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 const char* os::get_current_directory(char *buf, int buflen) {
1667 return getcwd(buf, buflen);
1668 }
1669
1670 // check if addr is inside libjvm.so
1671 bool os::address_is_in_vm(address addr) {
1672 static address libjvm_base_addr;
1673 Dl_info dlinfo;
1674
1675 if (libjvm_base_addr == NULL) {
1676 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1677 libjvm_base_addr = (address)dlinfo.dli_fbase;
1678 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1679 }
1680
1681 if (dladdr((void *)addr, &dlinfo)) {
1682 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1683 }
1684
1685 return false;
1686 }
1687
1688 bool os::dll_address_to_function_name(address addr, char *buf,
1689 int buflen, int *offset) {
1690 Dl_info dlinfo;
1691
1692 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1693 if (buf != NULL) {
1694 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1695 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1696 }
1697 }
1698 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1699 return true;
1700 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1701 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1702 buf, buflen, offset, dlinfo.dli_fname)) {
1703 return true;
1704 }
1705 }
1706
1707 if (buf != NULL) buf[0] = '\0';
1708 if (offset != NULL) *offset = -1;
1709 return false;
1710 }
1711
1712 struct _address_to_library_name {
1713 address addr; // input : memory address
1714 size_t buflen; // size of fname
1715 char* fname; // output: library name
1716 address base; // library base addr
1717 };
1718
1719 static int address_to_library_name_callback(struct dl_phdr_info *info,
1720 size_t size, void *data) {
1721 int i;
1722 bool found = false;
1723 address libbase = NULL;
1724 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1725
1726 // iterate through all loadable segments
1727 for (i = 0; i < info->dlpi_phnum; i++) {
1728 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1729 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1730 // base address of a library is the lowest address of its loaded
1731 // segments.
1732 if (libbase == NULL || libbase > segbase) {
1733 libbase = segbase;
1734 }
1735 // see if 'addr' is within current segment
1736 if (segbase <= d->addr &&
1737 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1738 found = true;
1739 }
1740 }
1741 }
1742
1743 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1744 // so dll_address_to_library_name() can fall through to use dladdr() which
1745 // can figure out executable name from argv[0].
1746 if (found && info->dlpi_name && info->dlpi_name[0]) {
1747 d->base = libbase;
1748 if (d->fname) {
1749 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1750 }
1751 return 1;
1752 }
1753 return 0;
1754 }
1755
1756 bool os::dll_address_to_library_name(address addr, char* buf,
1757 int buflen, int* offset) {
1758 Dl_info dlinfo;
1759 struct _address_to_library_name data;
1760
1761 // There is a bug in old glibc dladdr() implementation that it could resolve
1762 // to wrong library name if the .so file has a base address != NULL. Here
1763 // we iterate through the program headers of all loaded libraries to find
1764 // out which library 'addr' really belongs to. This workaround can be
1765 // removed once the minimum requirement for glibc is moved to 2.3.x.
1766 data.addr = addr;
1767 data.fname = buf;
1768 data.buflen = buflen;
1769 data.base = NULL;
1770 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1771
1772 if (rslt) {
1773 // buf already contains library name
1774 if (offset) *offset = addr - data.base;
1775 return true;
1776 } else if (dladdr((void*)addr, &dlinfo)){
1777 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1778 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1779 return true;
1780 } else {
1781 if (buf) buf[0] = '\0';
1782 if (offset) *offset = -1;
1783 return false;
1784 }
1785 }
1786
1787 // Loads .dll/.so and
1788 // in case of error it checks if .dll/.so was built for the
1789 // same architecture as Hotspot is running on
1790
1791
1792 // Remember the stack's state. The Linux dynamic linker will change
1793 // the stack to 'executable' at most once, so we must safepoint only once.
1794 bool os::Linux::_stack_is_executable = false;
1795
1796 // VM operation that loads a library. This is necessary if stack protection
1797 // of the Java stacks can be lost during loading the library. If we
1798 // do not stop the Java threads, they can stack overflow before the stacks
1799 // are protected again.
1800 class VM_LinuxDllLoad: public VM_Operation {
1801 private:
1802 const char *_filename;
1803 char *_ebuf;
1804 int _ebuflen;
1805 void *_lib;
1806 public:
1807 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1808 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1809 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1810 void doit() {
1811 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1812 os::Linux::_stack_is_executable = true;
1813 }
1814 void* loaded_library() { return _lib; }
1815 };
1816
1817 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1818 {
1819 void * result = NULL;
1820 bool load_attempted = false;
1821
1822 // Check whether the library to load might change execution rights
1823 // of the stack. If they are changed, the protection of the stack
1824 // guard pages will be lost. We need a safepoint to fix this.
1825 //
1826 // See Linux man page execstack(8) for more info.
1827 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1828 ElfFile ef(filename);
1829 if (!ef.specifies_noexecstack()) {
1830 if (!is_init_completed()) {
1831 os::Linux::_stack_is_executable = true;
1832 // This is OK - No Java threads have been created yet, and hence no
1833 // stack guard pages to fix.
1834 //
1835 // This should happen only when you are building JDK7 using a very
1836 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1837 //
1838 // Dynamic loader will make all stacks executable after
1839 // this function returns, and will not do that again.
1840 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1841 } else {
1842 warning("You have loaded library %s which might have disabled stack guard. "
1843 "The VM will try to fix the stack guard now.\n"
1844 "It's highly recommended that you fix the library with "
1845 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1846 filename);
1847
1848 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1849 JavaThread *jt = JavaThread::current();
1850 if (jt->thread_state() != _thread_in_native) {
1851 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1852 // that requires ExecStack. Cannot enter safe point. Let's give up.
1853 warning("Unable to fix stack guard. Giving up.");
1854 } else {
1855 if (!LoadExecStackDllInVMThread) {
1856 // This is for the case where the DLL has an static
1857 // constructor function that executes JNI code. We cannot
1858 // load such DLLs in the VMThread.
1859 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1860 }
1861
1862 ThreadInVMfromNative tiv(jt);
1863 debug_only(VMNativeEntryWrapper vew;)
1864
1865 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1866 VMThread::execute(&op);
1867 if (LoadExecStackDllInVMThread) {
1868 result = op.loaded_library();
1869 }
1870 load_attempted = true;
1871 }
1872 }
1873 }
1874 }
1875
1876 if (!load_attempted) {
1877 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1878 }
1879
1880 if (result != NULL) {
1881 // Successful loading
1882 return result;
1883 }
1884
1885 Elf32_Ehdr elf_head;
1886 int diag_msg_max_length=ebuflen-strlen(ebuf);
1887 char* diag_msg_buf=ebuf+strlen(ebuf);
1888
1889 if (diag_msg_max_length==0) {
1890 // No more space in ebuf for additional diagnostics message
1891 return NULL;
1892 }
1893
1894
1895 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1896
1897 if (file_descriptor < 0) {
1898 // Can't open library, report dlerror() message
1899 return NULL;
1900 }
1901
1902 bool failed_to_read_elf_head=
1903 (sizeof(elf_head)!=
1904 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1905
1906 ::close(file_descriptor);
1907 if (failed_to_read_elf_head) {
1908 // file i/o error - report dlerror() msg
1909 return NULL;
1910 }
1911
1912 typedef struct {
1913 Elf32_Half code; // Actual value as defined in elf.h
1914 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1915 char elf_class; // 32 or 64 bit
1916 char endianess; // MSB or LSB
1917 char* name; // String representation
1918 } arch_t;
1919
1920 #ifndef EM_486
1921 #define EM_486 6 /* Intel 80486 */
1922 #endif
1923
1924 static const arch_t arch_array[]={
1925 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1926 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1927 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1928 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1929 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1930 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1931 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1932 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1933 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1934 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1935 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1936 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1937 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1938 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1939 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1940 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1941 };
1942
1943 #if (defined IA32)
1944 static Elf32_Half running_arch_code=EM_386;
1945 #elif (defined AMD64)
1946 static Elf32_Half running_arch_code=EM_X86_64;
1947 #elif (defined IA64)
1948 static Elf32_Half running_arch_code=EM_IA_64;
1949 #elif (defined __sparc) && (defined _LP64)
1950 static Elf32_Half running_arch_code=EM_SPARCV9;
1951 #elif (defined __sparc) && (!defined _LP64)
1952 static Elf32_Half running_arch_code=EM_SPARC;
1953 #elif (defined __powerpc64__)
1954 static Elf32_Half running_arch_code=EM_PPC64;
1955 #elif (defined __powerpc__)
1956 static Elf32_Half running_arch_code=EM_PPC;
1957 #elif (defined ARM)
1958 static Elf32_Half running_arch_code=EM_ARM;
1959 #elif (defined S390)
1960 static Elf32_Half running_arch_code=EM_S390;
1961 #elif (defined ALPHA)
1962 static Elf32_Half running_arch_code=EM_ALPHA;
1963 #elif (defined MIPSEL)
1964 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1965 #elif (defined PARISC)
1966 static Elf32_Half running_arch_code=EM_PARISC;
1967 #elif (defined MIPS)
1968 static Elf32_Half running_arch_code=EM_MIPS;
1969 #elif (defined M68K)
1970 static Elf32_Half running_arch_code=EM_68K;
1971 #else
1972 #error Method os::dll_load requires that one of following is defined:\
1973 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1974 #endif
1975
1976 // Identify compatability class for VM's architecture and library's architecture
1977 // Obtain string descriptions for architectures
1978
1979 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1980 int running_arch_index=-1;
1981
1982 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1983 if (running_arch_code == arch_array[i].code) {
1984 running_arch_index = i;
1985 }
1986 if (lib_arch.code == arch_array[i].code) {
1987 lib_arch.compat_class = arch_array[i].compat_class;
1988 lib_arch.name = arch_array[i].name;
1989 }
1990 }
1991
1992 assert(running_arch_index != -1,
1993 "Didn't find running architecture code (running_arch_code) in arch_array");
1994 if (running_arch_index == -1) {
1995 // Even though running architecture detection failed
1996 // we may still continue with reporting dlerror() message
1997 return NULL;
1998 }
1999
2000 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2001 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2002 return NULL;
2003 }
2004
2005 #ifndef S390
2006 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2007 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2008 return NULL;
2009 }
2010 #endif // !S390
2011
2012 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2013 if ( lib_arch.name!=NULL ) {
2014 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2015 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2016 lib_arch.name, arch_array[running_arch_index].name);
2017 } else {
2018 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2019 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2020 lib_arch.code,
2021 arch_array[running_arch_index].name);
2022 }
2023 }
2024
2025 return NULL;
2026 }
2027
2028 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2029 void * result = ::dlopen(filename, RTLD_LAZY);
2030 if (result == NULL) {
2031 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2032 ebuf[ebuflen-1] = '\0';
2033 }
2034 return result;
2035 }
2036
2037 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2038 void * result = NULL;
2039 if (LoadExecStackDllInVMThread) {
2040 result = dlopen_helper(filename, ebuf, ebuflen);
2041 }
2042
2043 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2044 // library that requires an executable stack, or which does not have this
2045 // stack attribute set, dlopen changes the stack attribute to executable. The
2046 // read protection of the guard pages gets lost.
2047 //
2048 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2049 // may have been queued at the same time.
2050
2051 if (!_stack_is_executable) {
2052 JavaThread *jt = Threads::first();
2053
2054 while (jt) {
2055 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2056 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2057 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2058 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2059 warning("Attempt to reguard stack yellow zone failed.");
2060 }
2061 }
2062 jt = jt->next();
2063 }
2064 }
2065
2066 return result;
2067 }
2068
2069 /*
2070 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2071 * chances are you might want to run the generated bits against glibc-2.0
2072 * libdl.so, so always use locking for any version of glibc.
2073 */
2074 void* os::dll_lookup(void* handle, const char* name) {
2075 pthread_mutex_lock(&dl_mutex);
2076 void* res = dlsym(handle, name);
2077 pthread_mutex_unlock(&dl_mutex);
2078 return res;
2079 }
2080
2081
2082 static bool _print_ascii_file(const char* filename, outputStream* st) {
2083 int fd = ::open(filename, O_RDONLY);
2084 if (fd == -1) {
2085 return false;
2086 }
2087
2088 char buf[32];
2089 int bytes;
2090 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2091 st->print_raw(buf, bytes);
2092 }
2093
2094 ::close(fd);
2095
2096 return true;
2097 }
2098
2099 void os::print_dll_info(outputStream *st) {
2100 st->print_cr("Dynamic libraries:");
2101
2102 char fname[32];
2103 pid_t pid = os::Linux::gettid();
2104
2105 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2106
2107 if (!_print_ascii_file(fname, st)) {
2108 st->print("Can not get library information for pid = %d\n", pid);
2109 }
2110 }
2111
2112 void os::print_os_info_brief(outputStream* st) {
2113 os::Linux::print_distro_info(st);
2114
2115 os::Posix::print_uname_info(st);
2116
2117 os::Linux::print_libversion_info(st);
2118
2119 }
2120
2121 void os::print_os_info(outputStream* st) {
2122 st->print("OS:");
2123
2124 os::Linux::print_distro_info(st);
2125
2126 os::Posix::print_uname_info(st);
2127
2128 // Print warning if unsafe chroot environment detected
2129 if (unsafe_chroot_detected) {
2130 st->print("WARNING!! ");
2131 st->print_cr(unstable_chroot_error);
2132 }
2133
2134 os::Linux::print_libversion_info(st);
2135
2136 os::Posix::print_rlimit_info(st);
2137
2138 os::Posix::print_load_average(st);
2139
2140 os::Linux::print_full_memory_info(st);
2141 }
2142
2143 // Try to identify popular distros.
2144 // Most Linux distributions have /etc/XXX-release file, which contains
2145 // the OS version string. Some have more than one /etc/XXX-release file
2146 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2147 // so the order is important.
