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(PPC)
260 static char cpu_arch[] = "ppc";
261 #elif defined(SPARC)
262 # ifdef _LP64
263 static char cpu_arch[] = "sparcv9";
264 # else
265 static char cpu_arch[] = "sparc";
266 # endif
267 #else
268 #error Add appropriate cpu_arch setting
269 #endif
270
271
272 // pid_t gettid()
273 //
274 // Returns the kernel thread id of the currently running thread. Kernel
275 // thread id is used to access /proc.
276 //
277 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
278 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
279 //
280 pid_t os::Linux::gettid() {
281 int rslt = syscall(SYS_gettid);
282 if (rslt == -1) {
283 // old kernel, no NPTL support
284 return getpid();
285 } else {
286 return (pid_t)rslt;
287 }
288 }
289
290 // Most versions of linux have a bug where the number of processors are
291 // determined by looking at the /proc file system. In a chroot environment,
292 // the system call returns 1. This causes the VM to act as if it is
293 // a single processor and elide locking (see is_MP() call).
294 static bool unsafe_chroot_detected = false;
295 static const char *unstable_chroot_error = "/proc file system not found.\n"
296 "Java may be unstable running multithreaded in a chroot "
297 "environment on Linux when /proc filesystem is not mounted.";
298
299 void os::Linux::initialize_system_info() {
300 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
301 if (processor_count() == 1) {
302 pid_t pid = os::Linux::gettid();
303 char fname[32];
304 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
305 FILE *fp = fopen(fname, "r");
306 if (fp == NULL) {
307 unsafe_chroot_detected = true;
308 } else {
309 fclose(fp);
310 }
311 }
312 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
313 assert(processor_count() > 0, "linux error");
314 }
315
316 void os::init_system_properties_values() {
317 // char arch[12];
318 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
319
320 // The next steps are taken in the product version:
321 //
322 // Obtain the JAVA_HOME value from the location of libjvm.so.
323 // This library should be located at:
324 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
325 //
326 // If "/jre/lib/" appears at the right place in the path, then we
327 // assume libjvm.so is installed in a JDK and we use this path.
328 //
329 // Otherwise exit with message: "Could not create the Java virtual machine."
330 //
331 // The following extra steps are taken in the debugging version:
332 //
333 // If "/jre/lib/" does NOT appear at the right place in the path
334 // instead of exit check for $JAVA_HOME environment variable.
335 //
336 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
337 // then we append a fake suffix "hotspot/libjvm.so" to this path so
338 // it looks like libjvm.so is installed there
339 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
340 //
341 // Otherwise exit.
342 //
343 // Important note: if the location of libjvm.so changes this
344 // code needs to be changed accordingly.
345
346 // The next few definitions allow the code to be verbatim:
347 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
348 #define getenv(n) ::getenv(n)
349
350 /*
351 * See ld(1):
352 * The linker uses the following search paths to locate required
353 * shared libraries:
354 * 1: ...
355 * ...
356 * 7: The default directories, normally /lib and /usr/lib.
357 */
358 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
359 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
360 #else
361 #define DEFAULT_LIBPATH "/lib:/usr/lib"
362 #endif
363
364 #define EXTENSIONS_DIR "/lib/ext"
365 #define ENDORSED_DIR "/lib/endorsed"
366 #define REG_DIR "/usr/java/packages"
367
368 {
369 /* sysclasspath, java_home, dll_dir */
370 {
371 char *home_path;
372 char *dll_path;
373 char *pslash;
374 char buf[MAXPATHLEN];
375 os::jvm_path(buf, sizeof(buf));
376
377 // Found the full path to libjvm.so.
378 // Now cut the path to <java_home>/jre if we can.
379 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
380 pslash = strrchr(buf, '/');
381 if (pslash != NULL)
382 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
383 dll_path = malloc(strlen(buf) + 1);
384 if (dll_path == NULL)
385 return;
386 strcpy(dll_path, buf);
387 Arguments::set_dll_dir(dll_path);
388
389 if (pslash != NULL) {
390 pslash = strrchr(buf, '/');
391 if (pslash != NULL) {
392 *pslash = '\0'; /* get rid of /<arch> */
393 pslash = strrchr(buf, '/');
394 if (pslash != NULL)
395 *pslash = '\0'; /* get rid of /lib */
396 }
397 }
398
399 home_path = malloc(strlen(buf) + 1);
400 if (home_path == NULL)
401 return;
402 strcpy(home_path, buf);
403 Arguments::set_java_home(home_path);
404
405 if (!set_boot_path('/', ':'))
406 return;
407 }
408
409 /*
410 * Where to look for native libraries
411 *
412 * Note: Due to a legacy implementation, most of the library path
413 * is set in the launcher. This was to accomodate linking restrictions
414 * on legacy Linux implementations (which are no longer supported).
415 * Eventually, all the library path setting will be done here.
416 *
417 * However, to prevent the proliferation of improperly built native
418 * libraries, the new path component /usr/java/packages is added here.
419 * Eventually, all the library path setting will be done here.
420 */
421 {
422 char *ld_library_path;
423
424 /*
425 * Construct the invariant part of ld_library_path. Note that the
426 * space for the colon and the trailing null are provided by the
427 * nulls included by the sizeof operator (so actually we allocate
428 * a byte more than necessary).
429 */
430 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
431 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
432 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
433
434 /*
435 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
436 * should always exist (until the legacy problem cited above is
437 * addressed).
438 */
439 char *v = getenv("LD_LIBRARY_PATH");
440 if (v != NULL) {
441 char *t = ld_library_path;
442 /* That's +1 for the colon and +1 for the trailing '\0' */
443 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
444 sprintf(ld_library_path, "%s:%s", v, t);
445 }
446 Arguments::set_library_path(ld_library_path);
447 }
448
449 /*
450 * Extensions directories.
451 *
452 * Note that the space for the colon and the trailing null are provided
453 * by the nulls included by the sizeof operator (so actually one byte more
454 * than necessary is allocated).
455 */
456 {
457 char *buf = malloc(strlen(Arguments::get_java_home()) +
458 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
459 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
460 Arguments::get_java_home());
461 Arguments::set_ext_dirs(buf);
462 }
463
464 /* Endorsed standards default directory. */
465 {
466 char * buf;
467 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
468 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
469 Arguments::set_endorsed_dirs(buf);
470 }
471 }
472
473 #undef malloc
474 #undef getenv
475 #undef EXTENSIONS_DIR
476 #undef ENDORSED_DIR
477
478 // Done
479 return;
480 }
481
482 ////////////////////////////////////////////////////////////////////////////////
483 // breakpoint support
484
485 void os::breakpoint() {
486 BREAKPOINT;
487 }
488
489 extern "C" void breakpoint() {
490 // use debugger to set breakpoint here
491 }
492
493 ////////////////////////////////////////////////////////////////////////////////
494 // signal support
495
496 debug_only(static bool signal_sets_initialized = false);
497 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
498
499 bool os::Linux::is_sig_ignored(int sig) {
500 struct sigaction oact;
501 sigaction(sig, (struct sigaction*)NULL, &oact);
502 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
503 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
504 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
505 return true;
506 else
507 return false;
508 }
509
510 void os::Linux::signal_sets_init() {
511 // Should also have an assertion stating we are still single-threaded.
512 assert(!signal_sets_initialized, "Already initialized");
513 // Fill in signals that are necessarily unblocked for all threads in
514 // the VM. Currently, we unblock the following signals:
515 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
516 // by -Xrs (=ReduceSignalUsage));
517 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
518 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
519 // the dispositions or masks wrt these signals.
520 // Programs embedding the VM that want to use the above signals for their
521 // own purposes must, at this time, use the "-Xrs" option to prevent
522 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
523 // (See bug 4345157, and other related bugs).
524 // In reality, though, unblocking these signals is really a nop, since
525 // these signals are not blocked by default.
526 sigemptyset(&unblocked_sigs);
527 sigemptyset(&allowdebug_blocked_sigs);
528 sigaddset(&unblocked_sigs, SIGILL);
529 sigaddset(&unblocked_sigs, SIGSEGV);
530 sigaddset(&unblocked_sigs, SIGBUS);
531 sigaddset(&unblocked_sigs, SIGFPE);
532 sigaddset(&unblocked_sigs, SR_signum);
533
534 if (!ReduceSignalUsage) {
535 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
536 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
537 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
538 }
539 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
540 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
541 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
542 }
543 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
544 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
545 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
546 }
547 }
548 // Fill in signals that are blocked by all but the VM thread.
549 sigemptyset(&vm_sigs);
550 if (!ReduceSignalUsage)
551 sigaddset(&vm_sigs, BREAK_SIGNAL);
552 debug_only(signal_sets_initialized = true);
553
554 }
555
556 // These are signals that are unblocked while a thread is running Java.
557 // (For some reason, they get blocked by default.)
558 sigset_t* os::Linux::unblocked_signals() {
559 assert(signal_sets_initialized, "Not initialized");
560 return &unblocked_sigs;
561 }
562
563 // These are the signals that are blocked while a (non-VM) thread is
564 // running Java. Only the VM thread handles these signals.
565 sigset_t* os::Linux::vm_signals() {
566 assert(signal_sets_initialized, "Not initialized");
567 return &vm_sigs;
568 }
569
570 // These are signals that are blocked during cond_wait to allow debugger in
571 sigset_t* os::Linux::allowdebug_blocked_signals() {
572 assert(signal_sets_initialized, "Not initialized");
573 return &allowdebug_blocked_sigs;
574 }
575
576 void os::Linux::hotspot_sigmask(Thread* thread) {
577
578 //Save caller's signal mask before setting VM signal mask
579 sigset_t caller_sigmask;
580 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
581
582 OSThread* osthread = thread->osthread();
583 osthread->set_caller_sigmask(caller_sigmask);
584
585 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
586
587 if (!ReduceSignalUsage) {
588 if (thread->is_VM_thread()) {
589 // Only the VM thread handles BREAK_SIGNAL ...
590 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
591 } else {
592 // ... all other threads block BREAK_SIGNAL
593 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
594 }
595 }
596 }
597
598 //////////////////////////////////////////////////////////////////////////////
599 // detecting pthread library
600
601 void os::Linux::libpthread_init() {
602 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
603 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
604 // generic name for earlier versions.
605 // Define macros here so we can build HotSpot on old systems.
606 # ifndef _CS_GNU_LIBC_VERSION
607 # define _CS_GNU_LIBC_VERSION 2
608 # endif
609 # ifndef _CS_GNU_LIBPTHREAD_VERSION
610 # define _CS_GNU_LIBPTHREAD_VERSION 3
611 # endif
612
613 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
614 if (n > 0) {
615 char *str = (char *)malloc(n, mtInternal);
616 confstr(_CS_GNU_LIBC_VERSION, str, n);
617 os::Linux::set_glibc_version(str);
618 } else {
619 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
620 static char _gnu_libc_version[32];
621 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
622 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
623 os::Linux::set_glibc_version(_gnu_libc_version);
624 }
625
626 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
627 if (n > 0) {
628 char *str = (char *)malloc(n, mtInternal);
629 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
630 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
631 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
632 // is the case. LinuxThreads has a hard limit on max number of threads.
633 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
634 // On the other hand, NPTL does not have such a limit, sysconf()
635 // will return -1 and errno is not changed. Check if it is really NPTL.
636 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
637 strstr(str, "NPTL") &&
638 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
639 free(str);
640 os::Linux::set_libpthread_version("linuxthreads");
641 } else {
642 os::Linux::set_libpthread_version(str);
643 }
644 } else {
645 // glibc before 2.3.2 only has LinuxThreads.
646 os::Linux::set_libpthread_version("linuxthreads");
647 }
648
649 if (strstr(libpthread_version(), "NPTL")) {
650 os::Linux::set_is_NPTL();
651 } else {
652 os::Linux::set_is_LinuxThreads();
653 }
654
655 // LinuxThreads have two flavors: floating-stack mode, which allows variable
656 // stack size; and fixed-stack mode. NPTL is always floating-stack.
657 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
658 os::Linux::set_is_floating_stack();
659 }
660 }
661
662 /////////////////////////////////////////////////////////////////////////////
663 // thread stack
664
665 // Force Linux kernel to expand current thread stack. If "bottom" is close
666 // to the stack guard, caller should block all signals.
667 //
668 // MAP_GROWSDOWN:
669 // A special mmap() flag that is used to implement thread stacks. It tells
670 // kernel that the memory region should extend downwards when needed. This
671 // allows early versions of LinuxThreads to only mmap the first few pages
672 // when creating a new thread. Linux kernel will automatically expand thread
673 // stack as needed (on page faults).
674 //
675 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
676 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
677 // region, it's hard to tell if the fault is due to a legitimate stack
678 // access or because of reading/writing non-exist memory (e.g. buffer
679 // overrun). As a rule, if the fault happens below current stack pointer,
680 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
681 // application (see Linux kernel fault.c).
682 //
683 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
684 // stack overflow detection.
685 //
686 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
687 // not use this flag. However, the stack of initial thread is not created
688 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
689 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
690 // and then attach the thread to JVM.
691 //
692 // To get around the problem and allow stack banging on Linux, we need to
693 // manually expand thread stack after receiving the SIGSEGV.
694 //
695 // There are two ways to expand thread stack to address "bottom", we used
696 // both of them in JVM before 1.5:
697 // 1. adjust stack pointer first so that it is below "bottom", and then
698 // touch "bottom"
699 // 2. mmap() the page in question
700 //
701 // Now alternate signal stack is gone, it's harder to use 2. For instance,
702 // if current sp is already near the lower end of page 101, and we need to
703 // call mmap() to map page 100, it is possible that part of the mmap() frame
704 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
705 // That will destroy the mmap() frame and cause VM to crash.
706 //
707 // The following code works by adjusting sp first, then accessing the "bottom"
708 // page to force a page fault. Linux kernel will then automatically expand the
709 // stack mapping.
710 //
711 // _expand_stack_to() assumes its frame size is less than page size, which
712 // should always be true if the function is not inlined.
713
714 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
715 #define NOINLINE
716 #else
717 #define NOINLINE __attribute__ ((noinline))
718 #endif
719
720 static void _expand_stack_to(address bottom) NOINLINE;
721
722 static void _expand_stack_to(address bottom) {
723 address sp;
724 size_t size;
725 volatile char *p;
726
727 // Adjust bottom to point to the largest address within the same page, it
728 // gives us a one-page buffer if alloca() allocates slightly more memory.
729 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
730 bottom += os::Linux::page_size() - 1;
731
732 // sp might be slightly above current stack pointer; if that's the case, we
733 // will alloca() a little more space than necessary, which is OK. Don't use
734 // os::current_stack_pointer(), as its result can be slightly below current
735 // stack pointer, causing us to not alloca enough to reach "bottom".
736 sp = (address)&sp;
737
738 if (sp > bottom) {
739 size = sp - bottom;
740 p = (volatile char *)alloca(size);
741 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
742 p[0] = '\0';
743 }
744 }
745
746 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
747 assert(t!=NULL, "just checking");
748 assert(t->osthread()->expanding_stack(), "expand should be set");
749 assert(t->stack_base() != NULL, "stack_base was not initialized");
750
751 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
752 sigset_t mask_all, old_sigset;
753 sigfillset(&mask_all);
754 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
755 _expand_stack_to(addr);
756 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
757 return true;
758 }
759 return false;
760 }
761
762 //////////////////////////////////////////////////////////////////////////////
763 // create new thread
764
765 static address highest_vm_reserved_address();
766
767 // check if it's safe to start a new thread
768 static bool _thread_safety_check(Thread* thread) {
769 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
770 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
771 // Heap is mmap'ed at lower end of memory space. Thread stacks are
772 // allocated (MAP_FIXED) from high address space. Every thread stack
773 // occupies a fixed size slot (usually 2Mbytes, but user can change
774 // it to other values if they rebuild LinuxThreads).
775 //
776 // Problem with MAP_FIXED is that mmap() can still succeed even part of
777 // the memory region has already been mmap'ed. That means if we have too
778 // many threads and/or very large heap, eventually thread stack will
779 // collide with heap.
780 //
781 // Here we try to prevent heap/stack collision by comparing current
782 // stack bottom with the highest address that has been mmap'ed by JVM
783 // plus a safety margin for memory maps created by native code.
784 //
785 // This feature can be disabled by setting ThreadSafetyMargin to 0
786 //
787 if (ThreadSafetyMargin > 0) {
788 address stack_bottom = os::current_stack_base() - os::current_stack_size();
789
790 // not safe if our stack extends below the safety margin
791 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
792 } else {
793 return true;
794 }
795 } else {
796 // Floating stack LinuxThreads or NPTL:
797 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
798 // there's not enough space left, pthread_create() will fail. If we come
799 // here, that means enough space has been reserved for stack.
800 return true;
801 }
802 }
803
804 // Thread start routine for all newly created threads
805 static void *java_start(Thread *thread) {
806 // Try to randomize the cache line index of hot stack frames.
807 // This helps when threads of the same stack traces evict each other's
808 // cache lines. The threads can be either from the same JVM instance, or
809 // from different JVM instances. The benefit is especially true for
810 // processors with hyperthreading technology.
811 static int counter = 0;
812 int pid = os::current_process_id();
813 alloca(((pid ^ counter++) & 7) * 128);
814
815 ThreadLocalStorage::set_thread(thread);
816
817 OSThread* osthread = thread->osthread();
818 Monitor* sync = osthread->startThread_lock();
819
820 // non floating stack LinuxThreads needs extra check, see above
821 if (!_thread_safety_check(thread)) {
822 // notify parent thread
823 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
824 osthread->set_state(ZOMBIE);
825 sync->notify_all();
826 return NULL;
827 }
828
829 // thread_id is kernel thread id (similar to Solaris LWP id)
830 osthread->set_thread_id(os::Linux::gettid());
831
832 if (UseNUMA) {
833 int lgrp_id = os::numa_get_group_id();
834 if (lgrp_id != -1) {
835 thread->set_lgrp_id(lgrp_id);
836 }
837 }
838 // initialize signal mask for this thread
839 os::Linux::hotspot_sigmask(thread);
840
841 // initialize floating point control register
842 os::Linux::init_thread_fpu_state();
843
844 // handshaking with parent thread
845 {
846 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
847
848 // notify parent thread
849 osthread->set_state(INITIALIZED);
850 sync->notify_all();
851
852 // wait until os::start_thread()
853 while (osthread->get_state() == INITIALIZED) {
854 sync->wait(Mutex::_no_safepoint_check_flag);
855 }
856 }
857
858 // call one more level start routine
859 thread->run();
860
861 return 0;
862 }
863
864 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
865 assert(thread->osthread() == NULL, "caller responsible");
866
867 // Allocate the OSThread object
868 OSThread* osthread = new OSThread(NULL, NULL);
869 if (osthread == NULL) {
870 return false;
871 }
872
873 // set the correct thread state
874 osthread->set_thread_type(thr_type);
875
876 // Initial state is ALLOCATED but not INITIALIZED
877 osthread->set_state(ALLOCATED);
878
879 thread->set_osthread(osthread);
880
881 // init thread attributes
882 pthread_attr_t attr;
883 pthread_attr_init(&attr);
884 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
885
886 // stack size
887 if (os::Linux::supports_variable_stack_size()) {
888 // calculate stack size if it's not specified by caller
889 if (stack_size == 0) {
890 stack_size = os::Linux::default_stack_size(thr_type);
891
892 switch (thr_type) {
893 case os::java_thread:
894 // Java threads use ThreadStackSize which default value can be
895 // changed with the flag -Xss
896 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
897 stack_size = JavaThread::stack_size_at_create();
898 break;
899 case os::compiler_thread:
900 if (CompilerThreadStackSize > 0) {
901 stack_size = (size_t)(CompilerThreadStackSize * K);
902 break;
903 } // else fall through:
904 // use VMThreadStackSize if CompilerThreadStackSize is not defined
905 case os::vm_thread:
906 case os::pgc_thread:
907 case os::cgc_thread:
908 case os::watcher_thread:
909 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
910 break;
911 }
912 }
913
914 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
915 pthread_attr_setstacksize(&attr, stack_size);
916 } else {
917 // let pthread_create() pick the default value.
