern "C" JNIEXPORT int fork1() { return fork(); }
2867
2868
2869 // If we are running with libnuma version > 2, then we should
2870 // be trying to use symbols with versions 1.1
2871 // If we are running with earlier version, which did not have symbol versions,
2872 // we should use the base version.
2873 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2874 void *f = dlvsym(handle, name, "libnuma_1.1");
2875 if (f == NULL) {
2876 f = dlsym(handle, name);
2877 }
2878 return f;
2879 }
2880
2881 bool os::Linux::libnuma_init() {
2882 // sched_getcpu() should be in libc.
2883 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2884 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2885
2886 // If it's not, try a direct syscall.
2887 if (sched_getcpu() == -1)
2888 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2889
2890 if (sched_getcpu() != -1) { // Does it work?
2891 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2892 if (handle != NULL) {
2893 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2894 libnuma_dlsym(handle, "numa_node_to_cpus")));
2895 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2896 libnuma_dlsym(handle, "numa_max_node")));
2897 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2898 libnuma_dlsym(handle, "numa_available")));
2899 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2900 libnuma_dlsym(handle, "numa_tonode_memory")));
2901 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2902 libnuma_dlsym(handle, "numa_interleave_memory")));
2903
2904
2905 if (numa_available() != -1) {
2906 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2907 // Create a cpu -> node mapping
2908 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2909 rebuild_cpu_to_node_map();
2910 return true;
2911 }
2912 }
2913 }
2914 return false;
2915 }
2916
2917 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2918 // The table is later used in get_node_by_cpu().
2919 void os::Linux::rebuild_cpu_to_node_map() {
2920 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2921 // in libnuma (possible values are starting from 16,
2922 // and continuing up with every other power of 2, but less
2923 // than the maximum number of CPUs supported by kernel), and
2924 // is a subject to change (in libnuma version 2 the requirements
2925 // are more reasonable) we'll just hardcode the number they use
2926 // in the library.
2927 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2928
2929 size_t cpu_num = os::active_processor_count();
2930 size_t cpu_map_size = NCPUS / BitsPerCLong;
2931 size_t cpu_map_valid_size =
2932 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2933
2934 cpu_to_node()->clear();
2935 cpu_to_node()->at_grow(cpu_num - 1);
2936 size_t node_num = numa_get_groups_num();
2937
2938 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2939 for (size_t i = 0; i < node_num; i++) {
2940 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2941 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2942 if (cpu_map[j] != 0) {
2943 for (size_t k = 0; k < BitsPerCLong; k++) {
2944 if (cpu_map[j] & (1UL << k)) {
2945 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2946 }
2947 }
2948 }
2949 }
2950 }
2951 }
2952 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2953 }
2954
2955 int os::Linux::get_node_by_cpu(int cpu_id) {
2956 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2957 return cpu_to_node()->at(cpu_id);
2958 }
2959 return -1;
2960 }
2961
2962 GrowableArray<int>* os::Linux::_cpu_to_node;
2963 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2964 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2965 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2966 os::Linux::numa_available_func_t os::Linux::_numa_available;
2967 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2968 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2969 unsigned long* os::Linux::_numa_all_nodes;
2970
2971 bool os::pd_uncommit_memory(char* addr, size_t size) {
2972 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2973 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2974 return res != (uintptr_t) MAP_FAILED;
2975 }
2976
2977 // Linux uses a growable mapping for the stack, and if the mapping for
2978 // the stack guard pages is not removed when we detach a thread the
2979 // stack cannot grow beyond the pages where the stack guard was
2980 // mapped. If at some point later in the process the stack expands to
2981 // that point, the Linux kernel cannot expand the stack any further
2982 // because the guard pages are in the way, and a segfault occurs.
2983 //
2984 // However, it's essential not to split the stack region by unmapping
2985 // a region (leaving a hole) that's already part of the stack mapping,
2986 // so if the stack mapping has already grown beyond the guard pages at
2987 // the time we create them, we have to truncate the stack mapping.
2988 // So, we need to know the extent of the stack mapping when
2989 // create_stack_guard_pages() is called.
2990
2991 // Find the bounds of the stack mapping. Return true for success.
2992 //
2993 // We only need this for stacks that are growable: at the time of
2994 // writing thread stacks don't use growable mappings (i.e. those
2995 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2996 // only applies to the main thread.
2997
2998 static
2999 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
3000
3001 char buf[128];
3002 int fd, sz;
3003
3004 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
3005 return false;
3006 }
3007
3008 const char kw[] = "[stack]";
3009 const int kwlen = sizeof(kw)-1;
3010
3011 // Address part of /proc/self/maps couldn't be more than 128 bytes
3012 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
3013 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
3014 // Extract addresses
3015 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
3016 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
3017 if (sp >= *bottom && sp <= *top) {
3018 ::close(fd);
3019 return true;
3020 }
3021 }
3022 }
3023 }
3024
3025 ::close(fd);
3026 return false;
3027 }
3028
3029
3030 // If the (growable) stack mapping already extends beyond the point
3031 // where we're going to put our guard pages, truncate the mapping at
3032 // that point by munmap()ping it. This ensures that when we later
3033 // munmap() the guard pages we don't leave a hole in the stack
3034 // mapping. This only affects the main/initial thread, but guard
3035 // against future OS changes
3036 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3037 uintptr_t stack_extent, stack_base;
3038 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
3039 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
3040 assert(os::Linux::is_initial_thread(),
3041 "growable stack in non-initial thread");
3042 if (stack_extent < (uintptr_t)addr)
3043 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
3044 }
3045
3046 return os::commit_memory(addr, size, !ExecMem);
3047 }
3048
3049 // If this is a growable mapping, remove the guard pages entirely by
3050 // munmap()ping them. If not, just call uncommit_memory(). This only
3051 // affects the main/initial thread, but guard against future OS changes
3052 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3053 uintptr_t stack_extent, stack_base;
3054 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
3055 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
3056 assert(os::Linux::is_initial_thread(),
3057 "growable stack in non-initial thread");
3058
3059 return ::munmap(addr, size) == 0;
3060 }
3061
3062 return os::uncommit_memory(addr, size);
3063 }
3064
3065 static address _highest_vm_reserved_address = NULL;
3066
3067 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3068 // at 'requested_addr'. If there are existing memory mappings at the same
3069 // location, however, they will be overwritten. If 'fixed' is false,
3070 // 'requested_addr' is only treated as a hint, the return value may or
3071 // may not start from the requested address. Unlike Linux mmap(), this
3072 // function returns NULL to indicate failure.
3073 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3074 char * addr;
3075 int flags;
3076
3077 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3078 if (fixed) {
3079 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3080 flags |= MAP_FIXED;
3081 }
3082
3083 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3084 // touch an uncommitted page. Otherwise, the read/write might
3085 // succeed if we have enough swap space to back the physical page.
3086 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3087 flags, -1, 0);
3088
3089 if (addr != MAP_FAILED) {
3090 // anon_mmap() should only get called during VM initialization,
3091 // don't need lock (actually we can skip locking even it can be called
3092 // from multiple threads, because _highest_vm_reserved_address is just a
3093 // hint about the upper limit of non-stack memory regions.)
3094 if ((address)addr + bytes > _highest_vm_reserved_address) {
3095 _highest_vm_reserved_address = (address)addr + bytes;
3096 }
3097 }
3098
3099 return addr == MAP_FAILED ? NULL : addr;
3100 }
3101
3102 // Don't update _highest_vm_reserved_address, because there might be memory
3103 // regions above addr + size. If so, releasing a memory region only creates
3104 // a hole in the address space, it doesn't help prevent heap-stack collision.
3105 //
3106 static int anon_munmap(char * addr, size_t size) {
3107 return ::munmap(addr, size) == 0;
3108 }
3109
3110 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3111 size_t alignment_hint) {
3112 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3113 }
3114
3115 bool os::pd_release_memory(char* addr, size_t size) {
3116 return anon_munmap(addr, size);
3117 }
3118
3119 static address highest_vm_reserved_address() {
3120 return _highest_vm_reserved_address;
3121 }
3122
3123 static bool linux_mprotect(char* addr, size_t size, int prot) {
3124 // Linux wants the mprotect address argument to be page aligned.
3125 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3126
3127 // According to SUSv3, mprotect() should only be used with mappings
3128 // established by mmap(), and mmap() always maps whole pages. Unaligned
3129 // 'addr' likely indicates problem in the VM (e.g. trying to change
3130 // protection of malloc'ed or statically allocated memory). Check the
3131 // caller if you hit this assert.
3132 assert(addr == bottom, "sanity check");
3133
3134 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3135 return ::mprotect(bottom, size, prot) == 0;
3136 }
3137
3138 // Set protections specified
3139 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3140 bool is_committed) {
3141 unsigned int p = 0;
3142 switch (prot) {
3143 case MEM_PROT_NONE: p = PROT_NONE; break;
3144 case MEM_PROT_READ: p = PROT_READ; break;
3145 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3146 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3147 default:
3148 ShouldNotReachHere();
3149 }
3150 // is_committed is unused.
3151 return linux_mprotect(addr, bytes, p);
3152 }
3153
3154 bool os::guard_memory(char* addr, size_t size) {
3155 return linux_mprotect(addr, size, PROT_NONE);
3156 }
3157
3158 bool os::unguard_memory(char* addr, size_t size) {
3159 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3160 }
3161
3162 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3163 bool result = false;
3164 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
3165 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3166 -1, 0);
3167
3168 if (p != MAP_FAILED) {
3169 // We don't know if this really is a huge page or not.
3170 FILE *fp = fopen("/proc/self/maps", "r");
3171 if (fp) {
3172 while (!feof(fp)) {
3173 char chars[257];
3174 long x = 0;
3175 if (fgets(chars, sizeof(chars), fp)) {
3176 if (sscanf(chars, "%lx-%*x", &x) == 1
3177 && x == (long)p) {
3178 if (strstr (chars, "hugepage")) {
3179 result = true;
3180 break;
3181 }
3182 }
3183 }
3184 }
3185 fclose(fp);
3186 }
3187 munmap (p, page_size);
3188 if (result)
3189 return true;
3190 }
3191
3192 if (warn) {
3193 warning("HugeTLBFS is not supported by the operating system.");
3194 }
3195
3196 return result;
3197 }
3198
3199 /*
3200 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3201 *
3202 * From the coredump_filter documentation:
3203 *
3204 * - (bit 0) anonymous private memory
3205 * - (bit 1) anonymous shared memory
3206 * - (bit 2) file-backed private memory
3207 * - (bit 3) file-backed shared memory
3208 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3209 * effective only if the bit 2 is cleared)
3210 * - (bit 5) hugetlb private memory
3211 * - (bit 6) hugetlb shared memory
3212 */
3213 static void set_coredump_filter(void) {
3214 FILE *f;
3215 long cdm;
3216
3217 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3218 return;
3219 }
3220
3221 if (fscanf(f, "%lx", &cdm) != 1) {
3222 fclose(f);
3223 return;
3224 }
3225
3226 rewind(f);
3227
3228 if ((cdm & LARGEPAGES_BIT) == 0) {
3229 cdm |= LARGEPAGES_BIT;
3230 fprintf(f, "%#lx", cdm);
3231 }
3232
3233 fclose(f);
3234 }
3235
3236 // Large page support
3237
3238 static size_t _large_page_size = 0;
3239
3240 void os::large_page_init() {
3241 if (!UseLargePages) {
3242 UseHugeTLBFS = false;
3243 UseSHM = false;
3244 return;
3245 }
3246
3247 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3248 // If UseLargePages is specified on the command line try both methods,
3249 // if it's default, then try only HugeTLBFS.
