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 #define MAX_PATH (2 * K)
105
106 // for timer info max values which include all bits
107 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
108
109 #define LARGEPAGES_BIT (1 << 6)
110 ////////////////////////////////////////////////////////////////////////////////
111 // global variables
112 julong os::Linux::_physical_memory = 0;
113
114 address os::Linux::_initial_thread_stack_bottom = NULL;
115 uintptr_t os::Linux::_initial_thread_stack_size = 0;
116
117 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
118 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
119 Mutex* os::Linux::_createThread_lock = NULL;
120 pthread_t os::Linux::_main_thread;
121 int os::Linux::_page_size = -1;
122 const int os::Linux::_vm_default_page_size = (8 * K);
123 bool os::Linux::_is_floating_stack = false;
124 bool os::Linux::_is_NPTL = false;
125 bool os::Linux::_supports_fast_thread_cpu_time = false;
126 const char * os::Linux::_glibc_version = NULL;
127 const char * os::Linux::_libpthread_version = NULL;
128
129 static jlong initial_time_count=0;
130
131 static int clock_tics_per_sec = 100;
132
133 // For diagnostics to print a message once. see run_periodic_checks
134 static sigset_t check_signal_done;
135 static bool check_signals = true;;
136
137 static pid_t _initial_pid = 0;
138
139 /* Signal number used to suspend/resume a thread */
140
141 /* do not use any signal number less than SIGSEGV, see 4355769 */
142 static int SR_signum = SIGUSR2;
143 sigset_t SR_sigset;
144
145 /* Used to protect dlsym() calls */
146 static pthread_mutex_t dl_mutex;
147
148 #ifdef JAVASE_EMBEDDED
149 class MemNotifyThread: public Thread {
150 friend class VMStructs;
151 public:
152 virtual void run();
153
154 private:
155 static MemNotifyThread* _memnotify_thread;
156 int _fd;
157
158 public:
159
160 // Constructor
161 MemNotifyThread(int fd);
162
163 // Tester
164 bool is_memnotify_thread() const { return true; }
165
166 // Printing
167 char* name() const { return (char*)"Linux MemNotify Thread"; }
168
169 // Returns the single instance of the MemNotifyThread
170 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
171
172 // Create and start the single instance of MemNotifyThread
173 static void start();
174 };
175 #endif // JAVASE_EMBEDDED
176
177 // utility functions
178
179 static int SR_initialize();
180
181 julong os::available_memory() {
182 return Linux::available_memory();
183 }
184
185 julong os::Linux::available_memory() {
186 // values in struct sysinfo are "unsigned long"
187 struct sysinfo si;
188 sysinfo(&si);
189
190 return (julong)si.freeram * si.mem_unit;
191 }
192
193 julong os::physical_memory() {
194 return Linux::physical_memory();
195 }
196
197 ////////////////////////////////////////////////////////////////////////////////
198 // environment support
199
200 bool os::getenv(const char* name, char* buf, int len) {
201 const char* val = ::getenv(name);
202 if (val != NULL && strlen(val) < (size_t)len) {
203 strcpy(buf, val);
204 return true;
205 }
206 if (len > 0) buf[0] = 0; // return a null string
207 return false;
208 }
209
210
211 // Return true if user is running as root.
212
213 bool os::have_special_privileges() {
214 static bool init = false;
215 static bool privileges = false;
216 if (!init) {
217 privileges = (getuid() != geteuid()) || (getgid() != getegid());
218 init = true;
219 }
220 return privileges;
221 }
222
223
224 #ifndef SYS_gettid
225 // i386: 224, ia64: 1105, amd64: 186, sparc 143
226 #ifdef __ia64__
227 #define SYS_gettid 1105
228 #elif __i386__
229 #define SYS_gettid 224
230 #elif __amd64__
231 #define SYS_gettid 186
232 #elif __sparc__
233 #define SYS_gettid 143
234 #else
235 #error define gettid for the arch
236 #endif
237 #endif
238
239 // Cpu architecture string
240 #if defined(ZERO)
241 static char cpu_arch[] = ZERO_LIBARCH;
242 #elif defined(IA64)
243 static char cpu_arch[] = "ia64";
244 #elif defined(IA32)
245 static char cpu_arch[] = "i386";
246 #elif defined(AMD64)
247 static char cpu_arch[] = "amd64";
248 #elif defined(ARM)
249 static char cpu_arch[] = "arm";
250 #elif defined(PPC)
251 static char cpu_arch[] = "ppc";
252 #elif defined(SPARC)
253 # ifdef _LP64
254 static char cpu_arch[] = "sparcv9";
255 # else
256 static char cpu_arch[] = "sparc";
257 # endif
258 #else
259 #error Add appropriate cpu_arch setting
260 #endif
261
262
263 // pid_t gettid()
264 //
265 // Returns the kernel thread id of the currently running thread. Kernel
266 // thread id is used to access /proc.
267 //
268 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
269 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
270 //
271 pid_t os::Linux::gettid() {
272 int rslt = syscall(SYS_gettid);
273 if (rslt == -1) {
274 // old kernel, no NPTL support
275 return getpid();
276 } else {
277 return (pid_t)rslt;
278 }
279 }
280
281 // Most versions of linux have a bug where the number of processors are
282 // determined by looking at the /proc file system. In a chroot environment,
283 // the system call returns 1. This causes the VM to act as if it is
284 // a single processor and elide locking (see is_MP() call).
285 static bool unsafe_chroot_detected = false;
286 static const char *unstable_chroot_error = "/proc file system not found.\n"
287 "Java may be unstable running multithreaded in a chroot "
288 "environment on Linux when /proc filesystem is not mounted.";
289
290 void os::Linux::initialize_system_info() {
291 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
292 if (processor_count() == 1) {
293 pid_t pid = os::Linux::gettid();
294 char fname[32];
295 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
296 FILE *fp = fopen(fname, "r");
297 if (fp == NULL) {
298 unsafe_chroot_detected = true;
299 } else {
300 fclose(fp);
301 }
302 }
303 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
304 assert(processor_count() > 0, "linux error");
305 }
306
307 void os::init_system_properties_values() {
308 // char arch[12];
309 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
310
311 // The next steps are taken in the product version:
312 //
313 // Obtain the JAVA_HOME value from the location of libjvm.so.
314 // This library should be located at:
315 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
316 //
317 // If "/jre/lib/" appears at the right place in the path, then we
318 // assume libjvm.so is installed in a JDK and we use this path.
319 //
320 // Otherwise exit with message: "Could not create the Java virtual machine."
321 //
322 // The following extra steps are taken in the debugging version:
323 //
324 // If "/jre/lib/" does NOT appear at the right place in the path
325 // instead of exit check for $JAVA_HOME environment variable.
326 //
327 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
328 // then we append a fake suffix "hotspot/libjvm.so" to this path so
329 // it looks like libjvm.so is installed there
330 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
331 //
332 // Otherwise exit.
333 //
334 // Important note: if the location of libjvm.so changes this
335 // code needs to be changed accordingly.
336
337 // The next few definitions allow the code to be verbatim:
338 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
339 #define getenv(n) ::getenv(n)
340
341 /*
342 * See ld(1):
343 * The linker uses the following search paths to locate required
344 * shared libraries:
345 * 1: ...
346 * ...
347 * 7: The default directories, normally /lib and /usr/lib.
348 */
349 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
350 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
351 #else
352 #define DEFAULT_LIBPATH "/lib:/usr/lib"
353 #endif
354
355 #define EXTENSIONS_DIR "/lib/ext"
356 #define ENDORSED_DIR "/lib/endorsed"
357 #define REG_DIR "/usr/java/packages"
358
359 {
360 /* sysclasspath, java_home, dll_dir */
361 {
362 char *home_path;
363 char *dll_path;
364 char *pslash;
365 char buf[MAXPATHLEN];
366 os::jvm_path(buf, sizeof(buf));
367
368 // Found the full path to libjvm.so.
369 // Now cut the path to <java_home>/jre if we can.
370 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
371 pslash = strrchr(buf, '/');
372 if (pslash != NULL)
373 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
374 dll_path = malloc(strlen(buf) + 1);
375 if (dll_path == NULL)
376 return;
377 strcpy(dll_path, buf);
378 Arguments::set_dll_dir(dll_path);
379
380 if (pslash != NULL) {
381 pslash = strrchr(buf, '/');
382 if (pslash != NULL) {
383 *pslash = '\0'; /* get rid of /<arch> */
384 pslash = strrchr(buf, '/');
385 if (pslash != NULL)
386 *pslash = '\0'; /* get rid of /lib */
387 }
388 }
389
390 home_path = malloc(strlen(buf) + 1);
391 if (home_path == NULL)
392 return;
393 strcpy(home_path, buf);
394 Arguments::set_java_home(home_path);
395
396 if (!set_boot_path('/', ':'))
397 return;
398 }
399
400 /*
401 * Where to look for native libraries
402 *
403 * Note: Due to a legacy implementation, most of the library path
404 * is set in the launcher. This was to accomodate linking restrictions
405 * on legacy Linux implementations (which are no longer supported).
406 * Eventually, all the library path setting will be done here.
407 *
408 * However, to prevent the proliferation of improperly built native
409 * libraries, the new path component /usr/java/packages is added here.
410 * Eventually, all the library path setting will be done here.
411 */
412 {
413 char *ld_library_path;
414
415 /*
416 * Construct the invariant part of ld_library_path. Note that the
417 * space for the colon and the trailing null are provided by the
418 * nulls included by the sizeof operator (so actually we allocate
419 * a byte more than necessary).
420 */
421 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
422 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
423 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
424
425 /*
426 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
427 * should always exist (until the legacy problem cited above is
428 * addressed).
429 */
430 char *v = getenv("LD_LIBRARY_PATH");
431 if (v != NULL) {
432 char *t = ld_library_path;
433 /* That's +1 for the colon and +1 for the trailing '\0' */
434 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
435 sprintf(ld_library_path, "%s:%s", v, t);
436 }
437 Arguments::set_library_path(ld_library_path);
438 }
439
440 /*
441 * Extensions directories.
442 *
443 * Note that the space for the colon and the trailing null are provided
444 * by the nulls included by the sizeof operator (so actually one byte more
445 * than necessary is allocated).
446 */
447 {
448 char *buf = malloc(strlen(Arguments::get_java_home()) +
449 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
450 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
451 Arguments::get_java_home());
452 Arguments::set_ext_dirs(buf);
453 }
454
455 /* Endorsed standards default directory. */
456 {
457 char * buf;
458 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
459 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
460 Arguments::set_endorsed_dirs(buf);
461 }
462 }
463
464 #undef malloc
465 #undef getenv
466 #undef EXTENSIONS_DIR
467 #undef ENDORSED_DIR
468
469 // Done
470 return;
471 }
472
473 ////////////////////////////////////////////////////////////////////////////////
474 // breakpoint support
475
476 void os::breakpoint() {
477 BREAKPOINT;
478 }
479
480 extern "C" void breakpoint() {
481 // use debugger to set breakpoint here
482 }
483
484 ////////////////////////////////////////////////////////////////////////////////
485 // signal support
486
487 debug_only(static bool signal_sets_initialized = false);
488 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
489
490 bool os::Linux::is_sig_ignored(int sig) {
491 struct sigaction oact;
492 sigaction(sig, (struct sigaction*)NULL, &oact);
493 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
494 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
495 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
496 return true;
497 else
498 return false;
499 }
500
501 void os::Linux::signal_sets_init() {
502 // Should also have an assertion stating we are still single-threaded.
503 assert(!signal_sets_initialized, "Already initialized");
504 // Fill in signals that are necessarily unblocked for all threads in
505 // the VM. Currently, we unblock the following signals:
506 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
507 // by -Xrs (=ReduceSignalUsage));
508 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
509 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
510 // the dispositions or masks wrt these signals.
511 // Programs embedding the VM that want to use the above signals for their
512 // own purposes must, at this time, use the "-Xrs" option to prevent
513 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
514 // (See bug 4345157, and other related bugs).
515 // In reality, though, unblocking these signals is really a nop, since
516 // these signals are not blocked by default.
517 sigemptyset(&unblocked_sigs);
518 sigemptyset(&allowdebug_blocked_sigs);
519 sigaddset(&unblocked_sigs, SIGILL);
520 sigaddset(&unblocked_sigs, SIGSEGV);
521 sigaddset(&unblocked_sigs, SIGBUS);
522 sigaddset(&unblocked_sigs, SIGFPE);
523 sigaddset(&unblocked_sigs, SR_signum);
524
525 if (!ReduceSignalUsage) {
526 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
527 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
528 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
529 }
530 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
531 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
532 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
533 }
534 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
535 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
536 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
537 }
538 }
539 // Fill in signals that are blocked by all but the VM thread.
540 sigemptyset(&vm_sigs);
541 if (!ReduceSignalUsage)
542 sigaddset(&vm_sigs, BREAK_SIGNAL);
543 debug_only(signal_sets_initialized = true);
544
545 }
546
547 // These are signals that are unblocked while a thread is running Java.
548 // (For some reason, they get blocked by default.)
549 sigset_t* os::Linux::unblocked_signals() {
550 assert(signal_sets_initialized, "Not initialized");
551 return &unblocked_sigs;
552 }
553
554 // These are the signals that are blocked while a (non-VM) thread is
555 // running Java. Only the VM thread handles these signals.
556 sigset_t* os::Linux::vm_signals() {
557 assert(signal_sets_initialized, "Not initialized");
558 return &vm_sigs;
559 }
560
561 // These are signals that are blocked during cond_wait to allow debugger in
562 sigset_t* os::Linux::allowdebug_blocked_signals() {
563 assert(signal_sets_initialized, "Not initialized");
564 return &allowdebug_blocked_sigs;
565 }
566
567 void os::Linux::hotspot_sigmask(Thread* thread) {
568
569 //Save caller's signal mask before setting VM signal mask
570 sigset_t caller_sigmask;
571 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
572
573 OSThread* osthread = thread->osthread();
574 osthread->set_caller_sigmask(caller_sigmask);
575
576 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
577
578 if (!ReduceSignalUsage) {
579 if (thread->is_VM_thread()) {
580 // Only the VM thread handles BREAK_SIGNAL ...
581 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
582 } else {
583 // ... all other threads block BREAK_SIGNAL
584 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
585 }
586 }
587 }
588
589 //////////////////////////////////////////////////////////////////////////////
590 // detecting pthread library
591
592 void os::Linux::libpthread_init() {
593 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
594 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
595 // generic name for earlier versions.
596 // Define macros here so we can build HotSpot on old systems.
597 # ifndef _CS_GNU_LIBC_VERSION
598 # define _CS_GNU_LIBC_VERSION 2
599 # endif
600 # ifndef _CS_GNU_LIBPTHREAD_VERSION
601 # define _CS_GNU_LIBPTHREAD_VERSION 3
602 # endif
603
604 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
605 if (n > 0) {
606 char *str = (char *)malloc(n, mtInternal);
607 confstr(_CS_GNU_LIBC_VERSION, str, n);
608 os::Linux::set_glibc_version(str);
609 } else {
610 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
611 static char _gnu_libc_version[32];
612 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
613 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
614 os::Linux::set_glibc_version(_gnu_libc_version);
615 }
616
617 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
618 if (n > 0) {
619 char *str = (char *)malloc(n, mtInternal);
620 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
621 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
622 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
623 // is the case. LinuxThreads has a hard limit on max number of threads.
624 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
625 // On the other hand, NPTL does not have such a limit, sysconf()
626 // will return -1 and errno is not changed. Check if it is really NPTL.
627 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
628 strstr(str, "NPTL") &&
629 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
630 free(str);
631 os::Linux::set_libpthread_version("linuxthreads");
632 } else {
633 os::Linux::set_libpthread_version(str);
634 }
635 } else {
636 // glibc before 2.3.2 only has LinuxThreads.
637 os::Linux::set_libpthread_version("linuxthreads");
638 }
639
640 if (strstr(libpthread_version(), "NPTL")) {
641 os::Linux::set_is_NPTL();
642 } else {
643 os::Linux::set_is_LinuxThreads();
644 }
645
646 // LinuxThreads have two flavors: floating-stack mode, which allows variable
647 // stack size; and fixed-stack mode. NPTL is always floating-stack.
648 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
649 os::Linux::set_is_floating_stack();
650 }
651 }
652
653 /////////////////////////////////////////////////////////////////////////////
654 // thread stack
655
656 // Force Linux kernel to expand current thread stack. If "bottom" is close
657 // to the stack guard, caller should block all signals.
658 //
659 // MAP_GROWSDOWN:
660 // A special mmap() flag that is used to implement thread stacks. It tells
661 // kernel that the memory region should extend downwards when needed. This
662 // allows early versions of LinuxThreads to only mmap the first few pages
663 // when creating a new thread. Linux kernel will automatically expand thread
664 // stack as needed (on page faults).
665 //
666 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
667 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
668 // region, it's hard to tell if the fault is due to a legitimate stack
669 // access or because of reading/writing non-exist memory (e.g. buffer
670 // overrun). As a rule, if the fault happens below current stack pointer,
671 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
672 // application (see Linux kernel fault.c).
673 //
674 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
675 // stack overflow detection.
676 //
677 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
678 // not use this flag. However, the stack of initial thread is not created
679 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
680 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
681 // and then attach the thread to JVM.
682 //
683 // To get around the problem and allow stack banging on Linux, we need to
684 // manually expand thread stack after receiving the SIGSEGV.
685 //
686 // There are two ways to expand thread stack to address "bottom", we used
687 // both of them in JVM before 1.5:
688 // 1. adjust stack pointer first so that it is below "bottom", and then
689 // touch "bottom"
690 // 2. mmap() the page in question
691 //
692 // Now alternate signal stack is gone, it's harder to use 2. For instance,
693 // if current sp is already near the lower end of page 101, and we need to
694 // call mmap() to map page 100, it is possible that part of the mmap() frame
695 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
696 // That will destroy the mmap() frame and cause VM to crash.
697 //
698 // The following code works by adjusting sp first, then accessing the "bottom"
699 // page to force a page fault. Linux kernel will then automatically expand the
700 // stack mapping.
701 //
702 // _expand_stack_to() assumes its frame size is less than page size, which
703 // should always be true if the function is not inlined.
704
705 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
706 #define NOINLINE
707 #else
708 #define NOINLINE __attribute__ ((noinline))
709 #endif
710
711 static void _expand_stack_to(address bottom) NOINLINE;
712
713 static void _expand_stack_to(address bottom) {
714 address sp;
715 size_t size;
716 volatile char *p;
717
718 // Adjust bottom to point to the largest address within the same page, it
719 // gives us a one-page buffer if alloca() allocates slightly more memory.
720 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
721 bottom += os::Linux::page_size() - 1;
722
723 // sp might be slightly above current stack pointer; if that's the case, we
724 // will alloca() a little more space than necessary, which is OK. Don't use
725 // os::current_stack_pointer(), as its result can be slightly below current
726 // stack pointer, causing us to not alloca enough to reach "bottom".
727 sp = (address)&sp;
728
729 if (sp > bottom) {
730 size = sp - bottom;
731 p = (volatile char *)alloca(size);
732 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
733 p[0] = '\0';
734 }
735 }
736
737 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
738 assert(t!=NULL, "just checking");
739 assert(t->osthread()->expanding_stack(), "expand should be set");
740 assert(t->stack_base() != NULL, "stack_base was not initialized");
741
742 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
743 sigset_t mask_all, old_sigset;
744 sigfillset(&mask_all);
745 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
746 _expand_stack_to(addr);
747 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
748 return true;
749 }
750 return false;
751 }
752
753 //////////////////////////////////////////////////////////////////////////////
754 // create new thread
755
756 static address highest_vm_reserved_address();
757
758 // check if it's safe to start a new thread
759 static bool _thread_safety_check(Thread* thread) {
760 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
761 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
762 // Heap is mmap'ed at lower end of memory space. Thread stacks are
763 // allocated (MAP_FIXED) from high address space. Every thread stack
764 // occupies a fixed size slot (usually 2Mbytes, but user can change
765 // it to other values if they rebuild LinuxThreads).
766 //
767 // Problem with MAP_FIXED is that mmap() can still succeed even part of
768 // the memory region has already been mmap'ed. That means if we have too
769 // many threads and/or very large heap, eventually thread stack will
770 // collide with heap.
771 //
772 // Here we try to prevent heap/stack collision by comparing current
773 // stack bottom with the highest address that has been mmap'ed by JVM
774 // plus a safety margin for memory maps created by native code.
775 //
776 // This feature can be disabled by setting ThreadSafetyMargin to 0
777 //
778 if (ThreadSafetyMargin > 0) {
779 address stack_bottom = os::current_stack_base() - os::current_stack_size();
780
781 // not safe if our stack extends below the safety margin
782 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
783 } else {
784 return true;
785 }
786 } else {
787 // Floating stack LinuxThreads or NPTL:
788 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
789 // there's not enough space left, pthread_create() will fail. If we come
790 // here, that means enough space has been reserved for stack.