918 }
919
920 // glibc guard page
921 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
922
923 ThreadState state;
924
925 {
926 // Serialize thread creation if we are running with fixed stack LinuxThreads
927 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
928 if (lock) {
929 os::Linux::createThread_lock()->lock_without_safepoint_check();
930 }
931
932 pthread_t tid;
933 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
934
935 pthread_attr_destroy(&attr);
936
937 if (ret != 0) {
938 if (PrintMiscellaneous && (Verbose || WizardMode)) {
939 perror("pthread_create()");
940 }
941 // Need to clean up stuff we've allocated so far
942 thread->set_osthread(NULL);
943 delete osthread;
944 if (lock) os::Linux::createThread_lock()->unlock();
945 return false;
946 }
947
948 // Store pthread info into the OSThread
949 osthread->set_pthread_id(tid);
950
951 // Wait until child thread is either initialized or aborted
952 {
953 Monitor* sync_with_child = osthread->startThread_lock();
954 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
955 while ((state = osthread->get_state()) == ALLOCATED) {
956 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
957 }
958 }
959
960 if (lock) {
961 os::Linux::createThread_lock()->unlock();
962 }
963 }
964
965 // Aborted due to thread limit being reached
966 if (state == ZOMBIE) {
967 thread->set_osthread(NULL);
968 delete osthread;
969 return false;
970 }
971
972 // The thread is returned suspended (in state INITIALIZED),
973 // and is started higher up in the call chain
974 assert(state == INITIALIZED, "race condition");
975 return true;
976 }
977
978 /////////////////////////////////////////////////////////////////////////////
979 // attach existing thread
980
981 // bootstrap the main thread
982 bool os::create_main_thread(JavaThread* thread) {
983 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
984 return create_attached_thread(thread);
985 }
986
987 bool os::create_attached_thread(JavaThread* thread) {
988 #ifdef ASSERT
989 thread->verify_not_published();
990 #endif
991
992 // Allocate the OSThread object
993 OSThread* osthread = new OSThread(NULL, NULL);
994
995 if (osthread == NULL) {
996 return false;
997 }
998
999 // Store pthread info into the OSThread
1000 osthread->set_thread_id(os::Linux::gettid());
1001 osthread->set_pthread_id(::pthread_self());
1002
1003 // initialize floating point control register
1004 os::Linux::init_thread_fpu_state();
1005
1006 // Initial thread state is RUNNABLE
1007 osthread->set_state(RUNNABLE);
1008
1009 thread->set_osthread(osthread);
1010
1011 if (UseNUMA) {
1012 int lgrp_id = os::numa_get_group_id();
1013 if (lgrp_id != -1) {
1014 thread->set_lgrp_id(lgrp_id);
1015 }
1016 }
1017
1018 if (os::Linux::is_initial_thread()) {
1019 // If current thread is initial thread, its stack is mapped on demand,
1020 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1021 // the entire stack region to avoid SEGV in stack banging.
1022 // It is also useful to get around the heap-stack-gap problem on SuSE
1023 // kernel (see 4821821 for details). We first expand stack to the top
1024 // of yellow zone, then enable stack yellow zone (order is significant,
1025 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1026 // is no gap between the last two virtual memory regions.
1027
1028 JavaThread *jt = (JavaThread *)thread;
1029 address addr = jt->stack_yellow_zone_base();
1030 assert(addr != NULL, "initialization problem?");
1031 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1032
1033 osthread->set_expanding_stack();
1034 os::Linux::manually_expand_stack(jt, addr);
1035 osthread->clear_expanding_stack();
1036 }
1037
1038 // initialize signal mask for this thread
1039 // and save the caller's signal mask
1040 os::Linux::hotspot_sigmask(thread);
1041
1042 return true;
1043 }
1044
1045 void os::pd_start_thread(Thread* thread) {
1046 OSThread * osthread = thread->osthread();
1047 assert(osthread->get_state() != INITIALIZED, "just checking");
1048 Monitor* sync_with_child = osthread->startThread_lock();
1049 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1050 sync_with_child->notify();
1051 }
1052
1053 // Free Linux resources related to the OSThread
1054 void os::free_thread(OSThread* osthread) {
1055 assert(osthread != NULL, "osthread not set");
1056
1057 if (Thread::current()->osthread() == osthread) {
1058 // Restore caller's signal mask
1059 sigset_t sigmask = osthread->caller_sigmask();
1060 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1061 }
1062
1063 delete osthread;
1064 }
1065
1066 //////////////////////////////////////////////////////////////////////////////
1067 // thread local storage
1068
1069 int os::allocate_thread_local_storage() {
1070 pthread_key_t key;
1071 int rslt = pthread_key_create(&key, NULL);
1072 assert(rslt == 0, "cannot allocate thread local storage");
1073 return (int)key;
1074 }
1075
1076 // Note: This is currently not used by VM, as we don't destroy TLS key
1077 // on VM exit.
1078 void os::free_thread_local_storage(int index) {
1079 int rslt = pthread_key_delete((pthread_key_t)index);
1080 assert(rslt == 0, "invalid index");
1081 }
1082
1083 void os::thread_local_storage_at_put(int index, void* value) {
1084 int rslt = pthread_setspecific((pthread_key_t)index, value);
1085 assert(rslt == 0, "pthread_setspecific failed");
1086 }
1087
1088 extern "C" Thread* get_thread() {
1089 return ThreadLocalStorage::thread();
1090 }
1091
1092 //////////////////////////////////////////////////////////////////////////////
1093 // initial thread
1094
1095 // Check if current thread is the initial thread, similar to Solaris thr_main.
1096 bool os::Linux::is_initial_thread(void) {
1097 char dummy;
1098 // If called before init complete, thread stack bottom will be null.
1099 // Can be called if fatal error occurs before initialization.
1100 if (initial_thread_stack_bottom() == NULL) return false;
1101 assert(initial_thread_stack_bottom() != NULL &&
1102 initial_thread_stack_size() != 0,
1103 "os::init did not locate initial thread's stack region");
1104 if ((address)&dummy >= initial_thread_stack_bottom() &&
1105 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1106 return true;
1107 else return false;
1108 }
1109
1110 // Find the virtual memory area that contains addr
1111 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1112 FILE *fp = fopen("/proc/self/maps", "r");
1113 if (fp) {
1114 address low, high;
1115 while (!feof(fp)) {
1116 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1117 if (low <= addr && addr < high) {
1118 if (vma_low) *vma_low = low;
1119 if (vma_high) *vma_high = high;
1120 fclose (fp);
1121 return true;
1122 }
1123 }
1124 for (;;) {
1125 int ch = fgetc(fp);
1126 if (ch == EOF || ch == (int)'\n') break;
1127 }
1128 }
1129 fclose(fp);
1130 }
1131 return false;
1132 }
1133
1134 // Locate initial thread stack. This special handling of initial thread stack
1135 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1136 // bogus value for initial thread.
1137 void os::Linux::capture_initial_stack(size_t max_size) {
1138 // stack size is the easy part, get it from RLIMIT_STACK
1139 size_t stack_size;
1140 struct rlimit rlim;
1141 getrlimit(RLIMIT_STACK, &rlim);
1142 stack_size = rlim.rlim_cur;
1143
1144 // 6308388: a bug in ld.so will relocate its own .data section to the
1145 // lower end of primordial stack; reduce ulimit -s value a little bit
1146 // so we won't install guard page on ld.so's data section.
1147 stack_size -= 2 * page_size();
1148
1149 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1150 // 7.1, in both cases we will get 2G in return value.
1151 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1152 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1153 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1154 // in case other parts in glibc still assumes 2M max stack size.
1155 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1156 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1157 if (stack_size > 2 * K * K IA64_ONLY(*2))
1158 stack_size = 2 * K * K IA64_ONLY(*2);
1159 // Try to figure out where the stack base (top) is. This is harder.
1160 //
1161 // When an application is started, glibc saves the initial stack pointer in
1162 // a global variable "__libc_stack_end", which is then used by system
1163 // libraries. __libc_stack_end should be pretty close to stack top. The
1164 // variable is available since the very early days. However, because it is
1165 // a private interface, it could disappear in the future.
1166 //
1167 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1168 // to __libc_stack_end, it is very close to stack top, but isn't the real
1169 // stack top. Note that /proc may not exist if VM is running as a chroot
1170 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1171 // /proc/<pid>/stat could change in the future (though unlikely).
1172 //
1173 // We try __libc_stack_end first. If that doesn't work, look for
1174 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1175 // as a hint, which should work well in most cases.
1176
1177 uintptr_t stack_start;
1178
1179 // try __libc_stack_end first
1180 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1181 if (p && *p) {
1182 stack_start = *p;
1183 } else {
1184 // see if we can get the start_stack field from /proc/self/stat
1185 FILE *fp;
1186 int pid;
1187 char state;
1188 int ppid;
1189 int pgrp;
1190 int session;
1191 int nr;
1192 int tpgrp;
1193 unsigned long flags;
1194 unsigned long minflt;
1195 unsigned long cminflt;
1196 unsigned long majflt;
1197 unsigned long cmajflt;
1198 unsigned long utime;
1199 unsigned long stime;
1200 long cutime;
1201 long cstime;
1202 long prio;
1203 long nice;
1204 long junk;
1205 long it_real;
1206 uintptr_t start;
1207 uintptr_t vsize;
1208 intptr_t rss;
1209 uintptr_t rsslim;
1210 uintptr_t scodes;
1211 uintptr_t ecode;
1212 int i;
1213
1214 // Figure what the primordial thread stack base is. Code is inspired
1215 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1216 // followed by command name surrounded by parentheses, state, etc.
1217 char stat[2048];
1218 int statlen;
1219
1220 fp = fopen("/proc/self/stat", "r");
1221 if (fp) {
1222 statlen = fread(stat, 1, 2047, fp);
1223 stat[statlen] = '\0';
1224 fclose(fp);
1225
1226 // Skip pid and the command string. Note that we could be dealing with
1227 // weird command names, e.g. user could decide to rename java launcher
1228 // to "java 1.4.2 :)", then the stat file would look like
1229 // 1234 (java 1.4.2 :)) R ... ...
1230 // We don't really need to know the command string, just find the last
1231 // occurrence of ")" and then start parsing from there. See bug 4726580.
1232 char * s = strrchr(stat, ')');
1233
1234 i = 0;
1235 if (s) {
1236 // Skip blank chars
1237 do s++; while (isspace(*s));
1238
1239 #define _UFM UINTX_FORMAT
1240 #define _DFM INTX_FORMAT
1241
1242 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1243 /* 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 */
1244 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,
1245 &state, /* 3 %c */
1246 &ppid, /* 4 %d */
1247 &pgrp, /* 5 %d */
1248 &session, /* 6 %d */
1249 &nr, /* 7 %d */
1250 &tpgrp, /* 8 %d */
1251 &flags, /* 9 %lu */
1252 &minflt, /* 10 %lu */
1253 &cminflt, /* 11 %lu */
1254 &majflt, /* 12 %lu */
1255 &cmajflt, /* 13 %lu */
1256 &utime, /* 14 %lu */
1257 &stime, /* 15 %lu */
1258 &cutime, /* 16 %ld */
1259 &cstime, /* 17 %ld */
1260 &prio, /* 18 %ld */
1261 &nice, /* 19 %ld */
1262 &junk, /* 20 %ld */
1263 &it_real, /* 21 %ld */
1264 &start, /* 22 UINTX_FORMAT */
1265 &vsize, /* 23 UINTX_FORMAT */
1266 &rss, /* 24 INTX_FORMAT */
1267 &rsslim, /* 25 UINTX_FORMAT */
1268 &scodes, /* 26 UINTX_FORMAT */
1269 &ecode, /* 27 UINTX_FORMAT */
1270 &stack_start); /* 28 UINTX_FORMAT */
1271 }
1272
1273 #undef _UFM
1274 #undef _DFM
1275
1276 if (i != 28 - 2) {
1277 assert(false, "Bad conversion from /proc/self/stat");
1278 // product mode - assume we are the initial thread, good luck in the
1279 // embedded case.
1280 warning("Can't detect initial thread stack location - bad conversion");
1281 stack_start = (uintptr_t) &rlim;
1282 }
1283 } else {
1284 // For some reason we can't open /proc/self/stat (for example, running on
1285 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1286 // most cases, so don't abort:
1287 warning("Can't detect initial thread stack location - no /proc/self/stat");
1288 stack_start = (uintptr_t) &rlim;
1289 }
1290 }
1291
1292 // Now we have a pointer (stack_start) very close to the stack top, the
1293 // next thing to do is to figure out the exact location of stack top. We
1294 // can find out the virtual memory area that contains stack_start by
1295 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1296 // and its upper limit is the real stack top. (again, this would fail if
1297 // running inside chroot, because /proc may not exist.)
1298
1299 uintptr_t stack_top;
1300 address low, high;
1301 if (find_vma((address)stack_start, &low, &high)) {
1302 // success, "high" is the true stack top. (ignore "low", because initial
1303 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1304 stack_top = (uintptr_t)high;
1305 } else {
1306 // failed, likely because /proc/self/maps does not exist
1307 warning("Can't detect initial thread stack location - find_vma failed");
1308 // best effort: stack_start is normally within a few pages below the real
1309 // stack top, use it as stack top, and reduce stack size so we won't put
1310 // guard page outside stack.
1311 stack_top = stack_start;
1312 stack_size -= 16 * page_size();
1313 }
1314
1315 // stack_top could be partially down the page so align it
1316 stack_top = align_size_up(stack_top, page_size());
1317
1318 if (max_size && stack_size > max_size) {
1319 _initial_thread_stack_size = max_size;
1320 } else {
1321 _initial_thread_stack_size = stack_size;
1322 }
1323
1324 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1325 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1326 }
1327
1328 ////////////////////////////////////////////////////////////////////////////////
1329 // time support
1330
1331 // Time since start-up in seconds to a fine granularity.
1332 // Used by VMSelfDestructTimer and the MemProfiler.
1333 double os::elapsedTime() {
1334
1335 return (double)(os::elapsed_counter()) * 0.000001;
1336 }
1337
1338 jlong os::elapsed_counter() {
1339 timeval time;
1340 int status = gettimeofday(&time, NULL);
1341 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1342 }
1343
1344 jlong os::elapsed_frequency() {
1345 return (1000 * 1000);
1346 }
1347
1348 bool os::supports_vtime() { return true; }
1349 bool os::enable_vtime() { return false; }
1350 bool os::vtime_enabled() { return false; }
1351
1352 double os::elapsedVTime() {
1353 struct rusage usage;
1354 int retval = getrusage(RUSAGE_THREAD, &usage);
1355 if (retval == 0) {
1356 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);
1357 } else {
1358 // better than nothing, but not much
1359 return elapsedTime();
1360 }
1361 }
1362
1363 jlong os::javaTimeMillis() {
1364 timeval time;
1365 int status = gettimeofday(&time, NULL);
1366 assert(status != -1, "linux error");
1367 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1368 }
1369
1370 #ifndef CLOCK_MONOTONIC
1371 #define CLOCK_MONOTONIC (1)
1372 #endif
1373
1374 void os::Linux::clock_init() {
1375 // we do dlopen's in this particular order due to bug in linux
1376 // dynamical loader (see 6348968) leading to crash on exit
1377 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1378 if (handle == NULL) {
1379 handle = dlopen("librt.so", RTLD_LAZY);
1380 }
1381
1382 if (handle) {
1383 int (*clock_getres_func)(clockid_t, struct timespec*) =
1384 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1385 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1386 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1387 if (clock_getres_func && clock_gettime_func) {
1388 // See if monotonic clock is supported by the kernel. Note that some
1389 // early implementations simply return kernel jiffies (updated every
1390 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1391 // for nano time (though the monotonic property is still nice to have).
1392 // It's fixed in newer kernels, however clock_getres() still returns
1393 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1394 // resolution for now. Hopefully as people move to new kernels, this
1395 // won't be a problem.
1396 struct timespec res;
1397 struct timespec tp;
1398 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1399 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1400 // yes, monotonic clock is supported
1401 _clock_gettime = clock_gettime_func;
1402 } else {
1403 // close librt if there is no monotonic clock
1404 dlclose(handle);
1405 }
1406 }
1407 }
1408 }
1409
1410 #ifndef SYS_clock_getres
1411
1412 #if defined(IA32) || defined(AMD64)
1413 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1414 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1415 #else
1416 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1417 #define sys_clock_getres(x,y) -1
1418 #endif
1419
1420 #else
1421 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1422 #endif
1423
1424 void os::Linux::fast_thread_clock_init() {
1425 if (!UseLinuxPosixThreadCPUClocks) {
1426 return;
1427 }
1428 clockid_t clockid;
1429 struct timespec tp;
1430 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1431 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1432
1433 // Switch to using fast clocks for thread cpu time if
1434 // the sys_clock_getres() returns 0 error code.
1435 // Note, that some kernels may support the current thread
1436 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1437 // returned by the pthread_getcpuclockid().
1438 // If the fast Posix clocks are supported then the sys_clock_getres()
1439 // must return at least tp.tv_sec == 0 which means a resolution
1440 // better than 1 sec. This is extra check for reliability.
1441
1442 if(pthread_getcpuclockid_func &&
1443 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1444 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1445
1446 _supports_fast_thread_cpu_time = true;
1447 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1448 }
1449 }
1450
1451 jlong os::javaTimeNanos() {
1452 if (Linux::supports_monotonic_clock()) {
1453 struct timespec tp;
1454 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1455 assert(status == 0, "gettime error");
1456 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1457 return result;
1458 } else {
1459 timeval time;
1460 int status = gettimeofday(&time, NULL);
1461 assert(status != -1, "linux error");
1462 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1463 return 1000 * usecs;
1464 }
1465 }
1466
1467 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1468 if (Linux::supports_monotonic_clock()) {
1469 info_ptr->max_value = ALL_64_BITS;
1470
1471 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1472 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1473 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1474 } else {
1475 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1476 info_ptr->max_value = ALL_64_BITS;
1477
1478 // gettimeofday is a real time clock so it skips
1479 info_ptr->may_skip_backward = true;
1480 info_ptr->may_skip_forward = true;
1481 }
1482
1483 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1484 }
1485
1486 // Return the real, user, and system times in seconds from an
1487 // arbitrary fixed point in the past.
1488 bool os::getTimesSecs(double* process_real_time,
1489 double* process_user_time,
1490 double* process_system_time) {
1491 struct tms ticks;
1492 clock_t real_ticks = times(&ticks);
1493
1494 if (real_ticks == (clock_t) (-1)) {
1495 return false;
1496 } else {
1497 double ticks_per_second = (double) clock_tics_per_sec;
1498 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1499 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1500 *process_real_time = ((double) real_ticks) / ticks_per_second;
1501
1502 return true;
1503 }
1504 }
1505
1506
1507 char * os::local_time_string(char *buf, size_t buflen) {
1508 struct tm t;
1509 time_t long_time;
1510 time(&long_time);
1511 localtime_r(&long_time, &t);
1512 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1513 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1514 t.tm_hour, t.tm_min, t.tm_sec);
1515 return buf;
1516 }
1517
1518 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1519 return localtime_r(clock, res);
1520 }
1521
1522 ////////////////////////////////////////////////////////////////////////////////
1523 // runtime exit support
1524
1525 // Note: os::shutdown() might be called very early during initialization, or
1526 // called from signal handler. Before adding something to os::shutdown(), make
1527 // sure it is async-safe and can handle partially initialized VM.
1528 void os::shutdown() {
1529
1530 // allow PerfMemory to attempt cleanup of any persistent resources
1531 perfMemory_exit();
1532
1533 // needs to remove object in file system
1534 AttachListener::abort();
1535
1536 // flush buffered output, finish log files
1537 ostream_abort();
1538
1539 // Check for abort hook
1540 abort_hook_t abort_hook = Arguments::abort_hook();
1541 if (abort_hook != NULL) {
1542 abort_hook();
1543 }
1544
1545 }
1546
1547 // Note: os::abort() might be called very early during initialization, or
1548 // called from signal handler. Before adding something to os::abort(), make
1549 // sure it is async-safe and can handle partially initialized VM.
1550 void os::abort(bool dump_core) {
1551 os::shutdown();
1552 if (dump_core) {
1553 #ifndef PRODUCT
1554 fdStream out(defaultStream::output_fd());
1555 out.print_raw("Current thread is ");
1556 char buf[16];
1557 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1558 out.print_raw_cr(buf);
1559 out.print_raw_cr("Dumping core ...");
1560 #endif
1561 ::abort(); // dump core
1562 }
1563
1564 ::exit(1);
1565 }
1566
1567 // Die immediately, no exit hook, no abort hook, no cleanup.
1568 void os::die() {
1569 // _exit() on LinuxThreads only kills current thread
1570 ::abort();
1571 }
1572
1573 // unused on linux for now.
1574 void os::set_error_file(const char *logfile) {}
1575
1576
1577 // This method is a copy of JDK's sysGetLastErrorString
1578 // from src/solaris/hpi/src/system_md.c
1579
1580 size_t os::lasterror(char *buf, size_t len) {
1581
1582 if (errno == 0) return 0;
1583
1584 const char *s = ::strerror(errno);
1585 size_t n = ::strlen(s);
1586 if (n >= len) {
1587 n = len - 1;
1588 }
1589 ::strncpy(buf, s, n);
1590 buf[n] = '\0';
1591 return n;
1592 }
1593
1594 intx os::current_thread_id() { return (intx)pthread_self(); }
1595 int os::current_process_id() {
1596
1597 // Under the old linux thread library, linux gives each thread
1598 // its own process id. Because of this each thread will return
1599 // a different pid if this method were to return the result
1600 // of getpid(2). Linux provides no api that returns the pid
1601 // of the launcher thread for the vm. This implementation
1602 // returns a unique pid, the pid of the launcher thread
1603 // that starts the vm 'process'.