3250 if (FLAG_IS_DEFAULT(UseLargePages)) {
3251 UseHugeTLBFS = true;
3252 } else {
3253 UseHugeTLBFS = UseSHM = true;
3254 }
3255 }
3256
3257 if (LargePageSizeInBytes) {
3258 _large_page_size = LargePageSizeInBytes;
3259 } else {
3260 // large_page_size on Linux is used to round up heap size. x86 uses either
3261 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3262 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3263 // page as large as 256M.
3264 //
3265 // Here we try to figure out page size by parsing /proc/meminfo and looking
3266 // for a line with the following format:
3267 // Hugepagesize: 2048 kB
3268 //
3269 // If we can't determine the value (e.g. /proc is not mounted, or the text
3270 // format has been changed), we'll use the largest page size supported by
3271 // the processor.
3272
3273 #ifndef ZERO
3274 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3275 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3276 #endif // ZERO
3277
3278 FILE *fp = fopen("/proc/meminfo", "r");
3279 if (fp) {
3280 while (!feof(fp)) {
3281 int x = 0;
3282 char buf[16];
3283 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3284 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3285 _large_page_size = x * K;
3286 break;
3287 }
3288 } else {
3289 // skip to next line
3290 for (;;) {
3291 int ch = fgetc(fp);
3292 if (ch == EOF || ch == (int)'\n') break;
3293 }
3294 }
3295 }
3296 fclose(fp);
3297 }
3298 }
3299
3300 // print a warning if any large page related flag is specified on command line
3301 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3302
3303 const size_t default_page_size = (size_t)Linux::page_size();
3304 if (_large_page_size > default_page_size) {
3305 _page_sizes[0] = _large_page_size;
3306 _page_sizes[1] = default_page_size;
3307 _page_sizes[2] = 0;
3308 }
3309 UseHugeTLBFS = UseHugeTLBFS &&
3310 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3311
3312 if (UseHugeTLBFS)
3313 UseSHM = false;
3314
3315 UseLargePages = UseHugeTLBFS || UseSHM;
3316
3317 set_coredump_filter();
3318 }
3319
3320 #ifndef SHM_HUGETLB
3321 #define SHM_HUGETLB 04000
3322 #endif
3323
3324 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3325 // "exec" is passed in but not used. Creating the shared image for
3326 // the code cache doesn't have an SHM_X executable permission to check.
3327 assert(UseLargePages && UseSHM, "only for SHM large pages");
3328
3329 key_t key = IPC_PRIVATE;
3330 char *addr;
3331
3332 bool warn_on_failure = UseLargePages &&
3333 (!FLAG_IS_DEFAULT(UseLargePages) ||
3334 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3335 );
3336 char msg[128];
3337
3338 // Create a large shared memory region to attach to based on size.
3339 // Currently, size is the total size of the heap
3340 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3341 if (shmid == -1) {
3342 // Possible reasons for shmget failure:
3343 // 1. shmmax is too small for Java heap.
3344 // > check shmmax value: cat /proc/sys/kernel/shmmax
3345 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3346 // 2. not enough large page memory.
3347 // > check available large pages: cat /proc/meminfo
3348 // > increase amount of large pages:
3349 // echo new_value > /proc/sys/vm/nr_hugepages
3350 // Note 1: different Linux may use different name for this property,
3351 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3352 // Note 2: it's possible there's enough physical memory available but
3353 // they are so fragmented after a long run that they can't
3354 // coalesce into large pages. Try to reserve large pages when
3355 // the system is still "fresh".
3356 if (warn_on_failure) {
3357 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3358 warning(msg);
3359 }
3360 return NULL;
3361 }
3362
3363 // attach to the region
3364 addr = (char*)shmat(shmid, req_addr, 0);
3365 int err = errno;
3366
3367 // Remove shmid. If shmat() is successful, the actual shared memory segment
3368 // will be deleted when it's detached by shmdt() or when the process
3369 // terminates. If shmat() is not successful this will remove the shared
3370 // segment immediately.
3371 shmctl(shmid, IPC_RMID, NULL);
3372
3373 if ((intptr_t)addr == -1) {
3374 if (warn_on_failure) {
3375 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3376 warning(msg);
3377 }
3378 return NULL;
3379 }
3380
3381 if ((addr != NULL) && UseNUMAInterleaving) {
3382 numa_make_global(addr, bytes);
3383 }
3384
3385 // The memory is committed
3386 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3387
3388 return addr;
3389 }
3390
3391 bool os::release_memory_special(char* base, size_t bytes) {
3392 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3393 // detaching the SHM segment will also delete it, see reserve_memory_special()
3394 int rslt = shmdt(base);
3395 if (rslt == 0) {
3396 tkr.record((address)base, bytes);
3397 return true;
3398 } else {
3399 tkr.discard();
3400 return false;
3401 }
3402 }
3403
3404 size_t os::large_page_size() {
3405 return _large_page_size;
3406 }
3407
3408 // HugeTLBFS allows application to commit large page memory on demand;
3409 // with SysV SHM the entire memory region must be allocated as shared
3410 // memory.
3411 bool os::can_commit_large_page_memory() {
3412 return UseHugeTLBFS;
3413 }
3414
3415 bool os::can_execute_large_page_memory() {
3416 return UseHugeTLBFS;
3417 }
3418
3419 // Reserve memory at an arbitrary address, only if that area is
3420 // available (and not reserved for something else).
3421
3422 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3423 const int max_tries = 10;
3424 char* base[max_tries];
3425 size_t size[max_tries];
3426 const size_t gap = 0x000000;
3427
3428 // Assert only that the size is a multiple of the page size, since
3429 // that's all that mmap requires, and since that's all we really know
3430 // about at this low abstraction level. If we need higher alignment,
3431 // we can either pass an alignment to this method or verify alignment
3432 // in one of the methods further up the call chain. See bug 5044738.
3433 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3434
3435 // Repeatedly allocate blocks until the block is allocated at the
3436 // right spot. Give up after max_tries. Note that reserve_memory() will
3437 // automatically update _highest_vm_reserved_address if the call is
3438 // successful. The variable tracks the highest memory address every reserved
3439 // by JVM. It is used to detect heap-stack collision if running with
3440 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3441 // space than needed, it could confuse the collision detecting code. To
3442 // solve the problem, save current _highest_vm_reserved_address and
3443 // calculate the correct value before return.
3444 address old_highest = _highest_vm_reserved_address;
3445
3446 // Linux mmap allows caller to pass an address as hint; give it a try first,
3447 // if kernel honors the hint then we can return immediately.
3448 char * addr = anon_mmap(requested_addr, bytes, false);
3449 if (addr == requested_addr) {
3450 return requested_addr;
3451 }
3452
3453 if (addr != NULL) {
3454 // mmap() is successful but it fails to reserve at the requested address
3455 anon_munmap(addr, bytes);
3456 }
3457
3458 int i;
3459 for (i = 0; i < max_tries; ++i) {
3460 base[i] = reserve_memory(bytes);
3461
3462 if (base[i] != NULL) {
3463 // Is this the block we wanted?
3464 if (base[i] == requested_addr) {
3465 size[i] = bytes;
3466 break;
3467 }
3468
3469 // Does this overlap the block we wanted? Give back the overlapped
3470 // parts and try again.
3471
3472 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3473 if (top_overlap >= 0 && top_overlap < bytes) {
3474 unmap_memory(base[i], top_overlap);
3475 base[i] += top_overlap;
3476 size[i] = bytes - top_overlap;
3477 } else {
3478 size_t bottom_overlap = base[i] + bytes - requested_addr;
3479 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3480 unmap_memory(requested_addr, bottom_overlap);
3481 size[i] = bytes - bottom_overlap;
3482 } else {
3483 size[i] = bytes;
3484 }
3485 }
3486 }
3487 }
3488
3489 // Give back the unused reserved pieces.
3490
3491 for (int j = 0; j < i; ++j) {
3492 if (base[j] != NULL) {
3493 unmap_memory(base[j], size[j]);
3494 }
3495 }
3496
3497 if (i < max_tries) {
3498 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3499 return requested_addr;
3500 } else {
3501 _highest_vm_reserved_address = old_highest;
3502 return NULL;
3503 }
3504 }
3505
3506 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3507 return ::read(fd, buf, nBytes);
3508 }
3509
3510 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3511 // Solaris uses poll(), linux uses park().
3512 // Poll() is likely a better choice, assuming that Thread.interrupt()
3513 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3514 // SIGSEGV, see 4355769.
3515
3516 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3517 assert(thread == Thread::current(), "thread consistency check");
3518
3519 ParkEvent * const slp = thread->_SleepEvent ;
3520 slp->reset() ;
3521 OrderAccess::fence() ;
3522
3523 if (interruptible) {
3524 jlong prevtime = javaTimeNanos();
3525
3526 for (;;) {
3527 if (os::is_interrupted(thread, true)) {
3528 return OS_INTRPT;
3529 }
3530
3531 jlong newtime = javaTimeNanos();
3532
3533 if (newtime - prevtime < 0) {
3534 // time moving backwards, should only happen if no monotonic clock
3535 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3536 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3537 } else {
3538 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3539 }
3540
3541 if(millis <= 0) {
3542 return OS_OK;
3543 }
3544
3545 prevtime = newtime;
3546
3547 {
3548 assert(thread->is_Java_thread(), "sanity check");
3549 JavaThread *jt = (JavaThread *) thread;
3550 ThreadBlockInVM tbivm(jt);
3551 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3552
3553 jt->set_suspend_equivalent();
3554 // cleared by handle_special_suspend_equivalent_condition() or
3555 // java_suspend_self() via check_and_wait_while_suspended()
3556
3557 slp->park(millis);
3558
3559 // were we externally suspended while we were waiting?
3560 jt->check_and_wait_while_suspended();
3561 }
3562 }
3563 } else {
3564 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3565 jlong prevtime = javaTimeNanos();
3566
3567 for (;;) {
3568 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3569 // the 1st iteration ...
3570 jlong newtime = javaTimeNanos();
3571
3572 if (newtime - prevtime < 0) {
3573 // time moving backwards, should only happen if no monotonic clock
3574 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3575 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3576 } else {
3577 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3578 }
3579
3580 if(millis <= 0) break ;
3581
3582 prevtime = newtime;
3583 slp->park(millis);
3584 }
3585 return OS_OK ;
3586 }
3587 }
3588
3589 int os::naked_sleep() {
3590 // %% make the sleep time an integer flag. for now use 1 millisec.
3591 return os::sleep(Thread::current(), 1, false);
3592 }
3593
3594 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3595 void os::infinite_sleep() {
3596 while (true) { // sleep forever ...
3597 ::sleep(100); // ... 100 seconds at a time
3598 }
3599 }
3600
3601 // Used to convert frequent JVM_Yield() to nops
3602 bool os::dont_yield() {
3603 return DontYieldALot;
3604 }
3605
3606 void os::yield() {
3607 sched_yield();
3608 }
3609
3610 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3611
3612 void os::yield_all(int attempts) {
3613 // Yields to all threads, including threads with lower priorities
3614 // Threads on Linux are all with same priority. The Solaris style
3615 // os::yield_all() with nanosleep(1ms) is not necessary.