2148 void os::Linux::print_distro_info(outputStream* st) {
2149 if (!_print_ascii_file("/etc/mandrake-release", st) &&
2150 !_print_ascii_file("/etc/sun-release", st) &&
2151 !_print_ascii_file("/etc/redhat-release", st) &&
2152 !_print_ascii_file("/etc/SuSE-release", st) &&
2153 !_print_ascii_file("/etc/turbolinux-release", st) &&
2154 !_print_ascii_file("/etc/gentoo-release", st) &&
2155 !_print_ascii_file("/etc/debian_version", st) &&
2156 !_print_ascii_file("/etc/ltib-release", st) &&
2157 !_print_ascii_file("/etc/angstrom-version", st)) {
2158 st->print("Linux");
2159 }
2160 st->cr();
2161 }
2162
2163 void os::Linux::print_libversion_info(outputStream* st) {
2164 // libc, pthread
2165 st->print("libc:");
2166 st->print(os::Linux::glibc_version()); st->print(" ");
2167 st->print(os::Linux::libpthread_version()); st->print(" ");
2168 if (os::Linux::is_LinuxThreads()) {
2169 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2170 }
2171 st->cr();
2172 }
2173
2174 void os::Linux::print_full_memory_info(outputStream* st) {
2175 st->print("\n/proc/meminfo:\n");
2176 _print_ascii_file("/proc/meminfo", st);
2177 st->cr();
2178 }
2179
2180 void os::print_memory_info(outputStream* st) {
2181
2182 st->print("Memory:");
2183 st->print(" %dk page", os::vm_page_size()>>10);
2184
2185 // values in struct sysinfo are "unsigned long"
2186 struct sysinfo si;
2187 sysinfo(&si);
2188
2189 st->print(", physical " UINT64_FORMAT "k",
2190 os::physical_memory() >> 10);
2191 st->print("(" UINT64_FORMAT "k free)",
2192 os::available_memory() >> 10);
2193 st->print(", swap " UINT64_FORMAT "k",
2194 ((jlong)si.totalswap * si.mem_unit) >> 10);
2195 st->print("(" UINT64_FORMAT "k free)",
2196 ((jlong)si.freeswap * si.mem_unit) >> 10);
2197 st->cr();
2198 }
2199
2200 void os::pd_print_cpu_info(outputStream* st) {
2201 st->print("\n/proc/cpuinfo:\n");
2202 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2203 st->print(" <Not Available>");
2204 }
2205 st->cr();
2206 }
2207
2208 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2209 // but they're the same for all the linux arch that we support
2210 // and they're the same for solaris but there's no common place to put this.
2211 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2212 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2213 "ILL_COPROC", "ILL_BADSTK" };
2214
2215 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2216 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2217 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2218
2219 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2220
2221 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2222
2223 void os::print_siginfo(outputStream* st, void* siginfo) {
2224 st->print("siginfo:");
2225
2226 const int buflen = 100;
2227 char buf[buflen];
2228 siginfo_t *si = (siginfo_t*)siginfo;
2229 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2230 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2231 st->print("si_errno=%s", buf);
2232 } else {
2233 st->print("si_errno=%d", si->si_errno);
2234 }
2235 const int c = si->si_code;
2236 assert(c > 0, "unexpected si_code");
2237 switch (si->si_signo) {
2238 case SIGILL:
2239 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2240 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2241 break;
2242 case SIGFPE:
2243 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2244 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2245 break;
2246 case SIGSEGV:
2247 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2248 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2249 break;
2250 case SIGBUS:
2251 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2252 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2253 break;
2254 default:
2255 st->print(", si_code=%d", si->si_code);
2256 // no si_addr
2257 }
2258
2259 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2260 UseSharedSpaces) {
2261 FileMapInfo* mapinfo = FileMapInfo::current_info();
2262 if (mapinfo->is_in_shared_space(si->si_addr)) {
2263 st->print("\n\nError accessing class data sharing archive." \
2264 " Mapped file inaccessible during execution, " \
2265 " possible disk/network problem.");
2266 }
2267 }
2268 st->cr();
2269 }
2270
2271
2272 static void print_signal_handler(outputStream* st, int sig,
2273 char* buf, size_t buflen);
2274
2275 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2276 st->print_cr("Signal Handlers:");
2277 print_signal_handler(st, SIGSEGV, buf, buflen);
2278 print_signal_handler(st, SIGBUS , buf, buflen);
2279 print_signal_handler(st, SIGFPE , buf, buflen);
2280 print_signal_handler(st, SIGPIPE, buf, buflen);
2281 print_signal_handler(st, SIGXFSZ, buf, buflen);
2282 print_signal_handler(st, SIGILL , buf, buflen);
2283 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2284 print_signal_handler(st, SR_signum, buf, buflen);
2285 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2286 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2287 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2288 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2289 }
2290
2291 static char saved_jvm_path[MAXPATHLEN] = {0};
2292
2293 // Find the full path to the current module, libjvm.so
2294 void os::jvm_path(char *buf, jint buflen) {
2295 // Error checking.
2296 if (buflen < MAXPATHLEN) {
2297 assert(false, "must use a large-enough buffer");
2298 buf[0] = '\0';
2299 return;
2300 }
2301 // Lazy resolve the path to current module.
2302 if (saved_jvm_path[0] != 0) {
2303 strcpy(buf, saved_jvm_path);
2304 return;
2305 }
2306
2307 char dli_fname[MAXPATHLEN];
2308 bool ret = dll_address_to_library_name(
2309 CAST_FROM_FN_PTR(address, os::jvm_path),
2310 dli_fname, sizeof(dli_fname), NULL);
2311 assert(ret != 0, "cannot locate libjvm");
2312 char *rp = realpath(dli_fname, buf);
2313 if (rp == NULL)
2314 return;
2315
2316 if (Arguments::created_by_gamma_launcher()) {
2317 // Support for the gamma launcher. Typical value for buf is
2318 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2319 // the right place in the string, then assume we are installed in a JDK and
2320 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2321 // up the path so it looks like libjvm.so is installed there (append a
2322 // fake suffix hotspot/libjvm.so).
2323 const char *p = buf + strlen(buf) - 1;
2324 for (int count = 0; p > buf && count < 5; ++count) {
2325 for (--p; p > buf && *p != '/'; --p)
2326 /* empty */ ;
2327 }
2328
2329 if (strncmp(p, "/jre/lib/", 9) != 0) {
2330 // Look for JAVA_HOME in the environment.
2331 char* java_home_var = ::getenv("JAVA_HOME");
2332 if (java_home_var != NULL && java_home_var[0] != 0) {
2333 char* jrelib_p;
2334 int len;
2335
2336 // Check the current module name "libjvm.so".
2337 p = strrchr(buf, '/');
2338 assert(strstr(p, "/libjvm") == p, "invalid library name");
2339
2340 rp = realpath(java_home_var, buf);
2341 if (rp == NULL)
2342 return;
2343
2344 // determine if this is a legacy image or modules image
2345 // modules image doesn't have "jre" subdirectory
2346 len = strlen(buf);
2347 jrelib_p = buf + len;
2348 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2349 if (0 != access(buf, F_OK)) {
2350 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2351 }
2352
2353 if (0 == access(buf, F_OK)) {
2354 // Use current module name "libjvm.so"
2355 len = strlen(buf);
2356 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2357 } else {
2358 // Go back to path of .so
2359 rp = realpath(dli_fname, buf);
2360 if (rp == NULL)
2361 return;
2362 }
2363 }
2364 }
2365 }
2366
2367 strcpy(saved_jvm_path, buf);
2368 }
2369
2370 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2371 // no prefix required, not even "_"
2372 }
2373
2374 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2375 // no suffix required
2376 }
2377
2378 ////////////////////////////////////////////////////////////////////////////////
2379 // sun.misc.Signal support
2380
2381 static volatile jint sigint_count = 0;
2382
2383 static void
2384 UserHandler(int sig, void *siginfo, void *context) {
2385 // 4511530 - sem_post is serialized and handled by the manager thread. When
2386 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2387 // don't want to flood the manager thread with sem_post requests.
2388 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2389 return;
2390
2391 // Ctrl-C is pressed during error reporting, likely because the error
2392 // handler fails to abort. Let VM die immediately.
2393 if (sig == SIGINT && is_error_reported()) {
2394 os::die();
2395 }
2396
2397 os::signal_notify(sig);
2398 }
2399
2400 void* os::user_handler() {
2401 return CAST_FROM_FN_PTR(void*, UserHandler);
2402 }
2403
2404 extern "C" {
2405 typedef void (*sa_handler_t)(int);
2406 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2407 }
2408
2409 void* os::signal(int signal_number, void* handler) {
2410 struct sigaction sigAct, oldSigAct;
2411
2412 sigfillset(&(sigAct.sa_mask));
2413 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2414 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2415
2416 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2417 // -1 means registration failed
2418 return (void *)-1;
2419 }
2420
2421 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2422 }
2423
2424 void os::signal_raise(int signal_number) {
2425 ::raise(signal_number);
2426 }
2427
2428 /*
2429 * The following code is moved from os.cpp for making this
2430 * code platform specific, which it is by its very nature.
2431 */
2432
2433 // Will be modified when max signal is changed to be dynamic
2434 int os::sigexitnum_pd() {
2435 return NSIG;
2436 }
2437
2438 // a counter for each possible signal value
2439 static volatile jint pending_signals[NSIG+1] = { 0 };
2440
2441 // Linux(POSIX) specific hand shaking semaphore.
2442 static sem_t sig_sem;
2443
2444 void os::signal_init_pd() {
2445 // Initialize signal structures
2446 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2447
2448 // Initialize signal semaphore
2449 ::sem_init(&sig_sem, 0, 0);
2450 }
2451
2452 void os::signal_notify(int sig) {
2453 Atomic::inc(&pending_signals[sig]);
2454 ::sem_post(&sig_sem);
2455 }
2456
2457 static int check_pending_signals(bool wait) {
2458 Atomic::store(0, &sigint_count);
2459 for (;;) {
2460 for (int i = 0; i < NSIG + 1; i++) {
2461 jint n = pending_signals[i];
2462 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2463 return i;
2464 }
2465 }
2466 if (!wait) {
2467 return -1;
2468 }
2469 JavaThread *thread = JavaThread::current();
2470 ThreadBlockInVM tbivm(thread);
2471
2472 bool threadIsSuspended;
2473 do {
2474 thread->set_suspend_equivalent();
2475 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2476 ::sem_wait(&sig_sem);
2477
2478 // were we externally suspended while we were waiting?
2479 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2480 if (threadIsSuspended) {
2481 //
2482 // The semaphore has been incremented, but while we were waiting
2483 // another thread suspended us. We don't want to continue running
2484 // while suspended because that would surprise the thread that
2485 // suspended us.
2486 //
2487 ::sem_post(&sig_sem);
2488
2489 thread->java_suspend_self();
2490 }
2491 } while (threadIsSuspended);
2492 }
2493 }
2494
2495 int os::signal_lookup() {
2496 return check_pending_signals(false);
2497 }
2498
2499 int os::signal_wait() {
2500 return check_pending_signals(true);
2501 }
2502
2503 ////////////////////////////////////////////////////////////////////////////////
2504 // Virtual Memory
2505
2506 int os::vm_page_size() {
2507 // Seems redundant as all get out
2508 assert(os::Linux::page_size() != -1, "must call os::init");
2509 return os::Linux::page_size();
2510 }
2511
2512 // Solaris allocates memory by pages.
2513 int os::vm_allocation_granularity() {
2514 assert(os::Linux::page_size() != -1, "must call os::init");
2515 return os::Linux::page_size();
2516 }
2517
2518 // Rationale behind this function:
2519 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2520 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2521 // samples for JITted code. Here we create private executable mapping over the code cache
2522 // and then we can use standard (well, almost, as mapping can change) way to provide
2523 // info for the reporting script by storing timestamp and location of symbol
2524 void linux_wrap_code(char* base, size_t size) {
2525 static volatile jint cnt = 0;
2526
2527 if (!UseOprofile) {
2528 return;
2529 }
2530
2531 char buf[PATH_MAX+1];
2532 int num = Atomic::add(1, &cnt);
2533
2534 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2535 os::get_temp_directory(), os::current_process_id(), num);
2536 unlink(buf);
2537
2538 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2539
2540 if (fd != -1) {
2541 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2542 if (rv != (off_t)-1) {
2543 if (::write(fd, "", 1) == 1) {
2544 mmap(base, size,
2545 PROT_READ|PROT_WRITE|PROT_EXEC,
2546 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2547 }
2548 }
2549 ::close(fd);
2550 unlink(buf);
2551 }
2552 }
2553
2554 // NOTE: Linux kernel does not really reserve the pages for us.
2555 // All it does is to check if there are enough free pages
2556 // left at the time of mmap(). This could be a potential
2557 // problem.
2558 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2559 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2560 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2561 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2562 if (res != (uintptr_t) MAP_FAILED) {
2563 if (UseNUMAInterleaving) {
2564 numa_make_global(addr, size);
2565 }
2566 return true;
2567 }
2568 return false;
2569 }
2570
2571 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2572 #ifndef MAP_HUGETLB
2573 #define MAP_HUGETLB 0x40000
2574 #endif
2575
2576 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2577 #ifndef MADV_HUGEPAGE
2578 #define MADV_HUGEPAGE 14
2579 #endif
2580
2581 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2582 bool exec) {
2583 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2584 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2585 uintptr_t res =
2586 (uintptr_t) ::mmap(addr, size, prot,
2587 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2588 -1, 0);
2589 if (res != (uintptr_t) MAP_FAILED) {
2590 if (UseNUMAInterleaving) {
2591 numa_make_global(addr, size);
2592 }
2593 return true;
2594 }
2595 // Fall through and try to use small pages
2596 }
2597
2598 if (commit_memory(addr, size, exec)) {
2599 realign_memory(addr, size, alignment_hint);
2600 return true;
2601 }
2602 return false;
2603 }
2604
2605 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2606 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2607 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2608 // be supported or the memory may already be backed by huge pages.
2609 ::madvise(addr, bytes, MADV_HUGEPAGE);
2610 }
2611 }
2612
2613 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2614 // This method works by doing an mmap over an existing mmaping and effectively discarding
2615 // the existing pages. However it won't work for SHM-based large pages that cannot be
2616 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2617 // small pages on top of the SHM segment. This method always works for small pages, so we
2618 // allow that in any case.
2619 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2620 commit_memory(addr, bytes, alignment_hint, false);
2621 }
2622 }
2623
2624 void os::numa_make_global(char *addr, size_t bytes) {
2625 Linux::numa_interleave_memory(addr, bytes);
2626 }
2627
2628 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2629 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2630 }
2631
2632 bool os::numa_topology_changed() { return false; }
2633
2634 size_t os::numa_get_groups_num() {
2635 int max_node = Linux::numa_max_node();
2636 return max_node > 0 ? max_node + 1 : 1;
2637 }
2638
2639 int os::numa_get_group_id() {
2640 int cpu_id = Linux::sched_getcpu();
2641 if (cpu_id != -1) {
2642 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2643 if (lgrp_id != -1) {
2644 return lgrp_id;
2645 }
2646 }
2647 return 0;
2648 }
2649
2650 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2651 for (size_t i = 0; i < size; i++) {
2652 ids[i] = i;
2653 }
2654 return size;
2655 }
2656
2657 bool os::get_page_info(char *start, page_info* info) {
2658 return false;
2659 }
2660
2661 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2662 return end;
2663 }
2664
2665
2666 int os::Linux::sched_getcpu_syscall(void) {
2667 unsigned int cpu;
2668 int retval = -1;
2669
2670 #if defined(IA32)
2671 # ifndef SYS_getcpu
2672 # define SYS_getcpu 318
2673 # endif
2674 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2675 #elif defined(AMD64)
2676 // Unfortunately we have to bring all these macros here from vsyscall.h
2677 // to be able to compile on old linuxes.