1604
1605 // Under the NPTL, getpid() returns the same pid as the
1606 // launcher thread rather than a unique pid per thread.
1607 // Use gettid() if you want the old pre NPTL behaviour.
1608
1609 // if you are looking for the result of a call to getpid() that
1610 // returns a unique pid for the calling thread, then look at the
1611 // OSThread::thread_id() method in osThread_linux.hpp file
1612
1613 return (int)(_initial_pid ? _initial_pid : getpid());
1614 }
1615
1616 // DLL functions
1617
1618 const char* os::dll_file_extension() { return ".so"; }
1619
1620 // This must be hard coded because it's the system's temporary
1621 // directory not the java application's temp directory, ala java.io.tmpdir.
1622 const char* os::get_temp_directory() { return "/tmp"; }
1623
1624 static bool file_exists(const char* filename) {
1625 struct stat statbuf;
1626 if (filename == NULL || strlen(filename) == 0) {
1627 return false;
1628 }
1629 return os::stat(filename, &statbuf) == 0;
1630 }
1631
1632 bool os::dll_build_name(char* buffer, size_t buflen,
1633 const char* pname, const char* fname) {
1634 bool retval = false;
1635 // Copied from libhpi
1636 const size_t pnamelen = pname ? strlen(pname) : 0;
1637
1638 // Return error on buffer overflow.
1639 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1640 return retval;
1641 }
1642
1643 if (pnamelen == 0) {
1644 snprintf(buffer, buflen, "lib%s.so", fname);
1645 retval = true;
1646 } else if (strchr(pname, *os::path_separator()) != NULL) {
1647 int n;
1648 char** pelements = split_path(pname, &n);
1649 if (pelements == NULL) {
1650 return false;
1651 }
1652 for (int i = 0 ; i < n ; i++) {
1653 // Really shouldn't be NULL, but check can't hurt
1654 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1655 continue; // skip the empty path values
1656 }
1657 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1658 if (file_exists(buffer)) {
1659 retval = true;
1660 break;
1661 }
1662 }
1663 // release the storage
1664 for (int i = 0 ; i < n ; i++) {
1665 if (pelements[i] != NULL) {
1666 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1667 }
1668 }
1669 if (pelements != NULL) {
1670 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1671 }
1672 } else {
1673 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1674 retval = true;
1675 }
1676 return retval;
1677 }
1678
1679 // check if addr is inside libjvm.so
1680 bool os::address_is_in_vm(address addr) {
1681 static address libjvm_base_addr;
1682 Dl_info dlinfo;
1683
1684 if (libjvm_base_addr == NULL) {
1685 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1686 libjvm_base_addr = (address)dlinfo.dli_fbase;
1687 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1688 }
1689
1690 if (dladdr((void *)addr, &dlinfo)) {
1691 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1692 }
1693
1694 return false;
1695 }
1696
1697 bool os::dll_address_to_function_name(address addr, char *buf,
1698 int buflen, int *offset) {
1699 Dl_info dlinfo;
1700
1701 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1702 if (buf != NULL) {
1703 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1704 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1705 }
1706 }
1707 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1708 return true;
1709 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1710 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1711 buf, buflen, offset, dlinfo.dli_fname)) {
1712 return true;
1713 }
1714 }
1715
1716 if (buf != NULL) buf[0] = '\0';
1717 if (offset != NULL) *offset = -1;
1718 return false;
1719 }
1720
1721 struct _address_to_library_name {
1722 address addr; // input : memory address
1723 size_t buflen; // size of fname
1724 char* fname; // output: library name
1725 address base; // library base addr
1726 };
1727
1728 static int address_to_library_name_callback(struct dl_phdr_info *info,
1729 size_t size, void *data) {
1730 int i;
1731 bool found = false;
1732 address libbase = NULL;
1733 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1734
1735 // iterate through all loadable segments
1736 for (i = 0; i < info->dlpi_phnum; i++) {
1737 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1738 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1739 // base address of a library is the lowest address of its loaded
1740 // segments.
1741 if (libbase == NULL || libbase > segbase) {
1742 libbase = segbase;
1743 }
1744 // see if 'addr' is within current segment
1745 if (segbase <= d->addr &&
1746 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1747 found = true;
1748 }
1749 }
1750 }
1751
1752 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1753 // so dll_address_to_library_name() can fall through to use dladdr() which
1754 // can figure out executable name from argv[0].
1755 if (found && info->dlpi_name && info->dlpi_name[0]) {
1756 d->base = libbase;
1757 if (d->fname) {
1758 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1759 }
1760 return 1;
1761 }
1762 return 0;
1763 }
1764
1765 bool os::dll_address_to_library_name(address addr, char* buf,
1766 int buflen, int* offset) {
1767 Dl_info dlinfo;
1768 struct _address_to_library_name data;
1769
1770 // There is a bug in old glibc dladdr() implementation that it could resolve
1771 // to wrong library name if the .so file has a base address != NULL. Here
1772 // we iterate through the program headers of all loaded libraries to find
1773 // out which library 'addr' really belongs to. This workaround can be
1774 // removed once the minimum requirement for glibc is moved to 2.3.x.
1775 data.addr = addr;
1776 data.fname = buf;
1777 data.buflen = buflen;
1778 data.base = NULL;
1779 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1780
1781 if (rslt) {
1782 // buf already contains library name
1783 if (offset) *offset = addr - data.base;
1784 return true;
1785 } else if (dladdr((void*)addr, &dlinfo)){
1786 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1787 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1788 return true;
1789 } else {
1790 if (buf) buf[0] = '\0';
1791 if (offset) *offset = -1;
1792 return false;
1793 }
1794 }
1795
1796 // Loads .dll/.so and
1797 // in case of error it checks if .dll/.so was built for the
1798 // same architecture as Hotspot is running on
1799
1800
1801 // Remember the stack's state. The Linux dynamic linker will change
1802 // the stack to 'executable' at most once, so we must safepoint only once.
1803 bool os::Linux::_stack_is_executable = false;
1804
1805 // VM operation that loads a library. This is necessary if stack protection
1806 // of the Java stacks can be lost during loading the library. If we
1807 // do not stop the Java threads, they can stack overflow before the stacks
1808 // are protected again.
1809 class VM_LinuxDllLoad: public VM_Operation {
1810 private:
1811 const char *_filename;
1812 char *_ebuf;
1813 int _ebuflen;
1814 void *_lib;
1815 public:
1816 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1817 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1818 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1819 void doit() {
1820 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1821 os::Linux::_stack_is_executable = true;
1822 }
1823 void* loaded_library() { return _lib; }
1824 };
1825
1826 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1827 {
1828 void * result = NULL;
1829 bool load_attempted = false;
1830
1831 // Check whether the library to load might change execution rights
1832 // of the stack. If they are changed, the protection of the stack
1833 // guard pages will be lost. We need a safepoint to fix this.
1834 //
1835 // See Linux man page execstack(8) for more info.
1836 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1837 ElfFile ef(filename);
1838 if (!ef.specifies_noexecstack()) {
1839 if (!is_init_completed()) {
1840 os::Linux::_stack_is_executable = true;
1841 // This is OK - No Java threads have been created yet, and hence no
1842 // stack guard pages to fix.
1843 //
1844 // This should happen only when you are building JDK7 using a very
1845 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1846 //
1847 // Dynamic loader will make all stacks executable after
1848 // this function returns, and will not do that again.
1849 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1850 } else {
1851 warning("You have loaded library %s which might have disabled stack guard. "
1852 "The VM will try to fix the stack guard now.\n"
1853 "It's highly recommended that you fix the library with "
1854 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1855 filename);
1856
1857 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1858 JavaThread *jt = JavaThread::current();
1859 if (jt->thread_state() != _thread_in_native) {
1860 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1861 // that requires ExecStack. Cannot enter safe point. Let's give up.
1862 warning("Unable to fix stack guard. Giving up.");
1863 } else {
1864 if (!LoadExecStackDllInVMThread) {
1865 // This is for the case where the DLL has an static
1866 // constructor function that executes JNI code. We cannot
1867 // load such DLLs in the VMThread.
1868 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1869 }
1870
1871 ThreadInVMfromNative tiv(jt);
1872 debug_only(VMNativeEntryWrapper vew;)
1873
1874 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1875 VMThread::execute(&op);
1876 if (LoadExecStackDllInVMThread) {
1877 result = op.loaded_library();
1878 }
1879 load_attempted = true;
1880 }
1881 }
1882 }
1883 }
1884
1885 if (!load_attempted) {
1886 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1887 }
1888
1889 if (result != NULL) {
1890 // Successful loading
1891 return result;
1892 }
1893
1894 Elf32_Ehdr elf_head;
1895 int diag_msg_max_length=ebuflen-strlen(ebuf);
1896 char* diag_msg_buf=ebuf+strlen(ebuf);
1897
1898 if (diag_msg_max_length==0) {
1899 // No more space in ebuf for additional diagnostics message
1900 return NULL;
1901 }
1902
1903
1904 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1905
1906 if (file_descriptor < 0) {
1907 // Can't open library, report dlerror() message
1908 return NULL;
1909 }
1910
1911 bool failed_to_read_elf_head=
1912 (sizeof(elf_head)!=
1913 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1914
1915 ::close(file_descriptor);
1916 if (failed_to_read_elf_head) {
1917 // file i/o error - report dlerror() msg
1918 return NULL;
1919 }
1920
1921 typedef struct {
1922 Elf32_Half code; // Actual value as defined in elf.h
1923 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1924 char elf_class; // 32 or 64 bit
1925 char endianess; // MSB or LSB
1926 char* name; // String representation
1927 } arch_t;
1928
1929 #ifndef EM_486
1930 #define EM_486 6 /* Intel 80486 */
1931 #endif
1932
1933 static const arch_t arch_array[]={
1934 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1935 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1936 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1937 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1938 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1939 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1940 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1941 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1942 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1943 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1944 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1945 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1946 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1947 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1948 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1949 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1950 };
1951
1952 #if (defined IA32)
1953 static Elf32_Half running_arch_code=EM_386;
1954 #elif (defined AMD64)
1955 static Elf32_Half running_arch_code=EM_X86_64;
1956 #elif (defined IA64)
1957 static Elf32_Half running_arch_code=EM_IA_64;
1958 #elif (defined __sparc) && (defined _LP64)
1959 static Elf32_Half running_arch_code=EM_SPARCV9;
1960 #elif (defined __sparc) && (!defined _LP64)
1961 static Elf32_Half running_arch_code=EM_SPARC;
1962 #elif (defined __powerpc64__)
1963 static Elf32_Half running_arch_code=EM_PPC64;
1964 #elif (defined __powerpc__)
1965 static Elf32_Half running_arch_code=EM_PPC;
1966 #elif (defined ARM)
1967 static Elf32_Half running_arch_code=EM_ARM;
1968 #elif (defined S390)
1969 static Elf32_Half running_arch_code=EM_S390;
1970 #elif (defined ALPHA)
1971 static Elf32_Half running_arch_code=EM_ALPHA;
1972 #elif (defined MIPSEL)
1973 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1974 #elif (defined PARISC)
1975 static Elf32_Half running_arch_code=EM_PARISC;
1976 #elif (defined MIPS)
1977 static Elf32_Half running_arch_code=EM_MIPS;
1978 #elif (defined M68K)
1979 static Elf32_Half running_arch_code=EM_68K;
1980 #else
1981 #error Method os::dll_load requires that one of following is defined:\
1982 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1983 #endif
1984
1985 // Identify compatability class for VM's architecture and library's architecture
1986 // Obtain string descriptions for architectures
1987
1988 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1989 int running_arch_index=-1;
1990
1991 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1992 if (running_arch_code == arch_array[i].code) {
1993 running_arch_index = i;
1994 }
1995 if (lib_arch.code == arch_array[i].code) {
1996 lib_arch.compat_class = arch_array[i].compat_class;
1997 lib_arch.name = arch_array[i].name;
1998 }
1999 }
2000
2001 assert(running_arch_index != -1,
2002 "Didn't find running architecture code (running_arch_code) in arch_array");
2003 if (running_arch_index == -1) {
2004 // Even though running architecture detection failed
2005 // we may still continue with reporting dlerror() message
2006 return NULL;
2007 }
2008
2009 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2010 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2011 return NULL;
2012 }
2013
2014 #ifndef S390
2015 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2016 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2017 return NULL;
2018 }
2019 #endif // !S390
2020
2021 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2022 if ( lib_arch.name!=NULL ) {
2023 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2024 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2025 lib_arch.name, arch_array[running_arch_index].name);
2026 } else {
2027 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2028 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2029 lib_arch.code,
2030 arch_array[running_arch_index].name);
2031 }
2032 }
2033
2034 return NULL;
2035 }
2036
2037 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2038 void * result = ::dlopen(filename, RTLD_LAZY);
2039 if (result == NULL) {
2040 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2041 ebuf[ebuflen-1] = '\0';
2042 }
2043 return result;
2044 }
2045
2046 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2047 void * result = NULL;
2048 if (LoadExecStackDllInVMThread) {
2049 result = dlopen_helper(filename, ebuf, ebuflen);
2050 }
2051
2052 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2053 // library that requires an executable stack, or which does not have this
2054 // stack attribute set, dlopen changes the stack attribute to executable. The
2055 // read protection of the guard pages gets lost.
2056 //
2057 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2058 // may have been queued at the same time.
2059
2060 if (!_stack_is_executable) {
2061 JavaThread *jt = Threads::first();
2062
2063 while (jt) {
2064 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2065 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2066 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2067 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2068 warning("Attempt to reguard stack yellow zone failed.");
2069 }
2070 }
2071 jt = jt->next();
2072 }
2073 }
2074
2075 return result;
2076 }
2077
2078 /*
2079 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2080 * chances are you might want to run the generated bits against glibc-2.0
2081 * libdl.so, so always use locking for any version of glibc.
2082 */
2083 void* os::dll_lookup(void* handle, const char* name) {
2084 pthread_mutex_lock(&dl_mutex);
2085 void* res = dlsym(handle, name);
2086 pthread_mutex_unlock(&dl_mutex);
2087 return res;
2088 }
2089
2090
2091 static bool _print_ascii_file(const char* filename, outputStream* st) {
2092 int fd = ::open(filename, O_RDONLY);
2093 if (fd == -1) {
2094 return false;
2095 }
2096
2097 char buf[32];
2098 int bytes;
2099 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2100 st->print_raw(buf, bytes);
2101 }
2102
2103 ::close(fd);
2104
2105 return true;
2106 }
2107
2108 void os::print_dll_info(outputStream *st) {
2109 st->print_cr("Dynamic libraries:");
2110
2111 char fname[32];
2112 pid_t pid = os::Linux::gettid();
2113
2114 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2115
2116 if (!_print_ascii_file(fname, st)) {
2117 st->print("Can not get library information for pid = %d\n", pid);
2118 }
2119 }
2120
2121 void os::print_os_info_brief(outputStream* st) {
2122 os::Linux::print_distro_info(st);
2123
2124 os::Posix::print_uname_info(st);
2125
2126 os::Linux::print_libversion_info(st);
2127
2128 }
2129
2130 void os::print_os_info(outputStream* st) {
2131 st->print("OS:");
2132
2133 os::Linux::print_distro_info(st);
2134
2135 os::Posix::print_uname_info(st);
2136
2137 // Print warning if unsafe chroot environment detected
2138 if (unsafe_chroot_detected) {
2139 st->print("WARNING!! ");
2140 st->print_cr(unstable_chroot_error);
2141 }
2142
2143 os::Linux::print_libversion_info(st);
2144
2145 os::Posix::print_rlimit_info(st);
2146
2147 os::Posix::print_load_average(st);
2148
2149 os::Linux::print_full_memory_info(st);
2150 }
2151
2152 // Try to identify popular distros.
2153 // Most Linux distributions have /etc/XXX-release file, which contains
2154 // the OS version string. Some have more than one /etc/XXX-release file
2155 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2156 // so the order is important.
2157 void os::Linux::print_distro_info(outputStream* st) {
2158 if (!_print_ascii_file("/etc/mandrake-release", st) &&
2159 !_print_ascii_file("/etc/sun-release", st) &&
2160 !_print_ascii_file("/etc/redhat-release", st) &&
2161 !_print_ascii_file("/etc/SuSE-release", st) &&
2162 !_print_ascii_file("/etc/turbolinux-release", st) &&
2163 !_print_ascii_file("/etc/gentoo-release", st) &&
2164 !_print_ascii_file("/etc/debian_version", st) &&
2165 !_print_ascii_file("/etc/ltib-release", st) &&
2166 !_print_ascii_file("/etc/angstrom-version", st)) {
2167 st->print("Linux");
2168 }
2169 st->cr();
2170 }
2171
2172 void os::Linux::print_libversion_info(outputStream* st) {
2173 // libc, pthread
2174 st->print("libc:");
2175 st->print(os::Linux::glibc_version()); st->print(" ");
2176 st->print(os::Linux::libpthread_version()); st->print(" ");
2177 if (os::Linux::is_LinuxThreads()) {
2178 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2179 }
2180 st->cr();
2181 }
2182
2183 void os::Linux::print_full_memory_info(outputStream* st) {
2184 st->print("\n/proc/meminfo:\n");
2185 _print_ascii_file("/proc/meminfo", st);
2186 st->cr();
2187 }
2188
2189 void os::print_memory_info(outputStream* st) {
2190
2191 st->print("Memory:");
2192 st->print(" %dk page", os::vm_page_size()>>10);
2193
2194 // values in struct sysinfo are "unsigned long"
2195 struct sysinfo si;
2196 sysinfo(&si);
2197
2198 st->print(", physical " UINT64_FORMAT "k",
2199 os::physical_memory() >> 10);
2200 st->print("(" UINT64_FORMAT "k free)",
2201 os::available_memory() >> 10);
2202 st->print(", swap " UINT64_FORMAT "k",
2203 ((jlong)si.totalswap * si.mem_unit) >> 10);
2204 st->print("(" UINT64_FORMAT "k free)",
2205 ((jlong)si.freeswap * si.mem_unit) >> 10);
2206 st->cr();
2207 }
2208
2209 void os::pd_print_cpu_info(outputStream* st) {
2210 st->print("\n/proc/cpuinfo:\n");
2211 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2212 st->print(" <Not Available>");
2213 }
2214 st->cr();
2215 }
2216
2217 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2218 // but they're the same for all the linux arch that we support
2219 // and they're the same for solaris but there's no common place to put this.
2220 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2221 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2222 "ILL_COPROC", "ILL_BADSTK" };
2223
2224 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2225 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2226 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2227
2228 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2229
2230 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2231
2232 void os::print_siginfo(outputStream* st, void* siginfo) {
2233 st->print("siginfo:");
2234
2235 const int buflen = 100;
2236 char buf[buflen];
2237 siginfo_t *si = (siginfo_t*)siginfo;
2238 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2239 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2240 st->print("si_errno=%s", buf);
2241 } else {
2242 st->print("si_errno=%d", si->si_errno);
2243 }
2244 const int c = si->si_code;
2245 assert(c > 0, "unexpected si_code");
2246 switch (si->si_signo) {
2247 case SIGILL:
2248 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2249 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2250 break;
2251 case SIGFPE:
2252 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2253 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2254 break;
2255 case SIGSEGV:
2256 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2257 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2258 break;
2259 case SIGBUS:
2260 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2261 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2262 break;
2263 default:
2264 st->print(", si_code=%d", si->si_code);
2265 // no si_addr
2266 }
2267
2268 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2269 UseSharedSpaces) {
2270 FileMapInfo* mapinfo = FileMapInfo::current_info();
2271 if (mapinfo->is_in_shared_space(si->si_addr)) {
2272 st->print("\n\nError accessing class data sharing archive." \
2273 " Mapped file inaccessible during execution, " \
2274 " possible disk/network problem.");
2275 }
2276 }
2277 st->cr();
2278 }
2279
2280
2281 static void print_signal_handler(outputStream* st, int sig,
2282 char* buf, size_t buflen);
2283
2284 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2285 st->print_cr("Signal Handlers:");
2286 print_signal_handler(st, SIGSEGV, buf, buflen);
2287 print_signal_handler(st, SIGBUS , buf, buflen);
2288 print_signal_handler(st, SIGFPE , buf, buflen);
2289 print_signal_handler(st, SIGPIPE, buf, buflen);
2290 print_signal_handler(st, SIGXFSZ, buf, buflen);
2291 print_signal_handler(st, SIGILL , buf, buflen);
2292 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2293 print_signal_handler(st, SR_signum, buf, buflen);
2294 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2295 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2296 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2297 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2298 }
2299
2300 static char saved_jvm_path[MAXPATHLEN] = {0};
2301
2302 // Find the full path to the current module, libjvm.so
2303 void os::jvm_path(char *buf, jint buflen) {
2304 // Error checking.
2305 if (buflen < MAXPATHLEN) {
2306 assert(false, "must use a large-enough buffer");
2307 buf[0] = '\0';
2308 return;
2309 }
2310 // Lazy resolve the path to current module.