3616 sched_yield();
3617 }
3618
3619 // Called from the tight loops to possibly influence time-sharing heuristics
3620 void os::loop_breaker(int attempts) {
3621 os::yield_all(attempts);
3622 }
3623
3624 ////////////////////////////////////////////////////////////////////////////////
3625 // thread priority support
3626
3627 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3628 // only supports dynamic priority, static priority must be zero. For real-time
3629 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3630 // However, for large multi-threaded applications, SCHED_RR is not only slower
3631 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3632 // of 5 runs - Sep 2005).
3633 //
3634 // The following code actually changes the niceness of kernel-thread/LWP. It
3635 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3636 // not the entire user process, and user level threads are 1:1 mapped to kernel
3637 // threads. It has always been the case, but could change in the future. For
3638 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3639 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3640
3641 int os::java_to_os_priority[CriticalPriority + 1] = {
3642 19, // 0 Entry should never be used
3643
3644 4, // 1 MinPriority
3645 3, // 2
3646 2, // 3
3647
3648 1, // 4
3649 0, // 5 NormPriority
3650 -1, // 6
3651
3652 -2, // 7
3653 -3, // 8
3654 -4, // 9 NearMaxPriority
3655
3656 -5, // 10 MaxPriority
3657
3658 -5 // 11 CriticalPriority
3659 };
3660
3661 static int prio_init() {
3662 if (ThreadPriorityPolicy == 1) {
3663 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3664 // if effective uid is not root. Perhaps, a more elegant way of doing
3665 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3666 if (geteuid() != 0) {
3667 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3668 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3669 }
3670 ThreadPriorityPolicy = 0;
3671 }
3672 }
3673 if (UseCriticalJavaThreadPriority) {
3674 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3675 }
3676 return 0;
3677 }
3678
3679 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3680 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3681
3682 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3683 return (ret == 0) ? OS_OK : OS_ERR;
3684 }
3685
3686 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3687 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3688 *priority_ptr = java_to_os_priority[NormPriority];
3689 return OS_OK;
3690 }
3691
3692 errno = 0;
3693 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3694 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3695 }
3696
3697 // Hint to the underlying OS that a task switch would not be good.
3698 // Void return because it's a hint and can fail.
3699 void os::hint_no_preempt() {}
3700
3701 ////////////////////////////////////////////////////////////////////////////////
3702 // suspend/resume support
3703
3704 // the low-level signal-based suspend/resume support is a remnant from the
3705 // old VM-suspension that used to be for java-suspension, safepoints etc,
3706 // within hotspot. Now there is a single use-case for this:
3707 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3708 // that runs in the watcher thread.
3709 // The remaining code is greatly simplified from the more general suspension
3710 // code that used to be used.
3711 //
3712 // The protocol is quite simple:
3713 // - suspend:
3714 // - sends a signal to the target thread
3715 // - polls the suspend state of the osthread using a yield loop
3716 // - target thread signal handler (SR_handler) sets suspend state
3717 // and blocks in sigsuspend until continued
3718 // - resume:
3719 // - sets target osthread state to continue
3720 // - sends signal to end the sigsuspend loop in the SR_handler
3721 //
3722 // Note that the SR_lock plays no role in this suspend/resume protocol.
3723 //
3724
3725 static void resume_clear_context(OSThread *osthread) {
3726 osthread->set_ucontext(NULL);
3727 osthread->set_siginfo(NULL);
3728 }
3729
3730 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3731 osthread->set_ucontext(context);
3732 osthread->set_siginfo(siginfo);
3733 }
3734
3735 //
3736 // Handler function invoked when a thread's execution is suspended or
3737 // resumed. We have to be careful that only async-safe functions are
3738 // called here (Note: most pthread functions are not async safe and
3739 // should be avoided.)
3740 //
3741 // Note: sigwait() is a more natural fit than sigsuspend() from an
3742 // interface point of view, but sigwait() prevents the signal hander
3743 // from being run. libpthread would get very confused by not having
3744 // its signal handlers run and prevents sigwait()'s use with the
3745 // mutex granting granting signal.
3746 //
3747 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3748 //
3749 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3750 // Save and restore errno to avoid confusing native code with EINTR
3751 // after sigsuspend.
3752 int old_errno = errno;
3753
3754 Thread* thread = Thread::current();
3755 OSThread* osthread = thread->osthread();
3756 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3757
3758 os::SuspendResume::State current = osthread->sr.state();
3759 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3760 suspend_save_context(osthread, siginfo, context);
3761
3762 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3763 os::SuspendResume::State state = osthread->sr.suspended();
3764 if (state == os::SuspendResume::SR_SUSPENDED) {
3765 sigset_t suspend_set; // signals for sigsuspend()
3766
3767 // get current set of blocked signals and unblock resume signal
3768 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3769 sigdelset(&suspend_set, SR_signum);
3770
3771 sr_semaphore.signal();
3772 // wait here until we are resumed
3773 while (1) {
3774 sigsuspend(&suspend_set);
3775
3776 os::SuspendResume::State result = osthread->sr.running();
3777 if (result == os::SuspendResume::SR_RUNNING) {
3778 sr_semaphore.signal();
3779 break;
3780 }
3781 }
3782
3783 } else if (state == os::SuspendResume::SR_RUNNING) {
3784 // request was cancelled, continue
3785 } else {
3786 ShouldNotReachHere();
3787 }
3788
3789 resume_clear_context(osthread);
3790 } else if (current == os::SuspendResume::SR_RUNNING) {
3791 // request was cancelled, continue
3792 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3793 // ignore
3794 } else {
3795 // ignore
3796 }
3797
3798 errno = old_errno;
3799 }
3800
3801
3802 static int SR_initialize() {
3803 struct sigaction act;
3804 char *s;
3805 /* Get signal number to use for suspend/resume */
3806 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3807 int sig = ::strtol(s, 0, 10);
3808 if (sig > 0 || sig < _NSIG) {
3809 SR_signum = sig;
3810 }
3811 }
3812
3813 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3814 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3815
3816 sigemptyset(&SR_sigset);
3817 sigaddset(&SR_sigset, SR_signum);
3818
3819 /* Set up signal handler for suspend/resume */
3820 act.sa_flags = SA_RESTART|SA_SIGINFO;
3821 act.sa_handler = (void (*)(int)) SR_handler;
3822
3823 // SR_signum is blocked by default.
3824 // 4528190 - We also need to block pthread restart signal (32 on all
3825 // supported Linux platforms). Note that LinuxThreads need to block
3826 // this signal for all threads to work properly. So we don't have
3827 // to use hard-coded signal number when setting up the mask.
3828 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3829
3830 if (sigaction(SR_signum, &act, 0) == -1) {
3831 return -1;
3832 }
3833
3834 // Save signal flag
3835 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3836 return 0;
3837 }
3838
3839 static int sr_notify(OSThread* osthread) {
3840 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3841 assert_status(status == 0, status, "pthread_kill");
3842 return status;
3843 }
3844
3845 // "Randomly" selected value for how long we want to spin
3846 // before bailing out on suspending a thread, also how often
3847 // we send a signal to a thread we want to resume
3848 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3849 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3850
3851 // returns true on success and false on error - really an error is fatal
3852 // but this seems the normal response to library errors
3853 static bool do_suspend(OSThread* osthread) {
3854 assert(osthread->sr.is_running(), "thread should be running");
3855 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3856
3857 // mark as suspended and send signal
3858 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3859 // failed to switch, state wasn't running?
3860 ShouldNotReachHere();
3861 return false;
3862 }
3863
3864 if (sr_notify(osthread) != 0) {
3865 ShouldNotReachHere();
3866 }
3867
3868 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3869 while (true) {
3870 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3871 break;
3872 } else {
3873 // timeout
3874 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3875 if (cancelled == os::SuspendResume::SR_RUNNING) {
3876 return false;
3877 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3878 // make sure that we consume the signal on the semaphore as well
3879 sr_semaphore.wait();
3880 break;
3881 } else {
3882 ShouldNotReachHere();
3883 return false;
3884 }
3885 }
3886 }
3887
3888 guarantee(osthread->sr.is_suspended(), "Must be suspended");
3889 return true;
3890 }
3891
3892 static void do_resume(OSThread* osthread) {
3893 assert(osthread->sr.is_suspended(), "thread should be suspended");
3894 assert(!sr_semaphore.trywait(), "invalid semaphore state");
3895
3896 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3897 // failed to switch to WAKEUP_REQUEST
3898 ShouldNotReachHere();
3899 return;
3900 }
3901
3902 while (true) {
3903 if (sr_notify(osthread) == 0) {
3904 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3905 if (osthread->sr.is_running()) {
3906 return;
3907 }
3908 }
3909 } else {
3910 ShouldNotReachHere();
3911 }
3912 }
3913
3914 guarantee(osthread->sr.is_running(), "Must be running!");
3915 }
3916
3917 ////////////////////////////////////////////////////////////////////////////////
3918 // interrupt support
3919
3920 void os::interrupt(Thread* thread) {
3921 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3922 "possibility of dangling Thread pointer");
3923
3924 OSThread* osthread = thread->osthread();
3925
3926 if (!osthread->interrupted()) {
3927 osthread->set_interrupted(true);
3928 // More than one thread can get here with the same value of osthread,
3929 // resulting in multiple notifications. We do, however, want the store
3930 // to interrupted() to be visible to other threads before we execute unpark().
3931 OrderAccess::fence();
3932 ParkEvent * const slp = thread->_SleepEvent ;
3933 if (slp != NULL) slp->unpark() ;
3934 }
3935
3936 // For JSR166. Unpark even if interrupt status already was set
3937 if (thread->is_Java_thread())
3938 ((JavaThread*)thread)->parker()->unpark();
3939
3940 ParkEvent * ev = thread->_ParkEvent ;
3941 if (ev != NULL) ev->unpark() ;
3942
3943 }
3944
3945 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3946 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3947 "possibility of dangling Thread pointer");
3948
3949 OSThread* osthread = thread->osthread();
3950
3951 bool interrupted = osthread->interrupted();
3952
3953 if (interrupted && clear_interrupted) {
3954 osthread->set_interrupted(false);
3955 // consider thread->_SleepEvent->reset() ... optional optimization
3956 }
3957
3958 return interrupted;
3959 }
3960
3961 ///////////////////////////////////////////////////////////////////////////////////
3962 // signal handling (except suspend/resume)
3963
3964 // This routine may be used by user applications as a "hook" to catch signals.
3965 // The user-defined signal handler must pass unrecognized signals to this
3966 // routine, and if it returns true (non-zero), then the signal handler must
3967 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3968 // routine will never retun false (zero), but instead will execute a VM panic
3969 // routine kill the process.
3970 //
3971 // If this routine returns false, it is OK to call it again. This allows
3972 // the user-defined signal handler to perform checks either before or after
3973 // the VM performs its own checks. Naturally, the user code would be making
3974 // a serious error if it tried to handle an exception (such as a null check
3975 // or breakpoint) that the VM was generating for its own correct operation.
3976 //
3977 // This routine may recognize any of the following kinds of signals:
3978 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3979 // It should be consulted by handlers for any of those signals.