791 return true;
792 }
793 }
794
795 // Thread start routine for all newly created threads
796 static void *java_start(Thread *thread) {
797 // Try to randomize the cache line index of hot stack frames.
798 // This helps when threads of the same stack traces evict each other's
799 // cache lines. The threads can be either from the same JVM instance, or
800 // from different JVM instances. The benefit is especially true for
801 // processors with hyperthreading technology.
802 static int counter = 0;
803 int pid = os::current_process_id();
804 alloca(((pid ^ counter++) & 7) * 128);
805
806 ThreadLocalStorage::set_thread(thread);
807
808 OSThread* osthread = thread->osthread();
809 Monitor* sync = osthread->startThread_lock();
810
811 // non floating stack LinuxThreads needs extra check, see above
812 if (!_thread_safety_check(thread)) {
813 // notify parent thread
814 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
815 osthread->set_state(ZOMBIE);
816 sync->notify_all();
817 return NULL;
818 }
819
820 // thread_id is kernel thread id (similar to Solaris LWP id)
821 osthread->set_thread_id(os::Linux::gettid());
822
823 if (UseNUMA) {
824 int lgrp_id = os::numa_get_group_id();
825 if (lgrp_id != -1) {
826 thread->set_lgrp_id(lgrp_id);
827 }
828 }
829 // initialize signal mask for this thread
830 os::Linux::hotspot_sigmask(thread);
831
832 // initialize floating point control register
833 os::Linux::init_thread_fpu_state();
834
835 // handshaking with parent thread
836 {
837 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
838
839 // notify parent thread
840 osthread->set_state(INITIALIZED);
841 sync->notify_all();
842
843 // wait until os::start_thread()
844 while (osthread->get_state() == INITIALIZED) {
845 sync->wait(Mutex::_no_safepoint_check_flag);
846 }
847 }
848
849 // call one more level start routine
850 thread->run();
851
852 return 0;
853 }
854
855 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
856 assert(thread->osthread() == NULL, "caller responsible");
857
858 // Allocate the OSThread object
859 OSThread* osthread = new OSThread(NULL, NULL);
860 if (osthread == NULL) {
861 return false;
862 }
863
864 // set the correct thread state
865 osthread->set_thread_type(thr_type);
866
867 // Initial state is ALLOCATED but not INITIALIZED
868 osthread->set_state(ALLOCATED);
869
870 thread->set_osthread(osthread);
871
872 // init thread attributes
873 pthread_attr_t attr;
874 pthread_attr_init(&attr);
875 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
876
877 // stack size
878 if (os::Linux::supports_variable_stack_size()) {
879 // calculate stack size if it's not specified by caller
880 if (stack_size == 0) {
881 stack_size = os::Linux::default_stack_size(thr_type);
882
883 switch (thr_type) {
884 case os::java_thread:
885 // Java threads use ThreadStackSize which default value can be
886 // changed with the flag -Xss
887 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
888 stack_size = JavaThread::stack_size_at_create();
889 break;
890 case os::compiler_thread:
891 if (CompilerThreadStackSize > 0) {
892 stack_size = (size_t)(CompilerThreadStackSize * K);
893 break;
894 } // else fall through:
895 // use VMThreadStackSize if CompilerThreadStackSize is not defined
896 case os::vm_thread:
897 case os::pgc_thread:
898 case os::cgc_thread:
899 case os::watcher_thread:
900 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
901 break;
902 }
903 }
904
905 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
906 pthread_attr_setstacksize(&attr, stack_size);
907 } else {
908 // let pthread_create() pick the default value.
909 }
910
911 // glibc guard page
912 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
913
914 ThreadState state;
915
916 {
917 // Serialize thread creation if we are running with fixed stack LinuxThreads
918 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
919 if (lock) {
920 os::Linux::createThread_lock()->lock_without_safepoint_check();
921 }
922
923 pthread_t tid;
924 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
925
926 pthread_attr_destroy(&attr);
927
928 if (ret != 0) {
929 if (PrintMiscellaneous && (Verbose || WizardMode)) {
930 perror("pthread_create()");
931 }
932 // Need to clean up stuff we've allocated so far
933 thread->set_osthread(NULL);
934 delete osthread;
935 if (lock) os::Linux::createThread_lock()->unlock();
936 return false;
937 }
938
939 // Store pthread info into the OSThread
940 osthread->set_pthread_id(tid);
941
942 // Wait until child thread is either initialized or aborted
943 {
944 Monitor* sync_with_child = osthread->startThread_lock();
945 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
946 while ((state = osthread->get_state()) == ALLOCATED) {
947 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
948 }
949 }
950
951 if (lock) {
952 os::Linux::createThread_lock()->unlock();
953 }
954 }
955
956 // Aborted due to thread limit being reached
957 if (state == ZOMBIE) {
958 thread->set_osthread(NULL);
959 delete osthread;
960 return false;
961 }
962
963 // The thread is returned suspended (in state INITIALIZED),
964 // and is started higher up in the call chain
965 assert(state == INITIALIZED, "race condition");
966 return true;
967 }
968
969 /////////////////////////////////////////////////////////////////////////////
970 // attach existing thread
971
972 // bootstrap the main thread
973 bool os::create_main_thread(JavaThread* thread) {
974 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
975 return create_attached_thread(thread);
976 }
977
978 bool os::create_attached_thread(JavaThread* thread) {
979 #ifdef ASSERT
980 thread->verify_not_published();
981 #endif
982
983 // Allocate the OSThread object
984 OSThread* osthread = new OSThread(NULL, NULL);
985
986 if (osthread == NULL) {
987 return false;
988 }
989
990 // Store pthread info into the OSThread
991 osthread->set_thread_id(os::Linux::gettid());
992 osthread->set_pthread_id(::pthread_self());
993
994 // initialize floating point control register
995 os::Linux::init_thread_fpu_state();
996
997 // Initial thread state is RUNNABLE
998 osthread->set_state(RUNNABLE);
999
1000 thread->set_osthread(osthread);
1001
1002 if (UseNUMA) {
1003 int lgrp_id = os::numa_get_group_id();
1004 if (lgrp_id != -1) {
1005 thread->set_lgrp_id(lgrp_id);
1006 }
1007 }
1008
1009 if (os::Linux::is_initial_thread()) {
1010 // If current thread is initial thread, its stack is mapped on demand,
1011 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1012 // the entire stack region to avoid SEGV in stack banging.
1013 // It is also useful to get around the heap-stack-gap problem on SuSE
1014 // kernel (see 4821821 for details). We first expand stack to the top
1015 // of yellow zone, then enable stack yellow zone (order is significant,
1016 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1017 // is no gap between the last two virtual memory regions.
1018
1019 JavaThread *jt = (JavaThread *)thread;
1020 address addr = jt->stack_yellow_zone_base();
1021 assert(addr != NULL, "initialization problem?");
1022 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1023
1024 osthread->set_expanding_stack();
1025 os::Linux::manually_expand_stack(jt, addr);
1026 osthread->clear_expanding_stack();
1027 }
1028
1029 // initialize signal mask for this thread
1030 // and save the caller's signal mask
1031 os::Linux::hotspot_sigmask(thread);
1032
1033 return true;
1034 }
1035
1036 void os::pd_start_thread(Thread* thread) {
1037 OSThread * osthread = thread->osthread();
1038 assert(osthread->get_state() != INITIALIZED, "just checking");
1039 Monitor* sync_with_child = osthread->startThread_lock();
1040 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1041 sync_with_child->notify();
1042 }
1043
1044 // Free Linux resources related to the OSThread
1045 void os::free_thread(OSThread* osthread) {
1046 assert(osthread != NULL, "osthread not set");
1047
1048 if (Thread::current()->osthread() == osthread) {
1049 // Restore caller's signal mask
1050 sigset_t sigmask = osthread->caller_sigmask();
1051 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1052 }
1053
1054 delete osthread;
1055 }
1056
1057 //////////////////////////////////////////////////////////////////////////////
1058 // thread local storage
1059
1060 int os::allocate_thread_local_storage() {
1061 pthread_key_t key;
1062 int rslt = pthread_key_create(&key, NULL);
1063 assert(rslt == 0, "cannot allocate thread local storage");
1064 return (int)key;
1065 }
1066
1067 // Note: This is currently not used by VM, as we don't destroy TLS key
1068 // on VM exit.
1069 void os::free_thread_local_storage(int index) {
1070 int rslt = pthread_key_delete((pthread_key_t)index);
1071 assert(rslt == 0, "invalid index");
1072 }
1073
1074 void os::thread_local_storage_at_put(int index, void* value) {
1075 int rslt = pthread_setspecific((pthread_key_t)index, value);
1076 assert(rslt == 0, "pthread_setspecific failed");
1077 }
1078
1079 extern "C" Thread* get_thread() {
1080 return ThreadLocalStorage::thread();
1081 }
1082
1083 //////////////////////////////////////////////////////////////////////////////
1084 // initial thread
1085
1086 // Check if current thread is the initial thread, similar to Solaris thr_main.
1087 bool os::Linux::is_initial_thread(void) {
1088 char dummy;
1089 // If called before init complete, thread stack bottom will be null.
1090 // Can be called if fatal error occurs before initialization.
1091 if (initial_thread_stack_bottom() == NULL) return false;
1092 assert(initial_thread_stack_bottom() != NULL &&
1093 initial_thread_stack_size() != 0,
1094 "os::init did not locate initial thread's stack region");
1095 if ((address)&dummy >= initial_thread_stack_bottom() &&
1096 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1097 return true;
1098 else return false;
1099 }
1100
1101 // Find the virtual memory area that contains addr
1102 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1103 FILE *fp = fopen("/proc/self/maps", "r");
1104 if (fp) {
1105 address low, high;
1106 while (!feof(fp)) {
1107 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1108 if (low <= addr && addr < high) {
1109 if (vma_low) *vma_low = low;
1110 if (vma_high) *vma_high = high;
1111 fclose (fp);
1112 return true;
1113 }
1114 }
1115 for (;;) {
1116 int ch = fgetc(fp);
1117 if (ch == EOF || ch == (int)'\n') break;
1118 }
1119 }
1120 fclose(fp);
1121 }
1122 return false;
1123 }
1124
1125 // Locate initial thread stack. This special handling of initial thread stack
1126 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1127 // bogus value for initial thread.
1128 void os::Linux::capture_initial_stack(size_t max_size) {
1129 // stack size is the easy part, get it from RLIMIT_STACK
1130 size_t stack_size;
1131 struct rlimit rlim;
1132 getrlimit(RLIMIT_STACK, &rlim);
1133 stack_size = rlim.rlim_cur;
1134
1135 // 6308388: a bug in ld.so will relocate its own .data section to the
1136 // lower end of primordial stack; reduce ulimit -s value a little bit
1137 // so we won't install guard page on ld.so's data section.
1138 stack_size -= 2 * page_size();
1139
1140 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1141 // 7.1, in both cases we will get 2G in return value.
1142 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1143 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1144 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1145 // in case other parts in glibc still assumes 2M max stack size.
1146 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1147 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1148 if (stack_size > 2 * K * K IA64_ONLY(*2))
1149 stack_size = 2 * K * K IA64_ONLY(*2);
1150 // Try to figure out where the stack base (top) is. This is harder.
1151 //
1152 // When an application is started, glibc saves the initial stack pointer in
1153 // a global variable "__libc_stack_end", which is then used by system
1154 // libraries. __libc_stack_end should be pretty close to stack top. The
1155 // variable is available since the very early days. However, because it is
1156 // a private interface, it could disappear in the future.
1157 //
1158 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1159 // to __libc_stack_end, it is very close to stack top, but isn't the real
1160 // stack top. Note that /proc may not exist if VM is running as a chroot
1161 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1162 // /proc/<pid>/stat could change in the future (though unlikely).
1163 //
1164 // We try __libc_stack_end first. If that doesn't work, look for
1165 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1166 // as a hint, which should work well in most cases.
1167
1168 uintptr_t stack_start;
1169
1170 // try __libc_stack_end first
1171 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1172 if (p && *p) {
1173 stack_start = *p;
1174 } else {
1175 // see if we can get the start_stack field from /proc/self/stat
1176 FILE *fp;
1177 int pid;
1178 char state;
1179 int ppid;
1180 int pgrp;
1181 int session;
1182 int nr;
1183 int tpgrp;
1184 unsigned long flags;
1185 unsigned long minflt;
1186 unsigned long cminflt;
1187 unsigned long majflt;
1188 unsigned long cmajflt;
1189 unsigned long utime;
1190 unsigned long stime;
1191 long cutime;
1192 long cstime;
1193 long prio;
1194 long nice;
1195 long junk;
1196 long it_real;
1197 uintptr_t start;
1198 uintptr_t vsize;
1199 intptr_t rss;
1200 uintptr_t rsslim;
1201 uintptr_t scodes;
1202 uintptr_t ecode;
1203 int i;
1204
1205 // Figure what the primordial thread stack base is. Code is inspired
1206 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1207 // followed by command name surrounded by parentheses, state, etc.
1208 char stat[2048];
1209 int statlen;
1210
1211 fp = fopen("/proc/self/stat", "r");
1212 if (fp) {
1213 statlen = fread(stat, 1, 2047, fp);
1214 stat[statlen] = '\0';
1215 fclose(fp);
1216
1217 // Skip pid and the command string. Note that we could be dealing with
1218 // weird command names, e.g. user could decide to rename java launcher
1219 // to "java 1.4.2 :)", then the stat file would look like
1220 // 1234 (java 1.4.2 :)) R ... ...
1221 // We don't really need to know the command string, just find the last
1222 // occurrence of ")" and then start parsing from there. See bug 4726580.
1223 char * s = strrchr(stat, ')');
1224
1225 i = 0;
1226 if (s) {
1227 // Skip blank chars
1228 do s++; while (isspace(*s));
1229
1230 #define _UFM UINTX_FORMAT
1231 #define _DFM INTX_FORMAT
1232
1233 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1234 /* 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 */
1235 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,
1236 &state, /* 3 %c */
1237 &ppid, /* 4 %d */
1238 &pgrp, /* 5 %d */
1239 &session, /* 6 %d */
1240 &nr, /* 7 %d */
1241 &tpgrp, /* 8 %d */
1242 &flags, /* 9 %lu */
1243 &minflt, /* 10 %lu */
1244 &cminflt, /* 11 %lu */
1245 &majflt, /* 12 %lu */
1246 &cmajflt, /* 13 %lu */
1247 &utime, /* 14 %lu */
1248 &stime, /* 15 %lu */
1249 &cutime, /* 16 %ld */
1250 &cstime, /* 17 %ld */
1251 &prio, /* 18 %ld */
1252 &nice, /* 19 %ld */
1253 &junk, /* 20 %ld */
1254 &it_real, /* 21 %ld */
1255 &start, /* 22 UINTX_FORMAT */
1256 &vsize, /* 23 UINTX_FORMAT */
1257 &rss, /* 24 INTX_FORMAT */
1258 &rsslim, /* 25 UINTX_FORMAT */
1259 &scodes, /* 26 UINTX_FORMAT */
1260 &ecode, /* 27 UINTX_FORMAT */
1261 &stack_start); /* 28 UINTX_FORMAT */
1262 }
1263
1264 #undef _UFM
1265 #undef _DFM
1266
1267 if (i != 28 - 2) {
1268 assert(false, "Bad conversion from /proc/self/stat");
1269 // product mode - assume we are the initial thread, good luck in the
1270 // embedded case.
1271 warning("Can't detect initial thread stack location - bad conversion");
1272 stack_start = (uintptr_t) &rlim;
1273 }
1274 } else {
1275 // For some reason we can't open /proc/self/stat (for example, running on
1276 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1277 // most cases, so don't abort:
1278 warning("Can't detect initial thread stack location - no /proc/self/stat");
1279 stack_start = (uintptr_t) &rlim;
1280 }
1281 }
1282
1283 // Now we have a pointer (stack_start) very close to the stack top, the
1284 // next thing to do is to figure out the exact location of stack top. We
1285 // can find out the virtual memory area that contains stack_start by
1286 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1287 // and its upper limit is the real stack top. (again, this would fail if
1288 // running inside chroot, because /proc may not exist.)
1289
1290 uintptr_t stack_top;
1291 address low, high;
1292 if (find_vma((address)stack_start, &low, &high)) {
1293 // success, "high" is the true stack top. (ignore "low", because initial
1294 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1295 stack_top = (uintptr_t)high;
1296 } else {
1297 // failed, likely because /proc/self/maps does not exist
1298 warning("Can't detect initial thread stack location - find_vma failed");
1299 // best effort: stack_start is normally within a few pages below the real
1300 // stack top, use it as stack top, and reduce stack size so we won't put
1301 // guard page outside stack.
1302 stack_top = stack_start;
1303 stack_size -= 16 * page_size();
1304 }
1305
1306 // stack_top could be partially down the page so align it
1307 stack_top = align_size_up(stack_top, page_size());
1308
1309 if (max_size && stack_size > max_size) {
1310 _initial_thread_stack_size = max_size;
1311 } else {
1312 _initial_thread_stack_size = stack_size;
1313 }
1314
1315 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1316 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1317 }
1318
1319 ////////////////////////////////////////////////////////////////////////////////
1320 // time support
1321
1322 // Time since start-up in seconds to a fine granularity.
1323 // Used by VMSelfDestructTimer and the MemProfiler.
1324 double os::elapsedTime() {
1325
1326 return (double)(os::elapsed_counter()) * 0.000001;
1327 }
1328
1329 jlong os::elapsed_counter() {
1330 timeval time;
1331 int status = gettimeofday(&time, NULL);
1332 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1333 }
1334
1335 jlong os::elapsed_frequency() {
1336 return (1000 * 1000);
1337 }
1338
1339 // For now, we say that linux does not support vtime. I have no idea
1340 // whether it can actually be made to (DLD, 9/13/05).
1341
1342 bool os::supports_vtime() { return false; }
1343 bool os::enable_vtime() { return false; }
1344 bool os::vtime_enabled() { return false; }
1345 double os::elapsedVTime() {
1346 // better than nothing, but not much
1347 return elapsedTime();
1348 }
1349
1350 jlong os::javaTimeMillis() {
1351 timeval time;
1352 int status = gettimeofday(&time, NULL);
1353 assert(status != -1, "linux error");
1354 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1355 }
1356
1357 #ifndef CLOCK_MONOTONIC
1358 #define CLOCK_MONOTONIC (1)
1359 #endif
1360
1361 void os::Linux::clock_init() {
1362 // we do dlopen's in this particular order due to bug in linux
1363 // dynamical loader (see 6348968) leading to crash on exit
1364 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1365 if (handle == NULL) {
1366 handle = dlopen("librt.so", RTLD_LAZY);
1367 }
1368
1369 if (handle) {
1370 int (*clock_getres_func)(clockid_t, struct timespec*) =
1371 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1372 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1373 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1374 if (clock_getres_func && clock_gettime_func) {
1375 // See if monotonic clock is supported by the kernel. Note that some
1376 // early implementations simply return kernel jiffies (updated every
1377 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1378 // for nano time (though the monotonic property is still nice to have).
1379 // It's fixed in newer kernels, however clock_getres() still returns
1380 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1381 // resolution for now. Hopefully as people move to new kernels, this
1382 // won't be a problem.
1383 struct timespec res;
1384 struct timespec tp;
1385 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1386 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1387 // yes, monotonic clock is supported
1388 _clock_gettime = clock_gettime_func;
1389 } else {
1390 // close librt if there is no monotonic clock
1391 dlclose(handle);
1392 }
1393 }
1394 }
1395 }
1396
1397 #ifndef SYS_clock_getres
1398
1399 #if defined(IA32) || defined(AMD64)
1400 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1401 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1402 #else
1403 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1404 #define sys_clock_getres(x,y) -1
1405 #endif
1406
1407 #else
1408 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1409 #endif
1410
1411 void os::Linux::fast_thread_clock_init() {
1412 if (!UseLinuxPosixThreadCPUClocks) {
1413 return;
1414 }
1415 clockid_t clockid;
1416 struct timespec tp;
1417 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1418 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1419
1420 // Switch to using fast clocks for thread cpu time if
1421 // the sys_clock_getres() returns 0 error code.
1422 // Note, that some kernels may support the current thread
1423 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1424 // returned by the pthread_getcpuclockid().
1425 // If the fast Posix clocks are supported then the sys_clock_getres()
1426 // must return at least tp.tv_sec == 0 which means a resolution
1427 // better than 1 sec. This is extra check for reliability.