2311 if (saved_jvm_path[0] != 0) {
2312 strcpy(buf, saved_jvm_path);
2313 return;
2314 }
2315
2316 char dli_fname[MAXPATHLEN];
2317 bool ret = dll_address_to_library_name(
2318 CAST_FROM_FN_PTR(address, os::jvm_path),
2319 dli_fname, sizeof(dli_fname), NULL);
2320 assert(ret != 0, "cannot locate libjvm");
2321 char *rp = realpath(dli_fname, buf);
2322 if (rp == NULL)
2323 return;
2324
2325 if (Arguments::created_by_gamma_launcher()) {
2326 // Support for the gamma launcher. Typical value for buf is
2327 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2328 // the right place in the string, then assume we are installed in a JDK and
2329 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2330 // up the path so it looks like libjvm.so is installed there (append a
2331 // fake suffix hotspot/libjvm.so).
2332 const char *p = buf + strlen(buf) - 1;
2333 for (int count = 0; p > buf && count < 5; ++count) {
2334 for (--p; p > buf && *p != '/'; --p)
2335 /* empty */ ;
2336 }
2337
2338 if (strncmp(p, "/jre/lib/", 9) != 0) {
2339 // Look for JAVA_HOME in the environment.
2340 char* java_home_var = ::getenv("JAVA_HOME");
2341 if (java_home_var != NULL && java_home_var[0] != 0) {
2342 char* jrelib_p;
2343 int len;
2344
2345 // Check the current module name "libjvm.so".
2346 p = strrchr(buf, '/');
2347 assert(strstr(p, "/libjvm") == p, "invalid library name");
2348
2349 rp = realpath(java_home_var, buf);
2350 if (rp == NULL)
2351 return;
2352
2353 // determine if this is a legacy image or modules image
2354 // modules image doesn't have "jre" subdirectory
2355 len = strlen(buf);
2356 jrelib_p = buf + len;
2357 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2358 if (0 != access(buf, F_OK)) {
2359 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2360 }
2361
2362 if (0 == access(buf, F_OK)) {
2363 // Use current module name "libjvm.so"
2364 len = strlen(buf);
2365 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2366 } else {
2367 // Go back to path of .so
2368 rp = realpath(dli_fname, buf);
2369 if (rp == NULL)
2370 return;
2371 }
2372 }
2373 }
2374 }
2375
2376 strcpy(saved_jvm_path, buf);
2377 }
2378
2379 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2380 // no prefix required, not even "_"
2381 }
2382
2383 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2384 // no suffix required
2385 }
2386
2387 ////////////////////////////////////////////////////////////////////////////////
2388 // sun.misc.Signal support
2389
2390 static volatile jint sigint_count = 0;
2391
2392 static void
2393 UserHandler(int sig, void *siginfo, void *context) {
2394 // 4511530 - sem_post is serialized and handled by the manager thread. When
2395 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2396 // don't want to flood the manager thread with sem_post requests.
2397 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2398 return;
2399
2400 // Ctrl-C is pressed during error reporting, likely because the error
2401 // handler fails to abort. Let VM die immediately.
2402 if (sig == SIGINT && is_error_reported()) {
2403 os::die();
2404 }
2405
2406 os::signal_notify(sig);
2407 }
2408
2409 void* os::user_handler() {
2410 return CAST_FROM_FN_PTR(void*, UserHandler);
2411 }
2412
2413 class Semaphore : public StackObj {
2414 public:
2415 Semaphore();
2416 ~Semaphore();
2417 void signal();
2418 void wait();
2419 bool trywait();
2420 bool timedwait(unsigned int sec, int nsec);
2421 private:
2422 sem_t _semaphore;
2423 };
2424
2425
2426 Semaphore::Semaphore() {
2427 sem_init(&_semaphore, 0, 0);
2428 }
2429
2430 Semaphore::~Semaphore() {
2431 sem_destroy(&_semaphore);
2432 }
2433
2434 void Semaphore::signal() {
2435 sem_post(&_semaphore);
2436 }
2437
2438 void Semaphore::wait() {
2439 sem_wait(&_semaphore);
2440 }
2441
2442 bool Semaphore::trywait() {
2443 return sem_trywait(&_semaphore) == 0;
2444 }
2445
2446 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2447 struct timespec ts;
2448 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2449
2450 while (1) {
2451 int result = sem_timedwait(&_semaphore, &ts);
2452 if (result == 0) {
2453 return true;
2454 } else if (errno == EINTR) {
2455 continue;
2456 } else if (errno == ETIMEDOUT) {
2457 return false;
2458 } else {
2459 return false;
2460 }
2461 }
2462 }
2463
2464 extern "C" {
2465 typedef void (*sa_handler_t)(int);
2466 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2467 }
2468
2469 void* os::signal(int signal_number, void* handler) {
2470 struct sigaction sigAct, oldSigAct;
2471
2472 sigfillset(&(sigAct.sa_mask));
2473 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2474 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2475
2476 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2477 // -1 means registration failed
2478 return (void *)-1;
2479 }
2480
2481 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2482 }
2483
2484 void os::signal_raise(int signal_number) {
2485 ::raise(signal_number);
2486 }
2487
2488 /*
2489 * The following code is moved from os.cpp for making this
2490 * code platform specific, which it is by its very nature.
2491 */
2492
2493 // Will be modified when max signal is changed to be dynamic
2494 int os::sigexitnum_pd() {
2495 return NSIG;
2496 }
2497
2498 // a counter for each possible signal value
2499 static volatile jint pending_signals[NSIG+1] = { 0 };
2500
2501 // Linux(POSIX) specific hand shaking semaphore.
2502 static sem_t sig_sem;
2503 static Semaphore sr_semaphore;
2504
2505 void os::signal_init_pd() {
2506 // Initialize signal structures
2507 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2508
2509 // Initialize signal semaphore
2510 ::sem_init(&sig_sem, 0, 0);
2511 }
2512
2513 void os::signal_notify(int sig) {
2514 Atomic::inc(&pending_signals[sig]);
2515 ::sem_post(&sig_sem);
2516 }
2517
2518 static int check_pending_signals(bool wait) {
2519 Atomic::store(0, &sigint_count);
2520 for (;;) {
2521 for (int i = 0; i < NSIG + 1; i++) {
2522 jint n = pending_signals[i];
2523 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2524 return i;
2525 }
2526 }
2527 if (!wait) {
2528 return -1;
2529 }
2530 JavaThread *thread = JavaThread::current();
2531 ThreadBlockInVM tbivm(thread);
2532
2533 bool threadIsSuspended;
2534 do {
2535 thread->set_suspend_equivalent();
2536 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2537 ::sem_wait(&sig_sem);
2538
2539 // were we externally suspended while we were waiting?
2540 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2541 if (threadIsSuspended) {
2542 //
2543 // The semaphore has been incremented, but while we were waiting
2544 // another thread suspended us. We don't want to continue running
2545 // while suspended because that would surprise the thread that
2546 // suspended us.
2547 //
2548 ::sem_post(&sig_sem);
2549
2550 thread->java_suspend_self();
2551 }
2552 } while (threadIsSuspended);
2553 }
2554 }
2555
2556 int os::signal_lookup() {
2557 return check_pending_signals(false);
2558 }
2559
2560 int os::signal_wait() {
2561 return check_pending_signals(true);
2562 }
2563
2564 ////////////////////////////////////////////////////////////////////////////////
2565 // Virtual Memory
2566
2567 int os::vm_page_size() {
2568 // Seems redundant as all get out
2569 assert(os::Linux::page_size() != -1, "must call os::init");
2570 return os::Linux::page_size();
2571 }
2572
2573 // Solaris allocates memory by pages.
2574 int os::vm_allocation_granularity() {
2575 assert(os::Linux::page_size() != -1, "must call os::init");
2576 return os::Linux::page_size();
2577 }
2578
2579 // Rationale behind this function:
2580 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2581 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2582 // samples for JITted code. Here we create private executable mapping over the code cache
2583 // and then we can use standard (well, almost, as mapping can change) way to provide
2584 // info for the reporting script by storing timestamp and location of symbol
2585 void linux_wrap_code(char* base, size_t size) {
2586 static volatile jint cnt = 0;
2587
2588 if (!UseOprofile) {
2589 return;
2590 }
2591
2592 char buf[PATH_MAX+1];
2593 int num = Atomic::add(1, &cnt);
2594
2595 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2596 os::get_temp_directory(), os::current_process_id(), num);
2597 unlink(buf);
2598
2599 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2600
2601 if (fd != -1) {
2602 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2603 if (rv != (off_t)-1) {
2604 if (::write(fd, "", 1) == 1) {
2605 mmap(base, size,
2606 PROT_READ|PROT_WRITE|PROT_EXEC,
2607 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2608 }
2609 }
2610 ::close(fd);
2611 unlink(buf);
2612 }
2613 }
2614
2615 static bool recoverable_mmap_error(int err) {
2616 // See if the error is one we can let the caller handle. This
2617 // list of errno values comes from JBS-6843484. I can't find a
2618 // Linux man page that documents this specific set of errno
2619 // values so while this list currently matches Solaris, it may
2620 // change as we gain experience with this failure mode.
2621 switch (err) {
2622 case EBADF:
2623 case EINVAL:
2624 case ENOTSUP:
2625 // let the caller deal with these errors
2626 return true;
2627
2628 default:
2629 // Any remaining errors on this OS can cause our reserved mapping
2630 // to be lost. That can cause confusion where different data
2631 // structures think they have the same memory mapped. The worst
2632 // scenario is if both the VM and a library think they have the
2633 // same memory mapped.
2634 return false;
2635 }
2636 }
2637
2638 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2639 int err) {
2640 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2641 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2642 strerror(err), err);
2643 }
2644
2645 static void warn_fail_commit_memory(char* addr, size_t size,
2646 size_t alignment_hint, bool exec,
2647 int err) {
2648 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2649 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2650 alignment_hint, exec, strerror(err), err);
2651 }
2652
2653 // NOTE: Linux kernel does not really reserve the pages for us.
2654 // All it does is to check if there are enough free pages
2655 // left at the time of mmap(). This could be a potential
2656 // problem.
2657 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2658 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2659 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2660 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2661 if (res != (uintptr_t) MAP_FAILED) {
2662 if (UseNUMAInterleaving) {
2663 numa_make_global(addr, size);
2664 }
2665 return 0;
2666 }
2667
2668 int err = errno; // save errno from mmap() call above
2669
2670 if (!recoverable_mmap_error(err)) {
2671 warn_fail_commit_memory(addr, size, exec, err);
2672 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2673 }
2674
2675 return err;
2676 }
2677
2678 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2679 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2680 }
2681
2682 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2683 const char* mesg) {
2684 assert(mesg != NULL, "mesg must be specified");
2685 int err = os::Linux::commit_memory_impl(addr, size, exec);
2686 if (err != 0) {
2687 // the caller wants all commit errors to exit with the specified mesg:
2688 warn_fail_commit_memory(addr, size, exec, err);
2689 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2690 }
2691 }
2692
2693 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2694 #ifndef MAP_HUGETLB
2695 #define MAP_HUGETLB 0x40000
2696 #endif
2697
2698 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2699 #ifndef MADV_HUGEPAGE
2700 #define MADV_HUGEPAGE 14
2701 #endif
2702
2703 int os::Linux::commit_memory_impl(char* addr, size_t size,
2704 size_t alignment_hint, bool exec) {
2705 int err;
2706 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2707 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2708 uintptr_t res =
2709 (uintptr_t) ::mmap(addr, size, prot,
2710 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2711 -1, 0);
2712 if (res != (uintptr_t) MAP_FAILED) {
2713 if (UseNUMAInterleaving) {
2714 numa_make_global(addr, size);
2715 }
2716 return 0;
2717 }
2718
2719 err = errno; // save errno from mmap() call above
2720
2721 if (!recoverable_mmap_error(err)) {
2722 // However, it is not clear that this loss of our reserved mapping
2723 // happens with large pages on Linux or that we cannot recover
2724 // from the loss. For now, we just issue a warning and we don't
2725 // call vm_exit_out_of_memory(). This issue is being tracked by
2726 // JBS-8007074.
2727 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2728 // vm_exit_out_of_memory(size, OOM_MMAP_ERROR,
2729 // "committing reserved memory.");
2730 }
2731 // Fall through and try to use small pages
2732 }
2733
2734 err = os::Linux::commit_memory_impl(addr, size, exec);
2735 if (err == 0) {
2736 realign_memory(addr, size, alignment_hint);
2737 }
2738 return err;
2739 }
2740
2741 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2742 bool exec) {
2743 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2744 }
2745
2746 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2747 size_t alignment_hint, bool exec,
2748 const char* mesg) {
2749 assert(mesg != NULL, "mesg must be specified");
2750 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2751 if (err != 0) {
2752 // the caller wants all commit errors to exit with the specified mesg:
2753 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2754 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2755 }
2756 }
2757
2758 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2759 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2760 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2761 // be supported or the memory may already be backed by huge pages.
2762 ::madvise(addr, bytes, MADV_HUGEPAGE);
2763 }
2764 }
2765
2766 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2767 // This method works by doing an mmap over an existing mmaping and effectively discarding
2768 // the existing pages. However it won't work for SHM-based large pages that cannot be
2769 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2770 // small pages on top of the SHM segment. This method always works for small pages, so we
2771 // allow that in any case.
2772 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2773 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2774 }
2775 }
2776
2777 void os::numa_make_global(char *addr, size_t bytes) {
2778 Linux::numa_interleave_memory(addr, bytes);
2779 }
2780
2781 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2782 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2783 }
2784
2785 bool os::numa_topology_changed() { return false; }
2786
2787 size_t os::numa_get_groups_num() {
2788 int max_node = Linux::numa_max_node();
2789 return max_node > 0 ? max_node + 1 : 1;
2790 }
2791
2792 int os::numa_get_group_id() {
2793 int cpu_id = Linux::sched_getcpu();
2794 if (cpu_id != -1) {
2795 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2796 if (lgrp_id != -1) {
2797 return lgrp_id;
2798 }
2799 }
2800 return 0;
2801 }
2802
2803 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2804 for (size_t i = 0; i < size; i++) {
2805 ids[i] = i;
2806 }
2807 return size;
2808 }
2809
2810 bool os::get_page_info(char *start, page_info* info) {
2811 return false;
2812 }
2813
2814 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2815 return end;
2816 }
2817
2818
2819 int os::Linux::sched_getcpu_syscall(void) {
2820 unsigned int cpu;
2821 int retval = -1;
2822
2823 #if defined(IA32)
2824 # ifndef SYS_getcpu
2825 # define SYS_getcpu 318
2826 # endif
2827 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2828 #elif defined(AMD64)
2829 // Unfortunately we have to bring all these macros here from vsyscall.h
2830 // to be able to compile on old linuxes.
2831 # define __NR_vgetcpu 2
2832 # define VSYSCALL_START (-10UL << 20)
2833 # define VSYSCALL_SIZE 1024
2834 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2835 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2836 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2837 retval = vgetcpu(&cpu, NULL, NULL);
2838 #endif
2839
2840 return (retval == -1) ? retval : cpu;
2841 }
2842
2843 // Something to do with the numa-aware allocator needs these symbols
2844 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2845 extern "C" JNIEXPORT void numa_error(char *where) { }
2846 extern "C" JNIEXPORT int fork1() { return fork(); }
2847
2848
2849 // If we are running with libnuma version > 2, then we should
2850 // be trying to use symbols with versions 1.1
2851 // If we are running with earlier version, which did not have symbol versions,
2852 // we should use the base version.
2853 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2854 void *f = dlvsym(handle, name, "libnuma_1.1");
2855 if (f == NULL) {
2856 f = dlsym(handle, name);
2857 }
2858 return f;
2859 }
2860
2861 bool os::Linux::libnuma_init() {
2862 // sched_getcpu() should be in libc.
2863 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2864 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2865
2866 // If it's not, try a direct syscall.
2867 if (sched_getcpu() == -1)
2868 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2869
2870 if (sched_getcpu() != -1) { // Does it work?
2871 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2872 if (handle != NULL) {
2873 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2874 libnuma_dlsym(handle, "numa_node_to_cpus")));
2875 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2876 libnuma_dlsym(handle, "numa_max_node")));
2877 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2878 libnuma_dlsym(handle, "numa_available")));
2879 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2880 libnuma_dlsym(handle, "numa_tonode_memory")));
2881 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2882 libnuma_dlsym(handle, "numa_interleave_memory")));
2883
2884
2885 if (numa_available() != -1) {
2886 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2887 // Create a cpu -> node mapping
2888 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2889 rebuild_cpu_to_node_map();
2890 return true;
2891 }
2892 }
2893 }
2894 return false;
2895 }
2896
2897 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2898 // The table is later used in get_node_by_cpu().
2899 void os::Linux::rebuild_cpu_to_node_map() {
2900 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2901 // in libnuma (possible values are starting from 16,
2902 // and continuing up with every other power of 2, but less
2903 // than the maximum number of CPUs supported by kernel), and
2904 // is a subject to change (in libnuma version 2 the requirements
2905 // are more reasonable) we'll just hardcode the number they use
2906 // in the library.
2907 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2908
2909 size_t cpu_num = os::active_processor_count();
2910 size_t cpu_map_size = NCPUS / BitsPerCLong;
2911 size_t cpu_map_valid_size =
2912 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2913
2914 cpu_to_node()->clear();
2915 cpu_to_node()->at_grow(cpu_num - 1);
2916 size_t node_num = numa_get_groups_num();
2917
2918 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2919 for (size_t i = 0; i < node_num; i++) {
2920 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2921 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2922 if (cpu_map[j] != 0) {
2923 for (size_t k = 0; k < BitsPerCLong; k++) {
2924 if (cpu_map[j] & (1UL << k)) {
2925 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2926 }
2927 }
2928 }
2929 }
2930 }
2931 }
2932 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2933 }
2934
2935 int os::Linux::get_node_by_cpu(int cpu_id) {
2936 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2937 return cpu_to_node()->at(cpu_id);
2938 }
2939 return -1;
2940 }
2941
2942 GrowableArray<int>* os::Linux::_cpu_to_node;
2943 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2944 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2945 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2946 os::Linux::numa_available_func_t os::Linux::_numa_available;
2947 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2948 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2949 unsigned long* os::Linux::_numa_all_nodes;
2950
2951 bool os::pd_uncommit_memory(char* addr, size_t size) {
2952 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2953 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2954 return res != (uintptr_t) MAP_FAILED;
2955 }
2956
2957 // Linux uses a growable mapping for the stack, and if the mapping for
2958 // the stack guard pages is not removed when we detach a thread the
2959 // stack cannot grow beyond the pages where the stack guard was
2960 // mapped. If at some point later in the process the stack expands to
2961 // that point, the Linux kernel cannot expand the stack any further
2962 // because the guard pages are in the way, and a segfault occurs.
2963 //
2964 // However, it's essential not to split the stack region by unmapping
2965 // a region (leaving a hole) that's already part of the stack mapping,
2966 // so if the stack mapping has already grown beyond the guard pages at
2967 // the time we create them, we have to truncate the stack mapping.
2968 // So, we need to know the extent of the stack mapping when
2969 // create_stack_guard_pages() is called.
2970
2971 // Find the bounds of the stack mapping. Return true for success.
2972 //
2973 // We only need this for stacks that are growable: at the time of
2974 // writing thread stacks don't use growable mappings (i.e. those
2975 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2976 // only applies to the main thread.
2977
2978 static
2979 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2980
2981 char buf[128];
2982 int fd, sz;
2983
2984 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2985 return false;
2986 }
2987
2988 const char kw[] = "[stack]";
2989 const int kwlen = sizeof(kw)-1;
2990
2991 // Address part of /proc/self/maps couldn't be more than 128 bytes
2992 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2993 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2994 // Extract addresses
2995 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2996 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2997 if (sp >= *bottom && sp <= *top) {
2998 ::close(fd);
2999 return true;
3000 }
3001 }
3002 }
3003 }
3004
3005 ::close(fd);
3006 return false;
3007 }
3008
3009
3010 // If the (growable) stack mapping already extends beyond the point
3011 // where we're going to put our guard pages, truncate the mapping at
3012 // that point by munmap()ping it. This ensures that when we later
3013 // munmap() the guard pages we don't leave a hole in the stack
3014 // mapping. This only affects the main/initial thread, but guard
3015 // against future OS changes
3016 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3017 uintptr_t stack_extent, stack_base;
3018 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
3019 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
3020 assert(os::Linux::is_initial_thread(),
3021 "growable stack in non-initial thread");
3022 if (stack_extent < (uintptr_t)addr)
3023 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
3024 }
3025
3026 return os::commit_memory(addr, size, !ExecMem);
3027 }
3028
3029 // If this is a growable mapping, remove the guard pages entirely by
3030 // munmap()ping them. If not, just call uncommit_memory(). This only
3031 // affects the main/initial thread, but guard against future OS changes
3032 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3033 uintptr_t stack_extent, stack_base;
3034 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
3035 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
3036 assert(os::Linux::is_initial_thread(),
3037 "growable stack in non-initial thread");
3038
3039 return ::munmap(addr, size) == 0;
3040 }
3041
3042 return os::uncommit_memory(addr, size);
3043 }
3044
3045 static address _highest_vm_reserved_address = NULL;
3046
3047 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3048 // at 'requested_addr'. If there are existing memory mappings at the same
3049 // location, however, they will be overwritten. If 'fixed' is false,
3050 // 'requested_addr' is only treated as a hint, the return value may or
3051 // may not start from the requested address. Unlike Linux mmap(), this
3052 // function returns NULL to indicate failure.