3980 //
3981 // The caller of this routine must pass in the three arguments supplied
3982 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3983 // field of the structure passed to sigaction(). This routine assumes that
3984 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3985 //
3986 // Note that the VM will print warnings if it detects conflicting signal
3987 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3988 //
3989 extern "C" JNIEXPORT int
3990 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3991 void* ucontext, int abort_if_unrecognized);
3992
3993 void signalHandler(int sig, siginfo_t* info, void* uc) {
3994 assert(info != NULL && uc != NULL, "it must be old kernel");
3995 int orig_errno = errno; // Preserve errno value over signal handler.
3996 JVM_handle_linux_signal(sig, info, uc, true);
3997 errno = orig_errno;
3998 }
3999
4000
4001 // This boolean allows users to forward their own non-matching signals
4002 // to JVM_handle_linux_signal, harmlessly.
4003 bool os::Linux::signal_handlers_are_installed = false;
4004
4005 // For signal-chaining
4006 struct sigaction os::Linux::sigact[MAXSIGNUM];
4007 unsigned int os::Linux::sigs = 0;
4008 bool os::Linux::libjsig_is_loaded = false;
4009 typedef struct sigaction *(*get_signal_t)(int);
4010 get_signal_t os::Linux::get_signal_action = NULL;
4011
4012 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4013 struct sigaction *actp = NULL;
4014
4015 if (libjsig_is_loaded) {
4016 // Retrieve the old signal handler from libjsig
4017 actp = (*get_signal_action)(sig);
4018 }
4019 if (actp == NULL) {
4020 // Retrieve the preinstalled signal handler from jvm
4021 actp = get_preinstalled_handler(sig);
4022 }
4023
4024 return actp;
4025 }
4026
4027 static bool call_chained_handler(struct sigaction *actp, int sig,
4028 siginfo_t *siginfo, void *context) {
4029 // Call the old signal handler
4030 if (actp->sa_handler == SIG_DFL) {
4031 // It's more reasonable to let jvm treat it as an unexpected exception
4032 // instead of taking the default action.
4033 return false;
4034 } else if (actp->sa_handler != SIG_IGN) {
4035 if ((actp->sa_flags & SA_NODEFER) == 0) {
4036 // automaticlly block the signal
4037 sigaddset(&(actp->sa_mask), sig);
4038 }
4039
4040 sa_handler_t hand;
4041 sa_sigaction_t sa;
4042 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4043 // retrieve the chained handler
4044 if (siginfo_flag_set) {
4045 sa = actp->sa_sigaction;
4046 } else {
4047 hand = actp->sa_handler;
4048 }
4049
4050 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4051 actp->sa_handler = SIG_DFL;
4052 }
4053
4054 // try to honor the signal mask
4055 sigset_t oset;
4056 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4057
4058 // call into the chained handler
4059 if (siginfo_flag_set) {
4060 (*sa)(sig, siginfo, context);
4061 } else {
4062 (*hand)(sig);
4063 }
4064
4065 // restore the signal mask
4066 pthread_sigmask(SIG_SETMASK, &oset, 0);
4067 }
4068 // Tell jvm's signal handler the signal is taken care of.
4069 return true;
4070 }
4071
4072 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4073 bool chained = false;
4074 // signal-chaining
4075 if (UseSignalChaining) {
4076 struct sigaction *actp = get_chained_signal_action(sig);
4077 if (actp != NULL) {
4078 chained = call_chained_handler(actp, sig, siginfo, context);
4079 }
4080 }
4081 return chained;
4082 }
4083
4084 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4085 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4086 return &sigact[sig];
4087 }
4088 return NULL;
4089 }
4090
4091 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4092 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4093 sigact[sig] = oldAct;
4094 sigs |= (unsigned int)1 << sig;
4095 }
4096
4097 // for diagnostic
4098 int os::Linux::sigflags[MAXSIGNUM];
4099
4100 int os::Linux::get_our_sigflags(int sig) {
4101 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4102 return sigflags[sig];
4103 }
4104
4105 void os::Linux::set_our_sigflags(int sig, int flags) {
4106 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4107 sigflags[sig] = flags;
4108 }
4109
4110 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4111 // Check for overwrite.
4112 struct sigaction oldAct;
4113 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4114
4115 void* oldhand = oldAct.sa_sigaction
4116 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4117 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4118 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4119 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4120 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4121 if (AllowUserSignalHandlers || !set_installed) {
4122 // Do not overwrite; user takes responsibility to forward to us.
4123 return;
4124 } else if (UseSignalChaining) {
4125 // save the old handler in jvm
4126 save_preinstalled_handler(sig, oldAct);
4127 // libjsig also interposes the sigaction() call below and saves the
4128 // old sigaction on it own.
4129 } else {
4130 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4131 "%#lx for signal %d.", (long)oldhand, sig));
4132 }
4133 }
4134
4135 struct sigaction sigAct;
4136 sigfillset(&(sigAct.sa_mask));
4137 sigAct.sa_handler = SIG_DFL;
4138 if (!set_installed) {
4139 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4140 } else {
4141 sigAct.sa_sigaction = signalHandler;
4142 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4143 }
4144 // Save flags, which are set by ours
4145 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4146 sigflags[sig] = sigAct.sa_flags;
4147
4148 int ret = sigaction(sig, &sigAct, &oldAct);
4149 assert(ret == 0, "check");
4150
4151 void* oldhand2 = oldAct.sa_sigaction
4152 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4153 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4154 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4155 }
4156
4157 // install signal handlers for signals that HotSpot needs to
4158 // handle in order to support Java-level exception handling.
4159
4160 void os::Linux::install_signal_handlers() {
4161 if (!signal_handlers_are_installed) {
4162 signal_handlers_are_installed = true;
4163
4164 // signal-chaining
4165 typedef void (*signal_setting_t)();
4166 signal_setting_t begin_signal_setting = NULL;
4167 signal_setting_t end_signal_setting = NULL;
4168 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4169 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4170 if (begin_signal_setting != NULL) {
4171 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4172 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4173 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4174 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4175 libjsig_is_loaded = true;
4176 assert(UseSignalChaining, "should enable signal-chaining");
4177 }
4178 if (libjsig_is_loaded) {
4179 // Tell libjsig jvm is setting signal handlers
4180 (*begin_signal_setting)();
4181 }
4182
4183 set_signal_handler(SIGSEGV, true);
4184 set_signal_handler(SIGPIPE, true);
4185 set_signal_handler(SIGBUS, true);
4186 set_signal_handler(SIGILL, true);
4187 set_signal_handler(SIGFPE, true);
4188 set_signal_handler(SIGXFSZ, true);
4189
4190 if (libjsig_is_loaded) {
4191 // Tell libjsig jvm finishes setting signal handlers
4192 (*end_signal_setting)();
4193 }
4194
4195 // We don't activate signal checker if libjsig is in place, we trust ourselves
4196 // and if UserSignalHandler is installed all bets are off.
4197 // Log that signal checking is off only if -verbose:jni is specified.
4198 if (CheckJNICalls) {
4199 if (libjsig_is_loaded) {
4200 if (PrintJNIResolving) {
4201 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4202 }
4203 check_signals = false;
4204 }
4205 if (AllowUserSignalHandlers) {
4206 if (PrintJNIResolving) {
4207 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4208 }
4209 check_signals = false;
4210 }
4211 }
4212 }
4213 }
4214
4215 // This is the fastest way to get thread cpu time on Linux.
4216 // Returns cpu time (user+sys) for any thread, not only for current.
4217 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4218 // It might work on 2.6.10+ with a special kernel/glibc patch.
4219 // For reference, please, see IEEE Std 1003.1-2004:
4220 // http://www.unix.org/single_unix_specification
4221
4222 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4223 struct timespec tp;
4224 int rc = os::Linux::clock_gettime(clockid, &tp);
4225 assert(rc == 0, "clock_gettime is expected to return 0 code");
4226
4227 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4228 }
4229
4230 /////
4231 // glibc on Linux platform uses non-documented flag
4232 // to indicate, that some special sort of signal
4233 // trampoline is used.
4234 // We will never set this flag, and we should
4235 // ignore this flag in our diagnostic
4236 #ifdef SIGNIFICANT_SIGNAL_MASK
4237 #undef SIGNIFICANT_SIGNAL_MASK
4238 #endif
4239 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4240
4241 static const char* get_signal_handler_name(address handler,
4242 char* buf, int buflen) {
4243 int offset;
4244 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4245 if (found) {
4246 // skip directory names
4247 const char *p1, *p2;
4248 p1 = buf;
4249 size_t len = strlen(os::file_separator());
4250 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4251 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4252 } else {
4253 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4254 }
4255 return buf;
4256 }
4257
4258 static void print_signal_handler(outputStream* st, int sig,
4259 char* buf, size_t buflen) {
4260 struct sigaction sa;
4261
4262 sigaction(sig, NULL, &sa);
4263
4264 // See comment for SIGNIFICANT_SIGNAL_MASK define
4265 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4266
4267 st->print("%s: ", os::exception_name(sig, buf, buflen));
4268
4269 address handler = (sa.sa_flags & SA_SIGINFO)
4270 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4271 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4272
4273 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4274 st->print("SIG_DFL");
4275 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4276 st->print("SIG_IGN");
4277 } else {
4278 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4279 }
4280
4281 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4282
4283 address rh = VMError::get_resetted_sighandler(sig);
4284 // May be, handler was resetted by VMError?
4285 if(rh != NULL) {
4286 handler = rh;
4287 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4288 }
4289
4290 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
4291
4292 // Check: is it our handler?
4293 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4294 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4295 // It is our signal handler
4296 // check for flags, reset system-used one!