1428
1429 if(pthread_getcpuclockid_func &&
1430 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1431 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1432
1433 _supports_fast_thread_cpu_time = true;
1434 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1435 }
1436 }
1437
1438 jlong os::javaTimeNanos() {
1439 if (Linux::supports_monotonic_clock()) {
1440 struct timespec tp;
1441 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1442 assert(status == 0, "gettime error");
1443 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1444 return result;
1445 } else {
1446 timeval time;
1447 int status = gettimeofday(&time, NULL);
1448 assert(status != -1, "linux error");
1449 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1450 return 1000 * usecs;
1451 }
1452 }
1453
1454 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1455 if (Linux::supports_monotonic_clock()) {
1456 info_ptr->max_value = ALL_64_BITS;
1457
1458 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1459 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1460 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1461 } else {
1462 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1463 info_ptr->max_value = ALL_64_BITS;
1464
1465 // gettimeofday is a real time clock so it skips
1466 info_ptr->may_skip_backward = true;
1467 info_ptr->may_skip_forward = true;
1468 }
1469
1470 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1471 }
1472
1473 // Return the real, user, and system times in seconds from an
1474 // arbitrary fixed point in the past.
1475 bool os::getTimesSecs(double* process_real_time,
1476 double* process_user_time,
1477 double* process_system_time) {
1478 struct tms ticks;
1479 clock_t real_ticks = times(&ticks);
1480
1481 if (real_ticks == (clock_t) (-1)) {
1482 return false;
1483 } else {
1484 double ticks_per_second = (double) clock_tics_per_sec;
1485 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1486 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1487 *process_real_time = ((double) real_ticks) / ticks_per_second;
1488
1489 return true;
1490 }
1491 }
1492
1493
1494 char * os::local_time_string(char *buf, size_t buflen) {
1495 struct tm t;
1496 time_t long_time;
1497 time(&long_time);
1498 localtime_r(&long_time, &t);
1499 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1500 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1501 t.tm_hour, t.tm_min, t.tm_sec);
1502 return buf;
1503 }
1504
1505 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1506 return localtime_r(clock, res);
1507 }
1508
1509 ////////////////////////////////////////////////////////////////////////////////
1510 // runtime exit support
1511
1512 // Note: os::shutdown() might be called very early during initialization, or
1513 // called from signal handler. Before adding something to os::shutdown(), make
1514 // sure it is async-safe and can handle partially initialized VM.
1515 void os::shutdown() {
1516
1517 // allow PerfMemory to attempt cleanup of any persistent resources
1518 perfMemory_exit();
1519
1520 // needs to remove object in file system
1521 AttachListener::abort();
1522
1523 // flush buffered output, finish log files
1524 ostream_abort();
1525
1526 // Check for abort hook
1527 abort_hook_t abort_hook = Arguments::abort_hook();
1528 if (abort_hook != NULL) {
1529 abort_hook();
1530 }
1531
1532 }
1533
1534 // Note: os::abort() might be called very early during initialization, or
1535 // called from signal handler. Before adding something to os::abort(), make
1536 // sure it is async-safe and can handle partially initialized VM.
1537 void os::abort(bool dump_core) {
1538 os::shutdown();
1539 if (dump_core) {
1540 #ifndef PRODUCT
1541 fdStream out(defaultStream::output_fd());
1542 out.print_raw("Current thread is ");
1543 char buf[16];
1544 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1545 out.print_raw_cr(buf);
1546 out.print_raw_cr("Dumping core ...");
1547 #endif
1548 ::abort(); // dump core
1549 }
1550
1551 ::exit(1);
1552 }
1553
1554 // Die immediately, no exit hook, no abort hook, no cleanup.
1555 void os::die() {
1556 // _exit() on LinuxThreads only kills current thread
1557 ::abort();
1558 }
1559
1560 // unused on linux for now.
1561 void os::set_error_file(const char *logfile) {}
1562
1563
1564 // This method is a copy of JDK's sysGetLastErrorString
1565 // from src/solaris/hpi/src/system_md.c
1566
1567 size_t os::lasterror(char *buf, size_t len) {
1568
1569 if (errno == 0) return 0;
1570
1571 const char *s = ::strerror(errno);
1572 size_t n = ::strlen(s);
1573 if (n >= len) {
1574 n = len - 1;
1575 }
1576 ::strncpy(buf, s, n);
1577 buf[n] = '\0';
1578 return n;
1579 }
1580
1581 intx os::current_thread_id() { return (intx)pthread_self(); }
1582 int os::current_process_id() {
1583
1584 // Under the old linux thread library, linux gives each thread
1585 // its own process id. Because of this each thread will return
1586 // a different pid if this method were to return the result
1587 // of getpid(2). Linux provides no api that returns the pid
1588 // of the launcher thread for the vm. This implementation
1589 // returns a unique pid, the pid of the launcher thread
1590 // that starts the vm 'process'.
1591
1592 // Under the NPTL, getpid() returns the same pid as the
1593 // launcher thread rather than a unique pid per thread.
1594 // Use gettid() if you want the old pre NPTL behaviour.
1595
1596 // if you are looking for the result of a call to getpid() that
1597 // returns a unique pid for the calling thread, then look at the
1598 // OSThread::thread_id() method in osThread_linux.hpp file
1599
1600 return (int)(_initial_pid ? _initial_pid : getpid());
1601 }
1602
1603 // DLL functions
1604
1605 const char* os::dll_file_extension() { return ".so"; }
1606
1607 // This must be hard coded because it's the system's temporary
1608 // directory not the java application's temp directory, ala java.io.tmpdir.
1609 const char* os::get_temp_directory() { return "/tmp"; }
1610
1611 static bool file_exists(const char* filename) {
1612 struct stat statbuf;
1613 if (filename == NULL || strlen(filename) == 0) {
1614 return false;
1615 }
1616 return os::stat(filename, &statbuf) == 0;
1617 }
1618
1619 bool os::dll_build_name(char* buffer, size_t buflen,
1620 const char* pname, const char* fname) {
1621 bool retval = false;
1622 // Copied from libhpi
1623 const size_t pnamelen = pname ? strlen(pname) : 0;
1624
1625 // Return error on buffer overflow.
1626 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1627 return retval;
1628 }
1629
1630 if (pnamelen == 0) {
1631 snprintf(buffer, buflen, "lib%s.so", fname);
1632 retval = true;
1633 } else if (strchr(pname, *os::path_separator()) != NULL) {
1634 int n;
1635 char** pelements = split_path(pname, &n);
1636 if (pelements == NULL) {
1637 return false;
1638 }
1639 for (int i = 0 ; i < n ; i++) {
1640 // Really shouldn't be NULL, but check can't hurt
1641 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1642 continue; // skip the empty path values
1643 }
1644 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1645 if (file_exists(buffer)) {
1646 retval = true;
1647 break;
1648 }
1649 }
1650 // release the storage
1651 for (int i = 0 ; i < n ; i++) {
1652 if (pelements[i] != NULL) {
1653 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1654 }
1655 }
1656 if (pelements != NULL) {
1657 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1658 }
1659 } else {
1660 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1661 retval = true;
1662 }
1663 return retval;
1664 }
1665
1666 // check if addr is inside libjvm.so
1667 bool os::address_is_in_vm(address addr) {
1668 static address libjvm_base_addr;
1669 Dl_info dlinfo;
1670
1671 if (libjvm_base_addr == NULL) {
1672 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1673 libjvm_base_addr = (address)dlinfo.dli_fbase;
1674 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1675 }
1676
1677 if (dladdr((void *)addr, &dlinfo)) {
1678 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1679 }
1680
1681 return false;
1682 }
1683
1684 bool os::dll_address_to_function_name(address addr, char *buf,
1685 int buflen, int *offset) {
1686 Dl_info dlinfo;
1687
1688 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1689 if (buf != NULL) {
1690 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1691 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1692 }
1693 }
1694 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1695 return true;
1696 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1697 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1698 buf, buflen, offset, dlinfo.dli_fname)) {
1699 return true;
1700 }
1701 }
1702
1703 if (buf != NULL) buf[0] = '\0';
1704 if (offset != NULL) *offset = -1;
1705 return false;
1706 }
1707
1708 struct _address_to_library_name {
1709 address addr; // input : memory address
1710 size_t buflen; // size of fname
1711 char* fname; // output: library name
1712 address base; // library base addr
1713 };
1714
1715 static int address_to_library_name_callback(struct dl_phdr_info *info,
1716 size_t size, void *data) {
1717 int i;
1718 bool found = false;
1719 address libbase = NULL;
1720 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1721
1722 // iterate through all loadable segments
1723 for (i = 0; i < info->dlpi_phnum; i++) {
1724 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1725 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1726 // base address of a library is the lowest address of its loaded
1727 // segments.
1728 if (libbase == NULL || libbase > segbase) {
1729 libbase = segbase;
1730 }
1731 // see if 'addr' is within current segment
1732 if (segbase <= d->addr &&
1733 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1734 found = true;
1735 }
1736 }
1737 }
1738
1739 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1740 // so dll_address_to_library_name() can fall through to use dladdr() which
1741 // can figure out executable name from argv[0].
1742 if (found && info->dlpi_name && info->dlpi_name[0]) {
1743 d->base = libbase;
1744 if (d->fname) {
1745 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1746 }
1747 return 1;
1748 }
1749 return 0;
1750 }
1751
1752 bool os::dll_address_to_library_name(address addr, char* buf,
1753 int buflen, int* offset) {
1754 Dl_info dlinfo;
1755 struct _address_to_library_name data;
1756
1757 // There is a bug in old glibc dladdr() implementation that it could resolve
1758 // to wrong library name if the .so file has a base address != NULL. Here
1759 // we iterate through the program headers of all loaded libraries to find
1760 // out which library 'addr' really belongs to. This workaround can be
1761 // removed once the minimum requirement for glibc is moved to 2.3.x.
1762 data.addr = addr;
1763 data.fname = buf;
1764 data.buflen = buflen;
1765 data.base = NULL;
1766 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1767
1768 if (rslt) {
1769 // buf already contains library name
1770 if (offset) *offset = addr - data.base;
1771 return true;
1772 } else if (dladdr((void*)addr, &dlinfo)){
1773 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1774 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1775 return true;
1776 } else {
1777 if (buf) buf[0] = '\0';
1778 if (offset) *offset = -1;
1779 return false;
1780 }
1781 }
1782
1783 // Loads .dll/.so and
1784 // in case of error it checks if .dll/.so was built for the
1785 // same architecture as Hotspot is running on
1786
1787
1788 // Remember the stack's state. The Linux dynamic linker will change
1789 // the stack to 'executable' at most once, so we must safepoint only once.
1790 bool os::Linux::_stack_is_executable = false;
1791
1792 // VM operation that loads a library. This is necessary if stack protection
1793 // of the Java stacks can be lost during loading the library. If we
1794 // do not stop the Java threads, they can stack overflow before the stacks
1795 // are protected again.
1796 class VM_LinuxDllLoad: public VM_Operation {
1797 private:
1798 const char *_filename;
1799 char *_ebuf;
1800 int _ebuflen;
1801 void *_lib;
1802 public:
1803 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1804 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1805 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1806 void doit() {
1807 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1808 os::Linux::_stack_is_executable = true;
1809 }
1810 void* loaded_library() { return _lib; }
1811 };
1812
1813 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1814 {
1815 void * result = NULL;
1816 bool load_attempted = false;
1817
1818 // Check whether the library to load might change execution rights
1819 // of the stack. If they are changed, the protection of the stack
1820 // guard pages will be lost. We need a safepoint to fix this.
1821 //
1822 // See Linux man page execstack(8) for more info.
1823 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1824 ElfFile ef(filename);
1825 if (!ef.specifies_noexecstack()) {
1826 if (!is_init_completed()) {
1827 os::Linux::_stack_is_executable = true;
1828 // This is OK - No Java threads have been created yet, and hence no
1829 // stack guard pages to fix.
1830 //
1831 // This should happen only when you are building JDK7 using a very
1832 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1833 //
1834 // Dynamic loader will make all stacks executable after
1835 // this function returns, and will not do that again.
1836 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1837 } else {
1838 warning("You have loaded library %s which might have disabled stack guard. "
1839 "The VM will try to fix the stack guard now.\n"
1840 "It's highly recommended that you fix the library with "
1841 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1842 filename);
1843
1844 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1845 JavaThread *jt = JavaThread::current();
1846 if (jt->thread_state() != _thread_in_native) {
1847 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1848 // that requires ExecStack. Cannot enter safe point. Let's give up.
1849 warning("Unable to fix stack guard. Giving up.");
1850 } else {
1851 if (!LoadExecStackDllInVMThread) {
1852 // This is for the case where the DLL has an static
1853 // constructor function that executes JNI code. We cannot
1854 // load such DLLs in the VMThread.
1855 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1856 }
1857
1858 ThreadInVMfromNative tiv(jt);
1859 debug_only(VMNativeEntryWrapper vew;)
1860
1861 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1862 VMThread::execute(&op);
1863 if (LoadExecStackDllInVMThread) {
1864 result = op.loaded_library();
1865 }
1866 load_attempted = true;
1867 }
1868 }
1869 }
1870 }
1871
1872 if (!load_attempted) {
1873 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1874 }
1875
1876 if (result != NULL) {
1877 // Successful loading
1878 return result;
1879 }
1880
1881 Elf32_Ehdr elf_head;
1882 int diag_msg_max_length=ebuflen-strlen(ebuf);
1883 char* diag_msg_buf=ebuf+strlen(ebuf);
1884
1885 if (diag_msg_max_length==0) {
1886 // No more space in ebuf for additional diagnostics message
1887 return NULL;
1888 }
1889
1890
1891 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1892
1893 if (file_descriptor < 0) {
1894 // Can't open library, report dlerror() message
1895 return NULL;
1896 }
1897
1898 bool failed_to_read_elf_head=
1899 (sizeof(elf_head)!=
1900 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1901
1902 ::close(file_descriptor);
1903 if (failed_to_read_elf_head) {
1904 // file i/o error - report dlerror() msg
1905 return NULL;
1906 }
1907
1908 typedef struct {
1909 Elf32_Half code; // Actual value as defined in elf.h
1910 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1911 char elf_class; // 32 or 64 bit
1912 char endianess; // MSB or LSB
1913 char* name; // String representation
1914 } arch_t;
1915
1916 #ifndef EM_486
1917 #define EM_486 6 /* Intel 80486 */
1918 #endif
1919
1920 static const arch_t arch_array[]={
1921 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1922 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1923 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1924 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1925 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1926 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1927 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1928 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1929 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1930 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1931 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1932 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1933 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1934 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1935 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1936 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1937 };
1938
1939 #if (defined IA32)
1940 static Elf32_Half running_arch_code=EM_386;
1941 #elif (defined AMD64)
1942 static Elf32_Half running_arch_code=EM_X86_64;
1943 #elif (defined IA64)
1944 static Elf32_Half running_arch_code=EM_IA_64;
1945 #elif (defined __sparc) && (defined _LP64)
1946 static Elf32_Half running_arch_code=EM_SPARCV9;
1947 #elif (defined __sparc) && (!defined _LP64)
1948 static Elf32_Half running_arch_code=EM_SPARC;
1949 #elif (defined __powerpc64__)
1950 static Elf32_Half running_arch_code=EM_PPC64;
1951 #elif (defined __powerpc__)
1952 static Elf32_Half running_arch_code=EM_PPC;
1953 #elif (defined ARM)
1954 static Elf32_Half running_arch_code=EM_ARM;
1955 #elif (defined S390)
1956 static Elf32_Half running_arch_code=EM_S390;
1957 #elif (defined ALPHA)
1958 static Elf32_Half running_arch_code=EM_ALPHA;
1959 #elif (defined MIPSEL)
1960 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1961 #elif (defined PARISC)
1962 static Elf32_Half running_arch_code=EM_PARISC;
1963 #elif (defined MIPS)
1964 static Elf32_Half running_arch_code=EM_MIPS;
1965 #elif (defined M68K)
1966 static Elf32_Half running_arch_code=EM_68K;
1967 #else
1968 #error Method os::dll_load requires that one of following is defined:\
1969 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1970 #endif
1971
1972 // Identify compatability class for VM's architecture and library's architecture
1973 // Obtain string descriptions for architectures
1974
1975 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1976 int running_arch_index=-1;
1977
1978 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1979 if (running_arch_code == arch_array[i].code) {
1980 running_arch_index = i;
1981 }
1982 if (lib_arch.code == arch_array[i].code) {
1983 lib_arch.compat_class = arch_array[i].compat_class;
1984 lib_arch.name = arch_array[i].name;
1985 }
1986 }
1987
1988 assert(running_arch_index != -1,
1989 "Didn't find running architecture code (running_arch_code) in arch_array");
1990 if (running_arch_index == -1) {
1991 // Even though running architecture detection failed
1992 // we may still continue with reporting dlerror() message
1993 return NULL;
1994 }
1995
1996 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1997 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1998 return NULL;
1999 }
2000
2001 #ifndef S390
2002 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2003 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2004 return NULL;
2005 }
2006 #endif // !S390
2007
2008 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2009 if ( lib_arch.name!=NULL ) {
2010 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2011 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2012 lib_arch.name, arch_array[running_arch_index].name);
2013 } else {
2014 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2015 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2016 lib_arch.code,
2017 arch_array[running_arch_index].name);
2018 }
2019 }
2020
2021 return NULL;
2022 }
2023
2024 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2025 void * result = ::dlopen(filename, RTLD_LAZY);
2026 if (result == NULL) {
2027 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2028 ebuf[ebuflen-1] = '\0';
2029 }
2030 return result;
2031 }
2032
2033 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2034 void * result = NULL;
2035 if (LoadExecStackDllInVMThread) {
2036 result = dlopen_helper(filename, ebuf, ebuflen);
2037 }
2038
2039 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2040 // library that requires an executable stack, or which does not have this
2041 // stack attribute set, dlopen changes the stack attribute to executable. The
2042 // read protection of the guard pages gets lost.
2043 //
2044 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2045 // may have been queued at the same time.
2046
2047 if (!_stack_is_executable) {
2048 JavaThread *jt = Threads::first();
2049
2050 while (jt) {
2051 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2052 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2053 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2054 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2055 warning("Attempt to reguard stack yellow zone failed.");
2056 }
2057 }
2058 jt = jt->next();
2059 }
2060 }
2061
2062 return result;
2063 }
2064
2065 /*
2066 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2067 * chances are you might want to run the generated bits against glibc-2.0
2068 * libdl.so, so always use locking for any version of glibc.
2069 */
2070 void* os::dll_lookup(void* handle, const char* name) {
2071 pthread_mutex_lock(&dl_mutex);
2072 void* res = dlsym(handle, name);
2073 pthread_mutex_unlock(&dl_mutex);
2074 return res;
2075 }
2076
2077
2078 static bool _print_ascii_file(const char* filename, outputStream* st) {
2079 int fd = ::open(filename, O_RDONLY);
2080 if (fd == -1) {
2081 return false;
2082 }
2083
2084 char buf[32];
2085 int bytes;
2086 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2087 st->print_raw(buf, bytes);
2088 }
2089
2090 ::close(fd);
2091
2092 return true;
2093 }
2094
2095 void os::print_dll_info(outputStream *st) {
2096 st->print_cr("Dynamic libraries:");
2097
2098 char fname[32];
2099 pid_t pid = os::Linux::gettid();
2100
2101 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2102
2103 if (!_print_ascii_file(fname, st)) {
2104 st->print("Can not get library information for pid = %d\n", pid);
2105 }
2106 }
2107
2108 void os::print_os_info_brief(outputStream* st) {
2109 os::Linux::print_distro_info(st);
2110
2111 os::Posix::print_uname_info(st);
2112
2113 os::Linux::print_libversion_info(st);
2114
2115 }
2116
2117 void os::print_os_info(outputStream* st) {
2118 st->print("OS:");
2119
2120 os::Linux::print_distro_info(st);
2121
2122 os::Posix::print_uname_info(st);
2123
2124 // Print warning if unsafe chroot environment detected
2125 if (unsafe_chroot_detected) {
2126 st->print("WARNING!! ");
2127 st->print_cr(unstable_chroot_error);
2128 }
2129
2130 os::Linux::print_libversion_info(st);
2131
2132 os::Posix::print_rlimit_info(st);
2133
2134 os::Posix::print_load_average(st);
2135
2136 os::Linux::print_full_memory_info(st);
2137 }
2138
2139 // Try to identify popular distros.
2140 // Most Linux distributions have /etc/XXX-release file, which contains
2141 // the OS version string. Some have more than one /etc/XXX-release file
2142 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2143 // so the order is important.