3053 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3054 char * addr;
3055 int flags;
3056
3057 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3058 if (fixed) {
3059 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3060 flags |= MAP_FIXED;
3061 }
3062
3063 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3064 // touch an uncommitted page. Otherwise, the read/write might
3065 // succeed if we have enough swap space to back the physical page.
3066 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3067 flags, -1, 0);
3068
3069 if (addr != MAP_FAILED) {
3070 // anon_mmap() should only get called during VM initialization,
3071 // don't need lock (actually we can skip locking even it can be called
3072 // from multiple threads, because _highest_vm_reserved_address is just a
3073 // hint about the upper limit of non-stack memory regions.)
3074 if ((address)addr + bytes > _highest_vm_reserved_address) {
3075 _highest_vm_reserved_address = (address)addr + bytes;
3076 }
3077 }
3078
3079 return addr == MAP_FAILED ? NULL : addr;
3080 }
3081
3082 // Don't update _highest_vm_reserved_address, because there might be memory
3083 // regions above addr + size. If so, releasing a memory region only creates
3084 // a hole in the address space, it doesn't help prevent heap-stack collision.
3085 //
3086 static int anon_munmap(char * addr, size_t size) {
3087 return ::munmap(addr, size) == 0;
3088 }
3089
3090 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3091 size_t alignment_hint) {
3092 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3093 }
3094
3095 bool os::pd_release_memory(char* addr, size_t size) {
3096 return anon_munmap(addr, size);
3097 }
3098
3099 static address highest_vm_reserved_address() {
3100 return _highest_vm_reserved_address;
3101 }
3102
3103 static bool linux_mprotect(char* addr, size_t size, int prot) {
3104 // Linux wants the mprotect address argument to be page aligned.
3105 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3106
3107 // According to SUSv3, mprotect() should only be used with mappings
3108 // established by mmap(), and mmap() always maps whole pages. Unaligned
3109 // 'addr' likely indicates problem in the VM (e.g. trying to change
3110 // protection of malloc'ed or statically allocated memory). Check the
3111 // caller if you hit this assert.
3112 assert(addr == bottom, "sanity check");
3113
3114 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3115 return ::mprotect(bottom, size, prot) == 0;
3116 }
3117
3118 // Set protections specified
3119 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3120 bool is_committed) {
3121 unsigned int p = 0;
3122 switch (prot) {
3123 case MEM_PROT_NONE: p = PROT_NONE; break;
3124 case MEM_PROT_READ: p = PROT_READ; break;
3125 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3126 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3127 default:
3128 ShouldNotReachHere();
3129 }
3130 // is_committed is unused.
3131 return linux_mprotect(addr, bytes, p);
3132 }
3133
3134 bool os::guard_memory(char* addr, size_t size) {
3135 return linux_mprotect(addr, size, PROT_NONE);
3136 }
3137
3138 bool os::unguard_memory(char* addr, size_t size) {
3139 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3140 }
3141
3142 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3143 bool result = false;
3144 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
3145 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3146 -1, 0);
3147
3148 if (p != MAP_FAILED) {
3149 // We don't know if this really is a huge page or not.
3150 FILE *fp = fopen("/proc/self/maps", "r");
3151 if (fp) {
3152 while (!feof(fp)) {
3153 char chars[257];
3154 long x = 0;
3155 if (fgets(chars, sizeof(chars), fp)) {
3156 if (sscanf(chars, "%lx-%*x", &x) == 1
3157 && x == (long)p) {
3158 if (strstr (chars, "hugepage")) {
3159 result = true;
3160 break;
3161 }
3162 }
3163 }
3164 }
3165 fclose(fp);
3166 }
3167 munmap (p, page_size);
3168 if (result)
3169 return true;
3170 }
3171
3172 if (warn) {
3173 warning("HugeTLBFS is not supported by the operating system.");
3174 }
3175
3176 return result;
3177 }
3178
3179 /*
3180 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3181 *
3182 * From the coredump_filter documentation:
3183 *
3184 * - (bit 0) anonymous private memory
3185 * - (bit 1) anonymous shared memory
3186 * - (bit 2) file-backed private memory
3187 * - (bit 3) file-backed shared memory
3188 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3189 * effective only if the bit 2 is cleared)
3190 * - (bit 5) hugetlb private memory
3191 * - (bit 6) hugetlb shared memory
3192 */
3193 static void set_coredump_filter(void) {
3194 FILE *f;
3195 long cdm;
3196
3197 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3198 return;
3199 }
3200
3201 if (fscanf(f, "%lx", &cdm) != 1) {
3202 fclose(f);
3203 return;
3204 }
3205
3206 rewind(f);
3207
3208 if ((cdm & LARGEPAGES_BIT) == 0) {
3209 cdm |= LARGEPAGES_BIT;
3210 fprintf(f, "%#lx", cdm);
3211 }
3212
3213 fclose(f);
3214 }
3215
3216 // Large page support
3217
3218 static size_t _large_page_size = 0;
3219
3220 void os::large_page_init() {
3221 if (!UseLargePages) {
3222 UseHugeTLBFS = false;
3223 UseSHM = false;
3224 return;
3225 }
3226
3227 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3228 // If UseLargePages is specified on the command line try both methods,
3229 // if it's default, then try only HugeTLBFS.
3230 if (FLAG_IS_DEFAULT(UseLargePages)) {
3231 UseHugeTLBFS = true;
3232 } else {
3233 UseHugeTLBFS = UseSHM = true;
3234 }
3235 }
3236
3237 if (LargePageSizeInBytes) {
3238 _large_page_size = LargePageSizeInBytes;
3239 } else {
3240 // large_page_size on Linux is used to round up heap size. x86 uses either
3241 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3242 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3243 // page as large as 256M.
3244 //
3245 // Here we try to figure out page size by parsing /proc/meminfo and looking
3246 // for a line with the following format:
3247 // Hugepagesize: 2048 kB
3248 //
3249 // If we can't determine the value (e.g. /proc is not mounted, or the text
3250 // format has been changed), we'll use the largest page size supported by
3251 // the processor.
3252
3253 #ifndef ZERO
3254 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3255 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3256 #endif // ZERO
3257
3258 FILE *fp = fopen("/proc/meminfo", "r");
3259 if (fp) {
3260 while (!feof(fp)) {
3261 int x = 0;
3262 char buf[16];
3263 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3264 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3265 _large_page_size = x * K;
3266 break;
3267 }
3268 } else {
3269 // skip to next line
3270 for (;;) {
3271 int ch = fgetc(fp);
3272 if (ch == EOF || ch == (int)'\n') break;
3273 }
3274 }
3275 }
3276 fclose(fp);
3277 }
3278 }
3279
3280 // print a warning if any large page related flag is specified on command line
3281 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3282
3283 const size_t default_page_size = (size_t)Linux::page_size();
3284 if (_large_page_size > default_page_size) {
3285 _page_sizes[0] = _large_page_size;
3286 _page_sizes[1] = default_page_size;
3287 _page_sizes[2] = 0;
3288 }
3289 UseHugeTLBFS = UseHugeTLBFS &&
3290 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3291
3292 if (UseHugeTLBFS)
3293 UseSHM = false;
3294
3295 UseLargePages = UseHugeTLBFS || UseSHM;
3296
3297 set_coredump_filter();
3298 }
3299
3300 #ifndef SHM_HUGETLB
3301 #define SHM_HUGETLB 04000
3302 #endif
3303
3304 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3305 // "exec" is passed in but not used. Creating the shared image for
3306 // the code cache doesn't have an SHM_X executable permission to check.
3307 assert(UseLargePages && UseSHM, "only for SHM large pages");
3308
3309 key_t key = IPC_PRIVATE;
3310 char *addr;
3311
3312 bool warn_on_failure = UseLargePages &&
3313 (!FLAG_IS_DEFAULT(UseLargePages) ||
3314 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3315 );
3316 char msg[128];
3317
3318 // Create a large shared memory region to attach to based on size.
3319 // Currently, size is the total size of the heap
3320 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3321 if (shmid == -1) {
3322 // Possible reasons for shmget failure:
3323 // 1. shmmax is too small for Java heap.
3324 // > check shmmax value: cat /proc/sys/kernel/shmmax
3325 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3326 // 2. not enough large page memory.
3327 // > check available large pages: cat /proc/meminfo
3328 // > increase amount of large pages:
3329 // echo new_value > /proc/sys/vm/nr_hugepages
3330 // Note 1: different Linux may use different name for this property,
3331 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3332 // Note 2: it's possible there's enough physical memory available but
3333 // they are so fragmented after a long run that they can't
3334 // coalesce into large pages. Try to reserve large pages when
3335 // the system is still "fresh".
3336 if (warn_on_failure) {
3337 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3338 warning(msg);
3339 }
3340 return NULL;
3341 }
3342
3343 // attach to the region
3344 addr = (char*)shmat(shmid, req_addr, 0);
3345 int err = errno;
3346
3347 // Remove shmid. If shmat() is successful, the actual shared memory segment
3348 // will be deleted when it's detached by shmdt() or when the process
3349 // terminates. If shmat() is not successful this will remove the shared
3350 // segment immediately.
3351 shmctl(shmid, IPC_RMID, NULL);
3352
3353 if ((intptr_t)addr == -1) {
3354 if (warn_on_failure) {
3355 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3356 warning(msg);
3357 }
3358 return NULL;
3359 }
3360
3361 if ((addr != NULL) && UseNUMAInterleaving) {
3362 numa_make_global(addr, bytes);
3363 }
3364
3365 // The memory is committed
3366 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3367
3368 return addr;
3369 }
3370
3371 bool os::release_memory_special(char* base, size_t bytes) {
3372 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3373 // detaching the SHM segment will also delete it, see reserve_memory_special()
3374 int rslt = shmdt(base);
3375 if (rslt == 0) {
3376 tkr.record((address)base, bytes);
3377 return true;
3378 } else {
3379 tkr.discard();
3380 return false;
3381 }
3382 }
3383
3384 size_t os::large_page_size() {
3385 return _large_page_size;
3386 }
3387
3388 // HugeTLBFS allows application to commit large page memory on demand;
3389 // with SysV SHM the entire memory region must be allocated as shared
3390 // memory.
3391 bool os::can_commit_large_page_memory() {
3392 return UseHugeTLBFS;
3393 }
3394
3395 bool os::can_execute_large_page_memory() {
3396 return UseHugeTLBFS;
3397 }
3398
3399 // Reserve memory at an arbitrary address, only if that area is
3400 // available (and not reserved for something else).
3401
3402 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3403 const int max_tries = 10;
3404 char* base[max_tries];
3405 size_t size[max_tries];
3406 const size_t gap = 0x000000;
3407
3408 // Assert only that the size is a multiple of the page size, since
3409 // that's all that mmap requires, and since that's all we really know
3410 // about at this low abstraction level. If we need higher alignment,
3411 // we can either pass an alignment to this method or verify alignment
3412 // in one of the methods further up the call chain. See bug 5044738.
3413 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3414
3415 // Repeatedly allocate blocks until the block is allocated at the
3416 // right spot. Give up after max_tries. Note that reserve_memory() will
3417 // automatically update _highest_vm_reserved_address if the call is
3418 // successful. The variable tracks the highest memory address every reserved
3419 // by JVM. It is used to detect heap-stack collision if running with
3420 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3421 // space than needed, it could confuse the collision detecting code. To
3422 // solve the problem, save current _highest_vm_reserved_address and
3423 // calculate the correct value before return.
3424 address old_highest = _highest_vm_reserved_address;
3425
3426 // Linux mmap allows caller to pass an address as hint; give it a try first,
3427 // if kernel honors the hint then we can return immediately.
3428 char * addr = anon_mmap(requested_addr, bytes, false);
3429 if (addr == requested_addr) {
3430 return requested_addr;
3431 }
3432
3433 if (addr != NULL) {
3434 // mmap() is successful but it fails to reserve at the requested address
3435 anon_munmap(addr, bytes);
3436 }
3437
3438 int i;
3439 for (i = 0; i < max_tries; ++i) {
3440 base[i] = reserve_memory(bytes);
3441
3442 if (base[i] != NULL) {
3443 // Is this the block we wanted?
3444 if (base[i] == requested_addr) {
3445 size[i] = bytes;
3446 break;
3447 }
3448
3449 // Does this overlap the block we wanted? Give back the overlapped
3450 // parts and try again.
3451
3452 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3453 if (top_overlap >= 0 && top_overlap < bytes) {
3454 unmap_memory(base[i], top_overlap);
3455 base[i] += top_overlap;
3456 size[i] = bytes - top_overlap;
3457 } else {
3458 size_t bottom_overlap = base[i] + bytes - requested_addr;
3459 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3460 unmap_memory(requested_addr, bottom_overlap);
3461 size[i] = bytes - bottom_overlap;
3462 } else {
3463 size[i] = bytes;
3464 }
3465 }
3466 }
3467 }
3468
3469 // Give back the unused reserved pieces.
3470
3471 for (int j = 0; j < i; ++j) {
3472 if (base[j] != NULL) {
3473 unmap_memory(base[j], size[j]);
3474 }
3475 }
3476
3477 if (i < max_tries) {
3478 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3479 return requested_addr;
3480 } else {
3481 _highest_vm_reserved_address = old_highest;
3482 return NULL;
3483 }
3484 }
3485
3486 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3487 return ::read(fd, buf, nBytes);
3488 }
3489
3490 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3491 // Solaris uses poll(), linux uses park().
3492 // Poll() is likely a better choice, assuming that Thread.interrupt()
3493 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3494 // SIGSEGV, see 4355769.
3495
3496 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3497 assert(thread == Thread::current(), "thread consistency check");
3498
3499 ParkEvent * const slp = thread->_SleepEvent ;
3500 slp->reset() ;
3501 OrderAccess::fence() ;
3502
3503 if (interruptible) {
3504 jlong prevtime = javaTimeNanos();
3505
3506 for (;;) {
3507 if (os::is_interrupted(thread, true)) {
3508 return OS_INTRPT;
3509 }
3510
3511 jlong newtime = javaTimeNanos();
3512
3513 if (newtime - prevtime < 0) {
3514 // time moving backwards, should only happen if no monotonic clock
3515 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3516 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3517 } else {
3518 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3519 }
3520
3521 if(millis <= 0) {
3522 return OS_OK;
3523 }
3524
3525 prevtime = newtime;
3526
3527 {
3528 assert(thread->is_Java_thread(), "sanity check");
3529 JavaThread *jt = (JavaThread *) thread;
3530 ThreadBlockInVM tbivm(jt);
3531 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3532
3533 jt->set_suspend_equivalent();
3534 // cleared by handle_special_suspend_equivalent_condition() or
3535 // java_suspend_self() via check_and_wait_while_suspended()
3536
3537 slp->park(millis);
3538
3539 // were we externally suspended while we were waiting?
3540 jt->check_and_wait_while_suspended();
3541 }
3542 }
3543 } else {
3544 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3545 jlong prevtime = javaTimeNanos();
3546
3547 for (;;) {
3548 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3549 // the 1st iteration ...
3550 jlong newtime = javaTimeNanos();
3551
3552 if (newtime - prevtime < 0) {
3553 // time moving backwards, should only happen if no monotonic clock
3554 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3555 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3556 } else {
3557 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3558 }
3559
3560 if(millis <= 0) break ;
3561
3562 prevtime = newtime;
3563 slp->park(millis);
3564 }
3565 return OS_OK ;
3566 }
3567 }
3568
3569 int os::naked_sleep() {
3570 // %% make the sleep time an integer flag. for now use 1 millisec.
3571 return os::sleep(Thread::current(), 1, false);
3572 }
3573
3574 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3575 void os::infinite_sleep() {
3576 while (true) { // sleep forever ...
3577 ::sleep(100); // ... 100 seconds at a time
3578 }
3579 }
3580
3581 // Used to convert frequent JVM_Yield() to nops
3582 bool os::dont_yield() {
3583 return DontYieldALot;
3584 }
3585
3586 void os::yield() {
3587 sched_yield();
3588 }
3589
3590 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3591
3592 void os::yield_all(int attempts) {
3593 // Yields to all threads, including threads with lower priorities
3594 // Threads on Linux are all with same priority. The Solaris style
3595 // os::yield_all() with nanosleep(1ms) is not necessary.
3596 sched_yield();
3597 }
3598
3599 // Called from the tight loops to possibly influence time-sharing heuristics
3600 void os::loop_breaker(int attempts) {
3601 os::yield_all(attempts);
3602 }
3603
3604 ////////////////////////////////////////////////////////////////////////////////
3605 // thread priority support
3606
3607 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3608 // only supports dynamic priority, static priority must be zero. For real-time
3609 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3610 // However, for large multi-threaded applications, SCHED_RR is not only slower
3611 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3612 // of 5 runs - Sep 2005).
3613 //
3614 // The following code actually changes the niceness of kernel-thread/LWP. It
3615 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3616 // not the entire user process, and user level threads are 1:1 mapped to kernel
3617 // threads. It has always been the case, but could change in the future. For
3618 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3619 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3620
3621 int os::java_to_os_priority[CriticalPriority + 1] = {
3622 19, // 0 Entry should never be used
3623
3624 4, // 1 MinPriority
3625 3, // 2
3626 2, // 3
3627
3628 1, // 4
3629 0, // 5 NormPriority
3630 -1, // 6
3631
3632 -2, // 7
3633 -3, // 8
3634 -4, // 9 NearMaxPriority
3635
3636 -5, // 10 MaxPriority
3637
3638 -5 // 11 CriticalPriority
3639 };
3640
3641 static int prio_init() {
3642 if (ThreadPriorityPolicy == 1) {
3643 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3644 // if effective uid is not root. Perhaps, a more elegant way of doing
3645 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3646 if (geteuid() != 0) {
3647 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3648 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3649 }
3650 ThreadPriorityPolicy = 0;
3651 }
3652 }
3653 if (UseCriticalJavaThreadPriority) {
3654 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3655 }
3656 return 0;
3657 }
3658
3659 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3660 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3661
3662 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3663 return (ret == 0) ? OS_OK : OS_ERR;
3664 }
3665
3666 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3667 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3668 *priority_ptr = java_to_os_priority[NormPriority];
3669 return OS_OK;
3670 }
3671
3672 errno = 0;
3673 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3674 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3675 }
3676
3677 // Hint to the underlying OS that a task switch would not be good.
3678 // Void return because it's a hint and can fail.
3679 void os::hint_no_preempt() {}
3680
3681 ////////////////////////////////////////////////////////////////////////////////
3682 // suspend/resume support
3683
3684 // the low-level signal-based suspend/resume support is a remnant from the
3685 // old VM-suspension that used to be for java-suspension, safepoints etc,
3686 // within hotspot. Now there is a single use-case for this:
3687 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3688 // that runs in the watcher thread.
3689 // The remaining code is greatly simplified from the more general suspension
3690 // code that used to be used.
3691 //
3692 // The protocol is quite simple:
3693 // - suspend:
3694 // - sends a signal to the target thread
3695 // - polls the suspend state of the osthread using a yield loop
3696 // - target thread signal handler (SR_handler) sets suspend state
3697 // and blocks in sigsuspend until continued
3698 // - resume:
3699 // - sets target osthread state to continue
3700 // - sends signal to end the sigsuspend loop in the SR_handler
3701 //
3702 // Note that the SR_lock plays no role in this suspend/resume protocol.
3703 //
3704
3705 static void resume_clear_context(OSThread *osthread) {
3706 osthread->set_ucontext(NULL);
3707 osthread->set_siginfo(NULL);
3708 }
3709
3710 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3711 osthread->set_ucontext(context);
3712 osthread->set_siginfo(siginfo);
3713 }
3714
3715 //
3716 // Handler function invoked when a thread's execution is suspended or
3717 // resumed. We have to be careful that only async-safe functions are
3718 // called here (Note: most pthread functions are not async safe and
3719 // should be avoided.)
3720 //
3721 // Note: sigwait() is a more natural fit than sigsuspend() from an
3722 // interface point of view, but sigwait() prevents the signal hander
3723 // from being run. libpthread would get very confused by not having
3724 // its signal handlers run and prevents sigwait()'s use with the
3725 // mutex granting granting signal.
3726 //
3727 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3728 //
3729 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3730 // Save and restore errno to avoid confusing native code with EINTR
3731 // after sigsuspend.