4297 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4298 st->print(
4299 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4300 os::Linux::get_our_sigflags(sig));
4301 }
4302 }
4303 st->cr();
4304 }
4305
4306
4307 #define DO_SIGNAL_CHECK(sig) \
4308 if (!sigismember(&check_signal_done, sig)) \
4309 os::Linux::check_signal_handler(sig)
4310
4311 // This method is a periodic task to check for misbehaving JNI applications
4312 // under CheckJNI, we can add any periodic checks here
4313
4314 void os::run_periodic_checks() {
4315
4316 if (check_signals == false) return;
4317
4318 // SEGV and BUS if overridden could potentially prevent
4319 // generation of hs*.log in the event of a crash, debugging
4320 // such a case can be very challenging, so we absolutely
4321 // check the following for a good measure:
4322 DO_SIGNAL_CHECK(SIGSEGV);
4323 DO_SIGNAL_CHECK(SIGILL);
4324 DO_SIGNAL_CHECK(SIGFPE);
4325 DO_SIGNAL_CHECK(SIGBUS);
4326 DO_SIGNAL_CHECK(SIGPIPE);
4327 DO_SIGNAL_CHECK(SIGXFSZ);
4328
4329
4330 // ReduceSignalUsage allows the user to override these handlers
4331 // see comments at the very top and jvm_solaris.h
4332 if (!ReduceSignalUsage) {
4333 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4334 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4335 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4336 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4337 }
4338
4339 DO_SIGNAL_CHECK(SR_signum);
4340 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4341 }
4342
4343 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4344
4345 static os_sigaction_t os_sigaction = NULL;
4346
4347 void os::Linux::check_signal_handler(int sig) {
4348 char buf[O_BUFLEN];
4349 address jvmHandler = NULL;
4350
4351
4352 struct sigaction act;
4353 if (os_sigaction == NULL) {
4354 // only trust the default sigaction, in case it has been interposed
4355 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4356 if (os_sigaction == NULL) return;
4357 }
4358
4359 os_sigaction(sig, (struct sigaction*)NULL, &act);
4360
4361
4362 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4363
4364 address thisHandler = (act.sa_flags & SA_SIGINFO)
4365 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4366 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4367
4368
4369 switch(sig) {
4370 case SIGSEGV:
4371 case SIGBUS:
4372 case SIGFPE:
4373 case SIGPIPE:
4374 case SIGILL:
4375 case SIGXFSZ:
4376 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4377 break;
4378
4379 case SHUTDOWN1_SIGNAL:
4380 case SHUTDOWN2_SIGNAL:
4381 case SHUTDOWN3_SIGNAL:
4382 case BREAK_SIGNAL:
4383 jvmHandler = (address)user_handler();
4384 break;
4385
4386 case INTERRUPT_SIGNAL:
4387 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4388 break;
4389
4390 default:
4391 if (sig == SR_signum) {
4392 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4393 } else {
4394 return;
4395 }
4396 break;
4397 }
4398
4399 if (thisHandler != jvmHandler) {
4400 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4401 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4402 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4403 // No need to check this sig any longer
4404 sigaddset(&check_signal_done, sig);
4405 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4406 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4407 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4408 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4409 // No need to check this sig any longer
4410 sigaddset(&check_signal_done, sig);
4411 }
4412
4413 // Dump all the signal
4414 if (sigismember(&check_signal_done, sig)) {
4415 print_signal_handlers(tty, buf, O_BUFLEN);
4416 }
4417 }
4418
4419 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4420
4421 extern bool signal_name(int signo, char* buf, size_t len);
4422
4423 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4424 if (0 < exception_code && exception_code <= SIGRTMAX) {
4425 // signal
4426 if (!signal_name(exception_code, buf, size)) {
4427 jio_snprintf(buf, size, "SIG%d", exception_code);
4428 }
4429 return buf;
4430 } else {
4431 return NULL;
4432 }
4433 }
4434
4435 // this is called _before_ the most of global arguments have been parsed
4436 void os::init(void) {
4437 char dummy; /* used to get a guess on initial stack address */
4438 // first_hrtime = gethrtime();
4439
4440 // With LinuxThreads the JavaMain thread pid (primordial thread)
4441 // is different than the pid of the java launcher thread.
4442 // So, on Linux, the launcher thread pid is passed to the VM
4443 // via the sun.java.launcher.pid property.
4444 // Use this property instead of getpid() if it was correctly passed.
4445 // See bug 6351349.
4446 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4447
4448 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4449
4450 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4451
4452 init_random(1234567);
4453
4454 ThreadCritical::initialize();
4455
4456 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4457 if (Linux::page_size() == -1) {
4458 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4459 strerror(errno)));
4460 }
4461 init_page_sizes((size_t) Linux::page_size());
4462
4463 Linux::initialize_system_info();
4464
4465 // main_thread points to the aboriginal thread
4466 Linux::_main_thread = pthread_self();
4467
4468 Linux::clock_init();
4469 initial_time_count = os::elapsed_counter();
4470 pthread_mutex_init(&dl_mutex, NULL);
4471
4472 // If the pagesize of the VM is greater than 8K determine the appropriate
4473 // number of initial guard pages. The user can change this with the
4474 // command line arguments, if needed.
4475 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4476 StackYellowPages = 1;
4477 StackRedPages = 1;
4478 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4479 }
4480 }
4481
4482 // To install functions for atexit system call
4483 extern "C" {
4484 static void perfMemory_exit_helper() {
4485 perfMemory_exit();
4486 }
4487 }
4488
4489 // this is called _after_ the global arguments have been parsed
4490 jint os::init_2(void)
4491 {
4492 Linux::fast_thread_clock_init();
4493
4494 // Allocate a single page and mark it as readable for safepoint polling
4495 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4496 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4497
4498 os::set_polling_page( polling_page );
4499
4500 #ifndef PRODUCT
4501 if(Verbose && PrintMiscellaneous)
4502 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4503 #endif
4504
4505 if (!UseMembar) {
4506 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4507 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4508 os::set_memory_serialize_page( mem_serialize_page );
4509
4510 #ifndef PRODUCT
4511 if(Verbose && PrintMiscellaneous)
4512 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4513 #endif
4514 }
4515
4516 os::large_page_init();
4517
4518 // initialize suspend/resume support - must do this before signal_sets_init()
4519 if (SR_initialize() != 0) {
4520 perror("SR_initialize failed");
4521 return JNI_ERR;
4522 }
4523
4524 Linux::signal_sets_init();
4525 Linux::install_signal_handlers();
4526
4527 // Check minimum allowable stack size for thread creation and to initialize
4528 // the java system classes, including StackOverflowError - depends on page
4529 // size. Add a page for compiler2 recursion in main thread.
4530 // Add in 2*BytesPerWord times page size to account for VM stack during
4531 // class initialization depending on 32 or 64 bit VM.
4532 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4533 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4534 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4535
4536 size_t threadStackSizeInBytes = ThreadStackSize * K;
4537 if (threadStackSizeInBytes != 0 &&
4538 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4539 tty->print_cr("\nThe stack size specified is too small, "
4540 "Specify at least %dk",
4541 os::Linux::min_stack_allowed/ K);
4542 return JNI_ERR;
4543 }
4544
4545 // Make the stack size a multiple of the page size so that
4546 // the yellow/red zones can be guarded.
4547 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4548 vm_page_size()));
4549
4550 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4551
4552 Linux::libpthread_init();
4553 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4554 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4555 Linux::glibc_version(), Linux::libpthread_version(),
4556 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4557 }
4558
4559 if (UseNUMA) {
4560 if (!Linux::libnuma_init()) {
4561 UseNUMA = false;
4562 } else {
4563 if ((Linux::numa_max_node() < 1)) {
4564 // There's only one node(they start from 0), disable NUMA.
4565 UseNUMA = false;
4566 }
4567 }
4568 // With SHM large pages we cannot uncommit a page, so there's not way
4569 // we can make the adaptive lgrp chunk resizing work. If the user specified
4570 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4571 // disable adaptive resizing.
4572 if (UseNUMA && UseLargePages && UseSHM) {
4573 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4574 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4575 UseLargePages = false;
4576 } else {
4577 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4578 UseAdaptiveSizePolicy = false;
4579 UseAdaptiveNUMAChunkSizing = false;
4580 }
4581 } else {
4582 UseNUMA = false;
4583 }
4584 }
4585 if (!UseNUMA && ForceNUMA) {
4586 UseNUMA = true;
4587 }
4588 }
4589
4590 if (MaxFDLimit) {
4591 // set the number of file descriptors to max. print out error
4592 // if getrlimit/setrlimit fails but continue regardless.
4593 struct rlimit nbr_files;
4594 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4595 if (status != 0) {
4596 if (PrintMiscellaneous && (Verbose || WizardMode))
4597 perror("os::init_2 getrlimit failed");
4598 } else {
4599 nbr_files.rlim_cur = nbr_files.rlim_max;
4600 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4601 if (status != 0) {
4602 if (PrintMiscellaneous && (Verbose || WizardMode))
4603 perror("os::init_2 setrlimit failed");
4604 }
4605 }
4606 }
4607
4608 // Initialize lock used to serialize thread creation (see os::create_thread)
4609 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4610
4611 // at-exit methods are called in the reverse order of their registration.
4612 // atexit functions are called on return from main or as a result of a
4613 // call to exit(3C). There can be only 32 of these functions registered
4614 // and atexit() does not set errno.
4615
4616 if (PerfAllowAtExitRegistration) {
4617 // only register atexit functions if PerfAllowAtExitRegistration is set.
4618 // atexit functions can be delayed until process exit time, which
4619 // can be problematic for embedded VM situations. Embedded VMs should
4620 // call DestroyJavaVM() to assure that VM resources are released.
4621
4622 // note: perfMemory_exit_helper atexit function may be removed in
4623 // the future if the appropriate cleanup code can be added to the
4624 // VM_Exit VMOperation's doit method.
4625 if (atexit(perfMemory_exit_helper) != 0) {
4626 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4627 }
4628 }
4629
4630 // initialize thread priority policy
4631 prio_init();
4632
4633 return JNI_OK;
4634 }
4635
4636 // this is called at the end of vm_initialization
4637 void os::init_3(void) {
4638 #ifdef JAVASE_EMBEDDED
4639 // Start the MemNotifyThread
4640 if (LowMemoryProtection) {
4641 MemNotifyThread::start();
4642 }
4643 return;
4644 #endif
4645 }
4646
4647 // Mark the polling page as unreadable
4648 void os::make_polling_page_unreadable(void) {
4649 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4650 fatal("Could not disable polling page");
4651 };
4652
4653 // Mark the polling page as readable
4654 void os::make_polling_page_readable(void) {
4655 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4656 fatal("Could not enable polling page");
4657 }
4658 };
4659
4660 int os::active_processor_count() {
4661 // Linux doesn't yet have a (official) notion of processor sets,
4662 // so just return the number of online processors.
4663 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4664 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4665 return online_cpus;
4666 }
4667
4668 void os::set_native_thread_name(const char *name) {
4669 // Not yet implemented.
4670 return;
4671 }
4672
4673 bool os::distribute_processes(uint length, uint* distribution) {
4674 // Not yet implemented.
4675 return false;
4676 }
4677
4678 bool os::bind_to_processor(uint processor_id) {
4679 // Not yet implemented.
4680 return false;
4681 }
4682
4683 ///
4684
4685 void os::SuspendedThreadTask::internal_do_task() {
4686 if (do_suspend(_thread->osthread())) {
4687 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4688 do_task(context);
4689 do_resume(_thread->osthread());
4690 }
4691 }
4692
4693 class PcFetcher : public os::SuspendedThreadTask {
4694 public:
4695 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4696 ExtendedPC result();
4697 protected:
4698 void do_task(const os::SuspendedThreadTaskContext& context);
4699 private:
4700 ExtendedPC _epc;
4701 };
4702
4703 ExtendedPC PcFetcher::result() {
4704 guarantee(is_done(), "task is not done yet.");
4705 return _epc;
4706 }
4707
4708 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4709 Thread* thread = context.thread();
4710 OSThread* osthread = thread->osthread();
4711 if (osthread->ucontext() != NULL) {
4712 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4713 } else {
4714 // NULL context is unexpected, double-check this is the VMThread
4715 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4716 }
4717 }
4718
4719 // Suspends the target using the signal mechanism and then grabs the PC before
4720 // resuming the target. Used by the flat-profiler only
4721 ExtendedPC os::get_thread_pc(Thread* thread) {
4722 // Make sure that it is called by the watcher for the VMThread
4723 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4724 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4725
4726 PcFetcher fetcher(thread);
4727 fetcher.run();
4728 return fetcher.result();
4729 }
4730
4731 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4732 {
4733 if (is_NPTL()) {
4734 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4735 } else {
4736 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4737 // word back to default 64bit precision if condvar is signaled. Java
4738 // wants 53bit precision. Save and restore current value.