2144 void os::Linux::print_distro_info(outputStream* st) {
2145 if (!_print_ascii_file("/etc/mandrake-release", st) &&
2146 !_print_ascii_file("/etc/sun-release", st) &&
2147 !_print_ascii_file("/etc/redhat-release", st) &&
2148 !_print_ascii_file("/etc/SuSE-release", st) &&
2149 !_print_ascii_file("/etc/turbolinux-release", st) &&
2150 !_print_ascii_file("/etc/gentoo-release", st) &&
2151 !_print_ascii_file("/etc/debian_version", st) &&
2152 !_print_ascii_file("/etc/ltib-release", st) &&
2153 !_print_ascii_file("/etc/angstrom-version", st)) {
2154 st->print("Linux");
2155 }
2156 st->cr();
2157 }
2158
2159 void os::Linux::print_libversion_info(outputStream* st) {
2160 // libc, pthread
2161 st->print("libc:");
2162 st->print(os::Linux::glibc_version()); st->print(" ");
2163 st->print(os::Linux::libpthread_version()); st->print(" ");
2164 if (os::Linux::is_LinuxThreads()) {
2165 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2166 }
2167 st->cr();
2168 }
2169
2170 void os::Linux::print_full_memory_info(outputStream* st) {
2171 st->print("\n/proc/meminfo:\n");
2172 _print_ascii_file("/proc/meminfo", st);
2173 st->cr();
2174 }
2175
2176 void os::print_memory_info(outputStream* st) {
2177
2178 st->print("Memory:");
2179 st->print(" %dk page", os::vm_page_size()>>10);
2180
2181 // values in struct sysinfo are "unsigned long"
2182 struct sysinfo si;
2183 sysinfo(&si);
2184
2185 st->print(", physical " UINT64_FORMAT "k",
2186 os::physical_memory() >> 10);
2187 st->print("(" UINT64_FORMAT "k free)",
2188 os::available_memory() >> 10);
2189 st->print(", swap " UINT64_FORMAT "k",
2190 ((jlong)si.totalswap * si.mem_unit) >> 10);
2191 st->print("(" UINT64_FORMAT "k free)",
2192 ((jlong)si.freeswap * si.mem_unit) >> 10);
2193 st->cr();
2194 }
2195
2196 void os::pd_print_cpu_info(outputStream* st) {
2197 st->print("\n/proc/cpuinfo:\n");
2198 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2199 st->print(" <Not Available>");
2200 }
2201 st->cr();
2202 }
2203
2204 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2205 // but they're the same for all the linux arch that we support
2206 // and they're the same for solaris but there's no common place to put this.
2207 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2208 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2209 "ILL_COPROC", "ILL_BADSTK" };
2210
2211 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2212 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2213 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2214
2215 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2216
2217 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2218
2219 void os::print_siginfo(outputStream* st, void* siginfo) {
2220 st->print("siginfo:");
2221
2222 const int buflen = 100;
2223 char buf[buflen];
2224 siginfo_t *si = (siginfo_t*)siginfo;
2225 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2226 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2227 st->print("si_errno=%s", buf);
2228 } else {
2229 st->print("si_errno=%d", si->si_errno);
2230 }
2231 const int c = si->si_code;
2232 assert(c > 0, "unexpected si_code");
2233 switch (si->si_signo) {
2234 case SIGILL:
2235 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2236 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2237 break;
2238 case SIGFPE:
2239 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2240 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2241 break;
2242 case SIGSEGV:
2243 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2244 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2245 break;
2246 case SIGBUS:
2247 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2248 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2249 break;
2250 default:
2251 st->print(", si_code=%d", si->si_code);
2252 // no si_addr
2253 }
2254
2255 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2256 UseSharedSpaces) {
2257 FileMapInfo* mapinfo = FileMapInfo::current_info();
2258 if (mapinfo->is_in_shared_space(si->si_addr)) {
2259 st->print("\n\nError accessing class data sharing archive." \
2260 " Mapped file inaccessible during execution, " \
2261 " possible disk/network problem.");
2262 }
2263 }
2264 st->cr();
2265 }
2266
2267
2268 static void print_signal_handler(outputStream* st, int sig,
2269 char* buf, size_t buflen);
2270
2271 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2272 st->print_cr("Signal Handlers:");
2273 print_signal_handler(st, SIGSEGV, buf, buflen);
2274 print_signal_handler(st, SIGBUS , buf, buflen);
2275 print_signal_handler(st, SIGFPE , buf, buflen);
2276 print_signal_handler(st, SIGPIPE, buf, buflen);
2277 print_signal_handler(st, SIGXFSZ, buf, buflen);
2278 print_signal_handler(st, SIGILL , buf, buflen);
2279 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2280 print_signal_handler(st, SR_signum, buf, buflen);
2281 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2282 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2283 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2284 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2285 }
2286
2287 static char saved_jvm_path[MAXPATHLEN] = {0};
2288
2289 // Find the full path to the current module, libjvm.so
2290 void os::jvm_path(char *buf, jint buflen) {
2291 // Error checking.
2292 if (buflen < MAXPATHLEN) {
2293 assert(false, "must use a large-enough buffer");
2294 buf[0] = '\0';
2295 return;
2296 }
2297 // Lazy resolve the path to current module.
2298 if (saved_jvm_path[0] != 0) {
2299 strcpy(buf, saved_jvm_path);
2300 return;
2301 }
2302
2303 char dli_fname[MAXPATHLEN];
2304 bool ret = dll_address_to_library_name(
2305 CAST_FROM_FN_PTR(address, os::jvm_path),
2306 dli_fname, sizeof(dli_fname), NULL);
2307 assert(ret != 0, "cannot locate libjvm");
2308 char *rp = realpath(dli_fname, buf);
2309 if (rp == NULL)
2310 return;
2311
2312 if (Arguments::created_by_gamma_launcher()) {
2313 // Support for the gamma launcher. Typical value for buf is
2314 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2315 // the right place in the string, then assume we are installed in a JDK and
2316 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2317 // up the path so it looks like libjvm.so is installed there (append a
2318 // fake suffix hotspot/libjvm.so).
2319 const char *p = buf + strlen(buf) - 1;
2320 for (int count = 0; p > buf && count < 5; ++count) {
2321 for (--p; p > buf && *p != '/'; --p)
2322 /* empty */ ;
2323 }
2324
2325 if (strncmp(p, "/jre/lib/", 9) != 0) {
2326 // Look for JAVA_HOME in the environment.
2327 char* java_home_var = ::getenv("JAVA_HOME");
2328 if (java_home_var != NULL && java_home_var[0] != 0) {
2329 char* jrelib_p;
2330 int len;
2331
2332 // Check the current module name "libjvm.so".
2333 p = strrchr(buf, '/');
2334 assert(strstr(p, "/libjvm") == p, "invalid library name");
2335
2336 rp = realpath(java_home_var, buf);
2337 if (rp == NULL)
2338 return;
2339
2340 // determine if this is a legacy image or modules image
2341 // modules image doesn't have "jre" subdirectory
2342 len = strlen(buf);
2343 jrelib_p = buf + len;
2344 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2345 if (0 != access(buf, F_OK)) {
2346 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2347 }
2348
2349 if (0 == access(buf, F_OK)) {
2350 // Use current module name "libjvm.so"
2351 len = strlen(buf);
2352 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2353 } else {
2354 // Go back to path of .so
2355 rp = realpath(dli_fname, buf);
2356 if (rp == NULL)
2357 return;
2358 }
2359 }
2360 }
2361 }
2362
2363 strcpy(saved_jvm_path, buf);
2364 }
2365
2366 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2367 // no prefix required, not even "_"
2368 }
2369
2370 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2371 // no suffix required
2372 }
2373
2374 ////////////////////////////////////////////////////////////////////////////////
2375 // sun.misc.Signal support
2376
2377 static volatile jint sigint_count = 0;
2378
2379 static void
2380 UserHandler(int sig, void *siginfo, void *context) {
2381 // 4511530 - sem_post is serialized and handled by the manager thread. When
2382 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2383 // don't want to flood the manager thread with sem_post requests.
2384 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2385 return;
2386
2387 // Ctrl-C is pressed during error reporting, likely because the error
2388 // handler fails to abort. Let VM die immediately.
2389 if (sig == SIGINT && is_error_reported()) {
2390 os::die();
2391 }
2392
2393 os::signal_notify(sig);
2394 }
2395
2396 void* os::user_handler() {
2397 return CAST_FROM_FN_PTR(void*, UserHandler);
2398 }
2399
2400 extern "C" {
2401 typedef void (*sa_handler_t)(int);
2402 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2403 }
2404
2405 void* os::signal(int signal_number, void* handler) {
2406 struct sigaction sigAct, oldSigAct;
2407
2408 sigfillset(&(sigAct.sa_mask));
2409 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2410 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2411
2412 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2413 // -1 means registration failed
2414 return (void *)-1;
2415 }
2416
2417 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2418 }
2419
2420 void os::signal_raise(int signal_number) {
2421 ::raise(signal_number);
2422 }
2423
2424 /*
2425 * The following code is moved from os.cpp for making this
2426 * code platform specific, which it is by its very nature.
2427 */
2428
2429 // Will be modified when max signal is changed to be dynamic
2430 int os::sigexitnum_pd() {
2431 return NSIG;
2432 }
2433
2434 // a counter for each possible signal value
2435 static volatile jint pending_signals[NSIG+1] = { 0 };
2436
2437 // Linux(POSIX) specific hand shaking semaphore.
2438 static sem_t sig_sem;
2439
2440 void os::signal_init_pd() {
2441 // Initialize signal structures
2442 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2443
2444 // Initialize signal semaphore
2445 ::sem_init(&sig_sem, 0, 0);
2446 }
2447
2448 void os::signal_notify(int sig) {
2449 Atomic::inc(&pending_signals[sig]);
2450 ::sem_post(&sig_sem);
2451 }
2452
2453 static int check_pending_signals(bool wait) {
2454 Atomic::store(0, &sigint_count);
2455 for (;;) {
2456 for (int i = 0; i < NSIG + 1; i++) {
2457 jint n = pending_signals[i];
2458 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2459 return i;
2460 }
2461 }
2462 if (!wait) {
2463 return -1;
2464 }
2465 JavaThread *thread = JavaThread::current();
2466 ThreadBlockInVM tbivm(thread);
2467
2468 bool threadIsSuspended;
2469 do {
2470 thread->set_suspend_equivalent();
2471 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2472 ::sem_wait(&sig_sem);
2473
2474 // were we externally suspended while we were waiting?
2475 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2476 if (threadIsSuspended) {
2477 //
2478 // The semaphore has been incremented, but while we were waiting
2479 // another thread suspended us. We don't want to continue running
2480 // while suspended because that would surprise the thread that
2481 // suspended us.
2482 //
2483 ::sem_post(&sig_sem);
2484
2485 thread->java_suspend_self();
2486 }
2487 } while (threadIsSuspended);
2488 }
2489 }
2490
2491 int os::signal_lookup() {
2492 return check_pending_signals(false);
2493 }
2494
2495 int os::signal_wait() {
2496 return check_pending_signals(true);
2497 }
2498
2499 ////////////////////////////////////////////////////////////////////////////////
2500 // Virtual Memory
2501
2502 int os::vm_page_size() {
2503 // Seems redundant as all get out
2504 assert(os::Linux::page_size() != -1, "must call os::init");
2505 return os::Linux::page_size();
2506 }
2507
2508 // Solaris allocates memory by pages.
2509 int os::vm_allocation_granularity() {
2510 assert(os::Linux::page_size() != -1, "must call os::init");
2511 return os::Linux::page_size();
2512 }
2513
2514 // Rationale behind this function:
2515 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2516 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2517 // samples for JITted code. Here we create private executable mapping over the code cache
2518 // and then we can use standard (well, almost, as mapping can change) way to provide
2519 // info for the reporting script by storing timestamp and location of symbol
2520 void linux_wrap_code(char* base, size_t size) {
2521 static volatile jint cnt = 0;
2522
2523 if (!UseOprofile) {
2524 return;
2525 }
2526
2527 char buf[PATH_MAX+1];
2528 int num = Atomic::add(1, &cnt);
2529
2530 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2531 os::get_temp_directory(), os::current_process_id(), num);
2532 unlink(buf);
2533
2534 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2535
2536 if (fd != -1) {
2537 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2538 if (rv != (off_t)-1) {
2539 if (::write(fd, "", 1) == 1) {
2540 mmap(base, size,
2541 PROT_READ|PROT_WRITE|PROT_EXEC,
2542 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2543 }
2544 }
2545 ::close(fd);
2546 unlink(buf);
2547 }
2548 }
2549
2550 // NOTE: Linux kernel does not really reserve the pages for us.
2551 // All it does is to check if there are enough free pages
2552 // left at the time of mmap(). This could be a potential
2553 // problem.
2554 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2555 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2556 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2557 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2558 if (res != (uintptr_t) MAP_FAILED) {
2559 if (UseNUMAInterleaving) {
2560 numa_make_global(addr, size);
2561 }
2562 return true;
2563 }
2564 return false;
2565 }
2566
2567 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2568 #ifndef MAP_HUGETLB
2569 #define MAP_HUGETLB 0x40000
2570 #endif
2571
2572 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2573 #ifndef MADV_HUGEPAGE
2574 #define MADV_HUGEPAGE 14
2575 #endif
2576
2577 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2578 bool exec) {
2579 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2580 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2581 uintptr_t res =
2582 (uintptr_t) ::mmap(addr, size, prot,
2583 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2584 -1, 0);
2585 if (res != (uintptr_t) MAP_FAILED) {
2586 if (UseNUMAInterleaving) {
2587 numa_make_global(addr, size);
2588 }
2589 return true;
2590 }
2591 // Fall through and try to use small pages
2592 }
2593
2594 if (commit_memory(addr, size, exec)) {
2595 realign_memory(addr, size, alignment_hint);
2596 return true;
2597 }
2598 return false;
2599 }
2600
2601 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2602 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2603 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2604 // be supported or the memory may already be backed by huge pages.
2605 ::madvise(addr, bytes, MADV_HUGEPAGE);
2606 }
2607 }
2608
2609 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2610 // This method works by doing an mmap over an existing mmaping and effectively discarding
2611 // the existing pages. However it won't work for SHM-based large pages that cannot be
2612 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2613 // small pages on top of the SHM segment. This method always works for small pages, so we
2614 // allow that in any case.
2615 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2616 commit_memory(addr, bytes, alignment_hint, false);
2617 }
2618 }
2619
2620 void os::numa_make_global(char *addr, size_t bytes) {
2621 Linux::numa_interleave_memory(addr, bytes);
2622 }
2623
2624 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2625 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2626 }
2627
2628 bool os::numa_topology_changed() { return false; }
2629
2630 size_t os::numa_get_groups_num() {
2631 int max_node = Linux::numa_max_node();
2632 return max_node > 0 ? max_node + 1 : 1;
2633 }
2634
2635 int os::numa_get_group_id() {
2636 int cpu_id = Linux::sched_getcpu();
2637 if (cpu_id != -1) {
2638 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2639 if (lgrp_id != -1) {
2640 return lgrp_id;
2641 }
2642 }
2643 return 0;
2644 }
2645
2646 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2647 for (size_t i = 0; i < size; i++) {
2648 ids[i] = i;
2649 }
2650 return size;
2651 }
2652
2653 bool os::get_page_info(char *start, page_info* info) {
2654 return false;
2655 }
2656
2657 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2658 return end;
2659 }
2660
2661
2662 int os::Linux::sched_getcpu_syscall(void) {
2663 unsigned int cpu;
2664 int retval = -1;
2665
2666 #if defined(IA32)
2667 # ifndef SYS_getcpu
2668 # define SYS_getcpu 318
2669 # endif
2670 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2671 #elif defined(AMD64)
2672 // Unfortunately we have to bring all these macros here from vsyscall.h
2673 // to be able to compile on old linuxes.
2674 # define __NR_vgetcpu 2
2675 # define VSYSCALL_START (-10UL << 20)
2676 # define VSYSCALL_SIZE 1024
2677 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2678 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2679 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2680 retval = vgetcpu(&cpu, NULL, NULL);
2681 #endif
2682
2683 return (retval == -1) ? retval : cpu;
2684 }
2685
2686 // Something to do with the numa-aware allocator needs these symbols
2687 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2688 extern "C" JNIEXPORT void numa_error(char *where) { }
2689 extern "C" JNIEXPORT int fork1() { return fork(); }
2690
2691
2692 // If we are running with libnuma version > 2, then we should
2693 // be trying to use symbols with versions 1.1
2694 // If we are running with earlier version, which did not have symbol versions,
2695 // we should use the base version.
2696 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2697 void *f = dlvsym(handle, name, "libnuma_1.1");
2698 if (f == NULL) {
2699 f = dlsym(handle, name);
2700 }
2701 return f;
2702 }
2703
2704 bool os::Linux::libnuma_init() {
2705 // sched_getcpu() should be in libc.
2706 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2707 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2708
2709 // If it's not, try a direct syscall.
2710 if (sched_getcpu() == -1)
2711 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2712
2713 if (sched_getcpu() != -1) { // Does it work?
2714 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2715 if (handle != NULL) {
2716 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2717 libnuma_dlsym(handle, "numa_node_to_cpus")));
2718 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2719 libnuma_dlsym(handle, "numa_max_node")));
2720 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2721 libnuma_dlsym(handle, "numa_available")));
2722 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2723 libnuma_dlsym(handle, "numa_tonode_memory")));
2724 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2725 libnuma_dlsym(handle, "numa_interleave_memory")));
2726
2727
2728 if (numa_available() != -1) {
2729 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2730 // Create a cpu -> node mapping
2731 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2732 rebuild_cpu_to_node_map();
2733 return true;
2734 }
2735 }
2736 }
2737 return false;
2738 }
2739
2740 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2741 // The table is later used in get_node_by_cpu().
2742 void os::Linux::rebuild_cpu_to_node_map() {
2743 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2744 // in libnuma (possible values are starting from 16,
2745 // and continuing up with every other power of 2, but less
2746 // than the maximum number of CPUs supported by kernel), and
2747 // is a subject to change (in libnuma version 2 the requirements
2748 // are more reasonable) we'll just hardcode the number they use
2749 // in the library.
2750 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2751
2752 size_t cpu_num = os::active_processor_count();
2753 size_t cpu_map_size = NCPUS / BitsPerCLong;
2754 size_t cpu_map_valid_size =
2755 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2756
2757 cpu_to_node()->clear();
2758 cpu_to_node()->at_grow(cpu_num - 1);
2759 size_t node_num = numa_get_groups_num();
2760
2761 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2762 for (size_t i = 0; i < node_num; i++) {
2763 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2764 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2765 if (cpu_map[j] != 0) {
2766 for (size_t k = 0; k < BitsPerCLong; k++) {
2767 if (cpu_map[j] & (1UL << k)) {
2768 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2769 }
2770 }
2771 }
2772 }
2773 }
2774 }
2775 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2776 }
2777
2778 int os::Linux::get_node_by_cpu(int cpu_id) {
2779 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2780 return cpu_to_node()->at(cpu_id);
2781 }
2782 return -1;
2783 }
2784
2785 GrowableArray<int>* os::Linux::_cpu_to_node;
2786 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2787 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2788 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2789 os::Linux::numa_available_func_t os::Linux::_numa_available;
2790 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2791 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2792 unsigned long* os::Linux::_numa_all_nodes;
2793
2794 bool os::pd_uncommit_memory(char* addr, size_t size) {
2795 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2796 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2797 return res != (uintptr_t) MAP_FAILED;
2798 }
2799
2800 // Linux uses a growable mapping for the stack, and if the mapping for
2801 // the stack guard pages is not removed when we detach a thread the
2802 // stack cannot grow beyond the pages where the stack guard was
2803 // mapped. If at some point later in the process the stack expands to
2804 // that point, the Linux kernel cannot expand the stack any further
2805 // because the guard pages are in the way, and a segfault occurs.
2806 //
2807 // However, it's essential not to split the stack region by unmapping
2808 // a region (leaving a hole) that's already part of the stack mapping,
2809 // so if the stack mapping has already grown beyond the guard pages at
2810 // the time we create them, we have to truncate the stack mapping.
2811 // So, we need to know the extent of the stack mapping when
2812 // create_stack_guard_pages() is called.
2813
2814 // Find the bounds of the stack mapping. Return true for success.
2815 //
2816 // We only need this for stacks that are growable: at the time of
2817 // writing thread stacks don't use growable mappings (i.e. those
2818 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2819 // only applies to the main thread.