2678 # define __NR_vgetcpu 2
2679 # define VSYSCALL_START (-10UL << 20)
2680 # define VSYSCALL_SIZE 1024
2681 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2682 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2683 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2684 retval = vgetcpu(&cpu, NULL, NULL);
2685 #endif
2686
2687 return (retval == -1) ? retval : cpu;
2688 }
2689
2690 // Something to do with the numa-aware allocator needs these symbols
2691 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2692 extern "C" JNIEXPORT void numa_error(char *where) { }
2693 extern "C" JNIEXPORT int fork1() { return fork(); }
2694
2695
2696 // If we are running with libnuma version > 2, then we should
2697 // be trying to use symbols with versions 1.1
2698 // If we are running with earlier version, which did not have symbol versions,
2699 // we should use the base version.
2700 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2701 void *f = dlvsym(handle, name, "libnuma_1.1");
2702 if (f == NULL) {
2703 f = dlsym(handle, name);
2704 }
2705 return f;
2706 }
2707
2708 bool os::Linux::libnuma_init() {
2709 // sched_getcpu() should be in libc.
2710 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2711 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2712
2713 // If it's not, try a direct syscall.
2714 if (sched_getcpu() == -1)
2715 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2716
2717 if (sched_getcpu() != -1) { // Does it work?
2718 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2719 if (handle != NULL) {
2720 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2721 libnuma_dlsym(handle, "numa_node_to_cpus")));
2722 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2723 libnuma_dlsym(handle, "numa_max_node")));
2724 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2725 libnuma_dlsym(handle, "numa_available")));
2726 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2727 libnuma_dlsym(handle, "numa_tonode_memory")));
2728 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2729 libnuma_dlsym(handle, "numa_interleave_memory")));
2730
2731
2732 if (numa_available() != -1) {
2733 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2734 // Create a cpu -> node mapping
2735 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2736 rebuild_cpu_to_node_map();
2737 return true;
2738 }
2739 }
2740 }
2741 return false;
2742 }
2743
2744 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2745 // The table is later used in get_node_by_cpu().
2746 void os::Linux::rebuild_cpu_to_node_map() {
2747 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2748 // in libnuma (possible values are starting from 16,
2749 // and continuing up with every other power of 2, but less
2750 // than the maximum number of CPUs supported by kernel), and
2751 // is a subject to change (in libnuma version 2 the requirements
2752 // are more reasonable) we'll just hardcode the number they use
2753 // in the library.
2754 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2755
2756 size_t cpu_num = os::active_processor_count();
2757 size_t cpu_map_size = NCPUS / BitsPerCLong;
2758 size_t cpu_map_valid_size =
2759 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2760
2761 cpu_to_node()->clear();
2762 cpu_to_node()->at_grow(cpu_num - 1);
2763 size_t node_num = numa_get_groups_num();
2764
2765 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2766 for (size_t i = 0; i < node_num; i++) {
2767 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2768 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2769 if (cpu_map[j] != 0) {
2770 for (size_t k = 0; k < BitsPerCLong; k++) {
2771 if (cpu_map[j] & (1UL << k)) {
2772 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2773 }
2774 }
2775 }
2776 }
2777 }
2778 }
2779 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2780 }
2781
2782 int os::Linux::get_node_by_cpu(int cpu_id) {
2783 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2784 return cpu_to_node()->at(cpu_id);
2785 }
2786 return -1;
2787 }
2788
2789 GrowableArray<int>* os::Linux::_cpu_to_node;
2790 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2791 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2792 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2793 os::Linux::numa_available_func_t os::Linux::_numa_available;
2794 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2795 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2796 unsigned long* os::Linux::_numa_all_nodes;
2797
2798 bool os::pd_uncommit_memory(char* addr, size_t size) {
2799 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2800 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2801 return res != (uintptr_t) MAP_FAILED;
2802 }
2803
2804 // Linux uses a growable mapping for the stack, and if the mapping for
2805 // the stack guard pages is not removed when we detach a thread the
2806 // stack cannot grow beyond the pages where the stack guard was
2807 // mapped. If at some point later in the process the stack expands to
2808 // that point, the Linux kernel cannot expand the stack any further
2809 // because the guard pages are in the way, and a segfault occurs.
2810 //
2811 // However, it's essential not to split the stack region by unmapping
2812 // a region (leaving a hole) that's already part of the stack mapping,
2813 // so if the stack mapping has already grown beyond the guard pages at
2814 // the time we create them, we have to truncate the stack mapping.
2815 // So, we need to know the extent of the stack mapping when
2816 // create_stack_guard_pages() is called.
2817
2818 // Find the bounds of the stack mapping. Return true for success.
2819 //
2820 // We only need this for stacks that are growable: at the time of
2821 // writing thread stacks don't use growable mappings (i.e. those
2822 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2823 // only applies to the main thread.
2824
2825 static
2826 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2827
2828 char buf[128];
2829 int fd, sz;
2830
2831 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2832 return false;
2833 }
2834
2835 const char kw[] = "[stack]";
2836 const int kwlen = sizeof(kw)-1;
2837
2838 // Address part of /proc/self/maps couldn't be more than 128 bytes
2839 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2840 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2841 // Extract addresses
2842 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2843 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2844 if (sp >= *bottom && sp <= *top) {
2845 ::close(fd);
2846 return true;
2847 }
2848 }
2849 }
2850 }
2851
2852 ::close(fd);
2853 return false;
2854 }
2855
2856
2857 // If the (growable) stack mapping already extends beyond the point
2858 // where we're going to put our guard pages, truncate the mapping at
2859 // that point by munmap()ping it. This ensures that when we later
2860 // munmap() the guard pages we don't leave a hole in the stack
2861 // mapping. This only affects the main/initial thread, but guard
2862 // against future OS changes
2863 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2864 uintptr_t stack_extent, stack_base;
2865 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2866 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2867 assert(os::Linux::is_initial_thread(),
2868 "growable stack in non-initial thread");
2869 if (stack_extent < (uintptr_t)addr)
2870 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2871 }
2872
2873 return os::commit_memory(addr, size);
2874 }
2875
2876 // If this is a growable mapping, remove the guard pages entirely by
2877 // munmap()ping them. If not, just call uncommit_memory(). This only
2878 // affects the main/initial thread, but guard against future OS changes
2879 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2880 uintptr_t stack_extent, stack_base;
2881 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2882 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2883 assert(os::Linux::is_initial_thread(),
2884 "growable stack in non-initial thread");
2885
2886 return ::munmap(addr, size) == 0;
2887 }
2888
2889 return os::uncommit_memory(addr, size);
2890 }
2891
2892 static address _highest_vm_reserved_address = NULL;
2893
2894 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2895 // at 'requested_addr'. If there are existing memory mappings at the same
2896 // location, however, they will be overwritten. If 'fixed' is false,
2897 // 'requested_addr' is only treated as a hint, the return value may or
2898 // may not start from the requested address. Unlike Linux mmap(), this
2899 // function returns NULL to indicate failure.
2900 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2901 char * addr;
2902 int flags;
2903
2904 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2905 if (fixed) {
2906 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2907 flags |= MAP_FIXED;
2908 }
2909
2910 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2911 // to PROT_EXEC if executable when we commit the page.
2912 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2913 flags, -1, 0);
2914
2915 if (addr != MAP_FAILED) {
2916 // anon_mmap() should only get called during VM initialization,
2917 // don't need lock (actually we can skip locking even it can be called
2918 // from multiple threads, because _highest_vm_reserved_address is just a
2919 // hint about the upper limit of non-stack memory regions.)
2920 if ((address)addr + bytes > _highest_vm_reserved_address) {
2921 _highest_vm_reserved_address = (address)addr + bytes;
2922 }
2923 }
2924
2925 return addr == MAP_FAILED ? NULL : addr;
2926 }
2927
2928 // Don't update _highest_vm_reserved_address, because there might be memory
2929 // regions above addr + size. If so, releasing a memory region only creates
2930 // a hole in the address space, it doesn't help prevent heap-stack collision.
2931 //
2932 static int anon_munmap(char * addr, size_t size) {
2933 return ::munmap(addr, size) == 0;
2934 }
2935
2936 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2937 size_t alignment_hint) {
2938 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2939 }
2940
2941 bool os::pd_release_memory(char* addr, size_t size) {
2942 return anon_munmap(addr, size);
2943 }
2944
2945 static address highest_vm_reserved_address() {
2946 return _highest_vm_reserved_address;
2947 }
2948
2949 static bool linux_mprotect(char* addr, size_t size, int prot) {
2950 // Linux wants the mprotect address argument to be page aligned.
2951 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2952
2953 // According to SUSv3, mprotect() should only be used with mappings
2954 // established by mmap(), and mmap() always maps whole pages. Unaligned
2955 // 'addr' likely indicates problem in the VM (e.g. trying to change
2956 // protection of malloc'ed or statically allocated memory). Check the
2957 // caller if you hit this assert.
2958 assert(addr == bottom, "sanity check");
2959
2960 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2961 return ::mprotect(bottom, size, prot) == 0;
2962 }
2963
2964 // Set protections specified
2965 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2966 bool is_committed) {
2967 unsigned int p = 0;
2968 switch (prot) {
2969 case MEM_PROT_NONE: p = PROT_NONE; break;
2970 case MEM_PROT_READ: p = PROT_READ; break;
2971 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2972 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2973 default:
2974 ShouldNotReachHere();
2975 }
2976 // is_committed is unused.
2977 return linux_mprotect(addr, bytes, p);
2978 }
2979
2980 bool os::guard_memory(char* addr, size_t size) {
2981 return linux_mprotect(addr, size, PROT_NONE);
2982 }
2983
2984 bool os::unguard_memory(char* addr, size_t size) {
2985 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2986 }
2987
2988 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2989 bool result = false;
2990 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2991 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2992 -1, 0);
2993
2994 if (p != (void *) -1) {
2995 // We don't know if this really is a huge page or not.
2996 FILE *fp = fopen("/proc/self/maps", "r");
2997 if (fp) {
2998 while (!feof(fp)) {
2999 char chars[257];
3000 long x = 0;
3001 if (fgets(chars, sizeof(chars), fp)) {
3002 if (sscanf(chars, "%lx-%*x", &x) == 1
3003 && x == (long)p) {
3004 if (strstr (chars, "hugepage")) {
3005 result = true;
3006 break;
3007 }
3008 }
3009 }
3010 }
3011 fclose(fp);
3012 }
3013 munmap (p, page_size);
3014 if (result)
3015 return true;
3016 }
3017
3018 if (warn) {
3019 warning("HugeTLBFS is not supported by the operating system.");
3020 }
3021
3022 return result;
3023 }
3024
3025 /*
3026 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3027 *
3028 * From the coredump_filter documentation:
3029 *
3030 * - (bit 0) anonymous private memory
3031 * - (bit 1) anonymous shared memory
3032 * - (bit 2) file-backed private memory
3033 * - (bit 3) file-backed shared memory
3034 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3035 * effective only if the bit 2 is cleared)
3036 * - (bit 5) hugetlb private memory
3037 * - (bit 6) hugetlb shared memory
3038 */
3039 static void set_coredump_filter(void) {
3040 FILE *f;
3041 long cdm;
3042
3043 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3044 return;
3045 }
3046
3047 if (fscanf(f, "%lx", &cdm) != 1) {
3048 fclose(f);
3049 return;
3050 }
3051
3052 rewind(f);
3053
3054 if ((cdm & LARGEPAGES_BIT) == 0) {
3055 cdm |= LARGEPAGES_BIT;
3056 fprintf(f, "%#lx", cdm);
3057 }
3058
3059 fclose(f);
3060 }
3061
3062 // Large page support
3063
3064 static size_t _large_page_size = 0;
3065
3066 void os::large_page_init() {
3067 if (!UseLargePages) {
3068 UseHugeTLBFS = false;
3069 UseSHM = false;
3070 return;
3071 }
3072
3073 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3074 // If UseLargePages is specified on the command line try both methods,
3075 // if it's default, then try only HugeTLBFS.
3076 if (FLAG_IS_DEFAULT(UseLargePages)) {
3077 UseHugeTLBFS = true;
3078 } else {
3079 UseHugeTLBFS = UseSHM = true;
3080 }
3081 }
3082
3083 if (LargePageSizeInBytes) {
3084 _large_page_size = LargePageSizeInBytes;
3085 } else {
3086 // large_page_size on Linux is used to round up heap size. x86 uses either
3087 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3088 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3089 // page as large as 256M.
3090 //
3091 // Here we try to figure out page size by parsing /proc/meminfo and looking
3092 // for a line with the following format:
3093 // Hugepagesize: 2048 kB
3094 //
3095 // If we can't determine the value (e.g. /proc is not mounted, or the text
3096 // format has been changed), we'll use the largest page size supported by
3097 // the processor.
3098
3099 #ifndef ZERO
3100 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3101 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3102 #endif // ZERO
3103
3104 FILE *fp = fopen("/proc/meminfo", "r");
3105 if (fp) {
3106 while (!feof(fp)) {
3107 int x = 0;
3108 char buf[16];
3109 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3110 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3111 _large_page_size = x * K;
3112 break;
3113 }
3114 } else {
3115 // skip to next line
3116 for (;;) {
3117 int ch = fgetc(fp);
3118 if (ch == EOF || ch == (int)'\n') break;
3119 }
3120 }
3121 }
3122 fclose(fp);
3123 }
3124 }
3125
3126 // print a warning if any large page related flag is specified on command line
3127 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3128
3129 const size_t default_page_size = (size_t)Linux::page_size();
3130 if (_large_page_size > default_page_size) {
3131 _page_sizes[0] = _large_page_size;
3132 _page_sizes[1] = default_page_size;
3133 _page_sizes[2] = 0;
3134 }
3135 UseHugeTLBFS = UseHugeTLBFS &&
3136 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3137
3138 if (UseHugeTLBFS)
3139 UseSHM = false;
3140
3141 UseLargePages = UseHugeTLBFS || UseSHM;
3142
3143 set_coredump_filter();
3144 }
3145
3146 #ifndef SHM_HUGETLB
3147 #define SHM_HUGETLB 04000
3148 #endif
3149
3150 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3151 // "exec" is passed in but not used. Creating the shared image for
3152 // the code cache doesn't have an SHM_X executable permission to check.
3153 assert(UseLargePages && UseSHM, "only for SHM large pages");
3154
3155 key_t key = IPC_PRIVATE;
3156 char *addr;
3157
3158 bool warn_on_failure = UseLargePages &&
3159 (!FLAG_IS_DEFAULT(UseLargePages) ||
3160 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3161 );
3162 char msg[128];
3163
3164 // Create a large shared memory region to attach to based on size.
3165 // Currently, size is the total size of the heap
3166 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3167 if (shmid == -1) {
3168 // Possible reasons for shmget failure:
3169 // 1. shmmax is too small for Java heap.
3170 // > check shmmax value: cat /proc/sys/kernel/shmmax
3171 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3172 // 2. not enough large page memory.