4739 int fpu = get_fpu_control_word();
4740 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4741 set_fpu_control_word(fpu);
4742 return status;
4743 }
4744 }
4745
4746 ////////////////////////////////////////////////////////////////////////////////
4747 // debug support
4748
4749 bool os::find(address addr, outputStream* st) {
4750 Dl_info dlinfo;
4751 memset(&dlinfo, 0, sizeof(dlinfo));
4752 if (dladdr(addr, &dlinfo) != 0) {
4753 st->print(PTR_FORMAT ": ", addr);
4754 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4755 st->print("%s+%#x", dlinfo.dli_sname,
4756 addr - (intptr_t)dlinfo.dli_saddr);
4757 } else if (dlinfo.dli_fbase != NULL) {
4758 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4759 } else {
4760 st->print("<absolute address>");
4761 }
4762 if (dlinfo.dli_fname != NULL) {
4763 st->print(" in %s", dlinfo.dli_fname);
4764 }
4765 if (dlinfo.dli_fbase != NULL) {
4766 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4767 }
4768 st->cr();
4769
4770 if (Verbose) {
4771 // decode some bytes around the PC
4772 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4773 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4774 address lowest = (address) dlinfo.dli_sname;
4775 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4776 if (begin < lowest) begin = lowest;
4777 Dl_info dlinfo2;
4778 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4779 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4780 end = (address) dlinfo2.dli_saddr;
4781 Disassembler::decode(begin, end, st);
4782 }
4783 return true;
4784 }
4785 return false;
4786 }
4787
4788 ////////////////////////////////////////////////////////////////////////////////
4789 // misc
4790
4791 // This does not do anything on Linux. This is basically a hook for being
4792 // able to use structured exception handling (thread-local exception filters)
4793 // on, e.g., Win32.
4794 void
4795 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4796 JavaCallArguments* args, Thread* thread) {
4797 f(value, method, args, thread);
4798 }
4799
4800 void os::print_statistics() {
4801 }
4802
4803 int os::message_box(const char* title, const char* message) {
4804 int i;
4805 fdStream err(defaultStream::error_fd());
4806 for (i = 0; i < 78; i++) err.print_raw("=");
4807 err.cr();
4808 err.print_raw_cr(title);
4809 for (i = 0; i < 78; i++) err.print_raw("-");
4810 err.cr();
4811 err.print_raw_cr(message);
4812 for (i = 0; i < 78; i++) err.print_raw("=");
4813 err.cr();
4814
4815 char buf[16];
4816 // Prevent process from exiting upon "read error" without consuming all CPU
4817 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4818
4819 return buf[0] == 'y' || buf[0] == 'Y';
4820 }
4821
4822 int os::stat(const char *path, struct stat *sbuf) {
4823 char pathbuf[MAX_PATH];
4824 if (strlen(path) > MAX_PATH - 1) {
4825 errno = ENAMETOOLONG;
4826 return -1;
4827 }
4828 os::native_path(strcpy(pathbuf, path));
4829 return ::stat(pathbuf, sbuf);
4830 }
4831
4832 bool os::check_heap(bool force) {
4833 return true;
4834 }
4835
4836 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4837 return ::vsnprintf(buf, count, format, args);
4838 }
4839
4840 // Is a (classpath) directory empty?
4841 bool os::dir_is_empty(const char* path) {
4842 DIR *dir = NULL;
4843 struct dirent *ptr;
4844
4845 dir = opendir(path);
4846 if (dir == NULL) return true;
4847
4848 /* Scan the directory */
4849 bool result = true;
4850 char buf[sizeof(struct dirent) + MAX_PATH];
4851 while (result && (ptr = ::readdir(dir)) != NULL) {
4852 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4853 result = false;
4854 }
4855 }
4856 closedir(dir);
4857 return result;
4858 }
4859
4860 // This code originates from JDK's sysOpen and open64_w
4861 // from src/solaris/hpi/src/system_md.c
4862
4863 #ifndef O_DELETE
4864 #define O_DELETE 0x10000
4865 #endif
4866
4867 // Open a file. Unlink the file immediately after open returns
4868 // if the specified oflag has the O_DELETE flag set.
4869 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4870
4871 int os::open(const char *path, int oflag, int mode) {
4872
4873 if (strlen(path) > MAX_PATH - 1) {
4874 errno = ENAMETOOLONG;
4875 return -1;
4876 }
4877 int fd;
4878 int o_delete = (oflag & O_DELETE);
4879 oflag = oflag & ~O_DELETE;
4880
4881 fd = ::open64(path, oflag, mode);
4882 if (fd == -1) return -1;
4883
4884 //If the open succeeded, the file might still be a directory
4885 {
4886 struct stat64 buf64;
4887 int ret = ::fstat64(fd, &buf64);
4888 int st_mode = buf64.st_mode;
4889
4890 if (ret != -1) {
4891 if ((st_mode & S_IFMT) == S_IFDIR) {
4892 errno = EISDIR;
4893 ::close(fd);
4894 return -1;
4895 }
4896 } else {
4897 ::close(fd);
4898 return -1;
4899 }
4900 }
4901
4902 /*
4903 * All file descriptors that are opened in the JVM and not
4904 * specifically destined for a subprocess should have the
4905 * close-on-exec flag set. If we don't set it, then careless 3rd
4906 * party native code might fork and exec without closing all
4907 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4908 * UNIXProcess.c), and this in turn might:
4909 *
4910 * - cause end-of-file to fail to be detected on some file
4911 * descriptors, resulting in mysterious hangs, or
4912 *
4913 * - might cause an fopen in the subprocess to fail on a system
4914 * suffering from bug 1085341.
4915 *
4916 * (Yes, the default setting of the close-on-exec flag is a Unix
4917 * design flaw)
4918 *
4919 * See:
4920 * 1085341: 32-bit stdio routines should support file descriptors >255
4921 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4922 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4923 */
4924 #ifdef FD_CLOEXEC
4925 {
4926 int flags = ::fcntl(fd, F_GETFD);
4927 if (flags != -1)
4928 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4929 }
4930 #endif
4931
4932 if (o_delete != 0) {
4933 ::unlink(path);
4934 }
4935 return fd;
4936 }
4937
4938
4939 // create binary file, rewriting existing file if required
4940 int os::create_binary_file(const char* path, bool rewrite_existing) {
4941 int oflags = O_WRONLY | O_CREAT;
4942 if (!rewrite_existing) {
4943 oflags |= O_EXCL;
4944 }
4945 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4946 }
4947
4948 // return current position of file pointer
4949 jlong os::current_file_offset(int fd) {
4950 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4951 }
4952
4953 // move file pointer to the specified offset
4954 jlong os::seek_to_file_offset(int fd, jlong offset) {
4955 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4956 }
4957
4958 // This code originates from JDK's sysAvailable
4959 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4960
4961 int os::available(int fd, jlong *bytes) {
4962 jlong cur, end;
4963 int mode;
4964 struct stat64 buf64;
4965
4966 if (::fstat64(fd, &buf64) >= 0) {
4967 mode = buf64.st_mode;
4968 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4969 /*
4970 * XXX: is the following call interruptible? If so, this might
4971 * need to go through the INTERRUPT_IO() wrapper as for other
4972 * blocking, interruptible calls in this file.
4973 */
4974 int n;
4975 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4976 *bytes = n;
4977 return 1;
4978 }
4979 }
4980 }
4981 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4982 return 0;
4983 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4984 return 0;
4985 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4986 return 0;
4987 }
4988 *bytes = end - cur;
4989 return 1;
4990 }
4991
4992 int os::socket_available(int fd, jint *pbytes) {
4993 // Linux doc says EINTR not returned, unlike Solaris
4994 int ret = ::ioctl(fd, FIONREAD, pbytes);
4995
4996 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4997 // is expected to return 0 on failure and 1 on success to the jdk.
4998 return (ret < 0) ? 0 : 1;
4999 }
5000
5001 // Map a block of memory.
5002 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5003 char *addr, size_t bytes, bool read_only,
5004 bool allow_exec) {
5005 int prot;
5006 int flags = MAP_PRIVATE;
5007
5008 if (read_only) {
5009 prot = PROT_READ;
5010 } else {
5011 prot = PROT_READ | PROT_WRITE;
5012 }
5013
5014 if (allow_exec) {
5015 prot |= PROT_EXEC;
5016 }
5017
5018 if (addr != NULL) {
5019 flags |= MAP_FIXED;
5020 }
5021
5022 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5023 fd, file_offset);
5024 if (mapped_address == MAP_FAILED) {
5025 return NULL;
5026 }
5027 return mapped_address;
5028 }
5029
5030
5031 // Remap a block of memory.
5032 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5033 char *addr, size_t bytes, bool read_only,
5034 bool allow_exec) {
5035 // same as map_memory() on this OS
5036 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5037 allow_exec);
5038 }
5039
5040
5041 // Unmap a block of memory.
5042 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5043 return munmap(addr, bytes) == 0;
5044 }
5045
5046 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5047
5048 static clockid_t thread_cpu_clockid(Thread* thread) {
5049 pthread_t tid = thread->osthread()->pthread_id();
5050 clockid_t clockid;
5051
5052 // Get thread clockid
5053 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5054 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5055 return clockid;
5056 }
5057
5058 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5059 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5060 // of a thread.
5061 //
5062 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5063 // the fast estimate available on the platform.
5064
5065 jlong os::current_thread_cpu_time() {
5066 if (os::Linux::supports_fast_thread_cpu_time()) {
5067 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5068 } else {
5069 // return user + sys since the cost is the same
5070 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5071 }
5072 }
5073
5074 jlong os::thread_cpu_time(Thread* thread) {
5075 // consistent with what current_thread_cpu_time() returns
5076 if (os::Linux::supports_fast_thread_cpu_time()) {
5077 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5078 } else {
5079 return slow_thread_cpu_time(thread, true /* user + sys */);
5080 }
5081 }
5082
5083 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5084 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5085 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5086 } else {
5087 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5088 }
5089 }
5090
5091 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5092 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5093 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5094 } else {
5095 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5096 }
5097 }
5098
5099 //
5100 // -1 on error.
5101 //
5102
5103 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5104 static bool proc_task_unchecked = true;
5105 static const char *proc_stat_path = "/proc/%d/stat";
5106 pid_t tid = thread->osthread()->thread_id();
5107 char *s;
5108 char stat[2048];
5109 int statlen;
5110 char proc_name[64];
5111 int count;
5112 long sys_time, user_time;
5113 char cdummy;
5114 int idummy;
5115 long ldummy;
5116 FILE *fp;
5117
5118 // The /proc/<tid>/stat aggregates per-process usage on
5119 // new Linux kernels 2.6+ where NPTL is supported.
5120 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5121 // See bug 6328462.
5122 // There possibly can be cases where there is no directory
5123 // /proc/self/task, so we check its availability.
5124 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5125 // This is executed only once
5126 proc_task_unchecked = false;
5127 fp = fopen("/proc/self/task", "r");
5128 if (fp != NULL) {
5129 proc_stat_path = "/proc/self/task/%d/stat";
5130 fclose(fp);
5131 }
5132 }
5133
5134 sprintf(proc_name, proc_stat_path, tid);
5135 fp = fopen(proc_name, "r");
5136 if ( fp == NULL ) return -1;
5137 statlen = fread(stat, 1, 2047, fp);
5138 stat[statlen] = '\0';
5139 fclose(fp);
5140
5141 // Skip pid and the command string. Note that we could be dealing with
5142 // weird command names, e.g. user could decide to rename java launcher
5143 // to "java 1.4.2 :)", then the stat file would look like
5144 // 1234 (java 1.4.2 :)) R ... ...