3732 int old_errno = errno;
3733
3734 Thread* thread = Thread::current();
3735 OSThread* osthread = thread->osthread();
3736 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3737
3738 os::SuspendResume::State current = osthread->sr.state();
3739 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3740 suspend_save_context(osthread, siginfo, context);
3741
3742 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3743 os::SuspendResume::State state = osthread->sr.suspended();
3744 if (state == os::SuspendResume::SR_SUSPENDED) {
3745 sigset_t suspend_set; // signals for sigsuspend()
3746
3747 // get current set of blocked signals and unblock resume signal
3748 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3749 sigdelset(&suspend_set, SR_signum);
3750
3751 sr_semaphore.signal();
3752 // wait here until we are resumed
3753 while (1) {
3754 sigsuspend(&suspend_set);
3755
3756 os::SuspendResume::State result = osthread->sr.running();
3757 if (result == os::SuspendResume::SR_RUNNING) {
3758 sr_semaphore.signal();
3759 break;
3760 }
3761 }
3762
3763 } else if (state == os::SuspendResume::SR_RUNNING) {
3764 // request was cancelled, continue
3765 } else {
3766 ShouldNotReachHere();
3767 }
3768
3769 resume_clear_context(osthread);
3770 } else if (current == os::SuspendResume::SR_RUNNING) {
3771 // request was cancelled, continue
3772 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3773 // ignore
3774 } else {
3775 // ignore
3776 }
3777
3778 errno = old_errno;
3779 }
3780
3781
3782 static int SR_initialize() {
3783 struct sigaction act;
3784 char *s;
3785 /* Get signal number to use for suspend/resume */
3786 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3787 int sig = ::strtol(s, 0, 10);
3788 if (sig > 0 || sig < _NSIG) {
3789 SR_signum = sig;
3790 }
3791 }
3792
3793 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3794 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3795
3796 sigemptyset(&SR_sigset);
3797 sigaddset(&SR_sigset, SR_signum);
3798
3799 /* Set up signal handler for suspend/resume */
3800 act.sa_flags = SA_RESTART|SA_SIGINFO;
3801 act.sa_handler = (void (*)(int)) SR_handler;
3802
3803 // SR_signum is blocked by default.
3804 // 4528190 - We also need to block pthread restart signal (32 on all
3805 // supported Linux platforms). Note that LinuxThreads need to block
3806 // this signal for all threads to work properly. So we don't have
3807 // to use hard-coded signal number when setting up the mask.
3808 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3809
3810 if (sigaction(SR_signum, &act, 0) == -1) {
3811 return -1;
3812 }
3813
3814 // Save signal flag
3815 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3816 return 0;
3817 }
3818
3819 static int sr_notify(OSThread* osthread) {
3820 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3821 assert_status(status == 0, status, "pthread_kill");
3822 return status;
3823 }
3824
3825 // "Randomly" selected value for how long we want to spin
3826 // before bailing out on suspending a thread, also how often
3827 // we send a signal to a thread we want to resume
3828 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3829 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3830
3831 // returns true on success and false on error - really an error is fatal
3832 // but this seems the normal response to library errors
3833 static bool do_suspend(OSThread* osthread) {
3834 assert(osthread->sr.is_running(), "thread should be running");
3835 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3836
3837 // mark as suspended and send signal
3838 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3839 // failed to switch, state wasn't running?
3840 ShouldNotReachHere();
3841 return false;
3842 }
3843
3844 if (sr_notify(osthread) != 0) {
3845 ShouldNotReachHere();
3846 }
3847
3848 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3849 while (true) {
3850 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3851 break;
3852 } else {
3853 // timeout
3854 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3855 if (cancelled == os::SuspendResume::SR_RUNNING) {
3856 return false;
3857 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3858 // make sure that we consume the signal on the semaphore as well
3859 sr_semaphore.wait();
3860 break;
3861 } else {
3862 ShouldNotReachHere();
3863 return false;
3864 }
3865 }
3866 }
3867
3868 guarantee(osthread->sr.is_suspended(), "Must be suspended");
3869 return true;
3870 }
3871
3872 static void do_resume(OSThread* osthread) {
3873 assert(osthread->sr.is_suspended(), "thread should be suspended");
3874 assert(!sr_semaphore.trywait(), "invalid semaphore state");
3875
3876 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3877 // failed to switch to WAKEUP_REQUEST
3878 ShouldNotReachHere();
3879 return;
3880 }
3881
3882 while (true) {
3883 if (sr_notify(osthread) == 0) {
3884 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3885 if (osthread->sr.is_running()) {
3886 return;
3887 }
3888 }
3889 } else {
3890 ShouldNotReachHere();
3891 }
3892 }
3893
3894 guarantee(osthread->sr.is_running(), "Must be running!");
3895 }
3896
3897 ////////////////////////////////////////////////////////////////////////////////
3898 // interrupt support
3899
3900 void os::interrupt(Thread* thread) {
3901 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3902 "possibility of dangling Thread pointer");
3903
3904 OSThread* osthread = thread->osthread();
3905
3906 if (!osthread->interrupted()) {
3907 osthread->set_interrupted(true);
3908 // More than one thread can get here with the same value of osthread,
3909 // resulting in multiple notifications. We do, however, want the store
3910 // to interrupted() to be visible to other threads before we execute unpark().
3911 OrderAccess::fence();
3912 ParkEvent * const slp = thread->_SleepEvent ;
3913 if (slp != NULL) slp->unpark() ;
3914 }
3915
3916 // For JSR166. Unpark even if interrupt status already was set
3917 if (thread->is_Java_thread())
3918 ((JavaThread*)thread)->parker()->unpark();
3919
3920 ParkEvent * ev = thread->_ParkEvent ;
3921 if (ev != NULL) ev->unpark() ;
3922
3923 }
3924
3925 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3926 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3927 "possibility of dangling Thread pointer");
3928
3929 OSThread* osthread = thread->osthread();
3930
3931 bool interrupted = osthread->interrupted();
3932
3933 if (interrupted && clear_interrupted) {
3934 osthread->set_interrupted(false);
3935 // consider thread->_SleepEvent->reset() ... optional optimization
3936 }
3937
3938 return interrupted;
3939 }
3940
3941 ///////////////////////////////////////////////////////////////////////////////////
3942 // signal handling (except suspend/resume)
3943
3944 // This routine may be used by user applications as a "hook" to catch signals.
3945 // The user-defined signal handler must pass unrecognized signals to this
3946 // routine, and if it returns true (non-zero), then the signal handler must
3947 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3948 // routine will never retun false (zero), but instead will execute a VM panic
3949 // routine kill the process.
3950 //
3951 // If this routine returns false, it is OK to call it again. This allows
3952 // the user-defined signal handler to perform checks either before or after
3953 // the VM performs its own checks. Naturally, the user code would be making
3954 // a serious error if it tried to handle an exception (such as a null check
3955 // or breakpoint) that the VM was generating for its own correct operation.
3956 //
3957 // This routine may recognize any of the following kinds of signals:
3958 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3959 // It should be consulted by handlers for any of those signals.
3960 //
3961 // The caller of this routine must pass in the three arguments supplied
3962 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3963 // field of the structure passed to sigaction(). This routine assumes that
3964 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3965 //
3966 // Note that the VM will print warnings if it detects conflicting signal
3967 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3968 //
3969 extern "C" JNIEXPORT int
3970 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3971 void* ucontext, int abort_if_unrecognized);
3972
3973 void signalHandler(int sig, siginfo_t* info, void* uc) {
3974 assert(info != NULL && uc != NULL, "it must be old kernel");
3975 int orig_errno = errno; // Preserve errno value over signal handler.
3976 JVM_handle_linux_signal(sig, info, uc, true);
3977 errno = orig_errno;
3978 }
3979
3980
3981 // This boolean allows users to forward their own non-matching signals
3982 // to JVM_handle_linux_signal, harmlessly.
3983 bool os::Linux::signal_handlers_are_installed = false;
3984
3985 // For signal-chaining
3986 struct sigaction os::Linux::sigact[MAXSIGNUM];
3987 unsigned int os::Linux::sigs = 0;
3988 bool os::Linux::libjsig_is_loaded = false;
3989 typedef struct sigaction *(*get_signal_t)(int);
3990 get_signal_t os::Linux::get_signal_action = NULL;
3991
3992 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3993 struct sigaction *actp = NULL;
3994
3995 if (libjsig_is_loaded) {
3996 // Retrieve the old signal handler from libjsig
3997 actp = (*get_signal_action)(sig);
3998 }
3999 if (actp == NULL) {
4000 // Retrieve the preinstalled signal handler from jvm
4001 actp = get_preinstalled_handler(sig);
4002 }
4003
4004 return actp;
4005 }
4006
4007 static bool call_chained_handler(struct sigaction *actp, int sig,
4008 siginfo_t *siginfo, void *context) {
4009 // Call the old signal handler
4010 if (actp->sa_handler == SIG_DFL) {
4011 // It's more reasonable to let jvm treat it as an unexpected exception
4012 // instead of taking the default action.
4013 return false;
4014 } else if (actp->sa_handler != SIG_IGN) {
4015 if ((actp->sa_flags & SA_NODEFER) == 0) {
4016 // automaticlly block the signal
4017 sigaddset(&(actp->sa_mask), sig);
4018 }
4019
4020 sa_handler_t hand;
4021 sa_sigaction_t sa;
4022 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4023 // retrieve the chained handler
4024 if (siginfo_flag_set) {
4025 sa = actp->sa_sigaction;
4026 } else {
4027 hand = actp->sa_handler;
4028 }
4029
4030 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4031 actp->sa_handler = SIG_DFL;
4032 }
4033
4034 // try to honor the signal mask
4035 sigset_t oset;
4036 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4037
4038 // call into the chained handler
4039 if (siginfo_flag_set) {
4040 (*sa)(sig, siginfo, context);
4041 } else {
4042 (*hand)(sig);
4043 }
4044
4045 // restore the signal mask
4046 pthread_sigmask(SIG_SETMASK, &oset, 0);
4047 }
4048 // Tell jvm's signal handler the signal is taken care of.
4049 return true;
4050 }
4051
4052 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4053 bool chained = false;
4054 // signal-chaining
4055 if (UseSignalChaining) {
4056 struct sigaction *actp = get_chained_signal_action(sig);
4057 if (actp != NULL) {
4058 chained = call_chained_handler(actp, sig, siginfo, context);
4059 }
4060 }
4061 return chained;
4062 }
4063
4064 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4065 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4066 return &sigact[sig];
4067 }
4068 return NULL;
4069 }
4070
4071 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4072 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4073 sigact[sig] = oldAct;
4074 sigs |= (unsigned int)1 << sig;
4075 }
4076
4077 // for diagnostic
4078 int os::Linux::sigflags[MAXSIGNUM];
4079
4080 int os::Linux::get_our_sigflags(int sig) {
4081 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4082 return sigflags[sig];
4083 }
4084
4085 void os::Linux::set_our_sigflags(int sig, int flags) {
4086 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4087 sigflags[sig] = flags;
4088 }
4089
4090 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4091 // Check for overwrite.
4092 struct sigaction oldAct;
4093 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4094
4095 void* oldhand = oldAct.sa_sigaction
4096 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4097 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4098 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4099 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4100 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4101 if (AllowUserSignalHandlers || !set_installed) {
4102 // Do not overwrite; user takes responsibility to forward to us.
4103 return;
4104 } else if (UseSignalChaining) {
4105 // save the old handler in jvm
4106 save_preinstalled_handler(sig, oldAct);
4107 // libjsig also interposes the sigaction() call below and saves the
4108 // old sigaction on it own.
4109 } else {
4110 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4111 "%#lx for signal %d.", (long)oldhand, sig));
4112 }
4113 }
4114
4115 struct sigaction sigAct;
4116 sigfillset(&(sigAct.sa_mask));
4117 sigAct.sa_handler = SIG_DFL;
4118 if (!set_installed) {
4119 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4120 } else {
4121 sigAct.sa_sigaction = signalHandler;
4122 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4123 }
4124 // Save flags, which are set by ours
4125 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4126 sigflags[sig] = sigAct.sa_flags;
4127
4128 int ret = sigaction(sig, &sigAct, &oldAct);
4129 assert(ret == 0, "check");
4130
4131 void* oldhand2 = oldAct.sa_sigaction
4132 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4133 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4134 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4135 }
4136
4137 // install signal handlers for signals that HotSpot needs to
4138 // handle in order to support Java-level exception handling.
4139
4140 void os::Linux::install_signal_handlers() {
4141 if (!signal_handlers_are_installed) {
4142 signal_handlers_are_installed = true;
4143
4144 // signal-chaining
4145 typedef void (*signal_setting_t)();
4146 signal_setting_t begin_signal_setting = NULL;
4147 signal_setting_t end_signal_setting = NULL;
4148 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4149 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4150 if (begin_signal_setting != NULL) {
4151 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4152 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4153 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4154 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4155 libjsig_is_loaded = true;
4156 assert(UseSignalChaining, "should enable signal-chaining");
4157 }
4158 if (libjsig_is_loaded) {
4159 // Tell libjsig jvm is setting signal handlers
4160 (*begin_signal_setting)();
4161 }
4162
4163 set_signal_handler(SIGSEGV, true);
4164 set_signal_handler(SIGPIPE, true);
4165 set_signal_handler(SIGBUS, true);
4166 set_signal_handler(SIGILL, true);
4167 set_signal_handler(SIGFPE, true);
4168 set_signal_handler(SIGXFSZ, true);
4169
4170 if (libjsig_is_loaded) {
4171 // Tell libjsig jvm finishes setting signal handlers
4172 (*end_signal_setting)();
4173 }
4174
4175 // We don't activate signal checker if libjsig is in place, we trust ourselves
4176 // and if UserSignalHandler is installed all bets are off.
4177 // Log that signal checking is off only if -verbose:jni is specified.
4178 if (CheckJNICalls) {
4179 if (libjsig_is_loaded) {
4180 if (PrintJNIResolving) {
4181 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4182 }
4183 check_signals = false;
4184 }
4185 if (AllowUserSignalHandlers) {
4186 if (PrintJNIResolving) {
4187 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4188 }
4189 check_signals = false;
4190 }
4191 }
4192 }
4193 }
4194
4195 // This is the fastest way to get thread cpu time on Linux.
4196 // Returns cpu time (user+sys) for any thread, not only for current.
4197 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4198 // It might work on 2.6.10+ with a special kernel/glibc patch.
4199 // For reference, please, see IEEE Std 1003.1-2004:
4200 // http://www.unix.org/single_unix_specification
4201
4202 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4203 struct timespec tp;
4204 int rc = os::Linux::clock_gettime(clockid, &tp);
4205 assert(rc == 0, "clock_gettime is expected to return 0 code");
4206
4207 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4208 }
4209
4210 /////
4211 // glibc on Linux platform uses non-documented flag
4212 // to indicate, that some special sort of signal
4213 // trampoline is used.
4214 // We will never set this flag, and we should
4215 // ignore this flag in our diagnostic
4216 #ifdef SIGNIFICANT_SIGNAL_MASK
4217 #undef SIGNIFICANT_SIGNAL_MASK
4218 #endif
4219 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4220
4221 static const char* get_signal_handler_name(address handler,
4222 char* buf, int buflen) {
4223 int offset;
4224 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4225 if (found) {
4226 // skip directory names
4227 const char *p1, *p2;
4228 p1 = buf;
4229 size_t len = strlen(os::file_separator());
4230 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4231 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4232 } else {
4233 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4234 }
4235 return buf;
4236 }
4237
4238 static void print_signal_handler(outputStream* st, int sig,
4239 char* buf, size_t buflen) {
4240 struct sigaction sa;
4241
4242 sigaction(sig, NULL, &sa);
4243
4244 // See comment for SIGNIFICANT_SIGNAL_MASK define
4245 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4246
4247 st->print("%s: ", os::exception_name(sig, buf, buflen));
4248
4249 address handler = (sa.sa_flags & SA_SIGINFO)
4250 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4251 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4252
4253 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4254 st->print("SIG_DFL");
4255 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4256 st->print("SIG_IGN");
4257 } else {
4258 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4259 }
4260
4261 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4262
4263 address rh = VMError::get_resetted_sighandler(sig);
4264 // May be, handler was resetted by VMError?
4265 if(rh != NULL) {
4266 handler = rh;
4267 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4268 }
4269
4270 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
4271
4272 // Check: is it our handler?
4273 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4274 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4275 // It is our signal handler
4276 // check for flags, reset system-used one!
4277 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4278 st->print(
4279 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4280 os::Linux::get_our_sigflags(sig));
4281 }
4282 }
4283 st->cr();
4284 }
4285
4286
4287 #define DO_SIGNAL_CHECK(sig) \
4288 if (!sigismember(&check_signal_done, sig)) \
4289 os::Linux::check_signal_handler(sig)
4290
4291 // This method is a periodic task to check for misbehaving JNI applications
4292 // under CheckJNI, we can add any periodic checks here
4293
4294 void os::run_periodic_checks() {
4295
4296 if (check_signals == false) return;
4297
4298 // SEGV and BUS if overridden could potentially prevent
4299 // generation of hs*.log in the event of a crash, debugging
4300 // such a case can be very challenging, so we absolutely
4301 // check the following for a good measure:
4302 DO_SIGNAL_CHECK(SIGSEGV);
4303 DO_SIGNAL_CHECK(SIGILL);
4304 DO_SIGNAL_CHECK(SIGFPE);
4305 DO_SIGNAL_CHECK(SIGBUS);
4306 DO_SIGNAL_CHECK(SIGPIPE);
4307 DO_SIGNAL_CHECK(SIGXFSZ);
4308
4309
4310 // ReduceSignalUsage allows the user to override these handlers
4311 // see comments at the very top and jvm_solaris.h
4312 if (!ReduceSignalUsage) {
4313 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4314 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4315 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4316 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4317 }
4318
4319 DO_SIGNAL_CHECK(SR_signum);
4320 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4321 }
4322
4323 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4324
4325 static os_sigaction_t os_sigaction = NULL;
4326
4327 void os::Linux::check_signal_handler(int sig) {
4328 char buf[O_BUFLEN];
4329 address jvmHandler = NULL;
4330
4331
4332 struct sigaction act;
4333 if (os_sigaction == NULL) {
4334 // only trust the default sigaction, in case it has been interposed
4335 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4336 if (os_sigaction == NULL) return;
4337 }
4338
4339 os_sigaction(sig, (struct sigaction*)NULL, &act);
4340
4341
4342 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4343
4344 address thisHandler = (act.sa_flags & SA_SIGINFO)
4345 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4346 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4347
4348
4349 switch(sig) {
4350 case SIGSEGV:
4351 case SIGBUS:
4352 case SIGFPE:
4353 case SIGPIPE:
4354 case SIGILL:
4355 case SIGXFSZ:
4356 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4357 break;
4358
4359 case SHUTDOWN1_SIGNAL:
4360 case SHUTDOWN2_SIGNAL:
4361 case SHUTDOWN3_SIGNAL:
4362 case BREAK_SIGNAL:
4363 jvmHandler = (address)user_handler();
4364 break;
4365
4366 case INTERRUPT_SIGNAL:
4367 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4368 break;
4369
4370 default:
4371 if (sig == SR_signum) {
4372 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4373 } else {
4374 return;
4375 }
4376 break;
4377 }
4378
4379 if (thisHandler != jvmHandler) {
4380 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4381 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4382 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4383 // No need to check this sig any longer
4384 sigaddset(&check_signal_done, sig);
4385 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4386 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4387 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4388 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4389 // No need to check this sig any longer
4390 sigaddset(&check_signal_done, sig);
4391 }
4392
4393 // Dump all the signal
4394 if (sigismember(&check_signal_done, sig)) {
4395 print_signal_handlers(tty, buf, O_BUFLEN);
4396 }
4397 }
4398
4399 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4400
4401 extern bool signal_name(int signo, char* buf, size_t len);
4402
4403 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4404 if (0 < exception_code && exception_code <= SIGRTMAX) {
4405 // signal
4406 if (!signal_name(exception_code, buf, size)) {
4407 jio_snprintf(buf, size, "SIG%d", exception_code);
4408 }
4409 return buf;
4410 } else {
4411 return NULL;
4412 }
4413 }
4414
4415 // this is called _before_ the most of global arguments have been parsed
4416 void os::init(void) {
4417 char dummy; /* used to get a guess on initial stack address */
4418 // first_hrtime = gethrtime();
4419
4420 // With LinuxThreads the JavaMain thread pid (primordial thread)
4421 // is different than the pid of the java launcher thread.
4422 // So, on Linux, the launcher thread pid is passed to the VM
4423 // via the sun.java.launcher.pid property.
4424 // Use this property instead of getpid() if it was correctly passed.
4425 // See bug 6351349.
4426 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4427
4428 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4429
4430 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4431
4432 init_random(1234567);
4433
4434 ThreadCritical::initialize();
4435
4436 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4437 if (Linux::page_size() == -1) {
4438 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4439 strerror(errno)));
4440 }
4441 init_page_sizes((size_t) Linux::page_size());
4442
4443 Linux::initialize_system_info();
4444
4445 // main_thread points to the aboriginal thread
4446 Linux::_main_thread = pthread_self();
4447
4448 Linux::clock_init();
4449 initial_time_count = os::elapsed_counter();
4450 pthread_mutex_init(&dl_mutex, NULL);
4451
4452 // If the pagesize of the VM is greater than 8K determine the appropriate
4453 // number of initial guard pages. The user can change this with the
4454 // command line arguments, if needed.