3173 // > check available large pages: cat /proc/meminfo
3174 // > increase amount of large pages:
3175 // echo new_value > /proc/sys/vm/nr_hugepages
3176 // Note 1: different Linux may use different name for this property,
3177 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3178 // Note 2: it's possible there's enough physical memory available but
3179 // they are so fragmented after a long run that they can't
3180 // coalesce into large pages. Try to reserve large pages when
3181 // the system is still "fresh".
3182 if (warn_on_failure) {
3183 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3184 warning(msg);
3185 }
3186 return NULL;
3187 }
3188
3189 // attach to the region
3190 addr = (char*)shmat(shmid, req_addr, 0);
3191 int err = errno;
3192
3193 // Remove shmid. If shmat() is successful, the actual shared memory segment
3194 // will be deleted when it's detached by shmdt() or when the process
3195 // terminates. If shmat() is not successful this will remove the shared
3196 // segment immediately.
3197 shmctl(shmid, IPC_RMID, NULL);
3198
3199 if ((intptr_t)addr == -1) {
3200 if (warn_on_failure) {
3201 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3202 warning(msg);
3203 }
3204 return NULL;
3205 }
3206
3207 if ((addr != NULL) && UseNUMAInterleaving) {
3208 numa_make_global(addr, bytes);
3209 }
3210
3211 // The memory is committed
3212 address pc = CALLER_PC;
3213 MemTracker::record_virtual_memory_reserve((address)addr, bytes, pc);
3214 MemTracker::record_virtual_memory_commit((address)addr, bytes, pc);
3215
3216 return addr;
3217 }
3218
3219 bool os::release_memory_special(char* base, size_t bytes) {
3220 // detaching the SHM segment will also delete it, see reserve_memory_special()
3221 int rslt = shmdt(base);
3222 if (rslt == 0) {
3223 MemTracker::record_virtual_memory_uncommit((address)base, bytes);
3224 MemTracker::record_virtual_memory_release((address)base, bytes);
3225 return true;
3226 } else {
3227 return false;
3228 }
3229 }
3230
3231 size_t os::large_page_size() {
3232 return _large_page_size;
3233 }
3234
3235 // HugeTLBFS allows application to commit large page memory on demand;
3236 // with SysV SHM the entire memory region must be allocated as shared
3237 // memory.
3238 bool os::can_commit_large_page_memory() {
3239 return UseHugeTLBFS;
3240 }
3241
3242 bool os::can_execute_large_page_memory() {
3243 return UseHugeTLBFS;
3244 }
3245
3246 // Reserve memory at an arbitrary address, only if that area is
3247 // available (and not reserved for something else).
3248
3249 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3250 const int max_tries = 10;
3251 char* base[max_tries];
3252 size_t size[max_tries];
3253 const size_t gap = 0x000000;
3254
3255 // Assert only that the size is a multiple of the page size, since
3256 // that's all that mmap requires, and since that's all we really know
3257 // about at this low abstraction level. If we need higher alignment,
3258 // we can either pass an alignment to this method or verify alignment
3259 // in one of the methods further up the call chain. See bug 5044738.
3260 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3261
3262 // Repeatedly allocate blocks until the block is allocated at the
3263 // right spot. Give up after max_tries. Note that reserve_memory() will
3264 // automatically update _highest_vm_reserved_address if the call is
3265 // successful. The variable tracks the highest memory address every reserved
3266 // by JVM. It is used to detect heap-stack collision if running with
3267 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3268 // space than needed, it could confuse the collision detecting code. To
3269 // solve the problem, save current _highest_vm_reserved_address and
3270 // calculate the correct value before return.
3271 address old_highest = _highest_vm_reserved_address;
3272
3273 // Linux mmap allows caller to pass an address as hint; give it a try first,
3274 // if kernel honors the hint then we can return immediately.
3275 char * addr = anon_mmap(requested_addr, bytes, false);
3276 if (addr == requested_addr) {
3277 return requested_addr;
3278 }
3279
3280 if (addr != NULL) {
3281 // mmap() is successful but it fails to reserve at the requested address
3282 anon_munmap(addr, bytes);
3283 }
3284
3285 int i;
3286 for (i = 0; i < max_tries; ++i) {
3287 base[i] = reserve_memory(bytes);
3288
3289 if (base[i] != NULL) {
3290 // Is this the block we wanted?
3291 if (base[i] == requested_addr) {
3292 size[i] = bytes;
3293 break;
3294 }
3295
3296 // Does this overlap the block we wanted? Give back the overlapped
3297 // parts and try again.
3298
3299 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3300 if (top_overlap >= 0 && top_overlap < bytes) {
3301 unmap_memory(base[i], top_overlap);
3302 base[i] += top_overlap;
3303 size[i] = bytes - top_overlap;
3304 } else {
3305 size_t bottom_overlap = base[i] + bytes - requested_addr;
3306 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3307 unmap_memory(requested_addr, bottom_overlap);
3308 size[i] = bytes - bottom_overlap;
3309 } else {
3310 size[i] = bytes;
3311 }
3312 }
3313 }
3314 }
3315
3316 // Give back the unused reserved pieces.
3317
3318 for (int j = 0; j < i; ++j) {
3319 if (base[j] != NULL) {
3320 unmap_memory(base[j], size[j]);
3321 }
3322 }
3323
3324 if (i < max_tries) {
3325 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3326 return requested_addr;
3327 } else {
3328 _highest_vm_reserved_address = old_highest;
3329 return NULL;
3330 }
3331 }
3332
3333 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3334 return ::read(fd, buf, nBytes);
3335 }
3336
3337 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3338 // Solaris uses poll(), linux uses park().
3339 // Poll() is likely a better choice, assuming that Thread.interrupt()
3340 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3341 // SIGSEGV, see 4355769.
3342
3343 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3344 assert(thread == Thread::current(), "thread consistency check");
3345
3346 ParkEvent * const slp = thread->_SleepEvent ;
3347 slp->reset() ;
3348 OrderAccess::fence() ;
3349
3350 if (interruptible) {
3351 jlong prevtime = javaTimeNanos();
3352
3353 for (;;) {
3354 if (os::is_interrupted(thread, true)) {
3355 return OS_INTRPT;
3356 }
3357
3358 jlong newtime = javaTimeNanos();
3359
3360 if (newtime - prevtime < 0) {
3361 // time moving backwards, should only happen if no monotonic clock
3362 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3363 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3364 } else {
3365 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3366 }
3367
3368 if(millis <= 0) {
3369 return OS_OK;
3370 }
3371
3372 prevtime = newtime;
3373
3374 {
3375 assert(thread->is_Java_thread(), "sanity check");
3376 JavaThread *jt = (JavaThread *) thread;
3377 ThreadBlockInVM tbivm(jt);
3378 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3379
3380 jt->set_suspend_equivalent();
3381 // cleared by handle_special_suspend_equivalent_condition() or
3382 // java_suspend_self() via check_and_wait_while_suspended()
3383
3384 slp->park(millis);
3385
3386 // were we externally suspended while we were waiting?
3387 jt->check_and_wait_while_suspended();
3388 }
3389 }
3390 } else {
3391 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3392 jlong prevtime = javaTimeNanos();
3393
3394 for (;;) {
3395 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3396 // the 1st iteration ...
3397 jlong newtime = javaTimeNanos();
3398
3399 if (newtime - prevtime < 0) {
3400 // time moving backwards, should only happen if no monotonic clock
3401 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3402 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3403 } else {
3404 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3405 }
3406
3407 if(millis <= 0) break ;
3408
3409 prevtime = newtime;
3410 slp->park(millis);
3411 }
3412 return OS_OK ;
3413 }
3414 }
3415
3416 int os::naked_sleep() {
3417 // %% make the sleep time an integer flag. for now use 1 millisec.
3418 return os::sleep(Thread::current(), 1, false);
3419 }
3420
3421 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3422 void os::infinite_sleep() {
3423 while (true) { // sleep forever ...
3424 ::sleep(100); // ... 100 seconds at a time
3425 }
3426 }
3427
3428 // Used to convert frequent JVM_Yield() to nops
3429 bool os::dont_yield() {
3430 return DontYieldALot;
3431 }
3432
3433 void os::yield() {
3434 sched_yield();
3435 }
3436
3437 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3438
3439 void os::yield_all(int attempts) {
3440 // Yields to all threads, including threads with lower priorities
3441 // Threads on Linux are all with same priority. The Solaris style
3442 // os::yield_all() with nanosleep(1ms) is not necessary.
3443 sched_yield();
3444 }
3445
3446 // Called from the tight loops to possibly influence time-sharing heuristics
3447 void os::loop_breaker(int attempts) {
3448 os::yield_all(attempts);
3449 }
3450
3451 ////////////////////////////////////////////////////////////////////////////////
3452 // thread priority support
3453
3454 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3455 // only supports dynamic priority, static priority must be zero. For real-time
3456 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3457 // However, for large multi-threaded applications, SCHED_RR is not only slower
3458 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3459 // of 5 runs - Sep 2005).
3460 //
3461 // The following code actually changes the niceness of kernel-thread/LWP. It
3462 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3463 // not the entire user process, and user level threads are 1:1 mapped to kernel
3464 // threads. It has always been the case, but could change in the future. For
3465 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3466 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3467
3468 int os::java_to_os_priority[CriticalPriority + 1] = {
3469 19, // 0 Entry should never be used
3470
3471 4, // 1 MinPriority
3472 3, // 2
3473 2, // 3
3474
3475 1, // 4
3476 0, // 5 NormPriority
3477 -1, // 6
3478
3479 -2, // 7
3480 -3, // 8
3481 -4, // 9 NearMaxPriority
3482
3483 -5, // 10 MaxPriority
3484
3485 -5 // 11 CriticalPriority
3486 };
3487
3488 static int prio_init() {
3489 if (ThreadPriorityPolicy == 1) {
3490 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3491 // if effective uid is not root. Perhaps, a more elegant way of doing
3492 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3493 if (geteuid() != 0) {
3494 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3495 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3496 }
3497 ThreadPriorityPolicy = 0;
3498 }
3499 }
3500 if (UseCriticalJavaThreadPriority) {
3501 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3502 }
3503 return 0;
3504 }
3505
3506 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3507 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3508
3509 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3510 return (ret == 0) ? OS_OK : OS_ERR;
3511 }
3512
3513 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3514 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3515 *priority_ptr = java_to_os_priority[NormPriority];
3516 return OS_OK;
3517 }
3518
3519 errno = 0;
3520 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3521 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3522 }
3523
3524 // Hint to the underlying OS that a task switch would not be good.
3525 // Void return because it's a hint and can fail.
3526 void os::hint_no_preempt() {}
3527
3528 ////////////////////////////////////////////////////////////////////////////////
3529 // suspend/resume support
3530
3531 // the low-level signal-based suspend/resume support is a remnant from the
3532 // old VM-suspension that used to be for java-suspension, safepoints etc,
3533 // within hotspot. Now there is a single use-case for this:
3534 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3535 // that runs in the watcher thread.
3536 // The remaining code is greatly simplified from the more general suspension
3537 // code that used to be used.
3538 //
3539 // The protocol is quite simple:
3540 // - suspend:
3541 // - sends a signal to the target thread
3542 // - polls the suspend state of the osthread using a yield loop
3543 // - target thread signal handler (SR_handler) sets suspend state
3544 // and blocks in sigsuspend until continued
3545 // - resume:
3546 // - sets target osthread state to continue
3547 // - sends signal to end the sigsuspend loop in the SR_handler
3548 //
3549 // Note that the SR_lock plays no role in this suspend/resume protocol.
3550 //
3551
3552 static void resume_clear_context(OSThread *osthread) {
3553 osthread->set_ucontext(NULL);
3554 osthread->set_siginfo(NULL);
3555
3556 // notify the suspend action is completed, we have now resumed
3557 osthread->sr.clear_suspended();
3558 }
3559
3560 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3561 osthread->set_ucontext(context);
3562 osthread->set_siginfo(siginfo);
3563 }
3564
3565 //
3566 // Handler function invoked when a thread's execution is suspended or
3567 // resumed. We have to be careful that only async-safe functions are
3568 // called here (Note: most pthread functions are not async safe and
3569 // should be avoided.)
3570 //
3571 // Note: sigwait() is a more natural fit than sigsuspend() from an
3572 // interface point of view, but sigwait() prevents the signal hander
3573 // from being run. libpthread would get very confused by not having
3574 // its signal handlers run and prevents sigwait()'s use with the
3575 // mutex granting granting signal.
3576 //
3577 // Currently only ever called on the VMThread
3578 //
3579 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3580 // Save and restore errno to avoid confusing native code with EINTR
3581 // after sigsuspend.
3582 int old_errno = errno;
3583
3584 Thread* thread = Thread::current();
3585 OSThread* osthread = thread->osthread();
3586 assert(thread->is_VM_thread(), "Must be VMThread");
3587 // read current suspend action
3588 int action = osthread->sr.suspend_action();
3589 if (action == os::Linux::SuspendResume::SR_SUSPEND) {
3590 suspend_save_context(osthread, siginfo, context);
3591
3592 // Notify the suspend action is about to be completed. do_suspend()
3593 // waits until SR_SUSPENDED is set and then returns. We will wait
3594 // here for a resume signal and that completes the suspend-other
3595 // action. do_suspend/do_resume is always called as a pair from
3596 // the same thread - so there are no races
3597
3598 // notify the caller
3599 osthread->sr.set_suspended();
3600
3601 sigset_t suspend_set; // signals for sigsuspend()
3602
3603 // get current set of blocked signals and unblock resume signal
3604 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3605 sigdelset(&suspend_set, SR_signum);
3606
3607 // wait here until we are resumed
3608 do {
3609 sigsuspend(&suspend_set);
3610 // ignore all returns until we get a resume signal
3611 } while (osthread->sr.suspend_action() != os::Linux::SuspendResume::SR_CONTINUE);
3612
3613 resume_clear_context(osthread);
3614
3615 } else {
3616 assert(action == os::Linux::SuspendResume::SR_CONTINUE, "unexpected sr action");
3617 // nothing special to do - just leave the handler
3618 }
3619
3620 errno = old_errno;
3621 }
3622
3623
3624 static int SR_initialize() {
3625 struct sigaction act;
3626 char *s;
3627 /* Get signal number to use for suspend/resume */
3628 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3629 int sig = ::strtol(s, 0, 10);
3630 if (sig > 0 || sig < _NSIG) {
3631 SR_signum = sig;
3632 }
3633 }
3634
3635 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3636 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3637
3638 sigemptyset(&SR_sigset);
3639 sigaddset(&SR_sigset, SR_signum);
3640
3641 /* Set up signal handler for suspend/resume */
3642 act.sa_flags = SA_RESTART|SA_SIGINFO;
3643 act.sa_handler = (void (*)(int)) SR_handler;
3644
3645 // SR_signum is blocked by default.
3646 // 4528190 - We also need to block pthread restart signal (32 on all
3647 // supported Linux platforms). Note that LinuxThreads need to block
3648 // this signal for all threads to work properly. So we don't have
3649 // to use hard-coded signal number when setting up the mask.