2820
2821 static
2822 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2823
2824 char buf[128];
2825 int fd, sz;
2826
2827 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2828 return false;
2829 }
2830
2831 const char kw[] = "[stack]";
2832 const int kwlen = sizeof(kw)-1;
2833
2834 // Address part of /proc/self/maps couldn't be more than 128 bytes
2835 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2836 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2837 // Extract addresses
2838 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2839 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2840 if (sp >= *bottom && sp <= *top) {
2841 ::close(fd);
2842 return true;
2843 }
2844 }
2845 }
2846 }
2847
2848 ::close(fd);
2849 return false;
2850 }
2851
2852
2853 // If the (growable) stack mapping already extends beyond the point
2854 // where we're going to put our guard pages, truncate the mapping at
2855 // that point by munmap()ping it. This ensures that when we later
2856 // munmap() the guard pages we don't leave a hole in the stack
2857 // mapping. This only affects the main/initial thread, but guard
2858 // against future OS changes
2859 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2860 uintptr_t stack_extent, stack_base;
2861 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2862 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2863 assert(os::Linux::is_initial_thread(),
2864 "growable stack in non-initial thread");
2865 if (stack_extent < (uintptr_t)addr)
2866 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2867 }
2868
2869 return os::commit_memory(addr, size);
2870 }
2871
2872 // If this is a growable mapping, remove the guard pages entirely by
2873 // munmap()ping them. If not, just call uncommit_memory(). This only
2874 // affects the main/initial thread, but guard against future OS changes
2875 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2876 uintptr_t stack_extent, stack_base;
2877 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2878 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2879 assert(os::Linux::is_initial_thread(),
2880 "growable stack in non-initial thread");
2881
2882 return ::munmap(addr, size) == 0;
2883 }
2884
2885 return os::uncommit_memory(addr, size);
2886 }
2887
2888 static address _highest_vm_reserved_address = NULL;
2889
2890 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2891 // at 'requested_addr'. If there are existing memory mappings at the same
2892 // location, however, they will be overwritten. If 'fixed' is false,
2893 // 'requested_addr' is only treated as a hint, the return value may or
2894 // may not start from the requested address. Unlike Linux mmap(), this
2895 // function returns NULL to indicate failure.
2896 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2897 char * addr;
2898 int flags;
2899
2900 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2901 if (fixed) {
2902 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2903 flags |= MAP_FIXED;
2904 }
2905
2906 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
2907 // touch an uncommitted page. Otherwise, the read/write might
2908 // succeed if we have enough swap space to back the physical page.
2909 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
2910 flags, -1, 0);
2911
2912 if (addr != MAP_FAILED) {
2913 // anon_mmap() should only get called during VM initialization,
2914 // don't need lock (actually we can skip locking even it can be called
2915 // from multiple threads, because _highest_vm_reserved_address is just a
2916 // hint about the upper limit of non-stack memory regions.)
2917 if ((address)addr + bytes > _highest_vm_reserved_address) {
2918 _highest_vm_reserved_address = (address)addr + bytes;
2919 }
2920 }
2921
2922 return addr == MAP_FAILED ? NULL : addr;
2923 }
2924
2925 // Don't update _highest_vm_reserved_address, because there might be memory
2926 // regions above addr + size. If so, releasing a memory region only creates
2927 // a hole in the address space, it doesn't help prevent heap-stack collision.
2928 //
2929 static int anon_munmap(char * addr, size_t size) {
2930 return ::munmap(addr, size) == 0;
2931 }
2932
2933 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2934 size_t alignment_hint) {
2935 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2936 }
2937
2938 bool os::pd_release_memory(char* addr, size_t size) {
2939 return anon_munmap(addr, size);
2940 }
2941
2942 static address highest_vm_reserved_address() {
2943 return _highest_vm_reserved_address;
2944 }
2945
2946 static bool linux_mprotect(char* addr, size_t size, int prot) {
2947 // Linux wants the mprotect address argument to be page aligned.
2948 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2949
2950 // According to SUSv3, mprotect() should only be used with mappings
2951 // established by mmap(), and mmap() always maps whole pages. Unaligned
2952 // 'addr' likely indicates problem in the VM (e.g. trying to change
2953 // protection of malloc'ed or statically allocated memory). Check the
2954 // caller if you hit this assert.
2955 assert(addr == bottom, "sanity check");
2956
2957 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2958 return ::mprotect(bottom, size, prot) == 0;
2959 }
2960
2961 // Set protections specified
2962 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2963 bool is_committed) {
2964 unsigned int p = 0;
2965 switch (prot) {
2966 case MEM_PROT_NONE: p = PROT_NONE; break;
2967 case MEM_PROT_READ: p = PROT_READ; break;
2968 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2969 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2970 default:
2971 ShouldNotReachHere();
2972 }
2973 // is_committed is unused.
2974 return linux_mprotect(addr, bytes, p);
2975 }
2976
2977 bool os::guard_memory(char* addr, size_t size) {
2978 return linux_mprotect(addr, size, PROT_NONE);
2979 }
2980
2981 bool os::unguard_memory(char* addr, size_t size) {
2982 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2983 }
2984
2985 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2986 bool result = false;
2987 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2988 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2989 -1, 0);
2990
2991 if (p != (void *) -1) {
2992 // We don't know if this really is a huge page or not.
2993 FILE *fp = fopen("/proc/self/maps", "r");
2994 if (fp) {
2995 while (!feof(fp)) {
2996 char chars[257];
2997 long x = 0;
2998 if (fgets(chars, sizeof(chars), fp)) {
2999 if (sscanf(chars, "%lx-%*x", &x) == 1
3000 && x == (long)p) {
3001 if (strstr (chars, "hugepage")) {
3002 result = true;
3003 break;
3004 }
3005 }
3006 }
3007 }
3008 fclose(fp);
3009 }
3010 munmap (p, page_size);
3011 if (result)
3012 return true;
3013 }
3014
3015 if (warn) {
3016 warning("HugeTLBFS is not supported by the operating system.");
3017 }
3018
3019 return result;
3020 }
3021
3022 /*
3023 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3024 *
3025 * From the coredump_filter documentation:
3026 *
3027 * - (bit 0) anonymous private memory
3028 * - (bit 1) anonymous shared memory
3029 * - (bit 2) file-backed private memory
3030 * - (bit 3) file-backed shared memory
3031 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3032 * effective only if the bit 2 is cleared)
3033 * - (bit 5) hugetlb private memory
3034 * - (bit 6) hugetlb shared memory
3035 */
3036 static void set_coredump_filter(void) {
3037 FILE *f;
3038 long cdm;
3039
3040 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3041 return;
3042 }
3043
3044 if (fscanf(f, "%lx", &cdm) != 1) {
3045 fclose(f);
3046 return;
3047 }
3048
3049 rewind(f);
3050
3051 if ((cdm & LARGEPAGES_BIT) == 0) {
3052 cdm |= LARGEPAGES_BIT;
3053 fprintf(f, "%#lx", cdm);
3054 }
3055
3056 fclose(f);
3057 }
3058
3059 // Large page support
3060
3061 static size_t _large_page_size = 0;
3062
3063 void os::large_page_init() {
3064 if (!UseLargePages) {
3065 UseHugeTLBFS = false;
3066 UseSHM = false;
3067 return;
3068 }
3069
3070 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3071 // If UseLargePages is specified on the command line try both methods,
3072 // if it's default, then try only HugeTLBFS.
3073 if (FLAG_IS_DEFAULT(UseLargePages)) {
3074 UseHugeTLBFS = true;
3075 } else {
3076 UseHugeTLBFS = UseSHM = true;
3077 }
3078 }
3079
3080 if (LargePageSizeInBytes) {
3081 _large_page_size = LargePageSizeInBytes;
3082 } else {
3083 // large_page_size on Linux is used to round up heap size. x86 uses either
3084 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3085 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3086 // page as large as 256M.
3087 //
3088 // Here we try to figure out page size by parsing /proc/meminfo and looking
3089 // for a line with the following format:
3090 // Hugepagesize: 2048 kB
3091 //
3092 // If we can't determine the value (e.g. /proc is not mounted, or the text
3093 // format has been changed), we'll use the largest page size supported by
3094 // the processor.
3095
3096 #ifndef ZERO
3097 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3098 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3099 #endif // ZERO
3100
3101 FILE *fp = fopen("/proc/meminfo", "r");
3102 if (fp) {
3103 while (!feof(fp)) {
3104 int x = 0;
3105 char buf[16];
3106 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3107 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3108 _large_page_size = x * K;
3109 break;
3110 }
3111 } else {
3112 // skip to next line
3113 for (;;) {
3114 int ch = fgetc(fp);
3115 if (ch == EOF || ch == (int)'\n') break;
3116 }
3117 }
3118 }
3119 fclose(fp);
3120 }
3121 }
3122
3123 // print a warning if any large page related flag is specified on command line
3124 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3125
3126 const size_t default_page_size = (size_t)Linux::page_size();
3127 if (_large_page_size > default_page_size) {
3128 _page_sizes[0] = _large_page_size;
3129 _page_sizes[1] = default_page_size;
3130 _page_sizes[2] = 0;
3131 }
3132 UseHugeTLBFS = UseHugeTLBFS &&
3133 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3134
3135 if (UseHugeTLBFS)
3136 UseSHM = false;
3137
3138 UseLargePages = UseHugeTLBFS || UseSHM;
3139
3140 set_coredump_filter();
3141 }
3142
3143 #ifndef SHM_HUGETLB
3144 #define SHM_HUGETLB 04000
3145 #endif
3146
3147 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3148 // "exec" is passed in but not used. Creating the shared image for
3149 // the code cache doesn't have an SHM_X executable permission to check.
3150 assert(UseLargePages && UseSHM, "only for SHM large pages");
3151
3152 key_t key = IPC_PRIVATE;
3153 char *addr;
3154
3155 bool warn_on_failure = UseLargePages &&
3156 (!FLAG_IS_DEFAULT(UseLargePages) ||
3157 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3158 );
3159 char msg[128];
3160
3161 // Create a large shared memory region to attach to based on size.
3162 // Currently, size is the total size of the heap
3163 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3164 if (shmid == -1) {
3165 // Possible reasons for shmget failure:
3166 // 1. shmmax is too small for Java heap.
3167 // > check shmmax value: cat /proc/sys/kernel/shmmax
3168 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3169 // 2. not enough large page memory.
3170 // > check available large pages: cat /proc/meminfo
3171 // > increase amount of large pages:
3172 // echo new_value > /proc/sys/vm/nr_hugepages
3173 // Note 1: different Linux may use different name for this property,
3174 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3175 // Note 2: it's possible there's enough physical memory available but
3176 // they are so fragmented after a long run that they can't
3177 // coalesce into large pages. Try to reserve large pages when
3178 // the system is still "fresh".
3179 if (warn_on_failure) {
3180 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3181 warning(msg);
3182 }
3183 return NULL;
3184 }
3185
3186 // attach to the region
3187 addr = (char*)shmat(shmid, req_addr, 0);
3188 int err = errno;
3189
3190 // Remove shmid. If shmat() is successful, the actual shared memory segment
3191 // will be deleted when it's detached by shmdt() or when the process
3192 // terminates. If shmat() is not successful this will remove the shared
3193 // segment immediately.
3194 shmctl(shmid, IPC_RMID, NULL);
3195
3196 if ((intptr_t)addr == -1) {
3197 if (warn_on_failure) {
3198 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3199 warning(msg);
3200 }
3201 return NULL;
3202 }
3203
3204 if ((addr != NULL) && UseNUMAInterleaving) {
3205 numa_make_global(addr, bytes);
3206 }
3207
3208 // The memory is committed
3209 address pc = CALLER_PC;
3210 MemTracker::record_virtual_memory_reserve((address)addr, bytes, pc);
3211 MemTracker::record_virtual_memory_commit((address)addr, bytes, pc);
3212
3213 return addr;
3214 }
3215
3216 bool os::release_memory_special(char* base, size_t bytes) {
3217 // detaching the SHM segment will also delete it, see reserve_memory_special()
3218 int rslt = shmdt(base);
3219 if (rslt == 0) {
3220 MemTracker::record_virtual_memory_uncommit((address)base, bytes);
3221 MemTracker::record_virtual_memory_release((address)base, bytes);
3222 return true;
3223 } else {
3224 return false;
3225 }
3226 }
3227
3228 size_t os::large_page_size() {
3229 return _large_page_size;
3230 }
3231
3232 // HugeTLBFS allows application to commit large page memory on demand;
3233 // with SysV SHM the entire memory region must be allocated as shared
3234 // memory.
3235 bool os::can_commit_large_page_memory() {
3236 return UseHugeTLBFS;
3237 }
3238
3239 bool os::can_execute_large_page_memory() {
3240 return UseHugeTLBFS;
3241 }
3242
3243 // Reserve memory at an arbitrary address, only if that area is
3244 // available (and not reserved for something else).
3245
3246 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3247 const int max_tries = 10;
3248 char* base[max_tries];
3249 size_t size[max_tries];
3250 const size_t gap = 0x000000;
3251
3252 // Assert only that the size is a multiple of the page size, since
3253 // that's all that mmap requires, and since that's all we really know
3254 // about at this low abstraction level. If we need higher alignment,
3255 // we can either pass an alignment to this method or verify alignment
3256 // in one of the methods further up the call chain. See bug 5044738.
3257 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3258
3259 // Repeatedly allocate blocks until the block is allocated at the
3260 // right spot. Give up after max_tries. Note that reserve_memory() will
3261 // automatically update _highest_vm_reserved_address if the call is
3262 // successful. The variable tracks the highest memory address every reserved
3263 // by JVM. It is used to detect heap-stack collision if running with
3264 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3265 // space than needed, it could confuse the collision detecting code. To
3266 // solve the problem, save current _highest_vm_reserved_address and
3267 // calculate the correct value before return.
3268 address old_highest = _highest_vm_reserved_address;
3269
3270 // Linux mmap allows caller to pass an address as hint; give it a try first,
3271 // if kernel honors the hint then we can return immediately.
3272 char * addr = anon_mmap(requested_addr, bytes, false);
3273 if (addr == requested_addr) {
3274 return requested_addr;
3275 }
3276
3277 if (addr != NULL) {
3278 // mmap() is successful but it fails to reserve at the requested address
3279 anon_munmap(addr, bytes);
3280 }
3281
3282 int i;
3283 for (i = 0; i < max_tries; ++i) {
3284 base[i] = reserve_memory(bytes);
3285
3286 if (base[i] != NULL) {
3287 // Is this the block we wanted?
3288 if (base[i] == requested_addr) {
3289 size[i] = bytes;
3290 break;
3291 }
3292
3293 // Does this overlap the block we wanted? Give back the overlapped
3294 // parts and try again.
3295
3296 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3297 if (top_overlap >= 0 && top_overlap < bytes) {
3298 unmap_memory(base[i], top_overlap);
3299 base[i] += top_overlap;
3300 size[i] = bytes - top_overlap;
3301 } else {
3302 size_t bottom_overlap = base[i] + bytes - requested_addr;
3303 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3304 unmap_memory(requested_addr, bottom_overlap);
3305 size[i] = bytes - bottom_overlap;
3306 } else {
3307 size[i] = bytes;
3308 }
3309 }
3310 }
3311 }
3312
3313 // Give back the unused reserved pieces.
3314
3315 for (int j = 0; j < i; ++j) {
3316 if (base[j] != NULL) {
3317 unmap_memory(base[j], size[j]);
3318 }
3319 }
3320
3321 if (i < max_tries) {
3322 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3323 return requested_addr;
3324 } else {
3325 _highest_vm_reserved_address = old_highest;
3326 return NULL;
3327 }
3328 }
3329
3330 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3331 return ::read(fd, buf, nBytes);
3332 }
3333
3334 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3335 // Solaris uses poll(), linux uses park().
3336 // Poll() is likely a better choice, assuming that Thread.interrupt()
3337 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3338 // SIGSEGV, see 4355769.
3339
3340 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3341 assert(thread == Thread::current(), "thread consistency check");
3342
3343 ParkEvent * const slp = thread->_SleepEvent ;
3344 slp->reset() ;
3345 OrderAccess::fence() ;
3346
3347 if (interruptible) {
3348 jlong prevtime = javaTimeNanos();
3349
3350 for (;;) {
3351 if (os::is_interrupted(thread, true)) {
3352 return OS_INTRPT;
3353 }
3354
3355 jlong newtime = javaTimeNanos();
3356
3357 if (newtime - prevtime < 0) {
3358 // time moving backwards, should only happen if no monotonic clock
3359 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3360 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3361 } else {
3362 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3363 }
3364
3365 if(millis <= 0) {
3366 return OS_OK;
3367 }
3368
3369 prevtime = newtime;
3370
3371 {
3372 assert(thread->is_Java_thread(), "sanity check");
3373 JavaThread *jt = (JavaThread *) thread;
3374 ThreadBlockInVM tbivm(jt);
3375 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3376
3377 jt->set_suspend_equivalent();
3378 // cleared by handle_special_suspend_equivalent_condition() or
3379 // java_suspend_self() via check_and_wait_while_suspended()
3380
3381 slp->park(millis);
3382
3383 // were we externally suspended while we were waiting?
3384 jt->check_and_wait_while_suspended();
3385 }
3386 }
3387 } else {
3388 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3389 jlong prevtime = javaTimeNanos();
3390
3391 for (;;) {
3392 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3393 // the 1st iteration ...
3394 jlong newtime = javaTimeNanos();
3395
3396 if (newtime - prevtime < 0) {
3397 // time moving backwards, should only happen if no monotonic clock
3398 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3399 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3400 } else {
3401 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3402 }
3403
3404 if(millis <= 0) break ;
3405
3406 prevtime = newtime;
3407 slp->park(millis);
3408 }
3409 return OS_OK ;
3410 }
3411 }
3412
3413 int os::naked_sleep() {
3414 // %% make the sleep time an integer flag. for now use 1 millisec.
3415 return os::sleep(Thread::current(), 1, false);
3416 }
3417
3418 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3419 void os::infinite_sleep() {
3420 while (true) { // sleep forever ...
3421 ::sleep(100); // ... 100 seconds at a time
3422 }
3423 }
3424
3425 // Used to convert frequent JVM_Yield() to nops
3426 bool os::dont_yield() {
3427 return DontYieldALot;
3428 }
3429
3430 void os::yield() {
3431 sched_yield();
3432 }
3433
3434 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3435
3436 void os::yield_all(int attempts) {
3437 // Yields to all threads, including threads with lower priorities
3438 // Threads on Linux are all with same priority. The Solaris style
3439 // os::yield_all() with nanosleep(1ms) is not necessary.
3440 sched_yield();
3441 }
3442
3443 // Called from the tight loops to possibly influence time-sharing heuristics
3444 void os::loop_breaker(int attempts) {
3445 os::yield_all(attempts);
3446 }
3447
3448 ////////////////////////////////////////////////////////////////////////////////
3449 // thread priority support
3450
3451 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3452 // only supports dynamic priority, static priority must be zero. For real-time
3453 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3454 // However, for large multi-threaded applications, SCHED_RR is not only slower
3455 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3456 // of 5 runs - Sep 2005).
3457 //
3458 // The following code actually changes the niceness of kernel-thread/LWP. It
3459 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3460 // not the entire user process, and user level threads are 1:1 mapped to kernel
3461 // threads. It has always been the case, but could change in the future. For
3462 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3463 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3464
3465 int os::java_to_os_priority[CriticalPriority + 1] = {
3466 19, // 0 Entry should never be used
3467
3468 4, // 1 MinPriority
3469 3, // 2
3470 2, // 3
3471
3472 1, // 4
3473 0, // 5 NormPriority
3474 -1, // 6
3475
3476 -2, // 7
3477 -3, // 8
3478 -4, // 9 NearMaxPriority
3479
3480 -5, // 10 MaxPriority
3481
3482 -5 // 11 CriticalPriority
3483 };
3484
3485 static int prio_init() {
3486 if (ThreadPriorityPolicy == 1) {
3487 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3488 // if effective uid is not root. Perhaps, a more elegant way of doing
3489 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3490 if (geteuid() != 0) {
3491 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3492 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3493 }
3494 ThreadPriorityPolicy = 0;
3495 }
3496 }
3497 if (UseCriticalJavaThreadPriority) {
3498 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3499 }
3500 return 0;
3501 }
3502
3503 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3504 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3505
3506 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3507 return (ret == 0) ? OS_OK : OS_ERR;
3508 }
3509
3510 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3511 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3512 *priority_ptr = java_to_os_priority[NormPriority];
3513 return OS_OK;
3514 }
3515
3516 errno = 0;
3517 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3518 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3519 }
3520
3521 // Hint to the underlying OS that a task switch would not be good.
3522 // Void return because it's a hint and can fail.
3523 void os::hint_no_preempt() {}
3524
3525 ////////////////////////////////////////////////////////////////////////////////
3526 // suspend/resume support
3527
3528 // the low-level signal-based suspend/resume support is a remnant from the
3529 // old VM-suspension that used to be for java-suspension, safepoints etc,
3530 // within hotspot. Now there is a single use-case for this:
3531 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3532 // that runs in the watcher thread.
3533 // The remaining code is greatly simplified from the more general suspension
3534 // code that used to be used.
3535 //
3536 // The protocol is quite simple:
3537 // - suspend:
3538 // - sends a signal to the target thread
3539 // - polls the suspend state of the osthread using a yield loop
3540 // - target thread signal handler (SR_handler) sets suspend state
3541 // and blocks in sigsuspend until continued
3542 // - resume:
3543 // - sets target osthread state to continue
3544 // - sends signal to end the sigsuspend loop in the SR_handler
3545 //
3546 // Note that the SR_lock plays no role in this suspend/resume protocol.