4455 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4456 StackYellowPages = 1;
4457 StackRedPages = 1;
4458 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4459 }
4460 }
4461
4462 // To install functions for atexit system call
4463 extern "C" {
4464 static void perfMemory_exit_helper() {
4465 perfMemory_exit();
4466 }
4467 }
4468
4469 // this is called _after_ the global arguments have been parsed
4470 jint os::init_2(void)
4471 {
4472 Linux::fast_thread_clock_init();
4473
4474 // Allocate a single page and mark it as readable for safepoint polling
4475 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4476 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4477
4478 os::set_polling_page( polling_page );
4479
4480 #ifndef PRODUCT
4481 if(Verbose && PrintMiscellaneous)
4482 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4483 #endif
4484
4485 if (!UseMembar) {
4486 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4487 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4488 os::set_memory_serialize_page( mem_serialize_page );
4489
4490 #ifndef PRODUCT
4491 if(Verbose && PrintMiscellaneous)
4492 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4493 #endif
4494 }
4495
4496 os::large_page_init();
4497
4498 // initialize suspend/resume support - must do this before signal_sets_init()
4499 if (SR_initialize() != 0) {
4500 perror("SR_initialize failed");
4501 return JNI_ERR;
4502 }
4503
4504 Linux::signal_sets_init();
4505 Linux::install_signal_handlers();
4506
4507 // Check minimum allowable stack size for thread creation and to initialize
4508 // the java system classes, including StackOverflowError - depends on page
4509 // size. Add a page for compiler2 recursion in main thread.
4510 // Add in 2*BytesPerWord times page size to account for VM stack during
4511 // class initialization depending on 32 or 64 bit VM.
4512 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4513 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4514 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4515
4516 size_t threadStackSizeInBytes = ThreadStackSize * K;
4517 if (threadStackSizeInBytes != 0 &&
4518 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4519 tty->print_cr("\nThe stack size specified is too small, "
4520 "Specify at least %dk",
4521 os::Linux::min_stack_allowed/ K);
4522 return JNI_ERR;
4523 }
4524
4525 // Make the stack size a multiple of the page size so that
4526 // the yellow/red zones can be guarded.
4527 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4528 vm_page_size()));
4529
4530 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4531
4532 Linux::libpthread_init();
4533 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4534 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4535 Linux::glibc_version(), Linux::libpthread_version(),
4536 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4537 }
4538
4539 if (UseNUMA) {
4540 if (!Linux::libnuma_init()) {
4541 UseNUMA = false;
4542 } else {
4543 if ((Linux::numa_max_node() < 1)) {
4544 // There's only one node(they start from 0), disable NUMA.
4545 UseNUMA = false;
4546 }
4547 }
4548 // With SHM large pages we cannot uncommit a page, so there's not way
4549 // we can make the adaptive lgrp chunk resizing work. If the user specified
4550 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4551 // disable adaptive resizing.
4552 if (UseNUMA && UseLargePages && UseSHM) {
4553 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4554 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4555 UseLargePages = false;
4556 } else {
4557 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4558 UseAdaptiveSizePolicy = false;
4559 UseAdaptiveNUMAChunkSizing = false;
4560 }
4561 } else {
4562 UseNUMA = false;
4563 }
4564 }
4565 if (!UseNUMA && ForceNUMA) {
4566 UseNUMA = true;
4567 }
4568 }
4569
4570 if (MaxFDLimit) {
4571 // set the number of file descriptors to max. print out error
4572 // if getrlimit/setrlimit fails but continue regardless.
4573 struct rlimit nbr_files;
4574 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4575 if (status != 0) {
4576 if (PrintMiscellaneous && (Verbose || WizardMode))
4577 perror("os::init_2 getrlimit failed");
4578 } else {
4579 nbr_files.rlim_cur = nbr_files.rlim_max;
4580 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4581 if (status != 0) {
4582 if (PrintMiscellaneous && (Verbose || WizardMode))
4583 perror("os::init_2 setrlimit failed");
4584 }
4585 }
4586 }
4587
4588 // Initialize lock used to serialize thread creation (see os::create_thread)
4589 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4590
4591 // at-exit methods are called in the reverse order of their registration.
4592 // atexit functions are called on return from main or as a result of a
4593 // call to exit(3C). There can be only 32 of these functions registered
4594 // and atexit() does not set errno.
4595
4596 if (PerfAllowAtExitRegistration) {
4597 // only register atexit functions if PerfAllowAtExitRegistration is set.
4598 // atexit functions can be delayed until process exit time, which
4599 // can be problematic for embedded VM situations. Embedded VMs should
4600 // call DestroyJavaVM() to assure that VM resources are released.
4601
4602 // note: perfMemory_exit_helper atexit function may be removed in
4603 // the future if the appropriate cleanup code can be added to the
4604 // VM_Exit VMOperation's doit method.
4605 if (atexit(perfMemory_exit_helper) != 0) {
4606 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4607 }
4608 }
4609
4610 // initialize thread priority policy
4611 prio_init();
4612
4613 return JNI_OK;
4614 }
4615
4616 // this is called at the end of vm_initialization
4617 void os::init_3(void)
4618 {
4619 #ifdef JAVASE_EMBEDDED
4620 // Start the MemNotifyThread
4621 if (LowMemoryProtection) {
4622 MemNotifyThread::start();
4623 }
4624 return;
4625 #endif
4626 }
4627
4628 // Mark the polling page as unreadable
4629 void os::make_polling_page_unreadable(void) {
4630 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4631 fatal("Could not disable polling page");
4632 };
4633
4634 // Mark the polling page as readable
4635 void os::make_polling_page_readable(void) {
4636 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4637 fatal("Could not enable polling page");
4638 }
4639 };
4640
4641 int os::active_processor_count() {
4642 // Linux doesn't yet have a (official) notion of processor sets,
4643 // so just return the number of online processors.
4644 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4645 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4646 return online_cpus;
4647 }
4648
4649 void os::set_native_thread_name(const char *name) {
4650 // Not yet implemented.
4651 return;
4652 }
4653
4654 bool os::distribute_processes(uint length, uint* distribution) {
4655 // Not yet implemented.
4656 return false;
4657 }
4658
4659 bool os::bind_to_processor(uint processor_id) {
4660 // Not yet implemented.
4661 return false;
4662 }
4663
4664 ///
4665
4666 void os::SuspendedThreadTask::internal_do_task() {
4667 if (do_suspend(_thread->osthread())) {
4668 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4669 do_task(context);
4670 do_resume(_thread->osthread());
4671 }
4672 }
4673
4674 class PcFetcher : public os::SuspendedThreadTask {
4675 public:
4676 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4677 ExtendedPC result();
4678 protected:
4679 void do_task(const os::SuspendedThreadTaskContext& context);
4680 private:
4681 ExtendedPC _epc;
4682 };
4683
4684 ExtendedPC PcFetcher::result() {
4685 guarantee(is_done(), "task is not done yet.");
4686 return _epc;
4687 }
4688
4689 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4690 Thread* thread = context.thread();
4691 OSThread* osthread = thread->osthread();
4692 if (osthread->ucontext() != NULL) {
4693 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4694 } else {
4695 // NULL context is unexpected, double-check this is the VMThread
4696 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4697 }
4698 }
4699
4700 // Suspends the target using the signal mechanism and then grabs the PC before
4701 // resuming the target. Used by the flat-profiler only
4702 ExtendedPC os::get_thread_pc(Thread* thread) {
4703 // Make sure that it is called by the watcher for the VMThread
4704 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4705 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4706
4707 PcFetcher fetcher(thread);
4708 fetcher.run();
4709 return fetcher.result();
4710 }
4711
4712 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4713 {
4714 if (is_NPTL()) {
4715 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4716 } else {
4717 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4718 // word back to default 64bit precision if condvar is signaled. Java
4719 // wants 53bit precision. Save and restore current value.
4720 int fpu = get_fpu_control_word();
4721 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4722 set_fpu_control_word(fpu);
4723 return status;
4724 }
4725 }
4726
4727 ////////////////////////////////////////////////////////////////////////////////
4728 // debug support
4729
4730 bool os::find(address addr, outputStream* st) {
4731 Dl_info dlinfo;
4732 memset(&dlinfo, 0, sizeof(dlinfo));
4733 if (dladdr(addr, &dlinfo)) {
4734 st->print(PTR_FORMAT ": ", addr);
4735 if (dlinfo.dli_sname != NULL) {
4736 st->print("%s+%#x", dlinfo.dli_sname,
4737 addr - (intptr_t)dlinfo.dli_saddr);
4738 } else if (dlinfo.dli_fname) {
4739 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4740 } else {
4741 st->print("<absolute address>");
4742 }
4743 if (dlinfo.dli_fname) {
4744 st->print(" in %s", dlinfo.dli_fname);
4745 }
4746 if (dlinfo.dli_fbase) {
4747 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4748 }
4749 st->cr();
4750
4751 if (Verbose) {
4752 // decode some bytes around the PC
4753 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4754 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4755 address lowest = (address) dlinfo.dli_sname;
4756 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4757 if (begin < lowest) begin = lowest;
4758 Dl_info dlinfo2;
4759 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4760 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4761 end = (address) dlinfo2.dli_saddr;
4762 Disassembler::decode(begin, end, st);
4763 }
4764 return true;
4765 }
4766 return false;
4767 }
4768
4769 ////////////////////////////////////////////////////////////////////////////////
4770 // misc
4771
4772 // This does not do anything on Linux. This is basically a hook for being
4773 // able to use structured exception handling (thread-local exception filters)
4774 // on, e.g., Win32.
4775 void
4776 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4777 JavaCallArguments* args, Thread* thread) {
4778 f(value, method, args, thread);
4779 }
4780
4781 void os::print_statistics() {
4782 }
4783
4784 int os::message_box(const char* title, const char* message) {
4785 int i;
4786 fdStream err(defaultStream::error_fd());
4787 for (i = 0; i < 78; i++) err.print_raw("=");
4788 err.cr();
4789 err.print_raw_cr(title);
4790 for (i = 0; i < 78; i++) err.print_raw("-");
4791 err.cr();
4792 err.print_raw_cr(message);
4793 for (i = 0; i < 78; i++) err.print_raw("=");
4794 err.cr();
4795
4796 char buf[16];
4797 // Prevent process from exiting upon "read error" without consuming all CPU
4798 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4799
4800 return buf[0] == 'y' || buf[0] == 'Y';
4801 }
4802
4803 int os::stat(const char *path, struct stat *sbuf) {
4804 char pathbuf[MAX_PATH];
4805 if (strlen(path) > MAX_PATH - 1) {
4806 errno = ENAMETOOLONG;
4807 return -1;
4808 }
4809 os::native_path(strcpy(pathbuf, path));
4810 return ::stat(pathbuf, sbuf);
4811 }
4812
4813 bool os::check_heap(bool force) {
4814 return true;
4815 }
4816
4817 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4818 return ::vsnprintf(buf, count, format, args);
4819 }
4820
4821 // Is a (classpath) directory empty?
4822 bool os::dir_is_empty(const char* path) {
4823 DIR *dir = NULL;
4824 struct dirent *ptr;
4825
4826 dir = opendir(path);
4827 if (dir == NULL) return true;
4828
4829 /* Scan the directory */
4830 bool result = true;
4831 char buf[sizeof(struct dirent) + MAX_PATH];
4832 while (result && (ptr = ::readdir(dir)) != NULL) {
4833 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4834 result = false;
4835 }
4836 }
4837 closedir(dir);
4838 return result;
4839 }
4840
4841 // This code originates from JDK's sysOpen and open64_w
4842 // from src/solaris/hpi/src/system_md.c
4843
4844 #ifndef O_DELETE
4845 #define O_DELETE 0x10000
4846 #endif
4847
4848 // Open a file. Unlink the file immediately after open returns
4849 // if the specified oflag has the O_DELETE flag set.
4850 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4851
4852 int os::open(const char *path, int oflag, int mode) {
4853
4854 if (strlen(path) > MAX_PATH - 1) {
4855 errno = ENAMETOOLONG;
4856 return -1;
4857 }
4858 int fd;
4859 int o_delete = (oflag & O_DELETE);
4860 oflag = oflag & ~O_DELETE;
4861
4862 fd = ::open64(path, oflag, mode);
4863 if (fd == -1) return -1;
4864
4865 //If the open succeeded, the file might still be a directory
4866 {
4867 struct stat64 buf64;
4868 int ret = ::fstat64(fd, &buf64);
4869 int st_mode = buf64.st_mode;
4870
4871 if (ret != -1) {
4872 if ((st_mode & S_IFMT) == S_IFDIR) {
4873 errno = EISDIR;
4874 ::close(fd);
4875 return -1;
4876 }
4877 } else {
4878 ::close(fd);
4879 return -1;
4880 }
4881 }
4882
4883 /*
4884 * All file descriptors that are opened in the JVM and not
4885 * specifically destined for a subprocess should have the
4886 * close-on-exec flag set. If we don't set it, then careless 3rd
4887 * party native code might fork and exec without closing all
4888 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4889 * UNIXProcess.c), and this in turn might:
4890 *
4891 * - cause end-of-file to fail to be detected on some file
4892 * descriptors, resulting in mysterious hangs, or
4893 *
4894 * - might cause an fopen in the subprocess to fail on a system
4895 * suffering from bug 1085341.
4896 *
4897 * (Yes, the default setting of the close-on-exec flag is a Unix
4898 * design flaw)
4899 *
4900 * See:
4901 * 1085341: 32-bit stdio routines should support file descriptors >255
4902 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4903 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4904 */
4905 #ifdef FD_CLOEXEC
4906 {
4907 int flags = ::fcntl(fd, F_GETFD);
4908 if (flags != -1)
4909 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4910 }
4911 #endif
4912
4913 if (o_delete != 0) {
4914 ::unlink(path);
4915 }
4916 return fd;
4917 }
4918
4919
4920 // create binary file, rewriting existing file if required
4921 int os::create_binary_file(const char* path, bool rewrite_existing) {
4922 int oflags = O_WRONLY | O_CREAT;
4923 if (!rewrite_existing) {
4924 oflags |= O_EXCL;
4925 }
4926 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4927 }
4928
4929 // return current position of file pointer
4930 jlong os::current_file_offset(int fd) {
4931 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4932 }
4933
4934 // move file pointer to the specified offset
4935 jlong os::seek_to_file_offset(int fd, jlong offset) {
4936 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4937 }
4938
4939 // This code originates from JDK's sysAvailable
4940 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4941
4942 int os::available(int fd, jlong *bytes) {
4943 jlong cur, end;
4944 int mode;
4945 struct stat64 buf64;
4946
4947 if (::fstat64(fd, &buf64) >= 0) {
4948 mode = buf64.st_mode;
4949 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4950 /*
4951 * XXX: is the following call interruptible? If so, this might
4952 * need to go through the INTERRUPT_IO() wrapper as for other
4953 * blocking, interruptible calls in this file.
4954 */
4955 int n;
4956 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4957 *bytes = n;
4958 return 1;
4959 }
4960 }
4961 }
4962 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4963 return 0;
4964 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4965 return 0;
4966 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4967 return 0;
4968 }
4969 *bytes = end - cur;
4970 return 1;
4971 }
4972
4973 int os::socket_available(int fd, jint *pbytes) {
4974 // Linux doc says EINTR not returned, unlike Solaris
4975 int ret = ::ioctl(fd, FIONREAD, pbytes);
4976
4977 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4978 // is expected to return 0 on failure and 1 on success to the jdk.
4979 return (ret < 0) ? 0 : 1;
4980 }
4981
4982 // Map a block of memory.
4983 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4984 char *addr, size_t bytes, bool read_only,
4985 bool allow_exec) {
4986 int prot;
4987 int flags = MAP_PRIVATE;
4988
4989 if (read_only) {
4990 prot = PROT_READ;
4991 } else {
4992 prot = PROT_READ | PROT_WRITE;
4993 }
4994
4995 if (allow_exec) {
4996 prot |= PROT_EXEC;
4997 }
4998
4999 if (addr != NULL) {
5000 flags |= MAP_FIXED;
5001 }
5002
5003 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5004 fd, file_offset);
5005 if (mapped_address == MAP_FAILED) {
5006 return NULL;
5007 }
5008 return mapped_address;
5009 }
5010
5011
5012 // Remap a block of memory.
5013 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5014 char *addr, size_t bytes, bool read_only,
5015 bool allow_exec) {
5016 // same as map_memory() on this OS
5017 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5018 allow_exec);
5019 }
5020
5021
5022 // Unmap a block of memory.
5023 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5024 return munmap(addr, bytes) == 0;
5025 }
5026
5027 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5028
5029 static clockid_t thread_cpu_clockid(Thread* thread) {
5030 pthread_t tid = thread->osthread()->pthread_id();
5031 clockid_t clockid;
5032
5033 // Get thread clockid
5034 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5035 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5036 return clockid;
5037 }
5038
5039 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5040 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5041 // of a thread.
5042 //
5043 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5044 // the fast estimate available on the platform.
5045
5046 jlong os::current_thread_cpu_time() {
5047 if (os::Linux::supports_fast_thread_cpu_time()) {
5048 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5049 } else {
5050 // return user + sys since the cost is the same
5051 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5052 }
5053 }
5054
5055 jlong os::thread_cpu_time(Thread* thread) {
5056 // consistent with what current_thread_cpu_time() returns
5057 if (os::Linux::supports_fast_thread_cpu_time()) {
5058 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5059 } else {
5060 return slow_thread_cpu_time(thread, true /* user + sys */);
5061 }
5062 }
5063
5064 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5065 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5066 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5067 } else {
5068 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5069 }
5070 }
5071
5072 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5073 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5074 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5075 } else {
5076 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5077 }
5078 }
5079
5080 //
5081 // -1 on error.
5082 //
5083
5084 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5085 static bool proc_task_unchecked = true;
5086 static const char *proc_stat_path = "/proc/%d/stat";
5087 pid_t tid = thread->osthread()->thread_id();
5088 char *s;
5089 char stat[2048];
5090 int statlen;
5091 char proc_name[64];
5092 int count;
5093 long sys_time, user_time;
5094 char cdummy;
5095 int idummy;
5096 long ldummy;
5097 FILE *fp;
5098
5099 // The /proc/<tid>/stat aggregates per-process usage on
5100 // new Linux kernels 2.6+ where NPTL is supported.
5101 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5102 // See bug 6328462.
5103 // There possibly can be cases where there is no directory
5104 // /proc/self/task, so we check its availability.
5105 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5106 // This is executed only once
5107 proc_task_unchecked = false;
5108 fp = fopen("/proc/self/task", "r");
5109 if (fp != NULL) {
5110 proc_stat_path = "/proc/self/task/%d/stat";
5111 fclose(fp);
5112 }
5113 }
5114
5115 sprintf(proc_name, proc_stat_path, tid);
5116 fp = fopen(proc_name, "r");
5117 if ( fp == NULL ) return -1;
5118 statlen = fread(stat, 1, 2047, fp);
5119 stat[statlen] = '\0';
5120 fclose(fp);
5121
5122 // Skip pid and the command string. Note that we could be dealing with
5123 // weird command names, e.g. user could decide to rename java launcher
5124 // to "java 1.4.2 :)", then the stat file would look like
5125 // 1234 (java 1.4.2 :)) R ... ...
5126 // We don't really need to know the command string, just find the last
5127 // occurrence of ")" and then start parsing from there. See bug 4726580.
5128 s = strrchr(stat, ')');
5129 if (s == NULL ) return -1;
5130
5131 // Skip blank chars
5132 do s++; while (isspace(*s));
5133
5134 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5135 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5136 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5137 &user_time, &sys_time);
5138 if ( count != 13 ) return -1;
5139 if (user_sys_cpu_time) {
5140 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5141 } else {
5142 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5143 }
5144 }
5145
5146 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5147 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5148 info_ptr->may_skip_backward = false; // elapsed time not wall time
5149 info_ptr->may_skip_forward = false; // elapsed time not wall time
5150 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5151 }
5152
5153 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5154 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5155 info_ptr->may_skip_backward = false; // elapsed time not wall time
5156 info_ptr->may_skip_forward = false; // elapsed time not wall time
5157 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5158 }
5159
5160 bool os::is_thread_cpu_time_supported() {
5161 return true;
5162 }
5163
5164 // System loadavg support. Returns -1 if load average cannot be obtained.
5165 // Linux doesn't yet have a (official) notion of processor sets,
5166 // so just return the system wide load average.