3650 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3651
3652 if (sigaction(SR_signum, &act, 0) == -1) {
3653 return -1;
3654 }
3655
3656 // Save signal flag
3657 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3658 return 0;
3659 }
3660
3661
3662 // returns true on success and false on error - really an error is fatal
3663 // but this seems the normal response to library errors
3664 static bool do_suspend(OSThread* osthread) {
3665 // mark as suspended and send signal
3666 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_SUSPEND);
3667 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3668 assert_status(status == 0, status, "pthread_kill");
3669
3670 // check status and wait until notified of suspension
3671 if (status == 0) {
3672 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3673 os::yield_all(i);
3674 }
3675 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3676 return true;
3677 }
3678 else {
3679 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3680 return false;
3681 }
3682 }
3683
3684 static void do_resume(OSThread* osthread) {
3685 assert(osthread->sr.is_suspended(), "thread should be suspended");
3686 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_CONTINUE);
3687
3688 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3689 assert_status(status == 0, status, "pthread_kill");
3690 // check status and wait unit notified of resumption
3691 if (status == 0) {
3692 for (int i = 0; osthread->sr.is_suspended(); i++) {
3693 os::yield_all(i);
3694 }
3695 }
3696 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE);
3697 }
3698
3699 ////////////////////////////////////////////////////////////////////////////////
3700 // interrupt support
3701
3702 void os::interrupt(Thread* thread) {
3703 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3704 "possibility of dangling Thread pointer");
3705
3706 OSThread* osthread = thread->osthread();
3707
3708 if (!osthread->interrupted()) {
3709 osthread->set_interrupted(true);
3710 // More than one thread can get here with the same value of osthread,
3711 // resulting in multiple notifications. We do, however, want the store
3712 // to interrupted() to be visible to other threads before we execute unpark().
3713 OrderAccess::fence();
3714 ParkEvent * const slp = thread->_SleepEvent ;
3715 if (slp != NULL) slp->unpark() ;
3716 }
3717
3718 // For JSR166. Unpark even if interrupt status already was set
3719 if (thread->is_Java_thread())
3720 ((JavaThread*)thread)->parker()->unpark();
3721
3722 ParkEvent * ev = thread->_ParkEvent ;
3723 if (ev != NULL) ev->unpark() ;
3724
3725 }
3726
3727 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3728 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3729 "possibility of dangling Thread pointer");
3730
3731 OSThread* osthread = thread->osthread();
3732
3733 bool interrupted = osthread->interrupted();
3734
3735 if (interrupted && clear_interrupted) {
3736 osthread->set_interrupted(false);
3737 // consider thread->_SleepEvent->reset() ... optional optimization
3738 }
3739
3740 return interrupted;
3741 }
3742
3743 ///////////////////////////////////////////////////////////////////////////////////
3744 // signal handling (except suspend/resume)
3745
3746 // This routine may be used by user applications as a "hook" to catch signals.
3747 // The user-defined signal handler must pass unrecognized signals to this
3748 // routine, and if it returns true (non-zero), then the signal handler must
3749 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3750 // routine will never retun false (zero), but instead will execute a VM panic
3751 // routine kill the process.
3752 //
3753 // If this routine returns false, it is OK to call it again. This allows
3754 // the user-defined signal handler to perform checks either before or after
3755 // the VM performs its own checks. Naturally, the user code would be making
3756 // a serious error if it tried to handle an exception (such as a null check
3757 // or breakpoint) that the VM was generating for its own correct operation.
3758 //
3759 // This routine may recognize any of the following kinds of signals:
3760 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3761 // It should be consulted by handlers for any of those signals.
3762 //
3763 // The caller of this routine must pass in the three arguments supplied
3764 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3765 // field of the structure passed to sigaction(). This routine assumes that
3766 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3767 //
3768 // Note that the VM will print warnings if it detects conflicting signal
3769 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3770 //
3771 extern "C" JNIEXPORT int
3772 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3773 void* ucontext, int abort_if_unrecognized);
3774
3775 void signalHandler(int sig, siginfo_t* info, void* uc) {
3776 assert(info != NULL && uc != NULL, "it must be old kernel");
3777 int orig_errno = errno; // Preserve errno value over signal handler.
3778 JVM_handle_linux_signal(sig, info, uc, true);
3779 errno = orig_errno;
3780 }
3781
3782
3783 // This boolean allows users to forward their own non-matching signals
3784 // to JVM_handle_linux_signal, harmlessly.
3785 bool os::Linux::signal_handlers_are_installed = false;
3786
3787 // For signal-chaining
3788 struct sigaction os::Linux::sigact[MAXSIGNUM];
3789 unsigned int os::Linux::sigs = 0;
3790 bool os::Linux::libjsig_is_loaded = false;
3791 typedef struct sigaction *(*get_signal_t)(int);
3792 get_signal_t os::Linux::get_signal_action = NULL;
3793
3794 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3795 struct sigaction *actp = NULL;
3796
3797 if (libjsig_is_loaded) {
3798 // Retrieve the old signal handler from libjsig
3799 actp = (*get_signal_action)(sig);
3800 }
3801 if (actp == NULL) {
3802 // Retrieve the preinstalled signal handler from jvm
3803 actp = get_preinstalled_handler(sig);
3804 }
3805
3806 return actp;
3807 }
3808
3809 static bool call_chained_handler(struct sigaction *actp, int sig,
3810 siginfo_t *siginfo, void *context) {
3811 // Call the old signal handler
3812 if (actp->sa_handler == SIG_DFL) {
3813 // It's more reasonable to let jvm treat it as an unexpected exception
3814 // instead of taking the default action.
3815 return false;
3816 } else if (actp->sa_handler != SIG_IGN) {
3817 if ((actp->sa_flags & SA_NODEFER) == 0) {
3818 // automaticlly block the signal
3819 sigaddset(&(actp->sa_mask), sig);
3820 }
3821
3822 sa_handler_t hand;
3823 sa_sigaction_t sa;
3824 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3825 // retrieve the chained handler
3826 if (siginfo_flag_set) {
3827 sa = actp->sa_sigaction;
3828 } else {
3829 hand = actp->sa_handler;
3830 }
3831
3832 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3833 actp->sa_handler = SIG_DFL;
3834 }
3835
3836 // try to honor the signal mask
3837 sigset_t oset;
3838 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3839
3840 // call into the chained handler
3841 if (siginfo_flag_set) {
3842 (*sa)(sig, siginfo, context);
3843 } else {
3844 (*hand)(sig);
3845 }
3846
3847 // restore the signal mask
3848 pthread_sigmask(SIG_SETMASK, &oset, 0);
3849 }
3850 // Tell jvm's signal handler the signal is taken care of.
3851 return true;
3852 }
3853
3854 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3855 bool chained = false;
3856 // signal-chaining
3857 if (UseSignalChaining) {
3858 struct sigaction *actp = get_chained_signal_action(sig);
3859 if (actp != NULL) {
3860 chained = call_chained_handler(actp, sig, siginfo, context);
3861 }
3862 }
3863 return chained;
3864 }
3865
3866 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3867 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3868 return &sigact[sig];
3869 }
3870 return NULL;
3871 }
3872
3873 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3874 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3875 sigact[sig] = oldAct;
3876 sigs |= (unsigned int)1 << sig;
3877 }
3878
3879 // for diagnostic
3880 int os::Linux::sigflags[MAXSIGNUM];
3881
3882 int os::Linux::get_our_sigflags(int sig) {
3883 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3884 return sigflags[sig];
3885 }
3886
3887 void os::Linux::set_our_sigflags(int sig, int flags) {
3888 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3889 sigflags[sig] = flags;
3890 }
3891
3892 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3893 // Check for overwrite.
3894 struct sigaction oldAct;
3895 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3896
3897 void* oldhand = oldAct.sa_sigaction
3898 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3899 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3900 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3901 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3902 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3903 if (AllowUserSignalHandlers || !set_installed) {
3904 // Do not overwrite; user takes responsibility to forward to us.
3905 return;
3906 } else if (UseSignalChaining) {
3907 // save the old handler in jvm
3908 save_preinstalled_handler(sig, oldAct);
3909 // libjsig also interposes the sigaction() call below and saves the
3910 // old sigaction on it own.
3911 } else {
3912 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3913 "%#lx for signal %d.", (long)oldhand, sig));
3914 }
3915 }
3916
3917 struct sigaction sigAct;
3918 sigfillset(&(sigAct.sa_mask));
3919 sigAct.sa_handler = SIG_DFL;
3920 if (!set_installed) {
3921 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3922 } else {
3923 sigAct.sa_sigaction = signalHandler;
3924 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3925 }
3926 // Save flags, which are set by ours
3927 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3928 sigflags[sig] = sigAct.sa_flags;
3929
3930 int ret = sigaction(sig, &sigAct, &oldAct);
3931 assert(ret == 0, "check");
3932
3933 void* oldhand2 = oldAct.sa_sigaction
3934 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3935 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3936 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3937 }
3938
3939 // install signal handlers for signals that HotSpot needs to
3940 // handle in order to support Java-level exception handling.
3941
3942 void os::Linux::install_signal_handlers() {
3943 if (!signal_handlers_are_installed) {
3944 signal_handlers_are_installed = true;
3945
3946 // signal-chaining
3947 typedef void (*signal_setting_t)();
3948 signal_setting_t begin_signal_setting = NULL;
3949 signal_setting_t end_signal_setting = NULL;
3950 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3951 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3952 if (begin_signal_setting != NULL) {
3953 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3954 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3955 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3956 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3957 libjsig_is_loaded = true;
3958 assert(UseSignalChaining, "should enable signal-chaining");
3959 }
3960 if (libjsig_is_loaded) {
3961 // Tell libjsig jvm is setting signal handlers
3962 (*begin_signal_setting)();
3963 }
3964
3965 set_signal_handler(SIGSEGV, true);
3966 set_signal_handler(SIGPIPE, true);
3967 set_signal_handler(SIGBUS, true);
3968 set_signal_handler(SIGILL, true);
3969 set_signal_handler(SIGFPE, true);
3970 set_signal_handler(SIGXFSZ, true);
3971
3972 if (libjsig_is_loaded) {
3973 // Tell libjsig jvm finishes setting signal handlers
3974 (*end_signal_setting)();
3975 }
3976
3977 // We don't activate signal checker if libjsig is in place, we trust ourselves
3978 // and if UserSignalHandler is installed all bets are off.
3979 // Log that signal checking is off only if -verbose:jni is specified.
3980 if (CheckJNICalls) {
3981 if (libjsig_is_loaded) {
3982 if (PrintJNIResolving) {
3983 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3984 }
3985 check_signals = false;
3986 }
3987 if (AllowUserSignalHandlers) {
3988 if (PrintJNIResolving) {
3989 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3990 }
3991 check_signals = false;
3992 }
3993 }
3994 }
3995 }
3996
3997 // This is the fastest way to get thread cpu time on Linux.
3998 // Returns cpu time (user+sys) for any thread, not only for current.
3999 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4000 // It might work on 2.6.10+ with a special kernel/glibc patch.
4001 // For reference, please, see IEEE Std 1003.1-2004:
4002 // http://www.unix.org/single_unix_specification
4003
4004 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4005 struct timespec tp;
4006 int rc = os::Linux::clock_gettime(clockid, &tp);
4007 assert(rc == 0, "clock_gettime is expected to return 0 code");
4008
4009 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4010 }
4011
4012 /////
4013 // glibc on Linux platform uses non-documented flag
4014 // to indicate, that some special sort of signal
4015 // trampoline is used.
4016 // We will never set this flag, and we should
4017 // ignore this flag in our diagnostic
4018 #ifdef SIGNIFICANT_SIGNAL_MASK
4019 #undef SIGNIFICANT_SIGNAL_MASK
4020 #endif
4021 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4022
4023 static const char* get_signal_handler_name(address handler,
4024 char* buf, int buflen) {
4025 int offset;
4026 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4027 if (found) {
4028 // skip directory names
4029 const char *p1, *p2;
4030 p1 = buf;
4031 size_t len = strlen(os::file_separator());
4032 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4033 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4034 } else {
4035 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4036 }
4037 return buf;
4038 }
4039
4040 static void print_signal_handler(outputStream* st, int sig,
4041 char* buf, size_t buflen) {
4042 struct sigaction sa;
4043
4044 sigaction(sig, NULL, &sa);
4045
4046 // See comment for SIGNIFICANT_SIGNAL_MASK define
4047 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4048
4049 st->print("%s: ", os::exception_name(sig, buf, buflen));
4050
4051 address handler = (sa.sa_flags & SA_SIGINFO)
4052 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4053 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4054
4055 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4056 st->print("SIG_DFL");
4057 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4058 st->print("SIG_IGN");
4059 } else {
4060 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4061 }
4062
4063 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4064
4065 address rh = VMError::get_resetted_sighandler(sig);
4066 // May be, handler was resetted by VMError?
4067 if(rh != NULL) {
4068 handler = rh;
4069 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4070 }
4071
4072 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
4073
4074 // Check: is it our handler?
4075 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4076 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4077 // It is our signal handler
4078 // check for flags, reset system-used one!