3547 //
3548
3549 static void resume_clear_context(OSThread *osthread) {
3550 osthread->set_ucontext(NULL);
3551 osthread->set_siginfo(NULL);
3552
3553 // notify the suspend action is completed, we have now resumed
3554 osthread->sr.clear_suspended();
3555 }
3556
3557 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3558 osthread->set_ucontext(context);
3559 osthread->set_siginfo(siginfo);
3560 }
3561
3562 //
3563 // Handler function invoked when a thread's execution is suspended or
3564 // resumed. We have to be careful that only async-safe functions are
3565 // called here (Note: most pthread functions are not async safe and
3566 // should be avoided.)
3567 //
3568 // Note: sigwait() is a more natural fit than sigsuspend() from an
3569 // interface point of view, but sigwait() prevents the signal hander
3570 // from being run. libpthread would get very confused by not having
3571 // its signal handlers run and prevents sigwait()'s use with the
3572 // mutex granting granting signal.
3573 //
3574 // Currently only ever called on the VMThread
3575 //
3576 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3577 // Save and restore errno to avoid confusing native code with EINTR
3578 // after sigsuspend.
3579 int old_errno = errno;
3580
3581 Thread* thread = Thread::current();
3582 OSThread* osthread = thread->osthread();
3583 assert(thread->is_VM_thread(), "Must be VMThread");
3584 // read current suspend action
3585 int action = osthread->sr.suspend_action();
3586 if (action == os::Linux::SuspendResume::SR_SUSPEND) {
3587 suspend_save_context(osthread, siginfo, context);
3588
3589 // Notify the suspend action is about to be completed. do_suspend()
3590 // waits until SR_SUSPENDED is set and then returns. We will wait
3591 // here for a resume signal and that completes the suspend-other
3592 // action. do_suspend/do_resume is always called as a pair from
3593 // the same thread - so there are no races
3594
3595 // notify the caller
3596 osthread->sr.set_suspended();
3597
3598 sigset_t suspend_set; // signals for sigsuspend()
3599
3600 // get current set of blocked signals and unblock resume signal
3601 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3602 sigdelset(&suspend_set, SR_signum);
3603
3604 // wait here until we are resumed
3605 do {
3606 sigsuspend(&suspend_set);
3607 // ignore all returns until we get a resume signal
3608 } while (osthread->sr.suspend_action() != os::Linux::SuspendResume::SR_CONTINUE);
3609
3610 resume_clear_context(osthread);
3611
3612 } else {
3613 assert(action == os::Linux::SuspendResume::SR_CONTINUE, "unexpected sr action");
3614 // nothing special to do - just leave the handler
3615 }
3616
3617 errno = old_errno;
3618 }
3619
3620
3621 static int SR_initialize() {
3622 struct sigaction act;
3623 char *s;
3624 /* Get signal number to use for suspend/resume */
3625 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3626 int sig = ::strtol(s, 0, 10);
3627 if (sig > 0 || sig < _NSIG) {
3628 SR_signum = sig;
3629 }
3630 }
3631
3632 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3633 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3634
3635 sigemptyset(&SR_sigset);
3636 sigaddset(&SR_sigset, SR_signum);
3637
3638 /* Set up signal handler for suspend/resume */
3639 act.sa_flags = SA_RESTART|SA_SIGINFO;
3640 act.sa_handler = (void (*)(int)) SR_handler;
3641
3642 // SR_signum is blocked by default.
3643 // 4528190 - We also need to block pthread restart signal (32 on all
3644 // supported Linux platforms). Note that LinuxThreads need to block
3645 // this signal for all threads to work properly. So we don't have
3646 // to use hard-coded signal number when setting up the mask.
3647 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3648
3649 if (sigaction(SR_signum, &act, 0) == -1) {
3650 return -1;
3651 }
3652
3653 // Save signal flag
3654 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3655 return 0;
3656 }
3657
3658
3659 // returns true on success and false on error - really an error is fatal
3660 // but this seems the normal response to library errors
3661 static bool do_suspend(OSThread* osthread) {
3662 // mark as suspended and send signal
3663 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_SUSPEND);
3664 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3665 assert_status(status == 0, status, "pthread_kill");
3666
3667 // check status and wait until notified of suspension
3668 if (status == 0) {
3669 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3670 os::yield_all(i);
3671 }
3672 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3673 return true;
3674 }
3675 else {
3676 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3677 return false;
3678 }
3679 }
3680
3681 static void do_resume(OSThread* osthread) {
3682 assert(osthread->sr.is_suspended(), "thread should be suspended");
3683 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_CONTINUE);
3684
3685 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3686 assert_status(status == 0, status, "pthread_kill");
3687 // check status and wait unit notified of resumption
3688 if (status == 0) {
3689 for (int i = 0; osthread->sr.is_suspended(); i++) {
3690 os::yield_all(i);
3691 }
3692 }
3693 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3694 }
3695
3696 ////////////////////////////////////////////////////////////////////////////////
3697 // interrupt support
3698
3699 void os::interrupt(Thread* thread) {
3700 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3701 "possibility of dangling Thread pointer");
3702
3703 OSThread* osthread = thread->osthread();
3704
3705 if (!osthread->interrupted()) {
3706 osthread->set_interrupted(true);
3707 // More than one thread can get here with the same value of osthread,
3708 // resulting in multiple notifications. We do, however, want the store
3709 // to interrupted() to be visible to other threads before we execute unpark().
3710 OrderAccess::fence();
3711 ParkEvent * const slp = thread->_SleepEvent ;
3712 if (slp != NULL) slp->unpark() ;
3713 }
3714
3715 // For JSR166. Unpark even if interrupt status already was set
3716 if (thread->is_Java_thread())
3717 ((JavaThread*)thread)->parker()->unpark();
3718
3719 ParkEvent * ev = thread->_ParkEvent ;
3720 if (ev != NULL) ev->unpark() ;
3721
3722 }
3723
3724 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3725 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3726 "possibility of dangling Thread pointer");
3727
3728 OSThread* osthread = thread->osthread();
3729
3730 bool interrupted = osthread->interrupted();
3731
3732 if (interrupted && clear_interrupted) {
3733 osthread->set_interrupted(false);
3734 // consider thread->_SleepEvent->reset() ... optional optimization
3735 }
3736
3737 return interrupted;
3738 }
3739
3740 ///////////////////////////////////////////////////////////////////////////////////
3741 // signal handling (except suspend/resume)
3742
3743 // This routine may be used by user applications as a "hook" to catch signals.
3744 // The user-defined signal handler must pass unrecognized signals to this
3745 // routine, and if it returns true (non-zero), then the signal handler must
3746 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3747 // routine will never retun false (zero), but instead will execute a VM panic
3748 // routine kill the process.
3749 //
3750 // If this routine returns false, it is OK to call it again. This allows
3751 // the user-defined signal handler to perform checks either before or after
3752 // the VM performs its own checks. Naturally, the user code would be making
3753 // a serious error if it tried to handle an exception (such as a null check
3754 // or breakpoint) that the VM was generating for its own correct operation.
3755 //
3756 // This routine may recognize any of the following kinds of signals:
3757 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3758 // It should be consulted by handlers for any of those signals.
3759 //
3760 // The caller of this routine must pass in the three arguments supplied
3761 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3762 // field of the structure passed to sigaction(). This routine assumes that
3763 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3764 //
3765 // Note that the VM will print warnings if it detects conflicting signal
3766 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3767 //
3768 extern "C" JNIEXPORT int
3769 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3770 void* ucontext, int abort_if_unrecognized);
3771
3772 void signalHandler(int sig, siginfo_t* info, void* uc) {
3773 assert(info != NULL && uc != NULL, "it must be old kernel");
3774 int orig_errno = errno; // Preserve errno value over signal handler.
3775 JVM_handle_linux_signal(sig, info, uc, true);
3776 errno = orig_errno;
3777 }
3778
3779
3780 // This boolean allows users to forward their own non-matching signals
3781 // to JVM_handle_linux_signal, harmlessly.
3782 bool os::Linux::signal_handlers_are_installed = false;
3783
3784 // For signal-chaining
3785 struct sigaction os::Linux::sigact[MAXSIGNUM];
3786 unsigned int os::Linux::sigs = 0;
3787 bool os::Linux::libjsig_is_loaded = false;
3788 typedef struct sigaction *(*get_signal_t)(int);
3789 get_signal_t os::Linux::get_signal_action = NULL;
3790
3791 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3792 struct sigaction *actp = NULL;
3793
3794 if (libjsig_is_loaded) {
3795 // Retrieve the old signal handler from libjsig
3796 actp = (*get_signal_action)(sig);
3797 }
3798 if (actp == NULL) {
3799 // Retrieve the preinstalled signal handler from jvm
3800 actp = get_preinstalled_handler(sig);
3801 }
3802
3803 return actp;
3804 }
3805
3806 static bool call_chained_handler(struct sigaction *actp, int sig,
3807 siginfo_t *siginfo, void *context) {
3808 // Call the old signal handler
3809 if (actp->sa_handler == SIG_DFL) {
3810 // It's more reasonable to let jvm treat it as an unexpected exception
3811 // instead of taking the default action.
3812 return false;
3813 } else if (actp->sa_handler != SIG_IGN) {
3814 if ((actp->sa_flags & SA_NODEFER) == 0) {
3815 // automaticlly block the signal
3816 sigaddset(&(actp->sa_mask), sig);
3817 }
3818
3819 sa_handler_t hand;
3820 sa_sigaction_t sa;
3821 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3822 // retrieve the chained handler
3823 if (siginfo_flag_set) {
3824 sa = actp->sa_sigaction;
3825 } else {
3826 hand = actp->sa_handler;
3827 }
3828
3829 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3830 actp->sa_handler = SIG_DFL;
3831 }
3832
3833 // try to honor the signal mask
3834 sigset_t oset;
3835 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3836
3837 // call into the chained handler
3838 if (siginfo_flag_set) {
3839 (*sa)(sig, siginfo, context);
3840 } else {
3841 (*hand)(sig);
3842 }
3843
3844 // restore the signal mask
3845 pthread_sigmask(SIG_SETMASK, &oset, 0);
3846 }
3847 // Tell jvm's signal handler the signal is taken care of.
3848 return true;
3849 }
3850
3851 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3852 bool chained = false;
3853 // signal-chaining
3854 if (UseSignalChaining) {
3855 struct sigaction *actp = get_chained_signal_action(sig);
3856 if (actp != NULL) {
3857 chained = call_chained_handler(actp, sig, siginfo, context);
3858 }
3859 }
3860 return chained;
3861 }
3862
3863 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3864 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3865 return &sigact[sig];
3866 }
3867 return NULL;
3868 }
3869
3870 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3871 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3872 sigact[sig] = oldAct;
3873 sigs |= (unsigned int)1 << sig;
3874 }
3875
3876 // for diagnostic
3877 int os::Linux::sigflags[MAXSIGNUM];
3878
3879 int os::Linux::get_our_sigflags(int sig) {
3880 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3881 return sigflags[sig];
3882 }
3883
3884 void os::Linux::set_our_sigflags(int sig, int flags) {
3885 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3886 sigflags[sig] = flags;
3887 }
3888
3889 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3890 // Check for overwrite.
3891 struct sigaction oldAct;
3892 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3893
3894 void* oldhand = oldAct.sa_sigaction
3895 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3896 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3897 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3898 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3899 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3900 if (AllowUserSignalHandlers || !set_installed) {
3901 // Do not overwrite; user takes responsibility to forward to us.
3902 return;
3903 } else if (UseSignalChaining) {
3904 // save the old handler in jvm
3905 save_preinstalled_handler(sig, oldAct);
3906 // libjsig also interposes the sigaction() call below and saves the
3907 // old sigaction on it own.
3908 } else {
3909 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3910 "%#lx for signal %d.", (long)oldhand, sig));
3911 }
3912 }
3913
3914 struct sigaction sigAct;
3915 sigfillset(&(sigAct.sa_mask));
3916 sigAct.sa_handler = SIG_DFL;
3917 if (!set_installed) {
3918 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3919 } else {
3920 sigAct.sa_sigaction = signalHandler;
3921 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3922 }
3923 // Save flags, which are set by ours
3924 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3925 sigflags[sig] = sigAct.sa_flags;
3926
3927 int ret = sigaction(sig, &sigAct, &oldAct);
3928 assert(ret == 0, "check");
3929
3930 void* oldhand2 = oldAct.sa_sigaction
3931 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3932 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3933 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3934 }
3935
3936 // install signal handlers for signals that HotSpot needs to
3937 // handle in order to support Java-level exception handling.
3938
3939 void os::Linux::install_signal_handlers() {
3940 if (!signal_handlers_are_installed) {
3941 signal_handlers_are_installed = true;
3942
3943 // signal-chaining
3944 typedef void (*signal_setting_t)();
3945 signal_setting_t begin_signal_setting = NULL;
3946 signal_setting_t end_signal_setting = NULL;
3947 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3948 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3949 if (begin_signal_setting != NULL) {
3950 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3951 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3952 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3953 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3954 libjsig_is_loaded = true;
3955 assert(UseSignalChaining, "should enable signal-chaining");
3956 }
3957 if (libjsig_is_loaded) {
3958 // Tell libjsig jvm is setting signal handlers
3959 (*begin_signal_setting)();
3960 }
3961
3962 set_signal_handler(SIGSEGV, true);
3963 set_signal_handler(SIGPIPE, true);
3964 set_signal_handler(SIGBUS, true);
3965 set_signal_handler(SIGILL, true);
3966 set_signal_handler(SIGFPE, true);
3967 set_signal_handler(SIGXFSZ, true);
3968
3969 if (libjsig_is_loaded) {
3970 // Tell libjsig jvm finishes setting signal handlers
3971 (*end_signal_setting)();
3972 }
3973
3974 // We don't activate signal checker if libjsig is in place, we trust ourselves
3975 // and if UserSignalHandler is installed all bets are off.
3976 // Log that signal checking is off only if -verbose:jni is specified.
3977 if (CheckJNICalls) {
3978 if (libjsig_is_loaded) {
3979 if (PrintJNIResolving) {
3980 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3981 }
3982 check_signals = false;
3983 }
3984 if (AllowUserSignalHandlers) {
3985 if (PrintJNIResolving) {
3986 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3987 }
3988 check_signals = false;
3989 }
3990 }
3991 }
3992 }
3993
3994 // This is the fastest way to get thread cpu time on Linux.
3995 // Returns cpu time (user+sys) for any thread, not only for current.
3996 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3997 // It might work on 2.6.10+ with a special kernel/glibc patch.
3998 // For reference, please, see IEEE Std 1003.1-2004:
3999 // http://www.unix.org/single_unix_specification
4000
4001 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4002 struct timespec tp;
4003 int rc = os::Linux::clock_gettime(clockid, &tp);
4004 assert(rc == 0, "clock_gettime is expected to return 0 code");
4005
4006 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4007 }
4008
4009 /////
4010 // glibc on Linux platform uses non-documented flag
4011 // to indicate, that some special sort of signal
4012 // trampoline is used.
4013 // We will never set this flag, and we should
4014 // ignore this flag in our diagnostic
4015 #ifdef SIGNIFICANT_SIGNAL_MASK
4016 #undef SIGNIFICANT_SIGNAL_MASK
4017 #endif
4018 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4019
4020 static const char* get_signal_handler_name(address handler,
4021 char* buf, int buflen) {
4022 int offset;
4023 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4024 if (found) {
4025 // skip directory names
4026 const char *p1, *p2;
4027 p1 = buf;
4028 size_t len = strlen(os::file_separator());
4029 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4030 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4031 } else {
4032 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4033 }
4034 return buf;
4035 }
4036
4037 static void print_signal_handler(outputStream* st, int sig,
4038 char* buf, size_t buflen) {
4039 struct sigaction sa;
4040
4041 sigaction(sig, NULL, &sa);
4042
4043 // See comment for SIGNIFICANT_SIGNAL_MASK define
4044 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4045
4046 st->print("%s: ", os::exception_name(sig, buf, buflen));
4047
4048 address handler = (sa.sa_flags & SA_SIGINFO)
4049 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4050 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4051
4052 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4053 st->print("SIG_DFL");
4054 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4055 st->print("SIG_IGN");
4056 } else {
4057 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4058 }
4059
4060 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4061
4062 address rh = VMError::get_resetted_sighandler(sig);
4063 // May be, handler was resetted by VMError?
4064 if(rh != NULL) {
4065 handler = rh;
4066 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4067 }
4068
4069 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
4070
4071 // Check: is it our handler?
4072 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4073 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4074 // It is our signal handler
4075 // check for flags, reset system-used one!
4076 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4077 st->print(
4078 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4079 os::Linux::get_our_sigflags(sig));
4080 }
4081 }
4082 st->cr();
4083 }
4084
4085
4086 #define DO_SIGNAL_CHECK(sig) \
4087 if (!sigismember(&check_signal_done, sig)) \
4088 os::Linux::check_signal_handler(sig)
4089
4090 // This method is a periodic task to check for misbehaving JNI applications
4091 // under CheckJNI, we can add any periodic checks here
4092
4093 void os::run_periodic_checks() {
4094
4095 if (check_signals == false) return;
4096
4097 // SEGV and BUS if overridden could potentially prevent
4098 // generation of hs*.log in the event of a crash, debugging
4099 // such a case can be very challenging, so we absolutely
4100 // check the following for a good measure:
4101 DO_SIGNAL_CHECK(SIGSEGV);
4102 DO_SIGNAL_CHECK(SIGILL);
4103 DO_SIGNAL_CHECK(SIGFPE);
4104 DO_SIGNAL_CHECK(SIGBUS);
4105 DO_SIGNAL_CHECK(SIGPIPE);
4106 DO_SIGNAL_CHECK(SIGXFSZ);
4107
4108
4109 // ReduceSignalUsage allows the user to override these handlers
4110 // see comments at the very top and jvm_solaris.h
4111 if (!ReduceSignalUsage) {
4112 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4113 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4114 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4115 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4116 }
4117
4118 DO_SIGNAL_CHECK(SR_signum);
4119 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4120 }
4121
4122 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4123
4124 static os_sigaction_t os_sigaction = NULL;
4125
4126 void os::Linux::check_signal_handler(int sig) {
4127 char buf[O_BUFLEN];
4128 address jvmHandler = NULL;
4129
4130
4131 struct sigaction act;
4132 if (os_sigaction == NULL) {
4133 // only trust the default sigaction, in case it has been interposed
4134 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4135 if (os_sigaction == NULL) return;
4136 }
4137
4138 os_sigaction(sig, (struct sigaction*)NULL, &act);
4139
4140
4141 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4142
4143 address thisHandler = (act.sa_flags & SA_SIGINFO)
4144 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4145 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4146
4147
4148 switch(sig) {
4149 case SIGSEGV:
4150 case SIGBUS:
4151 case SIGFPE:
4152 case SIGPIPE:
4153 case SIGILL:
4154 case SIGXFSZ:
4155 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4156 break;
4157
4158 case SHUTDOWN1_SIGNAL:
4159 case SHUTDOWN2_SIGNAL:
4160 case SHUTDOWN3_SIGNAL:
4161 case BREAK_SIGNAL:
4162 jvmHandler = (address)user_handler();
4163 break;
4164
4165 case INTERRUPT_SIGNAL:
4166 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4167 break;
4168
4169 default:
4170 if (sig == SR_signum) {
4171 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4172 } else {
4173 return;
4174 }
4175 break;
4176 }
4177
4178 if (thisHandler != jvmHandler) {
4179 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4180 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4181 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4182 // No need to check this sig any longer
4183 sigaddset(&check_signal_done, sig);
4184 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4185 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4186 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4187 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4188 // No need to check this sig any longer
4189 sigaddset(&check_signal_done, sig);
4190 }
4191
4192 // Dump all the signal
4193 if (sigismember(&check_signal_done, sig)) {
4194 print_signal_handlers(tty, buf, O_BUFLEN);
4195 }
4196 }
4197
4198 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4199
4200 extern bool signal_name(int signo, char* buf, size_t len);
4201
4202 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4203 if (0 < exception_code && exception_code <= SIGRTMAX) {
4204 // signal
4205 if (!signal_name(exception_code, buf, size)) {
4206 jio_snprintf(buf, size, "SIG%d", exception_code);
4207 }
4208 return buf;
4209 } else {
4210 return NULL;
4211 }
4212 }
4213
4214 // this is called _before_ the most of global arguments have been parsed
4215 void os::init(void) {
4216 char dummy; /* used to get a guess on initial stack address */
4217 // first_hrtime = gethrtime();
4218
4219 // With LinuxThreads the JavaMain thread pid (primordial thread)
4220 // is different than the pid of the java launcher thread.
4221 // So, on Linux, the launcher thread pid is passed to the VM
4222 // via the sun.java.launcher.pid property.
4223 // Use this property instead of getpid() if it was correctly passed.
4224 // See bug 6351349.