5145 // We don't really need to know the command string, just find the last
5146 // occurrence of ")" and then start parsing from there. See bug 4726580.
5147 s = strrchr(stat, ')');
5148 if (s == NULL ) return -1;
5149
5150 // Skip blank chars
5151 do s++; while (isspace(*s));
5152
5153 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5154 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5155 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5156 &user_time, &sys_time);
5157 if ( count != 13 ) return -1;
5158 if (user_sys_cpu_time) {
5159 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5160 } else {
5161 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5162 }
5163 }
5164
5165 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5166 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5167 info_ptr->may_skip_backward = false; // elapsed time not wall time
5168 info_ptr->may_skip_forward = false; // elapsed time not wall time
5169 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5170 }
5171
5172 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5173 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5174 info_ptr->may_skip_backward = false; // elapsed time not wall time
5175 info_ptr->may_skip_forward = false; // elapsed time not wall time
5176 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5177 }
5178
5179 bool os::is_thread_cpu_time_supported() {
5180 return true;
5181 }
5182
5183 // System loadavg support. Returns -1 if load average cannot be obtained.
5184 // Linux doesn't yet have a (official) notion of processor sets,
5185 // so just return the system wide load average.
5186 int os::loadavg(double loadavg[], int nelem) {
5187 return ::getloadavg(loadavg, nelem);
5188 }
5189
5190 void os::pause() {
5191 char filename[MAX_PATH];
5192 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5193 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5194 } else {
5195 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5196 }
5197
5198 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5199 if (fd != -1) {
5200 struct stat buf;
5201 ::close(fd);
5202 while (::stat(filename, &buf) == 0) {
5203 (void)::poll(NULL, 0, 100);
5204 }
5205 } else {
5206 jio_fprintf(stderr,
5207 "Could not open pause file '%s', continuing immediately.\n", filename);
5208 }
5209 }
5210
5211
5212 // Refer to the comments in os_solaris.cpp park-unpark.
5213 //
5214 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5215 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5216 // For specifics regarding the bug see GLIBC BUGID 261237 :
5217 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5218 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5219 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5220 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5221 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5222 // and monitorenter when we're using 1-0 locking. All those operations may result in
5223 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5224 // of libpthread avoids the problem, but isn't practical.
5225 //
5226 // Possible remedies:
5227 //
5228 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5229 // This is palliative and probabilistic, however. If the thread is preempted
5230 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5231 // than the minimum period may have passed, and the abstime may be stale (in the
5232 // past) resultin in a hang. Using this technique reduces the odds of a hang
5233 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5234 //
5235 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5236 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5237 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5238 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5239 // thread.
5240 //
5241 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5242 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5243 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5244 // This also works well. In fact it avoids kernel-level scalability impediments
5245 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5246 // timers in a graceful fashion.
5247 //
5248 // 4. When the abstime value is in the past it appears that control returns
5249 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5250 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5251 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5252 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5253 // It may be possible to avoid reinitialization by checking the return
5254 // value from pthread_cond_timedwait(). In addition to reinitializing the
5255 // condvar we must establish the invariant that cond_signal() is only called
5256 // within critical sections protected by the adjunct mutex. This prevents
5257 // cond_signal() from "seeing" a condvar that's in the midst of being
5258 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5259 // desirable signal-after-unlock optimization that avoids futile context switching.
5260 //
5261 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5262 // structure when a condvar is used or initialized. cond_destroy() would
5263 // release the helper structure. Our reinitialize-after-timedwait fix
5264 // put excessive stress on malloc/free and locks protecting the c-heap.
5265 //
5266 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5267 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5268 // and only enabling the work-around for vulnerable environments.
5269
5270 // utility to compute the abstime argument to timedwait:
5271 // millis is the relative timeout time
5272 // abstime will be the absolute timeout time
5273 // TODO: replace compute_abstime() with unpackTime()
5274
5275 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5276 if (millis < 0) millis = 0;
5277 struct timeval now;
5278 int status = gettimeofday(&now, NULL);
5279 assert(status == 0, "gettimeofday");
5280 jlong seconds = millis / 1000;
5281 millis %= 1000;
5282 if (seconds > 50000000) { // see man cond_timedwait(3T)
5283 seconds = 50000000;
5284 }
5285 abstime->tv_sec = now.tv_sec + seconds;
5286 long usec = now.tv_usec + millis * 1000;
5287 if (usec >= 1000000) {
5288 abstime->tv_sec += 1;
5289 usec -= 1000000;
5290 }
5291 abstime->tv_nsec = usec * 1000;
5292 return abstime;
5293 }
5294
5295
5296 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5297 // Conceptually TryPark() should be equivalent to park(0).
5298
5299 int os::PlatformEvent::TryPark() {
5300 for (;;) {
5301 const int v = _Event ;
5302 guarantee ((v == 0) || (v == 1), "invariant") ;
5303 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5304 }
5305 }
5306
5307 void os::PlatformEvent::park() { // AKA "down()"
5308 // Invariant: Only the thread associated with the Event/PlatformEvent
5309 // may call park().
5310 // TODO: assert that _Assoc != NULL or _Assoc == Self
5311 int v ;
5312 for (;;) {
5313 v = _Event ;
5314 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5315 }
5316 guarantee (v >= 0, "invariant") ;
5317 if (v == 0) {
5318 // Do this the hard way by blocking ...
5319 int status = pthread_mutex_lock(_mutex);
5320 assert_status(status == 0, status, "mutex_lock");
5321 guarantee (_nParked == 0, "invariant") ;
5322 ++ _nParked ;
5323 while (_Event < 0) {
5324 status = pthread_cond_wait(_cond, _mutex);
5325 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5326 // Treat this the same as if the wait was interrupted
5327 if (status == ETIME) { status = EINTR; }
5328 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5329 }
5330 -- _nParked ;
5331
5332 _Event = 0 ;
5333 status = pthread_mutex_unlock(_mutex);
5334 assert_status(status == 0, status, "mutex_unlock");
5335 // Paranoia to ensure our locked and lock-free paths interact
5336 // correctly with each other.
5337 OrderAccess::fence();
5338 }
5339 guarantee (_Event >= 0, "invariant") ;
5340 }
5341
5342 int os::PlatformEvent::park(jlong millis) {
5343 guarantee (_nParked == 0, "invariant") ;
5344
5345 int v ;
5346 for (;;) {
5347 v = _Event ;
5348 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5349 }
5350 guarantee (v >= 0, "invariant") ;
5351 if (v != 0) return OS_OK ;
5352
5353 // We do this the hard way, by blocking the thread.
5354 // Consider enforcing a minimum timeout value.
5355 struct timespec abst;
5356 compute_abstime(&abst, millis);
5357
5358 int ret = OS_TIMEOUT;
5359 int status = pthread_mutex_lock(_mutex);
5360 assert_status(status == 0, status, "mutex_lock");
5361 guarantee (_nParked == 0, "invariant") ;
5362 ++_nParked ;
5363
5364 // Object.wait(timo) will return because of
5365 // (a) notification
5366 // (b) timeout
5367 // (c) thread.interrupt
5368 //
5369 // Thread.interrupt and object.notify{All} both call Event::set.
5370 // That is, we treat thread.interrupt as a special case of notification.
5371 // The underlying Solaris implementation, cond_timedwait, admits
5372 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5373 // JVM from making those visible to Java code. As such, we must
5374 // filter out spurious wakeups. We assume all ETIME returns are valid.
5375 //
5376 // TODO: properly differentiate simultaneous notify+interrupt.
5377 // In that case, we should propagate the notify to another waiter.
5378
5379 while (_Event < 0) {
5380 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5381 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5382 pthread_cond_destroy (_cond);
5383 pthread_cond_init (_cond, NULL) ;
5384 }
5385 assert_status(status == 0 || status == EINTR ||
5386 status == ETIME || status == ETIMEDOUT,
5387 status, "cond_timedwait");
5388 if (!FilterSpuriousWakeups) break ; // previous semantics
5389 if (status == ETIME || status == ETIMEDOUT) break ;
5390 // We consume and ignore EINTR and spurious wakeups.
5391 }
5392 --_nParked ;
5393 if (_Event >= 0) {
5394 ret = OS_OK;
5395 }
5396 _Event = 0 ;
5397 status = pthread_mutex_unlock(_mutex);
5398 assert_status(status == 0, status, "mutex_unlock");
5399 assert (_nParked == 0, "invariant") ;
5400 // Paranoia to ensure our locked and lock-free paths interact
5401 // correctly with each other.
5402 OrderAccess::fence();
5403 return ret;
5404 }
5405
5406 void os::PlatformEvent::unpark() {
5407 // Transitions for _Event:
5408 // 0 :=> 1
5409 // 1 :=> 1
5410 // -1 :=> either 0 or 1; must signal target thread
5411 // That is, we can safely transition _Event from -1 to either
5412 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5413 // unpark() calls.
5414 // See also: "Semaphores in Plan 9" by Mullender & Cox
5415 //
5416 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5417 // that it will take two back-to-back park() calls for the owning
5418 // thread to block. This has the benefit of forcing a spurious return
5419 // from the first park() call after an unpark() call which will help
5420 // shake out uses of park() and unpark() without condition variables.
5421
5422 if (Atomic::xchg(1, &_Event) >= 0) return;
5423
5424 // Wait for the thread associated with the event to vacate
5425 int status = pthread_mutex_lock(_mutex);
5426 assert_status(status == 0, status, "mutex_lock");
5427 int AnyWaiters = _nParked;
5428 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5429 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5430 AnyWaiters = 0;
5431 pthread_cond_signal(_cond);
5432 }
5433 status = pthread_mutex_unlock(_mutex);
5434 assert_status(status == 0, status, "mutex_unlock");
5435 if (AnyWaiters != 0) {
5436 status = pthread_cond_signal(_cond);
5437 assert_status(status == 0, status, "cond_signal");
5438 }
5439
5440 // Note that we signal() _after dropping the lock for "immortal" Events.
5441 // This is safe and avoids a common class of futile wakeups. In rare
5442 // circumstances this can cause a thread to return prematurely from
5443 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5444 // simply re-test the condition and re-park itself.
5445 }
5446
5447
5448 // JSR166
5449 // -------------------------------------------------------
5450
5451 /*
5452 * The solaris and linux implementations of park/unpark are fairly
5453 * conservative for now, but can be improved. They currently use a
5454 * mutex/condvar pair, plus a a count.
5455 * Park decrements count if > 0, else does a condvar wait. Unpark
5456 * sets count to 1 and signals condvar. Only one thread ever waits
5457 * on the condvar. Contention seen when trying to park implies that someone
5458 * is unparking you, so don't wait. And spurious returns are fine, so there
5459 * is no need to track notifications.
5460 */
5461
5462 #define MAX_SECS 100000000
5463 /*
5464 * This code is common to linux and solaris and will be moved to a
5465 * common place in dolphin.
5466 *
5467 * The passed in time value is either a relative time in nanoseconds
5468 * or an absolute time in milliseconds. Either way it has to be unpacked
5469 * into suitable seconds and nanoseconds components and stored in the
5470 * given timespec structure.
5471 * Given time is a 64-bit value and the time_t used in the timespec is only
5472 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5473 * overflow if times way in the future are given. Further on Solaris versions
5474 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5475 * number of seconds, in abstime, is less than current_time + 100,000,000.
5476 * As it will be 28 years before "now + 100000000" will overflow we can
5477 * ignore overflow and just impose a hard-limit on seconds using the value
5478 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5479 * years from "now".