4079 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4080 st->print(
4081 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4082 os::Linux::get_our_sigflags(sig));
4083 }
4084 }
4085 st->cr();
4086 }
4087
4088
4089 #define DO_SIGNAL_CHECK(sig) \
4090 if (!sigismember(&check_signal_done, sig)) \
4091 os::Linux::check_signal_handler(sig)
4092
4093 // This method is a periodic task to check for misbehaving JNI applications
4094 // under CheckJNI, we can add any periodic checks here
4095
4096 void os::run_periodic_checks() {
4097
4098 if (check_signals == false) return;
4099
4100 // SEGV and BUS if overridden could potentially prevent
4101 // generation of hs*.log in the event of a crash, debugging
4102 // such a case can be very challenging, so we absolutely
4103 // check the following for a good measure:
4104 DO_SIGNAL_CHECK(SIGSEGV);
4105 DO_SIGNAL_CHECK(SIGILL);
4106 DO_SIGNAL_CHECK(SIGFPE);
4107 DO_SIGNAL_CHECK(SIGBUS);
4108 DO_SIGNAL_CHECK(SIGPIPE);
4109 DO_SIGNAL_CHECK(SIGXFSZ);
4110
4111
4112 // ReduceSignalUsage allows the user to override these handlers
4113 // see comments at the very top and jvm_solaris.h
4114 if (!ReduceSignalUsage) {
4115 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4116 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4117 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4118 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4119 }
4120
4121 DO_SIGNAL_CHECK(SR_signum);
4122 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4123 }
4124
4125 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4126
4127 static os_sigaction_t os_sigaction = NULL;
4128
4129 void os::Linux::check_signal_handler(int sig) {
4130 char buf[O_BUFLEN];
4131 address jvmHandler = NULL;
4132
4133
4134 struct sigaction act;
4135 if (os_sigaction == NULL) {
4136 // only trust the default sigaction, in case it has been interposed
4137 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4138 if (os_sigaction == NULL) return;
4139 }
4140
4141 os_sigaction(sig, (struct sigaction*)NULL, &act);
4142
4143
4144 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4145
4146 address thisHandler = (act.sa_flags & SA_SIGINFO)
4147 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4148 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4149
4150
4151 switch(sig) {
4152 case SIGSEGV:
4153 case SIGBUS:
4154 case SIGFPE:
4155 case SIGPIPE:
4156 case SIGILL:
4157 case SIGXFSZ:
4158 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4159 break;
4160
4161 case SHUTDOWN1_SIGNAL:
4162 case SHUTDOWN2_SIGNAL:
4163 case SHUTDOWN3_SIGNAL:
4164 case BREAK_SIGNAL:
4165 jvmHandler = (address)user_handler();
4166 break;
4167
4168 case INTERRUPT_SIGNAL:
4169 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4170 break;
4171
4172 default:
4173 if (sig == SR_signum) {
4174 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4175 } else {
4176 return;
4177 }
4178 break;
4179 }
4180
4181 if (thisHandler != jvmHandler) {
4182 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4183 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4184 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4185 // No need to check this sig any longer
4186 sigaddset(&check_signal_done, sig);
4187 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4188 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4189 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4190 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4191 // No need to check this sig any longer
4192 sigaddset(&check_signal_done, sig);
4193 }
4194
4195 // Dump all the signal
4196 if (sigismember(&check_signal_done, sig)) {
4197 print_signal_handlers(tty, buf, O_BUFLEN);
4198 }
4199 }
4200
4201 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4202
4203 extern bool signal_name(int signo, char* buf, size_t len);
4204
4205 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4206 if (0 < exception_code && exception_code <= SIGRTMAX) {
4207 // signal
4208 if (!signal_name(exception_code, buf, size)) {
4209 jio_snprintf(buf, size, "SIG%d", exception_code);
4210 }
4211 return buf;
4212 } else {
4213 return NULL;
4214 }
4215 }
4216
4217 // this is called _before_ the most of global arguments have been parsed
4218 void os::init(void) {
4219 char dummy; /* used to get a guess on initial stack address */
4220 // first_hrtime = gethrtime();
4221
4222 // With LinuxThreads the JavaMain thread pid (primordial thread)
4223 // is different than the pid of the java launcher thread.
4224 // So, on Linux, the launcher thread pid is passed to the VM
4225 // via the sun.java.launcher.pid property.
4226 // Use this property instead of getpid() if it was correctly passed.
4227 // See bug 6351349.
4228 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4229
4230 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4231
4232 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4233
4234 init_random(1234567);
4235
4236 ThreadCritical::initialize();
4237
4238 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4239 if (Linux::page_size() == -1) {
4240 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4241 strerror(errno)));
4242 }
4243 init_page_sizes((size_t) Linux::page_size());
4244
4245 Linux::initialize_system_info();
4246
4247 // main_thread points to the aboriginal thread
4248 Linux::_main_thread = pthread_self();
4249
4250 Linux::clock_init();
4251 initial_time_count = os::elapsed_counter();
4252 pthread_mutex_init(&dl_mutex, NULL);
4253
4254 // If the pagesize of the VM is greater than 8K determine the appropriate
4255 // number of initial guard pages. The user can change this with the
4256 // command line arguments, if needed.
4257 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4258 StackYellowPages = 1;
4259 StackRedPages = 1;
4260 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4261 }
4262 }
4263
4264 // To install functions for atexit system call
4265 extern "C" {
4266 static void perfMemory_exit_helper() {
4267 perfMemory_exit();
4268 }
4269 }
4270
4271 // this is called _after_ the global arguments have been parsed
4272 jint os::init_2(void)
4273 {
4274 Linux::fast_thread_clock_init();
4275
4276 // Allocate a single page and mark it as readable for safepoint polling
4277 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4278 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4279
4280 os::set_polling_page( polling_page );
4281
4282 #ifndef PRODUCT
4283 if(Verbose && PrintMiscellaneous)
4284 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4285 #endif
4286
4287 if (!UseMembar) {
4288 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4289 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4290 os::set_memory_serialize_page( mem_serialize_page );
4291
4292 #ifndef PRODUCT
4293 if(Verbose && PrintMiscellaneous)
4294 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4295 #endif
4296 }
4297
4298 os::large_page_init();
4299
4300 // initialize suspend/resume support - must do this before signal_sets_init()
4301 if (SR_initialize() != 0) {
4302 perror("SR_initialize failed");
4303 return JNI_ERR;
4304 }
4305
4306 Linux::signal_sets_init();
4307 Linux::install_signal_handlers();
4308
4309 // Check minimum allowable stack size for thread creation and to initialize
4310 // the java system classes, including StackOverflowError - depends on page
4311 // size. Add a page for compiler2 recursion in main thread.
4312 // Add in 2*BytesPerWord times page size to account for VM stack during
4313 // class initialization depending on 32 or 64 bit VM.
4314 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4315 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4316 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4317
4318 size_t threadStackSizeInBytes = ThreadStackSize * K;
4319 if (threadStackSizeInBytes != 0 &&
4320 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4321 tty->print_cr("\nThe stack size specified is too small, "
4322 "Specify at least %dk",
4323 os::Linux::min_stack_allowed/ K);
4324 return JNI_ERR;
4325 }
4326
4327 // Make the stack size a multiple of the page size so that
4328 // the yellow/red zones can be guarded.
4329 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4330 vm_page_size()));
4331
4332 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4333
4334 Linux::libpthread_init();
4335 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4336 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4337 Linux::glibc_version(), Linux::libpthread_version(),
4338 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4339 }
4340
4341 if (UseNUMA) {
4342 if (!Linux::libnuma_init()) {
4343 UseNUMA = false;
4344 } else {
4345 if ((Linux::numa_max_node() < 1)) {
4346 // There's only one node(they start from 0), disable NUMA.
4347 UseNUMA = false;
4348 }
4349 }
4350 // With SHM large pages we cannot uncommit a page, so there's not way
4351 // we can make the adaptive lgrp chunk resizing work. If the user specified
4352 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4353 // disable adaptive resizing.
4354 if (UseNUMA && UseLargePages && UseSHM) {
4355 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4356 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4357 UseLargePages = false;
4358 } else {
4359 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4360 UseAdaptiveSizePolicy = false;
4361 UseAdaptiveNUMAChunkSizing = false;
4362 }
4363 } else {
4364 UseNUMA = false;
4365 }
4366 }
4367 if (!UseNUMA && ForceNUMA) {
4368 UseNUMA = true;
4369 }
4370 }
4371
4372 if (MaxFDLimit) {
4373 // set the number of file descriptors to max. print out error
4374 // if getrlimit/setrlimit fails but continue regardless.
4375 struct rlimit nbr_files;
4376 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4377 if (status != 0) {
4378 if (PrintMiscellaneous && (Verbose || WizardMode))
4379 perror("os::init_2 getrlimit failed");
4380 } else {
4381 nbr_files.rlim_cur = nbr_files.rlim_max;
4382 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4383 if (status != 0) {
4384 if (PrintMiscellaneous && (Verbose || WizardMode))
4385 perror("os::init_2 setrlimit failed");
4386 }
4387 }
4388 }
4389
4390 // Initialize lock used to serialize thread creation (see os::create_thread)
4391 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4392
4393 // at-exit methods are called in the reverse order of their registration.
4394 // atexit functions are called on return from main or as a result of a
4395 // call to exit(3C). There can be only 32 of these functions registered
4396 // and atexit() does not set errno.
4397
4398 if (PerfAllowAtExitRegistration) {
4399 // only register atexit functions if PerfAllowAtExitRegistration is set.
4400 // atexit functions can be delayed until process exit time, which
4401 // can be problematic for embedded VM situations. Embedded VMs should
4402 // call DestroyJavaVM() to assure that VM resources are released.
4403
4404 // note: perfMemory_exit_helper atexit function may be removed in
4405 // the future if the appropriate cleanup code can be added to the
4406 // VM_Exit VMOperation's doit method.
4407 if (atexit(perfMemory_exit_helper) != 0) {
4408 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4409 }
4410 }
4411
4412 // initialize thread priority policy
4413 prio_init();
4414
4415 return JNI_OK;
4416 }
4417
4418 // this is called at the end of vm_initialization
4419 void os::init_3(void)
4420 {
4421 #ifdef JAVASE_EMBEDDED
4422 // Start the MemNotifyThread
4423 if (LowMemoryProtection) {
4424 MemNotifyThread::start();
4425 }
4426 return;
4427 #endif
4428 }
4429
4430 // Mark the polling page as unreadable
4431 void os::make_polling_page_unreadable(void) {
4432 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4433 fatal("Could not disable polling page");
4434 };
4435
4436 // Mark the polling page as readable
4437 void os::make_polling_page_readable(void) {
4438 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4439 fatal("Could not enable polling page");
4440 }
4441 };
4442
4443 int os::active_processor_count() {
4444 // Linux doesn't yet have a (official) notion of processor sets,
4445 // so just return the number of online processors.
4446 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4447 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4448 return online_cpus;
4449 }
4450
4451 void os::set_native_thread_name(const char *name) {
4452 // Not yet implemented.
4453 return;
4454 }
4455
4456 bool os::distribute_processes(uint length, uint* distribution) {
4457 // Not yet implemented.
4458 return false;
4459 }
4460
4461 bool os::bind_to_processor(uint processor_id) {
4462 // Not yet implemented.
4463 return false;
4464 }
4465
4466 ///
4467
4468 // Suspends the target using the signal mechanism and then grabs the PC before
4469 // resuming the target. Used by the flat-profiler only
4470 ExtendedPC os::get_thread_pc(Thread* thread) {
4471 // Make sure that it is called by the watcher for the VMThread
4472 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4473 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4474
4475 ExtendedPC epc;
4476
4477 OSThread* osthread = thread->osthread();
4478 if (do_suspend(osthread)) {
4479 if (osthread->ucontext() != NULL) {
4480 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4481 } else {
4482 // NULL context is unexpected, double-check this is the VMThread
4483 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4484 }
4485 do_resume(osthread);
4486 }
4487 // failure means pthread_kill failed for some reason - arguably this is
4488 // a fatal problem, but such problems are ignored elsewhere
4489
4490 return epc;
4491 }
4492
4493 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4494 {
4495 if (is_NPTL()) {
4496 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4497 } else {
4498 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4499 // word back to default 64bit precision if condvar is signaled. Java
4500 // wants 53bit precision. Save and restore current value.
4501 int fpu = get_fpu_control_word();
4502 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4503 set_fpu_control_word(fpu);
4504 return status;
4505 }
4506 }
4507
4508 ////////////////////////////////////////////////////////////////////////////////
4509 // debug support
4510
4511 bool os::find(address addr, outputStream* st) {
4512 Dl_info dlinfo;
4513 memset(&dlinfo, 0, sizeof(dlinfo));
4514 if (dladdr(addr, &dlinfo)) {
4515 st->print(PTR_FORMAT ": ", addr);
4516 if (dlinfo.dli_sname != NULL) {
4517 st->print("%s+%#x", dlinfo.dli_sname,
4518 addr - (intptr_t)dlinfo.dli_saddr);
4519 } else if (dlinfo.dli_fname) {
4520 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4521 } else {
4522 st->print("<absolute address>");
4523 }
4524 if (dlinfo.dli_fname) {
4525 st->print(" in %s", dlinfo.dli_fname);
4526 }
4527 if (dlinfo.dli_fbase) {
4528 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4529 }
4530 st->cr();
4531
4532 if (Verbose) {
4533 // decode some bytes around the PC
4534 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4535 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4536 address lowest = (address) dlinfo.dli_sname;
4537 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4538 if (begin < lowest) begin = lowest;
4539 Dl_info dlinfo2;
4540 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4541 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4542 end = (address) dlinfo2.dli_saddr;
4543 Disassembler::decode(begin, end, st);
4544 }
4545 return true;
4546 }
4547 return false;
4548 }
4549
4550 ////////////////////////////////////////////////////////////////////////////////
4551 // misc
4552
4553 // This does not do anything on Linux. This is basically a hook for being
4554 // able to use structured exception handling (thread-local exception filters)
4555 // on, e.g., Win32.
4556 void
4557 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4558 JavaCallArguments* args, Thread* thread) {
4559 f(value, method, args, thread);
4560 }
4561
4562 void os::print_statistics() {
4563 }
4564
4565 int os::message_box(const char* title, const char* message) {
4566 int i;
4567 fdStream err(defaultStream::error_fd());
4568 for (i = 0; i < 78; i++) err.print_raw("=");
4569 err.cr();
4570 err.print_raw_cr(title);
4571 for (i = 0; i < 78; i++) err.print_raw("-");
4572 err.cr();
4573 err.print_raw_cr(message);
4574 for (i = 0; i < 78; i++) err.print_raw("=");
4575 err.cr();
4576
4577 char buf[16];
4578 // Prevent process from exiting upon "read error" without consuming all CPU
4579 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4580
4581 return buf[0] == 'y' || buf[0] == 'Y';
4582 }
4583
4584 int os::stat(const char *path, struct stat *sbuf) {
4585 char pathbuf[MAX_PATH];
4586 if (strlen(path) > MAX_PATH - 1) {
4587 errno = ENAMETOOLONG;
4588 return -1;
4589 }
4590 os::native_path(strcpy(pathbuf, path));
4591 return ::stat(pathbuf, sbuf);
4592 }
4593
4594 bool os::check_heap(bool force) {
4595 return true;
4596 }
4597
4598 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4599 return ::vsnprintf(buf, count, format, args);
4600 }
4601
4602 // Is a (classpath) directory empty?
4603 bool os::dir_is_empty(const char* path) {
4604 DIR *dir = NULL;
4605 struct dirent *ptr;
4606
4607 dir = opendir(path);
4608 if (dir == NULL) return true;
4609
4610 /* Scan the directory */
4611 bool result = true;
4612 char buf[sizeof(struct dirent) + MAX_PATH];
4613 while (result && (ptr = ::readdir(dir)) != NULL) {
4614 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4615 result = false;
4616 }
4617 }
4618 closedir(dir);
4619 return result;
4620 }
4621
4622 // This code originates from JDK's sysOpen and open64_w
4623 // from src/solaris/hpi/src/system_md.c
4624
4625 #ifndef O_DELETE
4626 #define O_DELETE 0x10000
4627 #endif
4628
4629 // Open a file. Unlink the file immediately after open returns
4630 // if the specified oflag has the O_DELETE flag set.