4225 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4226
4227 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4228
4229 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4230
4231 init_random(1234567);
4232
4233 ThreadCritical::initialize();
4234
4235 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4236 if (Linux::page_size() == -1) {
4237 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4238 strerror(errno)));
4239 }
4240 init_page_sizes((size_t) Linux::page_size());
4241
4242 Linux::initialize_system_info();
4243
4244 // main_thread points to the aboriginal thread
4245 Linux::_main_thread = pthread_self();
4246
4247 Linux::clock_init();
4248 initial_time_count = os::elapsed_counter();
4249 pthread_mutex_init(&dl_mutex, NULL);
4250
4251 // If the pagesize of the VM is greater than 8K determine the appropriate
4252 // number of initial guard pages. The user can change this with the
4253 // command line arguments, if needed.
4254 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4255 StackYellowPages = 1;
4256 StackRedPages = 1;
4257 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4258 }
4259 }
4260
4261 // To install functions for atexit system call
4262 extern "C" {
4263 static void perfMemory_exit_helper() {
4264 perfMemory_exit();
4265 }
4266 }
4267
4268 // this is called _after_ the global arguments have been parsed
4269 jint os::init_2(void)
4270 {
4271 Linux::fast_thread_clock_init();
4272
4273 // Allocate a single page and mark it as readable for safepoint polling
4274 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4275 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4276
4277 os::set_polling_page( polling_page );
4278
4279 #ifndef PRODUCT
4280 if(Verbose && PrintMiscellaneous)
4281 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4282 #endif
4283
4284 if (!UseMembar) {
4285 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4286 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4287 os::set_memory_serialize_page( mem_serialize_page );
4288
4289 #ifndef PRODUCT
4290 if(Verbose && PrintMiscellaneous)
4291 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4292 #endif
4293 }
4294
4295 os::large_page_init();
4296
4297 // initialize suspend/resume support - must do this before signal_sets_init()
4298 if (SR_initialize() != 0) {
4299 perror("SR_initialize failed");
4300 return JNI_ERR;
4301 }
4302
4303 Linux::signal_sets_init();
4304 Linux::install_signal_handlers();
4305
4306 // Check minimum allowable stack size for thread creation and to initialize
4307 // the java system classes, including StackOverflowError - depends on page
4308 // size. Add a page for compiler2 recursion in main thread.
4309 // Add in 2*BytesPerWord times page size to account for VM stack during
4310 // class initialization depending on 32 or 64 bit VM.
4311 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4312 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4313 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4314
4315 size_t threadStackSizeInBytes = ThreadStackSize * K;
4316 if (threadStackSizeInBytes != 0 &&
4317 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4318 tty->print_cr("\nThe stack size specified is too small, "
4319 "Specify at least %dk",
4320 os::Linux::min_stack_allowed/ K);
4321 return JNI_ERR;
4322 }
4323
4324 // Make the stack size a multiple of the page size so that
4325 // the yellow/red zones can be guarded.
4326 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4327 vm_page_size()));
4328
4329 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4330
4331 Linux::libpthread_init();
4332 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4333 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4334 Linux::glibc_version(), Linux::libpthread_version(),
4335 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4336 }
4337
4338 if (UseNUMA) {
4339 if (!Linux::libnuma_init()) {
4340 UseNUMA = false;
4341 } else {
4342 if ((Linux::numa_max_node() < 1)) {
4343 // There's only one node(they start from 0), disable NUMA.
4344 UseNUMA = false;
4345 }
4346 }
4347 // With SHM large pages we cannot uncommit a page, so there's not way
4348 // we can make the adaptive lgrp chunk resizing work. If the user specified
4349 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4350 // disable adaptive resizing.
4351 if (UseNUMA && UseLargePages && UseSHM) {
4352 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4353 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4354 UseLargePages = false;
4355 } else {
4356 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4357 UseAdaptiveSizePolicy = false;
4358 UseAdaptiveNUMAChunkSizing = false;
4359 }
4360 } else {
4361 UseNUMA = false;
4362 }
4363 }
4364 if (!UseNUMA && ForceNUMA) {
4365 UseNUMA = true;
4366 }
4367 }
4368
4369 if (MaxFDLimit) {
4370 // set the number of file descriptors to max. print out error
4371 // if getrlimit/setrlimit fails but continue regardless.
4372 struct rlimit nbr_files;
4373 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4374 if (status != 0) {
4375 if (PrintMiscellaneous && (Verbose || WizardMode))
4376 perror("os::init_2 getrlimit failed");
4377 } else {
4378 nbr_files.rlim_cur = nbr_files.rlim_max;
4379 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4380 if (status != 0) {
4381 if (PrintMiscellaneous && (Verbose || WizardMode))
4382 perror("os::init_2 setrlimit failed");
4383 }
4384 }
4385 }
4386
4387 // Initialize lock used to serialize thread creation (see os::create_thread)
4388 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4389
4390 // at-exit methods are called in the reverse order of their registration.
4391 // atexit functions are called on return from main or as a result of a
4392 // call to exit(3C). There can be only 32 of these functions registered
4393 // and atexit() does not set errno.
4394
4395 if (PerfAllowAtExitRegistration) {
4396 // only register atexit functions if PerfAllowAtExitRegistration is set.
4397 // atexit functions can be delayed until process exit time, which
4398 // can be problematic for embedded VM situations. Embedded VMs should
4399 // call DestroyJavaVM() to assure that VM resources are released.
4400
4401 // note: perfMemory_exit_helper atexit function may be removed in
4402 // the future if the appropriate cleanup code can be added to the
4403 // VM_Exit VMOperation's doit method.
4404 if (atexit(perfMemory_exit_helper) != 0) {
4405 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4406 }
4407 }
4408
4409 // initialize thread priority policy
4410 prio_init();
4411
4412 return JNI_OK;
4413 }
4414
4415 // this is called at the end of vm_initialization
4416 void os::init_3(void)
4417 {
4418 #ifdef JAVASE_EMBEDDED
4419 // Start the MemNotifyThread
4420 if (LowMemoryProtection) {
4421 MemNotifyThread::start();
4422 }
4423 return;
4424 #endif
4425 }
4426
4427 // Mark the polling page as unreadable
4428 void os::make_polling_page_unreadable(void) {
4429 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4430 fatal("Could not disable polling page");
4431 };
4432
4433 // Mark the polling page as readable
4434 void os::make_polling_page_readable(void) {
4435 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4436 fatal("Could not enable polling page");
4437 }
4438 };
4439
4440 int os::active_processor_count() {
4441 // Linux doesn't yet have a (official) notion of processor sets,
4442 // so just return the number of online processors.
4443 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4444 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4445 return online_cpus;
4446 }
4447
4448 void os::set_native_thread_name(const char *name) {
4449 // Not yet implemented.
4450 return;
4451 }
4452
4453 bool os::distribute_processes(uint length, uint* distribution) {
4454 // Not yet implemented.
4455 return false;
4456 }
4457
4458 bool os::bind_to_processor(uint processor_id) {
4459 // Not yet implemented.
4460 return false;
4461 }
4462
4463 ///
4464
4465 // Suspends the target using the signal mechanism and then grabs the PC before
4466 // resuming the target. Used by the flat-profiler only
4467 ExtendedPC os::get_thread_pc(Thread* thread) {
4468 // Make sure that it is called by the watcher for the VMThread
4469 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4470 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4471
4472 ExtendedPC epc;
4473
4474 OSThread* osthread = thread->osthread();
4475 if (do_suspend(osthread)) {
4476 if (osthread->ucontext() != NULL) {
4477 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4478 } else {
4479 // NULL context is unexpected, double-check this is the VMThread
4480 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4481 }
4482 do_resume(osthread);
4483 }
4484 // failure means pthread_kill failed for some reason - arguably this is
4485 // a fatal problem, but such problems are ignored elsewhere
4486
4487 return epc;
4488 }
4489
4490 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4491 {
4492 if (is_NPTL()) {
4493 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4494 } else {
4495 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4496 // word back to default 64bit precision if condvar is signaled. Java
4497 // wants 53bit precision. Save and restore current value.
4498 int fpu = get_fpu_control_word();
4499 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4500 set_fpu_control_word(fpu);
4501 return status;
4502 }
4503 }
4504
4505 ////////////////////////////////////////////////////////////////////////////////
4506 // debug support
4507
4508 bool os::find(address addr, outputStream* st) {
4509 Dl_info dlinfo;
4510 memset(&dlinfo, 0, sizeof(dlinfo));
4511 if (dladdr(addr, &dlinfo)) {
4512 st->print(PTR_FORMAT ": ", addr);
4513 if (dlinfo.dli_sname != NULL) {
4514 st->print("%s+%#x", dlinfo.dli_sname,
4515 addr - (intptr_t)dlinfo.dli_saddr);
4516 } else if (dlinfo.dli_fname) {
4517 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4518 } else {
4519 st->print("<absolute address>");
4520 }
4521 if (dlinfo.dli_fname) {
4522 st->print(" in %s", dlinfo.dli_fname);
4523 }
4524 if (dlinfo.dli_fbase) {
4525 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4526 }
4527 st->cr();
4528
4529 if (Verbose) {
4530 // decode some bytes around the PC
4531 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4532 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4533 address lowest = (address) dlinfo.dli_sname;
4534 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4535 if (begin < lowest) begin = lowest;
4536 Dl_info dlinfo2;
4537 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4538 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4539 end = (address) dlinfo2.dli_saddr;
4540 Disassembler::decode(begin, end, st);
4541 }
4542 return true;
4543 }
4544 return false;
4545 }
4546
4547 ////////////////////////////////////////////////////////////////////////////////
4548 // misc
4549
4550 // This does not do anything on Linux. This is basically a hook for being
4551 // able to use structured exception handling (thread-local exception filters)
4552 // on, e.g., Win32.
4553 void
4554 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4555 JavaCallArguments* args, Thread* thread) {
4556 f(value, method, args, thread);
4557 }
4558
4559 void os::print_statistics() {
4560 }
4561
4562 int os::message_box(const char* title, const char* message) {
4563 int i;
4564 fdStream err(defaultStream::error_fd());
4565 for (i = 0; i < 78; i++) err.print_raw("=");
4566 err.cr();
4567 err.print_raw_cr(title);
4568 for (i = 0; i < 78; i++) err.print_raw("-");
4569 err.cr();
4570 err.print_raw_cr(message);
4571 for (i = 0; i < 78; i++) err.print_raw("=");
4572 err.cr();
4573
4574 char buf[16];
4575 // Prevent process from exiting upon "read error" without consuming all CPU
4576 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4577
4578 return buf[0] == 'y' || buf[0] == 'Y';
4579 }
4580
4581 int os::stat(const char *path, struct stat *sbuf) {
4582 char pathbuf[MAX_PATH];
4583 if (strlen(path) > MAX_PATH - 1) {
4584 errno = ENAMETOOLONG;
4585 return -1;
4586 }
4587 os::native_path(strcpy(pathbuf, path));
4588 return ::stat(pathbuf, sbuf);
4589 }
4590
4591 bool os::check_heap(bool force) {
4592 return true;
4593 }
4594
4595 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4596 return ::vsnprintf(buf, count, format, args);
4597 }
4598
4599 // Is a (classpath) directory empty?
4600 bool os::dir_is_empty(const char* path) {
4601 DIR *dir = NULL;
4602 struct dirent *ptr;
4603
4604 dir = opendir(path);
4605 if (dir == NULL) return true;
4606
4607 /* Scan the directory */
4608 bool result = true;
4609 char buf[sizeof(struct dirent) + MAX_PATH];
4610 while (result && (ptr = ::readdir(dir)) != NULL) {
4611 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4612 result = false;
4613 }
4614 }
4615 closedir(dir);
4616 return result;
4617 }
4618
4619 // This code originates from JDK's sysOpen and open64_w
4620 // from src/solaris/hpi/src/system_md.c
4621
4622 #ifndef O_DELETE
4623 #define O_DELETE 0x10000
4624 #endif
4625
4626 // Open a file. Unlink the file immediately after open returns
4627 // if the specified oflag has the O_DELETE flag set.
4628 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4629
4630 int os::open(const char *path, int oflag, int mode) {
4631
4632 if (strlen(path) > MAX_PATH - 1) {
4633 errno = ENAMETOOLONG;
4634 return -1;
4635 }
4636 int fd;
4637 int o_delete = (oflag & O_DELETE);
4638 oflag = oflag & ~O_DELETE;
4639
4640 fd = ::open64(path, oflag, mode);
4641 if (fd == -1) return -1;
4642
4643 //If the open succeeded, the file might still be a directory
4644 {
4645 struct stat64 buf64;
4646 int ret = ::fstat64(fd, &buf64);
4647 int st_mode = buf64.st_mode;
4648
4649 if (ret != -1) {
4650 if ((st_mode & S_IFMT) == S_IFDIR) {
4651 errno = EISDIR;
4652 ::close(fd);
4653 return -1;
4654 }
4655 } else {
4656 ::close(fd);
4657 return -1;
4658 }
4659 }
4660
4661 /*
4662 * All file descriptors that are opened in the JVM and not
4663 * specifically destined for a subprocess should have the
4664 * close-on-exec flag set. If we don't set it, then careless 3rd
4665 * party native code might fork and exec without closing all
4666 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4667 * UNIXProcess.c), and this in turn might:
4668 *
4669 * - cause end-of-file to fail to be detected on some file
4670 * descriptors, resulting in mysterious hangs, or
4671 *
4672 * - might cause an fopen in the subprocess to fail on a system
4673 * suffering from bug 1085341.
4674 *
4675 * (Yes, the default setting of the close-on-exec flag is a Unix
4676 * design flaw)
4677 *
4678 * See:
4679 * 1085341: 32-bit stdio routines should support file descriptors >255
4680 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4681 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4682 */
4683 #ifdef FD_CLOEXEC
4684 {
4685 int flags = ::fcntl(fd, F_GETFD);
4686 if (flags != -1)
4687 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4688 }
4689 #endif
4690
4691 if (o_delete != 0) {
4692 ::unlink(path);
4693 }
4694 return fd;
4695 }
4696
4697
4698 // create binary file, rewriting existing file if required
4699 int os::create_binary_file(const char* path, bool rewrite_existing) {
4700 int oflags = O_WRONLY | O_CREAT;
4701 if (!rewrite_existing) {
4702 oflags |= O_EXCL;
4703 }
4704 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4705 }
4706
4707 // return current position of file pointer
4708 jlong os::current_file_offset(int fd) {
4709 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4710 }
4711
4712 // move file pointer to the specified offset
4713 jlong os::seek_to_file_offset(int fd, jlong offset) {
4714 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4715 }
4716
4717 // This code originates from JDK's sysAvailable
4718 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4719
4720 int os::available(int fd, jlong *bytes) {
4721 jlong cur, end;
4722 int mode;
4723 struct stat64 buf64;
4724
4725 if (::fstat64(fd, &buf64) >= 0) {
4726 mode = buf64.st_mode;
4727 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4728 /*
4729 * XXX: is the following call interruptible? If so, this might
4730 * need to go through the INTERRUPT_IO() wrapper as for other
4731 * blocking, interruptible calls in this file.
4732 */
4733 int n;
4734 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4735 *bytes = n;
4736 return 1;
4737 }
4738 }
4739 }
4740 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4741 return 0;
4742 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4743 return 0;
4744 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4745 return 0;
4746 }
4747 *bytes = end - cur;
4748 return 1;
4749 }
4750
4751 int os::socket_available(int fd, jint *pbytes) {
4752 // Linux doc says EINTR not returned, unlike Solaris
4753 int ret = ::ioctl(fd, FIONREAD, pbytes);
4754
4755 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4756 // is expected to return 0 on failure and 1 on success to the jdk.
4757 return (ret < 0) ? 0 : 1;
4758 }
4759
4760 // Map a block of memory.
4761 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4762 char *addr, size_t bytes, bool read_only,
4763 bool allow_exec) {
4764 int prot;
4765 int flags = MAP_PRIVATE;
4766
4767 if (read_only) {
4768 prot = PROT_READ;
4769 } else {
4770 prot = PROT_READ | PROT_WRITE;
4771 }
4772
4773 if (allow_exec) {
4774 prot |= PROT_EXEC;
4775 }
4776
4777 if (addr != NULL) {
4778 flags |= MAP_FIXED;
4779 }
4780
4781 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4782 fd, file_offset);
4783 if (mapped_address == MAP_FAILED) {
4784 return NULL;
4785 }
4786 return mapped_address;
4787 }
4788
4789
4790 // Remap a block of memory.
4791 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4792 char *addr, size_t bytes, bool read_only,
4793 bool allow_exec) {
4794 // same as map_memory() on this OS
4795 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4796 allow_exec);
4797 }
4798
4799
4800 // Unmap a block of memory.
4801 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4802 return munmap(addr, bytes) == 0;
4803 }
4804
4805 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4806
4807 static clockid_t thread_cpu_clockid(Thread* thread) {
4808 pthread_t tid = thread->osthread()->pthread_id();
4809 clockid_t clockid;
4810
4811 // Get thread clockid
4812 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4813 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4814 return clockid;
4815 }
4816
4817 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4818 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4819 // of a thread.
4820 //
4821 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4822 // the fast estimate available on the platform.
4823
4824 jlong os::current_thread_cpu_time() {
4825 if (os::Linux::supports_fast_thread_cpu_time()) {
4826 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4827 } else {
4828 // return user + sys since the cost is the same
4829 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4830 }
4831 }
4832
4833 jlong os::thread_cpu_time(Thread* thread) {
4834 // consistent with what current_thread_cpu_time() returns
4835 if (os::Linux::supports_fast_thread_cpu_time()) {
4836 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4837 } else {
4838 return slow_thread_cpu_time(thread, true /* user + sys */);
4839 }
4840 }
4841
4842 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4843 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4844 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4845 } else {
4846 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4847 }
4848 }
4849
4850 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4851 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4852 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4853 } else {
4854 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4855 }
4856 }
4857
4858 //
4859 // -1 on error.
4860 //
4861
4862 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4863 static bool proc_task_unchecked = true;
4864 static const char *proc_stat_path = "/proc/%d/stat";
4865 pid_t tid = thread->osthread()->thread_id();
4866 char *s;
4867 char stat[2048];
4868 int statlen;
4869 char proc_name[64];
4870 int count;
4871 long sys_time, user_time;
4872 char cdummy;
4873 int idummy;
4874 long ldummy;
4875 FILE *fp;
4876
4877 // The /proc/<tid>/stat aggregates per-process usage on
4878 // new Linux kernels 2.6+ where NPTL is supported.
4879 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4880 // See bug 6328462.
4881 // There possibly can be cases where there is no directory
4882 // /proc/self/task, so we check its availability.
4883 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4884 // This is executed only once
4885 proc_task_unchecked = false;
4886 fp = fopen("/proc/self/task", "r");
4887 if (fp != NULL) {
4888 proc_stat_path = "/proc/self/task/%d/stat";
4889 fclose(fp);
4890 }
4891 }
4892
4893 sprintf(proc_name, proc_stat_path, tid);
4894 fp = fopen(proc_name, "r");
4895 if ( fp == NULL ) return -1;
4896 statlen = fread(stat, 1, 2047, fp);
4897 stat[statlen] = '\0';
4898 fclose(fp);
4899
4900 // Skip pid and the command string. Note that we could be dealing with
4901 // weird command names, e.g. user could decide to rename java launcher
4902 // to "java 1.4.2 :)", then the stat file would look like
4903 // 1234 (java 1.4.2 :)) R ... ...
4904 // We don't really need to know the command string, just find the last
4905 // occurrence of ")" and then start parsing from there. See bug 4726580.
4906 s = strrchr(stat, ')');
4907 if (s == NULL ) return -1;
4908
4909 // Skip blank chars
4910 do s++; while (isspace(*s));
4911
4912 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4913 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4914 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4915 &user_time, &sys_time);
4916 if ( count != 13 ) return -1;
4917 if (user_sys_cpu_time) {
4918 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4919 } else {
4920 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4921 }
4922 }
4923
4924 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4925 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4926 info_ptr->may_skip_backward = false; // elapsed time not wall time
4927 info_ptr->may_skip_forward = false; // elapsed time not wall time
4928 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4929 }
4930
4931 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4932 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4933 info_ptr->may_skip_backward = false; // elapsed time not wall time
4934 info_ptr->may_skip_forward = false; // elapsed time not wall time
4935 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4936 }
4937
4938 bool os::is_thread_cpu_time_supported() {
4939 return true;
4940 }
4941
4942 // System loadavg support. Returns -1 if load average cannot be obtained.
4943 // Linux doesn't yet have a (official) notion of processor sets,
4944 // so just return the system wide load average.
4945 int os::loadavg(double loadavg[], int nelem) {
4946 return ::getloadavg(loadavg, nelem);
4947 }
4948
4949 void os::pause() {
4950 char filename[MAX_PATH];
4951 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4952 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4953 } else {
4954 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4955 }
4956
4957 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4958 if (fd != -1) {
4959 struct stat buf;
4960 ::close(fd);
4961 while (::stat(filename, &buf) == 0) {
4962 (void)::poll(NULL, 0, 100);
4963 }
4964 } else {
4965 jio_fprintf(stderr,
4966 "Could not open pause file '%s', continuing immediately.\n", filename);
4967 }
4968 }
4969
4970
4971 // Refer to the comments in os_solaris.cpp park-unpark.