5167 int os::loadavg(double loadavg[], int nelem) {
5168 return ::getloadavg(loadavg, nelem);
5169 }
5170
5171 void os::pause() {
5172 char filename[MAX_PATH];
5173 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5174 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5175 } else {
5176 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5177 }
5178
5179 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5180 if (fd != -1) {
5181 struct stat buf;
5182 ::close(fd);
5183 while (::stat(filename, &buf) == 0) {
5184 (void)::poll(NULL, 0, 100);
5185 }
5186 } else {
5187 jio_fprintf(stderr,
5188 "Could not open pause file '%s', continuing immediately.\n", filename);
5189 }
5190 }
5191
5192
5193 // Refer to the comments in os_solaris.cpp park-unpark.
5194 //
5195 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5196 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5197 // For specifics regarding the bug see GLIBC BUGID 261237 :
5198 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5199 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5200 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5201 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5202 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5203 // and monitorenter when we're using 1-0 locking. All those operations may result in
5204 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5205 // of libpthread avoids the problem, but isn't practical.
5206 //
5207 // Possible remedies:
5208 //
5209 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5210 // This is palliative and probabilistic, however. If the thread is preempted
5211 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5212 // than the minimum period may have passed, and the abstime may be stale (in the
5213 // past) resultin in a hang. Using this technique reduces the odds of a hang
5214 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5215 //
5216 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5217 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5218 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5219 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5220 // thread.
5221 //
5222 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5223 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5224 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5225 // This also works well. In fact it avoids kernel-level scalability impediments
5226 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5227 // timers in a graceful fashion.
5228 //
5229 // 4. When the abstime value is in the past it appears that control returns
5230 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5231 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5232 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5233 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5234 // It may be possible to avoid reinitialization by checking the return
5235 // value from pthread_cond_timedwait(). In addition to reinitializing the
5236 // condvar we must establish the invariant that cond_signal() is only called
5237 // within critical sections protected by the adjunct mutex. This prevents
5238 // cond_signal() from "seeing" a condvar that's in the midst of being
5239 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5240 // desirable signal-after-unlock optimization that avoids futile context switching.
5241 //
5242 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5243 // structure when a condvar is used or initialized. cond_destroy() would
5244 // release the helper structure. Our reinitialize-after-timedwait fix
5245 // put excessive stress on malloc/free and locks protecting the c-heap.
5246 //
5247 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5248 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5249 // and only enabling the work-around for vulnerable environments.
5250
5251 // utility to compute the abstime argument to timedwait:
5252 // millis is the relative timeout time
5253 // abstime will be the absolute timeout time
5254 // TODO: replace compute_abstime() with unpackTime()
5255
5256 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5257 if (millis < 0) millis = 0;
5258 struct timeval now;
5259 int status = gettimeofday(&now, NULL);
5260 assert(status == 0, "gettimeofday");
5261 jlong seconds = millis / 1000;
5262 millis %= 1000;
5263 if (seconds > 50000000) { // see man cond_timedwait(3T)
5264 seconds = 50000000;
5265 }
5266 abstime->tv_sec = now.tv_sec + seconds;
5267 long usec = now.tv_usec + millis * 1000;
5268 if (usec >= 1000000) {
5269 abstime->tv_sec += 1;
5270 usec -= 1000000;
5271 }
5272 abstime->tv_nsec = usec * 1000;
5273 return abstime;
5274 }
5275
5276
5277 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5278 // Conceptually TryPark() should be equivalent to park(0).
5279
5280 int os::PlatformEvent::TryPark() {
5281 for (;;) {
5282 const int v = _Event ;
5283 guarantee ((v == 0) || (v == 1), "invariant") ;
5284 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5285 }
5286 }
5287
5288 void os::PlatformEvent::park() { // AKA "down()"
5289 // Invariant: Only the thread associated with the Event/PlatformEvent
5290 // may call park().
5291 // TODO: assert that _Assoc != NULL or _Assoc == Self
5292 int v ;
5293 for (;;) {
5294 v = _Event ;
5295 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5296 }
5297 guarantee (v >= 0, "invariant") ;
5298 if (v == 0) {
5299 // Do this the hard way by blocking ...
5300 int status = pthread_mutex_lock(_mutex);
5301 assert_status(status == 0, status, "mutex_lock");
5302 guarantee (_nParked == 0, "invariant") ;
5303 ++ _nParked ;
5304 while (_Event < 0) {
5305 status = pthread_cond_wait(_cond, _mutex);
5306 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5307 // Treat this the same as if the wait was interrupted
5308 if (status == ETIME) { status = EINTR; }
5309 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5310 }
5311 -- _nParked ;
5312
5313 _Event = 0 ;
5314 status = pthread_mutex_unlock(_mutex);
5315 assert_status(status == 0, status, "mutex_unlock");
5316 // Paranoia to ensure our locked and lock-free paths interact
5317 // correctly with each other.
5318 OrderAccess::fence();
5319 }
5320 guarantee (_Event >= 0, "invariant") ;
5321 }
5322
5323 int os::PlatformEvent::park(jlong millis) {
5324 guarantee (_nParked == 0, "invariant") ;
5325
5326 int v ;
5327 for (;;) {
5328 v = _Event ;
5329 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5330 }
5331 guarantee (v >= 0, "invariant") ;
5332 if (v != 0) return OS_OK ;
5333
5334 // We do this the hard way, by blocking the thread.
5335 // Consider enforcing a minimum timeout value.
5336 struct timespec abst;
5337 compute_abstime(&abst, millis);
5338
5339 int ret = OS_TIMEOUT;
5340 int status = pthread_mutex_lock(_mutex);
5341 assert_status(status == 0, status, "mutex_lock");
5342 guarantee (_nParked == 0, "invariant") ;
5343 ++_nParked ;
5344
5345 // Object.wait(timo) will return because of
5346 // (a) notification
5347 // (b) timeout
5348 // (c) thread.interrupt
5349 //
5350 // Thread.interrupt and object.notify{All} both call Event::set.
5351 // That is, we treat thread.interrupt as a special case of notification.
5352 // The underlying Solaris implementation, cond_timedwait, admits
5353 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5354 // JVM from making those visible to Java code. As such, we must
5355 // filter out spurious wakeups. We assume all ETIME returns are valid.
5356 //
5357 // TODO: properly differentiate simultaneous notify+interrupt.
5358 // In that case, we should propagate the notify to another waiter.
5359
5360 while (_Event < 0) {
5361 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5362 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5363 pthread_cond_destroy (_cond);
5364 pthread_cond_init (_cond, NULL) ;
5365 }
5366 assert_status(status == 0 || status == EINTR ||
5367 status == ETIME || status == ETIMEDOUT,
5368 status, "cond_timedwait");
5369 if (!FilterSpuriousWakeups) break ; // previous semantics
5370 if (status == ETIME || status == ETIMEDOUT) break ;
5371 // We consume and ignore EINTR and spurious wakeups.
5372 }
5373 --_nParked ;
5374 if (_Event >= 0) {
5375 ret = OS_OK;
5376 }
5377 _Event = 0 ;
5378 status = pthread_mutex_unlock(_mutex);
5379 assert_status(status == 0, status, "mutex_unlock");
5380 assert (_nParked == 0, "invariant") ;
5381 // Paranoia to ensure our locked and lock-free paths interact
5382 // correctly with each other.
5383 OrderAccess::fence();
5384 return ret;
5385 }
5386
5387 void os::PlatformEvent::unpark() {
5388 // Transitions for _Event:
5389 // 0 :=> 1
5390 // 1 :=> 1
5391 // -1 :=> either 0 or 1; must signal target thread
5392 // That is, we can safely transition _Event from -1 to either
5393 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5394 // unpark() calls.
5395 // See also: "Semaphores in Plan 9" by Mullender & Cox
5396 //
5397 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5398 // that it will take two back-to-back park() calls for the owning
5399 // thread to block. This has the benefit of forcing a spurious return
5400 // from the first park() call after an unpark() call which will help
5401 // shake out uses of park() and unpark() without condition variables.
5402
5403 if (Atomic::xchg(1, &_Event) >= 0) return;
5404
5405 // Wait for the thread associated with the event to vacate
5406 int status = pthread_mutex_lock(_mutex);
5407 assert_status(status == 0, status, "mutex_lock");
5408 int AnyWaiters = _nParked;
5409 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5410 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5411 AnyWaiters = 0;
5412 pthread_cond_signal(_cond);
5413 }
5414 status = pthread_mutex_unlock(_mutex);
5415 assert_status(status == 0, status, "mutex_unlock");
5416 if (AnyWaiters != 0) {
5417 status = pthread_cond_signal(_cond);
5418 assert_status(status == 0, status, "cond_signal");
5419 }
5420
5421 // Note that we signal() _after dropping the lock for "immortal" Events.
5422 // This is safe and avoids a common class of futile wakeups. In rare
5423 // circumstances this can cause a thread to return prematurely from
5424 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5425 // simply re-test the condition and re-park itself.
5426 }
5427
5428
5429 // JSR166
5430 // -------------------------------------------------------
5431
5432 /*
5433 * The solaris and linux implementations of park/unpark are fairly
5434 * conservative for now, but can be improved. They currently use a
5435 * mutex/condvar pair, plus a a count.
5436 * Park decrements count if > 0, else does a condvar wait. Unpark
5437 * sets count to 1 and signals condvar. Only one thread ever waits
5438 * on the condvar. Contention seen when trying to park implies that someone
5439 * is unparking you, so don't wait. And spurious returns are fine, so there
5440 * is no need to track notifications.
5441 */
5442
5443 #define MAX_SECS 100000000
5444 /*
5445 * This code is common to linux and solaris and will be moved to a
5446 * common place in dolphin.
5447 *
5448 * The passed in time value is either a relative time in nanoseconds
5449 * or an absolute time in milliseconds. Either way it has to be unpacked
5450 * into suitable seconds and nanoseconds components and stored in the
5451 * given timespec structure.
5452 * Given time is a 64-bit value and the time_t used in the timespec is only
5453 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5454 * overflow if times way in the future are given. Further on Solaris versions
5455 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5456 * number of seconds, in abstime, is less than current_time + 100,000,000.
5457 * As it will be 28 years before "now + 100000000" will overflow we can
5458 * ignore overflow and just impose a hard-limit on seconds using the value
5459 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5460 * years from "now".
5461 */
5462
5463 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5464 assert (time > 0, "convertTime");
5465
5466 struct timeval now;
5467 int status = gettimeofday(&now, NULL);
5468 assert(status == 0, "gettimeofday");
5469
5470 time_t max_secs = now.tv_sec + MAX_SECS;
5471
5472 if (isAbsolute) {
5473 jlong secs = time / 1000;
5474 if (secs > max_secs) {
5475 absTime->tv_sec = max_secs;
5476 }
5477 else {
5478 absTime->tv_sec = secs;
5479 }
5480 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5481 }
5482 else {
5483 jlong secs = time / NANOSECS_PER_SEC;
5484 if (secs >= MAX_SECS) {
5485 absTime->tv_sec = max_secs;
5486 absTime->tv_nsec = 0;
5487 }
5488 else {
5489 absTime->tv_sec = now.tv_sec + secs;
5490 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5491 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5492 absTime->tv_nsec -= NANOSECS_PER_SEC;
5493 ++absTime->tv_sec; // note: this must be <= max_secs
5494 }
5495 }
5496 }
5497 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5498 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5499 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5500 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5501 }
5502
5503 void Parker::park(bool isAbsolute, jlong time) {
5504 // Ideally we'd do something useful while spinning, such
5505 // as calling unpackTime().
5506
5507 // Optional fast-path check:
5508 // Return immediately if a permit is available.
5509 // We depend on Atomic::xchg() having full barrier semantics
5510 // since we are doing a lock-free update to _counter.
5511 if (Atomic::xchg(0, &_counter) > 0) return;
5512
5513 Thread* thread = Thread::current();
5514 assert(thread->is_Java_thread(), "Must be JavaThread");
5515 JavaThread *jt = (JavaThread *)thread;
5516
5517 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5518 // Check interrupt before trying to wait
5519 if (Thread::is_interrupted(thread, false)) {
5520 return;
5521 }
5522
5523 // Next, demultiplex/decode time arguments
5524 timespec absTime;
5525 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5526 return;
5527 }
5528 if (time > 0) {
5529 unpackTime(&absTime, isAbsolute, time);
5530 }
5531
5532
5533 // Enter safepoint region
5534 // Beware of deadlocks such as 6317397.
5535 // The per-thread Parker:: mutex is a classic leaf-lock.
5536 // In particular a thread must never block on the Threads_lock while
5537 // holding the Parker:: mutex. If safepoints are pending both the
5538 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5539 ThreadBlockInVM tbivm(jt);
5540
5541 // Don't wait if cannot get lock since interference arises from
5542 // unblocking. Also. check interrupt before trying wait
5543 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5544 return;
5545 }
5546
5547 int status ;
5548 if (_counter > 0) { // no wait needed
5549 _counter = 0;
5550 status = pthread_mutex_unlock(_mutex);
5551 assert (status == 0, "invariant") ;
5552 // Paranoia to ensure our locked and lock-free paths interact
5553 // correctly with each other and Java-level accesses.
5554 OrderAccess::fence();
5555 return;
5556 }
5557
5558 #ifdef ASSERT
5559 // Don't catch signals while blocked; let the running threads have the signals.
5560 // (This allows a debugger to break into the running thread.)
5561 sigset_t oldsigs;
5562 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5563 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5564 #endif
5565
5566 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5567 jt->set_suspend_equivalent();
5568 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5569
5570 if (time == 0) {
5571 status = pthread_cond_wait (_cond, _mutex) ;
5572 } else {
5573 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5574 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5575 pthread_cond_destroy (_cond) ;
5576 pthread_cond_init (_cond, NULL);
5577 }
5578 }
5579 assert_status(status == 0 || status == EINTR ||
5580 status == ETIME || status == ETIMEDOUT,
5581 status, "cond_timedwait");
5582
5583 #ifdef ASSERT
5584 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5585 #endif
5586
5587 _counter = 0 ;
5588 status = pthread_mutex_unlock(_mutex) ;
5589 assert_status(status == 0, status, "invariant") ;
5590 // Paranoia to ensure our locked and lock-free paths interact
5591 // correctly with each other and Java-level accesses.
5592 OrderAccess::fence();
5593
5594 // If externally suspended while waiting, re-suspend
5595 if (jt->handle_special_suspend_equivalent_condition()) {
5596 jt->java_suspend_self();
5597 }
5598 }
5599
5600 void Parker::unpark() {
5601 int s, status ;
5602 status = pthread_mutex_lock(_mutex);
5603 assert (status == 0, "invariant") ;
5604 s = _counter;
5605 _counter = 1;
5606 if (s < 1) {
5607 if (WorkAroundNPTLTimedWaitHang) {
5608 status = pthread_cond_signal (_cond) ;
5609 assert (status == 0, "invariant") ;
5610 status = pthread_mutex_unlock(_mutex);
5611 assert (status == 0, "invariant") ;
5612 } else {
5613 status = pthread_mutex_unlock(_mutex);
5614 assert (status == 0, "invariant") ;
5615 status = pthread_cond_signal (_cond) ;
5616 assert (status == 0, "invariant") ;
5617 }
5618 } else {
5619 pthread_mutex_unlock(_mutex);
5620 assert (status == 0, "invariant") ;
5621 }
5622 }
5623
5624
5625 extern char** environ;
5626
5627 #ifndef __NR_fork
5628 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5629 #endif
5630
5631 #ifndef __NR_execve
5632 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5633 #endif
5634
5635 // Run the specified command in a separate process. Return its exit value,
5636 // or -1 on failure (e.g. can't fork a new process).
5637 // Unlike system(), this function can be called from signal handler. It
5638 // doesn't block SIGINT et al.
5639 int os::fork_and_exec(char* cmd) {
5640 const char * argv[4] = {"sh", "-c", cmd, NULL};
5641
5642 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5643 // pthread_atfork handlers and reset pthread library. All we need is a
5644 // separate process to execve. Make a direct syscall to fork process.
5645 // On IA64 there's no fork syscall, we have to use fork() and hope for
5646 // the best...
5647 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5648 IA64_ONLY(fork();)
5649
5650 if (pid < 0) {
5651 // fork failed
5652 return -1;
5653
5654 } else if (pid == 0) {
5655 // child process
5656
5657 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5658 // first to kill every thread on the thread list. Because this list is
5659 // not reset by fork() (see notes above), execve() will instead kill
5660 // every thread in the parent process. We know this is the only thread
5661 // in the new process, so make a system call directly.
5662 // IA64 should use normal execve() from glibc to match the glibc fork()
5663 // above.
5664 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5665 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5666
5667 // execve failed
5668 _exit(-1);
5669
5670 } else {
5671 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5672 // care about the actual exit code, for now.
5673
5674 int status;
5675
5676 // Wait for the child process to exit. This returns immediately if
5677 // the child has already exited. */
5678 while (waitpid(pid, &status, 0) < 0) {
5679 switch (errno) {
5680 case ECHILD: return 0;
5681 case EINTR: break;
5682 default: return -1;
5683 }
5684 }
5685
5686 if (WIFEXITED(status)) {
5687 // The child exited normally; get its exit code.
5688 return WEXITSTATUS(status);
5689 } else if (WIFSIGNALED(status)) {
5690 // The child exited because of a signal
5691 // The best value to return is 0x80 + signal number,
5692 // because that is what all Unix shells do, and because
5693 // it allows callers to distinguish between process exit and
5694 // process death by signal.
5695 return 0x80 + WTERMSIG(status);
5696 } else {
5697 // Unknown exit code; pass it through
5698 return status;
5699 }
5700 }
5701 }
5702
5703 // is_headless_jre()
5704 //
5705 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5706 // in order to report if we are running in a headless jre
5707 //
5708 // Since JDK8 xawt/libmawt.so was moved into the same directory
5709 // as libawt.so, and renamed libawt_xawt.so
5710 //
5711 bool os::is_headless_jre() {
5712 struct stat statbuf;
5713 char buf[MAXPATHLEN];
5714 char libmawtpath[MAXPATHLEN];
5715 const char *xawtstr = "/xawt/libmawt.so";
5716 const char *new_xawtstr = "/libawt_xawt.so";
5717 char *p;
5718
5719 // Get path to libjvm.so
5720 os::jvm_path(buf, sizeof(buf));
5721
5722 // Get rid of libjvm.so
5723 p = strrchr(buf, '/');
5724 if (p == NULL) return false;
5725 else *p = '\0';
5726
5727 // Get rid of client or server
5728 p = strrchr(buf, '/');
5729 if (p == NULL) return false;
5730 else *p = '\0';
5731
5732 // check xawt/libmawt.so
5733 strcpy(libmawtpath, buf);
5734 strcat(libmawtpath, xawtstr);
5735 if (::stat(libmawtpath, &statbuf) == 0) return false;
5736
5737 // check libawt_xawt.so
5738 strcpy(libmawtpath, buf);
5739 strcat(libmawtpath, new_xawtstr);
5740 if (::stat(libmawtpath, &statbuf) == 0) return false;
5741
5742 return true;
5743 }
5744
5745 // Get the default path to the core file
5746 // Returns the length of the string
5747 int os::get_core_path(char* buffer, size_t bufferSize) {
5748 const char* p = get_current_directory(buffer, bufferSize);
5749
5750 if (p == NULL) {
5751 assert(p != NULL, "failed to get current directory");
5752 return 0;
5753 }
5754
5755 return strlen(buffer);
5756 }
5757
5758 #ifdef JAVASE_EMBEDDED
5759 //
5760 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5761 //
5762 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5763
5764 // ctor
5765 //
5766 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5767 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5768 _fd = fd;
5769
5770 if (os::create_thread(this, os::os_thread)) {
5771 _memnotify_thread = this;
5772 os::set_priority(this, NearMaxPriority);
5773 os::start_thread(this);
5774 }
5775 }
5776
5777 // Where all the work gets done
5778 //
5779 void MemNotifyThread::run() {
5780 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5781
5782 // Set up the select arguments
5783 fd_set rfds;
5784 if (_fd != -1) {
5785 FD_ZERO(&rfds);
5786 FD_SET(_fd, &rfds);
5787 }
5788
5789 // Now wait for the mem_notify device to wake up
5790 while (1) {
5791 // Wait for the mem_notify device to signal us..
5792 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5793 if (rc == -1) {
5794 perror("select!\n");
5795 break;
5796 } else if (rc) {
5797 //ssize_t free_before = os::available_memory();
5798 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5799
5800 // The kernel is telling us there is not much memory left...
5801 // try to do something about that
5802
5803 // If we are not already in a GC, try one.
5804 if (!Universe::heap()->is_gc_active()) {
5805 Universe::heap()->collect(GCCause::_allocation_failure);
5806
5807 //ssize_t free_after = os::available_memory();
5808 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5809 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5810 }
5811 // We might want to do something like the following if we find the GC's are not helping...
5812 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5813 }
5814 }
5815 }
5816
5817 //
5818 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5819 //
5820 void MemNotifyThread::start() {
5821 int fd;
5822 fd = open ("/dev/mem_notify", O_RDONLY, 0);
5823 if (fd < 0) {
5824 return;
5825 }
5826
5827 if (memnotify_thread() == NULL) {
5828 new MemNotifyThread(fd);
5829 }
5830 }
5831
5832 #endif // JAVASE_EMBEDDED