4631 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4632
4633 int os::open(const char *path, int oflag, int mode) {
4634
4635 if (strlen(path) > MAX_PATH - 1) {
4636 errno = ENAMETOOLONG;
4637 return -1;
4638 }
4639 int fd;
4640 int o_delete = (oflag & O_DELETE);
4641 oflag = oflag & ~O_DELETE;
4642
4643 fd = ::open64(path, oflag, mode);
4644 if (fd == -1) return -1;
4645
4646 //If the open succeeded, the file might still be a directory
4647 {
4648 struct stat64 buf64;
4649 int ret = ::fstat64(fd, &buf64);
4650 int st_mode = buf64.st_mode;
4651
4652 if (ret != -1) {
4653 if ((st_mode & S_IFMT) == S_IFDIR) {
4654 errno = EISDIR;
4655 ::close(fd);
4656 return -1;
4657 }
4658 } else {
4659 ::close(fd);
4660 return -1;
4661 }
4662 }
4663
4664 /*
4665 * All file descriptors that are opened in the JVM and not
4666 * specifically destined for a subprocess should have the
4667 * close-on-exec flag set. If we don't set it, then careless 3rd
4668 * party native code might fork and exec without closing all
4669 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4670 * UNIXProcess.c), and this in turn might:
4671 *
4672 * - cause end-of-file to fail to be detected on some file
4673 * descriptors, resulting in mysterious hangs, or
4674 *
4675 * - might cause an fopen in the subprocess to fail on a system
4676 * suffering from bug 1085341.
4677 *
4678 * (Yes, the default setting of the close-on-exec flag is a Unix
4679 * design flaw)
4680 *
4681 * See:
4682 * 1085341: 32-bit stdio routines should support file descriptors >255
4683 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4684 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4685 */
4686 #ifdef FD_CLOEXEC
4687 {
4688 int flags = ::fcntl(fd, F_GETFD);
4689 if (flags != -1)
4690 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4691 }
4692 #endif
4693
4694 if (o_delete != 0) {
4695 ::unlink(path);
4696 }
4697 return fd;
4698 }
4699
4700
4701 // create binary file, rewriting existing file if required
4702 int os::create_binary_file(const char* path, bool rewrite_existing) {
4703 int oflags = O_WRONLY | O_CREAT;
4704 if (!rewrite_existing) {
4705 oflags |= O_EXCL;
4706 }
4707 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4708 }
4709
4710 // return current position of file pointer
4711 jlong os::current_file_offset(int fd) {
4712 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4713 }
4714
4715 // move file pointer to the specified offset
4716 jlong os::seek_to_file_offset(int fd, jlong offset) {
4717 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4718 }
4719
4720 // This code originates from JDK's sysAvailable
4721 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4722
4723 int os::available(int fd, jlong *bytes) {
4724 jlong cur, end;
4725 int mode;
4726 struct stat64 buf64;
4727
4728 if (::fstat64(fd, &buf64) >= 0) {
4729 mode = buf64.st_mode;
4730 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4731 /*
4732 * XXX: is the following call interruptible? If so, this might
4733 * need to go through the INTERRUPT_IO() wrapper as for other
4734 * blocking, interruptible calls in this file.
4735 */
4736 int n;
4737 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4738 *bytes = n;
4739 return 1;
4740 }
4741 }
4742 }
4743 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4744 return 0;
4745 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4746 return 0;
4747 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4748 return 0;
4749 }
4750 *bytes = end - cur;
4751 return 1;
4752 }
4753
4754 int os::socket_available(int fd, jint *pbytes) {
4755 // Linux doc says EINTR not returned, unlike Solaris
4756 int ret = ::ioctl(fd, FIONREAD, pbytes);
4757
4758 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4759 // is expected to return 0 on failure and 1 on success to the jdk.
4760 return (ret < 0) ? 0 : 1;
4761 }
4762
4763 // Map a block of memory.
4764 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4765 char *addr, size_t bytes, bool read_only,
4766 bool allow_exec) {
4767 int prot;
4768 int flags = MAP_PRIVATE;
4769
4770 if (read_only) {
4771 prot = PROT_READ;
4772 } else {
4773 prot = PROT_READ | PROT_WRITE;
4774 }
4775
4776 if (allow_exec) {
4777 prot |= PROT_EXEC;
4778 }
4779
4780 if (addr != NULL) {
4781 flags |= MAP_FIXED;
4782 }
4783
4784 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4785 fd, file_offset);
4786 if (mapped_address == MAP_FAILED) {
4787 return NULL;
4788 }
4789 return mapped_address;
4790 }
4791
4792
4793 // Remap a block of memory.
4794 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4795 char *addr, size_t bytes, bool read_only,
4796 bool allow_exec) {
4797 // same as map_memory() on this OS
4798 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4799 allow_exec);
4800 }
4801
4802
4803 // Unmap a block of memory.
4804 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4805 return munmap(addr, bytes) == 0;
4806 }
4807
4808 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4809
4810 static clockid_t thread_cpu_clockid(Thread* thread) {
4811 pthread_t tid = thread->osthread()->pthread_id();
4812 clockid_t clockid;
4813
4814 // Get thread clockid
4815 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4816 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4817 return clockid;
4818 }
4819
4820 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4821 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4822 // of a thread.
4823 //
4824 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4825 // the fast estimate available on the platform.
4826
4827 jlong os::current_thread_cpu_time() {
4828 if (os::Linux::supports_fast_thread_cpu_time()) {
4829 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4830 } else {
4831 // return user + sys since the cost is the same
4832 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4833 }
4834 }
4835
4836 jlong os::thread_cpu_time(Thread* thread) {
4837 // consistent with what current_thread_cpu_time() returns
4838 if (os::Linux::supports_fast_thread_cpu_time()) {
4839 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4840 } else {
4841 return slow_thread_cpu_time(thread, true /* user + sys */);
4842 }
4843 }
4844
4845 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4846 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4847 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4848 } else {
4849 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4850 }
4851 }
4852
4853 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4854 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4855 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4856 } else {
4857 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4858 }
4859 }
4860
4861 //
4862 // -1 on error.
4863 //
4864
4865 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4866 static bool proc_task_unchecked = true;
4867 static const char *proc_stat_path = "/proc/%d/stat";
4868 pid_t tid = thread->osthread()->thread_id();
4869 char *s;
4870 char stat[2048];
4871 int statlen;
4872 char proc_name[64];
4873 int count;
4874 long sys_time, user_time;
4875 char cdummy;
4876 int idummy;
4877 long ldummy;
4878 FILE *fp;
4879
4880 // The /proc/<tid>/stat aggregates per-process usage on
4881 // new Linux kernels 2.6+ where NPTL is supported.
4882 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4883 // See bug 6328462.
4884 // There possibly can be cases where there is no directory
4885 // /proc/self/task, so we check its availability.
4886 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4887 // This is executed only once
4888 proc_task_unchecked = false;
4889 fp = fopen("/proc/self/task", "r");
4890 if (fp != NULL) {
4891 proc_stat_path = "/proc/self/task/%d/stat";
4892 fclose(fp);
4893 }
4894 }
4895
4896 sprintf(proc_name, proc_stat_path, tid);
4897 fp = fopen(proc_name, "r");
4898 if ( fp == NULL ) return -1;
4899 statlen = fread(stat, 1, 2047, fp);
4900 stat[statlen] = '\0';
4901 fclose(fp);
4902
4903 // Skip pid and the command string. Note that we could be dealing with
4904 // weird command names, e.g. user could decide to rename java launcher
4905 // to "java 1.4.2 :)", then the stat file would look like
4906 // 1234 (java 1.4.2 :)) R ... ...
4907 // We don't really need to know the command string, just find the last
4908 // occurrence of ")" and then start parsing from there. See bug 4726580.
4909 s = strrchr(stat, ')');
4910 if (s == NULL ) return -1;
4911
4912 // Skip blank chars
4913 do s++; while (isspace(*s));
4914
4915 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4916 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4917 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4918 &user_time, &sys_time);
4919 if ( count != 13 ) return -1;
4920 if (user_sys_cpu_time) {
4921 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4922 } else {
4923 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4924 }
4925 }
4926
4927 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4928 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4929 info_ptr->may_skip_backward = false; // elapsed time not wall time
4930 info_ptr->may_skip_forward = false; // elapsed time not wall time
4931 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4932 }
4933
4934 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4935 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4936 info_ptr->may_skip_backward = false; // elapsed time not wall time
4937 info_ptr->may_skip_forward = false; // elapsed time not wall time
4938 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4939 }
4940
4941 bool os::is_thread_cpu_time_supported() {
4942 return true;
4943 }
4944
4945 // System loadavg support. Returns -1 if load average cannot be obtained.
4946 // Linux doesn't yet have a (official) notion of processor sets,
4947 // so just return the system wide load average.
4948 int os::loadavg(double loadavg[], int nelem) {
4949 return ::getloadavg(loadavg, nelem);
4950 }
4951
4952 void os::pause() {
4953 char filename[MAX_PATH];
4954 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4955 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4956 } else {
4957 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4958 }
4959
4960 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4961 if (fd != -1) {
4962 struct stat buf;
4963 ::close(fd);
4964 while (::stat(filename, &buf) == 0) {
4965 (void)::poll(NULL, 0, 100);
4966 }
4967 } else {
4968 jio_fprintf(stderr,
4969 "Could not open pause file '%s', continuing immediately.\n", filename);
4970 }
4971 }
4972
4973
4974 // Refer to the comments in os_solaris.cpp park-unpark.
4975 //
4976 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4977 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4978 // For specifics regarding the bug see GLIBC BUGID 261237 :
4979 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4980 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4981 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4982 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4983 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4984 // and monitorenter when we're using 1-0 locking. All those operations may result in
4985 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4986 // of libpthread avoids the problem, but isn't practical.
4987 //
4988 // Possible remedies:
4989 //
4990 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4991 // This is palliative and probabilistic, however. If the thread is preempted
4992 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4993 // than the minimum period may have passed, and the abstime may be stale (in the
4994 // past) resultin in a hang. Using this technique reduces the odds of a hang
4995 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4996 //
4997 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4998 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4999 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5000 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5001 // thread.
5002 //
5003 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5004 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5005 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5006 // This also works well. In fact it avoids kernel-level scalability impediments
5007 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5008 // timers in a graceful fashion.
5009 //
5010 // 4. When the abstime value is in the past it appears that control returns
5011 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5012 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5013 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5014 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5015 // It may be possible to avoid reinitialization by checking the return
5016 // value from pthread_cond_timedwait(). In addition to reinitializing the
5017 // condvar we must establish the invariant that cond_signal() is only called
5018 // within critical sections protected by the adjunct mutex. This prevents
5019 // cond_signal() from "seeing" a condvar that's in the midst of being
5020 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5021 // desirable signal-after-unlock optimization that avoids futile context switching.
5022 //
5023 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5024 // structure when a condvar is used or initialized. cond_destroy() would
5025 // release the helper structure. Our reinitialize-after-timedwait fix
5026 // put excessive stress on malloc/free and locks protecting the c-heap.
5027 //
5028 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5029 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5030 // and only enabling the work-around for vulnerable environments.
5031
5032 // utility to compute the abstime argument to timedwait:
5033 // millis is the relative timeout time
5034 // abstime will be the absolute timeout time
5035 // TODO: replace compute_abstime() with unpackTime()
5036
5037 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5038 if (millis < 0) millis = 0;
5039 struct timeval now;
5040 int status = gettimeofday(&now, NULL);
5041 assert(status == 0, "gettimeofday");
5042 jlong seconds = millis / 1000;
5043 millis %= 1000;
5044 if (seconds > 50000000) { // see man cond_timedwait(3T)
5045 seconds = 50000000;
5046 }
5047 abstime->tv_sec = now.tv_sec + seconds;
5048 long usec = now.tv_usec + millis * 1000;
5049 if (usec >= 1000000) {
5050 abstime->tv_sec += 1;
5051 usec -= 1000000;
5052 }
5053 abstime->tv_nsec = usec * 1000;
5054 return abstime;
5055 }
5056
5057
5058 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5059 // Conceptually TryPark() should be equivalent to park(0).
5060
5061 int os::PlatformEvent::TryPark() {
5062 for (;;) {
5063 const int v = _Event ;
5064 guarantee ((v == 0) || (v == 1), "invariant") ;
5065 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5066 }
5067 }
5068
5069 void os::PlatformEvent::park() { // AKA "down()"
5070 // Invariant: Only the thread associated with the Event/PlatformEvent
5071 // may call park().
5072 // TODO: assert that _Assoc != NULL or _Assoc == Self
5073 int v ;
5074 for (;;) {
5075 v = _Event ;
5076 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5077 }
5078 guarantee (v >= 0, "invariant") ;
5079 if (v == 0) {
5080 // Do this the hard way by blocking ...
5081 int status = pthread_mutex_lock(_mutex);
5082 assert_status(status == 0, status, "mutex_lock");
5083 guarantee (_nParked == 0, "invariant") ;
5084 ++ _nParked ;
5085 while (_Event < 0) {
5086 status = pthread_cond_wait(_cond, _mutex);
5087 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5088 // Treat this the same as if the wait was interrupted
5089 if (status == ETIME) { status = EINTR; }
5090 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5091 }
5092 -- _nParked ;
5093
5094 _Event = 0 ;
5095 status = pthread_mutex_unlock(_mutex);
5096 assert_status(status == 0, status, "mutex_unlock");
5097 // Paranoia to ensure our locked and lock-free paths interact
5098 // correctly with each other.
5099 OrderAccess::fence();
5100 }
5101 guarantee (_Event >= 0, "invariant") ;
5102 }
5103
5104 int os::PlatformEvent::park(jlong millis) {
5105 guarantee (_nParked == 0, "invariant") ;
5106
5107 int v ;
5108 for (;;) {
5109 v = _Event ;
5110 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5111 }
5112 guarantee (v >= 0, "invariant") ;
5113 if (v != 0) return OS_OK ;
5114
5115 // We do this the hard way, by blocking the thread.
5116 // Consider enforcing a minimum timeout value.
5117 struct timespec abst;
5118 compute_abstime(&abst, millis);
5119
5120 int ret = OS_TIMEOUT;
5121 int status = pthread_mutex_lock(_mutex);
5122 assert_status(status == 0, status, "mutex_lock");
5123 guarantee (_nParked == 0, "invariant") ;
5124 ++_nParked ;
5125
5126 // Object.wait(timo) will return because of
5127 // (a) notification
5128 // (b) timeout
5129 // (c) thread.interrupt
5130 //
5131 // Thread.interrupt and object.notify{All} both call Event::set.
5132 // That is, we treat thread.interrupt as a special case of notification.
5133 // The underlying Solaris implementation, cond_timedwait, admits
5134 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5135 // JVM from making those visible to Java code. As such, we must
5136 // filter out spurious wakeups. We assume all ETIME returns are valid.
5137 //
5138 // TODO: properly differentiate simultaneous notify+interrupt.
5139 // In that case, we should propagate the notify to another waiter.