4972 //
4973 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4974 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4975 // For specifics regarding the bug see GLIBC BUGID 261237 :
4976 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4977 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4978 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4979 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4980 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4981 // and monitorenter when we're using 1-0 locking. All those operations may result in
4982 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4983 // of libpthread avoids the problem, but isn't practical.
4984 //
4985 // Possible remedies:
4986 //
4987 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4988 // This is palliative and probabilistic, however. If the thread is preempted
4989 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4990 // than the minimum period may have passed, and the abstime may be stale (in the
4991 // past) resultin in a hang. Using this technique reduces the odds of a hang
4992 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4993 //
4994 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4995 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4996 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4997 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4998 // thread.
4999 //
5000 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5001 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5002 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5003 // This also works well. In fact it avoids kernel-level scalability impediments
5004 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5005 // timers in a graceful fashion.
5006 //
5007 // 4. When the abstime value is in the past it appears that control returns
5008 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5009 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5010 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5011 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5012 // It may be possible to avoid reinitialization by checking the return
5013 // value from pthread_cond_timedwait(). In addition to reinitializing the
5014 // condvar we must establish the invariant that cond_signal() is only called
5015 // within critical sections protected by the adjunct mutex. This prevents
5016 // cond_signal() from "seeing" a condvar that's in the midst of being
5017 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5018 // desirable signal-after-unlock optimization that avoids futile context switching.
5019 //
5020 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5021 // structure when a condvar is used or initialized. cond_destroy() would
5022 // release the helper structure. Our reinitialize-after-timedwait fix
5023 // put excessive stress on malloc/free and locks protecting the c-heap.
5024 //
5025 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5026 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5027 // and only enabling the work-around for vulnerable environments.
5028
5029 // utility to compute the abstime argument to timedwait:
5030 // millis is the relative timeout time
5031 // abstime will be the absolute timeout time
5032 // TODO: replace compute_abstime() with unpackTime()
5033
5034 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5035 if (millis < 0) millis = 0;
5036 struct timeval now;
5037 int status = gettimeofday(&now, NULL);
5038 assert(status == 0, "gettimeofday");
5039 jlong seconds = millis / 1000;
5040 millis %= 1000;
5041 if (seconds > 50000000) { // see man cond_timedwait(3T)
5042 seconds = 50000000;
5043 }
5044 abstime->tv_sec = now.tv_sec + seconds;
5045 long usec = now.tv_usec + millis * 1000;
5046 if (usec >= 1000000) {
5047 abstime->tv_sec += 1;
5048 usec -= 1000000;
5049 }
5050 abstime->tv_nsec = usec * 1000;
5051 return abstime;
5052 }
5053
5054
5055 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5056 // Conceptually TryPark() should be equivalent to park(0).
5057
5058 int os::PlatformEvent::TryPark() {
5059 for (;;) {
5060 const int v = _Event ;
5061 guarantee ((v == 0) || (v == 1), "invariant") ;
5062 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5063 }
5064 }
5065
5066 void os::PlatformEvent::park() { // AKA "down()"
5067 // Invariant: Only the thread associated with the Event/PlatformEvent
5068 // may call park().
5069 // TODO: assert that _Assoc != NULL or _Assoc == Self
5070 int v ;
5071 for (;;) {
5072 v = _Event ;
5073 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5074 }
5075 guarantee (v >= 0, "invariant") ;
5076 if (v == 0) {
5077 // Do this the hard way by blocking ...
5078 int status = pthread_mutex_lock(_mutex);
5079 assert_status(status == 0, status, "mutex_lock");
5080 guarantee (_nParked == 0, "invariant") ;
5081 ++ _nParked ;
5082 while (_Event < 0) {
5083 status = pthread_cond_wait(_cond, _mutex);
5084 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5085 // Treat this the same as if the wait was interrupted
5086 if (status == ETIME) { status = EINTR; }
5087 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5088 }
5089 -- _nParked ;
5090
5091 _Event = 0 ;
5092 status = pthread_mutex_unlock(_mutex);
5093 assert_status(status == 0, status, "mutex_unlock");
5094 // Paranoia to ensure our locked and lock-free paths interact
5095 // correctly with each other.
5096 OrderAccess::fence();
5097 }
5098 guarantee (_Event >= 0, "invariant") ;
5099 }
5100
5101 int os::PlatformEvent::park(jlong millis) {
5102 guarantee (_nParked == 0, "invariant") ;
5103
5104 int v ;
5105 for (;;) {
5106 v = _Event ;
5107 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5108 }
5109 guarantee (v >= 0, "invariant") ;
5110 if (v != 0) return OS_OK ;
5111
5112 // We do this the hard way, by blocking the thread.
5113 // Consider enforcing a minimum timeout value.
5114 struct timespec abst;
5115 compute_abstime(&abst, millis);
5116
5117 int ret = OS_TIMEOUT;
5118 int status = pthread_mutex_lock(_mutex);
5119 assert_status(status == 0, status, "mutex_lock");
5120 guarantee (_nParked == 0, "invariant") ;
5121 ++_nParked ;
5122
5123 // Object.wait(timo) will return because of
5124 // (a) notification
5125 // (b) timeout
5126 // (c) thread.interrupt
5127 //
5128 // Thread.interrupt and object.notify{All} both call Event::set.
5129 // That is, we treat thread.interrupt as a special case of notification.
5130 // The underlying Solaris implementation, cond_timedwait, admits
5131 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5132 // JVM from making those visible to Java code. As such, we must
5133 // filter out spurious wakeups. We assume all ETIME returns are valid.
5134 //
5135 // TODO: properly differentiate simultaneous notify+interrupt.
5136 // In that case, we should propagate the notify to another waiter.
5137
5138 while (_Event < 0) {
5139 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5140 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5141 pthread_cond_destroy (_cond);
5142 pthread_cond_init (_cond, NULL) ;
5143 }
5144 assert_status(status == 0 || status == EINTR ||
5145 status == ETIME || status == ETIMEDOUT,
5146 status, "cond_timedwait");
5147 if (!FilterSpuriousWakeups) break ; // previous semantics
5148 if (status == ETIME || status == ETIMEDOUT) break ;
5149 // We consume and ignore EINTR and spurious wakeups.
5150 }
5151 --_nParked ;
5152 if (_Event >= 0) {
5153 ret = OS_OK;
5154 }
5155 _Event = 0 ;
5156 status = pthread_mutex_unlock(_mutex);
5157 assert_status(status == 0, status, "mutex_unlock");
5158 assert (_nParked == 0, "invariant") ;
5159 // Paranoia to ensure our locked and lock-free paths interact
5160 // correctly with each other.
5161 OrderAccess::fence();
5162 return ret;
5163 }
5164
5165 void os::PlatformEvent::unpark() {
5166 // Transitions for _Event:
5167 // 0 :=> 1
5168 // 1 :=> 1
5169 // -1 :=> either 0 or 1; must signal target thread
5170 // That is, we can safely transition _Event from -1 to either
5171 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5172 // unpark() calls.
5173 // See also: "Semaphores in Plan 9" by Mullender & Cox
5174 //
5175 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5176 // that it will take two back-to-back park() calls for the owning
5177 // thread to block. This has the benefit of forcing a spurious return
5178 // from the first park() call after an unpark() call which will help
5179 // shake out uses of park() and unpark() without condition variables.
5180
5181 if (Atomic::xchg(1, &_Event) >= 0) return;
5182
5183 // Wait for the thread associated with the event to vacate
5184 int status = pthread_mutex_lock(_mutex);
5185 assert_status(status == 0, status, "mutex_lock");
5186 int AnyWaiters = _nParked;
5187 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5188 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5189 AnyWaiters = 0;
5190 pthread_cond_signal(_cond);
5191 }
5192 status = pthread_mutex_unlock(_mutex);
5193 assert_status(status == 0, status, "mutex_unlock");
5194 if (AnyWaiters != 0) {
5195 status = pthread_cond_signal(_cond);
5196 assert_status(status == 0, status, "cond_signal");
5197 }
5198
5199 // Note that we signal() _after dropping the lock for "immortal" Events.
5200 // This is safe and avoids a common class of futile wakeups. In rare
5201 // circumstances this can cause a thread to return prematurely from
5202 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5203 // simply re-test the condition and re-park itself.
5204 }
5205
5206
5207 // JSR166
5208 // -------------------------------------------------------
5209
5210 /*
5211 * The solaris and linux implementations of park/unpark are fairly
5212 * conservative for now, but can be improved. They currently use a
5213 * mutex/condvar pair, plus a a count.
5214 * Park decrements count if > 0, else does a condvar wait. Unpark
5215 * sets count to 1 and signals condvar. Only one thread ever waits
5216 * on the condvar. Contention seen when trying to park implies that someone
5217 * is unparking you, so don't wait. And spurious returns are fine, so there
5218 * is no need to track notifications.
5219 */
5220
5221 #define MAX_SECS 100000000
5222 /*
5223 * This code is common to linux and solaris and will be moved to a
5224 * common place in dolphin.
5225 *
5226 * The passed in time value is either a relative time in nanoseconds
5227 * or an absolute time in milliseconds. Either way it has to be unpacked
5228 * into suitable seconds and nanoseconds components and stored in the
5229 * given timespec structure.
5230 * Given time is a 64-bit value and the time_t used in the timespec is only
5231 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5232 * overflow if times way in the future are given. Further on Solaris versions
5233 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5234 * number of seconds, in abstime, is less than current_time + 100,000,000.
5235 * As it will be 28 years before "now + 100000000" will overflow we can
5236 * ignore overflow and just impose a hard-limit on seconds using the value
5237 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5238 * years from "now".
5239 */
5240
5241 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5242 assert (time > 0, "convertTime");
5243
5244 struct timeval now;
5245 int status = gettimeofday(&now, NULL);
5246 assert(status == 0, "gettimeofday");
5247
5248 time_t max_secs = now.tv_sec + MAX_SECS;
5249
5250 if (isAbsolute) {
5251 jlong secs = time / 1000;
5252 if (secs > max_secs) {
5253 absTime->tv_sec = max_secs;
5254 }
5255 else {
5256 absTime->tv_sec = secs;
5257 }
5258 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5259 }
5260 else {
5261 jlong secs = time / NANOSECS_PER_SEC;
5262 if (secs >= MAX_SECS) {
5263 absTime->tv_sec = max_secs;
5264 absTime->tv_nsec = 0;
5265 }
5266 else {
5267 absTime->tv_sec = now.tv_sec + secs;
5268 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5269 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5270 absTime->tv_nsec -= NANOSECS_PER_SEC;
5271 ++absTime->tv_sec; // note: this must be <= max_secs
5272 }
5273 }
5274 }
5275 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5276 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5277 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5278 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5279 }
5280
5281 void Parker::park(bool isAbsolute, jlong time) {
5282 // Ideally we'd do something useful while spinning, such
5283 // as calling unpackTime().
5284
5285 // Optional fast-path check:
5286 // Return immediately if a permit is available.
5287 // We depend on Atomic::xchg() having full barrier semantics
5288 // since we are doing a lock-free update to _counter.
5289 if (Atomic::xchg(0, &_counter) > 0) return;
5290
5291 Thread* thread = Thread::current();
5292 assert(thread->is_Java_thread(), "Must be JavaThread");
5293 JavaThread *jt = (JavaThread *)thread;
5294
5295 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5296 // Check interrupt before trying to wait
5297 if (Thread::is_interrupted(thread, false)) {
5298 return;
5299 }
5300
5301 // Next, demultiplex/decode time arguments
5302 timespec absTime;
5303 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5304 return;
5305 }
5306 if (time > 0) {
5307 unpackTime(&absTime, isAbsolute, time);
5308 }
5309
5310
5311 // Enter safepoint region
5312 // Beware of deadlocks such as 6317397.
5313 // The per-thread Parker:: mutex is a classic leaf-lock.
5314 // In particular a thread must never block on the Threads_lock while
5315 // holding the Parker:: mutex. If safepoints are pending both the
5316 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5317 ThreadBlockInVM tbivm(jt);
5318
5319 // Don't wait if cannot get lock since interference arises from
5320 // unblocking. Also. check interrupt before trying wait
5321 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5322 return;
5323 }
5324
5325 int status ;
5326 if (_counter > 0) { // no wait needed
5327 _counter = 0;
5328 status = pthread_mutex_unlock(_mutex);
5329 assert (status == 0, "invariant") ;
5330 // Paranoia to ensure our locked and lock-free paths interact
5331 // correctly with each other and Java-level accesses.
5332 OrderAccess::fence();
5333 return;
5334 }
5335
5336 #ifdef ASSERT
5337 // Don't catch signals while blocked; let the running threads have the signals.
5338 // (This allows a debugger to break into the running thread.)
5339 sigset_t oldsigs;
5340 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5341 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5342 #endif
5343
5344 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5345 jt->set_suspend_equivalent();
5346 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5347
5348 if (time == 0) {
5349 status = pthread_cond_wait (_cond, _mutex) ;
5350 } else {
5351 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5352 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5353 pthread_cond_destroy (_cond) ;
5354 pthread_cond_init (_cond, NULL);
5355 }
5356 }
5357 assert_status(status == 0 || status == EINTR ||
5358 status == ETIME || status == ETIMEDOUT,
5359 status, "cond_timedwait");
5360
5361 #ifdef ASSERT
5362 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5363 #endif
5364
5365 _counter = 0 ;
5366 status = pthread_mutex_unlock(_mutex) ;
5367 assert_status(status == 0, status, "invariant") ;
5368 // Paranoia to ensure our locked and lock-free paths interact
5369 // correctly with each other and Java-level accesses.
5370 OrderAccess::fence();
5371
5372 // If externally suspended while waiting, re-suspend
5373 if (jt->handle_special_suspend_equivalent_condition()) {
5374 jt->java_suspend_self();
5375 }
5376 }
5377
5378 void Parker::unpark() {
5379 int s, status ;
5380 status = pthread_mutex_lock(_mutex);
5381 assert (status == 0, "invariant") ;
5382 s = _counter;
5383 _counter = 1;
5384 if (s < 1) {
5385 if (WorkAroundNPTLTimedWaitHang) {
5386 status = pthread_cond_signal (_cond) ;
5387 assert (status == 0, "invariant") ;
5388 status = pthread_mutex_unlock(_mutex);
5389 assert (status == 0, "invariant") ;
5390 } else {
5391 status = pthread_mutex_unlock(_mutex);
5392 assert (status == 0, "invariant") ;
5393 status = pthread_cond_signal (_cond) ;
5394 assert (status == 0, "invariant") ;
5395 }
5396 } else {
5397 pthread_mutex_unlock(_mutex);
5398 assert (status == 0, "invariant") ;
5399 }
5400 }
5401
5402
5403 extern char** environ;
5404
5405 #ifndef __NR_fork
5406 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5407 #endif
5408
5409 #ifndef __NR_execve
5410 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5411 #endif
5412
5413 // Run the specified command in a separate process. Return its exit value,
5414 // or -1 on failure (e.g. can't fork a new process).
5415 // Unlike system(), this function can be called from signal handler. It
5416 // doesn't block SIGINT et al.
5417 int os::fork_and_exec(char* cmd) {
5418 const char * argv[4] = {"sh", "-c", cmd, NULL};
5419
5420 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5421 // pthread_atfork handlers and reset pthread library. All we need is a
5422 // separate process to execve. Make a direct syscall to fork process.
5423 // On IA64 there's no fork syscall, we have to use fork() and hope for
5424 // the best...
5425 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5426 IA64_ONLY(fork();)
5427
5428 if (pid < 0) {
5429 // fork failed
5430 return -1;
5431
5432 } else if (pid == 0) {
5433 // child process
5434
5435 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5436 // first to kill every thread on the thread list. Because this list is
5437 // not reset by fork() (see notes above), execve() will instead kill
5438 // every thread in the parent process. We know this is the only thread
5439 // in the new process, so make a system call directly.
5440 // IA64 should use normal execve() from glibc to match the glibc fork()
5441 // above.
5442 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5443 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5444
5445 // execve failed
5446 _exit(-1);
5447
5448 } else {
5449 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5450 // care about the actual exit code, for now.
5451
5452 int status;
5453
5454 // Wait for the child process to exit. This returns immediately if
5455 // the child has already exited. */
5456 while (waitpid(pid, &status, 0) < 0) {
5457 switch (errno) {
5458 case ECHILD: return 0;
5459 case EINTR: break;
5460 default: return -1;
5461 }
5462 }
5463
5464 if (WIFEXITED(status)) {
5465 // The child exited normally; get its exit code.
5466 return WEXITSTATUS(status);
5467 } else if (WIFSIGNALED(status)) {
5468 // The child exited because of a signal
5469 // The best value to return is 0x80 + signal number,
5470 // because that is what all Unix shells do, and because
5471 // it allows callers to distinguish between process exit and
5472 // process death by signal.
5473 return 0x80 + WTERMSIG(status);
5474 } else {
5475 // Unknown exit code; pass it through
5476 return status;
5477 }
5478 }
5479 }
5480
5481 // is_headless_jre()
5482 //
5483 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5484 // in order to report if we are running in a headless jre
5485 //
5486 // Since JDK8 xawt/libmawt.so was moved into the same directory
5487 // as libawt.so, and renamed libawt_xawt.so
5488 //
5489 bool os::is_headless_jre() {
5490 struct stat statbuf;
5491 char buf[MAXPATHLEN];
5492 char libmawtpath[MAXPATHLEN];
5493 const char *xawtstr = "/xawt/libmawt.so";
5494 const char *new_xawtstr = "/libawt_xawt.so";
5495 char *p;
5496
5497 // Get path to libjvm.so
5498 os::jvm_path(buf, sizeof(buf));
5499
5500 // Get rid of libjvm.so
5501 p = strrchr(buf, '/');
5502 if (p == NULL) return false;
5503 else *p = '\0';
5504
5505 // Get rid of client or server
5506 p = strrchr(buf, '/');
5507 if (p == NULL) return false;
5508 else *p = '\0';
5509
5510 // check xawt/libmawt.so
5511 strcpy(libmawtpath, buf);
5512 strcat(libmawtpath, xawtstr);
5513 if (::stat(libmawtpath, &statbuf) == 0) return false;
5514
5515 // check libawt_xawt.so
5516 strcpy(libmawtpath, buf);
5517 strcat(libmawtpath, new_xawtstr);
5518 if (::stat(libmawtpath, &statbuf) == 0) return false;
5519
5520 return true;
5521 }
5522
5523 // Get the default path to the core file
5524 // Returns the length of the string
5525 int os::get_core_path(char* buffer, size_t bufferSize) {
5526 const char* p = get_current_directory(buffer, bufferSize);
5527
5528 if (p == NULL) {
5529 assert(p != NULL, "failed to get current directory");
5530 return 0;
5531 }
5532
5533 return strlen(buffer);
5534 }
5535
5536 #ifdef JAVASE_EMBEDDED
5537 //
5538 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5539 //
5540 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5541
5542 // ctor
5543 //
5544 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5545 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5546 _fd = fd;
5547
5548 if (os::create_thread(this, os::os_thread)) {
5549 _memnotify_thread = this;
5550 os::set_priority(this, NearMaxPriority);
5551 os::start_thread(this);
5552 }
5553 }
5554
5555 // Where all the work gets done
5556 //
5557 void MemNotifyThread::run() {
5558 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5559
5560 // Set up the select arguments
5561 fd_set rfds;
5562 if (_fd != -1) {
5563 FD_ZERO(&rfds);
5564 FD_SET(_fd, &rfds);
5565 }
5566
5567 // Now wait for the mem_notify device to wake up
5568 while (1) {
5569 // Wait for the mem_notify device to signal us..
5570 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5571 if (rc == -1) {
5572 perror("select!\n");
5573 break;
5574 } else if (rc) {
5575 //ssize_t free_before = os::available_memory();
5576 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5577
5578 // The kernel is telling us there is not much memory left...
5579 // try to do something about that
5580
5581 // If we are not already in a GC, try one.
5582 if (!Universe::heap()->is_gc_active()) {
5583 Universe::heap()->collect(GCCause::_allocation_failure);
5584
5585 //ssize_t free_after = os::available_memory();
5586 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5587 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5588 }
5589 // We might want to do something like the following if we find the GC's are not helping...
5590 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5591 }
5592 }
5593 }
5594
5595 //
5596 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5597 //
5598 void MemNotifyThread::start() {
5599 int fd;
5600 fd = open ("/dev/mem_notify", O_RDONLY, 0);
5601 if (fd < 0) {
5602 return;
5603 }
5604
5605 if (memnotify_thread() == NULL) {
5606 new MemNotifyThread(fd);
5607 }
5608 }
5609 #endif // JAVASE_EMBEDDED