5480 */
5481
5482 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5483 assert (time > 0, "convertTime");
5484
5485 struct timeval now;
5486 int status = gettimeofday(&now, NULL);
5487 assert(status == 0, "gettimeofday");
5488
5489 time_t max_secs = now.tv_sec + MAX_SECS;
5490
5491 if (isAbsolute) {
5492 jlong secs = time / 1000;
5493 if (secs > max_secs) {
5494 absTime->tv_sec = max_secs;
5495 }
5496 else {
5497 absTime->tv_sec = secs;
5498 }
5499 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5500 }
5501 else {
5502 jlong secs = time / NANOSECS_PER_SEC;
5503 if (secs >= MAX_SECS) {
5504 absTime->tv_sec = max_secs;
5505 absTime->tv_nsec = 0;
5506 }
5507 else {
5508 absTime->tv_sec = now.tv_sec + secs;
5509 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5510 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5511 absTime->tv_nsec -= NANOSECS_PER_SEC;
5512 ++absTime->tv_sec; // note: this must be <= max_secs
5513 }
5514 }
5515 }
5516 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5517 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5518 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5519 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5520 }
5521
5522 void Parker::park(bool isAbsolute, jlong time) {
5523 // Ideally we'd do something useful while spinning, such
5524 // as calling unpackTime().
5525
5526 // Optional fast-path check:
5527 // Return immediately if a permit is available.
5528 // We depend on Atomic::xchg() having full barrier semantics
5529 // since we are doing a lock-free update to _counter.
5530 if (Atomic::xchg(0, &_counter) > 0) return;
5531
5532 Thread* thread = Thread::current();
5533 assert(thread->is_Java_thread(), "Must be JavaThread");
5534 JavaThread *jt = (JavaThread *)thread;
5535
5536 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5537 // Check interrupt before trying to wait
5538 if (Thread::is_interrupted(thread, false)) {
5539 return;
5540 }
5541
5542 // Next, demultiplex/decode time arguments
5543 timespec absTime;
5544 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5545 return;
5546 }
5547 if (time > 0) {
5548 unpackTime(&absTime, isAbsolute, time);
5549 }
5550
5551
5552 // Enter safepoint region
5553 // Beware of deadlocks such as 6317397.
5554 // The per-thread Parker:: mutex is a classic leaf-lock.
5555 // In particular a thread must never block on the Threads_lock while
5556 // holding the Parker:: mutex. If safepoints are pending both the
5557 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5558 ThreadBlockInVM tbivm(jt);
5559
5560 // Don't wait if cannot get lock since interference arises from
5561 // unblocking. Also. check interrupt before trying wait
5562 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5563 return;
5564 }
5565
5566 int status ;
5567 if (_counter > 0) { // no wait needed
5568 _counter = 0;
5569 status = pthread_mutex_unlock(_mutex);
5570 assert (status == 0, "invariant") ;
5571 // Paranoia to ensure our locked and lock-free paths interact
5572 // correctly with each other and Java-level accesses.
5573 OrderAccess::fence();
5574 return;
5575 }
5576
5577 #ifdef ASSERT
5578 // Don't catch signals while blocked; let the running threads have the signals.
5579 // (This allows a debugger to break into the running thread.)
5580 sigset_t oldsigs;
5581 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5582 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5583 #endif
5584
5585 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5586 jt->set_suspend_equivalent();
5587 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5588
5589 if (time == 0) {
5590 status = pthread_cond_wait (_cond, _mutex) ;
5591 } else {
5592 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5593 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5594 pthread_cond_destroy (_cond) ;
5595 pthread_cond_init (_cond, NULL);
5596 }
5597 }
5598 assert_status(status == 0 || status == EINTR ||
5599 status == ETIME || status == ETIMEDOUT,
5600 status, "cond_timedwait");
5601
5602 #ifdef ASSERT
5603 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5604 #endif
5605
5606 _counter = 0 ;
5607 status = pthread_mutex_unlock(_mutex) ;
5608 assert_status(status == 0, status, "invariant") ;
5609 // Paranoia to ensure our locked and lock-free paths interact
5610 // correctly with each other and Java-level accesses.
5611 OrderAccess::fence();
5612
5613 // If externally suspended while waiting, re-suspend
5614 if (jt->handle_special_suspend_equivalent_condition()) {
5615 jt->java_suspend_self();
5616 }
5617 }
5618
5619 void Parker::unpark() {
5620 int s, status ;
5621 status = pthread_mutex_lock(_mutex);
5622 assert (status == 0, "invariant") ;
5623 s = _counter;
5624 _counter = 1;
5625 if (s < 1) {
5626 if (WorkAroundNPTLTimedWaitHang) {
5627 status = pthread_cond_signal (_cond) ;
5628 assert (status == 0, "invariant") ;
5629 status = pthread_mutex_unlock(_mutex);
5630 assert (status == 0, "invariant") ;
5631 } else {
5632 status = pthread_mutex_unlock(_mutex);
5633 assert (status == 0, "invariant") ;
5634 status = pthread_cond_signal (_cond) ;
5635 assert (status == 0, "invariant") ;
5636 }
5637 } else {
5638 pthread_mutex_unlock(_mutex);
5639 assert (status == 0, "invariant") ;
5640 }
5641 }
5642
5643
5644 extern char** environ;
5645
5646 #ifndef __NR_fork
5647 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5648 #endif
5649
5650 #ifndef __NR_execve
5651 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5652 #endif
5653
5654 // Run the specified command in a separate process. Return its exit value,
5655 // or -1 on failure (e.g. can't fork a new process).
5656 // Unlike system(), this function can be called from signal handler. It
5657 // doesn't block SIGINT et al.
5658 int os::fork_and_exec(char* cmd) {
5659 const char * argv[4] = {"sh", "-c", cmd, NULL};
5660
5661 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5662 // pthread_atfork handlers and reset pthread library. All we need is a
5663 // separate process to execve. Make a direct syscall to fork process.
5664 // On IA64 there's no fork syscall, we have to use fork() and hope for
5665 // the best...
5666 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5667 IA64_ONLY(fork();)
5668
5669 if (pid < 0) {
5670 // fork failed
5671 return -1;
5672
5673 } else if (pid == 0) {
5674 // child process
5675
5676 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5677 // first to kill every thread on the thread list. Because this list is
5678 // not reset by fork() (see notes above), execve() will instead kill
5679 // every thread in the parent process. We know this is the only thread
5680 // in the new process, so make a system call directly.
5681 // IA64 should use normal execve() from glibc to match the glibc fork()
5682 // above.
5683 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5684 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5685
5686 // execve failed
5687 _exit(-1);
5688
5689 } else {
5690 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5691 // care about the actual exit code, for now.
5692
5693 int status;
5694
5695 // Wait for the child process to exit. This returns immediately if
5696 // the child has already exited. */
5697 while (waitpid(pid, &status, 0) < 0) {
5698 switch (errno) {
5699 case ECHILD: return 0;
5700 case EINTR: break;
5701 default: return -1;
5702 }
5703 }
5704
5705 if (WIFEXITED(status)) {
5706 // The child exited normally; get its exit code.
5707 return WEXITSTATUS(status);
5708 } else if (WIFSIGNALED(status)) {
5709 // The child exited because of a signal
5710 // The best value to return is 0x80 + signal number,
5711 // because that is what all Unix shells do, and because
5712 // it allows callers to distinguish between process exit and
5713 // process death by signal.
5714 return 0x80 + WTERMSIG(status);
5715 } else {
5716 // Unknown exit code; pass it through
5717 return status;
5718 }
5719 }
5720 }
5721
5722 // is_headless_jre()
5723 //
5724 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5725 // in order to report if we are running in a headless jre
5726 //
5727 // Since JDK8 xawt/libmawt.so was moved into the same directory
5728 // as libawt.so, and renamed libawt_xawt.so
5729 //
5730 bool os::is_headless_jre() {
5731 struct stat statbuf;
5732 char buf[MAXPATHLEN];
5733 char libmawtpath[MAXPATHLEN];
5734 const char *xawtstr = "/xawt/libmawt.so";
5735 const char *new_xawtstr = "/libawt_xawt.so";
5736 char *p;
5737
5738 // Get path to libjvm.so
5739 os::jvm_path(buf, sizeof(buf));
5740
5741 // Get rid of libjvm.so
5742 p = strrchr(buf, '/');
5743 if (p == NULL) return false;
5744 else *p = '\0';
5745
5746 // Get rid of client or server
5747 p = strrchr(buf, '/');
5748 if (p == NULL) return false;
5749 else *p = '\0';
5750
5751 // check xawt/libmawt.so
5752 strcpy(libmawtpath, buf);
5753 strcat(libmawtpath, xawtstr);
5754 if (::stat(libmawtpath, &statbuf) == 0) return false;
5755
5756 // check libawt_xawt.so
5757 strcpy(libmawtpath, buf);
5758 strcat(libmawtpath, new_xawtstr);
5759 if (::stat(libmawtpath, &statbuf) == 0) return false;
5760
5761 return true;
5762 }
5763
5764 // Get the default path to the core file
5765 // Returns the length of the string
5766 int os::get_core_path(char* buffer, size_t bufferSize) {
5767 const char* p = get_current_directory(buffer, bufferSize);
5768
5769 if (p == NULL) {
5770 assert(p != NULL, "failed to get current directory");
5771 return 0;
5772 }
5773
5774 return strlen(buffer);
5775 }
5776
5777 #ifdef JAVASE_EMBEDDED
5778 //
5779 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5780 //
5781 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5782
5783 // ctor
5784 //
5785 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5786 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5787 _fd = fd;
5788
5789 if (os::create_thread(this, os::os_thread)) {
5790 _memnotify_thread = this;
5791 os::set_priority(this, NearMaxPriority);
5792 os::start_thread(this);
5793 }
5794 }
5795
5796 // Where all the work gets done
5797 //
5798 void MemNotifyThread::run() {
5799 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5800
5801 // Set up the select arguments
5802 fd_set rfds;
5803 if (_fd != -1) {
5804 FD_ZERO(&rfds);
5805 FD_SET(_fd, &rfds);
5806 }
5807
5808 // Now wait for the mem_notify device to wake up
5809 while (1) {
5810 // Wait for the mem_notify device to signal us..
5811 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5812 if (rc == -1) {
5813 perror("select!\n");
5814 break;
5815 } else if (rc) {
5816 //ssize_t free_before = os::available_memory();
5817 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5818
5819 // The kernel is telling us there is not much memory left...
5820 // try to do something about that
5821
5822 // If we are not already in a GC, try one.
5823 if (!Universe::heap()->is_gc_active()) {
5824 Universe::heap()->collect(GCCause::_allocation_failure);
5825
5826 //ssize_t free_after = os::available_memory();
5827 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5828 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5829 }
5830 // We might want to do something like the following if we find the GC's are not helping...
5831 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5832 }
5833 }
5834 }
5835
5836 //
5837 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5838 //
5839 void MemNotifyThread::start() {
5840 int fd;
5841 fd = open ("/dev/mem_notify", O_RDONLY, 0);
5842 if (fd < 0) {
5843 return;
5844 }
5845
5846 if (memnotify_thread() == NULL) {
5847 new MemNotifyThread(fd);
5848 }
5849 }
5850
5851 #endif // JAVASE_EMBEDDED