5140
5141 while (_Event < 0) {
5142 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5143 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5144 pthread_cond_destroy (_cond);
5145 pthread_cond_init (_cond, NULL) ;
5146 }
5147 assert_status(status == 0 || status == EINTR ||
5148 status == ETIME || status == ETIMEDOUT,
5149 status, "cond_timedwait");
5150 if (!FilterSpuriousWakeups) break ; // previous semantics
5151 if (status == ETIME || status == ETIMEDOUT) break ;
5152 // We consume and ignore EINTR and spurious wakeups.
5153 }
5154 --_nParked ;
5155 if (_Event >= 0) {
5156 ret = OS_OK;
5157 }
5158 _Event = 0 ;
5159 status = pthread_mutex_unlock(_mutex);
5160 assert_status(status == 0, status, "mutex_unlock");
5161 assert (_nParked == 0, "invariant") ;
5162 // Paranoia to ensure our locked and lock-free paths interact
5163 // correctly with each other.
5164 OrderAccess::fence();
5165 return ret;
5166 }
5167
5168 void os::PlatformEvent::unpark() {
5169 // Transitions for _Event:
5170 // 0 :=> 1
5171 // 1 :=> 1
5172 // -1 :=> either 0 or 1; must signal target thread
5173 // That is, we can safely transition _Event from -1 to either
5174 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5175 // unpark() calls.
5176 // See also: "Semaphores in Plan 9" by Mullender & Cox
5177 //
5178 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5179 // that it will take two back-to-back park() calls for the owning
5180 // thread to block. This has the benefit of forcing a spurious return
5181 // from the first park() call after an unpark() call which will help
5182 // shake out uses of park() and unpark() without condition variables.
5183
5184 if (Atomic::xchg(1, &_Event) >= 0) return;
5185
5186 // Wait for the thread associated with the event to vacate
5187 int status = pthread_mutex_lock(_mutex);
5188 assert_status(status == 0, status, "mutex_lock");
5189 int AnyWaiters = _nParked;
5190 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5191 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5192 AnyWaiters = 0;
5193 pthread_cond_signal(_cond);
5194 }
5195 status = pthread_mutex_unlock(_mutex);
5196 assert_status(status == 0, status, "mutex_unlock");
5197 if (AnyWaiters != 0) {
5198 status = pthread_cond_signal(_cond);
5199 assert_status(status == 0, status, "cond_signal");
5200 }
5201
5202 // Note that we signal() _after dropping the lock for "immortal" Events.
5203 // This is safe and avoids a common class of futile wakeups. In rare
5204 // circumstances this can cause a thread to return prematurely from
5205 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5206 // simply re-test the condition and re-park itself.
5207 }
5208
5209
5210 // JSR166
5211 // -------------------------------------------------------
5212
5213 /*
5214 * The solaris and linux implementations of park/unpark are fairly
5215 * conservative for now, but can be improved. They currently use a
5216 * mutex/condvar pair, plus a a count.
5217 * Park decrements count if > 0, else does a condvar wait. Unpark
5218 * sets count to 1 and signals condvar. Only one thread ever waits
5219 * on the condvar. Contention seen when trying to park implies that someone
5220 * is unparking you, so don't wait. And spurious returns are fine, so there
5221 * is no need to track notifications.
5222 */
5223
5224 #define MAX_SECS 100000000
5225 /*
5226 * This code is common to linux and solaris and will be moved to a
5227 * common place in dolphin.
5228 *
5229 * The passed in time value is either a relative time in nanoseconds
5230 * or an absolute time in milliseconds. Either way it has to be unpacked
5231 * into suitable seconds and nanoseconds components and stored in the
5232 * given timespec structure.
5233 * Given time is a 64-bit value and the time_t used in the timespec is only
5234 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5235 * overflow if times way in the future are given. Further on Solaris versions
5236 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5237 * number of seconds, in abstime, is less than current_time + 100,000,000.
5238 * As it will be 28 years before "now + 100000000" will overflow we can
5239 * ignore overflow and just impose a hard-limit on seconds using the value
5240 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5241 * years from "now".
5242 */
5243
5244 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5245 assert (time > 0, "convertTime");
5246
5247 struct timeval now;
5248 int status = gettimeofday(&now, NULL);
5249 assert(status == 0, "gettimeofday");
5250
5251 time_t max_secs = now.tv_sec + MAX_SECS;
5252
5253 if (isAbsolute) {
5254 jlong secs = time / 1000;
5255 if (secs > max_secs) {
5256 absTime->tv_sec = max_secs;
5257 }
5258 else {
5259 absTime->tv_sec = secs;
5260 }
5261 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5262 }
5263 else {
5264 jlong secs = time / NANOSECS_PER_SEC;
5265 if (secs >= MAX_SECS) {
5266 absTime->tv_sec = max_secs;
5267 absTime->tv_nsec = 0;
5268 }
5269 else {
5270 absTime->tv_sec = now.tv_sec + secs;
5271 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5272 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5273 absTime->tv_nsec -= NANOSECS_PER_SEC;
5274 ++absTime->tv_sec; // note: this must be <= max_secs
5275 }
5276 }
5277 }
5278 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5279 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5280 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5281 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5282 }
5283
5284 void Parker::park(bool isAbsolute, jlong time) {
5285 // Ideally we'd do something useful while spinning, such
5286 // as calling unpackTime().
5287
5288 // Optional fast-path check:
5289 // Return immediately if a permit is available.
5290 // We depend on Atomic::xchg() having full barrier semantics
5291 // since we are doing a lock-free update to _counter.
5292 if (Atomic::xchg(0, &_counter) > 0) return;
5293
5294 Thread* thread = Thread::current();
5295 assert(thread->is_Java_thread(), "Must be JavaThread");
5296 JavaThread *jt = (JavaThread *)thread;
5297
5298 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5299 // Check interrupt before trying to wait
5300 if (Thread::is_interrupted(thread, false)) {
5301 return;
5302 }
5303
5304 // Next, demultiplex/decode time arguments
5305 timespec absTime;
5306 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5307 return;
5308 }
5309 if (time > 0) {
5310 unpackTime(&absTime, isAbsolute, time);
5311 }
5312
5313
5314 // Enter safepoint region
5315 // Beware of deadlocks such as 6317397.
5316 // The per-thread Parker:: mutex is a classic leaf-lock.
5317 // In particular a thread must never block on the Threads_lock while
5318 // holding the Parker:: mutex. If safepoints are pending both the
5319 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5320 ThreadBlockInVM tbivm(jt);
5321
5322 // Don't wait if cannot get lock since interference arises from
5323 // unblocking. Also. check interrupt before trying wait
5324 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5325 return;
5326 }
5327
5328 int status ;
5329 if (_counter > 0) { // no wait needed
5330 _counter = 0;
5331 status = pthread_mutex_unlock(_mutex);
5332 assert (status == 0, "invariant") ;
5333 // Paranoia to ensure our locked and lock-free paths interact
5334 // correctly with each other and Java-level accesses.
5335 OrderAccess::fence();
5336 return;
5337 }
5338
5339 #ifdef ASSERT
5340 // Don't catch signals while blocked; let the running threads have the signals.
5341 // (This allows a debugger to break into the running thread.)
5342 sigset_t oldsigs;
5343 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5344 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5345 #endif
5346
5347 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5348 jt->set_suspend_equivalent();
5349 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5350
5351 if (time == 0) {
5352 status = pthread_cond_wait (_cond, _mutex) ;
5353 } else {
5354 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5355 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5356 pthread_cond_destroy (_cond) ;
5357 pthread_cond_init (_cond, NULL);
5358 }
5359 }
5360 assert_status(status == 0 || status == EINTR ||
5361 status == ETIME || status == ETIMEDOUT,
5362 status, "cond_timedwait");
5363
5364 #ifdef ASSERT
5365 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5366 #endif
5367
5368 _counter = 0 ;
5369 status = pthread_mutex_unlock(_mutex) ;
5370 assert_status(status == 0, status, "invariant") ;
5371 // Paranoia to ensure our locked and lock-free paths interact
5372 // correctly with each other and Java-level accesses.
5373 OrderAccess::fence();
5374
5375 // If externally suspended while waiting, re-suspend
5376 if (jt->handle_special_suspend_equivalent_condition()) {
5377 jt->java_suspend_self();
5378 }
5379 }
5380
5381 void Parker::unpark() {
5382 int s, status ;
5383 status = pthread_mutex_lock(_mutex);
5384 assert (status == 0, "invariant") ;
5385 s = _counter;
5386 _counter = 1;
5387 if (s < 1) {
5388 if (WorkAroundNPTLTimedWaitHang) {
5389 status = pthread_cond_signal (_cond) ;
5390 assert (status == 0, "invariant") ;
5391 status = pthread_mutex_unlock(_mutex);
5392 assert (status == 0, "invariant") ;
5393 } else {
5394 status = pthread_mutex_unlock(_mutex);
5395 assert (status == 0, "invariant") ;
5396 status = pthread_cond_signal (_cond) ;
5397 assert (status == 0, "invariant") ;
5398 }
5399 } else {
5400 pthread_mutex_unlock(_mutex);
5401 assert (status == 0, "invariant") ;
5402 }
5403 }
5404
5405
5406 extern char** environ;
5407
5408 #ifndef __NR_fork
5409 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5410 #endif
5411
5412 #ifndef __NR_execve
5413 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5414 #endif
5415
5416 // Run the specified command in a separate process. Return its exit value,
5417 // or -1 on failure (e.g. can't fork a new process).
5418 // Unlike system(), this function can be called from signal handler. It
5419 // doesn't block SIGINT et al.
5420 int os::fork_and_exec(char* cmd) {
5421 const char * argv[4] = {"sh", "-c", cmd, NULL};
5422
5423 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5424 // pthread_atfork handlers and reset pthread library. All we need is a
5425 // separate process to execve. Make a direct syscall to fork process.
5426 // On IA64 there's no fork syscall, we have to use fork() and hope for
5427 // the best...
5428 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5429 IA64_ONLY(fork();)
5430
5431 if (pid < 0) {
5432 // fork failed
5433 return -1;
5434
5435 } else if (pid == 0) {
5436 // child process
5437
5438 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5439 // first to kill every thread on the thread list. Because this list is
5440 // not reset by fork() (see notes above), execve() will instead kill
5441 // every thread in the parent process. We know this is the only thread
5442 // in the new process, so make a system call directly.
5443 // IA64 should use normal execve() from glibc to match the glibc fork()
5444 // above.
5445 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5446 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5447
5448 // execve failed
5449 _exit(-1);
5450
5451 } else {
5452 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5453 // care about the actual exit code, for now.
5454
5455 int status;
5456
5457 // Wait for the child process to exit. This returns immediately if
5458 // the child has already exited. */
5459 while (waitpid(pid, &status, 0) < 0) {
5460 switch (errno) {
5461 case ECHILD: return 0;
5462 case EINTR: break;
5463 default: return -1;
5464 }
5465 }
5466
5467 if (WIFEXITED(status)) {
5468 // The child exited normally; get its exit code.
5469 return WEXITSTATUS(status);
5470 } else if (WIFSIGNALED(status)) {
5471 // The child exited because of a signal
5472 // The best value to return is 0x80 + signal number,
5473 // because that is what all Unix shells do, and because
5474 // it allows callers to distinguish between process exit and
5475 // process death by signal.
5476 return 0x80 + WTERMSIG(status);
5477 } else {
5478 // Unknown exit code; pass it through
5479 return status;
5480 }
5481 }
5482 }
5483
5484 // is_headless_jre()
5485 //
5486 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5487 // in order to report if we are running in a headless jre
5488 //
5489 // Since JDK8 xawt/libmawt.so was moved into the same directory
5490 // as libawt.so, and renamed libawt_xawt.so
5491 //
5492 bool os::is_headless_jre() {
5493 struct stat statbuf;
5494 char buf[MAXPATHLEN];
5495 char libmawtpath[MAXPATHLEN];
5496 const char *xawtstr = "/xawt/libmawt.so";
5497 const char *new_xawtstr = "/libawt_xawt.so";
5498 char *p;
5499
5500 // Get path to libjvm.so
5501 os::jvm_path(buf, sizeof(buf));
5502
5503 // Get rid of libjvm.so
5504 p = strrchr(buf, '/');
5505 if (p == NULL) return false;
5506 else *p = '\0';
5507
5508 // Get rid of client or server
5509 p = strrchr(buf, '/');
5510 if (p == NULL) return false;
5511 else *p = '\0';
5512
5513 // check xawt/libmawt.so
5514 strcpy(libmawtpath, buf);
5515 strcat(libmawtpath, xawtstr);
5516 if (::stat(libmawtpath, &statbuf) == 0) return false;
5517
5518 // check libawt_xawt.so
5519 strcpy(libmawtpath, buf);
5520 strcat(libmawtpath, new_xawtstr);
5521 if (::stat(libmawtpath, &statbuf) == 0) return false;
5522
5523 return true;
5524 }
5525
5526 // Get the default path to the core file
5527 // Returns the length of the string
5528 int os::get_core_path(char* buffer, size_t bufferSize) {
5529 const char* p = get_current_directory(buffer, bufferSize);
5530
5531 if (p == NULL) {
5532 assert(p != NULL, "failed to get current directory");
5533 return 0;
5534 }
5535
5536 return strlen(buffer);
5537 }
5538
5539 #ifdef JAVASE_EMBEDDED
5540 //
5541 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5542 //
5543 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5544
5545 // ctor
5546 //
5547 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5548 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5549 _fd = fd;
5550
5551 if (os::create_thread(this, os::os_thread)) {
5552 _memnotify_thread = this;
5553 os::set_priority(this, NearMaxPriority);
5554 os::start_thread(this);
5555 }
5556 }
5557
5558 // Where all the work gets done
5559 //
5560 void MemNotifyThread::run() {
5561 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5562
5563 // Set up the select arguments
5564 fd_set rfds;
5565 if (_fd != -1) {
5566 FD_ZERO(&rfds);
5567 FD_SET(_fd, &rfds);
5568 }
5569
5570 // Now wait for the mem_notify device to wake up
5571 while (1) {
5572 // Wait for the mem_notify device to signal us..
5573 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5574 if (rc == -1) {
5575 perror("select!\n");
5576 break;
5577 } else if (rc) {
5578 //ssize_t free_before = os::available_memory();
5579 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5580
5581 // The kernel is telling us there is not much memory left...
5582 // try to do something about that
5583
5584 // If we are not already in a GC, try one.
5585 if (!Universe::heap()->is_gc_active()) {
5586 Universe::heap()->collect(GCCause::_allocation_failure);
5587
5588 //ssize_t free_after = os::available_memory();
5589 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5590 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5591 }
5592 // We might want to do something like the following if we find the GC's are not helping...
5593 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5594 }
5595 }
5596 }
5597
5598 //
5599 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5600 //
5601 void MemNotifyThread::start() {
5602 int fd;
5603 fd = open ("/dev/mem_notify", O_RDONLY, 0);
5604 if (fd < 0) {
5605 return;
5606 }
5607
5608 if (memnotify_thread() == NULL) {
5609 new MemNotifyThread(fd);
5610 }
5611 }
5612 #endif // JAVASE_EMBEDDED